Volume 91 July 2006

Number 1

Published monthly ISSN 0022-3514

by the
American Psychological
Association

Journal of
Personality
and
Social Psychology
ATTITUDES AND SOCIAL COGNITION
Charles M. Judd, Editor
Dacher Keltner, Associate Editor
Anne Maass, Associate Editor
Bernd Wittenbrink, Associate Editor
Vincent Yzerbyt, Associate Editor
INTERPERSONAL RELATIONS AND GROUP PROCESSES
John F. Dovidio, Editor
Daphne Blunt Bugental, Associate Editor
Jacques-Philippe Leyens, Associate Editor
Antony Manstead, Associate Editor
Cynthia L. Pickett, Associate Editor
Jeffry A. Simpson, Associate Editor
Scott Tindale, Associate Editor
Jacquie D. Vorauer, Associate Editor
PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES
Charles S. Carver, Editor
Tim Kasser, Associate Editor
Mario Mikulincer, Associate Editor
Eva M. Pomerantz, Associate Editor
Richard W. Robins, Associate Editor
Gerard Saucier, Associate Editor
www.apa.org/journals/psp.html Thomas A. Widiger, Associate Editor
The Journal of Personality and Social Psychology publishes original
papers in all areas of personality and social psychology. It empha-
sizes empirical reports but may include specialized theoretical,
methodological, and review papers. The journal is divided into three
independently edited sections:
f ATTITUDES AND SOCIAL COGNITION addresses those domains of social
behavior in which cognition plays a major role, including the interface
of cognition with overt behavior, affect, and motivation. Among topics
covered are the formation, change, and utilization of attitudes, attribu-
tions, and stereotypes, person memory, self-regulation, and the origins
and consequences of moods and emotions insofar as these interact
with cognition. Of interest also is the influence of cognition and its
various interfaces on significant social phenomena such as persuasion,
communication, prejudice, social development, and cultural trends.
f INTERPERSONAL RELATIONS AND GROUP PROCESSES focuses on psycho-
logical and structural features of interaction in dyads and groups.
Appropriate to this section are papers on the nature and dynamics of
interactions and social relationships, including interpersonal attraction,
communication, emotion, and relationship development, and on group
and organizational processes such as social influence, group decision
making and task performance, intergroup relations, and aggression,
prosocial behavior and other types of social behavior.
f PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES publishes re-
search on all aspects of personality psychology. It includes studies of
individual differences and basic processes in behavior, emotions, cop-
ing, health, motivation, and other phenomena that reflect personality.
Articles in areas such as personality structure, personality develop-
ment, and personality assessment are also appropriate to this section
of the journal, as are studies of the interplay of culture and personality
and manifestations of personality in everyday behavior.
Manuscripts: Submit manuscripts to the appropriate section editor
according to the above definitions and according to the Instructions
to Authors. Section editors reserve the right to redirect papers among
themselves as appropriate unless an author specifically requests oth-
erwise. Rejection by one section editor is considered rejection by all;
therefore a manuscript rejected by one section editor should not be
submitted to another. The opinions and statements published are the
responsibility of the authors, and such opinions and statements do not
necessarily represent the policies of APA or the views of the editors.
Section editors’ addresses appear below:
ATTITUDES AND SOCIAL COGNITION
Charles M. Judd, Editor
c/o Laurie Hawkins
Department of Psychology
University of Colorado
UCB 345
Boulder, CO 80309
INTERPERSONAL RELATIONS AND GROUP
PROCESSES
John F. Dovidio, Editor
Department of Psychology
University of Connecticut
406 Babbidge Road
Storrs, CT 06269-1020
PERSONALITY PROCESSES AND INDIVIDUAL
DIFFERENCES
Charles S. Carver, Editor
ATTN: JPSP: PPID
Department of Psychology
University of Miami
P.O. Box 248185
Coral Gables, FL 33124-0751
Change of Address: Send change of address notice and a recent
mailing label to the attention of the Subscriptions Department, American
Psychological Association, 30 days prior to the actual change of address.
The paper in this journal meets or exceeds EPA guidelines for recycled paper. Since 1986, this journal has been printed on acid-free paper.
APA will not replace undelivered copies resulting from address changes;
journals will be forwarded only if subscribers notify the local post office
in writing that they will guarantee periodicals forwarding postage.
Electronic access: APA members who subscribe to this journal have
automatic access to a 3-year file of the journal in the PsycARTICLES姞
full-text database. See http://members.apa.org/access.
Reprints: Authors may order reprints of their articles from the
printer when they receive proofs.
Single Issues, Back Issues, and Back Volumes: For information
regarding back issues or back volumes write to Order Department,
American Psychological Association, 750 First Street, NE, Washing-
ton, DC 20002-4242.
Microform Editions: For information regarding microform editions,
write to University Microfilms, Ann Arbor, MI 48106.
Copyright and Permission: Those who wish to reuse APA-
copyrighted material in a non-APA publication must secure from APA
and the author of reproduced material written permission to reproduce
a journal article in full or journal text of more than 500 words. APA
normally grants permission contingent upon like permission of the
author, inclusion of the APA copyright notice on the first page of
reproduced material, and payment of a fee of $20 per page. Permission
from APA and fees are waived for those who wish to reproduce a single
table or figure from a journal for use in a print product, provided the
author’s permission is obtained and full credit is given to APA as
copyright holder and to the author through a complete citation. (Re-
questers requiring written permission for commercial use of a single
table or figure will be assessed a $25 service fee.) Permission and fees
are waived for authors who wish to reproduce their own material for
personal use; fees only are waived for authors who wish to use more
than a single table or figure of their own material commercially (but for
use in edited books, fees are waived for the author only if serving as the
book editor). Permission and fees are waived for the photocopying of
isolated journal articles for nonprofit classroom or library reserve use by
instructors and educational institutions. A permission fee may be
charged to the requester if students are charged for the material,
multiple articles are copied, or large-scale copying is involved (e.g., for
course packs). Access services may use unedited abstracts without
the permission of APA or the author. Libraries are permitted to photo-
copy beyond the limits of U.S. copyright law: (1) post-1977 articles,
provided the per-copy fee in the code for this journal (0022-3514/06/
$12.00) is paid through the Copyright Clearance Center, 222 Rose-
wood Drive, Danvers, MA 01923; (2) pre-1978 articles, provided that
the per-copy fee stated in the Publishers’ Fee List is paid through the
Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA
01923. Address requests for reprint permission to the Permissions
Office, American Psychological Association, 750 First Street, NE,
Washington, DC 20002-4242.
APA Journal Staff: Susan J. A. Harris, Senior Director, Journals Pro-
gram; Greg Long, Production Account Manager; Julie Palmer-
Hoffman, Manuscript Editor; Jodi Ashcraft, Advertising Sales Manager.
Journal of Personality and Social Psychology (ISSN 0022-3514)
is published monthly in two volumes per year by the American
Psychological Association, 750 First Street, NE, Washington, DC
20002-4242. Subscriptions are available on a calendar year basis
only (January through December). The 2006 rates follow: Non-
member Individual: $421 Domestic, $464 Foreign, $491 Air Mail.
Institutional: $1,249 Domestic, $1,340 Foreign, $1,367 Air Mail.
APA Member: $202. Write to Subscriptions Department, American
Psychological Association, 750 First Street, NE, Washington, DC
20002-4242. Printed in the U.S.A. Periodicals postage paid at
Washington, DC, and at additional mailing offices. POSTMASTER:
Send address changes to Journal of Personality and Social Psy-
chology, 750 First Street, NE, Washington, DC 20002-4242.
 APA DICTIONARY HANDBOOK OF COUNSELING
OF PSYCHOLOGY AND PSYCHOTHERAPY WITH
Editor-in-Chief: Gary R. VandenBos LESBIAN, GAY, BISEXUAL,
2006. 1,008 pages. Hardcover. AND TRANSGENDER CLIENTS
List: $49.95 • APA Member/Affiliate: $39.95 Second Edition
ISBN 1-59147-380-2
Edited by Kathleen J. Bieschke,
ISBN-13: 978-1-59147-380-0
Ruperto M. Perez, and Kurt A. DeBord
Item # 4311007
2007. 464 pages. Hardcover.
List: $79.95 • APA Member/Affiliate: $49.95
BECOMING CULTURALLY ISBN 1-59147-421-3
ORIENTED ISBN-13: 978-1-59147-421-0
Practical Advice for Psychologists Item # 4317113
and Educators
Nadya A. Fouad and Patricia Arredondo INTERVENING IN
2007. 208 pages. Hardcover.
List: $49.95 • APA Member/Affiliate: $39.95
CHILDREN’S LIVES
An Ecological, Family-Centered
ISBN 1-59147-424-8
ISBN-13: 978-1-59147-424-1 Approach to Mental Health Care
Item # 4317114 Thomas J. Dishion
and Elizabeth A. Stormshak
2007. 320 pages. Hardcover.
SPIRITUAL APPROACHES IN List: $69.95 • APA Member/Affiliate: $49.95
THE TREATMENT OF WOMEN ISBN 1-59147-428-0
WITH EATING DISORDERS
ISBN-13: 978-1-59147-428-9
P. Scott Richards, Randy K. Hardman,
Item # 4317115
and Michael E. Berrett
2007. 312 pages. Hardcover.
List: $59.95 • APA Member/Affiliate: $49.95 PREVENTING YOUTH
ISBN 1-59147-393-4 SUBSTANCE ABUSE
ISBN-13: 978-1-59147-393-0 Science-Based Programs
Item # 4317103 for Children and Adolescents
Edited by Patrick Tolan, José Szapocznik,
GRADUATE STUDY and Soledad Sambrano
IN PSYCHOLOGY, 2007 2007. 264 pages. Hardcover.
2007. 832 pages. Paperback. List: $69.95 • APA Member/Affiliate: $49.95
List: $24.95 • APA Member/Affiliate: $21.95 ISBN 1-59147-307-1
ISBN 1-59147-423-X ISBN-13: 978-1-59147-307-7
ISBN-13: 978-1-59147-423-4 Item # 4316058
Item # 4270090
SCIENTIFIC JURY SELECTION
DISORDERS OF THE SELF Joel D. Lieberman and Bruce D. Sales
A Personality-Guided Approach 2007. 264 pages. Hardcover.
Marshall L. Silverstein List: $79.95 • APA Member/Affiliate: $49.95
2007. 320 pages. Hardcover. ISBN 1-59147-427-2
List: $69.95 • APA Member/Affiliate: $49.95 ISBN-13: 978-1-59147-427-2
ISBN 1-59147-430-2 Item # 4316081
ISBN-13: 978-1-59147-430-2
Item # 4317116 PRIMATE PERSPECTIVES ON
BEHAVIOR AND COGNITION
CHILD DEVELOPMENT Edited by David A. Washburn
AND SOCIAL POLICY 2007. 368 pages. Hardcover.
Knowledge for Action List: $79.95 • APA Member/Affiliate: $49.95
Edited by J. Lawrence Aber, Sandra J. Bishop- ISBN 1-59147-422-1
Josef, Stephanie M. Jones, Kathryn Taaffe ISBN-13: 978-1-59147-422-7
McLearn, and Deborah A. Phillips Item # 4318035
2007. 352 pages. Hardcover.
List: $79.95 • APA Member/Affiliate: $49.95
ISBN 1-59147-425-6
ISBN-13: 978-1-59147-425-8
Item # 4318036
AD0456

1-800-374-2721 • WWW.APA.ORG/BOOKS
 Available

apa books
A Landmark Reference That Defines the Lexicon of Psychology
July 2006

APA Dictionary of Psychology

Editor-in-Chief: Gary R. VandenBos, PhD
The American Psychological Association is proud to announce the
publication of an invaluable addition to your reference shelf, one that
represents a major scholarly and editorial undertaking. With over 25,000
terms and definitions, the APA Dictionary of Psychology encompasses
all areas of research and application, and includes coverage of
concepts, processes, and therapies across all the major subdisciplines
of psychology. Ten years in the making and edited by a distinguished
editorial board of nearly 100 psychological scholars, researchers and
practitioners, the Dictionary is destined to become the most authoritative
reference of its kind.

Academicians, researchers, clinicians, undergraduates and graduate

students, and professionals in allied mental health, education, medicine,
and law, as well as academic and public libraries, will find the Dictionary
essential.

2006. Hardcover. 1,008 pages.

List: $49.95 | APA Member/Affiliate: $39.95
ISBN 1-59147-380-2 | Item # 4311007

The APA Dictionary of Psychology includes

• 25,000 entries offering clear and authoritative definitions

• Thousands of incisive cross-references directing the user to synonyms and

antonyms, acronyms and abbreviations, and related terms and concepts that
deepen the user’s understanding of related topics

• Balanced coverage of over 100 subject areas across the field of psychology
including clinical, experimental, neuropsychology, cognitive, personality
and social, developmental, health, psychopharmacology, methodology and
statistics, and many others

• Entries include nearly 8,000 terms from the APA’s Thesaurus of

Psychological Index Terms® which helps students and researchers refine
their APA database searches (such as the flagship PsycINFO® bibliographic
database’s 2+ million records)
800-374-2721
www.apa.org/books
• “A Guide to Use” and “Quick Guide to Format” that together explain
important stylistic and format features to help readers most effectively
use the Dictionary

• Each of four appendices gathers terms into a thematic summary listing,

covering (1) biographies; (2) institutions, associations and organizations;
(3) psychological therapies and interventions; and (4) psychological tests
and assessment instruments

AD0470
 Journal of

Personality
and

Social Psychology
www.apa.org/journals/psp.html

July 2006
Copyright © 2006 by the
VOLUME 91 NUMBER 1 American Psychological Association

Attitudes and Social Cognition

1 Alone but Feeling No Pain: Effects of Social Exclusion on Physical Pain Tolerance and Pain Threshold,
Affective Forecasting, and Interpersonal Empathy
C. Nathan DeWall and Roy F. Baumeister

16 The Single Category Implicit Association Test as a Measure of Implicit Social Cognition
Andrew Karpinski and Ross B. Steinman

33 Thinking Within the Box: The Relational Processing Style Elicited by Counterfactual Mind-Sets
Laura J. Kray, Adam D. Galinsky, and Elaine M. Wong

49 Self-Regulatory Processes Defend Against the Threat of Death: Effects of Self-Control Depletion and
Trait Self-Control on Thoughts and Fears of Dying
Matthew T. Gailliot, Brandon J. Schmeichel, and Roy F. Baumeister

Interpersonal Relations and Group Processes

63 Peacocks, Picasso, and Parental Investment: The Effects of Romantic Motives on Creativity
Vladas Griskevicius, Robert B. Cialdini, and Douglas T. Kenrick

77 Navigating the Interdependence Dilemma: Attachment Goals and the Use of Communal Norms With
Potential Close Others
Jennifer A. Bartz and John E. Lydon

97 Intergroup Helping as Status Relations: Effects of Status Stability, Identification, and Type of Help on
Receptivity to High-Status Group’s Help
Arie Nadler and Samer Halabi

111 Information Quantity and Quality Affect the Realistic Accuracy of Personality Judgment
Tera D. Letzring, Shannon M. Wells, and David C. Funder

124 Supplication and Appeasement in Conflict and Negotiation: The Interpersonal Effects of Disappointment,
Worry, Guilt, and Regret
Gerben A. Van Kleef, Carsten K. W. De Dreu, and Antony S. R. Manstead

Personality Processes and Individual Differences

143 Optimism in Close Relationships: How Seeing Things in a Positive Light Makes Them So
Sanjay Srivastava, Kelly M. McGonigal, Jane M. Richards, Emily A. Butler, and James J. Gross

(contents continue)
154 Discrepancies Between Explicit and Implicit Self-Concepts: Consequences for Information Processing
Pablo Briñol, Richard E. Petty, and S. Christian Wheeler

171 Investigating the Dopaminergic Basis of Extraversion in Humans: A Multilevel Approach

Jan Wacker, Mira-Lynn Chavanon, and Gerhard Stemmler

188 Possible Selves and Academic Outcomes: How and When Possible Selves Impel Action
Daphna Oyserman, Deborah Bybee, and Kathy Terry

Other
15 American Psychological Association Subscription Claims Information
ii E-Mail Notification of Your Latest Issue Online!
96 Instructions to Authors
62 Subscription Order Form

ii
154 Discrepancies Between Explicit and Implicit Self-Concepts: Consequences for Information Processing
Pablo Briñol, Richard E. Petty, and S. Christian Wheeler

171 Investigating the Dopaminergic Basis of Extraversion in Humans: A Multilevel Approach

Jan Wacker, Mira-Lynn Chavanon, and Gerhard Stemmler

188 Possible Selves and Academic Outcomes: How and When Possible Selves Impel Action
Daphna Oyserman, Deborah Bybee, and Kathy Terry

Other
15 American Psychological Association Subscription Claims Information
ii E-Mail Notification of Your Latest Issue Online!
96 Instructions to Authors
62 Subscription Order Form

ii
 ATTITUDES AND SOCIAL COGNITION
CHARLES M. JUDD, Editor
University of Colorado at Boulder
ASSOCIATE EDITORS ALICE H. EAGLY NIRA LIBERMAN LINDA SKITKA
Northwestern University Tel Aviv University, Tel Aviv, Israel University of Illinois at Chicago
DACHER KELTNER
University of California, Berkeley NICHOLAS EPLEY DIANE M. MACKIE JOHN SKOWRONSKI
University of Chicago University of California, Santa Barbara Northern Illinois University
ANNE MAASS
Università di Padova, Padova, Italy RUSSELL H. FAZIO NEIL MACRAE ELIOT R. SMITH
Ohio State University Dartmouth College Indiana University Bloomington
BERND WITTENBRINK
University of Chicago LISA FELDMAN BARRETT TONY MANSTEAD DIEDERIK STAPEL
Boston College Cardiff University, Cardiff, Wales University of Groningen, Groningen,
VINCENT YZERBYT
Catholic University of Louvain, SUSAN T. FISKE THOMAS MUSSWEILER the Netherlands
Louvain-la-Neuve, Belgium Princeton University Universität Köln, Cologne, Germany FRITZ STRACK
CONSULTING EDITORS BARBARA L. FREDRICKSON JAMES M. OLSON Universität Würzburg, Würzburg,
University of Michigan University of Western Ontario, Germany
ICEK AJZEN
WENDI GARDNER London, Ontario, Canada
University of Massachusetts ABRAHAM TESSER
Northwestern University BERNADETTE M. PARK University of Georgia
MAHZARIN BANAJI
DANIEL GILBERT University of Colorado at Boulder
Harvard University YAACOV TROPE
Harvard University RICHARD E. PETTY New York University
MONICA BIERNAT Ohio State University
University of Kansas THOMAS GILOVICH THERESA K. VESCIO
Cornell University NEAL J. ROESE Pennsylvania State University
IRENE V. BLAIR University of Illinois at Urbana–
University of Colorado at Boulder ANTHONY G. GREENWALD
University of Washington Champaign WILLIAM VON HIPPEL
GALEN V. BODENHAUSEN University of New South Wales,
DAVID L. HAMILTON MYRON ROTHBART Sydney, Australia
Northwestern University University of Oregon
University of California, Santa
MARKUS BRAUER Barbara DUANE T. WEGENER
LAURIE RUDMAN Purdue University
LAPSCO, Université Blaise Pascal Rutgers, The State University
EDWARD R. HIRT
Clermont-Ferrand, France of New Jersey DANIEL M. WEGNER
Indiana University Bloomington
MARILYNN B. BREWER Harvard University
TIFFANY ITO MARK SCHALLER
Ohio State University University of Colorado at Boulder University of British Columbia, DIRK WENTURA
JOHN T. CACIOPPO Vancouver, British Columbia, Canada Saarland University, Saarbrücken,
YOSHIHISA KASHIMA
University of Chicago TONI SCHMADER Germany
University of Melbourne, Victoria,
OLIVIER CORNEILLE Australia University of Arizona DANIEL WIGBOLDUS
Catholic University of Louvain, KARLE CHRISTOPHE KLAUER NORBERT SCHWARZ Radboud University Nijmegen,
Louvain-la-Neuve, Belgium Albrecht-Ludwigs-Universität University of Michigan Nijmegen, the Netherlands
PATRICIA DEVINE Freiburg, Freiburg, Germany TIMOTHY D. WILSON
GÜN R. SEMIN
University of Wisconsin—Madison ARIE W. KRUGLANSKI Free University, Amsterdam, the University of Virginia
University of Maryland Netherlands PIOTR WINKIELMEN
AP DIJKSTERHUIS
University of Amsterdam, ALAN LAMBERT JEFFREY W. SHERMAN University of California, San Diego
Amsterdam, the Netherlands Washington University in St. Louis University of California, Davis
MARK P. ZANNA
DAVID DUNNING JENNIFER LERNER STEVEN J. SHERMAN University of Waterloo, Waterloo,
Cornell University Carnegie Mellon University Indiana University Bloomington Ontario, Canada
ASSISTANT TO THE EDITOR—LAURIE HAWKINS

INTERPERSONAL RELATIONS AND GROUP PROCESSES

JOHN F. DOVIDIO, Editor
University of Connecticut
ASSOCIATE EDITORS XIMENA ARRIAGA LORNE CAMPBELL KLAUS FIEDLER
DAPHNE BLUNT BUGENTAL Purdue University University of Western Ontario, University of Heidelberg,
University of California, Santa Barbara London, Ontario, Canada Heidelberg, Germany
WINTON W. T. AU
JACQUES-PHILIPPE LEYENS The Chinese University of Hong Kong, SERENA CHEN GARTH FLETCHER
Catholic University of Louvain, Shatin, Hong Kong University of California, Berkeley University of Canterbury,
Louvain-la-Neuve, Belgium Christchurch, New Zealand
MARK BALDWIN MARGARET CLARK
ANTONY MANSTEAD McGill University, Montreal, Quebec, Yale University SHELLY GABLE
Cardiff University, Cardiff, Canada University of California, Los Angeles
CARSTEN DE DREU
United Kingdom University of Amsterdam, LOWELL GAERTNER
KIM BARTHOLOMEW
Amsterdam, the Netherlands University of Tennessee, Knoxville
CYNTHIA L. PICKETT Simon Fraser University, Burnaby,
University of California, Davis British Columbia, Canada STÉPHANIE DEMOULIN SAMUEL L. GAERTNER
Catholic University of Louvain University of Delaware
JEFFRY A. SIMPSON C. DANIEL BATSON
University of Minnesota, University of Kansas Louvain-la-Neuve, Belgium, and ADAM GALINSKY
Twin Cities Campus Belgan National Fund for Scientific Northwestern University
B. ANNE BETTENCOURT Research, Brussels, Belgium
SCOTT TINDALE University of Missouri—Columbia PETER GLICK
Loyola University Chicago DAVID DESTENO Lawrence University
GERD BOHNER
Northeastern University STEPHANIE A. GOODWIN
JACQUIE D. VORAUER Universität Bielefeld, Bielefeld,
University of Manitoba, Winnipeg, Germany STEVE DRIGOTAS Purdue University
Manitoba, Canada Johns Hopkins University MARTIE G. HASSELTON
NIALL BOLGER
CONSULTING EDITORS New York University ELISSA S. EPEL University of California, Los Angeles
DOMINIC ABRAMS University of California, San S. ALEXANDER HASLAM
NYLA R. BRANSCOMBE Francisco
University of Kent at Canterbury, University of Exeter, Exeter,
University of Kansas
Canterbury, England VICTORIA ESSES United Kingdom
CHRIS AGNEW JONATHON D. BROWN University of Western Ontario, VERLIN HINSZ
Purdue University University of Washington London, Ontario, Canada North Dakota State University
ARTHUR ARON RUPERT BROWN BEVERLY FEHR GORDON HODSON
State University of New York at The University of Kent at Canterbury, University of Winnipeg, Winnipeg, Brock University, St. Catherine’s,
Stony Brook Canterbury, England Manitoba, Canada Ontario, Canada
(editors continue)
MICHAEL A. HOGG LAURA J. KRAY SABINE OTTEN CHRISTINE SMITH
University of Queensland, University of California, Berkeley University of Gröningen, Grand Valley State University
Brisbane, Australia JAMES R. LARSON JR. Gröningen, the Netherlands HEATHER J. SMITH
ANDREA B. HOLLINGSHEAD University of Illinois at Chicago CRAIG D. PARKS Sonoma State University
University of Southern California COLIN WAYNE LEACH Washington State University
RUSSELL SPEARS
JOHN G. HOLMES University of Sussex, Sussex, United LOUIS A. PENNER Cardiff University, Cardiff, Wales
University of Waterloo, Waterloo, Kingdom Wayne State University
CHARLES STANGOR
Ontario, Canada PAULA PIETROMONACO
JOHN LEVINE University of Maryland
RICK H. HOYLE University of Pittsburgh University of Massachusetts at
University of Kentucky GARY L. STASSER
Amherst
JOHN E. LYDON Miami University—Ohio
JOLANDA JETTEN McGill University, Montreal, Quebec, TOM POSTMES
University of Exeter, Exeter, University of Exeter, Exeter, United WALTER STEPHAN
Canada
United Kingdom Kingdom New Mexico State University
JON K. MANER
JAMES D. JOHNSON FELICIA PRATTO WILLIAM B. SWANN JR.
Florida State University
University of North Carolina at University of Connecticut University of Texas at Austin
Wilmington BRENDA MAJOR
HARRY T. REIS JANET SWIM
University of California, Santa
TATSUYA KAMEDA University of Rochester Pennsylvania State University
Barbara
Hokkaido University, Sapporo, Japan LEIGH L. THOMPSON
CRAIG MCGARTY W. STEVEN RHOLES
BENJAMIN R. KARNEY Texas A&M University Northwestern University
Australian National University,
RAND Corporation, Santa Monica, Canberra, Australia TOM TYLER
JENNIFER A. RICHESON
California New York University
WENDY BERRY MENDES Northwestern University
YOSHI KASHIMA Harvard University JEROEN VAES
MARK SCHALLER
University of Melbourne, Victoria, Australia University of Padova, Padova, Italy
RICHARD MORELAND University of British Columbia,
DEBORAH A. KASHY Vancouver, British Columbia, Canada KEES VAN DEN BOS
University of Pittsburgh
Michigan State University University of Utrecht, Utrecht,
BRIAN MULLEN DAVID A. SCHROEDER
KERRY KAWAKAMI University of Arkansas the Netherlands
University of Kent at Canterbury,
York University, Toronto, Ontario, Canada PAUL A. M. VAN LANGE
Canterbury, England CONSTANTINE SEDIKIDES
JANICE R. KELLY University of Southampton, Free University, Amsterdam,
AMÉLIE MUMMENDEY
Purdue University Southampton, England Amsterdam, the Netherlands
Friedrich-Schiller-Universität, Jena,
DACHER KELTNER Jena, Germany PHILLIP R. SHAVER LAURIE R. WEINGART
University of California, Berkeley University of California, Davis Carnegie Mellon University
MARK MURAVEN
DAVID A. KENNY University at Albany, State University J. NICOLE SHELTON GWEN M. WITTENBAUM
University of Connecticut of New York Princeton University Michigan State University
DOUGLAS T. KENRICK SANDRA L. MURRAY MARGARET SHIH WENDY L. WOOD
Arizona State University State University of New York at Buffalo University of Michigan Texas A&M University
NORBERT L. KERR LISA A. NEFF STACEY SINCLAIR MICHAEL ZÁRATE
Michigan State University University of Toledo University of Virginia University of Texas at El Paso
ASSISTANT TO THE EDITOR—CHRISTINE KELLY

PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES

CHARLES S. CARVER, Editor
University of Miami
ASSOCIATE EDITORS GEORGE A. BONANNO EDDIE HARMON-JONES DANIEL W. RUSSELL
TIM KASSER Teachers College, Texas A&M University Iowa State University
Knox College Columbia University TODD HEATHERTON OLIVER C. SCHULTHEISS
MARIO MIKULINCER AVSHALOM CASPI Dartmouth College University of Michigan
Bar-Ilan University, Ramat-Gan, Israel King’s College, London JUTTA HECKHAUSEN SUZANNE C. SEGERSTROM
EDWARD C. CHANG University of California, Irvine University of Kentucky
EVA M. POMERANTZ
University of Illinois at Urbana– University of Michigan KENNON M. SHELDON
STEVEN J. HEINE
Champaign SERENA CHEN University of British Columbia, University of Missouri—Columbia
University of California, Berkeley Vancouver, British Columbia, Canada C. R. SNYDER
RICHARD W. ROBINS
University of California, Davis A. TIMOTHY CHURCH RICHARD KOESTNER University of Kansas
Washington State University McGill University SANJAY SRIVASTAVA
GERARD SAUCIER University of Oregon
JAMES COAN Montreal, Quebec, Canada
University of Oregon
University of Wisconsin—Madison DAVID LUBINSKI TIMOTHY STRAUMAN
THOMAS A. WIDIGER Vanderbilt University Duke University
University of Kentucky M. LYNNE COOPER
University of Missouri—Columbia RICHARD E. LUCAS MICHAEL J. STRUBE
CONSULTING EDITORS Michigan State University Washington University
MICHAEL EID
STEPHAN A. AHADI JERRY SULS
University of Geneva, Geneva, ROBERT R. MCCRAE
American Institutes for Research, University of Iowa
Switzerland National Institute on Aging, Baltimore
Washington, DC
ANDREW J. ELLIOT WENDY BERRY MENDES WILLIAM B. SWANN JR.
JAMIE ARNDT University of Rochester Harvard University University of Texas at Austin
University of Missouri—Columbia HOWARD TENNEN
LISA FELDMAN BARRETT RODOLFO MENDOZA-DENTON
JENS B. ASENDORPF University of Connecticut Health
Boston College University of California, Berkeley
Humboldt-Universität Berlin Center
Berlin, Germany WILLIAM FLEESON DANIEL K. MROCZEK SUZANNE THOMPSON
Wake Forest University Fordham University Pomona College
MICHAEL C. ASHTON
R. CHRIS FRALEY STEPHEN A. PETRILL ROBERT J. VALLERAND
Brock University, St. Catherines,
University of Illinois at Chicago Pennsylvania State University Université du Québec à Montréal
Ontario, Canada
ANTONIO L. FREITAS RALPH L. PIEDMONT Montreal, Quebec, Canada
OZLEM AYDUK State University of New York at Loyola College in Maryland
University of California, Berkeley KATHLEEN D. VOHS
Stony Brook E. ASHBY PLANT University of Minnesota
ROY F. BAUMEISTER DAVID C. FUNDER Florida State University DAVID WATSON
Florida State University University of California, Riverside University of Iowa
BRENT ROBERTS
VERÓNICA BENET-MARTÍNEZ STEVEN W. GANGESTAD University of Illinois at BARBARA WOIKE
University of California, Riverside University of New Mexico Urbana–Champaign Columbia University
APRIL L. BLESKE-RECHEK CAROL L. GOHM MICHAEL D. ROBINSON REX A. WRIGHT
University of Wisconsin—Eau Claire University of Mississippi North Dakota State University University of Alabama at Birmingham
ASSISTANT TO THE EDITOR—JESSICA LILLESAND
 ATTITUDES AND SOCIAL COGNITION

Alone but Feeling No Pain: Effects of Social Exclusion on Physical Pain

Tolerance and Pain Threshold, Affective Forecasting,
and Interpersonal Empathy
C. Nathan DeWall and Roy F. Baumeister
Florida State University

Prior findings of emotional numbness (rather than distress) among socially excluded persons led the
authors to investigate whether exclusion causes a far-reaching insensitivity to both physical and
emotional pain. Experiments 1– 4 showed that receiving an ostensibly diagnostic forecast of a lonesome
future life reduced sensitivity to physical pain, as indicated by both (higher) thresholds and tolerance.
Exclusion also caused emotional insensitivity, as indicated by reductions in affective forecasting of joy
or woe over a future football outcome (Experiment 3), as well as lesser empathizing with another
person’s suffering from either romantic breakup (Experiment 4) or a broken leg (Experiment 5). The
insensitivities to pain and emotion were highly intercorrelated.

Keywords: social exclusion, rejection, emotion, affective forecasting, empathy

People depend heavily on others for much of their physical and
mental well-being. With no fangs, no claws, an extremely pro-
longed childhood phase of dependency and vulnerability, and other
physical weaknesses, human beings are not well suited to living in
isolation from others. Given the importance of acquiring and
maintaining membership in social groups, it is therefore hardly
surprising that people would react strongly to any threat of social
exclusion. Multiple laboratory studies of social exclusion have
found, however, that people respond to social exclusion in a
seemingly detached and emotionally indifferent manner. Socially
excluded people often report emotional states that do not differ
significantly from participants in acceptance or control conditions
(Baumeister, Twenge, & Nuss, 2002; Gardner, Pickett, & Brewer,
2000; Twenge, Baumeister, Tice, & Stucke, 2001; Twenge &
Campbell, 2003; Twenge, Catanese, & Baumeister, 2002; Zadro,
Williams, & Richardson, 2004).
C. Nathan DeWall and Roy F. Baumeister, Department of Psychology,
Florida State University.
We gratefully acknowledge support by National Institute of Mental
Health Grant MH-65559.
The research reported in this article was part of C. Nathan DeWall’s
master’s thesis at Florida State University under the direction of Roy F.
Baumeister. We graciously thank committee members Thomas Joiner, Jon
K. Maner, and Dianne M. Tice for their helpful comments and suggestions.
We also thank Carey Morewedge for helpful comments on an earlier
version of this article.
Correspondence concerning this article should be addressed to C.
Nathan DeWall, Department of Psychology, Florida State University,
Tallahassee, FL 32306-1270. E-mail: dewall@darwin.psy.fsu.edu
Journal of Personality and Social Psychology, 2006, Vol. 91, No. 1, 1–15
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00
Social exclusion poses a serious threat to a person’s physical
and psychological well-being, as indicated by higher rates of
physical and mental illnesses among people who live alone as
compared with people who have strong social networks (Argyle,
1987; Cacioppo, Hawkley, & Berntson, 2003; Kiecolt-Glaser et
al., 1984; Lynch, 1979; Myers, 1992). These findings confirm that
social exclusion has potentially drastic and negative effects on
physical and psychological health, but they offer no explanation
for the finding that people respond to social exclusion with emo-
tional numbness (e.g., Twenge et al., 2001). Indeed, they under-
score the paradox in the fact that people would respond to a
potentially momentous event, such as social exclusion, with emo-
tional numbness and detachment.
The first purpose of the current investigation was to resolve the
paradox as to why rejected people report emotional numbness, by
identifying a possible mechanism by which people respond to
social exclusion. We propose that certain interpersonal events,
such as social rejection, activate the body’s pain response system
and potentially alter how it registers physical and emotional pain
(Eisenberger, Lieberman, & Williams, 2003). With regard to phys-
ical pain, social exclusion may disrupt the ability to respond to
physical pain in the same manner as people who have not expe-
rienced social exclusion. This would lead to increases in both pain
threshold (i.e., sensitivity to pain) and pain tolerance (i.e., with-
standing greater pain). Hence, the first goal of this research (tested
in Experiments 1– 4) was to demonstrate that socially excluded
people become relatively numb to physical pain.
The second goal of this work was to extend the idea of physical
numbness to emotional functioning. As suggested by MacDonald
and Leary (2005), if the body uses the same system to respond to
physical injury and interpersonal injury, then physical pain and
DOI: 10.1037/0022-3514.91.1.1

1
2 DEWALL AND BAUMEISTER
interpersonal emotions may be linked—and just as the body goes
numb to pain, it may also become less sensitive to emotion. If the
emotional system ceases to function normally in the aftermath of
social exclusion, people might show abnormalities not only in their
emotional reactions to current events but also in their forecasts of
emotional responses to future events. They might also lose their
empathy for the physical suffering of another person. Last, and
ironically, rejected people might even lose their empathy for the
suffering of someone else who has also suffered rejection. Exper-
iments 3–5 tested these predictions.
Social Exclusion and Emotional Responding: The Paradox
of Detachment and Numbness
A desire to form and maintain social bonds with others is a
fundamental and pervasive desire among humans (Baumeister &
Leary, 1995). Insofar as belongingness is a basic need rather than
a want, people experience considerable difficulties when their need
to belong is unsatisfied or frustrated. A lack of stable relationships
also has detrimental effects on one’s health (Cacioppo, Hawkley,
& Berntson, 2003; Hawkley, Burleson, Berntson, & Cacioppo,
2003; Lynch, 1979; see Uchino, Cacioppo, & Kiecolt-Glaser,
1996, for a review).
One might assume and hope that excluded people would show
adaptive responses of changing themselves so as to become more
socially attractive or to take better care of themselves as individ-
uals. The opposite is often found, however. Rejected people ex-
hibit decreased intellectual functioning (Baumeister et al., 2002),
become more aggressive toward others (Buckley, Winkel, &
Leary, 2004; Twenge et al., 2001; Twenge & Campbell, 2003), are
less willing to self-regulate (Baumeister, DeWall, Ciarocco, &
Twenge, 2005), and engage in various self-defeating behaviors,
such as risk taking and procrastination (Twenge et al., 2002). None
of these responses seem likely to foster interpersonal acceptance or
personal health and well-being.
As already noted, socially excluded participants frequently re-
port emotional states that do not differ significantly from socially
accepted and control participants (Baumeister et al., 2002; Gardner
et al., 2000; Twenge et al., 2002; Zadro et al., 2004). Even more
surprisingly, when differences in emotion are found, these differ-
ences do not mediate the behavioral consequences of social exclu-
sion. Even researchers who find that social exclusion increases
emotional distress have not found that the distress mediates the
behavioral consequences of social exclusion (Buckley et al., 2004;
Williams, Cheung, & Choi, 2000). Given the fundamental nature
of the need to belong, one might reasonably expect that real,
potential, or imagined social exclusion would result in severe
emotional reactions. It is therefore unclear why emotion plays such
a minor role in explaining behavioral responses to social exclusion.
One reason socially excluded people report feelings of emo-
tional numbness may be that exclusion leads to a defensive state of
cognitive deconstruction. The deconstructed state has been in-
voked to characterize presuicidal mental states (Baumeister, 1990),
and it is marked by emotional numbness, an altered perception of
time, thoughts of meaninglessness, lethargy, and avoidance of
self-focused attention. Twenge, Catanese, and Baumeister (2003)
showed that socially excluded people exhibited each of these
symptoms of cognitive deconstruction. Social exclusion leads to
behaviors that could preclude the possibility of gaining future
acceptance (e.g., aggression, self-defeating behavior), but the de-
constructed state may offer rejected people a temporary reprieve
from feeling the intense pain or distress that can accompany threats
to belongingness. Although escaping from aversive emotional
states and high self-awareness may be beneficial in the short-term,
such behaviors are characteristic of severe psychopathology in
clinical populations (Bancroft, Skirmshire, & Simkins, 1976; Haw-
ton, Cole, O’Grady, & Osborn, 1982) and of a variety of self-
destructive behaviors in nonclinical samples (see Baumeister &
Scher, 1988, for a review). The experiments reported in this article
sought to link emotional unresponsiveness with insensitivity to
physical pain, both of which could be natural defenses against
one’s own suffering and could contribute to the deconstructed
state.
If people have normal or natural defenses that help them escape
or minimize emotional distress, then social exclusion may tempo-
rarily impair the ability to experience emotions in a normal fash-
ion. The theory proposed in the current article is that the emotion
system temporarily ceases to function normally in response to
social exclusion. The resulting emotional numbness could be
linked to an insensitivity to physical pain, as the next section will
explain.
Social and Physical Pain
There is reason to believe that the pain of social exclusion shares
many of the same neural and psychological mechanisms as expe-
riences of physical pain. At a purely linguistic level, people fre-
quently use words connoting physical pain when describing emo-
tional responses to distressing events. For example, people may
report feeling “hurt” or “crushed” following the dissolution of a
meaningful relationship (Leary & Springer, 2000). MacDonald
and Leary (2005) have suggested that the similarity between
descriptions of social and physical pain extends beyond mere
metaphor. Specifically, they proposed that social pain and physical
pain operate via shared physiological mechanisms, including the
anterior cingulate cortex, periaqueductal gray (PAG) brain struc-
tures, and the opioid and oxytocin neuroendocrine systems (see
also Rossi, Pasternak, & Bodnar, 1993).
Decades ago, Panksepp and colleagues proposed a link between
social and physical pain (Herman & Panksepp, 1978; Panksepp,
Herman, Conner, Bishop, & Scott, 1978; Panksepp, Vilberg, Bean,
Coy, & Kastin, 1978). They suggested that as evolution prepared
animals for increasing social interaction, instead of creating en-
tirely new systems to react to social events, such as being rejected
or excluded, it piggybacked these responses onto the existing
systems that were hard-wired for responding to physical pain.
Social events might therefore activate the body’s pain response
system and potentially alter how it would register physical pain.
Recent neuroscience and functional MRI research has shown
that the anterior cingulate cortex (ACC) functions as a neural
alarm system to warn people that factors in their environment
threaten their goals (Bush, Luu, & Posner, 2000; Eisenberger &
Lieberman, 2004; Eisenberger et al., 2003; Kimbrell et al., 1999;
E. E. Nelson & Panksepp, 1998). Eisenberger et al. (2003) showed
that the ACC plays a prominent role in the detection of threats to
belongingness. In their functional MRI study, ostracized people
showed activation of the dorsal ACC. Ostracized people also
showed activation of the right ventral prefrontal cortex, which has
 SOCIAL EXCLUSION AND PAIN
been previously linked to the regulation of physical pain distress
and negative affect (Hariri, Bookheimer, & Mazziotta, 2000;
Petrovic, Kalso, Petersson, & Ingvar, 2002). Thus, preliminary
evidence suggests that Panksepp and colleagues (Herman & Pank-
sepp, 1978; Panksepp, Herman, et al., 1978; Panksepp, Vilberg, et
al., 1978) were correct in proposing that certain physiological
processes responsible for detection and regulation of physical pain
were co-opted to sense and respond to emotionally painful events,
such as being rejected or excluded.
Additional research has shown that the PAG brain structures,
which receive input from the body’s injury detection system (no-
ciceptive system) and ACC, are involved in the detection of
physical pain and are implicated in animal bonding behavior
(Craig & Dostrovsky, 1999). Panksepp (1998) demonstrated that
activation of the PAG elicited separation distress cries from rats,
and further evidence showed that lesions to the PAG led to reduced
separation distress cries (Wiedenmayer, Goodwin, & Barr, 2000).
Administration of the neuropeptide oxytocin and opioids such as
morphine have also been shown to reduce separation distress cries
in rats (Carden, Hernandez, & Hofer, 1996; Carden & Hofer, 1990;
Insel & Winslow, 1991). These findings indicate that some phys-
iological systems respond to both physical pain and socially dis-
tressing events.
Most relevant to the current investigation, however, is evidence
that separation from caregivers and isolation from conspecifics
results in decreased sensitivity to physical pain, also known as
analgesia. Several studies have shown that short-term isolation
produces reduced pain sensitivity in rat pups (Kehoe & Blass,
1986a, 1986b; Naranjo & Fuentes, 1985; Spear, Enters, Aswad, &
Louzan, 1985), mice (Konecka & Sroczynska, 1990), cows
(Rushen, Boissy, Terlouw, & de Passillé, 1999), and chicks (Sufka
& Hughes, 1990; Sufka & Weed, 1994; see MacDonald & Leary,
2005, for a review). Thus, threats to belongingness appear to
activate neural mechanisms associated with physical pain and the
regulation of pain in some nonhuman species (E. E. Nelson &
Panksepp, 1998; Panksepp, 1998).
Previous research has shown that social and physical pain share
common physiological mechanisms in some animals, but research
has not provided much evidence about whether a similar link exists
in humans. MacDonald, Kingsbury, and Shaw (2005) showed that
people high in rejection sensitivity responded to ostracism with
decreased sensitivity to physical pain (pain threshold) on the
cold-pressor task, but people low in rejection sensitivity did not
show the same pattern of responding. If threats to belongingness
activate basic neural mechanisms designed to regulate pain (Eisen-
berger et al., 2003), one would expect all ostracized participants to
exhibit decreased sensitivity to physical pain, regardless of their
propensity to anxiously expect and overreact to a threat to their
sense of belonging (Downey & Feldman, 1996). Those prelimi-
nary findings were also mute about another important issue,
namely, what relationship increased pain threshold and tolerance
may have to emotional responding, to which we turn in the next
section.
Effects of Increased Pain Threshold and Pain Tolerance
for Emotional Responding
If people respond to social exclusion with emotional numb-
ness, this temporary shutdown of the emotion system might be
3
beneficial in terms of immediately reducing a person’s suffer-
ing, in the same way that physical injuries often create an
analgesia that saves him or her from feeling acute, ongoing pain
for as long as the injury lasts. If the neural systems responsible
for detecting and regulating physical and emotional pain have a
common physiological basis, then social exclusion should also
influence how people respond to physically painful stimuli.
Thus, excluded people should become numb to physical pain,
and this physical numbness should be related to emotional
insensitivity (see Figure 1).
What would provide the best measures of emotional insensitiv-
ity? One method would be direct self-reports of emotion, which
(sure enough) tend to indicate no emotional reaction (e.g.,
Baumeister et al., 2005). Self-reported emotion may not provide
the best measure, however, because of social desirability biases
and other factors that could influence responding. For example,
excluded people might feel emotionally upset but refuse to admit
it. To avoid these methodological pitfalls of direct self-reports of
emotion, we measured two emotional phenomena in which people
seemingly rely on their current state and emotional simulation
(imagining how oneself would feel) to make judgments about
events distant from the present. The first of these phenomena is
affective forecasting; the other is empathy. If rejected people are
replete with emotion but reluctant to admit their distress out of
self-presentational concerns, they should be quite willing to predict
strong future emotional reactions and to empathize with others’
suffering. But if their emotional systems have really ceased to
function normally, excluded people should find it difficult to
imagine having strong emotions in response to hypothetical future
events, and they likewise might not empathize with another per-
son’s distress.
Affective forecasting refers to the ways in which people
predict their future feelings (see Wilson & Gilbert, 2003, for a
review). Research has shown consistently that people overesti-
mate the duration and intensity of their emotional responses to
future events. People overestimate how happy they will be after
winning a date on a simulated dating game (Wilson, Wheatley,
Kurtz, Dunn, & Gilbert, 2004), predict greater distress follow-
ing a romantic breakup or denial for academic tenure than
actually occurs (Gilbert, Pinel, Wilson, Blumberg, & Wheatley,
1998), and overrate what their quality of life will be if they
receive an important medical procedure (Jepson, Loewenstein,
& Ubel, 2001). Other research has shown that a person’s current
emotional state can bias his or her predictions of future emo-
tional responses and other decision-making processes (Gilbert,
Gill, & Wilson, 2002; Loewenstein, O’Donoghue, & Rabin, in
press; L. D. Nelson & Morrison, 2005; Van Boven & Loewen-
stein, 2003).
Emotional
Insensitivity
Social Exclusion
Physical
Insensitivity
Figure 1. Proposed model illustrating linked consequences of physical
and emotional insensitivity following social exclusion.
4 DEWALL AND BAUMEISTER
If social exclusion hampers the capacity for people to respond to
their own emotions, rejected people might show a different pattern
of affective forecasting than other participants. Specifically, re-
jected participants might report less happiness when forecasting
their emotional reaction to a positive event and less unhappiness
when forecasting their emotional reaction to a negative event
compared with participants who did not experience rejection.
Another process that may illustrate the intimate link between
social exclusion and emotional responding is the ability of rejected
people to show empathy toward people in various forms of phys-
ical and psychological distress. Previous research has shown that
rejected people tend to behave less prosocially than nonrejected
people (Twenge, Baumeister, DeWall, Ciarocco, & Bartels, 2006),
and work by Batson and colleagues has demonstrated that empathy
plays a prominent role in shaping prosocial behavior (Batson,
Klein, Highberger, & Shaw, 1995). It therefore seems plausible
that rejected participants might feel reduced empathy toward an-
other person’s suffering.
Present Research
The present experiments tested the theory that social exclusion
activates the body’s pain system and sets of responses that may
reduce sensitivity to both physical and emotional pain. Our pre-
diction had two parts. First, participants who experienced social
exclusion would show increased pain threshold and pain tolerance
compared with socially accepted and control participants. Second,
this insensitivity to physical pain should be related to emotional
insensitivity. Participants who experience social exclusion should
find it difficult to predict strong emotional responses to hypothet-
ical future events, and they should exhibit decreased empathy with
others’ suffering, compared with participants who do not experi-
ence rejection.
Thus, the current investigation sought to resolve the seeming
paradox as to why people respond to such a threatening experience
as rejection with emotional numbness. The first goal was to test the
hypothesis that people become numb to physical pain in the
aftermath of social exclusion. The second goal of this work was to
demonstrate that physical numbness has implications for emo-
tional responding, such as affective forecasting and interpersonal
empathy.
Experiments 1 and 2: Not Feeling My Own Pain
Experiments 1 and 2 tested the hypothesis that social exclusion
reduces sensitivity to physical pain. Social exclusion was manip-
ulated by having participants complete a personality test and
giving some participants bogus feedback stating that their person-
ality profile enabled the researchers to predict that they would
most likely end up alone in life. This future alone condition has
been used in past research to create a sense of social exclusion and
impending isolation, because people take it to mean that something
about their personalities will cause others to reject them (Twenge
et al., 2001). There were three control groups. One of these groups
received personality feedback stating that they had a personality
type that would lead to a future filled with several meaningful and
lasting relationships (future belonging). The second control group
was given no feedback regarding their personality or the implica-
tions their personality may have for their future belongingness
status (no feedback). The third control group (misfortune control),
added in Experiment 2, told people that their futures would be
marred by accidents. The purpose of this third control group was
to rule out the possibility that the effects of social exclusion were
simply due to receiving an unpleasant forecast about the future.
Hence, the misfortune control feedback predicted bad outcomes
that did not include social exclusion.
We predicted that participants in the future alone condition
would have higher pain thresholds and higher pain tolerance than
participants in the three control conditions. The inclusion of both
pain threshold and pain tolerance measures provided two separate
chances to confirm or disconfirm our hypothesis that social exclu-
sion leads to numbness to physical pain. We had no a priori reason
to think that socially excluded participants would show increases
in pain threshold but not in pain tolerance (or vice versa). If
socially excluded participants showed increases in both pain
threshold and pain tolerance, however, that would provide con-
verging support for our hypothesis.
Method
Participants. Thirty-three right-handed undergraduates (24 women)
participated in Experiment 1 for partial course credit. Experiment 2 had
30 participants (24 women). Of the participants in Experiment 1, 67%
were White and 33% were a racial minority. Of the participants in
Experiment 2, 83% were White and 17% were a racial minority. In
these and subsequent studies, we restricted participation to people who
were nonsmokers and right-handed because smoking has been shown to
reduce pain sensitivity (Kanarek & Carrington, 2004; Pomerleau, Turk,
& Fertig, 1984) and because left limbs show greater pain sensitivity
than right limbs, regardless of hand of preference (Murray & Hagan,
1973). Participants were also required to not have ingested any sugared
foods, alcoholic beverages, or pain medicine (including aspirin) for at
least 8 hr prior to participation in the experiments (Kanarek & Car-
rington, 2004; Mercer & Holder, 1997).
Materials and procedure. Participants arrived individually for a study
ostensibly concerned with personality and physical sensitivity. After giving
informed consent, participants in Experiment 1 but not 2 completed mea-
sures of rejection sensitivity (Downey & Feldman, 1996) and global
self-esteem (Rosenberg, 1965).
Baseline measurements of pain threshold and pain tolerance (two trials
each) were then taken by using a pressure algometer (Type II; Somedic,
Solletuna, Sweden). For pain threshold, participants were instructed to say
“now” when they first felt pain due to the pressure increase. For pain
tolerance, participants were instructed to say “stop” when the pain became
too uncomfortable to continue. At this point, the experimenter immediately
retracted the algometer. The digital display showed the value of pressure
applied at the moment the algometer was retracted. The algometer was
applied perpendicularly to the skin and lowered at a rate of approximately
5 kPA per second until pain threshold or tolerance was reached, as
indicated by participants’ verbal report. All pain measurements were taken
at the first dorsal interosseous muscle (i.e., behind the first knuckle of the
index finger) of the participant’s dominant hand. The order of the pain
tolerance and threshold measurements was counterbalanced across partic-
ipants. To prevent habituation, there was a 1.5-min interval between all
pain threshold and tolerance measurements (Orbach, Mikulincer, King,
Cohen, & Stein, 1997).
Participants then completed a brief demographic questionnaire and the
Eysenck Personality Questionnaire (Eysenck & Eysenck, 1975). To bolster
the credibility of the cover story, we gave participants accurate feedback
 SOCIAL EXCLUSION AND PAIN
regarding their extroversion score.1 Participants then received bogus feed-
back about the implications their extroversion score would have for their
future belongingness. Following a procedure developed by Twenge et al.
(2001), we randomly assigned participants in Experiment 1 to one of three
social feedback conditions: future belongingness, future alone, and no-
feedback control. For Experiment 2, a misfortune control group replaced
the no-feedback control group.
In the future belongingness condition, participants were given their
correct score on extroversion and then told, as additional feedback, that
they would have stable, rewarding relationships throughout life. In the
(crucial) future alone condition, participants were given their correct ex-
troversion score but then told that they would likely end up alone in life.
They were told they might have friends now, but these would drift away
after college, and once they were past the age at which people are
constantly making new friends, they would spend more and more time
alone. The experimenter said they might marry once or several times, but
these would be short-lived.
The no-feedback control condition gave neither an extroversion score
nor a forecast about the future. The misfortune control condition included
the extroversion score and a forecast that the participant would become
increasingly accident prone in future years, with many injuries and hospital
stays.
Participants then completed the Brief Mood Introspection Scale (BMIS;
Mayer & Gaschke, 1988). After completing the BMIS, measurements of
pain threshold and tolerance (five trials each) were taken. When partici-
pants had completed the pain threshold and tolerance measurements, they
were carefully and thoroughly debriefed. Care was taken to ensure that
participants recognized that the feedback they had received was based on
random assignment and had nothing to do with them, their personality, or
any form they had completed during the experimental session. The exper-
imenter also apologized for being misleading regarding the personality
feedback. When the experimenter was certain that participants understood
the true purpose of the experiment and the bogus nature of the personality
feedback, participants were given partial course credit, were thanked for
their time, and were dismissed.
Results
We computed both the omnibus F statistic and performed a
priori contrasts comparing socially excluded participants (i.e., fu-
ture alone) with nonsocially excluded participants (i.e., future
belongingness, no-feedback control, and misfortune control). Pain
threshold and tolerance scores were compared by using baseline
pain threshold and pain tolerance scores as a covariate. There were
no main effects or interactions on pain threshold and tolerance (or
any other dependent measure) involving participant gender or
ethnicity in any other of the studies reported in this article. Hence,
these variables will not be discussed further.
Pain threshold. As predicted, socially excluded participants
showed significantly higher pain thresholds than participants who
were led to anticipate a future filled with social acceptance or
future misfortune or those who received no personality feedback.
Results from an analysis of covariance (ANCOVA) on pain thresh-
old scores in Experiment 1 (using baseline score as covariate)
indicated that there was significant variation between the three
experimental groups, F(2, 30) ⫽ 17.85, p ⬍ .001. A 2⫺1⫺1 a
priori contrast confirmed that future alone participants had higher
pain thresholds compared with future belongingness and no-
feedback control participants, F(1, 30) ⫽ 33.63, p ⬍ .001. For
Experiment 2, the ANCOVA yielded the following: F(2, 27) ⫽
10.84, p ⬍ .001; and the 2–1–1 contrast showed that future alone
participants had substantially higher pain threshold scores com-
5
pared with future belonging and misfortune control participants,
F(1, 27) ⫽ 21.35, p ⬍ .001.
A one-cell t test confirmed that the difference between baseline
and post-feedback pain threshold measurement among future alone
participants was greater than zero in both Experiment 1, t(10) ⫽
⫺4.83, p ⫽ .001, and Experiment 2, t(9) ⫽ ⫺3.14, p ⫽ .01. In
Experiment 1, future belonging participants showed a significant
decrease in pain threshold from baseline, t(10) ⫽ 4.16, p ⫽ .002,
though this effect did not replicate in Experiment 2 (t ⬍ 1, ns).
There was no change from baseline in the no-feedback control
condition in Experiments 1 and 2 (both ts ⬍1, ns). These findings
suggest that the increase in pain sensitivity was not merely relative
to participants in the other two conditions but was an absolute
change from baseline measurements. The means and standard
deviations are presented in Table 1.
Pain tolerance. In both studies, socially excluded participants
showed significantly higher levels of pain tolerance than all other
participants. Results from an ANCOVA on the pain tolerance
scores indicated that there was significant variation between the
three experimental groups in Experiment 1, F(2, 30) ⫽ 20.36, p ⬍
.001, and Experiment 2, F(2, 27) ⫽ 9.66, p ⫽ .001. A 2⫺1⫺1 a
priori contrast confirmed that future alone participants had higher
overall pain tolerance than both other conditions in Experiment 1,
F(1, 30) ⫽ 40.60, p ⬍ .001, and Experiment 2, F(1, 26) ⫽ 19.21,
p ⬍ .001.
A one-cell t test confirmed that the difference between baseline
and post-feedback pain tolerance measurement among future alone
participants was greater than zero in Experiment 1, t(10) ⫽ ⫺4.94,
p ⫽ .001, and again in Experiment 2, t(10) ⫽ ⫺2.91, p ⬍ .02,
indicating that their tolerance was significantly higher after receiv-
ing the feedback than it had been beforehand. There was no change
from baseline in the other conditions (both ts ⬍1, ns).
Mood and emotion, rejection sensitivity, and self-esteem. To
investigate the possible role of mood and emotion in shaping the
observed effects, we tested for fluctuations in mood valence and
arousal as a result of personality feedback. For each experiment,
two one-way ANOVAs were carried out by using the Valence and
Arousal subscales of the emotion measure (BMIS) as dependent
measures. In Experiment 1, there was no significant variation
between the three social feedback groups in terms of their reported
mood valence, F(2, 30) ⫽ 2.21, p ⫽ .13. Pairwise comparisons
revealed that future alone participants (M ⫽ 18.18, SD ⫽ 8.15)
reported levels of emotional valence that did not differ signifi-
cantly from future belonging participants (M ⫽ 20.73, SD ⫽ 8.46;
F ⬍ 1, ns). No-feedback control participants (M ⫽ 11.73, SD ⫽
13.54) reported more negative moods than future belonging par-
ticipants (M ⫽ 20.73, SD ⫽ 8.46), F(1, 30) ⫽ 4.16, p ⫽ .05.
No-feedback control participants (M ⫽ 11.73, SD ⫽ 13.54) also
reported a somewhat less positive mood than future alone partic-
ipants (M ⫽ 18.18, SD ⫽ 8.15), although this difference did not
reach significance, F(1, 30) ⫽ 2.14, p ⫽ .15. There was also no
significant variation between the three social feedback groups in
terms of their reported mood arousal (F ⬍ 1, ns). Regression
analyses revealed that neither self-esteem nor social sensitivity
significantly moderated the effects of condition on pain threshold
1
It should be noted that individual differences on the Introversion–
Extroversion Scale did not influence the results in any way.
6 DEWALL AND BAUMEISTER

Table 1
Pain Threshold and Pain Tolerance, Experiments 1– 4

Future Future Misfortune Other

alone belonging control control

Experiment and measure M SD M SD M SD M SD

1
Pain threshold 31.73 12.46 13.27 12.68 19.01 12.17
Pain tolerance 199.99 50.23 129.40 49.65 133.20 49.88
2
Pain threshold 45.94 32.05 10.55 31.46 14.41 31.89
Pain tolerance 388.22 113.44 276.32 107.68 271.94 114.82
3
Pain threshold 74.57 61.92 19.17 61.81 12.52 61.87
Pain tolerance 392.53 134.59 197.66 136.34 216.58 133.93
4
Pain threshold 62.51 22.11 14.35 21.05 10.24 22.22
Pain tolerance 496.84 120.97 236.05 120.81 244.47 109.89

Note. In all experiments, baseline pain threshold and pain tolerance scores were entered as covariates.
Covariate values for each experiment were as follows: Experiment 1, Pain threshold ⫽ 20.35; Experiment 1, Pain
tolerance ⫽ 163.83; Experiment 2, Pain threshold ⫽ 16.18; Experiment 2, Pain tolerance ⫽ 312.03; Experiment
3, Pain threshold ⫽ 23.02; Experiment 3, Pain tolerance ⫽ 269.02; Experiment 4, Pain threshold ⫽ 25.03;
Experiment 4, Pain tolerance ⫽ 314.37.

or tolerance ( ps ⬎ .20). In Experiment 2, there was no significant
variation between the three groups in either mood valence or mood
arousal (both Fs ⬍ 1, ns). The means and standard deviations are
presented in Table 2. Thus, the observed increases in pain toler-
ance and threshold among socially excluded participants were not
due to differences in mood valence and arousal or individual
differences in self-esteem and rejection sensitivity.
Discussion
Social exclusion produced increases in both pain threshold
and pain tolerance in both studies, consistent with the hypoth-
esis that people become less sensitive to physical pain as a
result of having their need to belong thwarted. Participants who
anticipated a lonely future showed greater tolerance and less
sensitivity to physical pain than participants who experienced
social acceptance, received no personality feedback, or received
feedback forecasting future physical misfortunes. They also
showed significantly less sensitivity to physical pain than they
themselves had shown on the baseline measures. The estimated
effect sizes for the increases in pain threshold and pain toler-
ance among socially excluded participants were quite large
(Cohen, 1977), exceeding a full standard deviation in each case.

Table 2
Self-Reports of Emotion by Experimental Condition, Experiments 1–5

Future Future Misfortune Other

alone belonging control control

Experiment and measure M SD M SD M SD M SD

1
Mood Valence 18.18 8.15 20.73 8.46 11.73 13.54
Mood Arousal 25.64 5.63 26.91 7.56 28.91 7.42
2
Mood Valence 17.80 13.40 22.90 12.97 15.30 13.37
Mood Arousal 26.60 5.50 24.90 7.48 27.10 6.15
3
Mood Valence 10.40 12.13 18.70 9.78 15.20 12.93
Mood Arousal 28.40 7.14 26.40 6.29 26.10 6.76
4
Mood Valence 19.50 10.71 16.90 11.23 11.45 10.51
Mood Arousal 27.40 5.87 22.90 120.81 24.64 4.41
5
Negative Affect 12.55 3.52 12.50 3.51 13.91 4.39
Positive Affect 22.98 5.74 24.76 5.23 23.32 4.64

Note. Values represent mean scores on either the BMIS Mood Valence or Mood Arousal subscales (Mayer &
Gaschke, 1988) or the PANAS Positive Affect and Negative Affect subscales (Watson et al., 1988). BMIS ⫽
Brief Mood Introspection Scale; PANAS ⫽ Positive and Negative Affect Schedule.
 SOCIAL EXCLUSION AND PAIN
The results of Experiment 1 and 2 were not due to any differ-
ences in reported mood valence or mood arousal. Future alone
participants did not report moods that differed significantly from
those reported by future belonging or no-feedback control partic-
ipants. Neither rejection sensitivity nor trait self-esteem played a
moderating role in predicting the relationship between social ex-
clusion and physical pain threshold and pain tolerance. Thus, a
future diagnostic forecast of social exclusion led to substantial
increases in pain threshold and pain tolerance compared with
socially accepted and control participants.
Experiment 3: Not Feeling Future Pain
Experiment 3 tested whether social exclusion would have im-
plications for affective forecasting. Affective forecasting involves
simulating emotional reactions to possible future events. Exclusion
causes the emotion system to cease functioning properly, indeed
muting and numbing emotional reactions. It is therefore possible
that hypothetical emotional reactions to the future (i.e., imagining
how one would feel) will also be numbed. Accordingly, rejected
participants (who reported emotional states that did not differ from
participants in the other conditions in Experiments 1 and 2) should
make affective forecasts of relative emotional numbness compared
with accepted and control participants. That is, rejected partici-
pants should report less happiness when forecasting their emo-
tional reaction to a positive event and less sadness and distress
when forecasting their emotional reaction to a negative event.
Socially accepted and control participants, in contrast, should
exhibit the previously documented impact bias, in which people
overestimate their happiness and sadness to possible future events
(e.g., Wilson et al., 2004). The current study also tested whether
increases in pain threshold and pain tolerance were related to the
predicted unemotional affective forecasts among socially excluded
participants.
Method
Participants. Thirty right-handed undergraduates (19 women) partici-
pated for partial course credit. Of the participants, 77% were White and
23% were a racial minority. The average age was 18.6 years.
Materials and procedure. Participants arrived at the laboratory indi-
vidually for an experiment ostensibly concerned with the relationship
between personality, verbal and nonverbal behaviors, and physical sensi-
tivity. After giving informed consent, participants completed baseline
measurements of pain threshold and pain tolerance (two trials each).
Participants then completed a personality test and were exposed to the
same social exclusion manipulation used in Experiment 1. By random
assignment, participants were told that they had a personality type in which
they would likely end up alone (future alone) or that their personality type
was diagnostic of a future filled with many rewarding and lasting relation-
ships (future belonging). Another group was not given any feedback on
their personality test (no-feedback control).
After receiving their personality feedback, participants completed the
BMIS mood and affective forecasting measures. The latter were modeled
after previous research by Wilson and colleagues (Wilson, Wheatley,
Meyers, Gilbert, & Axsom, 2000). Participants were reminded that the
Florida State football team was scheduled to compete against its closest
rival in approximately 2 months and were asked to predict what their level
of overall happiness would be directly after a victory and after a loss.
Participants also reported to what degree they considered themselves a
Florida State University football fan and to what extent they cared whether
7
Florida State won this particular game. These two measures were highly
correlated (r ⫽ .50, p ⫽ .005) and were combined to create a Florida State
football “fan” index.
Last, participants completed five additional measurements of pain
threshold and tolerance. Participants were then debriefed, thanked for their
time, given partial course credit, and dismissed.
Results
Pain threshold and pain tolerance: Replication. As in Exper-
iments 1 and 2, a future diagnostic forecast of social exclusion
produced significant increases in pain threshold compared with
participants who anticipated a future filled with meaningful rela-
tionships or who received no feedback on their personality test. An
ANCOVA on the pain threshold scores, using baseline pain thresh-
old scores as a covariate, revealed significant variation between the
three experimental groups, F(2, 27) ⫽ 9.07, p ⫽ .001. A 2⫺1⫺1
a priori contrast also showed that future alone participants had
significantly higher pain threshold scores than future belonging
and no-feedback control participants, F(2, 27) ⫽ 17.96, p ⬍ .001.
A one-cell t test confirmed that the difference between baseline
and post-feedback pain threshold measurement among future alone
participants was greater than zero, t(9) ⫽ ⫺2.50, p ⫽ .03, whereas
there was no change from baseline in the other conditions (both ts
⬍1, ns). Socially excluded participants also showed increased pain
tolerance compared with participants who experienced social ac-
ceptance and participants who received no feedback on a person-
ality test. An ANCOVA on the pain tolerance scores revealed
significant variation between the three experimental groups, F(2,
27) ⫽ 18.91, p ⬍ .001. A 2⫺1⫺1 a priori contrast confirmed that
future alone participants had significantly higher pain tolerance
scores than both future belonging and no-feedback control partic-
ipants, F(2, 27) ⫽ 37.69, p ⬍ .001. A one-cell t test confirmed that
the difference between baseline and post-feedback pain threshold
measurement among future alone participants was greater than
zero, t(9) ⫽ ⫺2.93, p ⫽ .02. There was no change from baseline
in the other conditions (both ts ⬍ 1, ns). Thus, future alone
participants showed substantial and absolute increases in both pain
threshold and pain tolerance. The means and standard deviations
are presented in Table 1.
Affective forecasting. Replicating previous affective forecast-
ing research, the results indicated that future belonging and no-
feedback control participants predicted strong emotional responses
to a Florida State victory and defeat. Most relevant to the current
investigation, however, was the finding that participants in the
future alone condition predicted relatively neutral emotional reac-
tions to both football outcomes. A mixed-model analysis of vari-
ance (ANOVA) was conducted by using the personality feedback
condition as a between-subjects factor and responses to the ques-
tions “Please predict what your overall happiness would be right
after the game if Florida State beats Florida” and “Please predict
what your overall happiness would be right after the game if
Florida State loses to Florida” as two levels of a within-subjects
factor. Results revealed a significant Personality Feedback ⫻
Differential Predicted Happiness interaction, F(2, 27) ⫽ 16.22,
p ⬍ .001. A 2⫺1⫺1 a priori contrast confirmed that future alone
participants (M ⫽ 4.70, SD ⫽ .99) predicted a significantly lower
degree of happiness in response to Florida State defeating the
University of Florida compared with both future belonging (M ⫽
6.60, SD ⫽ 0.69) and no-feedback control participants (M ⫽ 6.70,
8 DEWALL AND BAUMEISTER
SD ⫽ .67), F(2, 27) ⫽ 41.42, p ⬍ .001. In addition, a 2⫺1⫺1 a
priori contrast confirmed that future alone participants (M ⫽ 3.50,
SD ⫽ 0.71) predicted greater happiness in response to Florida
State being beaten by the University of Florida than both future
belonging (M ⫽ 2.30, SD ⫽ 1.34) and no-feedback control (M ⫽
1.90, SD ⫽ 1.10) participants, F(2, 27) ⫽ 11.20, p ⫽ .002. Thus,
future alone participants predicted less happiness to a Florida State
victory and less sadness to a Florida State defeat compared with
no-feedback control and future belonging participants.
Supplementary analyses were conducted to test whether the
exclusion, acceptance, and control groups differed in the degree to
which they perceived themselves as Florida State football fans and
the degree to which they cared about the outcome of the upcoming
football game. Results from an ANOVA revealed no significant
variation between the three groups on the fan index, F(2, 27) ⫽
2.29, p ⫽ .12. Pairwise comparisons revealed that future alone
participants (M ⫽ 10.30, SD ⫽ 2.63) had scores on the fan index
that were not significantly different from future belonging partic-
ipants (M ⫽ 8.90, SD ⫽ 3.45), F(1, 27) ⫽ 1.04, p ⫽ .32, and not
different from no-feedback control participants (M ⫽ 11.60, SD ⫽
2.27), F(1, 27) ⫽ 1.40, p ⫽ .25. Future belonging participants
(M ⫽ 8.90, SD ⫽ 3.45) had scores on the fan index that were
marginally lower than those of the no-feedback participants (M ⫽
11.60, SD ⫽ 2.27), F(1, 27) ⫽ 4.28, p ⬍ .06. Thus, the attenuated
forecasts (in the future alone condition) of affective reactions to
the future football game outcome were not due to participants
feeling as less a fan or feeling less interested in the football game.
Mood and emotion. To assess difference in mood, we con-
ducted two one-way ANOVAs by using the Mood Valence and
Mood Arousal subscales of the BMIS (Mayer & Gaschke, 1988) as
dependent measures. Both ANOVAs revealed no significant vari-
ation between the three experimental groups (both Fs ⬍ 1.28, ps ⬎
.30; see Table 2). These results suggest that increases in pain
threshold and pain tolerance following social exclusion were not
the result of differences in mood.
Link between physical and emotional insensitivity. To test the
hypothesis that unemotional affective forecasts were related to
increases in pain threshold and pain tolerance, we conducted a
regression analysis in which the measures of physical insensitivity
predicted measures of emotional insensitivity. For ease of analysis,
we computed a differential predicted happiness index by subtract-
ing scores to the question “Please predict what your overall hap-
piness would be right after the game if Florida State beats Florida”
from scores to the question “Please predict what your overall
happiness would be right after the game if Florida State loses to
Florida.” Pain threshold scores predicted differential predicted
happiness index scores, t(27) ⫽ ⫺2.36, ␤ ⫽ ⫺.32, p ⬍ .03. Pain
tolerance scores were also a significant predictor of differential
predicted happiness scores, t(27) ⫽ ⫺2.68, ␤ ⫽ ⫺.31, p ⫽ .01.
These findings provide evidence in support of the hypothesis that
social exclusion has two linked consequences, namely, numbness
to physiological pain and emotional numbness to possible future
events.
Discussion
Experiment 3 provided support for the hypothesis that responses
to social exclusion operate via a shared psychological system that
has related consequences for both physical and emotional pain. As
in Experiments 1 and 2, socially excluded participants exhibited
increases in both pain threshold and pain tolerance compared with
accepted and control participants. Excluded participants also re-
ported emotional states that did not differ from accepted and
control participants. It is difficult to ascertain whether excluded
participants truly felt no emotion or whether they were simply
refusing to admit that they were upset. If the latter were true,
however, excluded participants should be willing to describe and
predict their future emotional reactions, such as being happy if
their university were to win an important football game 2 months
in the future. The findings of Experiment 3 provide evidence that
social exclusion causes the emotion system to cease functioning
normally. Socially excluded participants not only failed to report
much in the way of current emotion; they also predicted little in the
way of a future emotional response to a major meaningful event.
Excluded participants predicted less happiness when forecasting
their emotional reaction to a positive event and greater happiness
when forecasting their emotional reaction to a negative event,
compared with participants in the other conditions.
It would have been plausible for exclusion to cause people to
identify either more or less strongly with Florida State University
and its football team, but they did not. In any case, the failure of
rejected people to predict much in the way of an emotional
reaction to future football outcomes was not mediated by changes
in their identification with Florida State University or expressed
concern with the outcome of the football game. After hearing that
they were likely to be alone later in their lives, they retained their
(relatively strong and positive) identification with their university
(i.e., they acknowledged their identity as Florida State University
football fans who cared about the outcome of an important game),
but they did not think they would feel much emotion about the
outcome of its football game against a major rival. Thus, the
results of Experiment 3 demonstrated that the body responds to
social exclusion feedback with a kind of physical shock reaction
that includes numbness and insensitivity to current physical pain
and possible future emotional events (cf. Campbell, Baumeister,
Dhavale, & Tice, 2003).
Experiment 4: Not Feeling Someone Else’s Pain
If social exclusion causes the emotion system to cease function-
ing normally, one should be less capable of empathy, and so one’s
reaction to another person’s suffering should be muted. Partici-
pants in Experiment 4 were asked to empathize with someone who
was suffering over a recent romantic failure. The main predictions
were that social exclusion would lead to less empathy toward the
stimulus person who had suffered the romantic rejection and that
this emotional insensitivity would be related to increases in both
pain threshold and pain tolerance.
Method
Participants. Thirty-one undergraduates (21 women) participated in
this experiment in exchange for partial course credit. Right-handedness
was stipulated, but one left-handed participant signed up and participated
anyway. Of the participants, 71% were White and 29% were a racial
minority. The average age was 18.5 years.
Materials and procedure. The materials and procedure for Experiment
4 were similar to Experiments 1–3. Participants arrived at the laboratory
individually for an experiment ostensibly concerned with the relationship
 SOCIAL EXCLUSION AND PAIN
between different aspects of personality, verbal and nonverbal perfor-
mance, and physical sensitivity. After giving informed consent, partici-
pants completed baseline measurements of pain threshold and pain toler-
ance (two trials each) and the Eysenck Personality Questionnaire (Eysenck
& Eysenck, 1975). By random assignment, participants received future
alone feedback, future belonging feedback, or no feedback.
After completing the BMIS mood measure, participants completed a
measure of empathy. The experimenter explained that there was another
experiment being conducted in the laboratory in which 1 participant writes
an essay about something going on in his or her life and another participant
reads the essay and answers some questions about it. Participants were told
that the participant who was supposed to read and respond to the essay did
not show up for the experiment. The experimenter explained that the
participant would read and respond to the essay. Participants were then
handed a manila folder that contained a handwritten essay (in male or
female handwriting,2 to match the participant’s gender) and short ques-
tionnaire. The content of the essay was adapted from Batson et al. (1995).
The essay read (in part):
Two days ago I broke up with my (girlfriend) boyfriend. We’ve been
going together since our junior year in high school and have been
really close, and it’s been great being at FSU together. I thought (s)he
felt the same, but things have changed. Now, (s)he wants to date other
people. (S)He says (s)he still cares a lot about me, but (s)he doesn’t
want to be tied down to just one person. I’ve been real down. It’s all
I think about. My friends all tell me that I’ll meet other (girls)guys and
they say that all I need is for something good to happen to cheer me
up. I guess they’re right, but so far that hasn’t happened.
Participants reported how sympathetic, warm, compassionate, soft-
hearted, and tender they felt toward the author of the essay. These adjec-
tives have been used in previous research to measure empathy (Batson,
1987, 1991; Batson et al., 1995). The internal reliability for the empathy-
related adjectives was good (Cronbach’s alpha ⫽ .92), and therefore an
empathy index was created by summing responses to the five empathy
adjectives (sympathetic, warm, compassionate, softhearted, and tender).
When finished, participants placed the essay and the questionnaire in an
envelope, sealed it, and handed it back to the experimenter. The experi-
menter then left the laboratory ostensibly to return the envelope to the
experimenter for the other study. When the experimenter returned, the
participant completed additional measurements of pain threshold and pain
tolerance (five trials each). Participants were then thoroughly debriefed,
thanked for their time, given partial course credit, and dismissed.
Results
Pain threshold and pain tolerance: Replication. As in the
previous three experiments, a future diagnostic forecast of social
exclusion produced increases in pain threshold compared with
participants who anticipated a future filled with social acceptance
or who received no feedback on a personality test. An ANCOVA
on the pain threshold scores revealed significant variation between
the three experimental groups, F(2, 28) ⫽ 54.12, p ⬍ .001. A
2⫺1⫺1 a priori contrast confirmed that future alone participants
had significantly higher pain threshold scores compared with fu-
ture belonging and no-feedback control participants, F(2, 28) ⫽
108.15, p ⬍ .001. A one-cell t test confirmed that the difference
between baseline and post-feedback pain threshold measurement
among future alone participants was significant, t(9) ⫽ ⫺5.99, p ⬍
.001, whereas there was no change from baseline in the other
conditions (both ts ⬍1, ns).
Socially excluded participants also demonstrated a significantly
higher tolerance for pain than participants assigned to the other
9
experimental groups. An ANCOVA on the pain tolerance scores
showed significant variation between the three experimental
groups, F(2, 28) ⫽ 45.25, p ⬍ .001. A 2⫺1⫺1 a priori contrast
confirmed that future alone participants had significantly higher
pain tolerance scores than participants who did not experience
social exclusion (i.e., future belonging and no-feedback control),
F(2, 28) ⫽ 89.99, p ⬍ .001. A one-cell t test confirmed that the
difference between baseline and post-feedback pain threshold
measurement among future alone participants was significantly
greater than zero, t(10) ⫽ ⫺6.69, p ⬍ .001. There was no change
from baseline in the other conditions (both ts ⬍1, ns). The means
and standard deviations are presented in Table 1.
Empathy. Socially excluded participants showed less empathy
than other participants toward the romantically rejected stimulus
person. An ANOVA on the five-item empathic emotion index
revealed significant variation between the three experimental
groups, F(2, 28) ⫽ 10.10, p ⬍ .001. A 2⫺1⫺1 a priori contrast
demonstrated that future alone participants reported feeling less
sympathy, compassion, warmth, tenderness, and softheartedness
toward the participant who had recently experienced relationship
loss, compared with future belonging and no-feedback control
participants, F(2, 28) ⫽ 19.67, p ⬍ .001. Pairwise comparisons
using the ANOVA mean-square error showed that future alone
participants (M ⫽ 32.80, SD ⫽ 7.66) had significantly lower
scores on the empathy index compared with future belonging
participants (M ⫽ 42.00, SD ⫽ 5.79), F(1, 28) ⫽ 12.37, p ⫽ .002,
d ⫽ 1.35. Future alone participants (M ⫽ 32.80, SD ⫽ 7.66) also
had lower scores on the empathy index compared with no-
feedback control participants (M ⫽ 43.55, SD ⫽ 3.59), F(1, 28) ⫽
17.67, p ⬍ .001, d ⫽ 1.79. Future belonging participants did not
differ in their empathy scores compared with no-feedback control
participants (F ⬍ 1, ns). Thus, future alone participants expressed
less emotional responsiveness to another person who had experi-
enced exclusion compared with future belonging and no-feedback
control participants.
Mood and emotion. Two one-way ANOVAs were conducting
by using the Mood Valence and Mood Arousal subscales of the
BMIS as dependent measures. Both ANOVAs revealed no signif-
icant variation between the three experimental groups (both Fs ⬍
1.79, ps ⬎ .20; see Table 2).
Link between physical and emotional insensitivity. To exam-
ine whether the lack of empathic responding among socially ex-
cluded participants was related to increased pain tolerance and
pain threshold, a regression analysis was conducted in which pain
threshold and pain tolerance scores predicted empathy scores.
Results revealed that pain threshold scores were a significant
predictor of empathy scores, t(28) ⫽ ⫺3.90, ␤ ⫽ ⫺.59, p ⫽ .001.
Pain tolerance scores also predicted empathy scores, t(27) ⫽
⫺5.24, ␤ ⫽ ⫺.61, p ⬍ .001. Thus, the increase in tolerance and
threshold for physical pain following social exclusion was related
to lack of empathic concern to others who had also experienced
social exclusion.
2
The essay was handwritten by either a male or female research assis-
tant. No participants expressed suspicion about the gender of the author of
the essay. In fact, many participants commented during the debriefing that
the handwriting was indicative of either a male or female author.
10 DEWALL AND BAUMEISTER
Discussion
Experiment 4 provided further evidence that the emotion system
temporarily ceases to function normally following social exclu-
sion. As compared with socially accepted and no-feedback con-
trols, socially excluded participants showed very little empathy
toward another person who had suffered a recent romantic
breakup. In fact, 1 participant in the future alone condition com-
mented “tough shit” upon reading about the other participant’s
romantic woes—a comment that clearly indicates an uncharitable
and unsympathetic attitude toward another person’s heartbreak.
The increases in pain threshold and pain tolerance among so-
cially excluded participants were related to their lack of empathy
toward the other participant. This supports the broad view that
social exclusion causes both physical numbness and emotional
insensitivity to others.
Experiment 5: I Can’t Feel Another Person’s Physical
Pain
One possible boundary condition for Experiment 4 is that it
explored the effects of social exclusion on empathy toward some-
one else who had recently suffered romantic rejection. Although
the essay used in Experiment 4 has been used in previous research
to evoke general empathy (Batson et al., 1995), the lack of em-
pathic responding among socially excluded participants may have
been due to the author of the essay having also experienced a form
of social exclusion. Excluded participants may have refused to
empathize with another rejected person because the other person’s
rejection may have reminded them of their own recent social
exclusion. Therefore, Study 5 measured empathy toward someone
who had suffered a broken leg.
Exclusion was manipulated by having participants complete a
vivid recall task in which they completed a brief autobiographical
narrative recalling a time they experienced social rejection, social
acceptance, or an unrelated event. This was meant to prime par-
ticipants with thoughts of their relevant experience, which would
in turn influence responding on a questionnaire designed to mea-
sure empathy. Previous research has shown that real and imagined
events activate many of the same neural and psychological pro-
cesses (Kosslyn et al., 1999; McGuire, Shah, & Murray, 1993),
and other research has shown that the exclusion vivid recall task
evokes responses similar to those found in manipulations of im-
mediate rejection (Gardner et al., 2000; Pickett, Gardner, &
Knowles, 2004). It was predicted that the current manipulation of
rejection would produce the same emotional numbness as in ex-
cluded participants in Experiments 1– 4. The main prediction of
Experiment 5 was that socially rejected participants would express
less empathy toward the author of the essay compared with ac-
cepted and control participants.
Method
Participants. One-hundred twenty-five participants (84 women) par-
ticipated in this study in exchange for partial course credit. Results from 9
participants were discarded from all analyses because of incomplete ques-
tionnaire packets (thus, there were originally 134 participants). Of the
participants, 74% were White and 26% were a racial minority. The average
age was 18.5 years.
Materials and procedure. Participants arrived at a large classroom in
groups of 10 –20 for a study ostensibly concerned with how people under-
stand each other. After giving informed consent, participants completed a
short questionnaire packet that contained instructions for an autobiograph-
ical narrative, a mood questionnaire, and additional materials aimed at
measuring empathic responding. By random assignment, participants were
assigned to one of three autobiographical narrative conditions: social
rejection, social acceptance, and control.
The instructions for the social rejection narrative read as follows:
On the next pages, you will write an essay about a time when you
experienced rejection or exclusion by others. Think of a time when
you felt that others did not want to be in your company and when you
did not feel a strong sense of belongingness with another person or
group. Nearly everyone has experienced such an experience more than
once; please choose an especially important and memorable event.
The instructions urged participants to describe the rejection experience in
as thorough detail as possible and to provide the “full story.” Other
participants wrote about a childhood experience of social acceptance. The
instructions for the social acceptance narrative mirrored the social rejection
instructions and replaced only words that dealt directly with the experience
of rejection or acceptance. Participants assigned to the control condition
were instructed to write a detailed essay about one of the things he or she
did the previous day.
After participants completed writing their narrative, participants com-
pleted the Positive and Negative Affect Schedule (Watson, Clark, &
Tellegen, 1988). Participants then read a short vignette that depicted a
college student (of unknown gender) and was meant to evoke empathic
responding from participants. The vignette read (in part):
Two days ago I broke my leg playing intramural sports. I’ve been
playing on the same intramural team for the past three years and I’m
upset that my season has been cut short. I’m experiencing pain
because of my injury. I’m also having a tough time getting around
campus, as there are lots of hills and stairs that make it hard to use my
crutches on. The parking people won’t let me get a handicapped
permit because they said my injury was only temporary. I’ve been real
down. It’s all I think about.
Participants reported how sympathetic, warm, compassionate, soft-
hearted, and tender they felt toward the author of the essay. The internal
reliability of these five items was good (Cronbach’s alpha ⫽ .86), and
therefore an empathy index was created by summing responses to these
five items. When participants had finished the empathy rating task, they
were debriefed, given partial course credit, thanked for their time, and
dismissed.
Results
Empathy. Participants who wrote about a time they experi-
enced social rejection showed significantly less empathy toward
the author compared with participants in the other two conditions.
An ANOVA on the five-item empathic emotion index revealed
significant variation between the three experimental groups, F(2,
122) ⫽ 6.72, p ⫽ .002. A 2⫺1⫺1 a priori contrast confirmed that
socially rejected participants felt significantly less sympathy,
warmth, compassion, tenderness, and softheartedness toward the
author of the essay compared with both socially accepted and
control participants, F(1, 122) ⫽ 12.79, p ⫽ .001. Pairwise com-
parisons using the ANOVA mean-square error showed that re-
jected participants (M ⫽ 26.33, SD ⫽ 9.42) empathized with the
essay author significantly less than did accepted participants (M ⫽
31.82, SD ⫽ 8.65), F(1, 122) ⫽ 7.53, p ⫽ .007, d ⫽ .61. In
 SOCIAL EXCLUSION AND PAIN
addition, rejected participants (M ⫽ 26.33, SD ⫽ 9.42) showed
less empathic concern toward the essay author compared with
control participants (M ⫽ 32.96, SD ⫽ 8.43), F(1, 122) ⫽ 12.21,
p ⫽ .001, d ⫽ .74. Socially accepted and control participants did
not differ significantly in terms of how much they empathized with
the essay author (F ⬍ 1, ns). Thus, rejected participants were quite
unsympathetic toward a member of their peer group who had
suffered a physically painful injury and was having difficulty
adjusting to the changes brought about by the injury. These results
suggest that rejection brings about a lack of empathic responding
toward others, regardless of whether the person has experienced
rejection or another traumatic and physically painful event.
Mood and emotion. Two one-way ANOVAs were conducting
by using the Positive Affect and Negative Affect subscales of the
Positive and Negative Affect Schedule as dependent measures.
There was no significant variation between the three experimental
groups in terms of reported positive affect, F(2, 122) ⫽ 1.31, p ⫽
.27, or negative affect, F(2, 122) ⫽ 1.89, p ⫽ .16 (see Table 2).
Discussion
Experiment 5 provided evidence that rejected participants ex-
pressed less empathy than accepted and control participants toward
a person suffering physical pain and inconvenience from a broken
leg. These findings provide converging evidence that the emotion
system temporarily ceases to function normally following social
exclusion, which leads to emotional numbness toward another
person in physical pain. Thus, the excluded people’s lack of
empathy found in Experiment 4 appears to be part of a broader
pattern rather than being tied to the special case of refusing to
empathize with someone else who has suffered social exclusion.
General Discussion
Social exclusion threatens a fundamental human motivation for
strong and stable relationships. Such a threat strikes at the core of
human psychological and physical well-being. Social exclusion
can have destructive effects on mental and physical well-being, but
socially excluded people frequently report relatively numb emo-
tional states. This paradox leaves open the mystery of why socially
excluded people often respond to such a disturbing event as social
exclusion in an emotionally numb manner. One possible explana-
tion for this anomalous emotional responding following social
exclusion is that the emotion system temporarily ceases function-
ing normally, leaving a person momentarily unable to respond to
emotional events in a customary fashion. This should lead to
numbness to physically painful stimuli and emotional numbness to
current and possible future events.
Across four experiments, socially excluded participants consis-
tently showed both decreased sensitivity and increased tolerance to
physical pain, compared with socially accepted and control par-
ticipants and compared with their own baseline responses. The
estimated effect size of the difference in pain threshold and pain
tolerance between socially excluded participants and participants
in the other conditions was consistently large, exceeding the cri-
teria typically used to describe large effects (Cohen, 1977). Thus,
participants responded to the social exclusion feedback by growing
increasingly numb to physical pain stimuli, and these increases in
pain threshold and pain tolerance were large in comparison to
11
participants who experienced social acceptance, received a nega-
tively valenced future diagnostic forecast, or received no feedback
on their personality test.
The second goal of the current investigation was to examine
whether insensitivity to physical pain might be linked to emotional
numbness following social exclusion. In all five experiments,
excluded participants reported emotional states that did not differ
significantly from socially accepted and control participants. Ad-
mittedly, these results suffer from the uncertainties that accompany
the use of self-reports of emotional responding, especially the idea
that excluded participants might be reluctant to admit that they are
upset over the experimental feedback. But excluded participants
should be willing and able to predict future emotional reactions or
empathize with others. The results of Experiments 3–5 showed that
excluded participants failed to make the standard predictions about
their emotional reactions to future events and failed to empathize
with others’ suffering. Social exclusion feedback caused people to
predict less happiness over a future football victory and less
sadness over a possible defeat. Social exclusion also led partici-
pants to become less empathic toward another person who had just
suffered a relationship loss, which as a form of social exclusion
might have elicited high empathy from participants who them-
selves had also experienced exclusion. And recalling a past expe-
rience of social exclusion made people less empathic toward a
student who was suffering over a broken leg (which was unrelated
to social exclusion). These findings suggest that social exclusion
produces a sweeping shutdown of the emotional system.
Thus, the findings from the current experiments indicate that the
body responds to social exclusion in much the same manner as it
responds to physical injury. Exclusion may well produce a bio-
chemical reaction that leads to temporary numbness to physical
pain. This physical numbness is also linked to emotional respond-
ing, which is demonstrated in emotional insensitivity.
The findings of Experiment 3 make a novel contribution to the
affective forecasting literature by identifying a moderating vari-
able to the previously documented pattern of people overestimat-
ing the impact of future emotional events on their lives (e.g.,
Wilson & Gilbert, 2003). Exclusion temporarily disrupted the
capacity for the emotion system to function normally, and this led
excluded participants to predict a significantly less emotional
response to both future positive and negative events than partici-
pants who had experienced social acceptance or than control
participants. This finding is potentially intriguing in that socially
excluded people made affective forecasts that were likely more
realistic than participants in the other conditions (D. T. Gilbert,
personal communication, December 16, 2004), though we did not
collect data on participants’ actual emotional responses to the
game’s outcome. Regardless of accuracy, our finding that de-
creased sensitivity to pain was closely linked to affective forecasts
of indifference provides further reason to think that the physical
pain and social emotion systems may share a common physiolog-
ical basis (or at least are intimately linked).
In sum, the results of the current studies provide evidence of an
immediate physical numbing effect in the aftermath of social
exclusion, which is related to the degree of emotional numbness
that socially excluded people exhibit. These findings help to re-
solve the paradox in the rejection literature by showing that one
possible reason people report relative emotional numbness to
social exclusion is that they experience a physiological reaction
12 DEWALL AND BAUMEISTER
that creates numbness to physically and emotionally painful stim-
uli. It is not likely that these physical numbing effects are long
lasting. Instead, physical numbness likely functions as an imme-
diate buffer that wards off potentially threatening aspects of one’s
current surroundings. The possible long-term effects of repeated or
chronic social rejection remain to be explored.
Limitations and Alternative Explanations
The five experiments reported in this article provide consistent
evidence that social exclusion produces increases in pain threshold
and pain tolerance and that this physical numbness to pain is
related to emotional insensitivity. Despite the consistency of these
effects, however, several alternative explanations warrant further
consideration. A first possibility is that social exclusion simply
constitutes a form of bad news. From this perspective, receiving
any negatively valenced feedback should temporarily cause the
emotion system to cease functioning normally, resulting in numb-
ness to painful physical stimuli. The results of our misfortune
control condition in Experiment 2 provide evidence that is incon-
sistent with this perspective. Participants assigned to that condition
received a negatively valenced future diagnostic forecast that was
unrelated to their future belongingness status, namely, that they
would become accident prone later in life. Despite the relative
negativity of this diagnostic forecast, misfortune control partici-
pants responded to painful stimuli in a manner that was most
similar to participants who experienced social acceptance or who
received no personality feedback.
A second possibility is that the methods used for assessing pain
threshold assessed something other than pain threshold, such as a
general ability to report tactile stimulation. If the pain threshold
measures were tapping something other than physical numbness to
pain, the pain threshold scores in the current study should not be
associated with measures of tolerance for physical pain. Against
this line of reasoning, Time 2 pain threshold scores (controlling for
baseline measurement) were strongly correlated with pain toler-
ance scores (controlling for baseline measurement). The correla-
tions between pain threshold and pain tolerance were consistently
large and were reduced significantly when controlling for social
exclusion condition. The correlations were as follows: Experiment
1, r(29) ⫽ .43, p ⬍ .02; controlling for exclusion condition,
r(30) ⫽ ⫺.20, p ⫽ .30; Experiment 2, r(29) ⫽ .82, p ⬍ .001;
controlling for exclusion condition, r(27) ⫽ .67, p ⬍ .001; Exper-
iment 3, r(26) ⫽ .46, p ⫽ .01; controlling for exclusion condition,
r(25) ⫽ ⫺.06, p ⫽ .77; Experiment 4, r(27) ⫽ .85, p ⬍ .001;
controlling for exclusion condition, r(26) ⫽ .22, p ⫽ .27. Thus, it
appears that our measures of pain threshold were measuring one
component of physical pain sensation as a result of the social
exclusion manipulation, namely, sensitivity to physical pain.
Another possible limitation in the current studies is whether
there is any relationship between chronic levels of pain sensitivity
and emotional sensitivity not induced by the social exclusion
manipulation. To examine this possibility, we predicted Time 2
pain sensitivity scores from the emotional sensitivity measures in
Experiments 3 and 4, while simultaneously controlling for social
exclusion condition and baseline pain sensitivity scores. In Exper-
iment 3, results revealed no significant relationship between Time
2 pain threshold and tolerance scores and scores on the differential
predicted happiness index (␤s ⫽ .08 and .15, respectively; both
ps ⬎ .24). Similar results emerged in Experiment 4. After control-
ling for baseline pain sensitivity measurement and social exclusion
condition, we found that the relationship between Time 2 pain
threshold and tolerance scores and empathy scores was nonsignif-
icant (␤s ⫽ ⫺.18 and ⫺.16, respectively; both ps ⬎ .14). Thus,
these findings indicate that the changes induced by condition in
pain sensitivity were directly related to the differences in emo-
tional sensitivity as a result of the social exclusion manipulation
and that within each condition there were not any strong and
consistent relationships between the physical and emotional sen-
sitivity measures after controlling for baseline measurements and
social exclusion condition.
A final explanation is that increased pain tolerance and pain
threshold among socially excluded participants were due to in-
creased self-regulation instead of a temporary shutdown in the
emotion system. Pain tolerance is often used as a measure of
self-regulation (Hilgard et al., 1974), and participants may have
responded to social exclusion with increased self-regulation as a
means of improving their moods. The results of Experiments 1–5
provided consistent and converging evidence that socially ex-
cluded participants did not report moods that were significantly
different from socially accepted or control participants, which
casts doubt on the likelihood that excluded participants were trying
to improve their moods. The present data cannot easily rule out the
idea that reduced sensitivity and increased tolerance to pain re-
flects increased self-regulation in response to social exclusion. Past
work has clearly shown the opposite pattern, however, namely,
that social exclusion leads to poor self-regulation (Baumeister et
al., 2005; DeWall, Baumeister, & Vohs, 2004). If the current
results could be explained in terms of effective self-regulation,
excluded participants would have also probably shown high levels
of empathy toward others’ suffering. This was, however, not the
case.
Concluding Remarks
The purpose of the current investigation was to resolve the
seeming paradox that people respond to social exclusion with
emotional numbness. The need to belong is a fundamental and
pervasive human motivation, and the behavioral consequences of
social exclusion are often quite large. It was therefore a mystery as
to why people fail to show strong emotional responses in reaction
to such a seemingly momentous event as social exclusion.
The apparent solution to the mystery, as suggested by the
present findings, is that the body responds to social exclusion in
much the same way (and possibly by means of the same physio-
logical mechanisms) as it responds to painful physical injury.
Physical sensation and emotion reactivity become dulled and in-
sensitive. This may temporarily reduce suffering and enable the
individual to begin to cope without feeling acutely distressed each
moment.
Reduced suffering and potentially better coping would of course
be adaptive responses, but past work has suggested that the im-
mediate behavioral effects of rejection are often far from beneficial
(e.g., Baumeister et al., 2005; Twenge et al., 2001, 2002, 2003),
even extending to behaviors that would seemingly reduce chances
of gaining future acceptance. From the perspective of the present
findings, one possible implication is that people use their emo-
tional faculties to help understand their world. Emotional numb-
 SOCIAL EXCLUSION AND PAIN
ness may therefore impair the person’s ability to relate to others
and function effectively in a complex social world. The present
results showed that excluded people were seemingly unable to
furnish the typical (and presumably socially adaptive) patterns of
affective forecasting and empathy. Most likely there are other
cognitive tasks that social animals routinely perform with the aid
of their emotion systems and that, accordingly, may abruptly be
lost when social exclusion temporarily causes the emotional sys-
tem to cease functioning normally.
There was no sign that people realize how much they are
affected by rejection. When asked to forecast their future reactions,
our excluded participants did not say they were uncertain as to how
they would feel. Rather, they engaged in self-reflection and told us
they would not have a strong emotional response to whether their
football team won or lost. Recent work has increasingly shown that
some emotional capabilities are essential for good decision making
and other adaptive behavioral patterns (e.g., Bechara, Damasio,
Damasio, & Lee, 1999; Loewenstein, Weber, Hsee, & Welch,
2001). If normal emotional functioning is essential for effective
decision making and behaviors, then when the emotion system
ceases to function normally, the person may be given faulty
information with which to make decisions and interact with others,
which may have potentially costly results.
The emerging image of the socially rejected person is thus one
of someone who is emotionally and hence socially impaired— but
may not realize it. Without a functioning emotion system, the
person may not anticipate how his or her actions will affect others
or even the person’s own future outcomes. Without recognizing
this impairment, the person may even act on those faulty intuitions.
In that way, the short-term relief brought by the emotional numb-
ness caused by social exclusion may also bring destructive social
consequences.
References
Argyle, M. (1987). The psychology of happiness. London: Methuen.
Bancroft, J., Skirmshire, A., & Simkins, S. (1976). The reasons people give
for taking overdoses. British Journal of Psychiatry, 128, 538 –548.
Batson, C. D. (1987). Prosocial motivation: Is it ever truly altruistic? In L.
Berkowitz (Ed.), Advances in experimental social psychology (Vol. 20,
pp. 65–122). New York: Academic Press.
Batson, C. D. (1991). The altruism question: Toward a social psycholog-
ical answer. Hillsdale, NJ: Erlbaum.
Batson, C. D., Klein, T. R., Highberger, L., & Shaw, L. L. (1995).
Immorality from empathy-induced altruism: When compassion and jus-
tice conflict. Journal of Personality and Social Psychology, 68, 1042–
1054.
Baumeister, R. F. (1990). Suicide as escape from the self. Psychological
Review, 97, 90 –113.
Baumeister, R. F., DeWall, C. N., Ciaracco, N. J., & Twenge, J. M. (2005).
Social exclusion impairs self-regulation. Journal of Personality and
Social Psychology, 88, 589 – 604.
Baumeister, R. F., & Leary, M. R. (1995). The need to belong: Desire for
interpersonal attachments as a fundamental human motivation. Psycho-
logical Bulletin, 117, 497–529.
Baumeister, R. F., & Scher, S. J. (1988). Self-defeating behavior patterns
among normal individuals: Review and analysis of common self-
destructive tendencies. Psychological Bulletin, 104, 3–22.
Baumeister, R. F., Twenge, J. M., & Nuss, C. K. (2002). Effects of social
exclusion on cognitive processes: Anticipated aloneness reduces intelli-
gent thought. Journal of Personality and Social Psychology, 83, 817–
827.
13
Bechara, A., Damasio, H., Damasio, A., & Lee, G. P. (1999). Different
contributions of the human amygdala ventromedial prefrontal cortex to
decision-making. Journal of Neuroscience, 19, 5473–5481.
Buckley, K. E., Winkel, R. E., & Leary, M. R. (2004). Reactions to
acceptance and rejection: Effects of level and sequence of relational
evaluation. Journal of Experimental Social Psychology, 40, 14 –28.
Bush, G., Luu, P., & Posner, M. I. (2000). Cognitive and emotional
influences in anterior cingulate cortex. Trends in Cognitive Sciences, 4,
215–222.
Cacioppo, J. T., Hawkley, L. C., & Berntson, G. G. (2003). The anatomy
of loneliness. Current Directions in Psychological Science, 12, 71–74.
Campbell, W. K., Baumeister, R. F., Dhavale, D., & Tice, D. M. (2003).
Responding to major threats to self-esteem: A preliminary, narrative
study of ego-shock. Journal of Social and Clinical Psychology, 22,
79 –96.
Carden, S. E., Hernandez, N., & Hofer, M. A. (1996). The isolation and
companion comfort responses of 7- and 3-day-old rat pups are modu-
lated by drugs active at the opioid receptor. Behavioral Neuroscience,
110, 324 –330.
Carden, S. E., & Hofer, M. A. (1990). Socially mediated reduction of
isolation distress in rat pups is blocked by naltrexone but not by Ro
15–1788. Behavioral Neuroscience, 104, 457– 463.
Cohen, J. (1977). Statistical power analysis for the behavioral sciences.
New York: Academic Press.
Craig, A. D., & Dostrovsky, J. O. (1999). Medulla to thalamus. In P. Wall
& R. Melzack (Eds.), Textbook of pain (pp. 183–214). New York:
Churchill Livingstone.
DeWall, C. N., Baumeister, R. F., & Vohs, K. D. (2004). Recovering from
rejection: Undoing the self-regulation deficits stemming from social
exclusion. Manuscript submitted for publication.
Downey, G., & Feldman, S. I. (1996). Implications for rejection sensitivity
for intimate relationships. Journal of Personality and Social Psychology,
70, 1327–1343.
Eisenberger, N. I., & Lieberman, M. D. (2004). Why rejection hurts: The
neurocognitive overlap between physical and social pain. Trends in
Cognitive Sciences, 8, 294 –300.
Eisenberger, N. I., Lieberman, M. D., & Williams, K. D. (2003, October
10). Does rejection hurt? An fMRI study of social exclusion. Science,
302, 290 –292.
Eysenck, H. J., & Eysenck, S. B. G. (1975). Manual of the Eysenck
Personality Questionnaire. San Diego, CA: EDITS.
Gardner, W. L., Pickett, C. L., & Brewer, M. B. (2000). Social exclusion
and selective memory: How the need to belong influences memory for
social events. Personality and Social Psychology Bulletin, 26, 486 – 496.
Gilbert, D. T., Gill, M. J., & Wilson, T. D. (2002). The future is now:
Temporal correction in affective forecasting. Organization Behavior and
Human Decision Processes, 88, 430 – 444.
Gilbert, D. T., Pinel, E. C., Wilson, T. D., Blumberg, S. J., & Wheatley, T.
(1998). Immune neglect: A source of durability bias in affective fore-
casting. Journal of Personality and Social Psychology, 75, 617– 638.
Hariri, A. R., Bookheimer, S. Y., & Mazziotta, J. C. (2000). Modulating
emotional responses: Effects of a neocortical network on the limbic
system. Neuroreport, 11, 43– 48.
Hawkley, L. C., Burleson, M. H., Berntson, G. G., & Cacioppo, J. T.
(2003). Loneliness in everyday life: Cardiovascular activity, psychoso-
cial context, and health behaviors. Journal of Personality and Social
Psychology, 85, 105–120.
Hawton, K., Cole, D., O’Grady, J., & Osborn, M. (1982). Motivational
aspects of deliberate self-poisoning in adolescents. British Journal of
Psychiatry, 141, 286 –291.
Herman, B. H., & Panksepp, J. (1978). Effects of morphine and naloxone
on separation distress and approach attachment: Evidence for opiate
mediation of social affect. Pharmacology Biochemistry & Behavior, 9,
213–220.
14 DEWALL AND BAUMEISTER
Hilgard, E. R., Ruch, J. C., Lange, A. F., Lenox, J. R., Morgan, A. H., &
Sachs, L. B. (1974). The psychophysics of cold pressor pain and its
modification through hypnotic suggestion. American Journal of Psy-
chology, 87, 17–31.
Insel, T. R., & Winslow, J. T. (1991). Central administration of oxytocin
modulates the infant rat’s response to social isolation. European Journal
of Pharmacology, 203, 149 –152.
Jepson, C., Loewenstein, G., & Ubel, P. (2001). Actual versus estimated
differences in quality of life before and after renal transplant. Unpub-
lished manuscript, Department of Social and Decision Sciences, Carne-
gie Mellon University.
Kanarek, R. B., & Carrington, C. (2004). Sucrose consumption enhances
the analgesic effects of cigarette smoking in male and female smokers.
Psychopharmacology, 173, 57– 63.
Kehoe, P., & Blass, E. M. (1986a). Behaviorally functional opioid systems
in infant rats: II. Evidence for pharmacological, physiological, and
psychological mediation of pain and stress. Behavioral Neuroscience,
100, 624 – 630.
Kehoe, P., & Blass, E. M. (1986b). Opioid-mediation of separation distress
in 10-day-old rats: Reversal of stress with maternal stimuli. Develop-
mental Psychobiology, 19, 385–398.
Kiecolt-Glaser, J. K., Ricker, D., George, J., Messick, G., Speicher, C. E.,
Garner, W., & Glaser, R. (1984). Urinary cortisol levels, cellular immu-
nocompetency, and loneliness in psychiatric inpatients. Psychosomatic
Medicine, 46, 15–24.
Kimbrell, T. A., George, M. S., Parekh, P. I., Ketter, T. A., Podell, D. M.,
Danielson, A. L., et al. (1999). Regional brain activity during transient
self-induced anxiety and anger in healthy adults. Biological Psychiatry,
46, 454 – 465.
Konecka, A. M., & Sroczynska, I. (1990). Stressors and pain sensitivity in
CFW mice: Role of opioid peptides. Archives Internationales de Physi-
ologie et de Biochimie, 98, 245–252.
Kosslyn, S. M., Pascual-Leone, A., Felician, O., Camposano, S., Keenan,
J. P., Thompson, W. L., et al. (1999, May 7). The role of area 17 in
visual imagery: Convergent evidence from PET and rTMS. Science, 284,
167–170.
Leary, M. R., & Springer, C. A. (2000). Hurt feelings: The neglected
emotion. In R. Kowalski (Ed.), Behaving badly: Aversive behaviors in
interpersonal transgression (pp. 151–175). Washington, DC: American
Psychological Association.
Loewenstein, G., O’Donoghue, T., & Rabin, M. (in press). Projection bias
in predicting future utility. Quarterly Journal of Economics.
Loewenstein, G. F., Weber, E. U., Hsee, C. K., & Welch, N. (2001). Risk
as feelings. Psychological Bulletin, 127, 267–286.
Lynch, J. J. (1979). The broken heart: The medical consequences of
loneliness. New York: Basic Books.
MacDonald, G., Kingsbury, R., & Shaw, S. (2005). Adding insult to injury:
Social pain theory and response to social exclusion. In K. Williams, J.
Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social
exclusion, rejection, and bullying (pp. 77–90). New York: Psychology
Press.
MacDonald, G., & Leary, M. R. (2005). Why does social exclusion hurt?
The relationship between social and physical pain. Psychological Bul-
letin, 131, 202–223.
Mayer, J. D., & Gaschke, Y. N. (1988). The experience and meta-
experience of mood. Journal of Personality and Social Psychology, 55,
102–111.
McGuire, P. K., Shah, G. M. S., & Murray, R. M. (1993). Increased blood
flow in Broca’s area during auditory hallucinations in schizophrenia.
Lancet, 342, 703–706.
Mercer, M. E., & Holder, M. D. (1997). Antinociceptive effects of palat-
able sweet ingesta on human responsivity to pressure pain. Physiology &
Behavior, 61, 311–318.
Murray, F. S., & Hagan, B. C. (1973). Pain threshold and tolerance of
hands and feet. Journal of Comparative and Physiological Psychology,
84, 639 – 643.
Myers, D. (1992). The pursuit of happiness. New York: Morrow.
Naranjo, J. R., & Fuentes, J. A. (1985). Association between hypoalgesia
and hypertension in rats after short-term isolation. Neuropharmacology,
24, 167–171.
Nelson, E. E., & Panksepp, J. (1998). Brain substrates of infant–mother
attachment: Contributions of opioids, oxytocin, and norepinephrine.
Neuroscience & Biobehavioral Reviews, 22, 437– 452.
Nelson, L. D., & Morrison, E. L. (2005). The symptoms of resource
scarcity: Judgments of food and finances influence preferences for
potential partners. Psychological Science, 16, 167–173.
Orbach, I., Mikulincer, M., King, R., Cohen, D., & Stein, D. (1997).
Thresholds and tolerance of physical pain in suicidal and nonsuicidal
patients. Journal of Consulting and Clinical Psychology, 65, 646 – 652.
Panksepp, J. (1998). Affective neuroscience: The foundations of human and
animal emotions. London: Oxford University Press.
Panksepp, J., Herman, B. H., Conner, R., Bishop, P., & Scott, J. P. (1978).
The biology of social attachments: Opiates alleviate separation distress.
Biological Psychiatry, 13, 607– 618.
Panksepp, J., Vilberg, T., Bean, N. J., Coy, D. H., & Kastin, A. J. (1978).
Reduction of distress vocalization in chicks by opiate-like peptides.
Brain Research Bulletin, 3, 663– 667.
Petrovic, P., Kalso, E., Petersson, K. M., & Ingvar, M. (2002). Placebo and
opioid analgesia—imaging a shared neural network. Neuroimage, 16,
1142–1150.
Pickett, C. L., Gardner, W. L., & Knowles, M. (2004). Getting a cue: The
need to belong and enhanced sensitivity to social cues. Personality and
Social Psychology Bulletin, 30, 1095–1107.
Pomerleau, O. F., Turk, D. C., & Fertig, J. B. (1984). The effects of
cigarette smoking on pain and anxiety. Addictive Behaviors, 9, 265–271.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton,
NJ: Princeton University Press.
Rossi, G. C., Pasternak, G. W., & Bodnar, R. J. (1993). Synergistic
brainstem interactions for morphine analgesia. Brain Research, 624,
171–180.
Rushen, J., Boissy, A., Terlouw, E. M., & de Passillé, A. M. (1999). Opioid
peptides and behavioral and physiological responses of dairy cows to
social isolation in unfamiliar surroundings. Journal of Animal Science,
77, 2918 –2924.
Spear, L. P., Enters, E. K., Aswad, M. A., & Louzan, M. (1985). Drug and
environmentally induced manipulations of the opiate and serotonergic
systems alter nociception in neonatal rat pups. Behavioral and Neural
Biology, 44, 1–22.
Sufka, K. J., & Hughes, R. A. (1990). Dose and temporal parameters of
morphine-induced hyperalgesia in domestic fowl. Physiology & Behav-
ior, 47, 385–387.
Sufka, K. J., & Weed, N. C. (1994). Construct validation of behavioral
indices of isolation stress and inflammatory nociception in young do-
mestic fowl. Physiology & Behavior, 55, 741–746.
Twenge, J. M., Baumeister, R. F., DeWall, C. N., Ciarocco, N. J., &
Bartels, J. M. (2006). Social exclusion reduces prosocial behavior.
Unpublished manuscript.
Twenge, J. M., Baumeister, R. F., Tice, D. M., & Stucke, T. S. (2001). If
you can’t join them, beat them: Effects of social exclusion on aggressive
behavior. Journal of Personality and Social Psychology, 81, 1058 –1069.
Twenge, J. M., & Campbell, W. K. (2003). “Isn’t it fun to get the respect
that we’re going to deserve?” Narcissism, social rejection, and aggres-
sion. Personality and Social Psychology Bulletin, 29, 261–272.
Twenge, J. M., Catanese, K. R., & Baumeister, R. F. (2002). Social
exclusion causes self-defeating behavior. Journal of Personality and
Social Psychology, 83, 606 – 615.
Twenge, J. M., Catanese, K. R., & Baumeister, R. F. (2003). Social
exclusion and the deconstructed state: Time perception, meaningless-
 SOCIAL EXCLUSION AND PAIN
ness, lethargy, lack of emotion, and self-awareness. Journal of Person-
ality and Social Psychology, 85, 409 – 423.
Uchino, B. N., Cacioppo, J. T., & Kiecolt-Glaser, J. K. (1996). The
relationship between social support and physiological processes: A
review with emphasis on underlying mechanisms and implications for
health. Psychological Bulletin, 119, 488 –531.
Van Boven, L., & Loewenstein, G. (2003). Social projection of transient
drive states. Personality and Social Psychology Bulletin, 29, 1159 –
1168.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and vali-
dation of brief measures of positive and negative affect: The PANAS
scales. Journal of Personality and Social Psychology, 54, 1063–1070.
Wiedenmayer, C. P., Goodwin, G. A., & Barr, G. A. (2000). The effect of
periaqueductal gray lesions on responses to age-specific threats in infant
rats. Developmental Brain Research, 120, 191–198.
Williams, K. D., Cheung, C. K. T., & Choi, W. (2000). CyberOstracism:
Effects of being ignored over the Internet. Journal of Personality and
Social Psychology, 79, 748 –762.
15
Wilson, T. D., & Gilbert, D. T. (2003). Affective forecasting. In M. P.
Zanna (Ed.), Advances in experimental social psychology (Vol. 35, pp.
345– 411). San Diego, CA: Academic Press.
Wilson, T. D., Wheatley, T., Kurtz, J., Dunn, E., & Gilbert, D. T. (2004).
When to fire: Anticipatory versus post-event reconstrual of uncontrol-
lable events. Personality and Social Psychology Bulletin, 30, 340 –351.
Wilson, T. D., Wheatley, T., Meyers, J. M., Gilbert, D. T., & Axsom, D.
(2000). Focalism: A source of durability bias in affective forecasting.
Journal of Personality and Social Psychology, 78, 821– 836.
Zadro, L., Williams, K. D., & Richardson, R. (2004). How low can you go?
Ostracism by a computer is sufficient to lower self-reported levels of
belonging, control, self-esteem, and meaningful existence. Journal of
Experimental Social Psychology, 40, 560 –567.
Received February 24, 2005
Revision received August 15, 2005
Accepted August 25, 2005 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 111–123 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.111

Information Quantity and Quality Affect the Realistic Accuracy

of Personality Judgment

Tera D. Letzring Shannon M. Wells

University of California, Riverside Sanford Systems, Incorporated

David C. Funder
University of California, Riverside

Triads of unacquainted college students interacted in 1 of 5 experimental conditions that manipulated

information quantity (amount of information) and information quality (relevance of information to
personality), and they then made judgments of each others’ personalities. To determine accuracy, the
authors compared the ratings of each judge to a broad-based accuracy criterion composed of personality
ratings from 3 types of knowledgeable informants (the self, real-life acquaintances, and clinician-
interviewers). Results supported the hypothesis that information quantity and quality would be positively
related to objective knowledge about the targets and realistic accuracy. Interjudge consensus and
self– other agreement followed a similar pattern. These findings are consistent with expectations based
on models of the process of accurate judgment (D. C. Funder, 1995, 1999) and consensus (D. A. Kenny,
1994).

Keywords: personality judgment, judgment accuracy, realistic accuracy, information quantity, informa-
tion quality

Every day, people make personality judgments of other people.
The person making the judgments (the judge) has known some
targets for several years and has seen them in many different
contexts (e.g., a sibling or a spouse), whereas other targets are
people the judge has just met and has seen only in a relatively
limited context (e.g., a new coworker, a bank teller). Not surpris-
ingly, these judgments vary in how accurate they are, perhaps
because they are made on the basis of varying degrees of acquain-
tance and information that vary in how relevant they are to
personality.
These two aspects of the information available to a judge are
referred to as information quantity, the sheer amount of informa-
tion that is available, and information quality, the degree to which
the available information is relevant to personality (Blackman &
Funder, 1998; Funder, 1999). The concept of information quantity
is straightforward in that it assumes that judges who have access to
more information about the personality of the target will make
Tera D. Letzring and David C. Funder, Department of Psychology,
University of California, Riverside; Shannon M. Wells, Sanford Systems,
Incorporated, Lake Elsinore, California (now Key Data Systems).
This article is based, in part, on the doctoral dissertation by Shannon M.
Wells titled “The Effect of Information Quantity and Quality on the
Accuracy of Personality Judgments.” Some of these findings were also
presented at the annual meeting of the Society for Personality and Social
Psychology as part of a symposium titled “The Accuracy of Interpersonal
Judgment.” Data gathering for this article was supported by National
Institute of Mental Health Grant MH42427 to David C. Funder.
Correspondence concerning this article should be addressed to Tera D.
Letzring, who is now at the Department of Psychology, Idaho State
University, Pocatello, ID 83209. E-mail: tera.letzring@gmail.com
111
more accurate judgments of personality. The effect of information
quantity has also been referred to as the acquaintanceship effect
(Colvin & Funder, 1991; Kenny, 1991), because as people are
acquainted longer they are assumed to acquire more information
about each other. The concept of information quality is more
complex, but at its core refers to the likelihood that even when
information quantity is held constant, different contexts of ac-
quaintanceship might vary in the degree to which personality-
relevant information becomes available in them. For example, a
context in which everybody acts the same would yield only low-
quality information for personality judgment because information
about individual differences in personality would not be revealed,
whereas a context in which behavior is relatively free to vary
would yield higher quality information because people would be
more likely to behave in a way that reflects their personalities
(Snyder & Ickes, 1985). It is also possible that different kinds of
information—for example, thoughts and feelings as opposed to
hobbies and activities (Anderson, 1984)—are differentially infor-
mative about personality. When more personality-relevant infor-
mation is available, the accuracy of judgments of individual dif-
ferences in personality is expected to be higher. Information
quantity and quality are two aspects of good information, one of
four factors proposed to be related to judgment accuracy (Funder,
1999).
It is difficult to examine information quantity and quality inde-
pendently of each other in natural settings of acquaintanceship,
because people are likely to emit a broader range of cues to
personality and share more personality-relevant information with
acquaintances they have known longer. For example, according to
social penetration theory (Altman & Taylor, 1973), people share
information about a larger number of the aspects of their person-
112 LETZRING, WELLS, AND FUNDER
ality and share more personal information within these aspects as
a relationship progresses, which suggests that information in-
creases in both quantity and quality along with acquaintance.
In the current study, the quantity and quality of the information
available to judges were experimentally and independently manip-
ulated so that the effects of these two factors on the accuracy of
personality judgment could be examined, in order to achieve two
goals. The first goal was to further examine the effects of infor-
mation quantity on judgment accuracy by (a) testing for the ac-
quaintanceship effect in a new context and across different levels
of information quantity and by (b) examining the effects of infor-
mation quantity on a relatively new construct, realistic accuracy, in
addition to the more traditional constructs of consensus and self–
other agreement. The second goal of the present research was to
directly examine the effects of experimentally manipulated infor-
mation quality on realistic accuracy, consensus, and self– other
agreement, which is a more distinctive contribution, as this is one
of the first empirical studies to examine these effects.
Indicators of Judgment Accuracy
Three indicators of judgment accuracy were assessed: realistic
accuracy, consensus, and self– other agreement.
Realistic Accuracy
Realistic accuracy refers to a hypothetical construct representing
the level of agreement between a personality judgment and what a
target is really like (Funder, 1995). This construct cannot be
directly measured by any single personality or behavior rating, as
any single rating is highly uncertain as an indicator of what a
person is really like. Instead, the ideal of realistic accuracy can be
approached to the degree that multiple methods of measurement
are used and combined to form a broad-based accuracy criterion
for each target of judgment. The assumption is that this broad-
based criterion is closer to what the target is really like than is any
single rating, because random errors in the ratings should cancel
each other out as more ratings are combined. In the current study,
we formed an accuracy criterion by using a composite rating based
on judgments from real-life acquaintances, psychologists who
interviewed the targets, and the self.
The difficulty of the criterion problem has led many researchers
to shy away from studying the accuracy of judgments of real
people and to instead turn their attention to models of the cognitive
processing of artificial stimuli (Funder, 1987, 1995). As research
progresses, it behooves investigators to begin to respond to the
criterion problem instead of continuing to bypass it. The ultimate
goal should be an appropriate, broad-based accuracy criterion,
which would ideally derive from multiple modes of assessment,
including the target’s self-perspective, ratings from knowledgeable
informants, clinical judgments, behavioral measures, life outcome
data, and perhaps even biological information such as hormone
levels and functional MRI images (Funder, 1995). We propose that
research that seeks to go beyond single operationalizations of the
accuracy of personality judgment should refer to the outcome as
realistic accuracy in order to distinguish this construct from other
constructs that measure some other singular form of agreement or
accuracy in personality judgment.
Consensus
Consensus can be measured more directly than realistic accu-
racy and is more commonly studied. Consensus is the level of
agreement between the personality judgments rendered by two or
more people about another person. Consensus can be used as an
indicator of accuracy, but it is important to keep in mind that
consensus is not necessarily related to what the person is actually
like, because many people could agree about the personality of
someone else, and all could be wrong, as is the case when inac-
curate stereotypes are used to make judgments.
Self–Other Agreement
Self– other agreement can also be directly measured and has
been used frequently in previous research (Ambady, Hallahan, &
Rosenthal, 1995; Blackman & Funder, 1998). It refers to the level
of agreement between judgments made about another person and
that person’s self-judgments. The self-judgments are often implic-
itly assumed to reflect reality, and the judgments made by others
are assumed to be accurate to the degree that they are related to the
self-judgments. However, self– other agreement does not guaran-
tee highly accurate judgments, because the self-rating will not
adequately represent one’s real personality if people are unwilling
or unable to provide accurate judgments of themselves (Hofstee,
1994; Kolar, Funder, & Colvin, 1996). Despite this complication
of using self– other agreement to measure accuracy, many inves-
tigations refer to self– other agreement simply as accuracy (Am-
bady et al., 1995; Bernieri, Zuckerman, Koestner, & Rosenthal,
1994; Blackman & Funder, 1998). However, in the current article
the term accuracy is used more specifically to describe realistic
accuracy so that realistic accuracy can be differentiated from
consensus and self– other agreement.
We report findings concerning all three indicators of judgment
accuracy for two reasons. First, realistic accuracy, although related
to consensus and self– other agreement, is far from synonymous
with either of these operationalizations and could potentially yield
different results. Second, consensus and self– other agreement
have been used in many prior studies (e.g., Anderson, 1984;
Blackman & Funder, 1998; Kenny, Albright, Malloy, & Kashy,
1994), and our presentation of these constructs allows comparison
with the existing literature.
Methods of Personality Judgment
Realistic Accuracy Model
The Realistic Accuracy Model (RAM; Funder, 1995, 1999)
describes an interpersonal and cognitive process that results in
accurate personality judgment when all four of its stages are
successfully completed. First, the target of the judgment must
display cues or behaviors that are relevant to the characteristic
being judged in such a way and in contexts that are available to the
judge. Then, the judge must detect these cues and correctly use
them to make a judgment. RAM assumes that the four stages
combine in a multiplicative fashion so that a failure at any of the
four stages will make accuracy impossible. For example, cues can
only be detected if they are available.
Four factors are thought to influence the judgment process by
affecting one or more stages of RAM (Funder, 1995): the good
 INFORMATION AND PERSONALITY JUDGMENT
judge (Allport, 1937; Kolar, 1995; Taft, 1955; Vernon, 1933; Vogt
& Colvin, 2003), good target (Colvin, 1993), good trait (Borkenau
& Liebler, 1993; Funder & Dobroth, 1987; John & Robins, 1993;
Norman & Goldberg, 1966), and good information (Blackman &
Funder, 1998; Funder & Colvin, 1988; Funder, Kolar, & Black-
man, 1995). Two aspects of good information, information quan-
tity and information quality, are the focus of the current article.
A prediction regarding the relations between information quan-
tity and quality and all three indicators of accuracy can be gener-
ated on the basis of RAM. Even though RAM is a model for
accuracy, it can also be used to make predictions about consensus
if one assumes that personality is something real, in which case
two accurate judgments of personality must agree with each other,
and therefore consensus can be expected to be high at sufficiently
high levels of accuracy. A linear increase between information and
realistic accuracy, consensus, and self– other agreement was used
as the prediction based on RAM. This is a conservative prediction
because it is not possible to determine the degree to which the
manipulated levels of information quantity and quality are equally
spaced and, therefore, whether the increases in accuracy should
also be equally spaced.
Weighted-Average Model
The Weighted-Average Model (WAM; Kenny, 1991, 1994)
offers a detailed description of the basis of consensus in person-
ality judgment. The model predicts that consensus will not increase
with acquaintanceship when judges see completely overlapping
behaviors of the target and interpret what they see in exactly the
same way (Kenny, 1991). However, when these assumptions are
replaced by the more realistic expectations that different judges
will interpret behaviors in different ways and will not detect
exactly the same behaviors of the target, WAM predicts that
consensus will increase rapidly at low levels of acquaintance and
remain about the same across higher levels of acquaintance. This
reasoning led us to predict that consensus would increase between
low and medium levels of information quantity and quality but
would be the same between medium and high levels of information
quantity and quality.
The prediction based on WAM was tested only for consensus
because WAM is explicitly formulated as a model of the process
of consensus, and it has been suggested that the relations between
different indicators of accuracy and acquaintanceship are not nec-
essarily the same (Kenny, 1991). For example, Kenny (1991)
pointed out that accuracy can increase with acquaintance even as
consensus stays the same, as is the case when all judges become
more accurate but continue to agree with each other to the same
degree.
It was not feasible to determine whether the prediction based on
RAM was superior to the prediction based on WAM, as the two
predictions were highly correlated.1 However, it was still interest-
ing to determine how well the obtained data fit each prediction.
Past Findings and Rationale for Current Experiment
Information quantity has received a fair amount of research
attention, and although the findings have been mixed, the main
conclusion has been that, all other things being equal, to know
someone longer is to know him or her better. Information quality
113
has received less attention, although the available evidence sug-
gests that not all information is created equal and that some kinds
of information are more likely to be related to judgment accuracy
than others (e.g., Anderson, 1984).
Information Quantity
A fair amount of research has examined the relationships be-
tween information quantity and two aspects of personality judg-
ment, consensus and self– other agreement. To our knowledge,
research has not previously been published on the relationship
between information quantity and realistic accuracy. Past findings
on consensus and self– other agreement have been mixed.
In between-subjects designs, in which different participants are
involved at each level of acquaintance, findings concerning con-
sensus have both supported and not supported the acquaintance-
ship effect. Blackman and Funder (1998) experimentally manipu-
lated level of acquaintance by having some people observe a target
interacting for 5 or 10 min and other people observe a target for 25
or 30 min, and they found that consensus did not increase with
acquaintance. However, researchers have found that consensus is
higher among real-life acquaintances than among relative strang-
ers—that is, people who viewed dyadic interactions for either 5
min (Funder & Colvin, 1988) or 25 or 30 min (Blackman &
Funder, 1998). Furthermore, a meta-analysis suggests that for
agreeableness, conscientiousness, neuroticism, and openness to
experience, consensus is greater among long-term acquaintances
than among people who were unacquainted or had interacted only
once (Kenny et al., 1994). In general, higher levels of consensus
have been associated with higher acquaintance in studies with
between-subjects designs in which there was a relatively large
difference in the amount of available information between the high
and low acquaintance groups, but not when there was a relatively
small difference in acquaintance.
Between-subjects designs have generally found support for the
acquaintanceship effect when self– other agreement is the out-
come. For example, self– other agreement was higher among peo-
ple who watched videotaped unstructured dyadic interactions for
25 to 30 min versus 5 or 10 min (Blackman & Funder, 1998),
people who had been roommates for more than 10 months versus
less than 10 months (Bernieri at al., 1994), people who indicated
they knew the target of judgment extremely well versus not at all
(Paunonen, 1989), and real-life acquaintances versus people who
had watched 5-min unstructured dyadic interactions (Funder et al.,
1995). In between-subjects designs, information quantity appears
to have a positive linear relationship with self– other agreement
across all levels of acquaintance. However, a complication com-
mon to between-subjects designs of consensus and self– other
agreement is that real-life acquaintances have self-selected to
know a person for longer, which introduces the possibility that
confounds such as similarity to the target and liking of the target
influence accuracy in addition to length of acquaintance. This is
why, in the current study, we have experimentally manipulated
information quantity.
In within-subjects research designs in which the same partici-
pants are involved at each level of acquaintance, previous findings
With contrast weights of –1, 0, ⫹1 for RAM and –2, ⫹1, ⫹1 for
1
WAM, the correlation between the predictions was .87.
114 LETZRING, WELLS, AND FUNDER
suggest that as acquaintance increases, consensus generally re-
mains constant, whereas self– other agreement generally increases.
Consensus has been shown to stay about the same as acquaintance
increases across a variety of samples, situations, and lengths of
time, including previously unacquainted students who participated
in 30-min group interactions on 4 consecutive days (Park & Judd,
1989), students who participated in weekly 20-min group meetings
for 7 weeks (Paulhus & Bruce, 1992), and students living near
each other in a dorm from 2 to 33 weeks after the start of a
semester (Park, Kraus, & Ryan, 1997). Furthermore, a meta-
analysis suggested that consensus did not increase among people
who interacted for between 8 min and 2 hr in both laboratory and
naturalistic settings (Kenny et al., 1994).
On the other hand, an examination of self– other agreement
provided evidence that self– other agreement generally increased
among students who participated in weekly 20-min group meet-
ings for 7 weeks (Paulhus & Bruce, 1992). Also, findings by
Borkenau and colleagues (Borkenau, Mauer, Riemann, Spinath, &
Angleitner, 2004) suggest that acquaintance and accuracy are
positively related when accuracy is operationalized as the level of
agreement with ratings from the target, acquaintances of the target,
experimenters, and a confederate who interacted with the target;
and when acquaintance is operationalized as the number of behav-
ioral episodes observed and rated by different judges that were
included in a composite judgment.
One possible explanation for the inconsistent findings concern-
ing the acquaintanceship effect is that consensus was used as the
outcome, which is not necessarily the appropriate way to approach
this issue if one wishes to determine the extent to which judgments
become more accurate with increased acquaintance. It is possible
that the ratings of the judges in the studies mentioned previously
(Kenny et al., 1994; Park & Judd, 1989; Park et al., 1997; Paulhus
& Bruce, 1992) did become more accurate across time, but this
change might not be reflected in consensus if the ratings continued
to have the same level of similarity with each other (Blackman &
Funder, 1998).
Another possible limitation of past research is that researchers
have simply assumed that people who have known each other
longer have also acquired more information about each other,
which is not necessarily true. One way to determine whether
people in longer relationships have actually acquired more
personality-relevant information is to directly measure the amount
of objective information they know about the target. In the current
study, this concern was addressed by asking participants a series of
factual questions about their interaction partners that sampled from
the kinds of information people might learn in an initial interac-
tion. This sample of relatively objective questions was used as a
manipulation check to determine whether participants who interact
for a longer period of time acquire more information about each
other. If the manipulation is successful, then people who interact
longer will have acquired more of this information, concomitant to
picking up more information relevant to personality, and therefore
we expected to find a moderate positive relation between infor-
mation acquisition and realistic accuracy, consensus, and self–
other agreement.
The information quantity analyses are a conceptual replication
of a study by Blackman and Funder (1998). The current project
also went beyond Blackman and Funder’s study and other previous
research in several important ways. First, participants in Blackman
and Funder’s experiment were exposed to targets via videotaped
two-person interactions, whereas participants in the current study
actually interacted with each other in three-person groups before
making personality judgments. This method may have different
implications for the way judges detect and use information when
making personality judgments. For example, judges watching vid-
eotapes have no need to be concerned with how they are perceived
or what they will say next, whereas judges who are involved in an
interaction are likely to be concerned with these and other issues.
More important, many researchers, including Blackman and
Funder, have used self– other agreement as a proxy for accuracy,
whereas in the current study, we went beyond self– other agree-
ment and examined a construct closer to the ideal of realistic
accuracy by using a broad-based accuracy criterion.
Information Quality
Although a considerable amount of research has focused on the
relationship between information quantity and personality judg-
ment, surprisingly little research has focused on an arguably even
more important aspect of information—information quality, or the
personality relevance of the information that becomes available
within a given period of time. Intuitively, it seems apparent that
not all kinds of information contribute equally to the achievement
of an accurate judgment, as it is possible to be acquainted with
someone for only a short time but to know him or her very well,
or to be acquainted with someone for years and still know little
about that person. From a research perspective, the first step in
examining information quality is to determine the extent to which
it is related to indicators of accuracy, including realistic accuracy,
consensus, and self– other agreement, which was one goal of the
current study.
Anderson’s (1984) groundbreaking study supported the idea that
information quality is related to self– other agreement. Anderson
found that observing an interview in which questions about
thoughts and feelings were discussed yielded higher self– other
agreement of personality ratings than did observing an interview in
which hobbies and activities were discussed. One conclusion that
can be based on this finding is that thoughts and feelings are higher
quality information than hobbies and activities, and this is why the
judges who learned about thoughts and feelings were able to agree
more closely with the target’s self-judgments of personality than
were the judges who learned about hobbies and activities.
Another aspect that might influence the quality of the informa-
tion that is likely to be revealed is the situation’s “strength”
(Snyder & Ickes, 1985). “Strong” situations limit the range of
behavior that people display because they include explicit rules or
evoke implicit norms to which people generally adhere. At the
other extreme, “weak” situations allow for considerable behavioral
variation because there are few rules or norms for typical behavior.
Therefore, in comparison with a strong situation, a weak situation
should allow for more behavioral variation, the availability of
higher quality individuating information, the acquisition of more
objective information, and greater judgment accuracy.
The present study was designed to begin the investigation of
what occurs when people interact in situations with different levels
of information quality. A strong situation was created experimen-
tally by giving participants the specific task of answering a long set
of trivia questions. Two weaker situations were created by telling
 INFORMATION AND PERSONALITY JUDGMENT
participants to talk about whatever they would like or to get to
know one another as well as possible. We predicted that more
personality-relevant information would be revealed in the two
weak situations than in the strong situation and that realistic
accuracy, self– other agreement, and consensus would be higher in
the two weak situations. We also predicted that more personality-
relevant information would be available in the weak situation in
which participants were told to get to know each other than in the
weak situation in which participants were told to talk about what-
ever they would like, because the participants would be more
likely to reveal or ask about personality-relevant information as
they attempted to get to know each other than when they were
simply trying to pass the time by talking about whatever they liked.
Research Questions
With the theoretical basis of both the RAM (Funder, 1995) and
the WAM (Kenny, 1991) in mind, in the current study we exam-
ined how quantity and quality of available information affect
realistic accuracy, consensus, and self– other agreement.
Hypothesis 1: Participants who interact for longer periods of
time will achieve higher levels of realistic accuracy, consen-
sus, and self– other agreement than people who interact for
shorter periods of time.
To examine this issue, we compared personality judgments fol-
lowing three-person unstructured interactions of three lengths
(minimal interaction, 50 min, and 3 hr). On the basis of RAM
(Funder, 1995, 1999), we expected that all three indicators of
personality judgment—realistic accuracy, consensus, and self–
other agreement—would increase linearly across these experimen-
tal conditions. On the basis of WAM (Kenny, 1991), we expected
consensus to increase between the low- and medium-quantity
conditions and to stay about the same between the medium- and
high-quantity conditions. Thus, the predicted relationships be-
tween information quantity and consensus differ slightly depend-
ing on which model is used, and both predictions were tested.
Hypothesis 2: Participants who interact in situations in which
more personality-relevant information is likely to be available
will achieve higher levels of realistic accuracy, consensus,
and self– other agreement than participants who interact in
situations in which less personality-relevant information is
likely to be available.
In one of the first experimental investigations to directly examine
this issue, we compared the accuracy of personality judgments
made following one of three 50-min interactions that vary in the
amount of personality-relevant information likely to be available
in them. The low-quality condition was a strong situation that
allowed for little behavioral variation and was expected to elicit
the least amount of personality-relevant information and therefore
the lowest levels of realistic accuracy, consensus, and self– other
agreement. The medium- and high-quality conditions were both
relatively weak situations that were expected to elicit more behav-
ioral variation and therefore higher levels of all indicators of
accuracy than the strong situation. Between the two weak situa-
tions, participants who were given the objective to get to know
each other were expected to share a broader range of personality-
115
relevant information and to achieve higher levels of realistic ac-
curacy, consensus, and self– other agreement than participants who
were simply instructed to talk about whatever they would like.
Therefore, on the basis of RAM, we predicted a linear relationship
between information quality and all three indictors of accuracy.
We could also make a prediction based on WAM. Even though the
model does not directly consider information quality, we have
defined information quality as the availability of more personality-
relevant information with time held constant, so that the acquain-
tanceship parameter of WAM should increase with information
quality. Therefore, on the basis of WAM, we predicted that con-
sensus would increase between the low- and medium-quality con-
ditions and remain about the same between the medium- and
high-quality conditions.
Method
The current study was a between-subjects experimental design in which
each participant interacted in one of five conditions. The first independent
variable, information quantity, had three levels: low (minimal information
condition), medium (short unstructured condition), and high (long unstruc-
tured condition). The second independent variable, information quality,
also had three levels: low (trivia condition), medium (short unstructured
condition), and high (get to know condition). The effects of each indepen-
dent variable were examined for three dependent variables: realistic accu-
racy (agreement between personality judgments and a broad-based accu-
racy criterion), self– other agreement (agreement between judgments and
the target’s self-ratings), and consensus (agreement between two judges
about a target).
Participants
A total of 506 undergraduate students participated in the Riverside
Accuracy Project—Phase 2 (RAP–II) and were paid $10 per hour for their
time. A core group of 180 target participants (90 men and 90 women) were
recruited via announcements made in psychology classes and fliers placed
on bulletin boards advertising “research on personality.” These participants
were asked to recruit two close acquaintances (for a total of 326 acquain-
tance informants) to provide personality descriptions of themselves and the
target participants. The ethnic breakdown of the target participants was
38% Asian American, 20% Hispanic, 14% Caucasian, 12% African Amer-
ican, and 16% other or not specified. There were three groups of each
possible gender composition (female–female–female [FFF], male–male–
male [MMM], female–female–male [FFM], and female–male–male
[FMM]) within each experimental condition, and all of the gender com-
positions were analyzed together because of the balanced nature of the
design and because the number of participants within each gender compo-
sition would be too small to find reliable results if analyzed separately.
This is the second article to come out of the extensive RAP–II data set,
and the analyses do not overlap with the previous project, which focused
on the constructs of ego-control and ego-resiliency (see Letzring, Block, &
Funder, 2005), or with future planned projects.
Materials
California Adult Q (CAQ) set. The CAQ (Block, 1961, as modified for
use by nonprofessionals by Bem & Funder, 1978) consists of 100 carefully
formulated descriptive statements about personality (e.g., is critical/skep-
tical, is personally charming, is cheerful). In the Q-item rating, each item
is responded to using a 9-point Likert-type scale ranging from 1 (extremely
uncharacteristic) to 9 (extremely characteristic). Participants rated them-
selves and their group interaction partners using the Q-item rating, and
acquaintance informants also rated the target participants using the Q-item
116 LETZRING, WELLS, AND FUNDER
rating. Participants rated both group interaction partners at the same time
for each item of the CAQ, which implies that for each item the interaction
partners were directly compared with each other at the time of the judg-
ment. We expected this procedure to help the judges make individuating
and thoughtful ratings for each item and both targets. Clinicians rated the
target participants using the traditional Q-sort method, in which the 100
items are placed into a forced-choice, quasi-normal distribution so that
each category (1 through 9) contains a predetermined number of items.
Information fact sheet. An information fact sheet was constructed for
RAP–II that consisted of open-ended questions that were designed to
sample from the wide array of information that could become available
during the experimental interactions. The information fact sheet was used
as a manipulation check to determine whether participants in higher quan-
tity and quality interactions actually acquired more information about their
interaction partners than did participants in lower quantity or quality
interactions. Questions included information about age, place of birth,
political affiliation, future goals, family makeup, and so forth. No attempt
was made to assess the information that participants picked up in any
comprehensive fashion, or even to focus specifically on personality-
relevant information (because techniques for either goal are presently
unavailable). Rather, we aimed simply to obtain a small sample of the
information that one person might learn about another during a casual
interaction, with the assumption that scores on this measure would corre-
late with the amount of information, including personality-relevant infor-
mation, acquired overall.
The information fact sheet was completed by the target participants
about their group interaction partners immediately following the ratings of
personality and was completed on a later occasion by each participant
about him- or herself. Each question was scored correct (1 point) if the
answer was equivalent to the answer given by the self; half (.5 points) if the
answer was similar, but not equivalent, to the answer given by the self; or
incorrect (0 points) if the answer was different from the answer given by
the self or if there was no response. Each information fact sheet was scored
by two independent judges, and, as the decisions about the correctness of
an answer were relatively objective, any discrepancies in scoring were
reconciled by a third judge. The reliability of the 19 items was high enough
to justify using the sum as a score representing information acquisition
(Cronbach’s alpha ⫽ .77).
Procedures
Overview. Target participants came to the lab on three separate occa-
sions and also completed three packets of self-report questionnaires outside
of the lab, only some of which were used in the current analyses. During
the first session, participants interacted in three-person groups in one of
five experimental conditions and then made personality judgments of their
two interaction partners. During a subsequent session, each participant was
interviewed by a clinical psychologist who described the personality of the
participant after the conclusion of the interview. Additionally, acquain-
tance informants completed questionnaires regarding the target participant
with whom they were acquainted.
Presession. At the presession, participants were given the first take-
home packet of questionnaires and were scheduled into one of the five
experimental conditions. The times of the conditions were established in
advance, and the participants chose a session that fit with their schedules.
Although technically this procedure was only quasi-random, it did assign
participants to conditions independently of any aspect of the experimental
design, and we have no reason to believe that participants in the various
conditions differed from each other in any way that might have affected the
outcome of the study. Additionally, each participant was shown a folder
containing pictures of the other participants in the same group to determine
whether he or she had ever seen the other participants. If the participant
recognized another participant in the same group, the participant was
assigned to a different group. This procedure ensured that none of the
participants were acquainted with their interaction partners prior to the
experimental group interaction.
Experimental conditions. Target participants interacted in the labora-
tory in one of five experimental conditions designed to vary in the quantity
and quality of interpersonal information likely to become available in them.
Each of the groups consisted of three previously unacquainted target
participants in one of four possible gender compositions: all female (FFF),
all male (MMM), 2 females and 1 male (FFM), and 1 female and 2 males
(FMM). An equal number of the four gender compositions was assigned to
each experimental condition. A total sample size of 180 allowed for a
balanced design with 3 groups of each gender composition per condition,
for a total of 12 groups per condition.2
In all conditions, participants were seated in a room at a round table and
given verbal instructions. At the conclusion of the interaction, participants
rated their two partners using three measures, which included the rating
form of the CAQ and the information fact sheet.
Minimal information condition. Participants rated the other two par-
ticipants in the room immediately following a short set of instructions
explaining that they were studying first impressions and to do the best they
could describing their first impressions of the other two people in the room.
They were asked not to speak to each other but were in each others’
physical presence for as long as it took to complete the ratings (less than
1 hr). This situation provided an empirical estimate of the baseline of
realistic accuracy, consensus, and self– other agreement and was used as
the low-level condition for information quantity.
Trivia quiz condition. Participants were presented with a packet con-
taining 380 trivia-type questions, each of which had a single correct
answer. They were informed that they would have 50 min to work jointly
through the packet of questions and arrive unanimously at what they
believed to be the correct answers. This context was designed to be a strong
situation that would leave relatively little room for extraneous commentary
or the disclosure of personality-relevant information, as the group would
spend the entire interaction time responding to the trivia questions. This
condition was used as the low-level condition for information quality.
Short unstructured condition. Participants were told that they could
talk about anything they liked for the next 50 min. No attempt was made
to direct or suggest what should be done over the course of the interaction.
This context provided a weak situation in that participants were able to do
and say whatever they liked and to exhibit individual differences in social
behavior, which should have included at least some personality-relevant
information on which judgments could be based. This condition was used
as the mid-level condition for both information quantity and quality.
Long unstructured condition. Participants were told that they could
talk about anything they liked for the next 3 hr, and as in the short
unstructured condition, no attempt was made to direct or suggest what
should be done over the course of the interaction. The experimenter
returned to the room halfway through the interaction to give participants a
short break, at which time snacks were provided and a restroom break was
offered. As in the short unstructured condition, this context provided a
weak situation with very little structure and should have included at least
some personality-relevant information on which judgments could be based.
This condition was used as the high-level condition for information
quantity.
Get to know (GTK) condition. Participants were told that their task for
the next 50 min was to get to know each other as well as possible and to
learn as much as they could about what type of person each of them was.
2
On completion of data collection, it was found that 2 target participants
were inadvertently included in the study twice. For both participants, the
second condition that they were in was different from the first. Analyses
were done with these groups included and excluded. The differences
between these two sets of findings were within rounding error. To maintain
a fully balanced design, we report results from the data of all 60 groups.
 INFORMATION AND PERSONALITY JUDGMENT
This context provided a weak situation that was expected to yield a wide
range of behaviors relevant to personality, and the objective of getting to
know each other was expected to guide participants to probe for and to
remember information related to personality. This condition was used as
the high-level condition for information quality.
Life history interview. Participants were individually interviewed by
one of four professionally trained (i.e., MSW, MA in counseling, or PhD
in clinical psychology) interviewers who had experience with a college
population. With consent of the participants, all interviews were video-
taped. The clinicians conducted a 1-hr semistructured life history interview
adapted from a protocol used for many years by the Institute of Personality
Assessment and Research (Craik et al., 2002). The protocol used in the
current study was adapted to better apply to college students and sought to
capture a broad range of personality-relevant information without explic-
itly asking about sensitive topics and risky behaviors. Each interview
started with the clinician asking the participant, “Tell me something about
yourself,” and then covered a broad range of topics, including college and
academic experiences, future plans, interpersonal relationships, and child-
hood and family history. In conclusion, each participant was asked to
describe “a defining event in [their] life that had a significant impact on or
changed [their] life in some way.”3 Following the interview, the clinician
completed a Q-sort description of the target participant. Some of the
interviews were later viewed by a second clinician, and in these cases the
second clinician also described the participant’s personality using the
CAQ.4 When two ratings of the participant were available, a composite
score was computed for each item and used in subsequent analyses. The
average interclinician profile agreement was r ⫽ .50 (SD ⫽ 0.17),5 and the
coefficient of internal consistency based on a composite of the two raters
was .67.
Acquaintance–informant ratings. Participants were asked to provide
the names and contact information of the two people who knew them best
at the university. Acquaintances came into the lab and provided descrip-
tions of the target participant by whom they had been identified, using the
rating format of the CAQ. The average profile agreement for the 100 items
as rated by two acquaintances was r ⫽ .40 (SD ⫽ 0.19).6 The average
coefficient of internal consistency, or the dependability of the profile based
on a composite of the two acquaintances, was .57.
Analyses
Realistic accuracy criterion. Many researchers have used self– other
agreement as a proxy for accuracy, but in the current project we strived to
go beyond this simple operationalization by comparing judgments to a
broader based criterion composed of ratings by three types of knowledge-
able informants: the self, two acquaintances nominated by the target, and
a professionally trained clinician-interviewer. The 100 items of the CAQ
were rated by each type of rater, and these ratings were used to compute a
composite score for each item. First, when ratings from two acquaintances
or two clinicians were available, these ratings were averaged.7 Then, a
simple average for each item was computed across the three ratings from
the self, acquaintances, and clinician. The construct of realistic accuracy is
very new, so there is not an established level of reliability for determining
the adequacy of the accuracy composite. However, on the basis of the
magnitudes of correlations that are reported in the self– other agreement
literature (Bernieri et al., 1994; Funder, 1980), the mean alpha reliability of
the accuracy composite seemed high enough to justify the use of these
average ratings (M ⫽ 0.42, SD ⫽ 0.13, range ⫽ 0.08 – 0.72).8
Computation of realistic accuracy. Profile analysis allows for an ex-
amination of judgments in regard to a target’s overall personality by the use
of the entire set of 100 CAQ items in a single analysis. The computation
of a profile correlation simply involves correlating two sets of ratings for
the same target across the 100 items of the CAQ. First, to examine realistic
accuracy, we correlated the ratings by each judge of each interaction
partner with the accuracy composite for that interaction partner, resulting
in a profile correlation that represented the relation between the ratings of
117
Figure 1. Computation of profile realistic accuracy scores. Note that
profile correlations were computed for each of the three judges in each
experimental group, and these scores were Z transformed and averaged to
compute scores for each group; the scores of all 12 groups in each type of
experimental interaction were averaged to determine the accuracy for the
corresponding interaction type. CAQ ⫽ California Adult Q.
the judge and an approximation of what the target is really like. This
procedure yielded six scores for each group (two scores for each of three
judges). These scores could be transformed using Fisher’s r-to-z transfor-
mation and averaged so that each group had a single realistic accuracy
score (see Figure 1).
Computation of consensus. To examine consensus, we correlated the
ratings provided by the two interaction partners of each target with each
other, resulting in a profile correlation that represented the degree of
similarity in the partner’s ratings of that target. This analysis yielded three
consensus scores for each group (one for each target), which could be Z
transformed and averaged so that each group had one consensus score.
Computation of self– other agreement. Finally, to examine self– other
agreement, we correlated the ratings of each interaction partner with the
partner’s self-ratings, resulting in a profile correlation that represented the
relation between the ratings of the judge and the target’s self-ratings. As in
realistic accuracy, this procedure yielded six self– other agreement scores
3
The full protocol for the clinical interview is available on the Riverside
Accuracy Project’s Web site at http://www.rap.ucr.edu/interview.htm.
4
When the participant scheduled for a live interview was unable to keep
the appointment, clinicians were given the option to observe and rate a
randomly chosen, previously recorded interview. In this manner, 47 inter-
views were observed and rated by a second clinician.
5
This number can be compared with 30 random pairings of ratings by
clinicians who described different targets, which resulted in an average
profile correlation of .29.
6
This number can be compared with 30 random pairings of ratings by
acquaintances who described different targets, which resulted in an average
profile correlation of .24.
7
When ratings from only one acquaintance or clinician were available,
then there was no need to compute an average before averaging the ratings
with the other sources.
8
On the basis of correlations among types of raters, the accuracy
criterion is appropriate to use for computing profile correlations: self–
acquaintance (M ⫽ 0.48, SD ⫽ 0.17), self– clinician (M ⫽ 0.39, SD ⫽
0.14), acquaintance– clinician (M ⫽ 0.37, SD ⫽ 0.19); Spearman–Brown
for three raters ⫽ .68.
118 LETZRING, WELLS, AND FUNDER
per group, and these scores could be Z transformed and averaged so that
each group had a single self– other agreement score.
Components of accuracy scores. Cronbach (1955) was one of the first
to point out that profile correlations may be confounded with several other
elements besides differential accuracy, including elevation, differential
elevation, and stereotype accuracy, which result from the use of response
sets or a reliance on stereotypes. The effects of response sets were not of
concern for between-groups comparisons in the current experiment be-
cause participants were randomly assigned to one of five experimental
conditions, and therefore, participants with various response sets were
equally distributed among the conditions. The effects of stereotype accu-
racy also were not of concern because stereotype accuracy should have
equivalent effects on the magnitude of the profile correlations across all
conditions, and we were interested in the differences between conditions
and not in the absolute magnitude of accuracy. For this reason, we used the
raw profile correlations in our cross-group comparisons, as did Blackman
and Funder (1998) when they conducted similar analyses.
However, it is true that as a result of stereotype accuracy, the baseline
correlation between two sets of ratings was not expected to be zero, as
some items of the CAQ are generally rated higher than others, regardless
of the target (Blackman & Funder, 1998). To determine baseline profile
correlations within the current data, we computed the correlations among
30 random pairs of ratings. For realistic accuracy, the profile correlations
were computed between the ratings of one target and the accuracy com-
posite for a different target, which resulted in a baseline profile correlation
of .36. For consensus, the profile correlations were computed between two
interaction partners of different targets, which resulted in a baseline profile
correlation of .25. For self– other agreement, the profile correlations were
computed between the ratings of one target and the self-ratings of a
different target, which resulted in a baseline profile correlation of .28.
These baseline correlations may at first seem somewhat high, but when one
realizes that all participants in the sample were college students and that,
in general, college students are similar to each other in many ways, it
makes sense that even random pairings resulted in profile correlations of
moderate magnitude. Profile correlations greater than these baseline cor-
relations indicate differential accuracy (Blackman & Funder, 1998).
Contrast analysis. Contrast analysis was used to test focused questions
by assigning weights to experimental conditions that reflect a prediction
concerning the structure of the data (Rosenthal, Rosnow, & Rubin, 2000).
Once contrast weights were assigned to each condition, it was possible to
determine the extent to which the data were in line with the prediction.
When a t test was computed for a contrast analysis,9 the degrees of freedom
were the number of participants minus the number of groups.10 One-tailed
p values that are associated with the tcontrast are reported. It was also
possible to determine how much variance in the group means was ac-
counted for by the prediction by correlating the group means with the
contrast weights and squaring the obtained correlation, which resulted in an
r2alerting. A large r2alerting provides support for the prediction. Furthermore,
it was possible to test whether the leftover variance, or the variance
not accounted for by the prediction, was significant by computing a
tnoncontrast.11 A nonsignificant tnoncontrast also lends support to the a priori
prediction because little variance is left over that is not accounted for by the
prediction.
On the basis of RAM, we predicted that there would be linear relation-
ships among information quantity and quality and realistic accuracy, con-
sensus, and self– other agreement, for which the appropriate contrast
weights are ⫺1 (low information), 0 (medium information), and ⫹1 (high
information). On the basis of WAM, we predicted a slightly different
relationship between information quantity and quality and consensus,
which reflects an increase at low levels of acquaintance and then a leveling
off at higher levels of acquaintance, for which the appropriate contrast
weights are ⫺2 (low information), ⫹1 (medium information), and ⫹1
(high information).
Post hoc t tests. Contrast analyses are informative regarding how well
the obtained data fit an a priori prediction about how the data should be
ordered, but they do not provide information regarding which pairs of
groups are reliably different from each other. Therefore, post hoc
independent-samples t tests were used to determine which pairs of exper-
imental conditions had reliably different levels of realistic accuracy, con-
sensus, and self– other agreement. No adjustments were made to the p
values associated with these tests, so the conservative reader may wish to
adjust the p values to account for the number of comparisons in each
analysis.
Results
The focus of our analyses was on realistic accuracy, as this
project was one of the first to explore the construct and the first, to
our knowledge, to examine the relationships between realistic
accuracy and experimentally manipulated information quantity
and quality. We also provided evidence concerning our predictions
for consensus and self– other agreement. Before examining the
effects of information quantity and quality on judgment accuracy,
we first determined whether participants in longer and higher
quality interactions did indeed acquire more information (on the
basis of a sample of information possibly available in such inter-
actions), as a manipulation check. Then, we used contrast analysis
(Rosenthal et al., 2000) to examine how well the changes in
accuracy across experimental conditions fit the predictions based
on RAM for realistic accuracy, consensus, and self– other agree-
ment and based on WAM for consensus only. Finally, the degree
to which information acquisition was related to realistic accuracy,
consensus, and self– other agreement, regardless of the type of
group the participants interacted in, was examined.
There are several possible ways to analyze these data. We used
traditional statistical procedures to examine differences at the
group level by conducting a between-groups analysis to determine
how well the obtained data fit the predictions. Our predictions
were concerned with properties of the context, not of the individ-
ual, and a group-level analysis reflects this orientation. Although
none of the basic conclusions of this study change when data are
examined at different levels of analysis, an advantage of analyzing
data at the level of the group is that it allowed us to perform all
analyses in a consistent manner (at the same level of analysis).
Information Acquisition
To determine whether participants in interactions of greater
quantity and quality acquired more information about each other,
9
The formula used to compute tcontrast was
冑 2
ralerting ⫻ SSbetween
MSwithin
.
The formula used to compute rcontrast was
冑 t2
2
t
contrast
contrast⫹ df
.
10
In the current analyses, there were 12 participants in each of the 3
groups, so there were [(12)(3) – 3] ⫽ 33 degrees of freedom for each
contrast analysis.
11
tnoncontrast ⫽ 冑SSbetween ⫺ SScontrast
MSwithin
.
 INFORMATION AND PERSONALITY JUDGMENT
we compared the scores achieved on the information fact sheet
across experimental conditions. This can be considered a manip-
ulation check that determines the degree to which participants in
longer and higher quality interactions actually learned more about
each other.
Information quantity. A good fit was found between amount
of acquired information and a linear increase across the three
levels of information quantity: minimal information (M ⫽ 2.80,
SD ⫽ 1.10); short unstructured (M ⫽ 5.17, SD ⫽ 1.09); long
unstructured (M ⫽ 7.00, SD ⫽ 1.58); and rcontrast ⫽ .81,
tcontrast(33) ⫽ 8.02, p ⬍ . 0001 (see Figure 2). Furthermore, on the
basis of independent-samples t tests, we found that the scores of all
groups differed significantly from each other: minimal information
and short unstructured, t(22) ⫽ 5.29, p ⬍ .0001; minimal infor-
mation and long unstructured, t(22) ⫽ 7.54, p ⬍ .0001; and short
unstructured and long unstructured, t(22) ⫽ 3.29, p ⫽ .003.
Information quality. A good fit was also found between the
amount of acquired information and a linear increase across the
three levels of information quality: trivia (M ⫽ 1.65, SD ⫽ 1.04);
short unstructured (M ⫽ 5.17, SD ⫽ 1.09); GTK (M ⫽ 5.87, SD ⫽
1.57); and rcontrast ⫽ .82, tcontrast(33) ⫽ 8.22, p ⬍ .0001 (see Figure
3). Participants in both the short unstructured and GTK conditions
acquired reliably more information than did participants in the
trivia condition, t(22) ⫽ 8.08, p ⬍ .0001, and t(22) ⫽ 7.75, p ⬍
.0001, respectively, although information acquisition did not differ
reliably between the short unstructured and GTK conditions,
t(22) ⫽ 1.26, p ⫽ .22.
These findings concerning a small sample of the available
information imply that participants in longer and higher quality
interactions acquired more information about each other. The next
analyses addressed the question of whether personality judgments
were also more accurate in the longer and higher quality
interactions.
Personality Judgment
Information quantity. To test the prediction that judges would
be more accurate after interacting for longer periods of time, we
examined personality judgments at three levels of information
quantity: low (minimal information–no interaction), medium
(short unstructured–50-min interaction), and high (long
unstructured–3-hr interaction). First, profile correlations were
computed for each group, as described in the Analyses section. The
Figure 2. Mean information acquisition scores for experimental condi-
tions used in the information quantity analyses.
119
Figure 3. Mean information acquisition scores for experimental condi-
tions used in the information quality analyses. GTK ⫽ get to know
condition.
profile correlations were transformed using Fisher’s r-to-z trans-
formation, and then contrast analyses (Rosenthal et al., 2000) were
used to determine the degree to which the profile correlations
followed the predictions based on RAM and WAM.
As predicted, we found a good fit between realistic accuracy
scores and the prediction of a positive linear relationship with
information quantity: minimal information (mean r ⫽ .29, SD ⫽
0.20); short unstructured (mean r ⫽ .45, SD ⫽ 0.14); long un-
structured (mean r ⫽ .47, SD ⫽ 0.09); and rcontrast ⫽ .53,
tcontrast(33) ⫽ 3.56, p ⫽ .000058 (see Figure 4). The linear pre-
diction accounted for 86% of the variance in the group means, and
the variance not accounted for by the prediction did not reach the
.05 level of significance, tnoncontrast(33) ⫽ 1.41, p ⫽ .08. When the
groups were compared with each other, realistic accuracy was
reliably higher in both the short unstructured and long unstructured
conditions than in the minimal information condition, t(22) ⫽
2.57, p ⫽ .02, and t(22) ⫽ 3.46, p ⫽ .002, respectively, but
accuracy in the long unstructured condition was not reliably higher
than accuracy in the short unstructured condition, t(22) ⫽ 0.74,
p ⫽ .47.
Information quantity also had a good fit to the prediction of a
positive linear relationship with consensus: minimal information
(mean r ⫽ .25, SD ⫽ 0.18); short unstructured (mean r ⫽ .41,
SD ⫽ 0.20); long unstructured (mean r ⫽ .44, SD ⫽ 0.17); and
rcontrast ⫽ .44, tcontrast(33) ⫽ 2.83, p ⫽ .004. The linear prediction
accounted for 86% of the variance in the group means, and the
Figure 4. Mean realistic accuracy scores for experimental conditions
used in the information quantity analyses. Note that the random baseline
profile correlation was .36.
120 LETZRING, WELLS, AND FUNDER
variance not accounted for by the prediction was not significant,
tnoncontrast(33) ⫽ 1.14, p ⫽ .13. Judges in the short unstructured
and long unstructured conditions had reliably higher consensus
than judges in the minimal information condition, t(22) ⫽ 2.33,
p ⫽ .03, and t(22) ⫽ 2.98, p ⫽ .007, respectively, although
consensus did not differ for judges in the short unstructured and
long unstructured conditions, t(22) ⫽ 0.42, p ⫽ .68.
The consensus data were also in line with the prediction based
on WAM, in which an increase was predicted between the low-
and medium-quantity conditions but not between the medium- and
high-quantity conditions, rcontrast ⫽ .48, tcontrast(33) ⫽ 3.04, p ⫽
.0023. The nonlinear prediction accounted for 98% of the variance
in the group means, and the variance not accounted for by the
prediction was not significant, tnoncontrast(33) ⫽ 1.18, p ⫽ .12.
Finally, information quantity had a good fit to the prediction of
a positive linear increase with self– other agreement: minimal
information (mean r ⫽ .21, SD ⫽ 0.19); short unstructured (mean
r ⫽ .36, SD ⫽ 0.12); long unstructured (mean r ⫽ .39, SD ⫽ 0.12);
rcontrast ⫽ .49, tcontrast(33) ⫽ 3.24, p ⫽ .0014. The linear prediction
accounted for 86% of the variance in the group means, and the
variance not accounted for by the prediction was not significant,
tnoncontrast(33) ⫽ 1.29, p ⫽ .10. Judges in the short unstructured
and long unstructured conditions had reliably higher self– other
agreement than judges in the minimal information condition,
t(22) ⫽ 2.58, p ⫽ .02, and t(22) ⫽ 3.03, p ⫽ .006, respectively,
although self– other agreement did not differ for judges in the short
unstructured and long unstructured conditions, t(22) ⫽ 0.72, p ⫽
.55.
Note that for all indicators of accuracy, the mean profile corre-
lations for the minimal information condition were approximately
equal to or smaller than the random baseline correlations of .36 for
realistic accuracy, t(11) ⫽ 0.31, p ⫽ .31; .25 for consensus,
t(11) ⫽ 0.11, p ⫽ .91; and .28 for self– other agreement, t(11) ⫽
1.29, p ⫽ .22; but the mean profile correlations were all higher
than the random baseline correlations for the short unstructured
condition: realistic accuracy, t(11) ⫽ 2.94, p ⫽ .01; consensus,
t(11) ⫽ 3.22, p ⫽ .008; and self– other agreement, t(11) ⫽ 2.76,
p ⫽ .02; and for the long unstructured condition: realistic accu-
racy, t(11) ⫽ 6.31, p ⬍ .0001; consensus, t(11) ⫽ 4.44, p ⫽ .001;
and self– other agreement, t(11) ⫽ 3.53, p ⫽ .005. This pattern
suggests that even though realistic accuracy, consensus, and self–
other agreement can reach moderate strengths without any inter-
action, this strength is at or below chance levels. It is only when
people are allowed to interact that differential realistic accuracy,
consensus, and self– other agreement are achieved.
Information quality. Recall that information quality refers to
the relevance of information to personality. To test our prediction
that judgments would be more accurate when participants inter-
acted in higher quality situations, we had participants interact in
three experimental conditions in which information quantity was
held constant while information quality was manipulated. We
again used contrast weights reflecting the prediction of a linear
relationship between information quality and realistic accuracy,
consensus, and self– other agreement to test the prediction based
on RAM and a nonlinear relationship to test the prediction for
consensus based on WAM.
As predicted, we found that realistic accuracy had a good fit to
the prediction of a positive linear relationship with information
quality: trivia (mean r ⫽ .36, SD ⫽ 0.12); short unstructured
(mean r ⫽ .45, SD ⫽ 0.12); GTK (mean r ⫽ .49, SD ⫽ 0.16); and
rcontrast ⫽ .43, tcontrast(33) ⫽ 2.71, p ⫽ .0053 (see Figure 5). The
linear prediction accounted for 97% of the variance in the group
means, and the variance not accounted for by the prediction was
not significant, tnoncontrast(33) ⫽ 0.50, p ⫽ .31. Realistic accuracy
was reliably higher in the GTK condition than in the trivia condi-
tion, t(22) ⫽ 2.76, p ⫽ .01, and the difference between the trivia
and short unstructured conditions approached the conventional
level of significance, t(22) ⫽ 1.94, p ⫽ .07. However, the GTK
condition did not differ reliably from the short unstructured con-
dition, t(22) ⫽ 0.88, p ⫽ .39.
A good fit to the prediction of a linear increase was also found
with consensus: trivia (mean r ⫽ .29, SD ⫽ 0.17); short unstruc-
tured (mean r ⫽ .41, SD ⫽ 0.20); GTK (mean r ⫽ .41, SD ⫽
0.10); and rcontrast ⫽ .33, tcontrast(33) ⫽ 1.99, p ⫽ .028. The linear
prediction accounted for 74% of the variance in the group means,
and the variance not accounted for by the prediction was not
significant, tnoncontrast(33) ⫽ 1.19, p ⫽ .12. Judges in the GTK
condition had reliably higher consensus than judges in the trivia
condition, t(22) ⫽ 2.34, p ⫽ .03; the difference between consensus
in the short unstructured and trivia conditions approached signif-
icance, t(22) ⫽ 1.78, p ⫽ .09; and consensus did not differ for
judges in the short unstructured and GTK conditions, t(22) ⫽ 0.04,
p ⫽ .97.
The consensus data were also in line with the prediction based
on WAM, rcontrast ⫽ .37, tcontrast(33) ⫽ 2.31, p ⫽ .014. The
nonlinear prediction accounted for 99% of the variance in the
group means, and the variance not accounted for by the prediction
was not significant, tnoncontrast(33) ⫽ 0.001, p ⫽ .50.
Finally, a good fit to the prediction of a linear increase was
found with self– other agreement: trivia (mean r ⫽ .27, SD ⫽
0.09); short unstructured (mean r ⫽ .36, SD ⫽ 0.12); GTK (mean
r ⫽ .38, SD ⫽ 0.16); rcontrast ⫽ .39, tcontrast(33) ⫽ 2.40, p ⫽ .011.
The linear prediction accounted for 91% of the variance in the
group means, and the variance not accounted for by the prediction
was not significant, tnoncontrast ⫽ 0.75, p ⫽ .23. Judges in the short
unstructured and GTK conditions had reliably higher self– other
agreement than judges in the trivia condition, t(22) ⫽ 2.16, p ⫽
.04, and t(22) ⫽ 2.33, p ⫽ .03, respectively, although self– other
agreement did not differ for judges in the short unstructured and
GTK conditions, t(22) ⫽ 0.49, p ⫽ .63.
Note that again, the mean profile correlations for the trivia
condition were approximately equal to or smaller than the random
baseline correlations of .36 for realistic accuracy, t(11) ⫽ 0.48,
Figure 5. Mean realistic accuracy scores for experimental conditions
used in the information quality analyses. Note that the random baseline
profile correlation was .36. GTK ⫽ get to know condition.
 INFORMATION AND PERSONALITY JUDGMENT
p ⫽ .64; .25 for consensus, t(11) ⫽ 1.09, p ⫽ .30; and .28 for
self– other agreement, t(11) ⫽ 0.03, p ⫽ .98, but the mean profile
correlations were all higher than the random baseline correlations
for the short unstructured condition: realistic accuracy, t(11) ⫽
2.94, p ⫽ .01; consensus, t(11) ⫽ 3.22, p ⫽ .008; and self– other
agreement, t(11) ⫽ 2.76, p ⫽ .02; and for the GTK condition:
realistic accuracy, t(11) ⫽ 3.84, p ⫽ .003; consensus, t(11) ⫽ 6.45,
p ⬍ .0001; and self– other agreement, t(11) ⫽ 2.73, p ⫽ .02. This
pattern suggests that not just any context of interaction or obser-
vation will afford differential judgment but that differential real-
istic accuracy, consensus, and self– other agreement are only
achieved in the conditions that allow for the exchange of
moderate- to high-quality information.
Relation Between Information Acquisition and Accuracy
We could also examine the degree to which information acqui-
sition is related to realistic accuracy, self– other agreement, and
consensus across all groups. RAM predicts that people who ac-
quire more information that is relevant to personality will make
more accurate judgments of personality, and it is reasonable to
assume that participants who acquired more of the sample of
information asked about on the information fact sheet also ac-
quired more personality-relevant information, broadly speaking. If
this is true, then there should be a positive relationship between
scores on the information fact sheet and realistic accuracy, self–
other agreement, and consensus, regardless of the type of experi-
mental condition in which the participants interacted.12 This pre-
diction can be tested by correlating the average information fact
sheet score of each group with the average realistic accuracy,
consensus, and self– other agreement scores of each group across
all five experimental conditions. As predicted, moderate positive
relationships were found between information acquisition and re-
alistic accuracy (r ⫽ .31, p ⫽ .02), self– other agreement (r ⫽ .30,
p ⫽ .02), and consensus (r ⫽ .30, p ⫽ .02), indicating that
regardless of information quantity or quality, groups that acquired
more of the objective information that was sampled with the
information fact sheet also judged their partners with higher aver-
age levels of accuracy, consensus, and self– other agreement.
Discussion
The preceding analyses provide tests of the relationships be-
tween two aspects of good information and the accuracy of per-
sonality judgment. Experimentally manipulated information quan-
tity, or length of acquaintance, was positively related to realistic
accuracy, consensus, and self– other agreement. Additionally, in-
formation quality was manipulated while information quantity was
held constant, and support was found for the prediction that real-
istic accuracy, consensus, and self– other agreement are higher in
situations in which personality-relevant information is more likely
to be available.
Information Quantity
RAM predicts that accuracy will increase in relation to infor-
mation quantity (Funder, 1995), but previous research has only
tested this prediction using self– other agreement, which is a lim-
ited definition of accuracy. The current research confirms past
121
findings that self– other agreement increases with acquaintance
(Bernieri et al., 1994; Blackman & Funder, 1998; Funder et al.,
1995; Paulhus & Bruce, 1992; Paunonen, 1989) and provides one
more piece of evidence in favor of the existence of a positive
relationship between level of acquaintance and self– other
agreement.
In addition to confirming past findings, the current research
takes an additional step in the examination of the acquaintanceship
effect by using a broad-based accuracy criterion based on the
ratings of several people who knew the target well. Such a criterion
allows for a more precise test of the acquaintanceship effect than
was possible in past work. In line with our prediction, realistic
accuracy was positively related to information quantity.
Both RAM and WAM can be used to make predictions con-
cerning consensus. RAM predicts that realistic accuracy will in-
crease across all levels of acquaintance, and because people who
are highly accurate about something real must also be in agree-
ment, RAM predicts that consensus will also increase with ac-
quaintance (Funder, 1995). WAM is specifically aimed at explain-
ing consensus, and our prediction was that consensus would
increase at low levels of acquaintance and then stay about the same
even as acquaintance continued to increase (Kenny, 1991). These
predictions are fairly similar, and we found support for both
predictions. It is not possible to determine if one prediction is
superior to the other, for a couple of reasons. First, the predictions
themselves are highly similar, and so any data should fit both
predictions to about the same degree. Second, it was not possible
to determine whether the experimentally manipulated difference
between the low- and medium-information quantity conditions
was equal to the difference between the medium and high condi-
tions and, therefore, whether a linear or nonlinear prediction was
more appropriate. However, it is obvious from the means of the
conditions that the larger increase was between the low and me-
dium groups, but it is not possible to determine whether the
findings support the prediction based on WAM better than the
prediction based on RAM.
Overall, these findings support the existence of the acquain-
tanceship effect and the role of information quantity in accurate
judgment. This interpretation was bolstered by the additional find-
ing that participants learned more facts about each other in the
longer interactions and therefore had a larger quantity of informa-
tion available to them when making judgments of personality.
Information Quality
To our knowledge, the current article provides the first direct
evidence of the positive relationships between information quality
and realistic accuracy, consensus, and self– other agreement. In-
formation quality was manipulated by varying situational strength
and instructions that provided an objective for the interaction. For
12
Some astute readers may wonder why we did not perform a media-
tional analysis with amount of information acquired as a mediator of the
relations of information quantity and quality with accuracy. We did not
compute these analyses because the information fact sheet was designed to
be a manipulation check and is only a sample of some relatively objective
information that is likely to become available during an initial interaction,
and it is not a comprehensive measure of the amount of personality-
relevant information acquired.
122 LETZRING, WELLS, AND FUNDER
the low-quality interaction, participants interacted in a strong sit-
uation with specific instructions to complete a packet of trivia-type
questions. We expected participants in this condition to exhibit
little behavioral variation and to not share information about
thoughts and feelings, and therefore only limited amounts of
personality-relevant information would become available. For the
medium- and high-quality interactions, participants interacted in
relatively unstructured situations that allowed for a good deal of
behavioral variation and larger amounts of personality-relevant
information to become available. The difference in these two
conditions was the instructions informing the participants about
the objective of the interaction. Participants in the medium-quality
interaction were not provided with an objective, whereas partici-
pants in the high-quality interaction were given the specific ob-
jective of getting to know each other as well as possible. Partici-
pants in this latter condition were expected to reveal the most
personality-relevant information and to pay the best attention to
the interaction partners in an attempt to get to know them well,
which should have increased accuracy by affecting the availability
and detection stages of RAM. Results confirmed these expecta-
tions, with realistic accuracy, consensus, and self– other agreement
all increasing, along with factual knowledge, across the three
conditions of information quality. However, the present data do
not allow us to determine whether these effects should be attrib-
uted to what cognitive psychologists would call the encoding stage
or the decoding stage. It may be that more relevant information
was available to be detected (encoding), that the detected infor-
mation was remembered and utilized better (decoding), or both.
As noted in the Results section, even though the data showed a
good fit to a linear prediction, most of the increase in the indicators
of accuracy was between the low- and medium-quality conditions,
and the medium- and high-quality conditions were quite similar.
One explanation for this finding is that people naturally try to get
to know each other in an initial interaction, so even though we
explicitly instructed participants in the high-quality condition to
get to know each other and simply told participants in the medium-
quality condition to talk about whatever they liked, they may have
had similar interactions in which they learned approximately the
same amounts of personality-relevant information that could be
used when making judgments of personality. If this is the case,
then information quality may not have been very different between
our medium- and high-quality conditions, which would explain
why realistic accuracy, consensus, and self– other agreement were
also not very different in these conditions.
Limitations and Future Directions
As in any study, our experimental design has limitations that
may affect interpretation. First, the construct of realistic accuracy
is a philosophical ideal, and any measurement of it must neces-
sarily be an approximation. We attempted to approach realistic
accuracy by having several knowledgeable people, including ex-
pert judges, describe the personalities of the target participants.
This procedure was not likely to tell us exactly what the target is
like, but it was likely to bring us closer to reality than would
ratings provided by any single informant, even the self. In future
research, we would hope to see criteria for realistic accuracy
expanded even further to include direct behavioral measurements,
important life outcomes, and perhaps even biological markers.
The present results involving realistic accuracy, consensus, and
self– other agreement were in general quite similar to each other.
This finding could be interpreted as evidence that realistic accu-
racy cannot be meaningfully distinguished from the simpler and
more traditional operationalizations or that constructing multifac-
eted criteria is not worth the time and effort. We caution against
both conclusions. First, realistic accuracy is a hypothetical con-
struct that is in principle quite different from any specific criterion
for or measurement of it, and on theoretical grounds it is important
to be clear about the distinction. Second, it is easy to imagine
circumstances in which self– other agreement, consensus, and re-
alistic accuracy would diverge (such as when an absentminded job
applicant tries to appear conscientious), and at our present state of
knowledge it is wise to be alert to this possibility and to develop
tools to detect such differentiation between constructs when it
happens.
Even though the contrast analyses suggest that there are mod-
erate relationships between information quantity and quality and
indicators of accuracy, this effect was largely driven by differences
between the low conditions (minimal information and trivia) and
the medium and high conditions (short unstructured, GTK, and
long unstructured), with the medium and high conditions being
only slightly different, albeit consistently in the predicted direc-
tion. The pattern of findings would have been even more convinc-
ing had there been a larger difference in realistic accuracy between
the medium- and high-level conditions. We propose that the
present evidence supports the role of quantity and quality in
accuracy but that our highest conditions may not have increased
acquaintance enough or provided enough motivation to show a
substantial increase in indicators of accuracy beyond the medium
levels. Future research should experiment further with techniques
for providing participants large amounts of information or infor-
mation that is highly relevant to personality.
Another reasonable next step would be to increase the external
validity of our conclusions through a project in which interactions
take place in real social contexts (vs. in the lab, as was the case in
the current experiment). There are many situations in which pre-
viously unacquainted groups of people interact and get to know
each other over a period of time, such as 1st-year freshmen in
college dorms, participants in volunteer corps who live and/or
work together (i.e., Peace Corps, Jesuit Volunteer Corps), and
older adults who move into retirement communities. Information
quantity could be examined in such situations by obtaining judg-
ments about others in the judge’s living community at several
stages of acquaintance. Information quality could be examined by
obtaining information about the quality of interactions from both
the target and the judge, perhaps by asking for reports of types of
topics most often discussed and amount of time spent talking
versus engaging in other activities. Such a project is likely to be
informative about how information quantity and quality are related
to realistic accuracy in situations and contexts outside of the
laboratory, as well as to how these two aspects of information are
related to each other and personality judgment.
A third direction for future research is to determine the actual
behavioral events during an interaction that are related to the achieved
levels of realistic accuracy, self– other agreement, and consensus.
Behavioral coding of the experimental interactions might increase our
understanding of why information quantity and quality are related to
accuracy, self– other agreement, and consensus.
 INFORMATION AND PERSONALITY JUDGMENT
Conclusion
The current findings support several aspects of RAM, including
the implications of information for the relevance, availability, and
detection stages that make accurate judgment possible, along with
specific predictions of WAM concerning consensus. Our main
findings are that judgments of personality are more likely to
achieve higher levels of realistic accuracy, consensus, and self–
other agreement when judges have interacted with targets for
longer periods of time or in situations that allow for or encourage
people to reveal personality-relevant information.
References
Allport, G. W. (1937). The ability to judge people. In G. W. Allport,
Personality: A psychological interpretation (pp. 499 –522). New York:
Holt.
Altman, I., & Taylor, D. A. (1973). Social penetration: The development of
interpersonal relationships. New York: Holt, Rinehart & Winston.
Ambady, N., Hallahan, M., & Rosenthal, R. (1995). On judging and being
judged accurately in zero-acquaintance situations. Journal of Personal-
ity and Social Psychology, 69, 518 –529.
Anderson, S. M. (1984). Self-knowledge and social inference: II. The
diagnosticity of cognitive/affective and behavioral data. Journal of Per-
sonality and Social Psychology, 46, 294 –307.
Bem, D. J., & Funder, D. C. (1978). Predicting more of the people more of
the time: Assessing the personality of situations. Psychological Review,
85, 485–501.
Bernieri, F. J., Zuckerman, M., Koestner, R., & Rosenthal, R. (1994).
Measuring person perception accuracy: Another look at self– other
agreement. Personality and Social Psychology Bulletin, 20, 367–378.
Blackman, M. C., & Funder, D. C. (1998). The effect of information on
consensus and accuracy in personality judgment. Journal of Experimen-
tal Social Psychology, 34, 164 –181.
Block, J. (1961). The Q-sort method in personality assessment and psy-
chiatric research. Palo Alto, CA: Consulting Psychologists Press.
Borkenau, P., & Liebler, A. (1993). Consensus and self– other agreement
for trait inferences from minimal information. Journal of Personality,
61, 477– 496.
Borkenau, P., Mauer, N., Riemann, R., Spinath, F. M., & Angleitner, A.
(2004). Thin slices of behavior as cues of personality and intelligence.
Journal of Personality and Social Psychology, 86, 599 – 614.
Colvin, R. C. (1993). “Judgeable” people: Personality, behavior, and
competing explanations. Journal of Personality and Social Psychology,
64, 861– 873.
Colvin, R. C., & Funder, D. C. (1991). Predicting personality and behavior:
A boundary on the acquaintanceship effect. Journal of Personality and
Social Psychology, 60, 884 – 894.
Craik, K. H., Ware, A. P., Kamp, J., O’Reilly, C., III, Staw, B., & Zedeck,
S. (2002). Explorations of construct validity in a combined managerial
and personality assessment programme. Journal of Occupational and
Organizational Psychology, 75, 171–193.
Cronbach, L. J. (1955). Processes affecting scores on “understanding of
others” and “assumed similarity.” Psychological Bulletin, 52, 177–193.
Funder, D. C. (1980). On seeing ourselves as others see us: Self– other
agreement and discrepancy in personality ratings. Journal of Personal-
ity, 48, 473– 493.
Funder, D. C. (1987). Errors and mistakes: Evaluating the accuracy of
social judgment. Psychological Bulletin, 101, 75–90.
Funder, D. C. (1995). On the accuracy of personality judgment: A realistic
approach. Psychological Review, 102, 652– 670.
Funder, D. C. (1999). Personality judgment: A realistic approach to person
perception. San Diego, CA: Academic Press.
123
Funder, D. C., & Colvin, C. R. (1988). Friends and strangers: Acquain-
tanceship, agreement, and the accuracy of personality judgment. Journal
of Personality and Social Psychology, 55, 149 –158.
Funder, D. C., & Dobroth, K. M. (1987). Differences between traits:
Properties associated with interjudge agreement. Journal of Personality
and Social Psychology, 52, 409 – 418.
Funder, D. C., Kolar, D. C., & Blackman, M. C. (1995). Agreement among
judges of personality: Interpersonal relations, similarity, and acquain-
tanceship. Journal of Personality and Social Psychology, 69, 656 – 672.
Hofstee, W. K. B. (1994). Who should own the definition of personality?
European Journal of Personality, 8, 149 –162.
John, O. P., & Robins, R. W. (1993). Determinants of interjudge agreement
on personality traits: The Big Five domains, observability, evaluative-
ness, and the unique perspective of the self. Journal of Personality, 41,
521–551.
Kenny, D. A. (1991). A general model of consensus and accuracy in
interpersonal perception. Psychological Review, 98, 155–163.
Kenny, D. A. (1994). Interpersonal perception. New York: Guilford Press.
Kenny, D. A., Albright, L., Malloy, T. E., & Kashy, D. A. (1994).
Consensus in interpersonal perception: Acquaintance and the Big Five.
Psychological Bulletin, 116, 245–258.
Kolar, D. W. (1995). Individual differences in the ability to accurately
judge the personality characteristics of others. Unpublished doctoral
dissertation, University of California, Riverside.
Kolar, D. W., Funder, D. C., & Colvin, C. R. (1996). Comparing the
accuracy of personality judgments by the self and knowledgeable others.
Journal of Personality, 64, 311–337.
Letzring, T. D., Block, J., & Funder, D. C. (2005). Ego-control and
ego-resiliency: Generalization of self-report scales based on personality
descriptions from acquaintances, clinicians, and the self. Journal of
Research in Personality, 39, 395– 422.
Norman, W. T., & Goldberg, L. R. (1966). Raters, ratees, and randomness
in personality structure. Journal of Personality and Social Psychology,
4, 681– 691.
Park, B., & Judd, J. M. (1989). Agreement on initial impressions: Differ-
ences due to perceivers, trait dimensions, and target behaviors. Journal
of Personality and Social Psychology, 56, 493–505.
Park, B., Kraus, S., & Ryan, C. S. (1997). Longitudinal changes in
consensus as a function of acquaintance and agreement in liking. Journal
of Personality and Social Psychology, 72, 604 – 616.
Paulhus, D. L., & Bruce, M. N. (1992). The effect of acquaintanceship on
the validity of personality impressions: A longitudinal study. Journal of
Personality and Social Psychology, 63, 816 – 824.
Paunonen, S. V. (1989). Consensus in personality judgment: Moderating
effects of target–rater acquaintanceship and behavior observability.
Journal of Personality and Social Psychology, 56, 823– 833.
Rosenthal, R., Rosnow, R. L., & Rubin, D. B. (2000). Contrasts and effect
sizes in behavioral research: A correlational approach. New York:
Cambridge University Press.
Snyder, M., & Ickes, W. (1985). Personality and social behavior. In G.
Lindzey & E. Aronson (Eds.), Handbook of social psychology (3rd ed.,
Vol. 2, pp. 883–947). New York: Random House.
Taft, R. (1955). The ability to judge people. Psychological Bulletin, 52,
1–23.
Vernon, P. E. (1933). Some characteristics of the good judge of personal-
ity. Journal of Social Psychology, 4, 42–57.
Vogt, D. S., & Colvin, C. R. (2003). Interpersonal orientation and the
accuracy of personality judgments. Journal of Personality, 71, 267–295.
Received August 26, 2004
Revision received July 22, 2005
Accepted December 14, 2005 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 124 –142 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.124

Supplication and Appeasement in Conflict and Negotiation:

The Interpersonal Effects of Disappointment, Worry, Guilt, and Regret

Gerben A. Van Kleef and Carsten K. W. De Dreu Antony S. R. Manstead

University of Amsterdam Cardiff University

This study examined the social effects of emotions related to supplication and appeasement in conflict
and negotiation. In a computer-simulated negotiation, participants in Experiment 1 were confronted with
a disappointed or worried opponent (supplication), with a guilty or regretful opponent (appeasement), or
with a nonemotional opponent (control). Compared with controls, participants conceded more when the
other experienced supplication emotions and conceded less when the other experienced appeasement
emotions (especially guilt). Experiment 2 replicated the effects of disappointment and guilt and showed
that they are moderated by the perceiver’s dispositional trust: Negotiators high in trust conceded more to
a disappointed counterpart than to a happy one, but those with low trust were unaffected. In Experiment
3, trust was manipulated through information about the other’s personality (cooperative vs. competitive),
and a similar moderation was obtained.

Keywords: conflict, negotiation, emotion, supplication, appeasement

Social interactions can produce conflict at all levels of society.
One of the most common and constructive ways of resolving such
conflicts and conducting social and economic exchange is by
means of negotiation. Negotiation can be defined as a discussion
between two or more parties aimed at resolving a perceived
divergence of interests (Pruitt & Carnevale, 1993). People may
negotiate with a car dealer when buying a new car, work groups
may negotiate the allocation of organizational resources, and par-
ents may negotiate with their children about how to spend the
holidays. Emotions are inherent to negotiation and social conflict
(Davidson & Greenhalgh, 1999) and are crucial to understanding
how individuals behave within bargaining situations (Barry, 1999).
So far, empirical research on emotion in conflict and negotiation
has focused almost exclusively on the effects of anger and happi-
ness. In this article, we focus on the social effects of emotions
related to supplication (e.g., disappointment, worry) and appease-
ment (e.g., guilt, regret) in negotiation, examining the ways in
which negotiators respond to their opponent’s emotions.
Emotions in Conflict and Negotiation
There are multiple definitions of emotion, most of which point
to three distinct features of emotion: physiological reactions, ac-
tion tendencies, and subjective experience (Lazarus, 1991). Emo-
tions differ from moods in that they are discrete (Russell &
Feldman Barrett, 1999), of relatively high intensity and short
Gerben A. Van Kleef and Carsten K. W. De Dreu, Department of
Psychology, University of Amsterdam, Amsterdam, the Netherlands;
Antony S. R. Manstead, School of Psychology, Cardiff University, Cardiff,
United Kingdom.
Correspondence concerning this article should be addressed to Gerben
A. Van Kleef, Department of Psychology, University of Amsterdam,
Roetersstraat 15, 1018 WB Amsterdam, the Netherlands. E-mail:
g.a.vankleef@uva.nl
124
duration (Barry, 1999; Oatley & Jenkins, 1996), and intentional—
directed at an object, person, or event (Frijda, 1993). In this article,
we use the term emotion in the sense intended above, whereas
affect is used as a superordinate construct that encompasses both
moods and emotions (Barry & Oliver, 1996).
Research on emotion in conflict and negotiation can be roughly
divided into two categories: studies of intrapersonal effects and
studies of interpersonal effects. In the 1980s and 1990s, research-
ers focused mostly on the intrapersonal effects of moods and
emotions, investigating the influence of a negotiator’s emotional
state on his or her own cognitions and behavior. For example,
positive affect has been shown to increase concession making
(Baron, 1990), stimulate creative problem solving (Isen, Daubman,
& Nowicki, 1987), increase joint gains (Allred, Mallozzi, Matsui,
& Raia, 1997; Carnevale & Isen, 1986), increase preferences for
cooperation (Baron, Fortin, Frei, Hauver, & Shack, 1990), reduce
the use of contentious tactics (Carnevale & Isen, 1986), and
increase the use of cooperative negotiation strategies (Forgas,
1998). Conversely, negative affect has been shown to decrease
initial offers (Baron et al., 1990), decrease joint gains (Allred et al.,
1997), promote the rejection of ultimatum offers (Pillutla & Mur-
nighan, 1996), increase the use of competitive strategies (Forgas,
1998), and decrease the desire to work together in the future
(Allred et al., 1997).
Recently, scientific interest in the role of affect in conflict and
negotiation has shifted away from the intrapersonal effects of
moods and emotions. Recognizing that negotiation is a social
phenomenon—negotiators’ emotions influence not only them-
selves, but also their counterparts—several scholars have empha-
sized the importance of the interpersonal or social effects of
emotions in negotiations (Adler, Rosen, & Silverstein, 1998;
Barry, Fulmer, & Van Kleef, 2004; Davidson & Greenhalgh, 1999;
Morris & Keltner, 2000; Thompson, Medvec, Seiden, & Kopel-
man, 2001; Van Kleef, De Dreu, & Manstead, 2004a, b). The basic
premise is that emotions have important social functions and
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
consequences (Frijda & Mesquita, 1994; Keltner & Haidt, 1999;
Oatley & Jenkins, 1992; Parkinson, 1996). Most notable is that
emotions convey information (Carver & Scheier, 1990), for in-
stance about how one feels about things (Ekman, 1993), about
one’s social intentions (Fridlund, 1994), and about one’s orienta-
tion toward other people (Knutson, 1996). In this way, one’s
emotions may influence not only one’s own behavior, but also the
behavior of others (Levenson, 1994).
In line with this social functions perspective, recent research has
demonstrated the pervasive interpersonal effects of anger and
happiness in negotiations. In a computer-mediated negotiation task
with a simulated opponent, Van Kleef et al. (2004a) provided
participants with information about the opponent’s emotional state
(angry, happy, or no emotion) at three time points during the
negotiation. They found that participants with an angry opponent
placed lower demands and made larger concessions than did
participants with a nonemotional opponent, whereas participants
with a happy opponent placed higher demands and made smaller
concessions. Sinaceur and Tiedens (in press) examined the effects
of anger and happiness using a different paradigm and obtained
similar results. In face-to-face negotiations, they instructed one
negotiator in each dyad to show either anger or happiness. In
keeping with the results obtained by Van Kleef et al. (2004a),
Sinaceur and Tiedens found that participants conceded more to an
angry than to a happy counterpart. So results from research using
different procedures (i.e., computer mediated and face to face)
point to the social impact of anger and happiness on negotiation
behavior.
Given the pervasive interpersonal effects of anger and happiness
on negotiation behavior, it is worth considering whether other
emotions have the potential to affect negotiation behavior. Accord-
ing to the social functions perspective, emotions convey informa-
tion that is likely to influence other people’s behavior. For exam-
ple, Van Kleef et al. (2004a) demonstrated that negotiators
concede more to angry counterparts than to happy ones because
anger signals high limits, whereas happiness signals low limits.
Obviously, emotions can also convey other important information,
the strategic implications and interpretation of which are likely to
depend in part on observers’ appraisals of the cause of the emotion.
For instance, in a negotiation the distribution of resources can be
perceived as fair or unfair (Bazerman, Curhan, Moore, & Valley,
2000; Hegtvedt & Killian, 1999). Research on distributive justice
has shown that fair distributions give rise to positive emotions,
whereas unfair distributions give rise to negative emotions (De
Dreu, Lualhati, & McCusker, 1994; Loewenstein, Thompson, &
Bazerman, 1989). Individuals who receive what they expected
tend to experience happiness or satisfaction (Hegtvedt, 1990; Mes-
sick & Sentis, 1979; Sprecher, 1992). By contrast, individuals who
feel that they are getting too much or too little are likely to
experience negative emotions. Emotional reactions that are likely
to occur in individuals who feel underrewarded include disappoint-
ment, sadness, depression, anger, and resentment (Hegtvedt &
Killian, 1999), whereas those who feel overrewarded can be ex-
pected to experience guilt (Hegtvedt & Killian, 1999; Homans,
1974) or regret (van Dijk & Zeelenberg, 2002; Zeelenberg, van der
Pligt, & Manstead, 1998).
The purpose of the present research was to investigate the
interpersonal effects of emotions that may arise as a result of the
appraisal that one has taken too much or received too little.
125
Specifically, we focused on the interpersonal effects of disappoint-
ment, worry, guilt, and regret on demands and concessions in
negotiations. Below we review research pertaining to these emo-
tions, and we build on this research to advance a number of
hypotheses regarding the interpersonal effects of these emotions
on negotiation behavior.
When One Feels One Has Taken Too Much:
Appeasement
Emotions such as guilt, shame, embarrassment, and interper-
sonal regret serve an appeasement function (Baumeister, Stillwell,
& Heatherton, 1994; Keltner & Buswell, 1997). Of these emotions,
guilt is perhaps the most extensively researched. It entails a feeling
of self-reproach resulting from the belief that one has done some-
thing wrong (R. H. Smith, Webster, Parrott, & Eyre, 2002).
Baumeister et al. (1994) noted that “from an interpersonal per-
spective, the prototypical cause of guilt would be the infliction of
harm, loss, or distress on a relationship partner” (p. 245). The
experience of guilt is typically rooted in an interpersonal context
and often (although not necessarily) arises as a result of a per-
ceived transgression, in particular when the individual feels that he
or she has violated some expectation or norm (Leith & Baumeister,
1998; H. B. Lewis, 1971; Tangney, 1990, 1995, 1999).
Guilt is closely linked with reactions such as regret, self-
reproach, repentance, and remorse (R. H. Smith et al., 2002) and
tends to produce outwardly focused behaviors aimed at reducing
the damage caused by one’s behavior (Barrett, 1995; Tangney,
1995). Guilt is associated with perspective-taking, interpersonal
sensitivity, and improved relationship outcomes (Baumeister et al.,
1994; Leith & Baumeister, 1998). People experiencing guilt tend
to engage in behaviors aimed at repairing the social relationship
(Baumeister et al., 1994; M. Lewis, 2000). Research has shown
that transgressions and concomitant guilt increase subsequent
helping, compliance, and cooperation on the part of the transgres-
sor (Carlsmith & Gross, 1969; Freedman, Wallington, & Bless,
1967; Ketelaar & Au, 2003). Other research has found that guilt
motivates people to apologize and to make reparations or amends
(Friedman, 1985; Hoffman, 1982; H. B. Lewis, 1971). Further-
more, when the transgression has an interpersonal character, guilt
motivates people to compensate the victim (Berscheid & Walster,
1967; Wallace & Sadalla, 1966). Thus, if the transgressor displays
guilt, the victim may see this as an implicit commitment to rectify
the transgression by making amends and as a promise of better
treatment in the future (Baumeister et al., 1994; Manstead, 1991).
In an interpersonal context, social transgressions can also cause
feelings of regret (Zeelenberg et al., 1998). When the regret is
interpersonal in nature—that is, when one regrets a behavior that
has inflicted harm on another person (rather than on oneself)—it
shares a number of characteristics with guilt (Berndsen, van der
Pligt, Doosje, & Manstead, 2004; Roseman, Wiest, & Swartz,
1994). Because regret is an aversive state, people are motivated to
avoid it and, once they experience it, to take action to undo it
(Zeelenberg & Beattie, 1997; Zeelenberg, van Dijk, Manstead, &
van der Pligt, 2000). Gilovich and Medvec (1994, 1995) refer to
this undoing as “behavioral repair work” or “ameliorative behav-
ior.” In the case of interpersonal regret, this repair work typically
takes the form of apologizing to the person who has been affected
by the transgression (Steiner, 2000; Zeelenberg et al., 1998).
126 VAN KLEEF, DE DREU, AND MANSTEAD
By extrapolating these findings to the domain of conflict and
negotiation, we can advance the following general predictions.
First, a negotiator who deals with an opponent who appears to
experience emotions of appeasement (e.g., guilt or regret) is likely
to expect the opponent to make up for his or her previous “trans-
gression” (e.g., tough demands) by making a concession. Follow-
ing this line of reasoning, a negotiator faced with an opponent who
seems to be guilty or regretful can be expected to stand firm and
wait for the other to make a concession. Second, emotions of
appeasement are associated with interpersonal sensitivity and the
willingness to appreciate another person’s perspective, and they
share a number of relationship-repairing qualities (Baumeister et
al., 1994; Leith & Baumeister, 1998; Steiner, 2000). Guilt and
regret can therefore be expected to signal an outward focus and a
concern for the other and to have a beneficial effect on the
interpersonal relationship.
When One Feels One Has Received Too Little:
Supplication
Emotions such as sadness, disappointment, fear, and worry
serve a supplication function (M. S. Clark, Pataki, & Carver,
1996). These emotions communicate dependency and a need for
support (Eisenberg, 2000; Kennedy-Moore & Watson, 2001), and
they evoke empathy and helping behavior (M. S. Clark et al., 1996;
Eisenberg, 2000; Hill, Weary, & Williams, 1986; Parrott, 1993).
Sadness, for example, communicates to the self and to others that
one is in need of help (Tomkins, 1963). Expressions of sadness
have been demonstrated to increase perceptions of neediness and
dependency (M. A. Clark & Taraban, 1991) and to evoke helping
behavior in both children (Barnett, Howard, Melton, & Dino,
1982) and adults (M. S. Clark, Ouellette, Powell, & Milberg, 1987;
Yee & Greenberg, 1998). In a similar vein, crying serves a help-
seeking function (Labott, Martin, Eason, & Berkey, 1991). For
example, Cornelius (1984) demonstrated that (involuntary) crying
was an effective means of eliciting a positive and desired change
in other people’s behavior, in this case a shift from conflict to
support.
Similar effects on helping behavior have been observed for
expressions of worry and fear. Like sadness, worry and fear
communicate a need for assistance, and they elicit sympathetic and
supportive responses in others (Eisenberg, 2000; Kennedy-Moore
& Watson, 2001). For example, a study of reactions to crime
victims by Yee and Greenberg (1998) revealed that fear on the part
of the victims influences observers’ appraisals of need and in-
creases the inclination to help, especially if the observer and the
victim are in a communal rather than an exchange relationship. By
the same token, employees who display fear are likely to evoke
helpful, supportive responses from coworkers (Côté, 2005).
In contrast to the interpersonal effects of sadness and, to a lesser
degree, fear and worry, the interpersonal effects of disappointment
have not received much research attention. Carver and Scheier
(1990) suggested that disappointment signals that progress toward
a goal is below expectations. In other words, disappointment arises
when a desired outcome is not achieved (Bell, 1985; Frijda, 1986;
van Dijk & van der Pligt, 1997), and as such, it is highly relevant
to negotiation. Research on the effects of disappointment is sparse,
and it has focused predominantly on intrapersonal consequences
(e.g., engaging in behaviors aimed at minimizing future disap-
pointment; Bell, 1985; Loomes & Sugden, 1986; see Zeelenberg et
al., 2000, for an overview). Timmers, Fischer, and Manstead
(1998) reported evidence suggesting that under particular circum-
stances, people may deliberately express disappointment to change
the behavior of a target person in a direction that would be
beneficial for the expresser. We argue that, at the interpersonal
level, disappointment is likely to have effects similar to those of
other distress-related emotions (e.g., fear, worry, sadness), which
have generally been shown to facilitate prosocial behavior aimed
at easing the other’s pain (Barnett, King, & Howard, 1979; Batson,
1987; Eisenberg, Fabes, Miller, et al., 1989; Fabes, Eisenberg,
Karbon, Troyer, & Switzer, 1994; Morris & Keltner, 2000).
On the basis of the above considerations, it can be predicted that
a negotiator who is confronted with an opponent who appears to
experience emotions of supplication (e.g., disappointment or
worry) will try to relieve the other’s pain by making concessions.
Further, emotions of distress are associated with self-focus and
egoistic motivations (Eisenberg, Fabes, Schaller, & Miller, 1989).
By reverse analogy with previous research suggesting that ap-
peasement emotions enhance interpersonal relationships because
they signal interpersonal sensitivity and concern (Baumeister et al.,
1994; Leith & Baumeister, 1998), we argue that supplication
emotions lead to less positive impressions because they signal an
inward focus and preoccupation with the self.
Experiment 1
The purpose of Experiment 1 was to investigate the interper-
sonal effects of emotions related to supplication and appeasement
in negotiations. We predicted that participants with an opponent
who experiences disappointment or worry (i.e., emotions of sup-
plication) would make smaller demands than would participants
with a nonemotional opponent (Hypothesis 1a), whereas negotia-
tors with an opponent who experiences guilt or regret (i.e., emo-
tions of appeasement) would make larger demands (Hypothesis
1b). We also expected that participants with a guilty or regretful
opponent would perceive the opponent as more interpersonally
sensitive than would participants with a disappointed or worried
opponent (Hypothesis 2a) and that they would develop a more
favorable impression of the opponent (Hypothesis 2b). Finally, we
explored whether the effects of the opponent’s emotion on partic-
ipants’ impressions of the opponent are mediated by their percep-
tion of the opponent’s interpersonal sensitivity (Hypothesis 2c).
Method
Participants and Experimental Design
A total of 84 male and female undergraduate students at the University
of Amsterdam participated either in partial fulfillment of a course require-
ment or for monetary compensation (EUR 7, approximately US $8). The
experimental design included the opponent’s emotion (disappointment vs.
worry vs. guilt vs. regret vs. no emotion) as the independent variable and
demand level as the main dependent variable. Participants were randomly
assigned to the conditions, and the experimenters were blind to this
assignment.
Procedure
For each session, 6 to 8 participants were invited to the laboratory. On
arrival, participants were welcomed to the experiment and seated in sep-
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
arate cubicles in front of a computer. From that point onward all instruc-
tions, questionnaires, and experimental tasks were presented on the com-
puter screen. To facilitate the manipulation of the opponent’s emotion (see
below), we led participants to believe that the purpose of the study was to
find out how knowledge about one’s opponent’s intentions affects negoti-
ation processes in a situation in which the negotiating parties cannot see
each other. They were then told that they would engage in a computer-
mediated negotiation with another participant (whose behavior was in fact
simulated by the computer).
Negotiation task. The negotiation task was one used previously by Van
Kleef et al. (2004a, b; see also De Dreu & Van Lange, 1995; Hilty &
Carnevale, 1993). The task captures the main characteristics of real-life
negotiation (i.e., multiple issues differing in utility to the negotiator,
information about one’s own payoffs only, and the typical offer–
counteroffer sequence). In the current version, participants learned that
they would be assigned the role of buyer or seller of a consignment of
mobile phones and that their objective was to negotiate the price, the
warranty period, and the duration of the service contract of the phones.
They were then presented with a payoff chart (see Appendix A) that
showed which outcomes were most favorable to them, and they learned
that their objective was to earn as many points as possible. Level 9 on price
($110) yielded 0 points, and Level 1 ($150) yielded 400 points (i.e.,
increments of 50 points per level). For warranty period, Level 9 (9 months)
yielded 0 points, and Level 1 (1 month) yielded 120 points (i.e., increments
of 15 points per level). Finally, for duration of service contract, Level 9 (9
months) yielded 0 points, and Level 1 (1 month) yielded 240 points (i.e.,
increments of 30 points per level). Participants were told, “You can see that
the best deal for you is 1–1–1, for a total outcome of 760 points (400 ⫹
120 ⫹ 240).” The corresponding payoff table for the other party was not
displayed, and participants were told only that it differed from their own.
To enhance involvement in the negotiation task, we informed partici-
pants that points would be converted to lottery tickets at the end of the
experiment and that the more points earned, the more lottery tickets one
would obtain and the greater would be one’s chance of winning a EUR 50
(approximately US $64) prize. To emphasize the mixed-motive nature of
the negotiation, we told participants that only those who reached an
agreement would participate in the lottery. Thus, there were incentives both
to earn as many points as possible and to reach an agreement.
After a short pause during which the computer supposedly assigned
buyer and seller roles to the participants, all participants were assigned the
role of seller. They were told that the buyer (i.e., the opponent) would make
the first offer and that the negotiation would continue until an agreement
was reached or time ran out. Just before the negotiation started, participants
learned that an additional goal of the study was to examine the effects of
having versus not having information about the opposing negotiator’s
intentions. They read that the computer had randomly determined that they
would receive information about the intentions of the opponent and that the
opponent would not receive information about their intentions.
After these instructions, the negotiation started, and the buyer (i.e., the
computer) made a first offer. Over the negotiation rounds the buyer
proposed the following levels of agreement (for price–warranty–service):
8 –7– 8 (Round 1), 8 –7–7 (Round 2), 8 – 6 –7 (Round 3), 7– 6 –7 (Round 4),
7– 6 – 6 (Round 5), and 6 – 6 – 6 (Round 6). Past research has shown that this
preprogrammed strategy has face validity and is seen as intermediate in
cooperativeness and competitiveness (De Dreu & Van Lange, 1995). A
demand by the participant was accepted if it equaled or exceeded the offer
the computer was about to make in the next round. If no agreement was
reached by the sixth round, the negotiation was interrupted (see De Dreu &
Van Lange, 1995). Following the procedure of a study by Tripp and
Sondak (1992), we excluded from the sample participants who reached
agreement before Round 6 (n ⫽ 8) to allow for repeated-measures analy-
ses. (However, retaining these participants yielded a similar pattern of
results.)
127
Manipulation of the opponent’s emotion. We chose to manipulate the
opponent’s emotion in the context of a computer-mediated negotiation in
which parties could not see each other and communicated via computers
(see De Dreu & Van Kleef, 2004; De Dreu & Van Lange, 1995; Hilty &
Carnevale, 1993). Participants were led to believe that the purpose of the
study was to find out how knowledge about one’s opponent’s intentions
affects negotiation processes and outcomes. After the first, third, and fifth
negotiation rounds, participants received information about “the intentions
of the buyer,” which contained the manipulation of the buyer’s emotion.
Participants had to wait for about a minute and a half while the buyer was
supposedly asked to reveal what he or she intended to offer in the next
round, and why. After this short wait, participants received what appeared
to be the buyer’s answer. This was presented in a separate box in a
different font and contained some minor typing errors in order to
enhance experimental realism. The buyer’s intentions were held constant
across conditions and contained the buyer’s intended offer for the next
round. That is, after Round 1 the buyer wrote, “I think I will offer 8 –7–7,”
which would indeed be the buyer’s next offer. The buyer’s intention
information also contained an emotional statement that constituted the
experimental manipulation.
The emotion statements were pretested in a pilot study involving 64
psychology students, none of whom participated in the main study. We
tested seven statements designed to reflect disappointment, six statements
designed to reflect worry, seven statements designed to reflect interper-
sonal regret, and seven statements designed to reflect guilt. The statements
were pretested using a within-participants design. All participants rated a
selection of 13 or 14 out of the total of 27 statements, the order of
statements being randomized across participants. The statements were
distributed in such a way that each was rated by half of the participants. For
each statement, participants were asked to indicate on a 7-point scale how
comprehensible they found it (1 ⫽ very incomprehensible, 7 ⫽ very
comprehensible) and to what extent they felt it reflected disappointment,
worry, regret, and guilt (1 ⫽ not at all, 7 ⫽ to a great extent). We then
selected the statements that had the highest scores on the emotion they
were supposed to reflect and the lowest scores on the emotions that they
were not supposed to reflect, provided that they did not differ with respect
to their comprehensibility. We selected three statements for each emotion.
All selected statements were rated higher on the emotion they were
supposed to express than on the emotions they were not supposed to
express according to paired-samples t tests (5.09 ⬍ ts ⬍ 18.29, all ps ⬍
.01). Further, one-sample t tests showed that there was a significant effect
of all statements on the rating of the corresponding emotion (13.80 ⬍ ts ⬍
18.32, all ps ⬍ .01). Finally, paired-samples t tests revealed that the
statements did not differ with respect to comprehensibility (all ts ⬍ 1.12,
ns).
After the first negotiation round, participants in the disappointed oppo-
nent condition received the following information: “I am pretty disap-
pointed about this,” followed by the intention statement “I think I will offer
8 –7–7,” which was the same for all conditions. In the worried opponent
condition, participants read “This worries me quite a lot”; in the guilty
opponent condition, participants read “I feel guilty for not having conceded
more”; in the regret condition, participants read “I am sorry that I haven’t
conceded more.” In the control condition participants received only the
intention statement. After the third and fifth negotiation rounds, partici-
pants in the experimental conditions again received an emotional statement
and an intention, whereas those in the control condition simply received the
intention. Appendix B lists all statements used in the experiment.
Dependent measures. The main dependent variable was participants’
level of demand in Rounds 1 to 6. In addition, participants completed a
postnegotiation questionnaire that contained manipulation checks and
items designed to measure impressions of the opponent. To check the
adequacy of the emotion manipulation, we asked participants to indicate on
a 7-point scale how disappointed, worried, guilty, and regretful they
thought their opponent had been during the negotiation. Perceptions of the
128 VAN KLEEF, DE DREU, AND MANSTEAD

opponent’s disappointment were measured by three items (e.g., “The buyer
appeared to be disappointed during the negotiation,” 1 ⫽ totally disagree
to 7 ⫽ totally agree), which were combined into a single index of
perceived disappointment (␣ ⫽ .83). Perception of the opponent’s worry
was measured using five items (e.g., “The buyer appeared to be worried
during the negotiation”), which were averaged into an index of perceived
worry (␣ ⫽ .88). Perception of guilt was measured by three items (e.g.,
“The buyer appeared to feel guilty during the negotiation”), which were
combined into an index of perceived guilt (␣ ⫽ .93). Finally, perceptions
of the opponent’s regret were measured by three items (e.g., “The buyer
appeared to feel sorry during the negotiation”), which were averaged into
a scale of perceived regret (␣ ⫽ .91).
Impression of the opponent was assessed with seven items (“The buyer
strikes me as a sympathetic person”; “During the negotiation, the buyer
made a hostile impression,” reverse scored; “The buyer made a cooperative
impression”; “The buyer made a friendly impression”; “The buyer made a
competitive impression,” reverse scored; “The buyer made a stubborn
impression,” reverse scored; “I have developed a positive impression of the
buyer”; 1 ⫽ totally disagree to 7 ⫽ totally agree). The seven items were
combined into a measure of impression of the opponent (␣ ⫽ .79). We also
included three items measuring participants’ perceptions of the opponent’s
interpersonal sensitivity and perspective taking (“During the negotiation,
the buyer was self-centered,” reverse scored; “During the negotiation, the
buyer took my interests into consideration”; “During the negotiation, the
buyer appeared to be preoccupied with him- or herself,” reverse scored).
These three items were averaged into an index of the opponent’s interper-
sonal sensitivity (␣ ⫽ .82).
Results
Treatment of the Data
The offers made by participants in each round were transformed
into an index revealing the negotiator’s total level of demand for
each negotiation round (i.e., the number of points demanded in that
round, summed across the three negotiation issues of price, war-
ranty, and service; see Appendix A).
intended emotion than for the other emotions (2.27 ⬍ ts ⬍ 11.73,
all ps ⬍ .05).
Demand Level
Demand level in Rounds 1 to 6 was analyzed using a mixed-
model ANOVA with the opponent’s emotion (disappointment vs.
worry vs. guilt vs. regret vs. no emotion) as a between-participants
variable and negotiation round (1 to 6) as a repeated-measures
variable. Unsurprisingly, this analysis revealed a main effect of
round, F(5, 355) ⫽ 286.28, p ⬍ .01 (partial ␩2 ⫽ .80), indicating
that demands declined from Round 1 (M ⫽ 677, SD ⫽ 63) to
Round 6 (M ⫽ 484, SD ⫽ 93). More important, there was a
significant effect of the opponent’s emotion on average demands,
F(4, 71) ⫽ 7.76, p ⬍ .01 (partial ␩2 ⫽ .30), indicating that
participants’ demands were influenced by their adversaries’ emo-
tions (disappointment: M ⫽ 533, SD ⫽ 59; worry: M ⫽ 513, SD ⫽
76; guilt: M ⫽ 633, SD ⫽ 79; regret: M ⫽ 592, SD ⫽ 46; control:
M ⫽ 569, SD ⫽ 56). Finally, the main effects of the opponent’s
emotion and negotiation round were qualified by a significant
two-way interaction, F(20, 355) ⫽ 5.21, p ⬍ .01 (partial ␩2 ⫽ .23).
As can be seen in Figure 1, the influence of the opponent’s
emotion became more apparent as the negotiation progressed. In
Round 1, there were no differences among any of the conditions.
This is hardly surprising, given that the emotion manipulation
began only after this round. From the second round onward,
however, the different conditions started to diverge, the effect
becoming stronger after each consecutive round. Therefore, we
decided to use demands in Round 6 for our specific hypothesis
tests.
Hypotheses 1a and 1b were tested with planned comparisons.
The means and standard deviations as well as the specific contrasts

Manipulation Check
A 5 (opponent’s emotion: disappointment, worry, guilt, regret,
no emotion) ⫻ 4 (perception of the opponent’s emotion: disap-
pointment, worry, guilt, regret) analysis of variance (ANOVA)
with repeated measures on the last factor revealed a significant
interaction between the opponent’s emotion and participants’ per-
ceptions of the opponent’s emotion, F(12, 213) ⫽ 31.77, p ⬍ .01
(partial ␩2 ⫽ .59). Post hoc tests showed that participants in the
disappointed opponent condition rated the opponent as more dis-
appointed (M ⫽ 5.96, SD ⫽ 1.25) than did participants in all the
other conditions (2.84 ⬍ Ms ⬍ 4.95, 0.96 ⬍ SDs ⬍ 1.29).
Similarly, participants with a worried opponent rated the opponent
as more worried (M ⫽ 5.94, SD ⫽ 0.85) than did participants in the
other conditions (2.54 ⬍ Ms ⬍ 4.79, 0.83 ⬍ SDs ⬍ 1.68), and
participants with a guilty opponent rated the opponent as more
guilty (M ⫽ 5.85, SD ⫽ 1.39) than did those in the other conditions
(1.85 ⬍ Ms ⬍ 4.39, 0.82 ⬍ SDs ⬍ 1.85). Finally, participants with
a regretful opponent rated the other as more regretful (M ⫽ 5.03,
SD ⫽ 1.60) than did participants in all the other conditions (1.98 ⬍
Ms ⬍ 2.77, 0.68 ⬍ SDs ⬍ 1.04) except the guilty opponent
condition (M ⫽ 4.81, SD ⫽ 1.72). Furthermore, paired-samples t Figure 1. Demand level as a function of the opponent’s emotion and
tests showed that ratings within each condition were higher for the negotiation round in Experiment 1.
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
that were computed to test these hypotheses are shown in Table 1.
Four contrasts were computed. First, we tested whether partici-
pants made smaller demands to opponents who experienced sup-
plication emotions (i.e., disappointment and worry) than to non-
emotional opponents (Contrast 1). Second, we examined whether
participants responded differentially to disappointed versus wor-
ried counterparts in terms of their demands (Contrast 2). Third, we
examined whether participants made larger demands to opponents
engaging in appeasement (i.e., guilt or regret) than to nonemo-
tional opponents (Contrast 3). Fourth, we tested whether partici-
pants responded differentially to guilty versus regretful counter-
parts (Contrast 4). For Hypotheses 1a and 1b to be supported,
Contrasts 1 and 3 should be significant (participants should make
lower demands to opponents showing signs of supplication and
higher demands to opponents showing signs of appeasement).
Contrasts 2 and 4 speak to the question of whether the two
supplication emotions (disappointment and worry) have similar
effects on demands, and whether the two appeasement emotions
(guilt and regret) have similar effects on demands. The latter two
contrasts were computed for exploratory purposes.
In accordance with Hypothesis 1a, Contrast 1 was significant.
Participants made smaller demands to a disappointed or worried
opponent than to a nonemotional opponent, t(71) ⫽ 2.89, p ⬍ .01
(partial ␩2 ⫽ .15). Contrast 2 was nonsignificant, indicating that
participants did not make differential demands to disappointed
versus worried counterparts, t(71) ⬍ 1, ns. In support of Hypoth-
esis 1b, Contrast 3 was significant—participants made larger de-
mands to a guilty or regretful opponent than to a nonemotional
one, t(71) ⫽ 2.10, p ⬍ .05 (partial ␩2 ⫽ .08). Unexpectedly,
Contrast 4 was also significant, indicating that participants made
smaller demands to regretful than to guilty counterparts, t(71) ⫽
2.02, p ⬍ .05 (partial ␩2 ⫽ .07). These findings clearly support the
prediction that negotiators concede more to opponents who expe-
rience supplication emotions than to opponents who experience
appeasement emotions. Although the supplication emotions disap-
pointment and worry had similar effects on behavior, the effects of
the appeasement emotions guilt and regret differed.
Ratings of Interpersonal Sensitivity and Impression of the
Opponent
An ANOVA showed that participants’ ratings of the opponent’s
interpersonal sensitivity were influenced by the opponent’s emo-
tion, F(4, 71) ⫽ 5.62, p ⬍ .01 (partial ␩2 ⫽ .24). As predicted
Table 1
Means, Standard Deviations (in Parentheses), and Contrasts Computed to Test Hypothesis 1
(Experiment 1)
Demand and contrast Disappointment Worry
Demand in Round 6 435 (63) 422 (85)
Contrast 1* 1 1
Contrast 2 1 ⫺1
Contrast 3* 0 0
Contrast 4* 0 0
* p ⬍ .05.
129
(Hypothesis 2a), planned comparisons revealed that participants
with a guilty or regretful opponent judged the opponent to be more
interpersonally sensitive (M ⫽ 4.31, SD ⫽ 1.36 and M ⫽ 4.33,
SD ⫽ 1.39, respectively) than did participants with a worried or
disappointed opponent (M ⫽ 3.15, SD ⫽ 1.05 and M ⫽ 2.69, SD ⫽
1.14, respectively), t(71) ⫽ 4.37, p ⬍ .002 (partial ␩2 ⫽ .21).
Participants with a nonemotional opponent occupied an interme-
diate position (M ⫽ 3.86, SD ⫽ 1.15) that differed from the
disappointment and worry conditions, t(71) ⫽ 2.76, p ⬍ .01
(partial ␩2 ⫽ .15), but not from the guilt and regret conditions,
t(71) ⫽ 1.22, ns.
We also found a significant effect of emotion on participants’
impressions of the opponent, F(4, 71) ⫽ 4.87, p ⬍ .002 (partial
␩2 ⫽ .22). Consistent with Hypothesis 2b, planned comparisons
showed that participants developed more favorable impressions of
opponents who experienced guilt or regret (M ⫽ 4.48, SD ⫽ 0.99
and M ⫽ 4.24, SD ⫽ 0.87, respectively) than of opponents who
experienced disappointment or worry (M ⫽ 3.19, SD ⫽ 1.07 and
M ⫽ 3.78, SD ⫽ 0.75, respectively), t(71) ⫽ 3.64, p ⬍ .01 (partial
␩2 ⫽ .16). Again, participants with a nonemotional opponent
occupied an intermediate position (M ⫽ 4.02, SD ⫽ 0.79), differ-
ing from the disappointment and worry conditions, t(71) ⫽ 2.14,
p ⬍ .04 (partial ␩2 ⫽ .09), but not from the guilt and regret
conditions, t(71) ⫽ 1.28, ns.
To test the idea that participants developed more favorable
impressions of opponents who engaged in appeasement than of
opponents who engaged in supplication because they perceived the
former to be more interpersonally sensitive than the latter (Hy-
pothesis 2c), we conducted mediated regression analyses (see
Baron & Kenny, 1986). To this end, the opponent’s emotion was
dummy coded (0 for worry and disappointment and 1 for guilt and
regret). This dummy variable had a significant effect on partici-
pants’ impressions of the opponent (the dependent variable; ␤ ⫽
.45, p ⬍ .01) and on participants’ judgments of the opponent’s
interpersonal sensitivity (the mediator; ␤ ⫽ .51, p ⬍ .01). When
both variables were simultaneously entered into the regression, a
significant effect of interpersonal sensitivity on impression
emerged (␤ ⫽ .69, p ⬍ .01), and the originally significant effect of
the dummy variable on impression was reduced to nonsignificance
(␤ ⫽ .10, ns). The reduction of the direct path from the opponent’s
emotion to participants’ impressions of the opponent was signifi-
cant according to a Sobel test (Z ⫽ 3.74, p ⬍ .01; see Kenny,
Kashy, & Bolger, 1998, and for an updated version of the formula,
Condition
Guilt Regret No emotion
573 (100) 527 (59) 492 (70)
0 0 ⫺2
0 0 0
1 1 ⫺2
1 ⫺1 0
130 VAN KLEEF, DE DREU, AND MANSTEAD
see Kenny’s Web site at http://users.rcn.com/dakenny/mediate.
htm). These findings suggest that the effect of the opponent’s
emotion on impressions is indeed mediated by judgments of in-
terpersonal sensitivity, although it should be noted that both con-
structs were measured at the same time.
Discussion
The results of Experiment 1 are largely supportive of our pre-
dictions. In line with Hypothesis 1, participants whose opponents
experienced emotions of supplication (i.e., disappointment or
worry) made smaller demands than did participants whose oppo-
nents experienced emotions of appeasement (i.e., guilt or regret).
Negotiators who are faced with an opponent who experiences
emotions of appeasement appear to anticipate behavioral repair in
the form of concessions and stand firm. Conversely, negotiators
who deal with a counterpart who experiences emotions of suppli-
cation respond to the other’s discomfort by making concessions.
Thus, supplication and appeasement emotions have unique and
theoretically meaningful effects on negotiation behavior. Further,
and consistent with Hypotheses 2a to 2c, participants whose op-
ponents showed signs of appeasement developed a more positive
impression of the opponent than did those whose opponents en-
gaged in supplication, because the former were perceived as more
interpersonally sensitive than the latter. This finding is in line with
the presumed relationship-enhancing functions of expressing guilt
and regret.
One finding of Experiment 1 was unanticipated. On the basis of
the literature reviewed in the introduction, we expected guilt and
regret to have similar effects on demands. However, the present
findings suggest that guilt has a stronger effect on demands. A
possible explanation could be that expressions of regret are more
ambiguous. Expressing regret for not having conceded more could
indicate regret for hurting someone else or regret for not having
been more strategic and self-interested. Expressions of guilt are
more unequivocal in that they necessarily imply that one feels bad
about one’s behavior vis-à-vis someone else.
Another issue concerns the emotional statements that were used
to manipulate guilt and regret. In previous work (Van Kleef et al.,
2004a, b), we used rather straightforward expressions to manipu-
late anger and happiness (e.g., “This offer makes me really angry/
happy”). Although similar expressions could easily be used to
express disappointment and worry (e.g., “I am very disap-
pointed”), we felt that expressions of guilt and regret would appear
implausible unless they were accompanied by some kind of justi-
fication. We therefore decided to add a brief explanation to the first
guilt and regret statements for why the opponent felt guilty or
regretful (e.g., “I feel guilty for not having conceded more”).
However, in addition to making the emotion statement more cred-
ible, this also inadvertently introduced an implicit intention to
concede. Despite the fact that only the first of three guilt and regret
statements was accompanied by such an implicit intention, we
cannot be sure that guilt and regret were driving the effects. This
issue is dealt with in Experiment 1b.
Experiment 1b
The goal of Experiment 1b was to identify the unique interper-
sonal effects of guilt on demands by contrasting the guilt manip-
ulation of Experiment 1 with a “clean” manipulation that did not
include a justification for why the guilt was experienced. Both
guilt conditions were compared with a control condition. We
expected that participants in both guilt conditions would make
smaller concessions in the course of the negotiation than would
participants in the control condition and that the two guilt condi-
tions would not differ from each other. Extending the line of
argument presented earlier, we further predicted that participants
in both guilt conditions would expect larger concessions from the
opponent than would participants in the control condition (Hypoth-
esis 3).
Method
Participants and Design
Eighty-four undergraduate students at the University of Amsterdam
were assigned to one of three conditions: guilt with implicit intention to
concede, guilt without implicit intention to concede, or no emotion. In
return for participation, they received course credit or EUR 5 (approxi-
mately $6).
Procedure
The procedure was similar to the one used in Experiment 1, except for
the manipulation of guilt. In the guilt with intention condition, we used the
same guilt statements as in Experiment 1. Thus, after the first negotiation
round, participants received the following statement from the opponent
(see Appendix B): “I feel guilty for not having conceded more, I think I
will offer 8 –7–7.” In the guilt without intention condition, we omitted the
explanatory part. Thus, after the first round, participants simply read “I feel
guilty, I think I will offer 8 –7–7.” Participants who reached agreement
before Round 6 (n ⫽ 7) were excluded from the analyses (although
retaining these participants yielded similar results).
Dependent Measures
In addition to recording demands, we included a brief questionnaire to
measure participants’ expectations of the opponent’s future concessions
and to check the emotion manipulation. After Round 6, participants read
that the negotiation would be temporarily interrupted for some questions
and that the negotiation would resume later. Expectations regarding the
opponent’s future concessions were then measured by three items (“I
expect that the opponent will make large concessions in the next rounds”,
“I expect that the opponent will take a cooperative stance”, “I expect that
the opponent will be conciliatory”; 1 ⫽ totally disagree to 9 ⫽ totally
agree; ␣ ⫽ .85). The manipulation of guilt was checked using the same
scale as in Experiment 1 (␣ ⫽ .89).
Results
Manipulation Check
An ANOVA revealed a significant effect of the opponent’s
emotion on participants’ perceptions of the opponent’s emotion,
F(2, 74) ⫽ 37.99, p ⬍ .01 (partial ␩2 ⫽ .51). Planned comparisons
showed that participants in the guilt with intention (M ⫽ 6.36,
SD ⫽ 1.77) and guilt without intention (M ⫽ 6.26, SD ⫽ 2.01)
conditions rated the other as more guilty than did participants in
the no-emotion condition (M ⫽ 3.02, SD ⫽ 1.18), t(74) ⫽ 8.72,
p ⬍ .01 (partial ␩2 ⫽ .51). The two guilt conditions did not differ
from each other, t(74) ⫽ .21, ns.
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
Demand Level
As in Experiment 1, an ANOVA revealed a significant main
effect of round, F(5, 370) ⫽ 128.76, p ⬍ .01 (partial ␩2 ⫽ .76),
indicating that demands declined from Round 1 (M ⫽ 641, SD ⫽
72) to Round 6 (M ⫽ 489, SD ⫽ 100). More important, we found
a significant effect of the opponent’s emotion on average demands,
F(2, 74) ⫽ 4.53, p ⬍ .015 (partial ␩2 ⫽ .08), indicating that
participants’ demands were influenced by the opponent’s emotion
(guilt with intention: M ⫽ 575, SD ⫽ 78; guilt without intention:
M ⫽ 574, SD ⫽ 88; control: M ⫽ 521, SD ⫽ 66). These main
effects were qualified by a significant interaction between emotion
and negotiation round, F(10, 370) ⫽ 4.64, p ⬍ .01 (partial ␩2 ⫽
.14). As in Experiment 1, the influence of the opponent’s emotion
became more apparent as the negotiation progressed. In Round 1,
there were no differences among any of the conditions. From the
third round onward, the different conditions started to diverge, the
effect becoming stronger after each round. Planned comparisons of
demands in Round 6 revealed that both guilt conditions (guilt with
intention: M ⫽ 524, SD ⫽ 88; guilt without intention: M ⫽ 536,
SD ⫽ 107) differed significantly from the control condition (M ⫽
430, SD ⫽ 70), t(74) ⫽ 4.93, p ⬍ .01 (partial ␩2 ⫽ .19), and that
they did not differ significantly from each other, t(74) ⫽ .45, ns.
Expectations of Opponent’s Future Concessions
An ANOVA yielded an effect of emotion on participants’ ex-
pectations of the opponent’s future concessions, F(2, 74) ⫽ 6.35,
p ⬍ .01 (partial ␩2 ⫽ .14). Planned comparisons revealed that
participants in both guilt conditions expected more cooperation
from the opponent (guilt with intention: M ⫽ 5.45, SD ⫽ 1.19;
guilt without intention: M ⫽ 5.10, SD ⫽ 1.28) than did participants
in the control condition (M ⫽ 4.28, SD ⫽ 1.31), t(74) ⫽ 3.45, p ⬍
.01 (partial ␩2 ⫽ .06). Again, the guilt conditions did not differ
from each other, t(74) ⫽ .82, ns.
Discussion
These results show that the interpersonal effects of guilt as
observed in Experiment 1 also obtain when no justification for the
experienced guilt is provided or, in other words, no implicit
intention to concede is mentioned. In line with our theoretical
model, participants in both guilt conditions expected the opponent
to make large concessions, supporting Hypothesis 3. Furthermore,
participants in both guilt conditions made smaller concessions than
did participants in the control condition, providing additional sup-
port for Hypothesis 1b.
Experiment 2
The results thus far support the central proposition of the present
research: In a negotiation, emotions of appeasement elicit high
demands from one’s adversary, whereas emotions of supplication
elicit low demands. The objective of Experiment 2 was twofold.
The first objective was to replicate and extend the findings of
Experiment 1 by investigating the potential moderating role of
interpersonal trust. Trust plays an important role in negotiations,
and it is essential to the resolution of mixed-motive conflict
(Lindskold, 1978). It can be defined as “a psychological state
comprising the intention to accept vulnerability based upon posi-
131
tive expectations of the intentions or behaviors of another” (Rous-
seau, Sitkin, Burt, & Camerer, 1998, p. 395). Among other things,
level of trust has been shown to affect the exchange of information
regarding preferences and priorities in negotiations (Pruitt & Kim-
mel, 1977), the willingness to cooperate (Yamagishi, 1986), the
attainment of integrative solutions (De Dreu, Giebels, & Van de
Vliert, 1998; Kimmel, Pruitt, Magenau, Konar-Goldband, &
Carnevale, 1980), and the desire for future interaction (Naquin &
Paulson, 2003; for a review, see Ross & LaCroix, 1996).
Aside from its specific effects in the negotiation arena, trust has
a more generalized influence on the trustworthiness individuals
ascribe to others and on their willingness to accept information as
sincere and accurate (Parks, Henager, & Scamahorn, 1996; Rotter,
1980). This has important consequences for negotiators’ reactions
to their opponent’s emotions. Although the effects of trust on
responses to other people’s emotions have not been the explicit
focus of research attention, some indirect evidence exists for the
moderating role of trust. For instance, research has shown that the
interpersonal consequences of expressions of distress depend on
how receivers interpret such expressions (Kennedy-Moore &
Watson, 2001). Sometimes recipients respond to these expressions
in less than helpful ways because they misread the expresser’s
intent (L. F. Clark, 1993). When observers perceive that the
expression is untruthful or manipulative, they are less likely to
respond positively. On the other hand, when they view the expres-
sions as trustworthy or at least benign, they are more likely to
respond with social helping (Kennedy-Moore & Watson, 2001). In
other words, individuals’ reactions to others’ emotions are likely to
be moderated by interpersonal trust.
Variations in trust may be rooted in individual differences or
stem from characteristics of the situation (Yamagishi, 1986). The
present experiment is concerned with the potential moderating role
of individual differences in interpersonal trust. The aforemen-
tioned research suggests that a negotiator’s decision to adapt his or
her demands to the opponent’s emotions will be influenced by the
focal negotiator’s level of interpersonal trust. Negotiators with
high levels of trust can be expected to adapt their demands to the
opponent’s emotional state, whereas those with low trust may not
respond differentially to the other’s emotions. To test this idea, we
manipulated the opponent’s emotion (guilt vs. disappointment)
and measured participants’ level of dispositional trust. We pre-
dicted that individuals high in trust would make higher demands to
a guilty opponent (i.e., an opponent engaging in appeasement) than
to a disappointed opponent (i.e., an opponent engaging in suppli-
cation). By contrast, we predicted that individuals low in disposi-
tional trust would not respond differentially to their counterpart’s
emotions (Hypothesis 4). We further hypothesized that, compared
with high trust participants, low trust participants would be more
suspicious about the opponent’s emotion and more likely to dis-
count the other’s emotions rather than incorporate them into their
negotiation strategy (Hypothesis 5).
The second objective of Experiment 2 was to shed more light on
the process underlying the effects of supplication and appeasement
emotions on demands. According to the social functions perspec-
tive that was outlined in the introduction, appeasement emotions
such as guilt signal that behavioral repair can be anticipated,
whereas supplication emotions such as disappointment signal that
the other needs help. It seems reasonable to assume that such
considerations would affect participants’ negotiation objectives
132 VAN KLEEF, DE DREU, AND MANSTEAD
and strategy. In this context it is useful to consider the role of the
negotiator’s goal (Kelley, Beckman, & Fischer, 1967; Pruitt &
Carnevale, 1993; Siegel & Fouraker, 1960; Zetik & Stuhlmacher,
2002). Among other things, higher goals tend to produce larger
demands and greater resistance to concession making (Holmes,
Throop, & Strickland, 1971; Kelley et al., 1967; D. L. Smith,
Pruitt, & Carnevale, 1982; Yukl, 1974a, b). A negotiator who
expects to receive compensation in the form of large concessions
(e.g., in the case of a guilty opponent) can be expected to adopt
higher goals than a negotiator who does not anticipate large
concessions (e.g., in the case of a disappointed opponent). Thus, it
may be that negotiators concede more to a disappointed opponent
than to a guilty one because the other’s guilt raises their goals,
whereas disappointment lowers them. To examine this possibility,
we measured participants’ goals and hypothesized that participants
with a guilty opponent would adopt higher goals than would those
with a disappointed opponent (Hypothesis 6). We further predicted
that negotiators’ tendency to make higher demands to guilty op-
ponents than to disappointed ones would be mediated by the
ambitiousness of their goals (Hypothesis 7). Finally, we were
interested to see whether guilt and disappointment are indeed
interpreted to mean that someone has received too much versus too
little, as suggested by the literature reviewed in the introduction.
This issue was addressed in an exploratory fashion.
Method
Participants and Experimental Design
Participants were 154 undergraduate students at the University of Am-
sterdam. They took part in the experiment for course credit or for monetary
compensation (EUR 7, approximately U.S.$8). The experimental design
included the opponent’s emotion (disappointment vs. guilt) and the partic-
ipant’s dispositional trust (high vs. low) as the independent variables and
demand level as the main dependent variable. Participants were randomly
assigned to the conditions with the use of a double-blind procedure.
Procedure
The procedure was identical to the one used in Experiment 1, with one
major exception. In addition to manipulating the opponent’s emotion, we
also measured the participant’s level of dispositional trust (see below). As
in Experiment 1, participants who reached agreement before Round 6 (n ⫽
6) were excluded from the analyses (although retaining these participants
yielded a similar pattern of results).
Assessment of trust. Trust was assessed with Yamagishi’s (1986) trust
scale, which comprises the following five items: “Most people tell a lie if
they can benefit from doing so”, “Those devoted to unselfish causes are
often exploited by others”, “Many people do not cooperate because they
pursue only their own interests”, “Most people are basically honest”, and
“There will be more people who will not work if the social security system
is developed further” (1 ⫽ totally disagree to 7 ⫽ totally agree). The
internal consistency of the scale was good (␣ ⫽ .77). Following past
research (Parks & Hulbert, 1995), a median split (Mdn ⫽ 4.63) was
performed to classify participants as high or low in trust, yielding 35 to 40
participants per condition. Ratings on the trust scale did not differ across
emotion conditions (disappointment: M ⫽ 4.68, SD ⫽ 0.61; guilt: M ⫽
4.65, SD ⫽ 0.63), F(1, 146) ⬍ 1, ns. After the assessment of trust,
participants completed a 10-min filler task (consisting of a number of
unrelated scales) to reduce the likelihood of carryover effects from the trust
assessment to the negotiation task. After the filler task, participants pro-
ceeded to the negotiation task, which was identical to the one used in
Experiment 1.
Dependent measures. In addition to recording demands, we measured
participants’ goals, their interpretation of the other’s emotions, suspicion
regarding the trustworthiness of the opponent and his or her emotions, and
discounting of the other’s emotions. We used the same manipulation
checks as in Experiment 1 (disappointment: ␣ ⫽ .85; guilt: ␣ ⫽ .88).
Participants’ goals were measured with six items, two for each issue
(“On which level of [price/warranty/service] do you strive to reach an
agreement?” and “On which level of [price/warranty/service] do you hope
to reach an agreement?”). Responses could range from 1 (indicating an
extremely low goal) to 9 (indicating an extremely high goal; see Appendix
A). The items were averaged into a single index of participant’s goal (␣ ⫽
.79).
Interpretation of the opponent’s emotion was assessed by two semantic
differentials, both of which were introduced by the question, “How do you
interpret the emotions the buyer expressed during the negotiation?” The
two items participants responded to were “The buyer has received too
much/the buyer has received too little” and “The buyer has offered me too
much/the buyer has offered me too little” (both measured on a 7-point
scale). Because the correlation between the items was modest (r ⫽ ⫺.39),
we analyzed them separately.
Suspicion was measured with three items (“During the negotiation I was
suspicious”; “I distrusted the information I received from the buyer”; “The
information I received from the buyer made me suspicious”; 1 ⫽ totally
disagree, 7 ⫽ totally agree), which were combined into a single index
(␣ ⫽ .79). Discounting of the opponent’s emotion was also measured by
three items (“During the negotiation I did not take the information about
the buyer into account”; “During the negotiation, I paid serious attention to
the information I received about the buyer,” reverse scored; “I ignored the
information I received from the buyer”; 1 ⫽ totally disagree, 7 ⫽ totally
agree), which were also averaged into a single index (␣ ⫽ .67).
Results
Manipulation Check
A 2 (emotion of the opponent: disappointment vs. guilt) ⫻ 2
(dispositional trust: high vs. low) ⫻ 2 (perception of the oppo-
nent’s emotion: disappointed vs. guilty) ANOVA with repeated
measures on the last factor revealed a significant interaction be-
tween the opponent’s emotion and participants’ perceptions of the
opponent’s emotion, F(1, 144) ⫽ 300.54, p ⬍ .01 (partial ␩2 ⫽
.67). Participants with a disappointed opponent rated the opponent
as more disappointed (M ⫽ 5.25, SD ⫽ 1.42) than did those with
a guilty opponent (M ⫽ 3.29, SD ⫽ 1.16). Similarly, participants
with a guilty opponent rated the other as more guilty (M ⫽ 4.56,
SD ⫽ 1.54) than did those with a disappointed opponent (M ⫽
2.21, SD ⫽ 0.76). Paired-sample t tests showed that participants in
the disappointed-opponent condition rated the opponent as more
disappointed than guilty (M ⫽ 5.25 vs. M ⫽ 2.21), t(72) ⫽ 15.84,
p ⬍ .01, and that participants in the guilty-opponent condition
rated the opponent as more guilty than disappointed (M ⫽ 4.56 vs.
M ⫽ 3.29), t(74) ⫽ 7.79, p ⬍ .01. There were no effects of trust.
Demand Level
Demands in Rounds 1 to 6 were submitted to a 2 (opponent’s
emotion: disappointment vs. guilt) ⫻ 2 (participant’s trust: high vs.
low) mixed-model ANOVA with the opponent’s emotion and the
participant’s trust as between-participants variables and demands
in Rounds 1 to 6 as a repeated-measures variable. As in Experi-
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
ment 1, we first describe lower order effects, and we then turn to
the hypothesized interaction. The anticipated effect of round was
significant, F(5, 720) ⫽ 413.82, p ⬍ .01 (partial ␩2 ⫽ .74),
showing that demands declined from Round 1 (M ⫽ 643, SD ⫽
77) to Round 6 (M ⫽ 466, SD ⫽ 91). Results further revealed a
main effect of the opponent’s emotion on average demand, F(1,
144) ⫽ 10.30, p ⬍ .002 (partial ␩2 ⫽ .07), indicating that partic-
ipants’ demands were influenced by their adversaries’ emotions
(disappointment: M ⫽ 521, SD ⫽ 75; guilt: M ⫽ 560, SD ⫽ 84).
These effects were qualified by an Emotion ⫻ Round interaction,
F(5, 720) ⫽ 8.48, p ⬍ .01 (partial ␩2 ⫽ .06), which showed that
the effect of the opponent’s emotion on participants’ demands
increased from Round 1 (disappointment: M ⫽ 641, SD ⫽ 77;
guilt: M ⫽ 646, SD ⫽ 78) to Round 6 (disappointment: M ⫽ 439,
SD ⫽ 80; guilt: M ⫽ 491, SD ⫽ 93).
In line with Hypothesis 4, there was a significant interaction
between opponent’s emotion and participant’s trust, F(1, 144) ⫽
7.67, p ⬍ .01 (partial ␩2 ⫽ .05). As predicted, simple effects
analysis revealed that high trust participants made higher demands
to a guilty opponent (M ⫽ 586, SD ⫽ 66) than to a disappointed
one (M ⫽ 510, SD ⫽ 66), F(1, 144) ⫽ 16.96, p ⬍ .01 (partial ␩2 ⫽
.26), whereas those low in trust did not respond differentially to
their counterpart’s emotions (guilt: M ⫽ 537, SD ⫽ 92; disap-
pointment: M ⫽ 531, SD ⫽ 82), F(1, 144) ⬍ 1, ns. This interaction
was qualified by a significant three-way interaction between emo-
tion, trust, and round, F(5, 720) ⫽ 2.59, p ⬍ .03 (partial ␩2 ⫽ .08),
indicating that the interactive effects of emotion and trust became
more apparent as the negotiation progressed (see Figure 2). Simple
effects analysis revealed a significant Emotion ⫻ Round interac-
tion for high trusters, F(5, 720) ⫽ 8.07, p ⬍ .01 (partial ␩2 ⫽ .16),
but not for low trusters, F(5, 720) ⫽ 1.74, ns. As can be seen from
Figure 2, participants with high levels of trust made larger con-
cessions in the course of the negotiation when the opponent
experienced disappointment than when the opponent experienced
guilt, whereas participants with low levels of trust were not re-
sponsive to the opponent’s emotional state.
Interpretation of the Opponent’s Emotions
ANOVAs yielded significant main effects of emotion on the two
items tapping the participant’s interpretation of the opponent’s
Figure 2. Demand level as a function of the opponent’s emotion, participants’ trust, and negotiation round in
Experiment 2.
133
emotion. Participants who negotiated with a guilty opponent in-
terpreted the other’s emotion as signifying that the other had
obtained too much (M ⫽ 3.71, SD ⫽ 0.97), whereas those who
dealt with a disappointed opponent took the other’s emotion to
mean that the other had received too little (M ⫽ 4.47, SD ⫽ 1.17),
F(1, 144) ⫽ 19.33, p ⬍ .01 (partial ␩2 ⫽ .11). Conversely,
participants with a guilty opponent interpreted the other’s emotion
as a signal that the other had offered too little (M ⫽ 4.88, SD ⫽
1.09), whereas those with a disappointed partner were more likely
to believe that the other’s emotion revealed that he or she had
offered too much (M ⫽ 4.49, SD ⫽ 1.04), F(1, 144) ⫽ 5.28, p ⬍
.025 (partial ␩2 ⫽ .04). There was no significant effect of trust and
no interaction.
Suspicion and Discounting of the Opponent’s Emotions
According to Hypothesis 5, low trusters would be more suspi-
cious and more likely than high trusters to discount their counter-
part’s emotion. In line with this prediction, an ANOVA showed a
tendency for participants low in trust to be more suspicious (M ⫽
3.92, SD ⫽ 1.20) regarding the opponent’s emotions than were
participants high in trust (M ⫽ 3.50, SD ⫽ 1.31), F(1, 144) ⫽ 2.79,
p ⬍ .10 (partial ␩2 ⫽ .02). Low trusters were also more likely to
discount the opponent’s emotion than were high trusters (M ⫽
3.14, SD ⫽ 1.23 vs. M ⫽ 2.71, SD ⫽ 0.99), F(1, 144) ⫽ 5.14, p ⬍
.025 (partial ␩2 ⫽ .04). There were no effects of emotion and no
interaction (Fs ⬍ 1, ns).
Participants’ Goals
In line with Hypothesis 6, participants’ goals were influenced by
the opponent’s emotion. Participants who dealt with a guilty op-
ponent reported higher goals (M ⫽ 5.04, SD ⫽ 0.79) than did those
who dealt with a disappointed opponent (M ⫽ 5.38, SD ⫽ 0.79),
F(1, 144) ⫽ 7.73, p ⬍ .01, partial ␩2 ⫽ .05 (recall that low
numbers correspond to high goals; see Appendix A). This effect
was moderated by the participant’s level of trust, F(1, 144) ⫽ 4.63,
p ⬍ .04 (partial ␩2 ⫽ .03). Means and standard deviations per-
taining to this interaction are shown in Table 2. Simple effects
analyses revealed that high trusters adapted their goals to the
134 VAN KLEEF, DE DREU, AND MANSTEAD

Table 2
Participants’ Goals and Average Demands as a Function of the Opponent’s Emotion and the
Participant’s Trust (Experiment 2)

Participant’s trust

Low High

Condition Disappointment Guilt Disappointment Guilt

Goal 4.88a (0.86) 4.81a (0.70) 5.07a (0.67) 4.45b (0.82)

Demand 531a (82) 537a (92) 510a (66) 586b (66)

Note. Participants’ goals ranged from 1 (indicating a high goal) to 9 (indicating a low goal). Demand refers to
participants’ average demands in Rounds 1 to 6. Standard deviations are in parentheses. Means within the low
and high trust conditions with a different subscript differ at p ⬍ .05 according to simple-effects analyses.

opponent’s emotion, F(1, 144) ⫽ 11.54, p ⬍ .01 (partial ␩2 ⫽ .15),
whereas low trusters did not, F(1, 144) ⬍ 1, ns.
Mediation Analysis
To investigate whether the interactive effects of the opponent’s
emotion and the participant’s trust on demands were mediated by
the participant’s goals (as predicted in Hypothesis 7), we per-
formed mediated regression analyses, following Baron and Ken-
ny’s (1986) procedure. In Step 1 we entered emotion, trust, and
their interaction as the independent variables and demands as the
dependent variable. This produced a significant main effect of
emotion (␤ ⫽ .24, p ⬍ .003) and an interaction between emotion
and trust (␤ ⫽ .22, p ⬍ .01). In Step 2, we used the same
independent variables to predict the participant’s goals. This, too,
yielded a significant main effect of emotion (␤ ⫽ .21, p ⬍ .01) and
a significant interaction (␤ ⫽ .17, p ⬍ .04). In Step 3, we
simultaneously entered emotion and goals to predict demands.
This yielded a significant effect of goals on demands (␤ ⫽ .61, p ⬍
.01) and reduced the formerly significant Emotion ⫻ Trust inter-
action to nonsignificance (␤ ⫽ .11, ns). A Sobel test indicated that
the reduction of the direct path from the Emotion ⫻ Trust inter-
action to demands was significant (Z ⫽ 2.10, p ⬍ .04; see Kenny
et al., 1998). In support of Hypothesis 7, these results show that the
interactive effect of the opponent’s emotion and the participant’s
trust on demands is fully mediated by the participant’s goals.
Discussion
The results of Experiment 2 corroborate our hypotheses. As
predicted, negotiators’ reactions to their opponent’s emotions are
moderated by trust. Negotiators with high levels of dispositional
trust responded with high demands to an opponent who appeared
to experience guilt and with low demands to an opponent who
appeared to experience disappointment. By contrast, negotiators
with low trust did not respond differentially to the opponent’s
emotions. These findings replicate and extend those of Experiment
1 by showing that negotiators are more likely to act on their
opponent’s emotions to the extent that they trust the opponent and
see his or her emotions as trustworthy and reliable. We further
found that individuals with low trust were more suspicious regard-
ing the trustworthiness of the other’s emotions and were more
likely to discount them rather than take them into account when
designing their negotiation strategy. Furthermore, Experiment 2
sheds some light on the processes underlying the effects of guilt
and disappointment on demands. A mediation analysis revealed
that negotiators with high levels of trust made smaller demands to
a disappointed opponent than to a guilty one because the other’s
disappointment led them to lower their goals, whereas the other’s
guilt led them to raise their goals. Finally, results pertaining to
participants’ interpretation of their counterparts’ emotions are
compatible with the framework that was outlined in the introduc-
tion: Guilt is interpreted to mean that the other has claimed too
much, whereas disappointment is taken as a signal that the other
has received too little.
The fact that trust was measured rather than manipulated might
be regarded as a limitation of Experiment 2. Although the results
are consistent with our predictions, we cannot rule out the possi-
bility that they are caused by some unknown third variable that
was not taken into account. Another shortcoming of Experiment 2
is that there was no nonemotional control condition. Therefore, we
cannot draw firm conclusions as to whether disappointment led
participants to place lower demands, whether guilt led them to
place higher demands, or both. Although Experiment 1 demon-
strated that guilt and disappointment elicited higher and lower
demands, respectively, compared with a nonemotional control
condition, we felt that a replication of this effect would increase
confidence in its robustness. We therefore conducted a third ex-
periment in which trust was manipulated (through expectations
about the opponent’s personality) and a control condition was
included.
Experiment 3
Experiment 2 showed that individual differences in dispositional
trust moderate the interpersonal effects of supplication and ap-
peasement emotions in negotiations. As noted earlier, variations in
trust may also arise from features of the situation and/or the other
party (Yamagishi, 1986). Among other things, trust depends on a
negotiator’s expectations about the other’s cooperation or compe-
tition. Indeed, trust has been defined as the expectation that the
other will cooperate (Pruitt & Kimmel, 1977). Because trusting
another person is more risky to the extent that he or she can be
expected to take advantage of and exploit the trust (Ross &
LaCroix, 1996), people exhibit greater levels of trust in relation to
others whom they expect to be cooperative than others whom they
expect to be competitive (De Cremer, Snyder, & Dewitte, 2001;
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
De Dreu et al., 1998; Deutsch, 1960; Kee & Knox, 1970; Loomis,
1959; Pruitt & Kimmel, 1977; Rempel, Holmes, & Zanna, 1985;
for an overview of bases of trust in negotiation, see Ross &
LaCroix, 1996). On the basis of this research and the results of
Experiment 2, we predicted that negotiators would make higher
demands to a guilty opponent than to a disappointed one, but only
if they believed that the opponent was cooperative. If the opponent
was expected to be competitive, the trustworthiness of his or her
emotions should be decreased, and hence negotiators should not
adapt their demands to the other’s emotion (Hypothesis 8).
Method
Participants and Experimental Design
Ninety-four male and female undergraduate students at the University of
Amsterdam participated for course credit or monetary compensation (EUR
7, approximately US $8). The experimental design included the opponent’s
emotion (disappointment vs. guilt vs. no emotion) and the opponent’s
personality (cooperative vs. competitive) as the independent variables and
demand level as the main dependent variable. Participants were randomly
assigned to conditions, and the experimenters were blind to this
assignment.
Procedure
The procedure was the same as in Experiment 2, except that trust was
manipulated rather than measured. As in the preceding experiments, par-
ticipants who reached agreement before Round 6 (n ⫽ 2) were excluded
from the analyses (although including these participants in the analyses
yielded a similar pattern of findings).
Manipulation of trust. Trust was manipulated by varying participants’
expectations regarding the opponent’s cooperation versus competition with
a procedure that has been used successfully in past research (see De Dreu
& Van Kleef, 2004; Steinel & De Dreu, 2004; Van Kleef & De Dreu,
2002). At the beginning of the experiment, participants were asked to
complete a (fake) “Personality Test.” This questionnaire was described as
measuring collaboration skills and contained 20 items having to do with
cooperation in daily life (e.g., “In the bus, I vacate my seat for older
people”; “I enjoy working with other people”; “Love and respect are more
important than status and money”; “Winning is everything”; “I like situa-
tions in which it is me against someone else”). Participants were asked to
indicate their agreement on 5-point Likert scales (1 ⫽ strongly disagree to
5 ⫽ strongly agree).
After completion of this personality test, participants received the same
instructions as in Experiments 1 and 2. They then learned that on the basis
of the personality test, each participant had been classified as either
“cooperative” or “competitive” and that some participants would receive
this information about their opponent, whereas others would not. Next,
participants read that the computer had selected them to receive their
opponent’s personality information, but that the opponent would not re-
ceive their information. In the cooperative opponent condition, the out-
come of the opponent’s personality test was presented on the screen with
answers allegedly given by the opponent suggesting that he or she was very
cooperative. These answers were accompanied by a “general test result,”
which indicated that the other could best be classified as cooperative. In the
competitive opponent condition, participants were shown answers suggest-
ing that the opponent was very competitive, and the general test result
showed that the other could best be classified as competitive.
Dependent measures. The emotion check was the same as in Experi-
ment 2. To check participants’ expectations regarding the opponent’s
cooperative versus competitive orientation, we asked participants to rate
the opponent’s personality on five 7-point semantic differential scales (e.g.,
“cooperative– competitive,” “compliant– bossy,” “dominant– docile,” re-
135
verse scored). These items were averaged into a composite index of
judgment of the opponent’s cooperativeness versus competitiveness, rang-
ing from 1 ⫽ cooperative to 7 ⫽ competitive (␣ ⫽ .90). Additionally, we
assessed participants’ perceptions of the opponent’s trustworthiness using
three items (“The buyer is very trustworthy–not very trustworthy,” reverse
scored; “The buyer is very unreliable–very reliable”; “The buyer is very
honest–very dishonest”). These items were combined into an index of the
opponent’s trustworthiness, ranging from 1 ⫽ not very trustworthy to 7 ⫽
very trustworthy (␣ ⫽ .84).
Results
Manipulation Checks
Opponent’s emotion. As in the previous experiments, an
ANOVA showed the predicted interaction between the opponent’s
emotion and the participants’ perception of the opponent’s emo-
tion, F(2, 86) ⫽ 168.30, p ⬍ .01 (partial ␩2 ⫽ .80). Participants in
the disappointed opponent condition rated the opponent as signif-
icantly more disappointed (M ⫽ 6.00, SD ⫽ 0.93) than did
participants in the guilty or nonemotional opponent conditions
(M ⫽ 3.91, SD ⫽ 1.16; M ⫽ 2.81, SD ⫽ 1.00). Similarly,
participants with a guilty opponent rated the opponent as more
guilty (M ⫽ 5.67, SD ⫽ 0.85) than did participants in the other two
conditions (disappointment: M ⫽ 2.22, SD ⫽ 1.10; no emotion:
M ⫽ 2.57, SD ⫽ 0.89). Furthermore, paired-samples t tests showed
that ratings within the emotion conditions were higher for the
intended emotion than for the other emotion (8.13 ⬍ ts ⬍ 15.82;
both ps ⬍ .01). The ratings did not differ in the control condition.
Opponent’s cooperation/competition and trustworthiness. An
ANOVA revealed a significant effect of the manipulation of the
opponent’s orientation on participants’ judgments of the oppo-
nent’s cooperativeness versus competitiveness. Participants in the
cooperative opponent conditions rated the opponent as signifi-
cantly more cooperative (M ⫽ 3.67, SD ⫽ 1.08) than did those in
the competitive opponent conditions (M ⫽ 4.77, SD ⫽ 1.29), F(1,
86) ⫽ 19.63, p ⬍ .01 (partial ␩2 ⫽ .19). There was no main effect
of the opponent’s emotion and no interaction. We also obtained a
significant main effect of the opponent’s orientation on percep-
tions of the opponent’s trustworthiness, F(1, 86) ⫽ 18.69, p ⬍ .01
(partial ␩2 ⫽ .18). Participants who learned that the opponent had
a cooperative orientation judged the other as more trustworthy
(M ⫽ 5.14, SD ⫽ 0.98) than did those who learned that the other
had a competitive orientation (M ⫽ 4.17, SD ⫽ 1.23).
Demand Level
Demands were submitted to a 3 (opponent’s emotion: disap-
pointment vs. guilt vs. no emotion) ⫻ 2 (opponent’s personality:
cooperative vs. competitive) mixed-model ANOVA with the op-
ponent’s emotion and personality as between-participants vari-
ables and demands in Rounds 1 to 6 as a repeated-measures
variable. A main effect of round indicated that participants’ de-
mands declined over time (Round 1: M ⫽ 643, SD ⫽ 89; Round
6: M ⫽ 457, SD ⫽ 103), F(5, 430) ⫽ 143.22, p ⬍ .01 (partial ␩2 ⫽
.63). Furthermore, we found a main effect of emotion on average
demands, F(2, 86) ⫽ 4.41, p ⬍ .02 (partial ␩2 ⫽ .10). Planned
comparison results were consistent with those of the previous
experiments, indicating that participants with a disappointed op-
ponent made lower demands (M ⫽ 505, SD ⫽ 88) than did
136 VAN KLEEF, DE DREU, AND MANSTEAD
participants with a guilty opponent (M ⫽ 556, SD ⫽ 94), t(89) ⫽
1.88, p ⬍ .05 (partial ␩2 ⫽ .05). Participants with a nonemotional
opponent occupied an intermediate position (M ⫽ 541, SD ⫽ 70)
that did not differ significantly from the other two conditions (ts ⬍
1.50, ns).
We also found a significant interaction between the opponent’s
emotion and the opponent’s orientation, F(2, 86) ⫽ 4.74, p ⬍ .01
(partial ␩2 ⫽ .10). In line with our expectations, simple effects
analysis revealed a significant effect of the opponent’s emotion in
the cooperative opponent condition, indicating that average de-
mands were lower for participants who negotiated with a disap-
pointed opponent (M ⫽ 475, SD ⫽ 52) than for those who dealt
with a guilty or nonemotional opponent (M ⫽ 603, SD ⫽ 81 and
M ⫽ 558, SD ⫽ 77, respectively), F(2, 86) ⫽ 7.73, p ⬍ .01 (partial
␩2 ⫽ .37). In the competitive opponent condition, by contrast,
there was no effect of the opponent’s emotion on demands (523 ⬍
M ⬍ 525), F(2, 86) ⫽ 0.01, ns.
Finally, we found a significant three-way interaction between
opponent’s emotion, opponent’s orientation, and negotiation
round, F(10, 430) ⫽ 3.84, p ⬍ .025, partial ␩2 ⫽ .08 (see Figure
3). Results of simple effects analysis were consistent with Hypoth-
esis 8, revealing a significant Emotion ⫻ Round interaction in the
cooperative opponent condition, F(10, 430) ⫽ 5.62, p ⬍ .01
(partial ␩2 ⫽ .22), but not in the competitive opponent condition,
F(10, 430) ⫽ .31, ns. Planned comparisons of demands in Round
6 revealed that participants with a cooperative opponent made
higher demands when the opponent appeared to feel guilty than
when the opponent expressed no emotion (M ⫽ 553, SD ⫽ 130 vs.
M ⫽ 472, SD ⫽ 69), t(39) ⫽ 2.24, p ⬍ .03 (partial ␩2 ⫽ .06), and
that they made lower demands when the other appeared to be
disappointed (M ⫽ 393, SD ⫽ 59), t(39) ⫽ 3.06, p ⬍ .01 (partial
␩2 ⫽ .08). There were no differences in the competitive opponent
condition (439 ⬍ Ms ⬍ 449), ts ⬍ 1.0, ns.
Discussion
The results of Experiment 3 replicate those of Experiment 2
using a situational manipulation of trust. Participants who thought
that the opponent had a cooperative orientation (i.e., high trust)
conceded more to a disappointed opponent than to a nonemotional
one, and they tended to concede less to a guilty opponent. By
contrast, participants who believed the other to be competitive
Figure 3. Demand level as a function of the opponent’s emotion, the opponent’s orientation, and negotiation
round in Experiment 3.
(i.e., low trust) did not respond differentially to the other’s emo-
tions. Thus, the results of this experiment are consistent with the
findings reported in Experiment 2, and this suggests that trust is an
important prerequisite for the interpersonal effects of disappoint-
ment and guilt to obtain. Further, the inclusion of a nonemotional
control condition allows for specific conclusions regarding the
respective effects of disappointment and guilt. In line with the
results of Experiment 1, the present data show that participants
who expected cooperation made significantly smaller demands to
a disappointed opponent than to a nonemotional one, whereas they
made significantly larger demands to a guilty counterpart.
General Discussion
The results of the present experiments support our predictions.
Negotiators whose opponents appeared to experience emotions of
appeasement (i.e., guilt or regret) developed a positive impression
of their opponents, but they were nonconciliatory in the level of
their demands. By contrast, participants whose opponents experi-
enced emotions of supplication (i.e., disappointment or worry)
rated their opponents less positively, but they made larger conces-
sions in the course of the negotiation (Experiment 1). Experiment
1b showed that negotiators with a guilty opponent expected to
receive larger concessions from the other than did those with a
nonemotional opponent. Also in line with our theoretical frame-
work were results showing that the interpersonal effects of guilt
and disappointment on demands were mediated by the focal ne-
gotiator’s goals. Negotiators with a disappointed opponent lowered
their goals and made smaller demands, whereas those with a guilty
opponent raised their goals and made larger demands. The data
further showed that participants with a guilty opponent believed
that the other had claimed too much and offered too little, whereas
those with a disappointed counterpart believed that the other had
received too little and offered too much. This, too, is consistent
with our theoretical framework. Apparently, emotions of suppli-
cation signal that one is in need of compensation, which may lead
others to lower their goals and make concessions. Conversely,
emotions of appeasement appear to signal that one is willing to
compensate one’s counterpart, which may lead others to increase
their goals and stand firm.
The findings of Experiments 1 and 2 point to the pervasive
effects of emotions related to supplication and appeasement on
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
negotiation behavior and impressions, and the results of these
experiments shed some light on the processes underlying these
effects. In addition, the present research has identified an impor-
tant moderator of the interpersonal effects of appeasement and
supplication emotions on demands and concessions: interpersonal
trust. Experiment 2 showed that only negotiators with high levels
of trust adapted their goals and demands to their opponent’s
emotion. Participants with low trust reported more suspicion re-
garding the opponent’s emotions, were more likely to discount
them, and did not adapt their goals and demands to the other’s
emotion. In Experiment 3, trust was manipulated by varying par-
ticipants’ expectations regarding the opponent’s cooperative ver-
sus competitive orientation, and a similar effect was found. Par-
ticipants who expected a cooperative opponent had high trust and
made larger concessions to a disappointed opponent and smaller
concessions to a guilty opponent. By contrast, participants who
thought that the opponent had a competitive orientation had low
trust and did not respond differentially to the opponent’s disap-
pointment versus guilt in terms of their demands and concessions.
Together, the results of these experiments strongly support the
idea that emotions of appeasement and supplication have the
potential to influence negotiation behavior at the interpersonal
level. This conclusion has interesting implications for research on
conflict and negotiation and for our understanding of the social
consequences of emotions. Below we discuss these implications as
well as address some of the strengths and limitations of our
approach. We conclude by outlining some avenues for future
research.
Implications and Contributions
In exploring the interpersonal effects of appeasement and sup-
plication emotions in negotiations, the present work makes a
number of important contributions. First, prior research on emo-
tions in negotiation has focused almost exclusively on the role of
general positive versus negative affect or, in a few cases, discrete
emotions such as anger, happiness, and compassion (Allred et al.,
1997; Van Kleef et al., 2004a, b). The present research extends this
line of inquiry by examining the effects of guilt, regret, disappoint-
ment, and worry. Our findings demonstrate that these emotions,
like anger and happiness, can have a powerful impact on negoti-
ation behavior. This underscores the recent acknowledgment of the
importance of considering affective phenomena in conflict and
negotiation and indicates that scholarly attention should not remain
limited to the role of anger and happiness.
Second, the current research indicates that the role of emotion in
negotiation cannot be understood by merely classifying emotions
as positive or negative. Previous research has explored the inter-
personal effects of anger and happiness in negotiations, demon-
strating that negotiators concede more to angry opponents than to
happy ones (Van Kleef et al., 2004a, b). Although it is tempting to
explain this finding in terms of a positivity–negativity dimension,
the present results suggest that it would be unwise to do so. The
emotions that were investigated in this research— guilt, regret,
worry, and disappointment—are all negative in valence, yet they
have quite different effects on behavior: The effects of guilt and
regret were opposite to those of worry and disappointment. A more
fruitful approach, then, is to adopt a social–functional perspective
on emotion, which assumes that emotions have distinct social
137
functions and consequences (Frijda & Mesquita, 1994; Keltner &
Haidt, 1999; Morris & Keltner, 2000). This conclusion points to
the need for more research on the effects of discrete emotions
rather than nonspecific positive versus negative affect.
Third, the present findings contribute to a more thorough un-
derstanding of the social consequences of expressing guilt, regret,
disappointment, and worry. Although all of these emotions have
been the subject of research, most of this research has focused on
the intrapersonal effects of these emotions on the individual’s
cognitions and behavior. For example, research on guilt has ad-
dressed the question of how the experience of guilt influences the
guilty party’s behavior and thereby his or her relationship to the
interaction partner. Thus it has been shown that guilt motivates
people to make apologies and amends and to compensate the other
for one’s transgression (see Baumeister et al., 1994, for an over-
view of this research). Although this research contains an inter-
personal component (guilt may affect interpersonal relations), the
major focus has been on the impact of guilt on the individual’s
own motivations and behavior. Our findings move beyond the
intrapersonal effects of guilt by showing that interaction partners
anticipate compensation from the guilty party by making high
demands and small concessions. In a similar vein, previous re-
search on regret has established that regret instigates a desire to
undo one’s actions (Gilovich & Medvec, 1994, 1995; Zeelenberg
et al., 2000). The present research suggests that regret may not
only affect one’s own behavior but also that of others, although the
effects of regret appear to be weaker than those of guilt. A possible
explanation for this lies in the ambiguous nature of regret. An
opponent who regrets his or her behavior may do so because he or
she has harmed the other or because he or she should have been
more strategic and self-interested.
The present findings also increase our understanding of the
interpersonal effects of worry and disappointment. As is the case
with guilt and regret, most previous research on disappointment
has adopted an intrapersonal approach, investigating for instance
how the experience of disappointment motivates people to mini-
mize future disappointment (see Zeelenberg et al., 2000, for an
overview). Our research shows that disappointment can also in-
fluence behavior at the interpersonal level—it appears to be effec-
tive in eliciting concessions. The same holds for expressions of
worry. These findings are consistent with prior research on
distress-related emotions such as sadness, which have been shown
to facilitate prosocial behavior (Barnett et al., 1979; Batson, 1987;
Eisenberg, Fabes, Miller, et al., 1989; Fabes et al., 1994; Morris &
Keltner, 2000). Apparently, various emotions related to distress
and supplication have broadly comparable effects on others’
behavior.
Altogether, it appears that discrete emotions have distinct and
predictable effects in negotiations, which can be conceptualized
and understood in terms of the information they provide. For
example, guilt (and to a lesser degree regret) informs the adversary
that one has taken too much, and it signals that one is willing to
compensate for this. Disappointment and worry, on the other hand,
inform the other that one has received less than expected and
signal that one is in need of compensation. Because the informa-
tion conveyed by these and other emotions is similar across situ-
ations, we believe that parallel effects are to be expected in other
domains of social interaction.
138 VAN KLEEF, DE DREU, AND MANSTEAD
The identification of trust as a moderator of the interpersonal
effects of supplication and appeasement emotions on negotiation
behavior constitutes a fourth contribution of the present work. Two
different operationalizations of trust were used. In Experiment 2
we measured individual differences in trust; in Experiment 3 we
manipulated trust by providing information about the opponent’s
personality. Although both methods have been shown in prior
research to constitute valid operationalizations of trust (De Dreu &
Van Kleef, 2004; Steinel & De Dreu, 2004; Yamagishi, 1986),
they differ in a number of respects. Measures of dispositional trust
tap directly into the generalized tendency of individuals to trust
others, to believe in their honesty, and to accept what they say and
do as true and righteous (or not). By contrast, the method that was
used in Experiment 3 varied levels of state trust by manipulating
characteristics of the opponent, rather than assessing a character-
istic of the focal individual. Despite these differences, the two
methods similarly affected participants’ perceptions of the trust-
worthiness of the opponent and of his or her emotional expres-
sions, and both meaningfully moderated the interpersonal effects
of the other’s emotions.
Previous research has demonstrated the important role of infor-
mation processing motivation in determining the interpersonal
effects of anger and happiness on demands and concessions (Van
Kleef et al., 2004b). Negotiators concede more to angry opponents
than to happy ones, but only if they are sufficiently motivated to
consider the emotions of the other and to think about their impli-
cations for their own goal attainment. The present research shows
that trust has a similar moderating effect on the interpersonal
effects of supplication and appeasement emotions. Negotiators
tend to give in when the opponent experiences emotions of sup-
plication, but they stand firm when the opponent experiences
emotions of appeasement. However, this holds only for negotiators
with high levels of trust; those with low trust do not adapt their
demands to the other’s emotion.
The current findings point to an interesting dilemma facing
negotiators who anticipate future interaction. Expressions of dis-
appointment or worry (supplication) can elicit concessions from
others, but they may also contribute to a negative impression. By
contrast, expressions of guilt and regret (appeasement) may serve
to engender a more positive impression, but they may lead others
to stand firm and resist concession making. Thus, on the one hand,
negotiators may be motivated to express guilt or regret strategi-
cally to make a good impression and to induce or maintain a
positive interpersonal relationship, but this would be at the expense
of their personal negotiation outcomes. On the other hand, they
may choose to display disappointment or worry to get their oppo-
nents to comply with their wishes, thereby inadvertently spoiling
the interpersonal climate.
Limitations and Suggestions for Future Research
There are some limitations to our findings. First, there was no
face-to-face interaction. The primary purpose of this research was
to enhance our knowledge and understanding of the interpersonal
effects of discrete emotions in negotiation by generating and
testing new hypotheses about the effects of supplication and ap-
peasement emotions. In doing this, we made an explicit decision to
maintain as much experimental control as possible, and we chose
to use a computer-mediated negotiation paradigm to permit a
carefully controlled manipulation of the opponent’s emotion. As a
result, some caution is needed when generalizing the results. At the
very least, our findings pertain to computer-mediated negotiations.
Given the pervasiveness of negotiation as a form of social inter-
action and the increasing popularity of information technologies as
a communication medium, the question of how individuals react to
each other’s emotions in computer-mediated communication is
itself of great theoretical and practical importance (McGrath &
Hollingshead, 1994; McKersie & Fonstad, 1997; Moore,
Kurtzberg, Thompson, & Morris, 1999). However, considering
that our paradigm has previously yielded results that have also
been found in face-to-face settings (Sinaceur & Tiedens, in press)
we have no reason to suspect that our findings are restricted to the
domain of computer-mediated interaction. Future research could
shed more light on this issue by investigating the extent to which
the interpersonal effects of emotions relating to supplication and
appeasement generalize across settings.
Another issue concerns the “cognitive” nature of the emotion
manipulation that was used in the present experiments. The fact
that we used verbal manipulations of emotion raises the question
of whether our findings generalize to settings in which emotions
are communicated in a different manner (e.g., nonverbally). One
could argue that the effects would be different if people were
presented with behavioral rather than cognitive emotion cues. This
possibility cannot be ruled out on the basis of the present data.
However, previous research on anger and happiness in negotia-
tions has documented similar effects regardless of whether a verbal
(Van Kleef et al., 2004a, b) or nonverbal (Sinaceur & Tiedens, in
press) manipulation was used. We therefore have no reason to
doubt the generalizability of our findings. However, future re-
search is needed to explore this issue in greater depth.
A final issue concerns the way in which the information about
the opponent’s emotions and personality was presented. We told
participants that the computer had randomly determined that they
would receive information about the intentions or personality of
their counterpart, but that the other would not receive such infor-
mation about them. A possible downside of this procedure is that
it created an asymmetric situation in that participants believed that
they had more information about the opponent than vice versa. For
one thing, knowledge of this fact may have given participants a
sense of power over the opponent. Given that previous research
has found that high-power negotiators are less susceptible to their
counterpart’s emotions than are low-power negotiators (Sinaceur
& Tiedens, in press; Van Kleef, De Dreu, Pietroni, & Manstead, in
press), this asymmetry is likely to have contributed to a more
conservative test of our hypotheses. However, we cannot rule out
the possibility that the informational asymmetry may have influ-
enced our findings in one way or another. Future research might
explore possible differences in responses to the other’s emotions as
a function of the distribution of information.
After a decade of research on the intrapersonal effects of moods
and, occasionally, of emotions on the negotiator’s cognition and
behavior, there has been a recent upsurge of interest in the social
effects of discrete emotions in negotiations. This research has
begun to document the interpersonal effects of emotions relevant
to negotiation and conflict resolution. Although the results are
promising, many questions remain unanswered. How does the
expectation of future interaction with the same partner influence
the interpersonal effects of different emotions on negotiation be-
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
havior? Are there other variables, besides trust and information
processing, that moderate these effects? What happens when more
than one negotiator expresses certain emotions? What factors
determine whose emotions will have the bigger impact? How long
can a negotiator effectively continue to use anger or disappoint-
ment as a means of eliciting concessions? Do the effects of these
emotions on demands wear off or even reverse in the long run?
The investigation of these and other questions will allow the study
of emotion in social conflict to continue to advance.
Conclusion
The present research investigated the interpersonal effects of
guilt, regret, disappointment, and worry in negotiations. The re-
sults showed that negotiators tend to make larger concessions to
opponents who experience disappointment or worry (supplication
emotions) than to nonemotional opponents, whereas they make
smaller concessions to adversaries who experience guilt or regret
(appeasement emotions). This effect was shown to be mediated by
the focal negotiator’s goals (negotiators with a guilty opponent
adopted higher goals than did those with a disappointed opponent)
and moderated by interpersonal trust—negotiators with high levels
of trust adapted their demands to their counterpart’s emotion, but
those with low trust did not. These findings point to the pervasive
interpersonal effects of emotions in negotiations, and they stress
the need for more research on the role of emotion in conflict and
negotiation. Such research promises to enhance our understanding
of the negotiation process, of the factors that facilitate or hinder
constructive conflict resolution, and of the social consequences of
emotions in general.
References
Adler, R. S., Rosen, B., & Silverstein, E. M. (1998). Emotions in negoti-
ation: How to manage fear and anger. Negotiation Journal, 14, 161–180.
Allred, K. G., Mallozzi, J. S., Matsui, F., & Raia, C. P. (1997). The
influence of anger and compassion on negotiation performance. Orga-
nizational Behavior and Human Decision Processes, 70, 175–187.
Barnett, M. A., Howard, J. A., Melton, E. M., & Dino, G. A. (1982). Effect
of inducing sadness about self or other on helping behavior in high- and
low-empathic children. Child Development, 53, 920 –923.
Barnett, M. A., King, L. W., & Howard, J. A. (1979). Inducing affect about
self or other: Effects on generosity in children. Developmental Psychol-
ogy, 15, 164 –167.
Baron, R. A. (1990). Environmentally induced positive affect: Its impact
on self-efficacy, task performance, negotiation, and conflict. Journal of
Applied Social Psychology, 20, 368 –384.
Baron, R. A., Fortin, S. P., Frei, R. L., Hauver, L. A., & Shack, M. L.
(1990). Reducing organizational conflict: The role of socially-induced
positive affect. International Journal of Conflict Management, 1, 133–
152.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable
distinction in social psychological research: Conceptual, strategic, and
statistical considerations. Journal of Personality and Social Psychology,
51, 1173–1182.
Barrett, K. C. (1995). A functionalist approach to shame and guilt. In J. P.
Tangney & K. W. Fischer (Eds.), Self-conscious emotions: The psychol-
ogy of shame, guilt, embarrassment, and pride (pp. 25– 63). New York:
Guilford Press.
Barry, B. (1999). The tactical use of emotion in negotiation. Research on
Negotiation in Organizations, 7, 93–121.
Barry, B., Fulmer, I. S., & Van Kleef, G. A. (2004). I laughed, I cried, I
139
settled: The role of emotion in negotiation. In M. J. Gelfand & J. M.
Brett (Eds.), The handbook of negotiation and culture (pp. 71–94).
Stanford, CA: Stanford University Press.
Barry, B., & Oliver, R. L. (1996). Affect in dyadic negotiation: A model
and propositions. Organizational Behavior and Human Decision Pro-
cesses, 67, 127–143.
Batson, C. D. (1987). Prosocial motivation: Is it ever truly altruistic? In L.
Berkowitz (Ed.), Advances in experimental social psychology (Vol. 20,
pp. 65–122). New York: Academic Press.
Baumeister, R. F., Stillwell, A. M., & Heatherton, T. F. (1994). Guilt: An
interpersonal approach. Psychological Bulletin, 115, 243–267.
Bazerman, M. H., Curhan, J. R., Moore, D. A., & Valley, K. L. (2000).
Annual Review of Psychology, 51, 279 –314.
Bell, D. E. (1985). Disappointment in decision making under uncertainty.
Operations Research, 33, 1–27.
Berndsen, M., van der Pligt, J., Doosje, B., & Manstead, A. S. R. (2004).
Guilt and regret: The determining role of interpersonal and intrapersonal
harm. Cognition and Emotion, 18, 55–70.
Berscheid, E., & Walster, E. (1967). When does a harm-doer compensate
a victim? Journal of Personality and Social Psychology, 6, 435– 441.
Carlsmith, J. M., & Gross, A. (1969). Some effects of guilt on compliance.
Journal of Personality and Social Psychology, 11, 232–239.
Carnevale, P. J., & Isen, A. M. (1986). The influence of positive affect and
visual access on the discovery of integrative solutions in bilateral nego-
tiation. Organizational Behavior and Human Decision Processes, 37,
1–13.
Carver, C. S., & Scheier, M. F. (1990). Origins and functions of positive
and negative affect: A control-process view. Psychological Review, 97,
19 –35.
Clark, L. F. (1993). Stress and the cognitive– conversational benefits of
social interaction. Journal of Social and Clinical Psychology, 12, 25–55.
Clark, M. A., & Taraban, C. B. (1991). Reactions to and willingness to
express emotion in two types of relationships. Journal of Experimental
Social Psychology, 27, 324 –336.
Clark, M. S., Ouellette, R., Powell, M. C., & Milberg, S. (1987). Recipi-
ent’s mood, relationship type, and helping. Journal of Personality and
Social Psychology, 53, 94 –103.
Clark, M. S., Pataki, S. P., & Carver, V. H. (1996). Some thoughts and
findings on self-presentation of emotions in relationships. In G. J. O.
Fletcher & J. Fitness (Eds.), Knowledge structures in close relation-
ships: A social psychological approach (pp. 247–274). Mahwah, NJ:
Erlbaum.
Cornelius, R. R. (1984). A rule model of emotional expression. In C. Z.
Malatesta & C. E. Izard (Eds.), Emotion in adult development (pp.
213–233). Beverly Hills, CA: Sage.
Côté, S. (2005). A social interaction model of the effects of emotion
regulation on work strain. Academy of Management Review, 30, 509 –
530.
Davidson, M. N., & Greenhalgh, L. (1999). The role of emotion in
negotiation: The impact of anger and race. Research on Negotiation in
Organizations, 7, 3–26.
De Cremer, D., Snyder, M., & Dewitte, S. (2001). ‘The less I trust, the less
I contribute (or not?)’: The effects of trust, accountability and self-
monitoring in social dilemmas. European Journal of Social Psychology,
31, 93–107.
De Dreu, C. K. W., Giebels, E., & Van de Vliert, E. (1998). Social motives
and trust in integrative negotiation: The disruptive effects of punitive
capability. Journal of Applied Psychology, 83, 408 – 422.
De Dreu, C. K. W., Lualhati, J. C., & McCusker, C. (1994). Effects of
gain/loss frames on satisfaction with self/other outcome-differences.
European Journal of Social Psychology, 24, 497–510.
De Dreu, C. K. W., & Van Kleef, G. A. (2004). The influence of power on
the information search, impression formation, and demands in negotia-
tion. Journal of Experimental Social Psychology, 40, 303–319.
140 VAN KLEEF, DE DREU, AND MANSTEAD
De Dreu, C. K. W., & Van Lange, P. A. M. (1995). The impact of social
value orientations on negotiator cognition and behavior. Personality and
Social Psychology Bulletin, 21, 1178 –1188.
Deutsch, M. (1960). The effect of motivational orientation upon trust and
suspicion. Human Relations, 13, 123–139.
Eisenberg, N. (2000). Emotion, regulation, and moral development. Annual
Review of Psychology, 51, 665– 697.
Eisenberg, N., Fabes, R. A., Miller, P. A., Fultz, J., Mathy, R. M., Shell, R.,
& Reno, R. R. (1989). The relations of sympathy and personal distress
to prosocial behavior: A multimethod study. Journal of Personality and
Social Psychology, 57, 55– 66.
Eisenberg, N., Fabes, R. A., Schaller, M., & Miller, P. (1989). Sympathy
and personal distress: Development, gender differences, and interrela-
tions of indexes. In N. Eisenberg (Ed.), Empathy and related emotional
responses (pp. 107–126). San Francisco: Jossey-Bass.
Ekman, P. (1993). Facial expression and emotion. American Psychologist,
48, 384 –392.
Fabes, R. A., Eisenberg, N., Karbon, M., Troyer, D., & Switzer, G. (1994).
The relations of children’s emotion regulation to their vicarious emo-
tional responses and comforting behaviors. Child Development, 65,
1678 –1693.
Forgas, J. P. (1998). On feeling good and getting your way: Mood effects
on negotiator cognition and behavior. Journal of Personality and Social
Psychology, 74, 565–577.
Freedman, J. L., Wallington, S., & Bless, E. (1967). Compliance without
pressure: The effect of guilt. Journal of Personality and Social Psychol-
ogy, 7, 117–124.
Fridlund, A. J. (1994). Human facial expression: An evolutionary view. San
Diego, CA: Academic Press.
Friedman, M. (1985). Toward a reconceptualization of guilt. Contempo-
rary Psychoanalysis, 21, 501–547.
Frijda, N. H. (1986). The emotions. Cambridge, England: Cambridge
University Press.
Frijda, N. H. (1993). Moods, emotion episodes, and emotions. In M. Lewis
& J. M. Haviland (Eds.), Handbook of emotions (pp. 381– 403). New
York: Guilford Press.
Frijda, N. H., & Mesquita, B. (1994). The social roles and functions of
emotions. In S. Kitayama & H. S. Markus (Eds.), Emotion and culture:
Empirical studies of mutual influence (pp. 51– 87). Washington, DC:
American Psychological Association.
Gilovich, T., & Medvec, V. H. (1994). The temporal pattern to the
experience of regret. Journal of Personality and Social Psychology, 67,
357–365.
Gilovich, T., & Medvec, V. H. (1995). The experience of regret: What,
when, and why. Psychological Review, 102, 379 –395.
Hegtvedt, K. A. (1990). The effects of relationship structure on emotional
responses to inequity. Social Psychology Quarterly, 53, 214 –228.
Hegtvedt, K. A., & Killian, C. (1999). Fairness and emotions: Reactions to
the process and outcomes of negotiations. Social Forces, 78, 269 –302.
Hill, M. G., Weary, G., & Williams, J. (1986). Depression: A self-
presentation formulation. In R. Baumeister (Ed.), Public self and private
self (pp. 213–240). New York: Springer-Verlag.
Hilty, J. A., & Carnevale, P. J. (1993). Black-hat/white-hat strategy in
bilateral negotiation. Organizational Behavior and Human Decision
Processes, 55, 444 – 469.
Hoffman, M. L. (1982). Development of prosocial motivation: Empathy
and guilt. In N. Eisenberg (Ed.), The development of prosocial behavior
(pp. 281–313). New York: Academic Press.
Holmes, J. G., Throop, W. F., & Strickland, L. H. (1971). The effects of
prenegotiation expectations on the distributive bargaining process. Jour-
nal of Experimental Social Psychology, 7, 582–599.
Homans, G. C. (1974). Social behavior: Its elementary forms. Oxford,
England: Harcourt Brace Jovanovich.
Isen, A. M., Daubman, K. A., & Nowicki, G. P. (1987). Positive affect
facilitates creative problem solving. Journal of Personality and Social
Psychology, 52, 1122–1131.
Kee, H. W., & Knox, R. E. (1970). Conceptual and methodological
considerations in the study of trust and suspicion. Journal of Conflict
Resolution, 14, 357–366.
Kelley, H. H., Beckman, L. L., & Fischer, C. S. (1967). Negotiating the
division of a reward under incomplete information. Journal of Experi-
mental Social Psychology, 3, 361–398.
Keltner, D., & Buswell, B. N. (1997). Embarrassment: Its distinct form and
appeasement functions. Psychological Bulletin, 122, 250 –270.
Keltner, D., & Haidt, J. (1999). Social functions of emotions at four levels
of analysis. Cognition and Emotion, 13, 505–521.
Kennedy-Moore, E., & Watson, J. C. (2001). How and when does emo-
tional expression help? Review of General Psychology, 5, 187–212.
Kenny, D. A., Kashy, D. A., & Bolger, N. (1998). Data analysis in so-
cial psychology. In D. T. Gilbert, S. T. Fiske, & A. Lindzey (Eds.),
The handbook of social psychology (Vol. 1, pp. 233–265). Boston:
McGraw-Hill.
Ketelaar, T., & Au, W. T. (2003). The effects of feelings of guilt on the
behaviour of uncooperative individuals in repeated social bargaining
games: An affect-as-information interpretation of the role of emotion in
social interaction. Cognition and Emotion, 17, 429 – 453.
Kimmel, M. J., Pruitt, D. G., Magenau, J. M., Konar-Goldband, E., &
Carnevale, P. J. D. (1980). Effects of trust, aspiration, and gender on
negotiation tactics. Journal of Personality and Social Psychology, 38,
9 –22.
Knutson, B. (1996). Facial expressions of emotion influence interpersonal
trait inferences. Journal of Nonverbal Behavior, 20, 165–182.
Labott, S. M., Martin, R. B., Eason, P. S., & Berkey, E. Y. (1991). Social
reactions to the expression of emotion. Cognition and Emotion, 5,
397– 417.
Lazarus, R. S. (1991). Emotion and adaptation. Oxford, England: Oxford
University Press.
Leith, K. P., & Baumeister, R. F. (1998). Empathy, shame, guilt, and
narratives of interpersonal conflicts: Guilt-prone people are better at
perspective taking. Journal of Personality and Social Psychology, 66,
1–37.
Levenson, R. W. (1994). Human emotion: A functional view. In P. Ekman
& R. J. Davidson (Eds.), The nature of emotion: Fundamental questions
(pp. 123–126). New York: Oxford University Press.
Lewis, H. B. (1971). Shame and guilt in neurosis. New York: International
Universities Press.
Lewis, M. (2000). Self-conscious emotions: Embarrassment, pride, shame,
and guilt. In M. Lewis & J. M. Haviland-Jones (Eds.), Handbook of
emotions (2nd ed., pp. 623– 636). New York: Guilford Press.
Lindskold, S. (1978). Trust development, the GRIT proposal, and the
effects of conciliatory acts on conflict and cooperation. Psychological
Bulletin, 85, 772–793.
Loewenstein, G. F., Thompson, L., & Bazerman, M. H. (1989). Social
utility and decision making in interpersonal contexts. Journal of Per-
sonality and Social Psychology, 57, 426 – 441.
Loomes, G., & Sugden, R. (1986). Disappointment and dynamic inconsis-
tency in choice under uncertainty. Review of Economic Studies, 53,
271–282.
Loomis, J. L. (1959). Communication, the development of trust, and
cooperative behavior. Human Relations, 12, 305–315.
Manstead, A. S. R. (1991). Emotion in social life. Cognition and Emotion,
5, 353–362.
McGrath, J. E., & Hollingshead, A. B. (1994). Groups interacting with
technology: Ideas, evidence, issues, and an agenda. Thousand Oaks,
CA: Sage.
McKersie, R. B., & Fonstad, N. O. (1997). Teaching negotiation theory and
skills over the Internet. Negotiation Journal, 13, 363–368.
 SUPPLICATION AND APPEASEMENT IN NEGOTIATIONS
Messick, D. M., & Sentis, K. P. (1979). Fairness and preference. Journal
of Experimental Social Psychology, 15, 418 – 434.
Moore, D. A., Kurtzberg, T. R., Thompson, L. L., & Morris, M. W. (1999).
Long and short routes to success in electronically mediated negotiations:
Group affiliations and good vibrations. Organizational Behavior and
Human Decision Processes, 77, 22– 43.
Morris, M. W., & Keltner, D. (2000). How emotions work: An analysis of
the social functions of emotional expression in negotiations. Research in
Organizational Behavior, 22, 1–50.
Naquin, C. E., & Paulson, G. D. (2003). Online bargaining and interper-
sonal trust. Journal of Applied Psychology, 88, 113–120.
Oatley, K., & Jenkins, J. M. (1992). Human emotions: Functions and
dysfunctions. Annual Review of Psychology, 43, 55– 85.
Oatley, K., & Jenkins, J. M. (1996). Understanding emotions. Cambridge,
MA: Blackwell.
Parkinson, B. (1996). Emotions are social. British Journal of Psychology,
87, 663– 683.
Parks, C. D., Henager, R. F., & Scamahorn, S. D. (1996). Trust and
messages of intent in social dilemmas. Journal of Conflict Resolution,
40, 134 –151.
Parks, C. D., & Hulbert, L. G. (1995). High and low trusters’ responses to
fear in a payoff matrix. Journal of Conflict Resolution, 39, 718 –730.
Parrott, W. G. (1993). Beyond hedonism: Motives for inhibiting good
moods and for maintaining bad moods. In D. M. Wegner & J. W.
Pennebaker (Eds.), Handbook of mental control (pp. 278 –305). Engle-
wood Cliffs, NJ: Prentice Hall.
Pillutla, M. M., & Murnighan, J. K. (1996). Unfairness, anger, and spite:
Emotional rejections of ultimatum offers. Organizational Behavior and
Human Decision Processes, 68, 208 –224.
Pruitt, D. G., & Carnevale, P. J. (1993). Negotiation in social conflict.
Buckingham, England: Open University Press.
Pruitt, D. G., & Kimmel, M. J. (1977). Twenty years of experimental
gaming: Critique, synthesis, and suggestions for the future. Annual
Review of Psychology, 28, 363–392.
Rempel, J. K., Holmes, J. G., & Zanna, M. P. (1985). Trust in close
relationships. Journal of Personality and Social Psychology, 49, 95–112.
Roseman, I. J., Wiest, C., & Swartz, T. S. (1994). Phenomenology, behav-
iors, and goals differentiate discrete emotions. Journal of Personality
and Social Psychology, 67, 206 –221.
Ross, W., & LaCroix, J. (1996). Multiple meanings of trust in negotiation
theory and research: A literature review and integrative model. Interna-
tional Journal of Conflict Management, 7, 314 –360.
Rotter, J. B. (1980). Interpersonal trust, trustworthiness, and gullibility.
American Psychologist, 35, 1–7.
Rousseau, D. M., Sitkin, S. B., Burt, R. S., & Camerer, C. (1998). Not so
different after all: A cross-discipline view of trust. Academy of Man-
agement Review, 23, 393– 404.
Russell, J. A., & Feldman Barrett, L. (1999). Core affect, prototypical
emotional episodes, and other things called emotion: Dissecting the
elephant. Journal of Personality and Social Psychology, 76, 805– 819.
Siegel, S., & Fouraker, L. E. (1960). Bargaining and group decision
making: Experiments in bilateral monopoly. New York: McGraw-Hill.
Sinaceur, M., & Tiedens, L. Z. (in press). Get mad and get more than even:
When and why anger expression is effective in negotiations. Journal of
Experimental Social Psychology.
Smith, D. L., Pruitt, D. G., & Carnevale, P. J. (1982). Matching and
mismatching: The effect of own limit, other’s toughness, and time
pressure on concession rate in negotiation. Journal of Personality and
Social Psychology, 42, 876 – 883.
Smith, R. H., Webster, J. M., Parrott, W. G., & Eyre, H. L. (2002). The role
of public exposure in moral and nonmoral shame and guilt. Journal of
Personality and Social Psychology, 83, 138 –159.
Sprecher, S. (1992). How men and women expect to feel and behave in
141
response to inequity in close relations. Social Psychology Quarterly, 55,
57– 69.
Steinel, W., & De Dreu, C. K. W. (2004). Social motives and strategic
misrepresentation in social decision making. Journal of Personality and
Social Psychology, 86, 419 – 434.
Steiner, C. (2000). Apology: The transactional analysis of a fundamental
exchange. Transactional Analysis Journal, 30, 145–149.
Tangney, J. P. (1990). Assessing individual differences in proneness to
shame and guilt: Development of the self-conscious affect and attribu-
tion inventory. Journal of Personality and Social Psychology, 59, 102–
111.
Tangney, J. P. (1995). Shame and guilt in interpersonal relations. In J. P.
Tangney & K. W. Fischer (Eds.), Self-conscious emotions: The psychol-
ogy of shame, guilt, embarrassment, and pride (pp. 114 –139). New
York: Guilford Press.
Tangney, J. P. (1999). The self-conscious emotions: Shame, guilt, embar-
rassment, and pride. In T. Dalgleish & M. Power (Eds.), Handbook of
cognition and emotion (pp. 541–568). New York: Wiley.
Thompson, L., Medvec, V. H., Seiden, V., & Kopelman, S. (2001). Poker
face, smiley face, and rant ‘n’ rave: Myths and realities about emotion in
negotiation. In M. A. Hogg & R. S. Tindale (Eds.), Blackwell handbook
of social psychology: Group processes (pp. 139 –163). Malden, MA:
Blackwell.
Timmers, M., Fischer, A. H., & Manstead, A. S. R. (1998). Gender
differences in motives for regulating emotions. Personality and Social
Psychology Bulletin, 24, 974 –985.
Tomkins, S. S. (1963). Affect, imagery, consciousness. Vol. 2: The negative
affects. New York: Springer.
Tripp, T. M., & Sondak, H. (1992). An evaluation of dependent variables
in experimental negotiation studies: Impasse rates and Pareto efficiency.
Organizational Behavior and Human Decision Processes, 51, 273–295.
van Dijk, W. W., & van der Pligt, J. (1997). The impact of probability and
magnitude of outcome on disappointment and elation. Organizational
Behavior and Human Decision Processes, 69, 277–284.
van Dijk, W. W., & Zeelenberg, M. (2002). Investigating the appraisal
patterns of regret and disappointment. Motivation and Emotion, 26,
321–331.
Van Kleef, G. A., & De Dreu, C. K. W. (2002). Social value orientation
and impression formation: A test of two competing hypotheses about
information search in negotiation. International Journal of Conflict
Management, 13, 59 –77.
Van Kleef, G. A., De Dreu, C. K. W., & Manstead, A. S. R. (2004a). The
interpersonal effects of anger and happiness in negotiations. Journal of
Personality and Social Psychology, 86, 57–76.
Van Kleef, G. A., De Dreu, C. K. W., & Manstead, A. S. R. (2004b). The
interpersonal effects of emotions in negotiations: A motivated informa-
tion processing approach. Journal of Personality and Social Psychology,
87, 510 –528.
Van Kleef, G. A., De Dreu, C. K. W., Pietroni, D., & Manstead, A. S. R.
(in press). Power and emotion in negotiation: Power moderates the
interpersonal effects of anger and happiness on concession making.
European Journal of Social Psychology.
Wallace, J., & Sadalla, E. (1966). Behavioral consequences of transgres-
sion: I. The effects of social recognition. Journal of Experimental
Research in Personality, 1, 187–194.
Yamagishi, T. (1986). The provision of a sanctioning system as a public
good. Journal of Personality and Social Psychology, 50, 110 –116.
Yee, J. L., & Greenberg, M. S. (1998). Reactions to crime victims: Effects
of victims’ emotional state and type of friendship. Journal of Social and
Clinical Psychology, 17, 209 –226.
Yukl, G. A. (1974a). Effects of situational variables and opponent conces-
sions on a bargainer’s perception, aspirations, and concessions. Journal
of Personality and Social Psychology, 29, 227–236.
Yukl, G. A. (1974b). Effects of the opponent’s initial offer, concession
142 VAN KLEEF, DE DREU, AND MANSTEAD

magnitude, and concession frequency on bargaining behavior. Journal of
Personality and Social Psychology, 30, 323–335.
Zeelenberg, M., & Beattie, J. (1997). Consequences of regret aver-
sion 2: Additional evidence for effects of feedback on decision
making. Organizational Behavior and Human Decision Processes,
72, 63–78.
Zeelenberg, M., van der Pligt, J., & Manstead, A. S. R. (1998). Undoing
regret on Dutch television: Apologizing for interpersonal regrets involv-
ing actions or inactions. Personality and Social Psychology Bulletin, 24,
1113–1119.
Zeelenberg, M., van Dijk, W. W., Manstead, A. S. R., & van der Pligt, J.
(2000). On bad decisions and disconfirmed expectancies: The psychol-
ogy of regret and disappointment. Cognition and Emotion, 14, 521–541.
Zetik, D. C., & Stuhlmacher, A. F. (2002). Goal setting and negotiation
performance: A meta-analysis. Group Processes and Intergroup Rela-
tions, 5, 35–52.

Appendix A Appendix B

Participants’ Payoff Chart Statements Used for the Manipulation of the Opponent’s
Emotion
Price of phones Warranty period Service contract
Opponent’s emotion Statement
Level Price Payoff Warranty Payoff Service Payoff
After Round 1
1 $150 400 1 month 120 1 month 240
2 $145 350 2 months 105 2 months 210 Disappointment I am pretty disappointed about this, I think I will
3 $140 300 3 months 90 3 months 180 offer 8–7–7
4 $135 250 4 months 75 4 months 150 Worry This worries me quite a lot, I think I will offer
5 $130 200 5 months 60 5 months 120 8–7–7
6 $125 150 6 months 45 6 months 90 Guilt I feel guilty for not having conceded more, I
7 $120 100 7 months 30 7 months 60 think I will offer 8–7–7
8 $115 50 8 months 15 8 months 30 Regret I feel sorry that I haven’t conceded more, I think
9 $110 0 9 months 0 9 months 0 I will offer 8–7–7
No emotion I think I will offer 8–7–7
Note. Prices in Euros were converted to U.S. dollars and rounded to the
nearest U.S.$5. After Round 3
Disappointment This is going awry, I am very disappointed. I am
going to offer 7–6–7
Worry This is going awry, I am very worried. I am
going to offer 7–6–7
Guilt This is going awry, I feel pretty guilty. I am
going to offer 7–6–7
Regret This is going awry, I regret it. I am going to
offer 7–6–7
No emotion I am going to offer 7–6–7

After Round 5
Disappointment I am going to offer 6–6–6, because I am really
disappointed
Worry I am going to offer 6–6–6, because I am really
worried
Guilt I am going to offer 6–6–6, because I feel very
guilty
Regret I am going to offer 6–6–6, because I am very
sorry
No emotion I am going to offer 6–6–6

Note. Statements were pretested and have been translated from Dutch.
The typing errors deliberately included in the original versions are not
shown.

Received April 17, 2004

Revision received May 31, 2005
Accepted December 14, 2005 䡲
 PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES

Optimism in Close Relationships: How Seeing Things

in a Positive Light Makes Them So

Sanjay Srivastava Kelly M. McGonigal

University of Oregon Stanford University

Jane M. Richards Emily A. Butler

University of Texas at Austin University of Arizona

James J. Gross
Stanford University

Does expecting positive outcomes— especially in important life domains such as relationships—make
these positive outcomes more likely? In a longitudinal study of dating couples, the authors tested whether
optimists (who have a cognitive disposition to expect positive outcomes) and their romantic partners are
more satisfied in their relationships, and if so, whether this is due to optimists perceiving greater support
from their partners. In cross-sectional analyses, both optimists and their partners indicated greater
relationship satisfaction, an effect that was mediated by optimists’ greater perceived support. When the
couples engaged in a conflict conversation, optimists and their partners saw each other as engaging more
constructively during the conflict, which in turn led both partners to feel that the conflict was better
resolved 1 week later. In a 1-year follow-up, men’s optimism predicted relationship status. Effects of
optimism were mediated by the optimists’ perceived support, which appears to promote a variety of
beneficial processes in romantic relationships.

Keywords: optimism, relationship satisfaction, perceived support, close relationships

I have heard of reasons manifold
Why Love needs be blind,
But this the best of all I hold—
His eyes are in his mind.
—Samuel Taylor Coleridge (1811)
Individuals’ perceptions of the social world are more than just
objective reports of an external reality. Social perceptions are
shaped in the mind of the perceiver, a fact that can have very real
consequences for social life. Romantic relationships, in particular,
have long been observed by poets and writers to be substantially
Sanjay Srivastava, Department of Psychology, University of Oregon;
Kelly M. McGonigal and James J. Gross, Department of Psychology,
Stanford University; Jane M. Richards, Department of Psychology, Uni-
versity of Texas at Austin; Emily A. Butler, Department of Psychology,
University of Arizona.
This research was supported by National Institutes of Health Grant R01
M58147 awarded to James J. Gross. Kelly M. McGonigal was supported
by a National Science Foundation Graduate Research Fellowship.
Correspondence concerning this article should be addressed to Sanjay
Srivastava, Department of Psychology, 1227 University of Oregon, Eu-
gene, OR 97403-1227. E-mail: sanjay@uoregon.edu
affected by the cognitive dispositions of the individuals involved.
As Coleridge might say, lovers’ eyes are in their minds.
In this article, we present an investigation of the consequences
of one particular cognitive disposition, namely optimism, within
romantic relationships. Is optimism associated with happier and
longer lasting romantic relationships? To answer this question, our
research was designed to test two related hypotheses. First, we
tested the hypothesis that optimists and their partners would have
relationships that are more satisfying, are characterized by better
conflict resolution, and are longer lasting.1 Second, we tested the
hypothesis that the reason why optimists have better relationship
outcomes is that they perceive their partners as more supportive.
We tested these hypotheses in cross-sectional analyses of couples’
1
Optimism and pessimism can be conceptualized several different ways:
as opposite poles of a single dimension, as two distinct dimensions, or as
discrete categories. In this article we treat optimism both conceptually and
empirically as a single, bipolar dimension, an approach that was supported
by analyses of the data. To avoid cumbersome language, we have used the
term optimists in this article as a shorthand, meaning in effect, “individuals
who score higher in optimism, relative to those who score lower.” It is not
our intention to suggest that optimists are a discrete category.

Journal of Personality and Social Psychology, 2006, Vol. 91, No. 1, 143–153
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.143

143
144 SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS
reports about their relationships, in analyses of how couples re-
sponded to a conflict conversation, and in a 1-year follow-up of
relationship dissolution.
Optimism, Perceived Support, and Social Functioning
Optimism is defined as the cognitive disposition to expect fa-
vorable outcomes (Scheier & Carver, 1985). A substantial body of
research has linked optimism to effective coping and to positive
mental and physical health outcomes (e.g., Scheier, Carver, &
Bridges, 2001; Taylor, Kemeny, Reed, Bower, & Gruenewald,
2000). A smaller number of studies have also shown that optimism
leads to better social functioning. For example, optimism is asso-
ciated with lower social alienation (Scheier & Carver, 1985) and
with longer lasting friendships (Geers, Reilley, & Dember, 1998).
In romantic relationships, two prior studies have suggested that
optimism about a particular relationship predicts greater satisfac-
tion in that relationship and reduced likelihood of relationship
dissolution (Helgeson, 1994; Murray & Holmes, 1997), although
the mechanisms explaining such a relation were not directly tested.
Why might optimists have more positive experiences in rela-
tionships? As a cognitive disposition, optimism should influence
how individuals attend to and interpret others’ behaviors and
intentions. We propose that within a close relationship, this cog-
nitive disposition may manifest itself as perceived support, the
belief that one’s partner is able and willing to provide support if
necessary (Murray & Holmes, 1997). Perceived support could in
turn have a number of benefits: It has been shown to lead indi-
viduals to feel that their relationship facilitates their personal and
collective goals (Brunstein, Dangelmayer, & Schultheiss, 1996;
Kaplan & Maddux, 2002), and it may buffer against stress and
negative affect in relationships (Dehle, Larsen, & Landers, 2001).
This latter effect may be particularly important in close
relationships.
Research on perceived support indicates that, like optimism, it is
moderately stable over time (Sarason, Sarason, & Shearin, 1986),
and it appears to be something more than simply a direct reflection
of others’ actual supportive behaviors (Barrera, 1986; Belsher &
Costello, 1991; Newcomb, 1990). Yet despite the agreement
among many researchers that perceived support is influenced by
personality variables, Lakey, McCabe, Fisicaro, and Drew (1996)
wrote that “surprisingly, there has been very little research on the
personality factors that predict the development of perceived sup-
port” (p. 1278).
Among personality factors that might promote perceived sup-
port, optimism seems to be a likely candidate. Perceived support is
associated with positive biases in evaluating and remembering
supportive behaviors in specific interactions and relationships
(Lakey et al., 1996; Lakey & Cassady, 1990; Pierce, Sarason, &
Sarason, 1992). Furthermore, the proposed mechanisms of per-
ceived support—positive affect, coping self-efficacy, and adaptive
coping—are all robustly associated with optimism (Chang, 2001;
Cozarelli, 1993; Scheier et al., 2001). Optimists are better liked by
others, which may reinforce their expectations about how others
will treat them (Carver, Kus, & Scheier, 1994). In relationships, we
expect that optimists would be more likely to perceive others’
behaviors as supportive and to respond accordingly.
A few studies have offered some evidence directly linking
optimism to perceived support. Associations between optimism
and perceived support have been found among air crash rescue
workers (Dougall, Hyman, Hayward, McFeeley, & Baum, 2001),
bereaved men (Park & Folkman, 1997), and college students
(Sarason, Levine, Basham, & Sarason, 1983). In a longitudinal
investigation, Brissette, Scheier, and Carver (2002) investigated
the relationship between optimism and perceived social support. In
a sample of incoming college students, optimism was associated
with concurrent reports of perceived support and number of close
friendships at the beginning of college and with increases in
perceived support over the course of the semester. The increases in
perceived support mediated the effect of optimism on depression,
though not the effect of optimism on stress. Brissette and col-
leagues’ findings are important and suggestive, but they were not
able to examine relational outcomes such as relationship satisfac-
tion or conflict resolution; their study also did not examine the
effects of an individual’s optimism on relationship partners.
The Present Study
The available evidence suggests that optimism is associated
with positive outcomes in relationships in general, possibly as a
result of processes that promote and maintain perceived support.
Our particular interest was in examining these processes in the
context of close relationships. Optimism and perceived support are
often studied in terms of their consequences for social life in
general; an examination of close relationships offers several dis-
tinct opportunities to complement this research. For researchers
who study close relationships, studying optimism and perceived
support can potentially provide insights into the cognitive pro-
cesses that maintain security and closeness between partners. For
researchers who study optimism, close relationships are an impor-
tant life domain for which optimism may have meaningful
consequences.
Studies of perceived support also suggest that there may be
important processes taking place in the context of dyadic relation-
ships that could be missed in broad-bandwidth studies of social
life. Although individuals do differ in their general tendency to
perceive all others as supportive, perceived support also draws
substantially on relationship-specific perceptions (Lakey et al.,
1996). That is, individuals form distinct judgments about the
supportiveness of other individuals, above and beyond their broad
judgments about others in general. Although much research on
social support has focused broadly on social networks, this finding
suggests that it is also important to examine the consequences of
perceived support in the context of specific relationships.
In developing our questions and hypotheses, we organized our
investigation around two guiding questions. First, what conse-
quences, if any, does optimism have for satisfaction in close
relationships, both for the optimist and for the optimist’s partner?
Second, does perceived support explain the relation between op-
timism and relationship satisfaction?
Because of the complexity of the research design, we present the
findings in three parts (see Table 1). Part 1 examines the cross-
sectional relations among both partners’ optimism, perceived sup-
port, and relationship satisfaction at Time 1. Part 2 reports a closer
examination of how the couples reacted to conflict (Time 2) and
how well they felt the conflict was resolved 1 week later (Time 3).
Part 3 examines an objective outcome, relationship maintenance
versus dissolution, 1 year later (Time 4).
 OPTIMISM IN CLOSE RELATIONSHIPS 145

Table 1
Overview of Design, Procedures, and Measures

Timepoint (time since Time 1) Procedure Major measure

Part 1

Time 1 Questionnaire packet Optimism, perceived support,

relationship satisfaction

Part 2

Time 2 (1 week) Laboratory-based conflict Positive conflict engagement

conversation
Time 3 (2 weeks) Post-conflict follow-up Conflict resolution

Part 3

Time 4 (1 year) One-year follow-up Relationship status

Part 1: Optimism and Perceived Support
in a Dating Relationship
In Part 1, we examined partners’ reports regarding their dating
relationship in general. We hypothesized that optimism would be
associated with greater relationship satisfaction. Furthermore, we
expected not only that optimists themselves would report greater
relationship satisfaction than would pessimists but also that the
partners of optimists would report greater relationship satisfaction
than the partners of pessimists. Such an effect would indicate that
the positive relational consequences of optimism are not just “in
the head” of the optimists. We further hypothesized that the effects
of optimism would be mediated by optimists’ tendency to perceive
their partners as supportive in the relationship.
To rule out possible confounds, we conducted several additional
analyses. One possible confounding variable was partner invest-
ment: Perhaps optimists attract more supportive partners, in which
case an effect of optimism on perceived support could simply
reflect an accurate appraisal rather than a perceptual disposition.
Thus, we also obtained reports from each partner of offered sup-
port in the relationship to use as control variables. If optimists have
a global tendency to see their partners as supportive, that relation
should be independent of the actual amount of support offered by
their partners.
Finally, some studies have suggested that optimism may be
correlated with the personality traits of Neuroticism, Extraversion,
or self-esteem (see Scheier et al., 2001). Thus, we conducted
additional control analyses to ensure that the effects of optimism
were independent of these other dimensions of individual differ-
ences, as well as the individuals’ ages, the length of the relation-
ship, and whether the partners were living together.
Method
Procedure and Participants
We examined data from a study of dating couples assessed at multiple
time points over a 1-year period (see Table 1). For the analyses presented
in this article as Parts 1 and 2, we included couples from the original
sample who completed all measures at Times 1, 2, and 3 (but not neces-
sarily Time 4); this left us with 108 couples (N ⫽ 216) for the present
report.2 In Part 1 we analyze data from Time 1, when participants com-
pleted measures of personality, social support, and the dating relationship.
At least one member of each couple was an undergraduate recruited
from one of three northern California universities. Couples were exclusive
and had been dating for at least 6 months at the start of the study, with a
median relationship length of 16 months; 12% of couples were cohabiting.
Participant ages ranged from 18 to 25 years, with a mean age of 20.4 years.
The ethnic and racial composition of this sample was 2.1% African
American, 23.8% Asian, 56.3% Caucasian, 14.6% Latino/Hispanic, 0.8%
Native American, and 2.5% other. Participants were paid $15/hr for their
participation.
Measures
Optimism. The Life Orientation Test (LOT; Scheier & Carver, 1985) is
an eight-item self-report measure of general outcome expectancies. Sample
items include “In uncertain times, I usually expect the best” and the
reverse-coded item “If something can go wrong for me, it will.” Responses
range from 1 (strongly disagree) to 5 (strongly agree). We rescaled scores
of all individual difference measures to percent of maximum possible
(POMP) metric, which sets the theoretical range of a scale from 0 to 100.
POMP scoring is a linear transformation of raw scores and thus does not
affect standardized analyses, but it can aid in interpretation of raw scores
by putting them on an intuitive metric (Cohen, Cohen, Aiken, & West,
1999). Actual scores on the LOT, in POMP metric, ranged from 22 to 100;
means and standard deviations for the LOT and other major variables are
reported in Table 2. Alpha reliability coefficients were .80 for men and .86
for women, and factor analysis indicated a unidimensional structure. All of
our data analyses controlled for possible confounding due to partner
similarity on optimism. However, it is worth noting that the correlation
between partners’ optimism was r ⫽ .12, p ⫽ .22. In other words, there was
not a strong or reliable tendency for optimists to be partnered with other
optimists.
Perceived support. To assess perceived support in the dating relation-
ship, we used the Maintenance Questionnaire (MQ; Stafford & Canary,
1991). Participants rated 24 statements concerning their partner’s behaviors
on a scale from 1 (strongly disagree) to 7 (strongly agree). The MQ has
five subscales that cover a broad range of supportive behaviors: (a) posi-
tivity (e.g., “Does not criticize me”), (b) openness (e.g., “Encourages me to
2
We compared the 108 couples included in this report with the 12
couples who did not return at Times 2 or 3. Analyses indicated no
differences on optimism for the men or women of these couples (rs ⬍ .07,
ps ⬎ .52); however, the men in the 108 included couples were somewhat
higher in perceived support (r ⫽ .26, p ⫽ .004) and higher in relationship
satisfaction (r ⫽ .19, p ⫽ .04). The women in these couples did not differ
significantly on those dimensions (rs ⬍ .13, ps ⬎ .16).
146 SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS

Table 2
Correlations, Means, and Standard Deviations for Main Measures from Parts 1 and 2

Variable 1. MLOT 2. MMQ 3. MCSS 4. MCONV 5. MRES 6. FLOT 7. FMQ 8. FCSS 9. FCONV 10. FRES

1. MLOT —
2. MMQ .36 —
3. MCSS .32 .68 —
4. MCONV .19 .44 .49 —
5. MRES .21 .42 .40 .46 —
6. FLOT .12 .21 .26 .23 .26 —
7. FMQ .09 .41 .50 .51 .33 .26 —
8. FCSS .15 .29 .36 .30 .29 .27 .62 —
9. FCONV .16 .46 .41 .71 .49 .30 .49 .31 —
10. FRES .23 .25 .26 .40 .48 .32 .23 .32 .43 —
M 66.2 81.4 80.4 0.0 5.0 66.8 79.1 82.1 0.0 5.5
SD 15.9 12.1 16.11 0.7 2.5 17.8 12.8 16.4 0.7 2.5

Note. N ⫽ 108 couples. Effect sizes greater than .20 are significant at p ⬍ .05. The first letter in the variable name indicates gender (M ⫽ male, F ⫽
female). LOT ⫽ Life Orientation Test; MQ ⫽ Maintenance Questionnaire; CSS ⫽ Couple Satisfaction Survey; CONV ⫽ positive engagement in conflict
conversation (average of z-scored self-reports and partner-reports); RES ⫽ resolution of conflict.

disclose my thoughts and feelings to him/her”), (c) assurances (e.g.,
“Stresses his or her commitment to me”), (d) social network (e.g., “Focuses
on common friends and affiliations”), and (e) sharing tasks (e.g., “Helps
equally with tasks that need to be done”). The five subscales were all
positively correlated (mean r ⫽ .38, ranging from .20 to .58), so we
averaged the five scales and converted to POMP metric to create a global
measure of perceived support from the dating partner. Scores ranged from
35 to 100. Alphas (computed at the item level) were .91 for men and .92
for women.
Relationship satisfaction. To measure relationship satisfaction, we
used the Couple Satisfaction Scale (CSS; Cowan & Cowan, 1990). The
CSS includes eight items that are rated on scale from 1 (very dissatisfied)
to 5 (very satisfied). A sample item is, “In general, how do you feel about
the closeness and distance in your relationship with your partner now?”
Whereas the MQ, our measure of perceived support, asks members of
couples to report what their partners do, the CSS asks individuals how they
feel about the relationship. CSS scores, computed in POMP metric, ranged
from 9 to 100. Alphas for the CSS were .89 for men and .89 for women.
Control measure: Offered support. We used a subset of 10 items from
the Investment Scale (IS; Lund, 1985), which asks the participant to rate
“how much you feel you have invested in your relationship in each of the
following ways” on a scale from 1 (not invested) to 7 (very invested). Items
were selected to match the subscales of the MQ, for example, “Trying to
encourage and support your partner” (positivity), “Telling your partner
your true feelings about the relationship” (openness), “Integrating your
partner into your family” (social network), “Making formal agreements
about your relationship, such as deciding to go steady, get engaged, or get
married” (assurances), and “Doing favors for or helping your partner, such
as lending money or doing errands” (tasks). The items were summed and
converted to POMP metric to create a global self-report measure of offered
support. Scores ranged from 39 to 100; means were 76.8 (SD ⫽ 13.2) for
men and 76.7 (SD ⫽ 11.1) for women. Alphas were .80 for men and .72
for women.
Control measures: Extraversion, Neuroticism, self-esteem, and demo-
graphics. Extraversion and Neuroticism were measured with eight-item
scales from the Big Five Inventory (John & Srivastava, 1999). Alphas for
Extraversion were .89 for men and .88 for women; alphas for Neuroticism
were .77 for men and .82 for women. Self-esteem was measured with the
10-item Rosenberg Self-Esteem Scale (Rosenberg, 1965); alphas were .88
for men and .90 for women. We also measured each partner’s age, how
long the couple had been together, and whether they were cohabiting.
Discriminant validity among optimism and relational measures. Con-
ceptually, the measures of optimism, perceived support, offered support,
and relationship satisfaction are all supposed to measure different things.
However, it was important to establish discriminant validity; a possible
counterhypothesis was that the measures simply reflected a general rela-
tional positivity factor. To test this counterhypothesis, we ran a confirma-
tory factor analysis in which all four of the men’s measures loaded on a
latent “men’s positivity” factor, all of the women’s measures loaded on a
latent “women’s positivity” factor, and the men’s and women’s factors
were allowed to correlate.3 The analysis showed that the counterhypothesis
did not fit the data, ␹2(19, N ⫽ 108) ⫽ 52.9, p ⬍ .001; normed fit index
(NFI) ⫽ .80; root-mean-square error of approximation (RMSEA) ⫽ .13.
Analyses of reduced sets of variables, created by eliminating optimism or
offered support, did not show substantially better fit.
Results and Discussion
For our analyses we were interested in estimating both within-
person and between-persons effects—for example, how an indi-
vidual’s optimism relates to his or her own relationship satisfaction
(a within-person effect) and to his or her partner’s relationship
satisfaction (a between-persons effect). Both of these kinds of
questions are addressed by the actor–partner interdependence
model (APIM; Kashy & Kenny, 1997), a data analysis procedure
for dyads. The APIM was also designed to deal with the violations
of statistical independence associated with dyadic data. Thus, we
adopted the APIM as our basic data-analytic strategy.
The APIM estimates two kinds of effects: actor effects and
partner effects. Actor effects are within-person effects: They rep-
resent the influence of an individual’s level of a predictor variable
on that individual’s level of an outcome variable. Partner effects
are between-person effects: They represent the influence of an
individual’s level of a predictor on that individual’s partner’s level
of the outcome variable. APIM estimates also control for con-
founding due to partner similarity.
The APIM is rooted in regression (Kashy & Kenny, 1997). As
with regression, it is possible to extend the APIM to include
moderators, control variables, and mediators. We had a substantive
interest in taking advantage of all of these possibilities. One
important question was whether gender moderated the actor and
partner effects. In the APIM, actor and partner effects are aggre-
3
We thank an anonymous reviewer for suggesting this analysis.
 OPTIMISM IN CLOSE RELATIONSHIPS
gated across both members of the couple. When members of
couples are distinguishable on some variable—such as gender, in
the case of our heterosexual dating couples—it is possible to ask
whether actor and partner effects are moderated by gender. All of
the analyses we report were tested for moderator effects of gender.
Unless reported otherwise, such effects were not significant, and
thus results apply to both men and women.4
The basic APIM can also be elaborated to test models with
multiple predictors (for control analyses) or with mediated paths.
Shrout and Bolger (2003) reported that more sensitive tests of
mediation can be conducted by using bootstrap analyses, as com-
pared with other methods. Thus, we ran our analyses in Amos 4.01
(Arbuckle, 1999), which can conduct bootstrap analyses.
Do Optimists and Their Partners Report Greater
Relationship Satisfaction?
We expected that optimists and their partners would experience
their relationships as more satisfying. To test this hypothesis, we
performed an APIM analysis using optimism to predict relation-
ship satisfaction. The results indicated that optimists reported
greater relationship satisfaction. The standardized actor effect was
.27, p ⬍ .001, with a 95% confidence interval (CI) ranging from
.15 to .38. (The p values and CIs reported for all APIM analyses
are bias-corrected values from bootstrap analyses.) Furthermore,
optimists’ partners also reported greater relationship satisfaction,
indicating that the positive relational consequences of optimism
were not just “in the head” of the optimists: standardized partner
effect ⫽ .18, p ⫽ .006, 95% CI ⫽ (.06, .30).
Does Perceived Support Mediate Relations Between
Optimism and Relationship Satisfaction?
Having established that optimism was related to relationship
satisfaction, we then tested whether this relation was mediated by
perceived support. Following Shrout and Bolger’s (2003) proce-
dure (the logic of which is modeled on Baron & Kenny, 1986), this
required four further steps. Each step must produce a significant
result to proceed to the next. First, we tested whether optimism
predicts perceived support. Second, we tested whether perceived
support predicts relationship satisfaction when controlling for op-
timism. Third, we tested the mediated paths from optimism via
perceived support to relationship satisfaction; a significant boot-
strap test would support mediation. This bootstrap test is a more
powerful replacement for the Sobel test used in conventional
mediation analysis. Fourth, we tested the direct paths from opti-
mism to relationship satisfaction when controlling for perceived
support; this last step would indicate whether mediation was
partial or complete.
Did optimism predict perceived support? The results indicated
that optimists perceived greater support from their partners: actor
effect ⫽ .29, p ⬍ .001, 95% CI ⫽ (.17, .41). Optimists’ partners
had marginally higher levels of perceived support: partner effect ⫽
.12, p ⫽ .07, 95% CI ⫽ (–.01, .24).
Did perceived support predict relationship satisfaction? The
effect of perceived support on an actor’s own relationship satis-
faction was substantial: actor effect ⫽ .58, p ⫽ .001, 95% CI ⫽
(.44, .70). Individuals who perceived greater support also had
more satisfied partners: partner effect ⫽ .16, p ⫽ .003, 95% CI ⫽
(.07, .27).
147
Were the mediated paths significant? The bootstrap tests in-
dicated that the actor effect of optimism on relationship satisfac-
tion, reported earlier, was significantly mediated by perceived
support: mediated actor effect ⫽ .18, p ⫽ .001, 95% CI ⫽ (.09,
.27). Likewise, the effect of optimism on a partner’s relationship
satisfaction was also significantly mediated by the optimist’s per-
ceived support: mediated partner effect ⫽ .10, p ⫽ .003, 95% CI ⫽
(.03, .18).
Did the direct effects indicate full or partial mediation? If the
direct effect of optimism on an actor’s own relationship satisfac-
tion was still significant, that would indicate partial (rather than
full) mediation of the actor effect. This effect was not significant:
direct actor effect ⫽ .10, p ⫽ .15, 95% CI ⫽ (–.03, .24). Nor was
the direct partner effect significant: direct partner effect ⫽ .07, p ⫽
.17, 95% CI ⫽ (–.03, .18). Thus, the analyses indicated the effects
of an individual’s optimism on both the individual’s own relation-
ship satisfaction and on a partner’s satisfaction could be explained
by the optimist’s perceived support.
Control Analyses
To ensure that the effect of optimism on global perceived
support was not a result of optimists attracting more supportive
partners, we conducted an APIM analysis testing the effect of
optimism on perceived support while controlling for offered sup-
port. The effects of optimism were virtually unchanged: actor
effect ⫽ .28, p ⫽ .001, 95% CI ⫽ (.16, .40); partner effect ⫽ .10,
p ⫽ .13, 95% CI ⫽ (–.02, .22). Thus, optimists’ perceptions of
their partners’ supportiveness could not be explained away by
them attracting genuinely more supportive partners.
We also wanted to ensure that the effects of optimism on
relationship satisfaction were specific to optimism rather than
being attributable to related traits. To test this, we conducted
APIM analyses with covariates, controlling for individual differ-
ences in Extraversion, Neuroticism, and self-esteem, as well as
both partners’ ages, the length of the relationship, and whether the
couple was living together; covariates were tested one by one
because of concerns about multicollinearity. Pitted against each
covariate, optimism always was a significant predictor; further-
more, no covariate had a significant effect on relationship satis-
faction after controlling for optimism (all absolute effects ⬍ .12;
all ps ⬎ .16). Thus, we felt fairly confident that the effects of
optimism on relationship satisfaction were not confounded with
broader personality traits, with self-esteem, or with the demo-
graphic and background variables we examined.
Part 1 thus shows that the romantic relationships of optimists are
characterized by greater relationship satisfaction than the relation-
ships of those who are less optimistic. The mediation analyses
suggested that optimists’ general tendency to see their partners as
supportive mediated these positive relationship outcomes. Not
only did optimists report greater relationship satisfaction, but so
did their (not necessarily optimistic) partners, suggesting that the
4
The APIM can be specified as a path model with equality constraints
between members of the dyad; in this study, the APIM was specified by
setting men’s parameter estimates equal to women’s. The unconstrained or
“saturated” model produces separate parameter estimates for men and for
women. Thus, the chi-square test of model fit (which compares the con-
strained model to a saturated model) is, in the present context, a test of
moderation by gender.
148 SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS
positive relationship outcomes are not merely a Pollyanna-like
fantasy of the optimists. Because Part 1 was based on cross-
sectional data, however, the ordering of variables in the media-
tional model was based on conceptual considerations rather than
on the design of the study. In Part 2, we adopted a design in which
the temporal structure of the design gave us a stronger basis to
evaluate whether the relationship benefits of optimism are ex-
plained by perceived support.
Part 1 focused on global perceptions and feelings about the
relationship. In Part 2, we wanted to move beyond this global level
of analyses and take a closer look at the role of optimism in
relationship processes. To do this, we brought the same couples
from Part 1 into the laboratory and facilitated a conflict conver-
sation. We assessed whether optimists and their partners perceived
each other as supportive during the conflict, and whether that
perception of support contributed to both partners’ reports of how
well the conflict was resolved 1 week later.
Part 2: The Conflict Conversation
In dating relationships, a common stressor is disagreement be-
tween partners, such as disagreement about finances or time spent
together. How members of a couple perceive and react to disagree-
ments can be important for the health of the relationship (Bradbury
& Fincham, 1990; Carstensen, Gottman, & Levenson, 1995; Ga-
ble, Reis, & Downey, 2003). In Part 2, we examined how the
couples in our study responded to conflict by bringing them into
the laboratory to have a conversation about the most stressful area
of current disagreement in their relationship. Immediately after the
interaction, we asked each member of the couple to report how
positively and constructively they had engaged in the conflict, and
how positively and constructively their partner had engaged. One
week later, we asked each member of the couple how well they had
resolved the conflict as a couple. In measuring positive conflict
engagement, we believed that it was critical to take advantage of
the participants’ position as informants within their relationships to
tell us about how effective they were in addressing the conflict
together. In essence, we were interested in the shared reality of the
relationship—the couple’s joint construal of how effectively they
mutually dealt with the conflict.
We hypothesized that optimists and their partners would both
see the conflict as better resolved 1 week after their conversation.
We also hypothesized that this resolution would be explained, at
least in part, by a shared perception that both partners engaged
positively and constructively during the conversation. These hy-
potheses were brought together in a double-mediation model. In
Part 1, we found that optimists had a global tendency to perceive
their partners as supportive. Now, we anticipated that this global
tendency toward perceived support would be manifested in the
more specific context of the conversation through positive engage-
ment and that this positive engagement would be recognized by
both partners. More positive engagement in the conflict would, in
turn, lead to a better resolution to the conflict in the eyes of both
partners.
In examining whether optimists and their partners reported
better conflict resolution, we considered the alternative explana-
tion that optimists’ relationships might be characterized by rela-
tively low-intensity conflicts. That is, optimists might appear to be
good at resolving conflict, but only because their conflicts are
relatively easy to resolve. To address this possibility, we also
examined the participants’ ratings of how intensely they disagree
about various topics in their relationship, including the one dis-
cussed; this measure was taken before the conversation took place.
Method
Procedure
The conflict conversation procedure was modeled after that used by
Carstensen et al. (1995).5 On arrival at the laboratory at Time 2, a female
experimenter gave participants an overview of the study. Participants were
told that the study was about “how couples talk to each other about
important conflicts or areas of disagreement in the relationship.” Thus, they
would need to talk to each other for 10 min and complete questionnaires
concerning their reactions to the conversation. Both members of the couple
had separately reported on their area of greatest current disagreement in the
Time 1 questionnaire set using the Couple Problem Inventory (Gottman,
Markman, & Notarius, 1977). In this questionnaire, participants indicated
how much they disagree with their partner in a number of preestablished
areas (money, jealousy, recreation, etc.) and also had the opportunity to list
additional areas of disagreement. After rating disagreement across all
areas, participants then filled in a response to the question, “Which is
currently the greatest area of disagreement in your relationship?” (addi-
tional questions asked for the second and third greatest areas). Prior to the
Time 2 session, the experimenter randomly selected either the male’s or
female’s area of greatest disagreement. The experimenter then raised the
topic and asked each partner to describe (a) more specifically how the
problem area was relevant to their relationship, (b) the last time this
problem came up between them, (c) his or her emotions surrounding this
specific incident, and (d) why he or she experienced these emotions. If the
partners’ responses indicated that the topic was not likely to be appropriate
for the experiment, the experimenter selected a different area of disagree-
ment. Disagreement in the area chosen for discussion, rated by participants
on a scale from 0 (don’t disagree at all) to 100 (disagree very much),
averaged 60 for men and 65 for women. The couples discussed the topic for
10 min. At the end of the conversation, participants completed question-
naires about positive conflict resolution behaviors.
One week later (at Time 3), participants returned to the laboratory to
complete questionnaires about the conflict topic and the conversation they
had in the laboratory. Participants rated the degree to which the conflict had
been resolved since the conversation.
Measures
Intensity of disagreements. On the Couple Problem Inventory, each
partner rated the intensity of their disagreement in each potential area of
disagreement, using a scale from 0 (don’t disagree at all) to 100 (disagree
very much). We computed disagreement scores by averaging ratings across
the 13 preestablished areas and, if applicable, the 1 or 2 additional areas
identified by the participant (for men, M ⫽ 23, SD ⫽ 14; for women, M ⫽
23, SD ⫽ 13). Alpha reliabilities of the disagreement composite, computed
across the 13 preselected topics that every participant rated, were .81 for
men and .74 for women.
Positive engagement in conflict. After the conversation, participants
reported the extent to which they and their partners engaged in positive or
5
The data presented in this article were originally collected as part of an
experimental study of emotions and physiology. For the purposes of the
experimental study, physiological measurements were taken in the Time 2
laboratory session, and couples were randomly assigned to have one
member suppress or reappraise his or her emotions or to a control condition
(Richards, Butler, & Gross, 2003). Our present focus is on individual
differences rather than the effects of the experimental manipulation, and
none of the effects reported in this article interacted with experimental
condition.
 OPTIMISM IN CLOSE RELATIONSHIPS
supportive behaviors during the conflict conversation. Sample items in-
clude, “During the conversation, to what extent were you [was your
partner] a good listener?” “During the conversation, to what extent did you
[your partner] try to understand your partner’s [your] point of view?” and
“During the conversation, to what extent did you [your partner] criticize
your partner [you]?” (reverse scored). A total of 17 items were rated on a
scale from 0 (none/not at all) to 10 (a great deal/extremely). We averaged
the items to create composites. Self-reports ranged from 3.8 to 9.7; partner
reports ranged from 3.0 to 9.4. Means (and SDs) were as follows: men’s
reports of women, 6.8 (1.3); women’s reports of men, 6.9 (1.3); men’s
self-reports, 6.9 (1.2); women’s self-reports, 6.9 (1.2). Alphas were .87 for
men’s reports of women, .86 for women’s reports of men, .83 for men’s
reports of their own behavior, and .83 for women’s reports of their own
behavior.
To simplify the analyses, we created a positive engagement variable for
each individual that aggregated across data sources. That is, the positive
engagement variable for men was an average of the men’s self-reports with
women’s reports of their male partners, and vice versa to create an
aggregate for women; variables were converted to z scores before averag-
ing. These aggregates were justified by the substantial (though not perfect)
agreement between self-reports and partner reports: agreement between
men’s self-reports and women’s partner reports, indexed as an alpha
coefficient, was .64; agreement between women’s self-reports and men’s
partner reports was .53. To make sure that we were measuring the shared
reality of the relationship and not merely the positive biases of optimists,
we also examined the reports that relied on a single data source (either self
or partner) and attempted to replicate all of the analyses with the single-
reporter variables. In the model run with partner reports, optimists’ own
positive engagement would be reported by their partners, and thus the actor
effects would be immune from any positive perceptual bias “in the heads”
of optimists. In the model run with individuals’ construals of their own
behavior, the optimists’ partners’ behavior would be reported by the
partners rather than the optimists; thus the partner effects would be un-
tainted by optimists’ internal biases. If these analyses replicated the find-
ings with the aggregated variables, that would ensure that the effects
reflected the shared reality of the relationship rather than the idiosyncratic
views of one person.
Conflict resolution. Two items, rated on a scale from 0 (none/not at
all) to 10 (a great deal/extremely), were used to assess both partners’
feelings about how well the conflict was resolved 1 week after the con-
versation: “At this point, to what extent is the conflict you talked about in
your previous session resolved?” and “At this point, to what extent have
you and your partner moved in the right direction to resolve the conflict
you talked about in your previous session?” These two items were aver-
aged to create conflict resolution scores. Actual scores covered the full
range of the scale. Alphas were .80 for men and .81 for women.
Controls. As in Part 1, we examined Extraversion, Neuroticism, self-
esteem, both partners’ ages, length of relationship, and cohabitation status
as control variables. We also included a measure of relationship satisfac-
tion (the CSS from Part 1) to rule out the possibility that participants were
simply saying good things about how they resolved conflict because they
were generally satisfied with their relationships.
Results and Discussion
Did Optimists Have Less Intense Disagreements?
We examined the effect of optimism on both individuals’ ratings
of the intensity of disagreement in their relationships. Optimists
and their partners described their disagreements as somewhat less
intense: actor effect ⫽ ⫺.15, p ⫽ .02, 95% CI ⫽ (–.29, ⫺.02);
partner effect ⫽ ⫺.16, p ⫽ .01, 95% CI ⫽ (–.30, ⫺.03). Thus, we
included intensity of disagreement as a control variable when
testing the effects of optimism on conflict resolution.
149
Did Optimists and Their Partners See the Conflict as
Better Resolved?
Were optimists’ relationships characterized by better positive
conflict resolution, as perceived by both partners? In an APIM
analysis entering optimism and intensity of disagreement simulta-
neously to predict conflict resolution, the actor effect of optimism
was .17, p ⫽ .018, 95% CI ⫽ (.05, .31), and the partner effect of
optimism was .15, p ⫽ .02, 95% CI ⫽ (.02, .28). Individuals who
rated their disagreements as relatively intense did report poorer
conflict resolution, actor effect ⫽ ⫺.20, p ⫽ .01, 95% CI ⫽ (⫺.35,
⫺.04); the partner effect was not significant ( p ⫽ .13). From this
analysis it can be concluded that both optimists and their partners
agreed that their conflicts had reached a more satisfactory resolu-
tion 1 week later and that this effect could not be explained away
by baseline differences in the intensity of their disagreements.
Do Perceived Support and Positive Engagement Mediate
the Benefits of Optimism?
We hypothesized that optimists’ global perceived support would
promote positive engagement in the conflict conversation, as rec-
ognized by both partners, and that this would explain the effects of
optimism on achieving a more satisfactory conflict resolution. Part
1 already demonstrated the effect of optimism on global perceived
support; here we present evidence testing the remaining elements
of the double-mediation hypothesis.
Did perceived support promote positive engagement? We ran
APIMs testing the effects of global perceived support on the
aggregated positive engagement measure. The analyses showed
that individuals with greater perceived support were seen as en-
gaging more positively in the conflict: actor effect ⫽ .32, p ⫽ .004,
95% CI ⫽ (.22, .42). In follow-up analyses in which we analyzed
the self-reports and partner reports separately, this effect was
significant regardless of whose reports of positive engagement we
analyzed: Individuals with higher global perceived support saw
themselves as engaging more positively in the conflict, and their
partners saw them that way as well.
The analyses also showed that an individual’s perceived support
predicted the partner’s positive engagement: partner effect ⫽ .35,
p ⫽ .001, 95% CI ⫽ (.26, .45). Again, the follow-up analyses
indicated that this effect was significant with both data sources:
Individuals who were high in global perceived support saw their
partners as engaging more positively in the conflict, and their
partners shared that perception.
In an additional follow-up analysis that included intensity of
disagreement as a control variable, perceived support still had
significant actor and partner effects on positive engagement. In-
tensity of disagreement did not have significant actor or partner
effects in this analysis ( ps ⬎ .25).
Did positive engagement predict better resolution 1 week later?
Individuals who engaged more positively in the conflict conver-
sation reported better conflict resolution 1 week later: actor ef-
fect ⫽ .27, p ⫽ .002, 95% CI ⫽ (.13, .41). Their partners also saw
the conflict as better resolved: partner effect ⫽ .26, p ⫽ .002, 95%
CI ⫽ (.11, .40). These analyses held up regardless of the data
source for the positive engagement variable.
Was the effect of optimism on conflict resolution mediated by
perceived support and positive conflict resolution? We tested for
mediation by evaluating whether the mediated paths from opti-
150 SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS
mism, through perceived support, through positive engagement, to
conflict resolution were significant. The compound mediated path
from optimism to the optimist’s own report of conflict resolution
was significant: mediated actor effect ⫽ .07, p ⫽ .001, 95% CI ⫽
(.03, .12). The compound mediated path from optimism to the
partner’s conflict resolution was also significant: mediated partner
effect ⫽ .07, p ⫽ .001, 95% CI ⫽ (.04, .12). This analysis
supported the mediation hypothesis.
Was the mediation full or partial? In the double-mediation
model, there are six ways that direct effects could have “bypassed”
the mediated pathways. Optimism could have had a direct effect on
an actor’s own conflict resolution or on a partner’s conflict reso-
lution that was not mediated by perceived support or positive
engagement. Optimism could have had direct actor or partner
effects on positive engagement that were not mediated by per-
ceived support. Additionally, perceived support could have had
direct actor or partner effects on conflict resolution that were not
mediated by positive engagement.
To test these various possibilities together, we took advantage of
Amos’s model-comparison capabilities to test models with full and
partial mediation. In the full-mediation model, depicted in Figure
1, we allowed only effects from optimism to perceived support,
perceived support to positive engagement, and positive engage-
ment to conflict resolution. (As noted earlier, we tested for mod-
erating effects of gender by constraining men’s and women’s paths
to be equal; thus a ⫽ a’, b ⫽ b’, and so on. No gender moderation
was found, so we report the results of the analysis with equality
constraints.) The second model, called the partial-mediation
model, was a less-restricted model that added all of the previously
described indirect paths to the model depicted in Figure 1.
On its own, the full-mediation model was a good fit to the data:
␹2(22, N ⫽ 108) ⫽ 25.1, p ⫽ .29, NFI ⫽ .90, RMSEA ⫽ .036.
However, compared with the partial-mediation model, the full-
mediation model’s fit was slightly worse: ⌬␹2(6, N ⫽ 108) ⫽ 12.7,
p ⫽ .05. When we examined the individual paths in the partial-
mediation model, we found that all of the effects specified in the
full-mediation model were still significant. In addition, however,
optimism had a direct effect on a partner’s conflict resolution (i.e.,
its effect was partially but not wholly explained by the mediating
variables): direct partner effect ⫽ .12, p ⫽ .04, 95% CI ⫽ (.01,
.26).
Control Analyses
We separately analyzed each link in the double-mediation
model controlling for Extraversion, Neuroticism, self-esteem, part-
Figure 1. The full-mediation model for Part 2. Error variances are not shown; men’s and women’s error
variances for the same measure were allowed to covary.
ners’ ages, length of relationship, cohabitation status, relationship
satisfaction, intensity of disagreement, and which partner’s topic
was (randomly) selected by the experimenter. All of the links in
the double-mediation model remained significant when each of
these control variables was included.
Part 2 showed that optimism was associated not just with global
relationship satisfaction but also with how well both partners
perceived their engagement and resolution of a significant area of
conflict in a relationship. This effect seemed to be partially driven
by optimists’ tendency to perceive their partners as supportive,
which not only led optimists to engage more positively in discuss-
ing the conflict (according to both optimists and their partners) but
also elicited more positive engagement from their partners as well.
Our emphasis on the participants’ reports of engagement in the
conflict discussion and resolution of the conflict allowed us to gain
valuable insights into their relationships. By asking both partners
for their assessments of their own engagement, their partners’
engagement, and the conflict resolution, we were able to assess the
shared social reality of these intimate relationships. Nevertheless,
if optimism affects the shared reality of a relationship, then at some
point that shared reality might affect outcomes that are objectively
verifiable. For Part 3, we examined what is literally the ultimate
relationship outcome: relationship dissolution.
Part 3: The 1-Year Follow-Up
Parts 1 and 2 demonstrated that optimism was associated with
relationship satisfaction and subjective conflict resolution, largely
owing to the association between optimism and the perception of
greater social support. These findings suggest that optimism influ-
ences relationship processes relevant to relationship maintenance
and survival. In Part 3, we examined whether the dating couples
were still together 1 year after the initial phases of the study.
Because dating relationships are not as enduring as marriages, it
was reasonable to expect that enough relationships would have
ended that we could test for effects of optimism on relationship
longevity. We hypothesized that optimism would be associated
with relationship status at the 1-year follow-up and that this effect
would be mediated by perceived support.
Method
We attempted to contact all original participants via e-mail 1 year after
their participation in Part 2. Because data collection for Part 2 spanned
several months, we contacted couples within 1 week of the 1-year anni-
versary of their participation in Part 2. If neither member of a couple
 OPTIMISM IN CLOSE RELATIONSHIPS
responded within that week, we contacted them a second and third time, via
e-mail and phone. Through this procedure, we were able to obtain rela-
tionship status information from at least one member of 101 (94%) of the
couples. Analyses showed that members of responding couples and non-
responding couples did not differ significantly on measures of optimism or
relationship satisfaction.
We asked all participants whether they were still in an exclusive dating
relationship with their partner. Couples responding yes were coded as still
together at 1 year (1), and couples reporting no were coded as having
broken up (0). In the responding sample, 67 couples (66%) were still
together at the 1-year follow-up, and 34 had broken up.
Results and Discussion
Did Optimism Predict Relationship Status at 1 Year?
We hypothesized that greater optimism would be associated
with a higher probability of being together at a 1-year follow-up.
To test this hypothesis, we performed a logistic regression with
couple as the unit of analysis, using both male and female opti-
mism to predict the couple’s 1-year status. Greater male optimism
predicted relationship survival, B ⫽ 0.03, Wald ⫽ 7.53, p ⫽ .006,
but female optimism did not predict relationship survival, B ⫽
0.01, Wald ⫽ 0.48, p ⫽ .49. To illustrate this effect, we split the
couples into two groups according to the male optimism median
and examined survival for each group. We found that 75% of
couples with men at or above the median were still together at 1
year, contrasted with 54% of couples with men below the median.
Did Perceived Support Mediate the Effects of Optimism
on Relationship Longevity?
To evaluate whether perceived support mediated the effect of
optimism on relationship longevity, we added male and female
perceived support as predictors in the logistic regression model. In
this second model, male perceived support was a significant pre-
dictor, B ⫽ .06, Wald ⫽ 6.21, p ⫽ .01, the effect of male optimism
was reduced, B ⫽ .02, Wald ⫽ 3.28, p ⫽ .07, and female optimism
and perceived support were not significant predictors ( ps ⬎ .75).
We evaluated the reduction in the effect of male optimism by
computing a bootstrap confidence interval for the difference be-
tween the male optimism effect in the first and second models; the
confidence interval did not include a null effect, 95% CI ⫽ (.001,
.024), consistent with mediation. The effects of men’s optimism
and perceived support remained significant when controlling for
both partners’ Extraversion, Neuroticism, and self-esteem.
Analyses controlling for relationship satisfaction were consis-
tent with our main conclusions, though they yielded somewhat
more complicated results. In a logistic regression in which men’s
and women’s relationship satisfaction were the only predictors,
there was a significant effect of men’s relationship satisfaction,
B ⫽ .04, Wald ⫽ 7.59, p ⫽ .006, but not of women’s relationship
satisfaction, B ⫽ ⫺.01, Wald ⫽ 0.27, p ⫽ .61. In a regression that
included both optimism and relationship satisfaction, there were
significant effects both of men’s optimism ( p ⫽ .03) and men’s
relationship satisfaction ( p ⫽ .03), indicating that men’s relation-
ship satisfaction did not account for the effect of men’s optimism
on longevity. When we ran the full mediational model with a
control for relationship satisfaction, the results were in the gener-
ally expected direction but were not as clear as without the con-
trols. In this analysis, neither men’s nor women’s relationship
151
satisfaction was a significant predictor of relationship longevity
( ps ⬎ .42). The effect of men’s perceived support was in the
expected direction but was not statistically significant ( p ⫽ .10).
Given the sample size, it should perhaps not be surprising that in
a model with six moderately intercorrelated predictors, the pre-
dicted effect was only marginally significant.
Part 3 showed that optimism is associated with an important
social outcome: relationship survival. Intriguingly, this analysis
showed a sex difference, with men’s optimism being the important
predictor; this was in contrast to the other analyses, which indi-
cated that men’s and women’s optimism did not have different
effects on perceived support, relationship satisfaction, or conflict
resolution processes.
Why might male optimism play a more important role than
female optimism in predicting relationship survival? One possible
explanation has to do with the nature of men’s and women’s social
support networks. A number of studies have suggested that men
tend to rely more heavily on romantic partners for social support,
whereas women tend to draw on a wider network of family and
friends (e.g., Taylor et al., 2000; Voss, Markiewicz, & Doyle,
1999; Walen & Lachman, 2000). Thus, the tendency of male
pessimists to perceive their partners as less supportive might be
especially likely to produce shorter lived relationships, because for
men, negative perceptions of their partners would implicate their
entire support system and give them greater incentive to terminate
the relationship.
General Discussion
In a longitudinal study of dating couples, we found that opti-
mism was associated with better relationship outcomes in a num-
ber of domains. Part 1 found that optimists and their partners both
experienced greater overall relationship satisfaction; Part 2 found
that optimists and their partners saw themselves and each other as
engaging more positively in the conflict and as reaching a better
resolution; and Part 3 found that the relationships of male optimists
lasted longer than the relationships of male pessimists. Further-
more, all of the relationship consequences of optimism were me-
diated by optimists’ tendency to perceive their partners as
supportive.
How Does Perceived Support Affect the Relational
Environment?
Why should optimism be an asset in close relationships? This
study provided some insight into why optimism may lead to more
satisfying and longer lasting relationships by identifying perceived
support as a mediator. Perceived support was hypothesized to be
an important relational mediator because it creates a more adaptive
relational environment. We believe that perceived support proba-
bly helps relationships in a variety of ways. First, optimists’
tendency to perceive their partners as supportive may act as a
buffer against negative attributions. Relationships in which indi-
viduals attribute their partners’ negative behaviors to global, sta-
ble, voluntary dispositions rather than narrow and temporary in-
clinations tend to be marked by lower relationship satisfaction and
other maladaptive outcomes (Bradbury & Fincham, 1990). Opti-
mists may attribute specific instances of unsupportive or ambigu-
ous behavior to temporary and situationally limited states. Second,
optimists’ positive views of their partners may prevent or interrupt
152 SRIVASTAVA, MCGONIGAL, RICHARDS, BUTLER, AND GROSS
cycles of negative reciprocity by refocusing optimists’ attention on
the constructive things that their partners do and say, instead of on
their partners’ negative affect (Gottman, 1998). Third, optimists
may be better at acting as a “secure base” (Bowlby, 1988; Hazan
& Shaver, 1987), providing their partners with a reliable source of
support. As a result, optimists and their partners may be more
satisfied because they feel that their relationship helps them pur-
sue their personal goals (Brunstein et al., 1996; Kaplan &
Maddux, 2002).
Optimism, Shared Reality, and Positive Illusions
Did optimists and their partners benefit from positive illusions
(Taylor & Brown, 1988)? In conceptualizing this study, we have
sidestepped questions about accuracy and bias, instead focusing on
perceptions of support and on both partners’ assessment of positive
engagement and conflict resolution. Partners’ beliefs and percep-
tions of themselves and each other define the shared reality of a
relationship, a reality that is important independent of any objec-
tive analysis of accuracy (Gable et al., 2003). In support of the
value of such an approach, Part 3 suggests that shared reality can
have very real consequences for the long-term success of a
relationship.
How might we apply a different perspective to our findings?
Drawing on a positive illusions perspective, we could interpret the
results in Part 1 as stemming from positive illusions that optimists
hold about their relationships. In Part 2, we could conclude that
such illusions drive optimists to practice and elicit “objectively”
better conflict-related behavior; or alternatively, perhaps both part-
ners share an illusion about how they handle conflict. We believe
that the former interpretation is more compatible with other find-
ings about positive illusions in close relationships (e.g., Murray &
Holmes, 1997), though the present data cannot speak strongly to
questions of illusion or accuracy.
Relatedly, perceived support may function as a self-fulfilling
prophecy: By virtue of optimists’ general tendency to see their
partners as supportive, they may elicit actual support from their
partners. That would explain why optimists report using social
support as a coping strategy in general (Scheier, Carver, &
Bridges, 1994) but not in response to specific, everyday stressors
(Aspinwall & Taylor, 1992; Brissette et al., 2002): Optimists
expect to receive support from others, but they do not directly ask
for it.
Limitations and Future Directions
Throughout this article, we have adopted the perspective that
optimism leads to perceived support, which leads to positive
relationship outcomes. This was based in large part on theoretical
considerations: We ordered the constructs from optimism, the most
general and broad-based construct, to perceived support, which is
more domain-specific but still fairly broad as an individual differ-
ence (Sarason et al., 1986), to outcomes relevant to a specific
relationship (Part 1) or specific events within that relationship
(Parts 2 and 3), which are the most contextual. There is also a
temporal logic to the order of precedence: Optimism is a mean-
ingful construct even for an individual who has no experience or
beliefs about close relationships, and general perceptions of sup-
port can preexist specific experiences in relationships. In Part 2,
this temporal ordering was reflected in the design of the study and
was strengthened by an analysis controlling for Time 1 levels of
relationship satisfaction. In Part 3, relationship dissolution obvi-
ously is an outcome that temporally follows all other characteris-
tics of the relationship.
The strengths of our naturalistic approach are balanced against
the limitations of a nonexperimental design, however, and we
acknowledge that this sequencing is not airtight. It could be ar-
gued, for example, that perceptions of support are a consequence,
rather than a cause, of higher quality relationships (Metts, Geist, &
Gray, 1994). From this perspective, individuals may form their
perceptions of their partner’s supportiveness based on some other
aspect of the relationship. We partially addressed this concern by
controlling for relationship satisfaction and other relationship char-
acteristics in Part 2, but we cannot fully rule out the possibility that
some other, unmeasured feature of the relationship acted as a third
variable.
Our reliance on partners’ reports about the conflict conversation
in Part 2 might be regarded as a double-edged sword. As implied
earlier, this approach gave us insight into the shared reality of the
relationship, and the results showed that both self- and partner
reports from both members of the couple led to the same conclu-
sions about conversation processes. In fact, if an “objective”
observer failed to corroborate the positive conversation processes
evident in optimists’ relationships, we might have cause to suspect
the observer rather than the couple. Nevertheless, it would be
interesting in future research to examine more objectively the
specific processes that we believe are being promoted by opti-
mists: adaptive attributions, interruption of cycles of negative
reciprocity, and use of the relationship as a secure base. Such
research would elucidate the mechanisms that link optimists’ pos-
itive expectations to the fulfillment of these expectations.
References
Arbuckle, J. L. (1999). Amos 4.0. Chicago: SmallWaters Corporation.
Aspinwall, L. G., & Taylor, S. E. (1992). Modeling cognitive adaptation:
A longitudinal investigation of the impact of individual differences and
coping on college adjustment and performance. Journal of Personality
and Social Psychology, 63, 989 –1003.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable
distinction in social psychological research: Conceptual, strategic and
statistical considerations. Journal of Personality and Social Psychology,
51, 1173–1182.
Barrera, M. Jr. (1986). Distinctions between social support concepts,
measures, and models. American Journal of Community Psychology, 14,
413– 455.
Belsher, G., & Costello, C. G. (1991). Do confidants of depressed women
provide less social support than confidants of nondepressed women?
Journal of Abnormal Psychology, 100, 516 –525.
Bowlby, J. (1988). A secure base: Parent– child attachment and healthy
human development. New York: Basic Books.
Bradbury, T. N., & Fincham, F. D. (1990). Attributions in marriage:
Review and critique. Psychological Bulletin, 107, 3–33.
Brissette, I., Scheier, M. F., & Carver, C. S. (2002). The role of optimism
in social network development, coping, and psychological adjustment
during a life transition. Journal of Personality and Social Psychology,
82, 102–111.
Brunstein, J. C., Dangelmayer, G., & Schultheiss, O. C. (1996). Personal
goals and social support in close relationships: Effects on relationship
mood and marital satisfaction. Journal of Personality and Social Psy-
chology, 71, 1006 –1019.
Carstensen, L. L., Gottman, J. M., & Levenson, R. W. (1995). Emotional
behavior in long-term marriage. Psychology and Aging, 10, 140 –149.
 OPTIMISM IN CLOSE RELATIONSHIPS
Carver, C. S., Kus, L. A., & Scheier, M. F. (1994). Effects of good versus
bad mood and optimistic versus pessimistic outlook on social acceptance
versus rejection. Journal of Social and Clinical Psychology, 13, 138 –
151.
Chang, E. C. (2001). Optimism and pessimism: Implications for theory,
research, and practice. Washington, DC: American Psychological
Association.
Cohen, P., Cohen, J., Aiken, L. S., & West, S. G. (1999). The problem of
units and the circumstance for POMP. Multivariate Behavioral Re-
search, 34, 315–346.
Coleridge, S. T. (1881). Reason for love’s blindness. In Bartlett’s familiar
quotations. Retrieved from http://www.bartleby.com/100/
Cowan, C. P., & Cowan, P. A. (1990). Couple Communication Question-
naire. Unpublished questionnaire, University of California.
Cozarelli, C. (1993). Personality and self-efficacy as predictors of coping
with abortion. Journal of Personality and Social Psychology, 65, 1224 –
1236.
Dehle, C., Larsen, D., & Landers, J. E. (2001). Social support in marriage.
American Journal of Family Therapy, 29, 307–324.
Dougall, A. L., Hyman, K. B., Hayward, M. C., McFeeley, S., & Baum, A.
(2001). Optimism and traumatic stress: The importance of social support
and coping. Journal of Applied Social Psychology, 31, 223–245.
Gable, S. L., Reis, H. T., & Downey, G. (2003). He said, she said: A
quasi-signal detection analysis of daily interactions between close rela-
tionship partners. Psychological Science, 14, 100 –105.
Geers, A. L., Reilley, S. P., & Dember, W. N. (1998). Optimism, pessi-
mism, and friendship. Current Psychology: Developmental, Learning,
Personality, Social, 17, 3–19.
Gottman, J. M. (1998). Psychology and the study of the marital processes.
Annual Review of Psychology, 49, 169 –197.
Gottman, J. M., Markman, H., & Notarius, C. (1977). The topography of
marital conflict: A sequential analysis of verbal and nonverbal behavior.
Journal of Marriage and the Family, 39, 461– 477.
Hazan, C., & Shaver, P. (1987). Romantic love conceptualized as an
attachment process. Journal of Personality and Social Psychology, 52,
511–524.
Helgeson, V. S. (1994). The effects of self-beliefs and relationship beliefs
on adjustment to a relationship stressor. Personal Relationships, 1,
241–258.
John, O. P., & Srivastava, S. (1999). The Big-Five trait taxonomy: History,
measurement, and theoretical perspectives. In L. A. Pervin & O. P. John
(Eds.), Handbook of personality: Theory and research (2nd ed., pp.
102–138). New York: Guilford Press.
Kaplan, M., & Maddux, J. E. (2002). Goals and marital satisfaction:
Perceived support for personal goals and collective efficacy for collec-
tive goals. Journal of Social and Clinical Psychology, 21, 157–164.
Kashy, D. A., & Kenny, D. A. (1997). The analysis of data from dyads and
groups. In H. Reis & C. M. Judd (Eds.), Handbook of research methods
in social psychology (pp. 451– 477). New York: Cambridge University
Press.
Lakey, B., & Cassady, P. B. (1990). Cognitive processes in perceived
support. Journal of Personality and Social Psychology, 59, 337–348.
Lakey, B., McCabe, K. M., Fisicaro, S. A., & Drew, J. B. (1996). Envi-
ronmental and personal determinants of support perceptions: Three
generalizability studies. Journal of Personality and Social Psychology,
70, 1270 –1280.
Lund, M. (1985). The development of investment and commitment scales
for predicting continuity of personal relationships. Journal of Social and
Personal Relationships, 2, 3–23.
Metts, S., Geist, P., & Gray, J. L. (1994). The role of relationship charac-
teristics in the provision and effectiveness of supportive messages
153
among nursing professionals. In B. R. Burleson, T. L. Albrecht, & I. G.
Sarason (Eds.), Communication of social support (pp. 113–135). Thou-
sand Oaks, CA: Sage.
Murray, S. L., & Holmes, J. G. (1997). A leap of faith? Positive illusions
in romantic relationships. Personality and Social Psychology Bulletin,
23, 586 – 604.
Newcomb, M. D. (1990). What structural equation modeling can tell us
about social support. In B. R. Sarason, I. G. Sarason, & G. R. Pierce
(Eds.), Social support: An interactionist view (pp. 26 – 63). New York:
Wiley.
Park, C. L., & Folkman, S. (1997). Stability and change in psychosocial
resources during caregiving and bereavement in partners of men with
AIDS. Journal of Personality, 65, 421– 447.
Pierce, G. R., Sarason, B. R., & Sarason, I. G. (1992). General and specific
support expectations and stress as predictors of perceived supportive-
ness: An experimental study. Journal of Personality and Social Psychol-
ogy, 63, 297–307.
Richards, J. M., Butler, E. A., & Gross, J. J. (2003). Emotion regulation in
romantic relationships: The cognitive consequences of concealing feel-
ings. Journal of Social and Personal Relationships, 20, 599 – 620.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton,
NJ: Princeton University Press.
Sarason, I. G., Levine, H. M., Basham, R. B., & Sarason, B. R. (1983).
Assessing social support: The Social Support Questionnaire. Journal of
Personality and Social Psychology, 44, 127–139.
Sarason, I. G., Sarason, B. R., Shearin, E. N. (1986). Social support as an
individual difference variable: Its stability, origins, and relational as-
pects. Journal of Personality and Social Psychology, 50, 845– 855.
Scheier, M. F., & Carver, C. S. (1985). Optimism, coping, and health:
Assessment and implications of generalized outcome expectancies.
Health Psychology, 4, 219 –247.
Scheier, M. F., Carver, C. S., & Bridges, M. W. (1994). Distinguishing
optimism from neuroticism (and trait anxiety, self-mastery, and self-
esteem): A reevaluation of the Life Orientation Test. Journal of Person-
ality and Social Psychology, 67, 1063–1078.
Scheier, M. F., Carver, C. S., & Bridges, M. W. (2001). Optimism,
pessimism, and psychological well-being. In E. C. Chang (Ed.), Opti-
mism and pessimism: Implications for theory, research, and practice
(pp. 189 –216). Washington, DC: American Psychological Association.
Shrout, P. E., & Bolger, N. (2003). Mediation in experimental and non-
experimental studies: New procedures and recommendations. Psycho-
logical Methods, 7, 422– 445.
Stafford, L., & Canary, D. J. (1991). Maintenance strategies and romantic
relationship type, gender and relational characteristics. Journal of Social
and Personal Relationships, 8, 217–242.
Taylor, S. E., & Brown, J. D. (1988). Illusion and well-being. A social
psychological perspective on mental health. Psychological Bulletin, 103,
193–210.
Taylor, S. E., Kemeny, M. E., Reed, G. M., Bower, J. E., & Gruenewald,
T. L. (2000). Psychological resources, positive illusions, and health.
American Psychologist, 55, 99 –109.
Voss, K., Markiewicz, D., Doyle, A. B. (1999). Friendship, marriage and
self-esteem. Journal of Social and Personal Relationships, 16, 103–122.
Walen, H. R., & Lachman, M. E. (2000). Social support and strain from
partner, family, and friends: Costs and benefits for men and women in
adulthood. Journal of Social and Personal Relationships, 17, 5–30.
Received August 11, 2003
Revision received February 6, 2006
Accepted February 7, 2006 䡲
 SOCIAL EXCLUSION AND PAIN
ness, lethargy, lack of emotion, and self-awareness. Journal of Person-
ality and Social Psychology, 85, 409 – 423.
Uchino, B. N., Cacioppo, J. T., & Kiecolt-Glaser, J. K. (1996). The
relationship between social support and physiological processes: A
review with emphasis on underlying mechanisms and implications for
health. Psychological Bulletin, 119, 488 –531.
Van Boven, L., & Loewenstein, G. (2003). Social projection of transient
drive states. Personality and Social Psychology Bulletin, 29, 1159 –
1168.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and vali-
dation of brief measures of positive and negative affect: The PANAS
scales. Journal of Personality and Social Psychology, 54, 1063–1070.
Wiedenmayer, C. P., Goodwin, G. A., & Barr, G. A. (2000). The effect of
periaqueductal gray lesions on responses to age-specific threats in infant
rats. Developmental Brain Research, 120, 191–198.
Williams, K. D., Cheung, C. K. T., & Choi, W. (2000). CyberOstracism:
Effects of being ignored over the Internet. Journal of Personality and
Social Psychology, 79, 748 –762.
15
Wilson, T. D., & Gilbert, D. T. (2003). Affective forecasting. In M. P.
Zanna (Ed.), Advances in experimental social psychology (Vol. 35, pp.
345– 411). San Diego, CA: Academic Press.
Wilson, T. D., Wheatley, T., Kurtz, J., Dunn, E., & Gilbert, D. T. (2004).
When to fire: Anticipatory versus post-event reconstrual of uncontrol-
lable events. Personality and Social Psychology Bulletin, 30, 340 –351.
Wilson, T. D., Wheatley, T., Meyers, J. M., Gilbert, D. T., & Axsom, D.
(2000). Focalism: A source of durability bias in affective forecasting.
Journal of Personality and Social Psychology, 78, 821– 836.
Zadro, L., Williams, K. D., & Richardson, R. (2004). How low can you go?
Ostracism by a computer is sufficient to lower self-reported levels of
belonging, control, self-esteem, and meaningful existence. Journal of
Experimental Social Psychology, 40, 560 –567.
Received February 24, 2005
Revision received August 15, 2005
Accepted August 25, 2005 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 154 –170 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.154

Discrepancies Between Explicit and Implicit Self-Concepts: Consequences

for Information Processing

Pablo Briñol Richard E. Petty

Universidad Autónoma de Madrid Ohio State University

S. Christian Wheeler
Stanford University

Individuals with discrepancies among their explicit beliefs often engage in greater elaboration of
discrepancy-related information in a presumed attempt to reduce the discrepancy. The authors predicted
that individuals with discrepancies between explicit and implicit self-conceptions might similarly be
motivated to engage in processing of discrepancy-related information, even though they might not be
aware of the discrepancy. Four studies were conducted in which various self-dimensions were assessed
with explicit and implicit measures. Across several different self-dimensions (e.g., need to evaluate,
self-esteem), the authors found that as the discrepancy between the explicit and implicit measure
increased (regardless of direction), people engaged in more thinking about information framed as related
to the self-dimension on which the discrepancy existed. This research suggests that individuals might be
motivated to examine relevant information as a strategy to minimize the implicit doubt that accompanies
an inconsistency between explicit and implicit self-conceptions.

Keywords: discrepancy, explicit–implicit, self, persuasion, attitude change

The psychological literature has clearly documented that people
can simultaneously hold incompatible explicit beliefs, attitudes,
feelings, and behavioral tendencies regarding oneself and others
(e.g., Abelson & Rosenberg, 1958; Bem & Allen, 1974; Brehm &
Cohen, 1962; Cacioppo, Gardner, & Berntson, 1997; Heider, 1958;
Higgins, 1987; Kaplan, 1972; Newcomb, 1968; Norton, 1975;
Osgood & Tannenbaum, 1955; Priester & Petty, 1996). Virtually
every relevant theory holds that such internal discrepancies tend to
be unpleasant and can result in psychologically undesirable out-
comes (e.g., Campbell, 1990; Carver & Scheier, 1990; Greenier,
Kernis, & Waschull, 1995; Higgins, 1987; Kernis & Waschull,
1995). Because of this, people often attempt to resolve these
internal discrepancies. Perhaps the most common approach to
addressing discrepancy is enhanced thinking or information pro-
cessing (e.g., Abelson et al., 1968; Aronson, 1969; Festinger,
1957; Heider, 1958; Hass, Katz, Rizzo, Bailey, & Moore, 1992).
By considering additional information, individuals may hope to
gain enough information for one or the other side of the discrep-
ancy to resolve or minimize the inconsistency, or at least the
Pablo Briñol, Department of Social Psychology and Methodology, Uni-
versidad Autónoma de Madrid, Madrid, Spain; Richard E. Petty, Depart-
ment of Psychology, Ohio State University; S. Christian Wheeler, Graduate
School of Business, Stanford University.
We thank all of the members of the 2000 –2004 Groups for Attitudes and
Persuasion at Ohio State University for providing feedback on the exper-
iments reported in this article. We also thank Javier Horcajo and Isamel
Gallardo for their help with data collection.
Correspondence concerning this article should be addressed to Richard
E. Petty, Department of Psychology, Ohio State University, 1835 Neil
Avenue Mall, Columbus, OH 43210-1222. E-mail: petty.1@osu.edu
subjective discomfort that results from the discrepancy (e.g.,
Hänze, 2001; Hodson, Maio, & Esses, 2001; Jonas, Diehl, &
Bromer, 1997; Katz, Wackenhut, & Hass, 1986).
For example, Maio, Bell, and Esses (1996) measured partic-
ipants’ ambivalence regarding the issue of immigration to Can-
ada (i.e., the extent to which they had both positive and negative
reactions to the issue) and then exposed them to a discrepancy-
related message favoring immigration from Hong Kong to
Canada that contained either strong or weak arguments. The
extent to which participants processed the message information
was assessed by examining the extent to which the quality of
the arguments affected postmessage immigration attitudes
(Petty, Wells, & Brock, 1976). When people are thinking care-
fully about information, they should be affected by the quality
of the arguments a message contains (see Petty & Cacioppo,
1986). As expected, Maio et al. (1996) found that individuals
who had ambivalent attitudes toward immigration were more
influenced by argument quality than were unambivalent indi-
viduals, suggesting that they engaged in enhanced scrutiny of
the information.
Although research has focused extensively on explicit discrepan-
cies, relatively little work has examined the potential existence of and
consequences of discrepancies in which one cognitive element may
not be easily reported. Theory suggests that just as people can hold
conscious, explicit self-beliefs, they may also hold less conscious (or
explicitly denied) automatic self-associations that can conflict with the
more consciously endorsed ones (e.g., Greenwald et al., 2002). The
present research examines the information-processing consequences
of discrepancies between self-conceptions as assessed with explicit
and implicit measures of individual differences.

154
 EXPLICIT–IMPLICIT DISCREPANCY
The Relationship Between the Implicit and the Explicit
Self-Concept
Compared with an explicit or conscious self-conception, an
implicit self-view is generally defined as one that is held
outside of conscious awareness, or at least is an association that
is not endorsed consciously (Fazio & Olson, 2003; Petty,
Wheeler, & Tormala, 2003). Such self-conceptions can influ-
ence judgment and action automatically without the need for
reflection and may only be apparent on disguised or implicit
measures. Considerable research has examined the predictive
utility of such measures in the domains of motivation (e.g.,
McClelland, 1985), memory (e.g., Lindsay & Johnson, 1989),
personality (e.g., Bornstein, 1995), self-esteem (e.g., Green-
wald & Banaji, 1995; Hetts, Sakuma, & Pelham, 1999), atti-
tudes (e.g., Greenwald, McGhee, & Schwartz, 1998; Fazio,
Jackson, Dunton, & Williams, 1995), and stereotypes (e.g.,
Blair & Banaji, 1996; Dovidio, Kawakami, & Beach, 2001).
Currently the most popular implicit measures of self-conception
are based on reaction times, for example, the Implicit Associ-
ation Test (IAT; Greenwald et al., 1998) and the Automatic
Evaluation Task (Fazio et al., 1995). These measures capitalize
on automatic associations between the self and other con-
structs—reactions that may occur too quickly to come under
deliberative control (see Petty, Fazio, & Briñol, in press, for
reviews).1
Sometimes, what is assessed with implicit and explicit measures
is the same, suggesting that there is just one underlying motive,
attitude, personality, or self-conception that is open to conscious
awareness. However, explicit and implicit assessments are not
always congruent. Although there are various explanations for this
incongruency, some researchers have developed theoretical frame-
works that account for the divergence by proposing that explicit
and implicit constructs sometimes stem from two independent
systems that operate in different contexts and influence different
types of behavior. For example, Dovidio et al. (2001) have argued
that self-report (explicit) and response latency (implicit) measures
of attitudes can tap into different evaluations that predict behaviors
in different situations (spontaneous vs. deliberative; see also Carl-
ston & Skowronski, 1986; Gawronski & Bodenhausen, in press;
Wilson, Lindsey, & Schooler, 2000). A similar distinction has
been made in the domain of motivation (see, e.g., McClelland,
1985; Patten & White, 1977).
Indeed, Carver (2005) documented that psychologists have pro-
mulgated a wide array of dual system approaches (e.g., Lieberman,
2000; Metcalfe & Mischel, 1999; Shastri & Ajjanagadde, 1993;
Sloman, 1996; Smith & DeCoster, 2000; Smolensky, 1988). Al-
though the various frameworks differ in some details, each has
tended to emphasize the fact that some behavior is guided in a
relatively automatic and unconscious way, whereas other behavior
is guided in a more deliberative and conscious way. For example,
cognitive– experiential self-theory (Epstein, 1973, 2003) argues
that humans are characterized by two independent, interactive
systems of thinking that jointly determine behavior: a preconscious
experiential system and a conscious rational system. Similarly, the
reflective–impulsive model (Strack & Deutsch, 2004) postulates
two systems— one thoughtful and one more automatic—that op-
erate in parallel and interact with one another.
155
Simultaneous Activity of Explicit–Implicit Self-Discrepant
Dimensions
If discrepant explicit and implicit constructs always operated in
different situations (e.g., spontaneous vs. deliberative), then they
should lead to little conflict in any given situation. But if the two
self-conceptions were ever jointly activated or operative, there
would be the possibility of some consequences that mirror incon-
gruency between discrepant explicit constructs (Petty et al., 2006).
In fact, there is some evidence suggesting that possessing discrep-
ant implicit and explicit self-concepts can be consequential. For
example, Shedler, Mayman, and Manis (1993) studied individuals
who reported minimal emotional disturbance on Eysenck’s Neu-
roticism scale (explicit measure) but who were simultaneously
judged on the basis of their early memories (implicit measure) to
be relatively disturbed. In comparison with participants in the
genuine self-esteem group (who scored as healthy on both mea-
sures), those with explicit–implicit discrepancies were signifi-
cantly more reactive on a combined index of heart rate and blood
pressure and scored higher on behavioral indices of anxiety (see
also Weinberger & Haradaway, 1990).
In conceptually similar research, Zelenski and Larsen (2003)
found that having incongruent explicit (e.g., self-ratings) and im-
plicit (measured by the Thematic Apperception Test; Proshansky,
1943) motive profiles was associated with reduced emotional
well-being. Other recent research has demonstrated that people
who scored relatively high on an explicit measure of self-esteem,
but relatively low on an implicit measure (the IAT), exhibited the
most self-aggrandizement across different indices (Bosson,
Brown, Zeigler-Hill, & Swann, 2003), which is the main charac-
teristic of a narcissistic personality (e.g., Wing & Gough, 1990).
Additionally, individuals with the combination of relatively high
scores on explicit measures of self-esteem and relatively low
scores on implicit measures have been shown to be particularly
defensive (Jordan, Spencer, Zanna, Hoshino-Browne, & Correll,
2003).
Finally, we have found that discrepancy between explicit and
implicit self-esteem scores is associated with implicit but not
explicit self-doubt (Briñol, Petty, & Wheeler, 2003). Specifically,
as explicit–implicit self-esteem discrepancy (as assessed using the
absolute value of the difference between participants’ standardized
explicit and implicit self-esteem scores) increased, the strength
with which participants automatically associated doubt words with
self-words on an IAT also increased. However, increased discrep-
ancy was not associated with explicit reports of self-uncertainty.
Similarly, recent research has shown that when people’s attitudes
change on explicit measures, they show more doubt on an implicit
but not explicit measure of confidence regarding the attitude object
compared with people whose attitudes have not changed (Petty et
al., 2006). Because recent research also shows that when attitudes
1
When not compatible with the endorsed (explicit) self-conception, the
self-associations measured with contemporary implicit measures can rep-
resent many things (e.g., repressed evaluations of oneself, prior evaluations
of oneself, others’ evaluations of oneself, hopes for oneself, strong societal
prescriptions for self-conduct, and so forth). We postulate that regardless of
the particular source of the association, inconsistency with one’s con-
sciously endorsed self-view can lead to psychological conflict (see Petty, in
press; Petty, Tormala, Briñol, & Jarvis, 2006; Priester & Petty, 2001).
156 BRIÑOL, PETTY, AND WHEELER
change, a residue of the prior attitude may still be available on an
implicit (automatic) measure (e.g., Gregg, Seibt, & Banaji, 2006),
the enhanced implicit doubt that accompanies attitude change is
plausibly due to a discrepancy between old (implicit/automatic)
and new (explicit/deliberative) evaluations.
Together, these studies suggest that having discrepant explicit
and implicit self-dimensions is associated with numerous conse-
quences that often appear to be negative, unpleasant, or dysfunc-
tional. Because of this, people with such discrepancies should be
motivated to engage in some discrepancy resolution. Specifically,
we postulate that like possession of an explicit discrepancy (e.g.,
attitudinal ambivalence), possession of a discrepancy between
implicit and explicit self-conceptions will be associated with at-
tempts at discrepancy reduction, and that this would occur regard-
less of the direction of the discrepancy. As described earlier, by
thinking about information presumed relevant to the issue on
which there is a discrepancy, individuals with explicit–implicit
discrepancies can possibly restore internal consistency. Thus, the
purpose of the present research was to examine the influence of
explicit–implicit discrepancies on information processing. We ex-
amined conditions under which the discrepancy results both from
one’s explicit self-conception being relatively higher (more posi-
tive) than one’s implicit self-conception within the sample distri-
bution as well as from one’s explicit self-conception being rela-
tively lower (less positive) than one’s implicit self-conception.
Overview of the Present Research
We hypothesized that to reduce a discrepancy between explicit
and implicit self-conceptions, individuals with such discrepancies
would engage in more effortful elaboration of information pre-
sumed relevant to the self-dimension on which the discrepancy
exists. To test this hypothesis, we conducted four studies in which
explicit and implicit assessments of the self along four different
dimensions were collected. In each study we used a standard
self-report as the explicit measure and an IAT as the implicit
measure. The IAT was selected because (a) it taps into automatic
associations that are less subject to conscious control and (b)
previously validated procedures were available for some of the
dimensions of interest. We then examined the impact of explicit–
implicit divergence on the extent of processing of information
framed as discrepancy related.
The logic was similar across the four studies. In each study, an
index of the relative explicit–implicit discrepancy was formed for
each participant following procedures used in prior research (see
Kehr, 2004, for a discussion). Then, all participants were exposed
to a persuasive message. The extent to which participants pro-
cessed the information in the message was assessed by using an
information quality manipulation. As noted earlier, the impact that
the quality of information in a message has on resulting attitudes
is a widely used indicator of the extent to which individuals
carefully attend to and think about the information to which they
are exposed (Petty & Cacioppo, 1986).
Finally, across studies we varied whether the information con-
tained in the message was actually relevant (Study 1) or simply
framed as relevant (Studies 2, 3, and 4) to the self-dimension on
which discrepancy existed. In two studies (Studies 3 and 4), we
manipulated the extent to which participants expected the message
to be relevant to the issue on which discrepancy existed, even
though the message actually contained discrepancy-unrelated in-
formation. This manipulation was expected to moderate the impact
of discrepancy on information processing. If thinking is aimed at
reducing the discrepancy, then explicit–implicit incongruence
should be associated with enhanced thinking only when partici-
pants expect the message to be related to the self-dimension on
which there is a discrepancy (i.e., when it presumably could help
reduce the discrepancy). When the same information is framed as
unrelated, it should be useless in resolving the discrepancy.
Thus, in studies in which the information was presumed relevant
to the discrepancy (Studies 1 and 2), we predicted a two-way
interaction on attitudes between size of the explicit–implicit dis-
crepancy and information quality. Information quality should have
a greater impact on attitudes as discrepancy increases. In studies in
which we manipulated the presumed relevance of the information
to the discrepancy, a three-way interaction on attitudes was ex-
pected. Specifically, discrepancy should affect information pro-
cessing primarily when the information is framed as relevant. We
did not expect these results to depend on the direction of the
discrepancy (i.e., whether implicit scores were relatively higher or
lower than explicit scores). To the extent these hypotheses are
supported, it would suggest that the presence of discrepant explicit
and implicit self-dimensions, regardless of direction, is associated
with enhanced thinking. The enhanced thinking presumably re-
flects an attempt at discrepancy reduction.
Experiment 1
Experiment 1 was conducted to provide an initial test of the
notion that divergence between explicit and implicit self-views can
influence the extent of processing of information relevant to those
self-views. In our first experiment, we used the trait of shyness as
a specific dimension of personality. There were several reasons for
this selection. First, shyness is a personality trait that is well
represented in common language and lay psychology, is easily
judged by oneself, and is readily observable by others (Asendorpf,
1987, 1989). Second, because of its observability and ubiquity in
lay psychology, the self-concept dimension of shyness is relatively
accessible (Asendorpf, 1990). Third, it is easy to select shyness-
descriptive adjectives for both explicit self-ratings and an IAT for
shyness, because previous research has pretested instruments for
its assessment (Asendorpf, 1987, 1989). Last, research conducted
by Asendorpf, Banse, and Mücke (2002) has demonstrated that the
implicit measure of shyness uniquely predicted spontaneous (but
not controlled) shy behavior, whereas the explicit ratings uniquely
predicted controlled (but not spontaneous) shy behavior. That is,
Asendorpf et al. validated the implicit and explicit shyness mea-
sures used in Study 1 as useful independent indicators of behavior.
The explicit shyness measure consisted of a series of self-
reported responses to shyness-related adjectives, whereas the im-
plicit measure was a shyness IAT (see Asendorpf et al., 2002). In
this study, all of the participants were exposed to a persuasive
message containing either strong or weak arguments directly re-
lated to shyness. After reading the message, participants were
asked to report their attitudes toward the proposal in the message.
We expected participants with a large explicit–implicit discrep-
ancy to be more attentive to the persuasive message than those
with a small discrepancy. This enhanced thinking would be evi-
 EXPLICIT–IMPLICIT DISCREPANCY
denced in greater attitudinal responsiveness to the argument qual-
ity manipulation.
Method
Participants and Design
Eighty-one undergraduate psychology students at the Universidad Au-
tónoma de Madrid in Madrid, Spain, participated in partial fulfillment of a
course requirement. The students were randomly assigned to the argument
quality conditions (strong or weak). Additionally, explicit and implicit
shyness were assessed for all the participants to form an index of explicit–
implicit discrepancy. The independent variables thereby constituted an
Argument Quality (strong vs. weak) ⫻ Explicit–Implicit Discrepancy
(continuously scored) ⫻ Direction of Discrepancy (higher on explicit or
implicit measure) design.
Procedure
Upon arrival, participants were seated at individual computer stations
and completed the IAT for shyness. After the IAT task, all of the partic-
ipants reported their explicit shyness and completed several other ancillary
questions. Participants were then told that because of extra time remaining
in the session, they would also be participating in another experiment
designed to examine personality characteristics that were good for being a
psychologist. Participants received a message arguing that being shy was a
positive trait for a psychologist. The message contained either strong or
weak arguments. All of the participants were told in advance that the
message they were about to read had to do with shyness. After reading the
message, participants were told that it was important to know what their
personal views were on the benefits of shyness. Thus, they completed a
measure of their attitudes toward shyness as a trait. Finally, the participants
were thanked, asked for permission for their responses to be analyzed (all
gave permission), and given an appointment for a meeting to provide them
with feedback about their results.
Independent Variables
Argument quality. Participants were exposed to a message containing
information directly relevant to shyness and thus to any explicit–implicit
discrepancy on this trait. The shyness-related message participants received
contained either strong or weak arguments. This manipulation was de-
signed to assess the extent to which people were attentive to the content of
the message (Petty & Cacioppo, 1986). The arguments selected were
pretested and were shown to produce the appropriate pattern of cognitive
responding. That is, the strong arguments elicited mostly favorable
thoughts and the weak arguments elicited mostly unfavorable thoughts
when people were instructed to think carefully about them. In brief, 58
students were asked to list their thoughts for each version of the message.
Analysis of the thoughts listed revealed that, on average, the strong
message elicited more favorable (M ⫽ 1.86, SD ⫽ 1.80) than unfavorable
(M ⫽ 0.65, SD ⫽ 1.07) thoughts, t(28) ⫽ 5.54, p ⬍ .001. For the weak
version of the message, participants generated more unfavorable (M ⫽
2.58, SD ⫽ 1.93) than favorable (M ⫽ 0.86, SD ⫽ 1.57) thoughts, t(29) ⫽
7.19, p ⬍ .001. The gist of one strong argument in favor of shyness was
that shy people have more introspective ability, a quality that was highly
valuable in the workplace. The gist of another strong argument was that shy
people have been rated as better friends and partners, because they tend to
have interpersonal relationships that are more sincere, committed, stable,
and satisfactory. In contrast, the gist of one weak argument in favor of
shyness was that shy people tend to talk less than extraverted interviewers,
making other shy people feel more comfortable. The gist of another weak
argument was that some students’ parents prefer that their sons and
157
daughters choose shy people as friends and roommates in college because
they tend to be more self-controlled.
It is important to note that both the strong and weak arguments argued
in favor of shyness, but the strong arguments provided more compelling
reasons than did the weak arguments. This manipulation can be distin-
guished from other forms of message variations, such as arguing either in
favor of or against the proposal. Because the argument manipulation is
used to assess how much thinking people are doing about the message, all
arguments need to argue for the same position— but only with high or low
convincingness. Because both sets of arguments are ostensibly in favor of
the issue, they may be equally persuasive if people do not think about their
implications. Individuals not thinking about the message carefully may
respond simply to the number of arguments presented or their initial gut
reaction to the proposal (e.g., Petty & Cacioppo, 1984; see also Petty &
Wegener, 1998). The more attention paid to the information provided,
however, the greater the difference in subsequent attitudes to strong versus
weak arguments.
Explicit measure of shyness. Shyness was assessed by asking partici-
pants to rate the extent to which 10 adjectives described them on 4-point
scales anchored by extremely uncharacteristic of me (1) and extremely
characteristic of me (4). The adjectives were presented in random order
and included words such as inhibited, insecure, timid, reticent, reserved,
daring, candid, open, secure, and assertive. These words were taken from
Asendorpf et al. (2002) and were also the items used for the implicit
measure in the present research. Ratings on the scale items were highly
consistent with each other (␣ ⫽ .86) and were averaged (reverse scoring
where appropriate) to form a single index of explicit shyness for each
participant.
Implicit measure of shyness. In the IAT measure of shyness, partici-
pants classified target concepts (represented by me or other) and attributes
(represented by shy or nonshy categories of words) using two designated
keys on a computer. The me category was represented by the words I, self,
my, me, and own, whereas the other category was represented by the words
they, them, your, you, and other. Attributes related to shyness were selected
from the items on the explicit measure and included the words inhibited,
insecure, timid, reticent, and reserved. In contrast, nonshy attributes in-
cluded the words daring, candid, open, secure, and assertive. The test was
similar to the original IATs used by Greenwald et al. (1998) and paralleled
the one used by Asendorpf et al. (2002). There were seven blocks of trials.
Blocks 1, 2, and 5 were practice blocks for which participants made single
categorizations (me vs. other, and shy vs. nonshy). In the remaining blocks,
participants discriminated shy versus nonshy words and me versus other
words on separate trials within the same block. In Block 4, participants
used one response key to indicate if a word belonged to the nonshy or other
categories and the other key if the word belonged to the shy or me
categories. In Block 7, participants used one response key to indicate if a
word belonged to the nonshy or me categories and the other key if the word
belonged to the shy or other categories. Blocks 3 and 6 were combined
blocks that served as practice for Blocks 4 and 7. Only data from Blocks
4 and 7 were used to compute IAT scores. The main dependent variable
(IAT score) was computed by subtracting participants’ average response
latencies during Block 4 from their average response latencies during
Block 7. Positive differences in this index indicated faster automatic
associations between me and shy than between others and shy.
It is evident from the above description that the IAT is a relative
measure. For example, in the present studies, the category me is contrasted
with the category others. Indeed, opposing me with others might make the
IAT scores difficult to interpret in certain cases because they can reflect
associations with the self, with the others, or a combination of the two (e.g.,
Karpinski, 2004). Particularly relevant to the present research, however,
the explicit self-report scale also likely involved a similar relative judgment
compared with others. Most subjective judgments about the self or others
require reference to some standard—typically provided by others. For
example, as Mussweiler (2003) noted, “To characterize oneself as athletic
158 BRIÑOL, PETTY, AND WHEELER
. . . implies that one is more athletic than others and is thus, in essence, a
comparative statement” (p. 472). In fact, it would be hard for people to
develop their self-conceptions without comparing with others (Festinger,
1954). Thus, it is plausible to argue that both the implicit and the explicit
measures of shyness might include a comparative component.
Following typical procedures, stimuli in the IAT task appeared within a
centered white window. Reminder labels were positioned on top of the
stimuli on the left and right side. These reminders read me and other for
single target-classification blocks, shy and nonshy for single attribute-
classification blocks. Mixed target ⫹ attribute blocks were also accompa-
nied by appropriate labels (e.g., shy or other; nonshy or me). Incorrect
classifications were followed by error feedback (i.e., the word ERROR).
Summary feedback was provided at the end of each practice block inform-
ing participants about their average response latency and percentage of
errors for that block. All practice trials in the IAT were administered in five
blocks. Data-collection trials, consisting of combined target ⫹ attribute
classifications, were collected in four blocks. Within each block, stimuli
were randomly selected without replacement, and no more than two con-
secutively presented stimuli belonged to the same category. To correct for
anticipatory responses and momentary inattention, we recorded latencies
shorter than 300 ms and longer than 3,000 ms as 300 and 3,000 ms,
respectively. Response latencies were then log transformed to normalize
the distribution. Further details about the IAT procedure are provided by
Greenwald et al. (1998).
Explicit–implicit discrepancy. The explicit and implicit measures of
shyness were significantly correlated (r ⫽ .40, p ⬍ .001). This correlation
is similar to those obtained by Asendorpf et al. (2002) using similar explicit
and implicit measures in a different sample.2 An index of explicit–implicit
discrepancy was formed as the absolute value of the difference between the
standardized explicit and implicit measures of shyness. The discrepancy
index considers where people fall within the distribution of participants in
the study on the implicit versus explicit measures. A zero on the index
indicates that the person’s place in the distribution is exactly the same on
the implicit and explicit measures (e.g., high in the distribution on both,
low in the distribution on both, middling on both, and so forth). Discrep-
ancies can be in either direction. That is, people can be higher in the sample
distribution on the explicit measure than the implicit measure (a positive
discrepancy) or they can be lower in the distribution on the explicit
measure than the implicit measure (a negative discrepancy). In this study,
38 participants had positive discrepancies and 39 participants had negative
discrepancies. In addition to including size of the discrepancy, analyses
also included a factor for whether the discrepancy was positive or negative.
Dependent Measure: Attitudes
After reading the message relevant to shyness, participants’ attitudes
toward shyness were assessed using a single-item 9-point semantic differ-
ential scale ranging from bad (1) to good (9) on which they rated how
favorably they viewed shyness.
Results
Attitudes toward shyness were submitted to a hierarchical re-
gression analysis, with extent of discrepancy (continuous vari-
able), manipulated argument quality (strong–weak; dummy
coded), and direction of discrepancy (explicit ⬎ implicit vs. ex-
plicit ⬍ implicit; dummy coded) as the independent variables.
Scores on the discrepancy index were centered by subtracting the
mean from each person’s score (Aiken & West, 1991). Main
effects were interpreted in the first step of the regression, the
two-way interactions in the second step, and the three-way inter-
action in the final step (Cohen & Cohen, 1983). Responses to the
attitude scales were scored so that higher values represented more
favorable opinions toward shyness.
Results of this analysis revealed a significant main effect for
extent of discrepancy, ␤ ⫽ ⫺.25, t(73) ⫽ ⫺2.19, p ⫽ .03, showing
that participants’ attitudes were less favorable toward shyness as
explicit–implicit discrepancy increased (all ␤s reported are stan-
dardized coefficients). In addition, a significant interaction be-
tween argument quality and direction of the discrepancy emerged,
␤ ⫽ .43, t(70) ⫽ 2.42, p ⫽ .02, revealing that argument quality had
a larger effect on attitudes for the explicit ⬎ implicit direction than
the implicit ⬎ explicit direction. This effect was not obtained in
any of the other studies and is not discussed further.
More critical to our hypothesis, a significant interaction between
argument quality and extent of discrepancy was evident, ␤ ⫽ .46,
t(70) ⫽ 3.17, p ⫽ .002, revealing that as discrepancy increased,
argument quality had a larger effect on attitudes. Specifically,
decomposition of this interaction by recentering discrepancy at one
standard deviation (SD) above and below the mean (Aiken &
West, 1991) indicated that there was a significant effect of argu-
ment quality among participants with high discrepancy, ␤ ⫽ .36,
t(69) ⫽ 2.37, p ⫽ .02, but not among those with relatively low
discrepancy, ␤ ⫽ ⫺.25, t(69) ⫽ ⫺1.69, p ⫽ .10. The three-way
interaction between argument quality, extent of discrepancy, and
direction of the discrepancy was not significant, ␤ ⫽ .26, t(69) ⫽
1.14, p ⫽ .26, indicating that the effects of explicit–implicit
discrepancy on elaboration were not restricted to any particular
direction of the discrepancy.
Discussion
Experiment 1 demonstrated that as the discrepancy between
implicit and explicit measures of shyness increases, people are
more likely to think carefully about a shyness-related persuasive
message. This conclusion was supported by the finding that the
attitudes of relatively discrepant individuals were more reflective
of the quality of the persuasive message that they received about
shyness than were the attitudes of individuals who were low in
discrepancy. These findings suggest that participants with a high
explicit–implicit discrepancy paid more careful attention to the
message than those with a low discrepancy, presumably in an
attempt to resolve the discrepancy.
Experiment 1 demonstrated that explicit–implicit discrepancies
in shyness can lead to greater thinking about information directly
relevant to the dimension on which the discrepancy exists. As
noted earlier, shyness is a general and somewhat broad personality
self-dimension. To extend the utility and generality of our find-
ings, we sought to test whether the same effects would emerge
when the explicit–implicit discrepancy concerned a more specific
dimension of the self: a person’s need to evaluate (Jarvis & Petty,
1996).
Experiment 2
Our second study was designed to provide a conceptual repli-
cation and extension of the first. In Study 2, we used the same
paradigm to assess the extent to which participants engaged in
2
Asendorpf et al. (2002) reported correlations between explicit and
implicit shyness that ranged from .20 to .44 depending on the format of the
explicit measure. For example, the correlation between the IAT and the
explicit measure based on the shyness adjectives scale was .40.
 EXPLICIT–IMPLICIT DISCREPANCY
effortful elaboration (i.e., an argument quality manipulation).
However, several changes were introduced. First, instead of as-
sessing explicit and implicit shyness as in Study 1, we focused on
a motivational self-concept dimension, need to evaluate (NE;
Jarvis & Petty, 1996; Petty & Jarvis, 1996), which refers to
individual differences in people’s tendencies to engage in evalu-
ative thought. People who are high in NE tend to spontaneously
assess whether things are good or bad (e.g., Tormala & Petty,
2001; see also the “need to assess”; Kruglanski et al., 2000).
Knowing whether things in the world are good or bad helps people
to understand the environment. Probably because of this and other
functions (e.g., Maio & Olson, 2000), people tend to form attitudes
about nearly everything (e.g., Bargh, Chaiken, Govender, & Pratto,
1992; Roskos-Ewoldsen & Fazio, 1992). Nevertheless, some peo-
ple are more chronic and spontaneous than others in their tendency
to evaluate, and the NE scale assesses this. In contrast to the
explicit self-report measure typically used to assess NE, we de-
veloped an implicit NE measure using an IAT. Similar to the
previous study, an index of self-discrepancy was formed as the
absolute value of the difference between the standardized values of
the explicit and implicit measures.
Second, to generalize our results across topics, we used a
different persuasive issue. Instead of presenting information di-
rectly related to the discrepancy dimension as in Study 1, we used
a message containing substantive information that actually was not
relevant to the issue of the discrepancy. However, for all partici-
pants, the message was framed to appear relevant to NE. That is,
participants were told before reading the message that the research
concerned their evaluations, and so all participants had an expec-
tation that the message would contain information directly relevant
to the domain of the discrepancy.
The topic of the message the participants were to evaluate was
increasing the amount of vegetables in the diet. Comparing the
persuasive effect of strong and weak arguments tested the effects
of explicit–implicit self-discrepancy on information processing.
We expected participants with a relatively high discrepancy be-
tween their explicit and implicit NE to think about information
expected to be discrepancy related to a greater extent than indi-
viduals with a relatively low discrepancy. That is, we expected
argument quality to have a larger impact on attitudes for partici-
pants with a high discrepancy between explicit and implicit NE
compared with participants with a low discrepancy. Thus, as in
Study 1, we expected to find a Discrepancy ⫻ Argument Quality
interaction on the measure of attitudes that was unmoderated by
the direction of the discrepancy.
Method
Participants and Design
Ninety-nine undergraduates in psychology courses at the Universidad
Autónoma de Madrid participated in partial fulfillment of a course require-
ment. We randomly assigned the students to the argument quality condi-
tions (strong or weak) and assessed their explicit and implicit (IAT) NE to
form an index of explicit–implicit self-discrepancy. Direction of discrep-
ancy (higher on explicit or implicit measure) was also coded.
Procedure
The procedure was similar to Study 1. Participants were seated at
individual computer stations and were told that they were going to partic-
159
ipate in two different research projects. First, participants completed the
IAT for the NE measure, ostensibly as part of a research project in the
cognitive psychology program. After the IAT task, participants were told
that because of extra time remaining in the session, they would also be
participating in another experiment designed to assess their attitudes to-
ward a particular issue. To create a context of evaluation, all of the
participants were explicitly told that the goal of the second research project
was to measure their opinions and evaluations of a current commercial
campaign. Participants received a persuasive message in favor of the
consumption of vegetables containing either strong or weak arguments.
Then, participants were told that it was important to know more specifi-
cally what their opinions about the consumption of vegetables were. After
reporting their attitudes toward the proposal, participants completed the
explicit NE scale and several ancillary measures.
Independent Variables
Argument quality. The message about vegetable consumption con-
tained either strong or weak arguments in favor of this topic. This manip-
ulation was designed to influence the favorability of participants’ cognitive
responses if participants were thinking about the message (Petty & Ca-
cioppo, 1986). The gist of one of the strong arguments in favor of vegetable
consumption was that vegetables have more vitamins than the majority of
vitamin supplements on the market, making them especially appropriate
during exams and workout periods. The gist of one of the weak arguments
was that vegetables are becoming more popular for wedding celebrations
because they are colorful and look beautiful on plates. The argument
quality of the two messages was pretested to ensure that the strong version
of the message produced mostly favorable thoughts, whereas the weak one
produced mostly negative thoughts when people were instructed to think
carefully about them (e.g., Briñol, Horcajo, Becerra, Falces, & Sierra,
2002).
Explicit measure of need to evaluate. Participants completed the 16-
item NE scale (Jarvis & Petty, 1996). This scale assesses the chronic
tendency to engage in evaluative responding. Jarvis and Petty (1996)
demonstrated that, compared with people low in NE, those high in the NE
are more likely to form attitudes toward a variety of social and political
issues (see also Bizer et al., 2004), and the attitudes of high NE individuals
tend to be more accessible (Hermans, DeHouwer, & Eelen, 2001; for a
review, see Briñol & Petty, 2005). The NE scale contains statements such
as “I enjoy strongly liking and disliking new things” and “I am pretty much
indifferent to many important issues” (reverse scored). Participants re-
sponded to each statement on a 5-point scale anchored by extremely
uncharacteristic of me (1) and extremely characteristic of me (5). The
items on this self-concept dimension were intercorrelated (␣ ⫽ .79), so
responses to each item were summed to form a composite score of NE.
Participants’ NE scores were not affected by the argument quality manip-
ulation, F(1, 98) ⫽ 0.44, p ⫽ .5.
Implicit measure of need to evaluate. As in Study 1, we used an IAT
procedure as the implicit assessment of this self-conception (Greenwald et
al., 1998). The NE IAT was administered at the beginning of the experi-
mental session and was presented as part of a research project designed to
study how taxonomies are represented in people’s minds. In this IAT,
participants classified target concepts (represented by me or other) and
attributes (represented by neutral or extreme) using two designated keys.
The words extreme and neutral were pretested as representative of high and
low evaluation categories, respectively. Although the NE scale was de-
signed to measure the tendency to engage in evaluation per se rather than
the tendency to engage in extreme evaluation, a number of items on the NE
scale clearly refer to extremity. According to Jarvis and Petty (1996, p.
190), the reason those items were included was to maximize the variance
in participants’ scores on the NE scale. Also for that reason, we selected the
words that were related to extremity to increase the variability in the IAT.
The me category was represented by the words I, me, mine, my, and self,
whereas the other category was represented by the words they, others,
160 BRIÑOL, PETTY, AND WHEELER
them, theirs, and their. The extreme category included words such as
extremity, limit, radical, total, and extreme. In contrast, the neutral cate-
gory included words as moderate, caution, prudent, temperate, and neutral.
These words were equally familiar to participants and were selected based
on some of the items of the NE scale (e.g., “It bothers me to remain
neutral,” “I prefer to avoid taking extreme positions” [reversed], and “I
often prefer to remain neutral about complex issues” [reversed]). The
difference in response latencies for (me ⫹ extreme and other ⫹ neutral)
versus (other ⫹ extreme and me ⫹ neutral) provided our implicit measure
of NE. Regarding the combination of blocks, random assignment of stim-
uli, incorrect classifications, practice trials, anticipatory responses, momen-
tary inattention, and data transformation, we followed the standard IAT
procedures described in our first study (see also Greenwald et al., 1998).
Explicit–implicit discrepancy. Explicit and implicit NE were not cor-
related (r ⫽ ⫺.13, p ⫽ .17). An index of discrepancy was formed as the
absolute value of the difference between the standardized explicit and
standardized implicit measures. Higher scores on that variable reflected
greater differences between the explicit and the implicit measures, and thus
higher explicit–implicit discrepancy in NE. As in the previous study,
direction of discrepancy was symmetrical, with 50 participants showing a
positive discrepancy (i.e., higher in the sample distribution on the explicit
measure than the implicit measure) and 48 participants showing a negative
discrepancy (i.e., lower in the distribution on the explicit than the implicit
measure).
Dependent Measure: Attitudes
Similar to Study 1, participants’ postmessage attitudes toward vegetable
consumption were assessed using a single item 9-point semantic differen-
tial scale anchored at bad (1) and good (9).
Results
Attitudes toward vegetables were submitted to the same hierar-
chical regression analysis used in Study 1. Thus, the independent
variables included extent of discrepancy (continuous variable),
manipulated argument quality (strong–weak; dummy coded), and
direction of discrepancy (explicit ⬎ implicit vs. explicit ⬍ im-
plicit; dummy coded). Scores on the discrepancy index were
centered by subtracting the mean from each person’s score (Aiken
& West, 1991). Main effects were interpreted in the first step of the
regression, the two-way interactions in the second step, and the
three-way interaction in the final step (Cohen & Cohen, 1983).
Responses to the attitude scales were scored so that higher values
represented more favorable opinions toward the proposal.
Participants’ attitudes were more favorable toward consuming
vegetables after receiving the strong (M ⫽ 8.29, SD ⫽ 0.71) than
the weak (M ⫽ 7.98, SD ⫽ 0.79) message, ␤ ⫽ .20, t(95) ⫽ 1.96,
p ⫽ .05. No main effect for extent of discrepancy emerged ( p ⫽
.66). More important and consistent with expectations, the main
effect of argument quality was qualified by a marginally signifi-
cant interaction between argument quality and extent of discrep-
ancy, ␤ ⫽ .28, t(92) ⫽ 1.83, p ⫽ .07. This interaction revealed that
as discrepancy increased, argument quality had a larger effect on
attitudes. That is, there was a significant effect of argument quality
among participants with relatively high discrepancy (recentered at
⫹1 SD), ␤ ⫽ .38, t(91) ⫽ 2.64, p ⫽ .01, but not among those with
relatively low discrepancy (recentered at –1 SD), ␤ ⫽ .005,
t(91) ⫽ 0.03, p ⫽ .97. It is important to note that the three-way
interaction between argument quality, extent of discrepancy, and
direction of the discrepancy was not significant, ␤ ⫽ ⫺.28, t(91) ⫽
⫺1.42, p ⫽ .16, revealing that the effects of explicit–implicit
discrepancy on elaboration were not restricted to any particular
direction of the discrepancy.3
Discussion
Experiment 2 conceptually replicated the first experiment by
showing that participants with relatively high explicit–implicit
discrepancy in their NE processed the message more carefully than
participants with relatively low discrepancy. That is, compared
with those with a small discrepancy between their explicit and
implicit NE, individuals with a high discrepancy were more influ-
enced by the quality of the arguments in the message than were
those with a low discrepancy.
Although the message in the present study did not actually
contain information directly related to the issue on which the
explicit–implicit discrepancies existed (i.e., a person’s own NE),
the whole context of the study was framed as dealing with the
participants’ opinions. We argue that emphasizing that the task
was related to evaluation was sufficient for participants with high
explicit–implicit discrepancies in NE to engage in more extensive
thinking. From our first two studies, it appears that information can
be directly related to the issue of discrepancy (Study 1) or simply
framed as related to the dimension in which the discrepancy exists
(Study 2).
We argue that the mere existence of a discrepancy does not
result in the indiscriminate processing of any information present
in the situation. Enhanced thinking is expected only if the dimen-
sion on which the discrepancy exists is activated by leading people
to believe that the message is going to pertain or be relevant to that
dimension. Without that, there should be no differential process-
ing. To address this issue, our third study included a manipulation
of the message frame designed to induce participants to expect the
message to be related or unrelated to the issue on which there was
an explicit–implicit discrepancy. If having an explicit–implicit
discrepancy enhances information processing in general, then high
discrepancy individuals should be equally likely to process mes-
sages framed as relevant or irrelevant to the issue on which the
discrepancy exists. If the information processing is in service of
discrepancy reduction, however, individuals with high discrepan-
3
Because of possible concerns about the use of difference scores in our
analyses (i.e., the discrepancy index), we also conducted an alternative
analysis treating implicit and explicit measures separately. For maximum
power, data from both Studies 1 and 2 were combined and submitted to
another hierarchical regression analysis, with the implicit and explicit
measures (continuous variables) and manipulated argument quality
(dummy coded) as independent variables. Study was also included as a
factor in this analysis so that we could examine whether the results
generalized across the study differences. As expected, this analysis re-
vealed a significant main effect for argument quality, ␤ ⫽ .14, t(175) ⫽
1.97, p ⫽ .05, which was qualified by a three-way interaction between
argument quality, explicit self-concept, and implicit self-concept, ␤ ⫽
⫺.16, t(175) ⫽ ⫺2.08, p ⫽ .03. Also importantly, this significant three-
way interaction was not moderated by the study independent variable, as is
evident in the absence of a four-way interaction ( p ⫽ .64). To facilitate
ease of presentation (i.e., interpreting two-way rather than three-way in-
teractions) and matching more closely our conceptual variable (i.e., psy-
chological discrepancy), we present the discrepancy analysis in the text and
figures.
 EXPLICIT–IMPLICIT DISCREPANCY
cies should be more likely to process the messages framed as
relevant than irrelevant.
Experiment 3
Studies 1 and 2 provided initial evidence supporting the notion
that divergence between explicit and implicit self-views can influ-
ence the extent of information processing of ostensibly
discrepancy-related information. Experiment 3 was conducted to
replicate and extend the findings from these studies. Thus, several
changes were introduced. First, instead of examining a general
dimension of personality (shyness) or motivation (NE), we focused
on a more specific dimension of the self-concept: one’s beliefs
concerning one’s own resistance to persuasion.
Beliefs regarding one’s resistance to change play a central role
in people’s values and identities. For example, Schwartz’s (1992)
theory about universal human values is organized by two motiva-
tional dimensions: the self-transcendence/self-enhancement di-
mension and the openness to change/conservation dimension. This
work implies that almost everyone might have beliefs about their
own resistance to change and that such beliefs might be an integral
part of the self-concept. In fact, resistance to change constitutes
one of the most basic dimensions of personality according to the
Big Five framework (e.g., McCrae & Costa, 1985; Wiggins &
Trapnell, 1997). Although the construct of resistance to change can
be conceptualized quite differently in terms of personality, cogni-
tive ability, psychic structure, openness to experience, or openness
as culture (McCrae & Costa, 1997), we focus specifically on
personal perceptions of resistance to persuasion. In this experi-
ment, the explicit self-dimension of resistance to persuasion was
assessed using the Resistance to Persuasion Scale (Briñol, Rucker,
Tormala, & Petty, 2004), and implicit resistance was assessed
using an IAT developed for this study. As in our previous studies,
an index of explicit–implicit discrepancy was created as the abso-
lute value of the difference between the standardized explicit and
implicit scores.
Second, to further generalize our results across topics, we used
a different persuasive issue. Instead of presenting information
about a relatively proattitudinal topic (i.e., increasing the amount
of vegetables in the diet; Study 2) or a rather neutral one (i.e., the
benefits of shyness; Study 1), we moved to a relatively counter-
attitudinal topic (i.e., institution of required exams for college
students).
Finally, and most importantly, we manipulated the presumed
discrepancy-relevance of the message information. Because the
explicit–implicit discrepancy was related to individual differences
in susceptibility to change in attitudes and opinions, half of the
participants were told that the study had to do with their attitudes
and opinions (similar to the frame in Study 2). This condition was
compared with a discrepancy-unrelated frame in which partici-
pants were told that the experiment was part of a text comprehen-
sion study. Thus, all of the participants were exposed to a persua-
sive message containing either strong or weak arguments that were
framed as either related or unrelated to the dimension on which the
discrepancy existed. After reading the message, participants were
asked to report their attitudes toward the proposal. We expected
participants with a large explicit–implicit discrepancy to think
more about the persuasive message than those with a small dis-
crepancy, but only when the message was framed to appear dis-
161
crepancy related. As in previous studies, this enhanced thinking
would be evidenced by greater attitudinal responsiveness to the
manipulation of argument quality.
Method
Participants and Design
One hundred seventy-three introductory psychology students at Ohio
State University participated in partial fulfillment of a course requirement.
The students were randomly assigned to the argument quality conditions
(strong or weak) and the message context conditions (discrepancy related
or unrelated), which were manipulated orthogonally. Additionally, we
measured participants’ explicit and implicit resistance to persuasion to
form an index of explicit–implicit discrepancy. The independent variables
thereby constituted an Argument Quality (strong vs. weak) ⫻ Message
Frame (discrepancy-related vs. discrepancy-unrelated) ⫻ Extent of
Explicit–Implicit Discrepancy (continuously scored) ⫻ Direction of Dis-
crepancy (higher on implicit or explicit measure) design.
Procedure
Upon arrival, participants were seated at individual computer stations
and were presented with all of the materials on the computer using
MediaLab software (Jarvis, 2000). All of the participants were told that
they were going to participate in two different research projects. First,
participants completed the IAT for resistance to persuasion, ostensibly as
part of a research project on semantic recognition and categorization
conducted by the cognitive psychology program. After the IAT task,
participants were told that because of extra time remaining in the session,
they would also be participating in another experiment designed to assess
possible changes in university policies. They read about a new school
policy and were told that students’ opinions about this policy were of
importance to the university. Participants received a message in favor of
instituting senior comprehensive exams that contained either strong or
weak arguments. After reading the message, participants were told that it
was important for the Board of Trustees to know what their opinions on the
topic were. Thus, they completed measures of their attitudes toward the
comprehensive exam policy. Finally, participants completed the Resistance
to Persuasion Scale (Briñol et al., 2004) and several ancillary questions.
Independent Variables
Argument quality. The comprehensive exam message participants re-
ceived contained either strong or weak arguments. The arguments selected
were adopted from previous research and have been shown many times to
produce the appropriate pattern of cognitive responding (see Petty &
Cacioppo, 1986). That is, the strong arguments elicited mostly favorable
thoughts and the weak arguments elicited mostly unfavorable thoughts
when people were instructed to think carefully about them. The gist of
some strong arguments in favor of the exam policy were that students’
grades would improve if the exams were adopted and that the average
starting salary of graduates would increase. The gist of some weak argu-
ments in favor of the exam policy were that implementing the exams would
allow the university to take part in a national trend and that the exams
would give students the opportunity to compare their scores with those of
students at other universities.
Message frame. The frame of the message was manipulated to appear
related or unrelated to participants’ opinions. That is, because the explicit–
implicit discrepancy was related to persuasion (openness vs. resistance to
changing one’s opinions), we framed the study as related (or not) to
opinions and attitudes. Thus, the message was introduced as part of an
opinion-related study (discrepancy-related frame) or as part of a text
162 BRIÑOL, PETTY, AND WHEELER
comprehension study (unrelated frame). Participants in the discrepancy-
related frame condition read the following:
OPINION SURVEY
This part of the study consists of a survey designed to obtain your
opinions and thoughts about a campus issue. The message we are
going to ask you to read is based on the transcript of an editorial from
a college radio station. The editorial was introduced on the radio in
order to familiarize students with this important issue. Please pay
attention because your opinions will be measured.
In the comprehension frame, participants were told that the message was
designed to measure their appreciation of new information. Comprehend-
ing a text would presumably not trigger the idea that the study involved any
explicit or blatant persuasive attempt. In this condition, participants read
the following:
TEXT COMPREHENSION
This part of the study has been designed to measure your appreciation
of new information. The information we are going to ask you to read
and to examine is based on a transcript of a class project broadcast on
a college radio station. We want you to examine the content of the
information presented carefully because the extent to which you
understand the text will be measured.
It is important to note that although attitude change was not salient for
the text comprehension frame, opinions toward the proposal were still
measured following the message. At the time opinions were measured, the
idea of persuasibility may have become apparent to these participants, but
by this time the extent of processing of the message was already deter-
mined. That is, in the discrepancy-relevant case, participants already
expected to give their opinion while they were reading the message
because the study was about opinions. In the discrepancy-irrelevant case,
they thought that the study was on text comprehension while reading and
processing the message, so forming opinions was not salient. We pretested
the effectiveness of the induction by randomly assigning 50 students to one
of the two frames and asking them what kind of information they expected
to receive after the frame. Specifically, we asked participants to respond on
a 5-point scale to the question: “To what extent do you expect to receive
any information that might be related to your resistance to persuasion?” As
expected, participants who received the discrepancy-related frame reported
significantly higher expectations to receive persuasion-related information
(M ⫽ 3.14, SD ⫽ 1.00) than those who received the discrepancy-unrelated
frame (M ⫽ 2.68, SD ⫽ 1.21), t(48) ⫽ 10.39, p ⬍ .0001. It is also
important to note that although the message was framed to be related or not
to the basis of the self-discrepancy, the actual message participants re-
ceived was identical.
Explicit measure of resistance. The Resistance to Persuasion Scale was
used to assess participants’ explicit perceptions of their resistance to
persuasion. This scale measures individuals’ perceptions and beliefs about
their own vulnerability to persuasion, willingness to change, and motiva-
tion and ability to resist persuasion. In prior research validating the scale,
it has predicted the number of counterarguments people generate to a
message and how resistant they were to influence (Briñol et al., 2004). The
scale contains 11 statements such as “My attitudes are open to change” and
“It is hard for me to change my ideas.” Participants responded to each
statement on a 5-point scale anchored by extremely uncharacteristic of me
(1) and extremely characteristic of me (5). Ratings on the scale items were
highly consistent with each other (␣ ⫽ .86) and were averaged to form a
single index of resistance to persuasion for each participant. Resistance to
persuasion scores were not affected by the argument quality manipulation,
F(1, 173) ⫽ 0.26, p ⫽ .61; the frame manipulation, F(1, 173) ⫽ 0.68, p ⫽
.41; or the interaction of the two, F(1, 173) ⫽ 0.02, p ⫽ .87.
Implicit measure of resistance. In the IAT measure of resistance,
participants classified target concepts (represented by me or other) and
attributes (represented by easy to be persuaded or hard to be persuaded
categories of words) using two designated keys. The me and other cate-
gories included the same words used in Study 2. Attributes related to
persuasibility were selected from the items on the explicit scale and
included the words easy, flexible, open, variable, and changeable. In
contrast, resistant attributes included the words resistant, stable, hard,
consistent, and committed. The difference in response latencies for me–
easy (me ⫹ easy and other ⫹ hard) versus me– hard (other ⫹ easy and
me ⫹ hard) responses provided a measure of relative self-association with
resistance (i.e., the IAT effect). Regarding the combination of blocks,
random assignment of stimuli, incorrect classifications, practice trials,
anticipatory responses, momentary inattention, and data transformation, we
followed the standard IAT procedures described in our previous studies
(see also Greenwald et al., 1998).
Explicit–implicit discrepancy. The explicit and implicit measures of
personal resistance were uncorrelated (r ⫽ ⫺.10, p ⫽ .16). An index of
explicit–implicit discrepancy was formed as the absolute value of the
difference between the standardized explicit and the implicit measures of
resistance. Higher scores on that variable reflected greater differences
between the explicit and the implicit measures (i.e., higher discrepancy).
The distributions of scores using this index revealed roughly equal num-
bers of people on each side of the discrepancy, with 87 participants
showing positive discrepancy (i.e., higher in the sample distribution on the
explicit measure than the implicit measure) and 86 showing negative
discrepancy (i.e., lower in the distribution on the explicit than the implicit
measure). Perhaps more relevant for the present research, the number of
participants in each direction was also equivalent in the conditions for
which the information was presented as discrepancy related, with 42
participants revealing positive discrepancies and also 42 participants show-
ing negative discrepancies.
Dependent Measure: Attitudes
Participants’ attitudes toward the proposal were assessed using a series
of five 9-point semantic differential scales ranging from 1 to 9 (i.e.,
bad– good, unfavorable–favorable, pro–against, foolish–wise, harmful–
beneficial) on which they rated the comprehensive exam policy. Ratings on
these items were highly intercorrelated (␣ ⫽ .84), so they were averaged to
form one overall attitude index.
Results
Attitudes were submitted to a hierarchical regression analysis,
with extent of discrepancy (continuous variable), manipulated
argument quality (strong–weak; dummy coded), message frame
(relevant–irrelevant; dummy coded), and direction of discrepancy
(explicit ⬎ implicit vs. explicit ⬍ implicit; dummy coded) as the
independent variables. Scores on the discrepancy index were cen-
tered by subtracting the mean from each person’s score (Aiken &
West, 1991). Main effects were interpreted in the first step of the
regression, two-way interactions in the second step, three-way
interactions in the third step, and the four-way interaction in the
final step (Cohen & Cohen, 1983). Responses to the attitude scales
were scored so that higher values represented more favorable
opinions toward the proposal.
Participants’ attitudes were more favorable toward the proposal
after receiving the strong (M ⫽ 4.46, SD ⫽ 2.34) than the weak
(M ⫽ 3.68, SD ⫽ 2.09) message, ␤ ⫽ .18, t(168) ⫽ 2.35, p ⫽ .02.
A significant interaction between argument quality and extent of
discrepancy also emerged, ␤ ⫽ .33, t(162) ⫽ 3.31, p ⫽ .001,
revealing that as discrepancy increased, argument quality had a
larger effect on attitudes. An interaction between frame and extent
 EXPLICIT–IMPLICIT DISCREPANCY
of discrepancy also emerged, ␤ ⫽ .23, t(162) ⫽ 1.97, p ⫽ .05. This
interaction suggested that attitudes tended to become more positive
as discrepancy increased for the relevant frame but more negative
as discrepancy increased for the irrelevant frame. This interaction
was only obtained in this study and is not discussed further.
More critical to our primary concerns, the predicted three-way
interaction between argument quality, extent of discrepancy, and
message frame was significant, ␤ ⫽ .42, t(158) ⫽ 2.84, p ⫽ .005.
As depicted in Figure 1, this three-way interaction indicated that
the two-way interaction between discrepancy and argument quality
was only significant for the discrepancy-related frame, ␤ ⫽ .56,
t(157) ⫽ 4.26, p ⬍ .0001, but not for the discrepancy-unrelated
frame, ␤ ⫽ ⫺.02, t(157) ⫽ ⫺.14, p ⫽ .89. For the discrepancy-
related frame, there was a significant effect of argument quality
among participants with relatively high discrepancy (analyzed at
⫹1 SD), ␤ ⫽ .72, t(157) ⫽ 4.86, p ⬍ .0001, but not among those
with relatively low discrepancy (analyzed at –1 SD), ␤ ⫽ ⫺.13,
t(157) ⫽ ⫺.93, p ⫽ .36. The four-way interaction was not signif-
icant, ␤ ⫽ ⫺.12, t(157) ⫽ ⫺.51, p ⫽ .61, revealing that the effects
of explicit–implicit discrepancy on elaboration were symmetrical
and not restricted to any particular direction of the discrepancy.
Discussion
Experiment 3 conceptually replicated our previous findings by
showing that people who have a large discrepancy between their
Figure 1. Attitudes as a function of argument quality, message frame, and
extent of explicit–implicit (e-i) discrepancy graphed at –1 standard devia-
tion (low discrepancy) and ⫹1 standard deviation (high discrepancy) in
Experiment 3.
163
implicit and explicit views of their resistance to persuasion are
more likely to think carefully about a discrepancy-related persua-
sive message than are people who have a small discrepancy
between their implicit and explicit self-conceptions. As in Studies
1 and 2, this conclusion was supported by the finding that the
attitudes of relatively discrepant individuals were more reflective
of the quality of the discrepancy-related persuasive message than
were the attitudes of less discrepant individuals. The fact that the
enhanced information processing only occurred for individuals
with large discrepancies when the message was framed so as to
seem related to the basis of the discrepancy (i.e., an opinion
context) rather than unrelated to the discrepancy (i.e., a compre-
hension context) is consistent with the idea that the purpose of the
processing was to resolve the discrepancy.
Experiment 3 demonstrated that explicit–implicit discrepancies
in resistance to persuasion can lead to greater thinking about
information framed as related to the discrepancy. However, resis-
tance to change is a very specific and descriptive self-dimension.
Thus, Studies 1, 2, and 3 all focused on descriptive (shyness, NE,
and persuasibility, respectively) rather than evaluative self-
dimensions. To extend the utility and generality of our findings, in
a final study we sought to test whether the same effects can emerge
when the explicit–implicit discrepancy concerns a broad, global
evaluation of the self. Thus, in our fourth study we used a similar
paradigm to test whether the information-processing effects can be
found for explicit–implicit divergences on a general evaluative
self-dimension: a person’s self-esteem.
Experiment 4
Our fourth study was designed to extend the previous findings to
the domain of self-evaluation. That is, in this study we focused on
an evaluative dimension of the self-concept, self-esteem, and as-
sessed it with both explicit and implicit measures. Implicit self-
esteem typically has been defined as an evaluation of the self that
occurs automatically and unintentionally and can differ from one’s
more controlled and deliberative self-assessments (e.g., Farnham,
Greenwald, & Banaji, 1999; Greenwald & Banaji, 1995; Green-
wald & Farnham, 2000; Hetts & Pelham, 2001; Koole, Dijkster-
huis, & van Knippenberg, 2001).
In Study 4, we used the same paradigm to assess the extent to
which participants engaged in effortful thinking (i.e., an argument
quality manipulation). Similar to Study 3, we also manipulated the
ostensible discrepancy relatedness of the message information by
framing it as related or unrelated to the basis of the discrepancy.
Finally, in this study we used the relatively proattitudinal topic
used in Study 2 (i.e., increasing the amount of vegetables in the
diet). We expected participants with a relatively large discrepancy
between their explicit and implicit self-esteem to elaborate the
information more than individuals with a relatively small discrep-
ancy, but only when the information was framed to seem discrep-
ancy related. That is, we expected argument quality to have a larger
impact on attitudes for participants with a large discrepancy between
explicit and implicit self-esteem compared with participants with a
small discrepancy, but only when the message was framed to seem
related to the discrepancy (i.e., when the message was framed to seem
related to their self-concept). More specifically, as in Study 3 we
expected to find a Discrepancy ⫻ Argument Quality ⫻ Message
Frame interaction on attitudes toward the message proposal.
164 BRIÑOL, PETTY, AND WHEELER
Method
Participants and Design
One hundred seventeen undergraduate psychology students at the Uni-
versidad Autónoma de Madrid were randomly assigned to the argument
quality (strong or weak) and the message frame (discrepancy-related or
discrepancy-unrelated) conditions, which were manipulated orthogonally.
Additionally, we measured participants’ explicit and implicit self-esteem to
form an index of explicit–implicit discrepancy and coded for the direction
of discrepancy. The independent variables thereby constituted an Argu-
ment Quality (strong vs. weak) ⫻ Message Frame (discrepancy-related vs.
discrepancy-unrelated) ⫻ Extent of Explicit–Implicit Discrepancy in Self-
Esteem (continuously scored) ⫻ Direction of Discrepancy (relatively
higher in explicit or implicit self-esteem) design.
Procedure
Upon arrival, participants were seated at individual computer stations
and were told that they were going to participate in two different research
projects. As in previous studies, participants first completed an IAT,
ostensibly as part of a research project on categorization conducted by the
cognitive psychology program. After the IAT task, participants were told
that because of extra time remaining in the session, they would also be
participating in another experiment designed to assess their attitudes to-
ward an opinion topic. Half of the participants were told that the message
they were about to read had to do with plants and vegetables and their
qualities and properties (discrepancy-unrelated frame). The rest of the
participants were told that the message concerned their personal habits and
the way they interact with the world (discrepancy-related frame). Then,
participants received a message containing strong or weak arguments in
favor of the consumption of vegetables. After reading the message, all of
the participants were told that it was important to know more specifically
what their opinions about the consumption of vegetables were. Finally,
after reporting their attitudes toward the proposal, participants completed
the Rosenberg Self-Esteem Scale (Rosenberg, 1965).
Independent Variables
Argument quality. The message about vegetable consumption con-
tained either strong or weak arguments in favor of this topic. This manip-
ulation was identical to that used in Study 2.
Message frame. The introduction to the message was framed to seem
as though the message would contain information unrelated or related to
the discrepancy domain. In the discrepancy-unrelated frame condition,
participants were told, “You are about to read an article about the charac-
teristics and properties of different plants and vegetables.” In the
discrepancy-related frame condition, they were told, “You are about to read
an article about your self-concept, your personal diet habits, and the way
you see your world.” To further enhance the discrepancy relevance, in the
discrepancy-unrelated frame, the title of the message was “Research About
Vegetables” and in the discrepancy-related frame, the title of the message
was “Research About the Self-Concept.”
It is important to note that, as in Experiments 2 and 3, we deliberately
did not attempt to provide any real information about students’ self-
conceptions. Rather, the manipulation was oriented to influence what
participants perceived was going to be the content of the message (not the
actual content itself). Thus, there would be no differences in ability to
process the message across conditions or other irrelevant confounds. We
pretested the effectiveness of the induction by randomly assigning 58
students to one of the two frames and asked them what kind of information
they expected to receive after that. Specifically, we asked participants to
respond on a 5-point scale to the question, “To what extent do you expect
to receive any information that might be relevant to your self-concept and
self-esteem?” As expected, participants who received the discrepancy-
related frame reported significantly higher expectations to receive self-
related information (M ⫽ 3.70, SD ⫽ 0.91) than those who received the
discrepancy-unrelated frame (M ⫽ 1.82, SD ⫽ 0.94), t(56) ⫽ 7.68, p ⬍
.0001.
Explicit self-esteem. We used the Rosenberg Self-Esteem Scale as our
explicit measure (Rosenberg, 1965). Ratings on the items were highly
consistent with each other (␣ ⫽ .76) and were averaged to form a single
index of self-esteem for each participant. Self-esteem scores were not
affected by the argument quality manipulation, F(1, 117) ⫽ 0.08, p ⫽ .77;
the frame manipulation, F(1, 117) ⫽ 1.41, p ⫽ .24; or the interaction of the
two, F(1, 117) ⫽ 0.11, p ⫽ .73.
Implicit self-esteem. As in the prior studies, we used an IAT procedure
to assess participants’ implicit self-evaluations (Greenwald et al., 1998).
The self-esteem IAT was administered at the beginning of the experimental
session and was presented as part of a research project designed to study
how taxonomies are represented in people’s minds. In the IAT, participants
classified target concepts (represented by me or other) and attributes
(represented by good and bad) using two designated keys. The me and
other categories included the same words used in Studies 2 and 3. The good
category words included freedom, peace, love, cheer, and paradise, and the
bad category words included poison, cancer, death, vomit, and disaster
(selected from Greenwald et al., 1998). The difference in response latencies
for me ⫹ bad and other ⫹ good trials versus other ⫹ bad and me ⫹ good
trials provided a measure of relative automatic self-esteem (i.e., the IAT
effect). Comparable implicit measures have been used in prior research and
have been shown to be effective in predicting a variety of thoughts and
behaviors relevant to self-evaluations, especially under low thinking (spon-
taneous) circumstances (e.g., Greenwald & Farnham, 2000; Hetts & Pel-
ham, 2001; Jones, Pelham, Mirenberg, & Hetts, 2002; Koole et al., 2001;
see also Bosson, Swann, & Pennebaker, 2000).
Explicit–implicit discrepancy. Explicit and implicit self-esteem
showed a small negative correlation (r ⫽ ⫺.19, p ⫽ .04). Previous research
has also shown negative correlations between explicit and implicit self-
esteem (e.g., Bosson et al., 2000; Hetts et al., 1999; Karpinski, 2004;
Kitayama & Uchida, 2003). An index of discrepancy was formed as the
absolute value of the difference between the standardized explicit and
implicit measures. Higher scores on that variable reflected greater differ-
ences between the explicit and the implicit measures, and thus higher
explicit–implicit self-esteem discrepancy. As in all previous studies, the
distributions of scores revealed equivalent numbers of participants on each
side of the discrepancy, with 54 participants showing positive discrepan-
cies and 63 showing negative discrepancies. This was also true for the
conditions in which the information was presented as discrepancy related,
with 26 participants revealing positive discrepancies and 31 showing
negative discrepancies.
Dependent Measure: Attitudes
As in the previous studies, participants were instructed to report their
attitudes following the message. Participants’ attitudes toward the proposal
(i.e., the increased consumption of vegetables) were assessed using a series
of five 9-point semantic differential scales anchored by 1 and 9 (i.e.,
bad– good, unfavorable–favorable, pro–against, foolish–wise, harmful–
beneficial). Ratings on these items were highly intercorrelated (␣ ⫽ .84)
and were averaged to form one overall attitude index. Responses to the
attitude scales were scored so that higher values represented more favor-
able opinions toward increasing consumption of vegetables.
Results
Attitudes were submitted to a hierarchical regression analysis,
with extent of discrepancy (continuous variable), manipulated
argument quality, message frame, and direction of discrepancy as
the independent variables. Scores on the discrepancy index were
 EXPLICIT–IMPLICIT DISCREPANCY
centered by subtracting the mean from each person’s score (Aiken
& West, 1991). Main effects were interpreted in the first step of the
regression, two-way interactions in the second step, three-way
interactions in the third step, and the four-way interaction in the
final step (Cohen & Cohen, 1983).
The regression procedure described above revealed a significant
main effect for direction of discrepancy, ␤ ⫽ .21, t(110) ⫽ 2.23,
p ⫽ .03, showing that participants’ attitudes were more favorable
toward the proposal in one direction of discrepancy (explicit ⬎
implicit) than the other (implicit ⬎ explicit). This effect was not
obtained in any of the other studies and is not discussed further. In
addition, a significant interaction between argument quality and
extent of discrepancy was obtained, ␤ ⫽ .31, t(104) ⫽ 2.54, p ⫽
.01, showing that argument quality had a greater impact on atti-
tudes as discrepancy increased.
Of most interest, the predicted three-way interaction between
argument quality, extent of discrepancy, and frame was significant,
␤ ⫽ .37, t(100) ⫽ 2.94, p ⫽ .004. To examine the basis of this
interaction, we decomposed the interaction by frame conditions.
As depicted in Figure 2, the two-way interaction between discrep-
ancy and argument quality was only significant for the
discrepancy-related frame, ␤ ⫽ .86, t(99) ⫽ 3.76, p ⫽ .0002, but
not for the discrepancy-unrelated frame, ␤ ⫽ .11, t(99) ⫽ 0.64,
p ⫽ .52. More specifically, for the discrepancy-related frame, there
Figure 2. Attitudes as a function of argument quality, message frame, and
extent of explicit–implicit (e-i) discrepancy graphed at –1 standard devia-
tion (low discrepancy) and ⫹1 standard deviation (high discrepancy) in
Experiment 4.
165
was a significant effect of argument quality among participants
with relatively high discrepancy (analyzed at ⫹1 SD), ␤ ⫽ .82,
t(99) ⫽ 4.37, p ⬍ .0001. In contrast, those with relatively low
discrepancy (analyzed at –1 SD) showed no effect of argument
quality, ␤ ⫽ ⫺.26, t(99) ⫽ ⫺1.34, p ⫽ .18. The effects of
explicit–implicit discrepancy on information processing were not
restricted to any particular direction of the discrepancy, as indi-
cated by a nonsignificant four-way interaction, ␤ ⫽ .06, t(99) ⫽
.29, p ⫽ .77.4
Discussion
Experiment 4 conceptually replicated the earlier experiments by
showing that participants with relatively high explicit–implicit
discrepancy in their self-esteem processed more carefully the in-
formation presumably related to the discrepancy dimension than
participants with relatively low discrepancy. That is, individuals
with a large discrepancy found strong arguments to be more
persuasive than weak ones whereas those with a small discrepancy
did not, but only when the messages were potentially relevant to
the discrepancy. Those with a high discrepancy presumably de-
voted more cognitive resources to new information framed as
discrepancy related to reduce the uncertainty or subjective discom-
fort that might result from holding simultaneously favorable and
unfavorable implicit and explicit self-evaluations.
General Discussion
Previous research has shown that there are many possible
sources of internal discrepancies (e.g., attitudes, motives, self-
conceptions), and these discrepancies are often associated with
negative affect and undesirable psychological outcomes. Although
most of the discrepancy research has relied on explicit and delib-
erative self-reports, previous studies identifying people who pos-
sess specific explicit and implicit self-discrepant dimensions have
also found that discrepancy is associated with some notable diffi-
culties in functioning.
The present research extends previous literature by showing that
explicit–implicit discrepancy is also associated with enhanced
4
Similar to Studies 1 and 2, alternative analyses were conducted in
which implicit and explicit measures were treated separately. For maxi-
mum power, data from both Studies 3 and 4 were combined and submitted
to another hierarchical regression analysis, with the implicit and explicit
measures (continuous variables) and manipulated argument quality and
message frame (dummy coded) as independent variables. Study was also
included as a factor in this analysis so that we could examine whether the
results generalized across the study differences. As expected, this analysis
revealed a significant main effect for argument quality, ␤ ⫽ .18, t(250) ⫽
3.61, p ⫽ .001, which was qualified by a three-way interaction between
argument quality, explicit self-concept, and implicit self-concept, ␤ ⫽
⫺.16, t(250) ⫽ ⫺2.31, p ⫽ .02. Most important for our concerns, a
significant four-way interaction also emerged, ␤ ⫽ ⫺.26, t(250) ⫽ 3.27,
p ⫽ .001. As predicted, this four-way interaction indicates that the three-
way interaction between argument quality, explicit, and implicit self-
concept was only significant for the discrepancy-related frame, ␤ ⫽ ⫺.38,
t(127) ⫽ ⫺4.0, p ⬍ .0001, but not for the discrepancy-unrelated frame,
␤ ⫽ ⫺.06, t(122) ⫽ ⫺0.58, p ⫽ .56. Also importantly, this significant
four-way interaction was not moderated by the study independent variable,
as is evident in the absence of a five-way interaction ( p ⫽ .44).
166 BRIÑOL, PETTY, AND WHEELER
processing of ostensibly discrepancy-related information, presum-
ably in an attempt at inconsistency reduction. The present studies
also demonstrate that the tendency to process discrepancy-related
information applies to both narrow (Study 3) and relatively broad
(Studies 1, 2, and 4), as well as to descriptive (Studies 1, 2, and 3)
and evaluative (Study 4), dimensions of the self. Across different
types of self-concept elements, we have demonstrated that indi-
viduals for whom explicit and implicit self-dimensions are incon-
gruent process information that either is (Study 1) or is framed as
(Studies 2, 3, 4) discrepancy related more carefully than individ-
uals with relatively congruent explicit–implicit self-dimensions.
By thinking about information that is presumably related to the
explicit–implicit dimension of conflict, people may be attempting
to reduce the discrepancy between the implicit and explicit dimen-
sions, perhaps unconsciously.
Symmetric Versus Asymmetric Discrepancy
In prior research, most researchers interested in explicit–implicit
discrepancies have concentrated their studies in just one of the
possible directions of discrepancy: people with more positive
explicit than implicit self-views. For example, studies on narcis-
sism (e.g., Bosson et al., 2003) and defensive self-esteem (Jordan
et al., 2003; Shedler et al., 1993) have particularly focused on
cases in which explicit self-esteem is higher than implicit self-
esteem and compared this particular discrepancy with cases in
which both implicit and explicit self-esteem are relatively high.
When looking at these asymmetric discrepancies, prior research
has usually divided participants on the basis of relative scores
within the sample distribution, as we did in the present research
(see Weinberger & Hardaway, 1990, for an exception). Several
different procedures have been used for dividing people into
discrepancy groups, such as by splitting people at the median,
tertiles, or quartiles of the sample distribution (Barger, Kircher, &
Croyle, 1997; Emmons & Colby, 1995; Mendolia, Moore, &
Tesser, 1996; Newman, Duff, & Baumeister, 1997; Weinberger,
1990) and by recentering at one SD above and below the mean of
the distribution (e.g., Bosson et al., 2003; Jordan et al., 2003;
McGregor & Marigold, 2003). Other research (e.g., Kehr, 2004),
like ours, used a procedure that created a continuous score by
taking the difference between standardized explicit and implicit
measures of the self-construct. It is important to note that we had
comparable numbers of individuals on each side of the discrepancy
(implicit ⬎ explicit and explicit ⬎ implicit), and we were able to
show that direction of discrepancy did not moderate the effects of
amount of discrepancy on information processing.
For exploration, we tried an alternative but conceptually similar
analysis in which we used the raw discrepancy scores rather than
the absolute values and tested for curvilinear argument quality
effects across levels of raw discrepancy. In this alternative analy-
sis, both high-positive and high-negative discrepancy individuals
should process the relevant frame message more than people who
have little discrepancy. That is, because the direction of discrep-
ancy does not matter according to our conceptualization, both
high-positive and high-negative discrepancies should look the
same. When analyzed in this manner, the results from our exper-
iments look exactly as predicted. Figure 3 shows the significant
interaction between discrepancy (quadratic term) and argument
quality, ␤ ⫽ .70, t(315) ⫽ 6.87, p ⬍ .0001, collapsing across
Figure 3. Attitudes (standardized scores) as a function of argument
quality and raw discrepancy scores across studies (only relevant-frame
conditions are included for Studies 3 and 4). Actual discrepancy scores in
the studies range from ⫺5.01 to ⫹ 4.52. E-i ⫽ explicit–implicit.
studies and controlling for study as a factor (which did not mod-
erate the critical interaction).
Although our discrepancy index does a good job of capturing
people’s relative standing in the distribution (as in most prior
research), our index can only deal with the range of scores ob-
served in the sample. Thus, by an absolute criterion, it could be the
case that all of the participants are relatively high in both explicit
and implicit self-esteem, high in explicit self-esteem and low in
implicit self-esteem, and so forth. If this were true, it would still be
the case that some people were higher than others on each measure
and that the relative discrepancies our index captures is successful
in predicting the extent of information processing. Nevertheless,
the generality of our conclusions would be limited if all of our
studies had the same profile of responses on the explicit and
implicit measures (e.g., all absolutely high on both, all high on one
and low on the other, etc.).
To explore this issue further, we created an “absolute” discrep-
ancy index in each study by using the middle point of the explicit
scale (e.g., 3 in a 5-point scale) and the zero point in the IAT to
approximate absolute differences between participants high and
low in each of the self-concept dimensions. This categorization
revealed that only Studies 1 and 3 had reasonable numbers of
individuals with discrepancies in both directions by this criterion
(e.g., implicit measure suggests shyness but explicit measure sug-
gests not shy; implicit measure suggests not shy but explicit
measure suggests shy). Notably, these studies produced the same
pattern of results as the studies in which the discrepancies were
present mostly on a relative basis.
Our speculation is that the size of the discrepancy matters more
than whether the discrepancy is relative or absolute. For example,
an individual with trivially positive self-esteem on an explicit
measure and trivially negative self-esteem on an implicit measure
 EXPLICIT–IMPLICIT DISCREPANCY
would be categorized as discrepant in absolute terms, whereas an
individual with trivially positive self-esteem on an explicit mea-
sure and extremely positive self-esteem on an implicit measure
would not, even though the difference between implicit and ex-
plicit scores of the latter individual would be much larger and more
consequential in our view. Nevertheless, future research might
profitably recruit larger samples and use alternative measures so
that the issue of absolute versus relative discrepancies can be
examined more systematically.
Motivation to Reduce Self-Inconsistency
As noted earlier, explicit discrepancies are often associated with
doubt. Campbell (1990) found that self-concept instability corre-
lated with lower subjective confidence in one’s own trait ratings
and also with slower “me/not me” responses to trait adjectives (see
Wright, 2001, for a review). Previous literature has also related
explicit evaluative inconsistency (attitudinal ambivalence) to lack
of confidence. For example, Jonas et al. (1997) provided empirical
evidence that evaluative inconsistency evokes elaboration of re-
lated information to achieve a sufficient level of confidence with
respect to the overall evaluation of the object. Bargh et al. (1992)
suggested that evaluative inconsistency might be related to doubt,
because response latencies (i.e., attitude accessibility) were found
to be slower for ambivalent participants (see also Costelo, Rice, &
Schoenfeld, 1974; Gilmore, 1982). Indeed, one function particular
to ambivalent attitudes—and, perhaps, also to explicit–implicit
discrepant selves—seems to be reducing action readiness and
promoting further and elaborated thinking about related informa-
tion to amplify confidence and knowledge about the target (e.g.,
Hänze, 2001; Hodson et al., 2001; Jonas et al., 1997).
On the basis of this research, it seems reasonable that explicit–
implicit self-discrepancies might also be associated with uncer-
tainty or doubt (Briñol et al., 2003). If explicit–implicit self-
discrepancies are associated with doubt, the enhanced information-
processing effects we observed might be due in part to this
uncertainty. Individuals who are induced to doubt before receiving
a message have been shown to engage in greater thinking (Tiedens
& Linton, 2001; Weary & Jacobson, 1997). For example, in one
study Tiedens and Linton (2001) had participants write about a sad
experience in which they felt uncertain about what was happening
or a sad experience in which they felt certain. Following this doubt
induction, participants received a message containing strong or
weak message arguments. The primary result was that uncertain
participants engaged in greater information processing (i.e., greater
attitudinal differentiation between strong and weak arguments)
than certain participants. We speculate that just as explicit uncer-
tainty can guide information processing, so too might implicit
uncertainty (see also Petty et al., 2006).
Future Directions
Future research might examine other consequences of explicit–
implicit discrepancies apart from enhanced elaboration. In partic-
ular, research might explore other ways in which people could
reduce these discrepancies. These could include changing self-
discrepant elements (e.g., Festinger, 1957; Harmon-Jones & Mills,
1999), minimizing the salience or perceived (explicit or implicit)
importance of the dimension on which the inconsistency exists
167
(e.g., Steele, Southwick, & Critchlow, 1981), or affirming oneself
by expressing important values (e.g., Steele & Liu, 1983). Addi-
tionally, ignoring or defensively avoiding discrepancy-related in-
formation might constitute another strategy when exposure does
not automatically occur, as in the present experiments. Different
mechanisms of reducing self-inconsistency might be substitutable
for each other (e.g., Tesser, 2001). If such mechanisms were
interchangeable, then future research might profitably explore the
potential differential impact of each of the discrepancy reduction
strategies.
A final avenue for further examination concerns the possibility
of actual resolution of the explicit–implicit discrepancy. For ex-
ample, in our Study 1 participants received information directly
relevant to the dimension (shyness) on which the discrepancy
existed. Thus, this information might have influenced their explicit
or implicit (or both) self-dimensions, thereby affecting the subse-
quent explicit–implicit discrepancy. Because follow-up measures
related to the self-dimension were not included in that study after
message processing, such an influence on the resulting discrepancy
could not be assessed. In the other studies of the present research,
it was less likely that message processing would help participants
to resolve their internal conflict toward their implicit or their
explicit self-view because the information provided to them was
not actually relevant to resolving the discrepancy. The experiments
in this article examine whether people are motivated to process
information when they have an explicit–implicit discrepancy,
rather than examining the direction or means of discrepancy res-
olution. This issue constitutes an intriguing question for future
consideration.
References
Abelson, R. P., Aronson, E., McGuire, W. J., Newcomb, T. M., Rosenberg,
M. J., & Tannenbaum, P. H. (1968). Theories of cognitive consistency:
A sourcebook. Chicago: Rand McNally.
Abelson, R. P., & Rosenberg, M. J. (1958). Symbolic psycho-logic: A
model of attitudinal cognition. Behavioral Science, 3, 1–13.
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and
interpreting interactions. Thousand Oaks, CA: Sage.
Aronson, E. (1969). Cognitive dissonance: A current perspective. In L.
Bercowitz (Ed.), Advances in experimental social psychology (Vol. 4,
pp. 1–34). New York: Academic Press.
Asendorpf, J. B. (1987). Videotape reconstruction of emotions and cogni-
tions related to shyness. Journal of Personality and Social Psychology,
53, 542–549.
Asendorpf, J. B. (1989). Shyness as a final common pathway for two
different kinds of inhibition. Journal of Personality and Social Psychol-
ogy, 57, 481– 492.
Asendorpf, J. B. (1990). The expression of shyness and embarrassment. In
W. R. Crozier (Ed.), Shyness and embarrassment: Perspectives from
social psychology (pp. 87–118). New York: Cambridge University
Press.
Asendorpf, J. B., Banse, R., & Mücke, D. (2002). Double dissociation
between implicit and explicit personality self-concept: The case of shy
behavior. Journal of Personality and Social Psychology, 83, 380 –393.
Barger, S. D., Kircher, J. C., & Croyle, R. T. (1997). The effects of social
context and defensiveness on the physiological responses of repressive
copers. Journal of Personality and Social Psychology, 73, 1118 –1128.
Bargh, J. A., Chaiken, S., Govender, R., & Pratto, F. (1992). The generality
of the automatic attitude activation effect. Journal of Personality and
Social Psychology, 62, 893–912.
Bem, D. J., & Allen, A. (1974). On predicting some of the people some of
168 BRIÑOL, PETTY, AND WHEELER
the time: The search for cross-situational consistencies in behavior.
Psychological Review, 81, 506 –520.
Bizer, G. Y., Krosnick, J. A., Holbrook, A. L., Petty, R. E., Wheeler, S. C.,
& Rucker, D. D. (2004). The impact of personality on cognitive, behav-
ioral, and affective political processes: The effects of need to evaluate.
Journal of Personality, 72, 995–1027.
Blair, I. V., & Banaji, M. R. (1996). Automatic and controlled processes in
stereotype priming. Journal of Personality and Social Psychology, 70,
1142–1163.
Bornstein, R. F. (1995). Sex differences in objective and projective depen-
dency tests: A meta-analytic review. Assessment, 2, 319 –331.
Bosson, J. K., Brown, R. P., Zeigler-Hill, V., & Swann, W. B., Jr. (2003).
Self-enhancement tendencies among people with high self-esteem: The
moderating role of implicit self-esteem. Self and Identity, 2, 169 –187.
Bosson, J. K., Swann, W. B., Jr., & Pennebaker, J. W. (2000). Stalking the
perfect measure of implicit self-esteem: The blind men and the elephant
revisited? Journal of Personality and Social Psychology, 79, 631– 643.
Brehm, J. W., & Cohen, A. R. (1962). Explorations in cognitive disso-
nance. New York: Wiley.
Briñol, P., Horcajo, J., Becerra, A., Falces, C., & Sierra, B. (2002). Cambio
de actitudes implı́citas [Implicit attitude change]. Psicothema, 14, 771–
775.
Briñol, P., & Petty, R. E. (2005). Individual differences in persuasion. In D.
Albarracı́n, B. T. Johnson, & M. P. Zanna (Eds.), The handbook of
attitudes and attitude change (pp. 575– 615). Hillsdale, NJ: Erlbaum.
Briñol, P., Petty, R. E., & Wheeler, S. C. (2003, May). Implicit ambiva-
lence: Implications for attitude change. Paper presented at the Attitudi-
nal Incongruence and Information Processing Symposium, University of
Amsterdam, Amsterdam, the Netherlands.
Briñol, P., Rucker, D., Tormala, Z. L., & Petty, R. E. (2004). Individual
differences in resistance to persuasion: The role of beliefs and meta-
beliefs. In E. S. Knowles & J. A. Linn (Eds.), Resistance and persuasion
(pp. 83–104). Mahwah, NJ: Erlbaum.
Cacioppo, J. T., Gardner, W. L., & Berntson, G. G. (1997). Beyond bipolar
conceptualizations and measures: The case of attitudes and evaluative
space. Personality and Social Psychology Review, 1, 3–25.
Campbell, J. D. (1990). Self-esteem and clarity of the self-concept. Journal
of Personality and Social Psychology, 59, 538 –549.
Carlston, D. E., & Skowronski, J. J. (1986). Trait memory and behavior
memory: The effects of alternative pathways on impression judgment
response times. Journal of Personality and Social Psychology, 50, 5–13.
Carver, C. S. (2005). Impulse and constraint: Some perspectives from
personality psychology, convergence with theory in other areas, and
potential for integration. Personality and Social Psychology Review, 9,
312–333.
Carver, C. S., & Scheier, M. F. (1990). Origins and functions of positive
and negative affect: A control-process view. Psychological Review, 97,
19 –35.
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation
analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum.
Costelo, R. M., Rice, D. P., & Schoenfeld, L. S. (1974). Attitudinal
ambivalence with alcoholic respondents. Journal of Consulting and
Clinical Psychology, 42, 303–304.
Dovidio, J. F., Kawakami, K., & Beach, K. (2001). Implicit and explicit
attitudes: Examination of the relationship between measures of inter-
group bias. In R. Brown & S. L. Gaertner (Eds.), Blackwell handbook of
social psychology: Vol. 4. Intergroup relations (pp. 175–197). Oxford,
England: Blackwell.
Emmons, R. A., & Colby, P. M. (1995). Emotional conflict and well-being:
Relation to perceived availability, daily utilization, and observer reports
of social support. Journal of Personality and Social Psychology, 68,
947–959.
Epstein, S. (1973). The self-concept revisited: Or a theory of a theory.
American Psychologist, 28, 404 – 416.
Epstein, S. (2003). Cognitive-Experiential Self-Theory of personality. In T.
Millon & M. J. Lerner (Eds.), Handbook of psychology (Vol. 5, pp.
159 –184). Hoboken, NJ: Wiley.
Farnham, S. D., Greenwald, A. G., & Banaji, M. R. (1999). Implicit
self-esteem. In D. Abrams & M. A. Hogg (Eds.), Social identity and
social cognition (pp. 230 –248). Oxford, England: Blackwell.
Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995).
Variability in automatic activation as an unobtrusive measure of racial
attitudes: A bona fide pipeline? Journal of Personality and Social
Psychology, 69, 1013–1027.
Fazio, R. H., & Olson, M. (2003). Implicit measures in social cognition
research: Their meaning and uses. Annual Review of Psychology, 54,
297–327.
Festinger, L. (1954). A theory of social comparison processes. Human
Relations, 7, 117–140.
Festinger, L. (1957). A theory of cognitive dissonance. Stanford, CA:
Stanford University Press.
Gawronski, B., & Bodenhausen, G. V. (in press). Associative and propo-
sitional processes in evaluation: An integrative review of implicit and
explicit attitude change. Psychological Bulletin.
Gilmore, R. F. (1982). Conflict or dissonance: A reinterpretation of coun-
terattitudinal essay writing. Dissertation Abstracts International, 43,
1659.
Greenier, K. D., Kernis, M. H., & Waschull, S. B. (1995). Not all high (or
low) self-esteem people are the same: Theory and research on stability
of self-esteem. In M. H. Kernis (Ed.), Efficacy, agency, and self-esteem
(pp. 51–71). New York: Plenum Press.
Greenwald, A. G., & Banaji, M. (1995). Implicit social cognition: Atti-
tudes, self-esteem, and stereotypes. Psychological Review, 102, 4 –27.
Greenwald, A. G., Banaji, M. R., Rudman, L. A., Farnham, S. D., Nosek,
B. A., & Mellott, D. S. (2002). A unified theory of implicit attitudes,
stereotypes, self-esteem, and self-concept. Psychological Review, 109,
3–25.
Greenwald, A. G., & Farnham, S. D. (2000). Using the Implicit Associa-
tion Test to measure self-esteem and self-concept. Journal of Personality
and Social Psychology, 79, 1022–1038.
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring
individual differences in implicit cognition: The implicit association
task. Journal of Personality and Social Psychology, 74, 1464 –1480.
Gregg, A. P., Seibt, B., & Banaji, M. R. (2006). Easier done than undone:
Asymmetry in the malleability of implicit preferences. Journal of Per-
sonality and Social Psychology, 90, 1–20.
Hänze, M. (2001). Ambivalence, conflict, and decision making: Attitudes
and feelings in Germany towards NATO’s military intervention in the
Kosovo war. European Journal of Social Psychology, 31, 693–706.
Harmon-Jones, E., & Mills, J. S. (Eds.). (1999). Cognitive dissonance:
Progress on a pivotal theory in social psychology. Washington, DC:
American Psychological Association.
Hass, R. G., Katz, I., Rizzo, N., Bailey, J., & Moore, L. (1992). When
racial ambivalence evokes negative affect, using a disguised measure of
mood. Personality and Social Psychology Bulletin, 18, 786 –797.
Heider, F. (1958). The psychology of interpersonal relations. New York:
Wiley.
Hermans, D., De Houwer, J., & Eelen, P. (2001). A time course analysis of
the affective priming effect. Cognition and Emotion, 15, 143–165.
Hetts, J. J., & Pelham, B. W. (2001). A case for the nonconscious self-
concept. In G. B. Moskowitz (Ed.), Cognitive social psychology: The
Princeton symposium on the legacy and future of social cognition (pp.
105–123). Washington, DC: American Psychological Association.
Hetts, J. J., Sakuma, M., & Pelham, B. W. (1999). Two roads to positive
regard: Implicit and explicit self-evaluation and culture. Journal of
Experimental Social Psychology, 35, 512–559.
Higgins, E. T. (1987). Self-discrepancy: A theory relating self and affect.
Psychological Review, 94, 319 –340.
 EXPLICIT–IMPLICIT DISCREPANCY
Hodson, G., Maio, G. R., & Esses, V. M. (2001). The role of attitudinal
ambivalence in susceptibility to consensus informative. Basic and Ap-
plied Social Psychology, 23, 197–205.
Jarvis, W. B. G. (2000). MediaLab (Version Beta 2000) [Computer soft-
ware]. Columbus, OH: Empirisoft.
Jarvis, W. B. G., & Petty, R. E. (1996). The need to evaluate. Journal of
Personality and Social Psychology, 70, 172–194.
Jonas, K., Diehl, M., & Bromer, P. (1997). Effects of attitudinal ambiva-
lence on information processing and attitude-intention consistency.
Journal of Experimental Social Psychology, 33, 190 –210.
Jones, J. T., Pelham, B. W., Mirenberg, M. C., & Hetts, J. J. (2002). Name
letter preferences are not merely mere exposure: Implicit egotism as
self-regulation. Journal of Experimental Social Psychology, 38, 170 –
177.
Jordan, C. H., Spencer, S. J., Zanna, M. P., Hoshino-Browne, E., & Correll,
J. (2003). Secure and defensive high self-esteem. Journal of Personality
and Social Psychology, 85, 969 –978.
Kaplan, K. J. (1972). On the ambivalence-indifference problem in attitude
theory and measurement: A suggested modification of the semantic
differential technique. Psychological Bulletin, 77, 361–372.
Karpinski, A. (2004). Measuring self-esteem using the Implicit Association
Test: The role of the other. Personality and Social Psychology Bulletin,
30, 22–34.
Katz, I., Wackenhut, J., & Hass, R. G. (1986). Racial ambivalence, value
duality, and behavior. In J. F. Dovidio & S. L. Gaertner (Eds.), Preju-
dice, discrimination, and racism (pp. 35–59). New York: Academic
Press.
Kehr, H. M. (2004). Implicit/explicit motive discrepancies and volitional
depletion among managers. Personality and Social Psychology Bulletin,
30, 315–327.
Kernis, M. H., & Waschull, S. B. (1995). The interactive roles of stability
and level of self-esteem: Research and theory. In M. P. Zanna (Ed.),
Advances in experimental social psychology (Vol. 27, pp. 93–141). San
Diego, CA: Academic Press.
Kitayama, S., & Uchida, Y. (2003). Explicit self-criticism and implicit
self-regard: Evaluating self and friend in two cultures. Journal of Ex-
perimental Social Psychology, 39, 476 – 482.
Koole, S. L., Dijksterhuis, A., & van Knippenberg, A. (2001). What’s in a
name: Implicit self-esteem and the automatic self. Journal of Personality
and Social Psychology, 80, 669 – 685.
Kruglanski, A. W., Thompson, E. P., Higgins, E. T., Atash, M. N., Pierro,
A., Shah, J. Y., & Spiegel, S. (2000). To “do the right thing” or to “just
do it”: Locomotion and assessment as distinct self-regulatory impera-
tives. Journal of Personality and Social Psychology, 79, 793– 815.
Lieberman, M. D. (2000). Intuition: A social cognitive neuroscience ap-
proach. Psychological Bulletin, 126, 109 –137.
Lindsay, D. S., & Johnson, M. K. (1989). The eyewitness suggestibility
effect and memory for source. Memory & Cognition, 17, 349 –358.
Maio, G. R., Bell, D. E., & Esses, V. M. (1996). Ambivalence and
persuasion: The processing of messages about immigrant groups. Jour-
nal of Experimental Social Psychology, 32, 513–536.
Maio, G. R., & Olson, J. M. (Eds.). (2000). Why we evaluate: Functions of
attitudes. Mahwah, NJ: Erlbaum.
McClelland, D. C. (1985). How motives, skills, and values determine what
people do. American Psychologist, 40, 812– 825.
McCrae, R. R., & Costa, P. T., Jr. (1985). Openness to experience. In R.
Hogan & W. H. Jones (Eds.), Perspectives in personality (Vol. 1, pp.
145–172). Greenwich, CT: JAI Press.
McCrae, R. R., & Costa, P. T., Jr. (1997). Conceptions and correlates of
openness to experience. In R. Hogan, J. Johnson, & S. Briggs (Eds.),
Handbook of personality psychology (Vol. 1, pp. 825– 847). New York:
Academic Press.
McGregor, I., & Marigold, D. C. (2003). Defensive zeal and the uncertain
169
self: What makes you so sure? Journal of Personality and Social
Psychology, 85, 838 – 852.
Mendolia, M., Moore, J., & Tesser, A. (1996). Dispositional and situational
determinants of repression. Journal of Personality and Social Psychol-
ogy, 70, 856 – 867.
Metcalfe, J., & Mischel, W. (1999). A hot/cool-system analysis of delay of
gratification: Dynamics of willpower. Psychological Review, 106, 3–19.
Mussweiler, T. (2003). Comparison processes in social judgment: Mech-
anisms and consequences. Psychological Review, 110, 472– 489.
Newcomb, T. M. (1968). Interpersonal balance. In R. Abelson et al. (Eds.),
Theories of cognitive consistency: A sourcebook (pp. 28 –51). Chicago:
Rand McNally.
Newman, L. S., Duff, K. J., & Baumeister, R. F. (1997). A new look at
defensive projection: Thought suppression, accessibility, and biased
person perception. Journal of Personality and Social Psychology, 72,
980 –1001.
Norton, G. R. (1975). Measurement of ambiguity tolerance. Journal of
Personality Assessment, 39, 607– 619.
Osgood, C. E., & Tannenbaum, P. H. (1955). The principle of congruity in
the prediction of attitude change. Psychological Review, 62, 42–55.
Patten, R. L., & White, L. A. (1977). Independent effects of achievement
behavior. Motivation and Emotion, 1, 39 –59.
Petty, R. E. (in press). A meta-cognitive model of attitudes. Journal of
Consumer Research.
Petty, R. E., & Cacioppo, J. T. (1984). The effects of involvement on
responses to argument quantity and quality: Central and peripheral
routes to persuasion. Journal of Personality and Social Psychology, 46,
69 – 81.
Petty, R. E., & Cacioppo, J. T. (1986). Communication and persuasion:
Central and peripheral routes to attitude change. New York: Springer-
Verlag.
Petty, R. E., Fazio, R. H., & Briñol, P. (Eds.). (in press). Attitudes: Insights
from the new implicit measures. Mahwah, NJ: Erlbaum.
Petty, R. E., & Jarvis, B. G. (1996). An individual differences perspective
on assessing cognitive processes. In N. Schwarz & S. Sudman (Eds.),
Answering questions: Methodology for determining cognitive and com-
municative processes in survey research (pp. 221–257). San Francisco:
Jossey-Bass.
Petty, R. E., Tormala, Z. L., Briñol, P., & Jarvis, W. B. G. (2006). Implicit
ambivalence from attitude change: An exploration of the PAST model.
Journal of Personality and Social Psychology, 90, 21– 41.
Petty, R. E., & Wegener, D. T. (1998). Attitude change: Multiple roles for
persuasion variables. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.). The
handbook of social psychology (4th ed., Vol. 1, pp. 323–390). New
York: McGraw-Hill.
Petty, R. E., Wells, G. L., & Brock, T. C. (1976). Distraction can enhance
or reduce yielding to propaganda: Thought disruption versus effort
justification. Journal of Personality and Social Psychology, 34, 874 –
884.
Petty, R. E., Wheeler, S. C., & Tormala, Z. L. (2003). Persuasion and
attitude change. In T. Millon & M. J. Lerner (Eds.), Handbook of
psychology: Vol. 5. Personality and social psychology (pp. 353–382).
Hoboken, NJ: Wiley.
Priester, J. R., & Petty, R. E. (1996). The gradual threshold model of
ambivalence: Relating the positive and negative bases of attitudes to
subjective ambivalence. Journal of Personality and Social Psychology,
71, 431– 449.
Priester, J. R., & Petty, R. E. (2001). Extending the bases of subjective
attitudinal ambivalence: Interpersonal and intrapersonal antecedents of
evaluative tension. Journal of Personality and Social Psychology, 80,
19 –34.
Proshansky, H. M. (1943). A projective method for the study of attitudes.
Journal of Abnormal and Social Psychology, 38, 393–395.
170 BRIÑOL, PETTY, AND WHEELER
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton,
NJ: Princeton University Press.
Roskos-Ewoldsen, D. R., & Fazio, R. H. (1992). On the orienting value of
attitudes: Attitude accessibility as a determinant of an object’s attraction
of visual attention. Journal of Personality and Social Psychology, 63,
198 –211.
Schwartz, S. H. (1992). Universals in the content and structure of values:
Theoretical advances and empirical test in 20 countries. In M. P. Zanna
(Ed.), Advances in experimental social psychology (Vol. 24, pp. 1– 65).
San Diego, CA: Academic Press.
Shastri, L., & Ajjanagadde, V. (1993). From simple associations to sys-
tematic reasoning: A connectionist representation of rules, variables, and
dynamic bindings using temporal synchrony. Behavioral and Brain
Sciences, 16, 417– 494.
Shedler, J., Mayman, M., & Manis, M. (1993). The illusion of mental
health. American Psychologist, 48, 1117–1131.
Sloman, S. A., (1996). The empirical case for two forms of reasoning.
Psychological Bulletin, 119, 3–22.
Smith, E. R., & DeCoster, J. (2000). Dual process models in social and
cognitive psychology: Conceptual integration and links to underlying
memory systems. Personality and Social Psychology Review, 4, 108 –
131.
Smolensky, P. (1988). On the proper treatment of connectionism. Behav-
ioral and Brain Sciences, 11, 1–23.
Steele, C. M., & Liu, T. J. (1983). Dissonance processes as self-
affirmation. Journal of Personality and Social Psychology, 45, 5–19.
Steele, C. M., Southwick, L., & Critchlow, B. (1981). Dissonance and
alcohol: Drinking your troubles away. Journal of Personality and Social
Psychology, 41, 831– 846.
Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of
social behavior. Personality and Social Psychology Review, 8, 220 –247.
Tesser, A. (2001). On the plasticity of self-defense. Current Directions in
Psychological Science, 10, 66 – 69.
Tiedens, L. Z., & Linton, S. (2001). Judgment under emotional certainty
and uncertainty: The effects of specific emotions on information pro-
cessing. Journal of Personality and Social Psychology, 81, 973–988.
Tormala, Z. L., & Petty, R. E. (2001). On-line versus memory-based
processing: The role of “need to evaluate” in person perception. Per-
sonality and Social Psychology Bulletin, 27, 1599 –1612.
Weary, G., & Jacobson, J. A. (1997). Causal uncertainty beliefs and
diagnostic information seeking. Journal of Personality and Social Psy-
chology, 73, 839 – 848.
Weinberger, D. M. (1990). The construct validity of the repressive coping
style. In J. L. Singer (Ed.), Repression and dissociation: Implications for
personality, psychopathology, and health (pp. 337–386). Chicago: Uni-
versity of Chicago Press.
Weinberger, J., & Hardaway, R. (1990). Separating science from myth in
subliminal psychodynamic activation. Clinical Psychology Review, 10,
727–756.
Wiggins, J. S., & Trapnell, P. D. (1997). Personality structure: The return
of the Big Five. In R. Hogan, J. Johnson, & S. Briggs (Eds.), Handbook
of personality psychology (Vol. 1, pp. 737–765). San Diego, CA: Aca-
demic Press.
Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000). A model of dual
attitudes. Psychological Review, 107, 101–126.
Wing, P., & Gough, H. G. (1990). New narcissism scales for the California
Psychological Inventory and MMPI. Journal of Personality Assessment,
54, 3– 4.
Wright, R. (2001). Self-certainty and self-esteem. In T. J. Owens, S.
Stryker, & N. Goodman (Eds.), Extending self-esteem theory and re-
search: Sociological and psychological currents (pp. 101–134). Cam-
bridge, England: Cambridge University Press.
Zelenski, J. M., & Larsen, R. J. (2003, February). The importance of
knowing what we want: How implicit and explicit motives interact to
predict well-being. Paper presented at the meeting of the Society for
Personality and Social Psychology, Los Angeles, CA.
Received June 2, 2003
Revision received February 9, 2006
Accepted February 17, 2006 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 16 –32 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.16

The Single Category Implicit Association Test as a Measure of Implicit

Social Cognition
Andrew Karpinski and Ross B. Steinman
Temple University

The Single Category Implicit Association Test (SC-IAT) is a modification of the Implicit Association
Test that measures the strength of evaluative associations with a single attitude object. Across 3 different
attitude domains—soda brand preferences, self-esteem, and racial attitudes—the authors found evidence
that the SC-IAT is internally consistent and makes unique contributions in the ability to understand
implicit social cognition. In a 4th study, the authors investigated the susceptibility of the SC-IAT to
faking or self-presentational concerns. Once participants with high error rates were removed, no
significant self-presentation effect was observed. These results provide initial evidence for the reliability
and validity of the SC-IAT as an individual difference measure of implicit social cognition.

Keywords: implicit social cognition, associative processes, individual differences

Over the past 20 years, there has been an increasing awareness
that much social cognition occurs outside of conscious awareness
or conscious control (Bargh & Ferguson, 2000; Greenwald, 1992;
Kihlstrom, 1990). Implicit social cognition may be inaccessible to
conscious introspection, and thus it is necessary to develop mea-
sures that do not rely on introspection and self-report in order to
understand and measure these processes (Greenwald & Banaji,
1995). One particularly fruitful approach to measuring implicit
social cognition has been the development of individual difference
measures of associative strength.
Although a large number of these association-based measures of
implicit social cognition have been developed (for a review, see
Fazio & Olson, 2003), the Implicit Association Test (IAT; Green-
wald, McGhee, & Schwartz, 1998) has become the most com-
monly used among the implicit measurement techniques because it
is reliable, easy to administer, and robust and produces large effect
sizes, particularly in comparison to other implicit measures of
social cognition (Greenwald et al., 1998; Greenwald & Nosek,
2001). The IAT is unique among the recently developed
association-based measures of implicit social cognition because it
cannot reveal the evaluative associations with a single target concept.
Because it uses complementary pairs of concepts and attributes, the
IAT is limited to measuring the relative strengths of pairs of associ-
ations rather than absolute strengths of single associations. In practice,
however, the IAT can nevertheless be effectively used because many
socially significant categories form complementary pairs, such as
positive–negative (valence), self– other, male–female, Jewish–Christian,
Andrew Karpinski and Ross B. Steinman, Department of Psychology,
Temple University.
Ross B. Steinman is now at the Department of Psychology, Widener
University.
We thank Shelley Keiper and Jennifer Steinberg for their helpful comments.
Correspondence concerning this article should be addressed to Andrew
Karpinski, Department of Psychology, Temple University, 1701 North
13th Street, Philadelphia, PA 19122-6085. E-mail: andykarp@temple.edu
young– old, weak–strong, warm– cold, liberal– conservative, aggressive–
peaceful, and so forth. (Greenwald & Farnham, 2000, p. 1023)
This property is both a strength and a limitation of the measure. As
highlighted by Greenwald and Farnham (2000), many attitude
objects have a complementary category, and it makes sense to
consider these attitude objects relative to another category.
Yet for some research questions, evaluative associations with a
single target concept may be of interest. For example, to measure
self-esteem by using the IAT, researchers have measured the
positive and negative associations a person has with the self in
comparison to an unspecified other (or with me in comparison to
not me). At the same time, an alternative approach to measuring
self-esteem would be to measure only evaluative associations with
the self with no complementary category (see Karpinski, 2004).
This approach is not possible within the standard IAT paradigm.
Furthermore, there are also instances in which the choice of a
complement is not obvious. Consider a researcher interested in
predicting President Bush’s job approval rating. This researcher
may want to obtain a measure of evaluative associations with
President Bush, but relative to whom? In such instances, it may be
useful to have an IAT-type task that does not require the use of a
complementary category (see also Blanton & Jaccard, 2006; Blan-
ton, Jaccard, Gonzales, & Christie, 2006; De Houwer, 2002).
In addition, greater information may be obtained by measuring
the evaluative associations with two concept domains indepen-
dently rather than examining only comparative associations.
Whereas measures of two concept domains can reveal two dimen-
sions of information, the IAT provides only one. For example, on
a Black–White IAT, scores are interpreted as a comparison of
one’s positive White associations and/or negative Black associa-
tions with one’s negative White associations and/or positive Black
associations. A high score could indicate (a) the presence of many
positive White associations, (b) the presence of many negative
Black associations, (c) the lack of negative White associations,
and/or (d) the lack of positive Black associations. From the single
IAT score, it is impossible to determine which of these factors, or
which combination of these factors, contributes to the overall score

16
 SINGLE CATEGORY IMPLICIT ASSOCIATION TEST 17

(Blanton & Jaccard, 2006; Blanton et al., 2006; Nosek, Greenwald,
& Banaji, 2005). If a single category IAT-type task were available,
then a measure of the evaluative associations with Whites and of
the evaluative associations with Blacks could be obtained inde-
pendently, to eliminate some of the ambiguity in the interpretation
of IAT scores.
Several implicit social cognition measures have been developed
to assess evaluative associations with a single attitude object.
Priming-based measures (see Dovidio, Kawakami, Johnson, John-
son, & Howard, 1997; Fazio, Jackson, Dunton, & Williams, 1995;
Wittenbrink, Judd, & Park, 1997), the Go/No-Go Association Task
(Nosek & Banaji, 2001), and the Extrinsic Affective Simon Task
(De Houwer, 2003) can all be used to assess associations with a
single attitude object. Each of these measures may be used to
assess the evaluative associations with a single target; however, the
reliability of these measures has been very low (see Bosson,
Swann, & Pennebaker, 2000; De Houwer, 2003; Nosek & Banaji,
200l; Olson & Fazio, 2003; Teige, Schnable, Banse, & Asendorpf,
2004). Thus, there is still the need for a reliable individual-
difference measure of the evaluative associations with a single
attitude object.
The Single Category IAT (SC-IAT)
Another possible type of measure to assess the strength of
evaluative associations with a single attitude object is a modified
version of the IAT that eliminates the need for the second contrast
category. We designed the SC-IAT as a two-stage modification of
IAT procedure to measure the evaluative associations with a single
category or attitude object. Because the SC-IAT is a modification
of IAT procedure, it shares many properties with the IAT, includ-
ing its ease of use and interpretation. In each stage, target words
associated with the attitude object and an evaluative dimension are
presented in random order. In the first stage, good words and
attitude object words are categorized on one response key, and bad
words are categorized on a different key. In the second stage, bad
words and attitude object words are categorized on one response
key, and good words are categorized on a different key (see Table
1 for a comparison of a self– other IAT to a self-SC-IAT).
Independently, other researchers have developed similar modi-
fications of the IAT to assess evaluative associations with a single
attitude object. The Single Target IAT (ST-IAT; Wigboldus, Hol-
land, & van Knippenberg, 2005) is conceptually identical to our
SC-IAT, differing only in minor procedural details. In an initial
test of its utility, a Christian ST-IAT was found to correlate with
explicit attitude questions about Christianity, and an Islamic ST-
IAT was found to correlate with explicit attitude questions about
Islam. Neither of the implicit measures correlated with a compar-
ative Christianity versus Islam explicit attitude measure. Addition-
ally, the difference between the Christian ST-IAT and Islamic
ST-IAT correlated weakly and only marginally significantly with
a Christian–Islamic IAT (Wigboldus et al., 2005). These findings
provide some empirical evidence that a measure of associations
with a single category or target may reveal different information
than a comparative IAT; however, additional research is needed to
determine whether single-category or target IAT measures are
reliable, valid across content domains, and relatively impervious to
self-presentation.
Overview of Studies
In the four studies that follow, we examined the reliability,
validity, and susceptibility to faking of the SC-IAT in four differ-
ent concept domains. For all four studies, we also examined the
IAT and explicit measures of attitudes. In cases where the concept
of interest was a comparative domain, a difference between two
SC-IATs (such as a White SC-IAT and a Black SC-IAT) was
computed to obtain a comparative SC-IAT score. We expected an
IAT and a comparative SC-IAT to reveal similar findings in terms
of known groups validity, correlations with explicit measures, and
predictive validity. Unlike the IAT, a comparative SC-IAT can be
decomposed into its components, and thus, the SC-IATs may
allow for specific conclusions in these domains, to supplement the
overall findings obtained from the IAT and comparative SC-IAT.
In cases where measurement of associations with a single category
may be preferable to a comparative measure, compared with the
IAT, we expected the SC-IAT to provide unique and meaningful
information about the category of interest.
We determined a fixed order of measures for each study in order
to reduce the large sample size requirements if we were to coun-
terbalance the presentation of all the measures. We presented all
implicit measures of attitudes prior to the explicit measures of

Table 1
Comparison of the Implicit Association Test (IAT) and Single Category IAT (SC-IAT)

IAT SC-IAT

Block Trials Function Left-key response Right-key response Block Trials Function Left-key response Right-key response

1 30 Practice Pleasant words Unpleasant words

2 30 Practice Self words Other words
3a 30 Practice Pleasant words ⫹ Unpleasant words ⫹ 1c 24 Practice Good words ⫹ self Bad words
self words other words words
4a 30 Test Pleasant words ⫹ Unpleasant words ⫹ 2c 72 Test Good words ⫹ self Bad words
self words other words words
5 30 Practice Other words Self words
6b 30 Practice Pleasant words ⫹ Unpleasant words ⫹ 3d 24 Practice Good words Bad words ⫹ self words
other words self words
7b 30 Test Pleasant words ⫹ Unpleasant words ⫹ 4d 72 Test Good words Bad words ⫹ self words
other words self words

Note. Blocks with a common subscript were experienced as one continuous block.
18 KARPINSKI AND STEINMAN
attitudes. Recently, Nosek and colleagues (2005) have argued that
the order of presentation of implicit and explicit measures does not
alter their psychometric properties or their intercorrelation. Other
studies have found evidence that the completion of explicit mea-
sures of attitudes affects subsequent responses on implicit mea-
sures and artificially increases the observed correlation between
implicit and explicit measures of attitudes (Bosson et al., 2000).
We know of no studies where completing the implicit measures
has significantly affected the responses on subsequent explicit
measures. Thus, to be conservative, we always presented the
implicit measures prior to the explicit self-report measures.
Study 1
The goal of Study 1 was to examine the SC-IAT in a situation
where a comparative measure, such as the IAT, would be useful.
We investigated attitudes toward soda brands, with the goal of
predicting whether participants preferred Coke products to Pepsi
products. Previous studies using the IAT to examine evaluative
associations with Coke and Pepsi have measured associations at
the product level (such as associations with the beverages Coca-
Cola and Pepsi-Cola; see Maison, Greenwald, & Bruin, 2004). In
Study 1, we used the IAT and SC-IAT to measure associations at
the brand level. The Coke brand included the beverages Coke and
Diet Coke; the Pepsi brand included the beverages Pepsi and Pepsi
One. As a consequence, the results of this study were expected to
be similar, but not identical to, the results of previously conducted
studies.
Because this study was the first test of the SC-IAT methodol-
ogy, we examined whether the SC-IAT revealed known soda
attitudes. That is, the SC-IAT should reveal that Coke product
drinkers have more favorable associations with Coke products than
with Pepsi products and that Pepsi product drinkers have more
favorable associations with Pepsi products than with Coke prod-
ucts. When participants choose between a Coke and a Pepsi
product, the outcome variable is a dichotomous choice. Because
the outcome variable is a comparative, dichotomous choice, a
comparative Coke–Pepsi IAT would seem to be an ideal attitude
measure. As a consequence, we expected a soda SC-IAT (the
difference between a Coke SC-IAT and a Pepsi SC-IAT) to be
similar in its effects to a Coke–Pepsi IAT, with regards to known
group differences, correlations with explicit attitude measures, and
prediction of soda choice.
Method
Participants
Fifty-six students (41 women, 15 men) enrolled in an introductory
psychology course at Temple University participated in this experiment.
All participants received course credit for their participation.
Procedure
Participants were tested in groups of up to 3 at a time. Each participant
was seated at a desk with a Gateway 1.5 Gz Pentium 4 desktop computer
using Medialab and Direct RT software. All tasks were presented on the
computer, and all participants completed the tasks in the same order: a
Coke–Pepsi IAT measure of soda associations, a Coke SC-IAT, a Pepsi
SC-IAT, and explicit measures of soda preferences. At the conclusion of
the session, the participants were thanked and completely debriefed.
IAT measure of soda brand associations. A Coke–Pepsi IAT proce-
dure followed the standard IAT paradigm (see Greenwald et al., 1998) with
minor modifications. Specifically, participants completed seven stages in
the same order (see Table 1 for an example). In each stage, participants
responded to 30 target presentations, and the target stimuli were selected
randomly, without replacement. The evaluative dimension was labeled
pleasant and unpleasant, and the object dimension was labeled Coke and
Pepsi. Five target words were used for each of the evaluative dimensions
(pleasant: brilliant, diamond, joy, truth, and sunrise; unpleasant: awkward,
hate, failure, slum, and stink). All target words were presented in lowercase
letters. Five target pictures were also selected to be associated with the
Coke brand (pictures of two-liter bottles and six-packs of cans of Coke and
Diet Coke) and with the Pepsi brand (pictures of two-liter bottles and
six-packs of cans of Pepsi and Pepsi One).
The procedure of the IAT was similar to the SC-IAT with five excep-
tions. First, participants responded by using the a key and the 5 key on the
number keypad to categorize target words and pictures. Second, the target
word remained on the screen until the participants responded. Third,
participants were not given feedback regarding the accuracy of their
responses. Fourth, the category reminder labels were appropriately posi-
tioned in the center of the screen, immediately to the left or right side of the
target word. Fifth, the evaluative dimension was labeled pleasant and
unpleasant.
SC-IAT measure of Coke and Pepsi brand associations. The Coke
SC-IAT consisted of two stages, which all participants completed in the
same order. Each stage consisted of 24 practice trials immediately followed
by 72 test trials (three blocks of 24 trials each). In the first stage (Coke ⫹
good), Coke pictures and good words were categorized on the z key, and
bad words were categorized on the 2 key on the numeric keypad. In an
attempt to prevent a response bias from developing, Coke pictures, good
words, and bad words were not presented at equal frequency, but were
presented in a 7:7:10 ratio so that 58% of correct responses were on the z
key and 42% of correct responses were on the 2 key. In the second stage
(Coke ⫹ bad), good words were categorized on the z key, and Coke
pictures and bad words were categorized on the 2 key on the numeric key-
pad. Coke pictures, good words, and bad words were presented in a 7:10:7
ratio so that 42% of correct responses were on the z key and 58% of correct
responses were on the 2 key.
The evaluative dimension was labeled good and bad, and the object
dimension was labeled Coke. Twenty-one target words were used for each
of the evaluative dimensions (see Appendix), and all target words were
presented in lowercase letters. Seven target pictures were selected to be
associated with Coke (pictures of six-packs and two-liter bottles of Coke
and Diet Coke). Within each category, words and pictures were selected
randomly without replacement.
Each stage was preceded by a set of instructions concerning the dimen-
sions of the categorization task and the appropriate key responses. Each
target word or picture appeared centered on the screen. Category reminder
labels were appropriately positioned on the bottom fourth of the screen.
The target word remained on the screen until the participants responded or
for 1,500 ms. If participants failed to respond within 1,500 ms, a reminder
to “Please respond more quickly!” appeared for 500 ms. This response
window is largely window dressing; pilot testing revealed that the response
window truncates less than 1% of all critical responses. Nevertheless, the
response window creates a sense of urgency and may decrease the likeli-
hood that participants engage in controlled processing during the task.
Following each response, participants were given feedback regarding the
accuracy of their response. A green O in the center of the screen for 150
ms followed correct responses; a red X in the center of the screen for 150
ms followed each incorrect response.
For the Pepsi SC-IAT, the procedure was repeated with the target
category Pepsi and target pictures of Pepsi products (pictures of six-packs
and two-liter bottles of Pepsi and Pepsi One). All participants completed
the Pepsi ⫹ good task followed by the Pepsi ⫹ bad task.
 SINGLE CATEGORY IMPLICIT ASSOCIATION TEST
Explicit measures of soda preferences. Next, participants completed
semantic differential, feeling thermometer, and rating scale measures re-
garding their attitudes toward the Coke and Pepsi brands. For the semantic
differential, participants rated the Coke and Pepsi brands on five bipolar
dimensions: ugly– beautiful, bad– good, unpleasant–pleasant, foolish–
wise, and awful–nice. Each dimension was rated on a 7-point scale ranging
from ⫺3 (the negative pole) to 3 (the positive pole), and participants were
instructed to circle zero if the anchoring adjectives were irrelevant to the
concept (Coke, ␣ ⫽ .88; Pepsi, ␣ ⫽ .92). A semantic differential measure
of soda brand preference was computed by subtracting semantic differen-
tial ratings of the Pepsi brand from semantic differential ratings of the Coke
brand. For the feeling thermometer, participants were asked to rate how
positive or negative they found the Coke and Pepsi brands on a scale from
0 (extremely negative) to 100 (extremely positive). A feeling thermometer
measure of soda brand preference was computed by subtracting feeling
thermometer ratings of Pepsi from thermometer ratings of Coke. A rating
scale measure of Coke brand enjoyment was obtained by asking partici-
pants to indicate their agreement or disagreement on a 6-point scale with
the following statements: “I enjoy drinking Coke (and Coke products)” and
“Coke products satisfy my thirst.” Higher numbers indicated more agree-
ment with the statement (Coke, ␣ ⫽ .49). These questions were repeated
with Pepsi as the target brand (Pepsi, ␣ ⫽ .67). A rating scale measure of
soda preference was computed by subtracting ratings of Pepsi from ratings
of Coke. All three explicit brand attitude measures correlated strongly with
each other (Coke attitudes, ␣ ⫽ .81; Pepsi attitudes, ␣ ⫽ .90; and soda
[Coke – Pepsi] attitudes, ␣ ⫽ .92). As a result, the three explicit measures
were standardized and averaged to create standardized explicit brand
attitude ratings of Coke, Pepsi, and soda (Coke – Pepsi).
Finally, participants answered a behavioral intention question. Partici-
pants indicated whether they would prefer a free Coke or Pepsi product
(indifference and refusal were also response options). On the basis of this
question, participants were defined as Coke drinkers (n ⫽ 17), Pepsi
drinkers (n ⫽ 30), or neither (n ⫽ 6).
Results
IAT and SC-IAT Data Reduction
Compared with the IAT, error rates were significantly higher on
both the Coke SC-IAT, t(55) ⫽ 4.19, p ⬍ .01, and the Pepsi
SC-IAT, t(55) ⫽ 3.61, p ⬍ .01. This result is not surprising given
that the response window in the SC-IAT procedure was included to
facilitate quick responding, and quicker responding is likely to be
accompanied by increased error rates. Participants with an error
rate greater than 20% on the soda IAT, the Coke SC-IAT, or the
Pepsi SC-IAT were excluded from analysis, resulting in the elim-
ination of 3 participants (average error rates: Coke SC-IAT ⫽
6.60%; Pepsi SC-IAT ⫽ 6.43%; soda IAT ⫽ 3.25%).
IAT scores were computed by using the newer D-score algo-
rithm for IAT data (Greenwald, Nosek, & Banaji, 2003). For the
resulting IAT scores, higher numbers indicated a bias for Coke
compared with Pepsi. For the SC-IAT, a scoring algorithm was
modeled on the D-score algorithm used for the IAT data. Because
the 24 practice trials in each stage were truly practice, data from
the practice blocks were discarded (Blocks 1 and 3). Responses
less than 350 ms were eliminated, nonresponses were eliminated,
and error responses were replaced with the block mean plus an
error penalty of 400 ms. The average response times of Block 2
(e.g., Coke ⫹ good) were subtracted from the average response
times of Block 4 (e.g., Coke ⫹ bad). This quantity was divided by
the standard deviation of all correct response times within Blocks
2 and 4. Thus, Coke SC-IAT and Pepsi SC-IAT D scores indicate
19
more positive than negative associations with Coke and Pepsi.
Finally, a soda SC-IAT D score was computed by subtracting the
Pepsi SC-IAT D score from the Coke SC-IAT D score.
Reliability of the SC-IAT and IAT
To determine the reliability of the SC-IAT, we divided each
SC-IAT into thirds (blocks of 24 test trials) and calculated a
SC-IAT score separately for each third of the trials without divid-
ing by the standard deviation of correct response times. A measure
of internal consistency was obtained by calculating the average
intercorrelation among these scores. Dividing the task into thirds
(or halves) underestimates the reliability of the entire measure.
Fortunately, the Spearman–Brown correction can be applied to
compensate for this underestimate of the true internal consistency
for the entire measure (designated adjusted r; Nunnally, 1978). All
internal consistency correlations reported in this article have been
adjusted by using the Spearman–Brown correction. These adjusted
reliability coefficients are conceptually equivalent and directly
comparable to the Cronbach’s alphas computed for the explicit
measures.
A reliability analysis on the SC-IAT measures from Study 1
revealed a reasonable level of internal consistency (Coke, adjusted
r ⫽ .61; Pepsi, adjusted r ⫽ .69). For the IAT, a reliability
correlation was computed by correlating the IAT score computed
from the practice trials with an IAT score computed from the test
trials (following the procedure outlined by Greenwald et al., 2003;
adjusted r ⫽ .82). Overall, the reliability of the SC-IAT is some-
what low compared with the IAT. However, the reliability of the
SC-IAT is similar to the reliability typically observed for IAT
measures (see Greenwald et al., 2003; Nosek et al., 2005) and
higher than the reliability of other implicit measures (see Bosson et
al., 2000; Olson & Fazio, 2003).
Implicit and Explicit Measures of Soda Attitudes
First, we divided the sample into Coke drinkers and Pepsi
drinkers, on the basis of the behavioral choices. The predicted
differences emerged on all the comparative soda measures (see top
of Table 2). Coke drinkers displayed a greater bias for Coke
compared with Pepsi than did Pepsi drinkers on the IAT, SC-IAT,
and explicit attitude measures ( ps ⬍ .02, ds ⱖ 0.87).
Single category measures also tended to reveal the expected
group differences (see bottom of Table 2). Coke drinkers had more
favorable explicit attitudes toward Coke than did Pepsi drinkers
(d ⫽ 0.69), and Pepsi drinkers had more favorable attitudes toward
Pepsi than did Coke drinkers (d ⫽ 2.40). A Coke SC-IAT revealed
no significant difference in evaluative Coke associations for Coke
and Pepsi drinkers (d ⫽ 0.20), but a Pepsi SC-IAT revealed that
Pepsi drinkers had more positive associations with Pepsi than did
Coke drinkers (d ⫽ 0.94).
Correlational Analyses
Soda IAT scores were significantly correlated with soda SC-
IAT scores, r(51) ⫽ .29, p ⫽ .04, suggesting that the soda brand
associations measured by the SC-IAT were significantly related to
the soda brand associations measured by the IAT, as expected.
However, soda IAT scores were unrelated to explicit soda ratings,
20 KARPINSKI AND STEINMAN

Table 2
Study 1: Summary Statistics for the IAT and Explicit Attitude Measures by Soda Preference

Coke drinkers Pepsi drinkers

(n ⫽ 17) (n ⫽ 30) Difference

Attitude measure M SD M SD d t(45) p

Comparative measures

Soda IAT 0.35 0.21 0.02 0.44 0.87 2.91 ⬍.01

Soda SC-IAT 0.34 0.44 ⫺0.03 0.53 0.74 2.47 .02
Explicit soda attitudes 0.91 0.72 ⫺0.56 0.66 2.11 7.08 ⬍.01

Single category measures

SC-IAT
Coke SC-IAT 0.25 0.30 0.18 0.35 0.20 0.68 .50
Pepsi SC-IAT ⫺0.09 0.33 0.22 0.31 0.94 3.16 ⬍.01
Explicit attitudes
Explicit Coke attitude 0.42 0.83 ⫺0.15 0.80 0.69 2.32 .03
Explicit Pepsi attitude ⫺0.79 0.77 0.60 0.42 2.40 8.08 ⬍.01

Note. For comparative soda measures, positive values indicate a bias or preference for Coke over Pepsi. For
single category Coke and Pepsi measures, positive values indicate a bias or preference for Coke or Pepsi,
respectively. IAT ⫽ Implicit Association Test; SC-IAT ⫽ Single Category Implicit Association Test.

r(51) ⫽ .18, p ⫽ .20. Follow-up analyses revealed that soda IAT
scores were also unrelated to explicit Coke or explicit Pepsi
attitudes, r(51) ⫽ .21, p ⫽ .14, and r(51) ⫽ ⫺.05, p ⫽ .74,
respectively.
As expected, soda SC-IAT scores were positively correlated
with explicit soda attitudes, r(51) ⫽ .29, p ⫽ .04. Follow-up
analyses revealed that Coke SC-IAT scores were positively corre-
lated with explicit Coke attitudes, r(51) ⫽ .27, p ⫽ .05, and
uncorrelated with explicit Pepsi attitudes, r(51) ⫽ ⫺.01, p ⫽ .97.
Similarly, Pepsi SC-IAT scores were positively correlated with
explicit Pepsi attitudes, r(51) ⫽ .26, p ⫽ .06, and uncorrelated with
explicit Coke attitudes, r(51) ⫽ ⫺.07, p ⫽ .61. These results
suggest that Coke and Pepsi SC-IAT scores reveal specific infor-
mation pertaining to Coke and Pepsi preferences, respectively, and
not general information about soda preferences.
Additionally, attitudes and associations with Coke were unre-
lated to attitudes and associations with Pepsi on both explicit
ratings, r(51) ⫽ ⫺.01, p ⫽ .93, and the SC-IAT measures, r(51) ⫽
⫺.06, p ⫽ .66.
Prediction of Soda Choice
Next, we examined the ability of the IAT, SC-IAT, and explicit
measures to predict soda choice. For the soda choice outcome
variable, participants who chose a Coke product were given a
value of 1, participants who indicated no preference were given a
value of 0, and participants who chose a Pepsi product were given
a value of ⫺1. First, the Coke SC-IAT, Pepsi SC-IAT, and soda
IAT were entered in a simultaneous regression predicting soda
choice (see top of Table 3). These results indicated that both the
IAT and Pepsi SC-IAT significantly predicted soda choice, but the
Coke SC-IAT was unrelated to soda choice. Next, the analysis was
repeated with explicit soda attitudes added to the model (see
bottom of Table 3). Explicit attitudes were strong predictors of
soda choice. However, even once explicit preferences were con-
trolled, the IAT, Coke SC-IAT, and Pepsi SC-IAT significantly
predicted soda choice.1
An alternative method of comparing the predictive validity of
the SC-IAT and the IAT is to examine the percentage of variance
accounted for by SC-IAT and IAT measures in soda choice. A
series of regression analyses revealed that the Coke and Pepsi
SC-IATs predicted 17% of the variance in soda choice (7% beyond
the variance explained by explicit measures). A soda IAT pre-
dicted 12% of the variance in soda choice (4.9% beyond the
variance explained by explicit measures). Taken together, these
results provide strong evidence for the utility of the SC-IAT above
and beyond the IAT and explicit attitude measures.
Information Obtained From the SC-IAT and IAT
One advantage to using two single attitude measures as opposed
to one comparative attitude is that a greater amount of information
is obtained from the two single attitude measures (see Figure 1).
On the top panel of Figure 1, Coke SC-IAT scores are plotted
against Pepsi SC-IAT scores. From this panel, information about
Coke, Pepsi, and soda (Coke – Pepsi) associations may be ob-
tained. Some participants had favorable associations with both
Coke and Pepsi (the upper right quadrant), some had unfavorable
associations with both Coke and Pepsi (the lower left quadrant),
and some had favorable associations toward one soda and unfa-
1
If the 6 individuals who did not indicate a preference for a Coke or
Pepsi product are eliminated from the analysis, then these relationships can
also be examined by using a logistic regression predicting soda choice. For
a logistic regression with implicit measures only, the results closely parallel
the standard regression analysis. IAT scores ( p ⫽ .05) and Pepsi SC-IAT
scores ( p ⫽ .06) predicted the soda choice, whereas Coke SC-IAT scores
did not ( p ⫽ .63). However, when explicit soda attitudes were added to the
equation, none of the implicit measures or the explicit measure uniquely
predicted soda choice (all ps ⬎ .28).
 SINGLE CATEGORY IMPLICIT ASSOCIATION TEST 21

Table 3 it is somewhat surprising that the soda SC-IAT measure correlated

Study 1: Predicting Soda Choice significantly with explicit soda preferences, whereas the IAT failed
to correlate with explicit measures of soda attitudes. Overall, these
Prediction of soda choice results suggest that the SC-IAT may have some utility above and
Predictor ␤ t p
beyond the IAT even in situations where the outcome variable is
comparative.
Implicit measures only Although the Coke and Pepsi SC-IATs performed well on a
number of aspects, there were a couple of curious aspects of these
Coke SC-IAT .06 0.47 .64
results. First, the reliability of the SC-IAT measures was lower
Pepsi SC-IAT ⫺.35 2.74 ⬍.01
IAT .28 2.10 .04 than the reliability observed for the IAT. One possibility is that not
having a comparative category for the attitude object may result in
Implicit and explicit measures extra error variance in responses. If this were true, then SC-IAT
measures would inherently have lower reliabilities than IAT mea-
Coke SC-IAT .16 1.76 .09
Pepsi SC-IAT ⫺.22 2.33 .02 sures. A second possible explanation for the low reliability of the
IAT .23 2.43 .02 SC-IATs is that the 1,500-ms response window not only resulted
Explicit soda attitudes .67 7.37 ⬍.01 in a higher error rate, but also increased unreliability for the
SC-IAT. A final possibility is that the order of the tasks adversely
Note. N ⫽ 52. All variables were entered simultaneously. SC-IAT ⫽
Single Category Implicit Association Test; IAT ⫽ Implicit Association
affected the reliability of the SC-IAT. For the IAT, all responses to
Test. the attitude objects of interest are comparative. After completing
an IAT, participants may continue to think about the attitude
objects in a comparative manner. If a SC-IAT were to follow an
vorable associations toward the other soda (the remaining two IAT, participants’ responses may be influenced by the comparative
quadrants). Information about participants’ relative soda associa- mind-set induced by the IAT, perhaps resulting in increased error
tions can be obtained by observing where the participant falls in
reference to the dashed diagonal line. Participants with SC-IAT
scores falling above the line had more positive and/or less negative
associations with Coke than Pepsi; participants with SC-IAT
scores falling below the line had more positive and/or less negative
associations with Pepsi than Coke. In the bottom panel of Figure 1,
soda IAT scores are plotted. Only information about participants’
relative soda associations is available from this panel. Participants
with scores on the left side of the graph (scores less than zero) had
more positive and/or less negative associations with Pepsi than
Coke, whereas participants with scores on the right side of the
graph (scores greater than zero) had more positive and/or less
negative associations with Coke than Pepsi.

Discussion
These results provide initial support for the SC-IAT as a valid
measure of evaluative associations with a single attitude object.
The validity of the SC-IAT was established on multiple levels:
known groups validity, convergent validity, and predictive valid-
ity. First, along with the IAT and explicit soda preferences, the
soda SC-IAT discriminated between Coke and Pepsi drinkers.
Second, the SC-IAT measure of soda preferences correlated sig-
nificantly with explicit soda preferences. The Coke and Pepsi
SC-IATs were specific in their measurement of associations. The
Pepsi SC-IAT revealed that Pepsi drinkers had more positive than
negative associations with Pepsi, compared with Coke drinkers.
The Pepsi SC-IAT also correlated with explicit Pepsi attitudes, but
not with explicit Coke attitudes. Likewise, the Coke SC-IAT
correlated with explicit Coke attitudes, but not with explicit Pepsi
attitudes. Third, the Pepsi SC-IAT, and to a lesser extent the Coke
SC-IAT, reliably predicted intended soda choice, even when con-
trolling for IAT scores and explicit attitude ratings. Figure 1. A comparison of information available from independent Coke
This study was ideally suited for a comparative IAT style of and Pepsi Single Category Implicit Association Tests (SC-IATs) and a
measurement, and as expected, the IAT discriminated between Coke–Pepsi Implicit Association Test (IAT). IAT and SC-IAT scores were
Coke and Pepsi drinkers and predicted intended soda choice. Thus, calculated by using the D-score algorithm (N ⫽ 53).
22 KARPINSKI AND STEINMAN
variance in the SC-IAT. The issue of the reliability of the SC-IAT
is one that will be revisited in subsequent studies.
A second curious finding in Study 1 is that all Pepsi measures
(SC-IAT and explicit measures) discriminated between Coke and
Pepsi drinkers better and predicted soda choice better than did
Coke measures. We are puzzled by this finding, but it may be the
result of the Temple University environment. Temple University
has an exclusive agreement with Pepsi to provide soft drinks on
campus. All vending machines and restaurants on campus serve
only Pepsi products. Thus, Temple University students may re-
ceive more information regarding Pepsi products than Coke prod-
ucts, and this environmental bias may lead to greater predictive
validity and known-groups validity for the Pepsi SC-IAT than for
the Coke SC-IAT. Regardless, evidence for the validity of the
Coke SC-IAT was obtained; the Coke SC-IAT correlated with
explicit Coke attitudes and not explicit Pepsi attitudes.
Study 2
A comparative attitude or associative measure is not ideally
suited for all contexts. For example, when investigators wish to
predict smoking behavior, there is no clear complementary behav-
ior to smoking; it depends on the nature of the research question.
Similarly, self-esteem researchers may be interested in assessing
evaluative associations with the self and not comparative self–
other associations (see Karpinski, 2004). In these situations, a
SC-IAT measure of associations may provide a more specific
measure of the evaluative associations in question than an IAT.
Because self-report measures of self-esteem are not explicitly
comparative, we expected to find that a self-SC-IAT would cor-
relate more strongly with explicit measures of self-esteem than
would a self– other IAT.
Study 2 was designed to investigate the use of a SC-IAT as a
measure of self-associations. The procedure of Study 2 closely
followed the procedure of Study 1, with one key exception. To
examine the possibility that taking a (comparative) IAT prior to a
SC-IAT adversely affected the reliability of the SC-IAT, we
switched the order of the SC-IAT and the IAT. Thus, participants
completed a self-SC-IAT, a self– other IAT, and then explicit
measures of self-esteem.
Method
Participants
Sixty-six students (16 men, 41 women, 9 unknown) enrolled in an
introductory psychology course at Temple University participated in this
experiment. All participants received course credit for their participation.
Procedure
All tasks were presented on the computer, and all participants completed
the tasks in the same order: a SC-IAT measure of self-associations, an IAT
measure of self– other associations, and explicit measures of self-esteem.
At the conclusion of the session, the participants were thanked and com-
pletely debriefed.
SC-IAT measure of self-associations. The self-SC-IAT was identical to
the Coke and Pepsi SC-IATs used in Study 1, with the exception of the
target words and category labels. The evaluative dimension remained
labeled good and bad, and the object dimension was labeled self. Five
target words were selected to be associated with the category self ( partic-
ipant’s first name, participant’s last name, me, I, and myself). The good
and bad target words were identical to the words used in Study 1. All target
and category words were presented in lowercase letters. Participants first
completed the self ⫹ positive blocks, followed by the self ⫹ negative
blocks.
IAT measure of self– other associations. The IAT procedure was iden-
tical to the procedure used in Study 1, with the exception of the target
words and category labels. The category labels self and other replaced the
labels of Coke and Pepsi, respectively. Five target words were also selected
to be associated with each of the attitude objects (self: participant’s first
name, participant’s last name, me, I, and myself; other: he, her, his, hers,
and person).
Explicit measures of self-esteem. Participants next completed three
explicit measures of self-esteem: a self semantic differential, a self feeling
thermometer, and the Rosenberg Self-Esteem Scale (Rosenberg, 1965). For
the semantic differential, participants rated self on five bipolar dimensions:
ugly– beautiful, bad– good, unpleasant–pleasant, foolish–wise, and awful–
nice. Each dimension was rated on a 7-point scale ranging from ⫺3
(negative pole) to 3 (positive pole). The five items were summed to form
a semantic differential measure of self-esteem (␣ ⫽ .55). The self feeling
thermometer consisted of a single item, with participants rating themselves
on a thermometer ranging from 0 (cold or unfavorable) to 100 (warm or
favorable). For presentation purposes, feeling thermometer scores have
been rescaled to range from ⫺50 to 50 so that zero indicates neutral
self-attitudes. For the Rosenberg scale, participants responded to each item
on a 7-point scale ranging from 1 (disagree strongly) to 7 (agree strongly).
The 10 items were averaged to compute a measure of self-esteem (␣ ⫽
.86). These three explicit measures of self-esteem were strongly interre-
lated (␣ ⫽ .80), and thus, the three measures were standardized and
averaged to compute a single explicit measure of self-esteem.
Results
IAT and SC-IAT Data Reduction and Reliability Analysis
Because of a computer error, the explicit attitude data were lost
for 9 participants. Replicating the results of Study 1, error rates
were significantly higher for the self-SC-IAT than for the self–
other IAT, t(65) ⫽ 3.15, p ⬍ .01. Participants with an IAT error
rate greater than 20% were excluded from analysis, resulting in the
elimination of 5 participants. Participants with a SC-IAT error rate
greater than 20% were also excluded, resulting in elimination of 9
additional participants. The resulting error rates were consistent
with those observed in Study 1 (self-SC-IAT ⫽ 6.60%, self– other
IAT ⫽ 4.53%).
IAT and SC-IAT scores were computed by using the scoring
algorithms described for Study 1. Self– other IAT scores were
computed so that higher numbers indicate more positive associa-
tions with the self (and/or negative associations with an other) than
negative associations with the self (and/or positive associations
with an other), and the self-SC-IAT scores were such that higher
scores indicate greater positive than negative associations with the
self.
To determine the reliability of the SC-IAT, we divided each
SC-IAT into thirds and calculated a SC-IAT score separately for
each block of 24 test trials. In this study, the SC-IAT displayed a
level of reliability that is similar to the level of reliability typically
found for the IAT (adjusted r ⫽ .73). For the IAT, a reliability
correlation was computed by correlating the IAT score computed
from the practice trials with the IAT score computed from the test
trials. The observed reliability coefficient was on the lower end of
what is typically observed in IAT data (adjusted r ⫽ .58).
 SINGLE CATEGORY IMPLICIT ASSOCIATION TEST 23

Average Levels of Self-Esteem self-SC-IAT revealed more positive than negative self-
associations, suggesting positive self-esteem in the sample. Unlike
Overall, all measures showed a pattern consistent with an inter- the self– other IAT, the self-SC-IAT correlated significantly with
pretation of high self-esteem (see Table 4). All three explicit explicit measures of self-esteem. These correlations are larger than
measures of self-esteem revealed the presence of positive self- correlations typically observed between implicit and explicit mea-
evaluations (all ds ⱖ 1.35). Implicit measures revealed a similar sures of self-esteem (see Bosson et al., 2000; Greenwald & Farn-
pattern. The IAT revealed that participants had more positive ham, 2000) but are similar in size to correlations found between an
self-associations and/or negative other associations than negative affective priming measure of self-esteem and explicit measures of
self-associations and/or positive other associations (d ⫽ 1.84). The self-esteem (Wentura, Kulfanek, & Greve, 2005).
SC-IAT revealed that participants had more positive self- In addition, a small and nonsignificant relationship was ob-
associations than negative self-associations (d ⫽ 1.13). No signif- served between self-SC-IAT scores and self– other IAT scores.
icant gender differences were observed on any of the measures of Karpinski (2004) hypothesized that the evaluative self– other as-
self-esteem (all ps ⱖ .13). sociations measured by the IAT may be qualitatively different
from evaluative self-associations, and these results provide support
Relationship Between Implicit and Explicit Measures of for this claim. The self– other IAT has proven to be a useful
Self-Esteem measure of self-esteem in many contexts, yet the current findings
suggest that a self-SC-IAT may provide additional information
It is interesting to note that the self-SC-IAT and self– other IAT
about implicit self-esteem that is not captured by the self– other
scores were only marginally correlated, r(50) ⫽ .25, p ⫽ .07. In
IAT. However, the correlation we observed between explicit mea-
other words, the self-associations measured by the SC-IAT were
sures of self-esteem and the self– other IAT is lower than what is
only weakly related to the self– other associations measured by the
typically reported (e.g., see Bosson et al., 2000). The reasons for
self– other IAT. The self– other IAT failed to correlate with the
this discrepancy are unclear, but the end result is that this study
standardized explicit measure of self-esteem, r(42) ⫽ .01, p ⫽ .93.
may underestimate the IAT– explicit self-esteem relationship and
Conversely, a significant positive correlation was observed be-
the SC-IAT/IAT relationship.
tween the self-SC-IAT and the explicit measure of self-esteem,
The results of Study 2 provided stronger evidence for the
r(42) ⫽ .38, p ⫽ .01.
reliability of the SC-IAT than did Study 1. In this study, the
A regression analysis was conducted to investigate whether the
SC-IAT displayed a level of internal consistency similar to the
self-SC-IAT predicted unique variance in explicit reports of self-
reliabilities that are typically found by using the IAT. One differ-
esteem. Consistent with the correlation findings, the regression
ence between this study and Study 1 is that in the current study, the
analysis revealed that self-SC-IAT scores were uniquely predictive
SC-IAT measure was obtained prior to the IAT measure. This
of explicit self-esteem (␤ ⫽ .40, p ⫽ .01), whereas IAT scores
result provides indirect support to the hypothesis that the comple-
were not uniquely predictive of explicit self-esteem (␤ ⫽ ⫺.09,
tion of a (comparative) IAT may interfere with ensuing SC-IAT
p ⫽ .54).
measures. Thus, for subsequent studies, we will present the SC-
IAT measures prior to the IAT measures.
Discussion
Study 2 provided additional evidence for the reliability and Study 3
validity of the SC-IAT as a measure of implicit social cognition.
Like the self– other IAT and explicit measures of self-esteem, the One of the more interesting applications of the IAT is its use as
a measure of implicit racial associations. Studies using a Black–
White IAT have typically found that White participants display a
large racial bias in favor of Whites and/or against Blacks (Das-
Table 4 gupta, McGhee, Greenwald, & Banaji, 2000; Greenwald et al.,
Study 2: Descriptive Statistics 1998; Monteith, Voils, & Ashburn-Nardo, 2001; Nosek, Banaji, &
Difference from midpoint
Greenwald, 2002). One possible interpretation of this effect is that
“virtually all White Americans may have automatic negative as-
Measure M SD t p d sociations to African American names” (Greenwald et al., 1998, p.
1475). Yet, because of the comparative nature of the IAT, there are
Implicit measures
multiple interpretations of this IAT race bias. For example, a
Self–other IAT 0.58 0.32 t(51) ⫽ 13.24 ⬍.01 1.84 person who has no evaluative associations with Blacks, and mostly
Self-SC-IAT 0.45 0.40 t(51) ⫽ 8.13 ⬍.01 1.13 positive associations with Whites, would display the typical IAT
race bias. Likewise, an IAT racial bias may emerge for participants
Explicit measures who have mostly positive associations with Blacks but also have
Rosenberg SE Scale 5.39 1.03 t(43) ⫽ 8.94 ⬍.01 1.35 more positive associations with Whites (Blanton & Jaccard, 2006;
Self semantic differential 8.74 2.93 t(43) ⫽ 19.75 ⬍.01 2.98 Blanton et al., 2006; Gehring, Karpinski, & Hilton, 2003). In these
Self feeling thermometer 27.70 18.54 t(43) ⫽ 9.91 ⬍.01 1.49 alternative interpretations of the IAT race bias, participants would
not have automatic negative associations with Blacks.
Note. Midpoint is the middle point of the scale or the point of the scale
at which a person has neutral self-associations. IAT ⫽ Implicit Association
We hypothesized that the SC-IAT could be used to help inter-
Test; SC-IAT ⫽ Single Category Implicit Association Test; SE ⫽ Self- pret meaning of the IAT race bias. For Study 3, separate Black and
Esteem. White SC-IATs were used to assess the strength of evaluative
24 KARPINSKI AND STEINMAN
associations with Blacks and Whites independently. We expected
a race bias SC-IAT (White SC-IAT–Black SC-IAT) to reveal a
similar race bias as the IAT; White SC-IAT scores were hypoth-
esized to be higher than Black SC-IAT scores. If this effect is due
to out-group prejudice, then participants would also have negative
Black SC-IAT scores. However, if the SC-IAT and IAT race
biases are a result of in-group favoritism, then we would expect to
observe positive evaluative associations on a White SC-IAT and
less positive (but not negative) evaluative associations on a Black
SC-IAT. Furthermore, we explored differences in automatic race
biases between White and Black participants. Previous studies
have found that Black participants tend to show either no IAT race
bias or a small IAT race bias in favor of Whites and/or against
Blacks (Nosek et al., 2002).
In Study 3, participants completed a White SC-IAT, a Black
SC-IAT, a Black–White IAT, and explicit measures of racial
attitudes. Several modifications were made to both the procedures
of the IAT and SC-IAT. First, participants taking the IAT received
response feedback, seeing a green O following a correct response
and a red X following an incorrect response. This modification is
in line with standard versions of the IAT that typically include
response feedback. Second, we reduced the number of critical
trials in each SC-IAT from 72 to 48 in order to reduce the time
required for each SC-IAT. Third, we increased the response win-
dow for the SC-IAT from 1,500 ms to 2,000 ms. If the 1,500-ms
response window was contributing to high error rates on the
SC-IAT, then increasing the response window may alleviate this
problem.
Method
Participants
Eighty-one White students (24 men, 57 women) and 37 Black students
(9 men, 28 women) enrolled in an introductory psychology course at
Temple University participated in this experiment. All participants re-
ceived course credit for their participation.
Procedure
The procedure mirrored the procedure of Study 1, with Black American
and White American as the categories of interest. All tasks were presented
on the computer, and all participants completed the tasks in the same order:
a White-SC-IAT, a Black-SC-IAT, a Black–White IAT, and explicit mea-
sures of racial attitudes. At the conclusion of the session, the participants
were thanked and completely debriefed.
SC-IAT measure of Black and White associations. The Black SC-IAT
and White SC-IAT were similar to the soda SC-IATs used in Study 1, with
three exceptions. First, the target words and category labels were changed
to be relevant for the categories Black and White. The evaluative dimen-
sion remained labeled good and bad; the category dimension was labeled
White American for the White SC-IAT and Black American for the Black
SC-IAT. Six Black faces and six White faces were selected to be associated
with the categories Black and White, respectively. These face stimuli were
identical to those used by Nosek et al. (2002). Second, the response
window was lengthened from 1,500 ms to 2,000 ms. If participants did not
respond within 2,000 ms of the presentation of the target word or face, they
were reminded to respond more quickly. Third, the number of critical trials
was reduced from 72 trials to 48 trials.
IAT measure of Black–White associations. The IAT procedure was
identical to the procedure used in Study 1, with two exceptions. First, the
target words and pictures and category labels were modified to be appro-
priate for a Black–White IAT. The category labels White American and
Black American replaced the labels of Coke and Pepsi, respectively. Six
target faces were also selected to be associated with each of the attitude
objects (three male and three female for each race, identical to the stimuli
used for the SC-IATs). In addition, several of the unpleasant target words
were changed to avoid overlap with the Black stereotype (see Appendix).
Finally, unlike Studies 1 and 2, participants received response feedback for
correct and incorrect responses. Participants viewed a green O for 150 ms
following a correct response and a red X for 150 ms following an incorrect
response.
Explicit measures of race attitudes. Participants next completed two
explicit measures of White and Black attitudes: a semantic differential
(White, ␣ ⫽ .84; Black, ␣ ⫽ .89) and a feeling thermometer. These
measures were identical to those used in the previous studies, but the target
category labels White American and Black American were used to assess
attitudes toward Whites and Blacks, respectively. These two explicit atti-
tudes measures were strongly related (White, ␣ ⫽ .64; Black, ␣ ⫽ .81), and
thus they were standardized and averaged to form composite explicit
attitude measures of Whites and Blacks. Explicit measures of comparative
racial attitudes were obtained by subtracting the explicit measures of
Blacks from the explicit measures of Whites, such that higher numbers
indicated more favorable attitudes of Whites compared with Blacks.
Results
IAT and SC-IAT Data Reduction and Reliability Analysis
Compared with the IAT, error rates were significantly higher on
the White SC-IAT, t(117) ⫽ 7.93, p ⬍ .01, and the Black SC-IAT,
t(117) ⫽ 7.40, p ⬍ .01. Participants with an error rate greater than
20% on the Black SC-IAT, the White SC-IAT, or race IAT were
excluded from analysis, resulting in the elimination of 5 partici-
pants. The resulting average error rates were consistent with error
rates observed in previous studies (White SC-IAT ⫽ 6.15%; Black
SC-IAT ⫽ 5.97%; race IAT ⫽ 3.08%). These findings suggest that
increasing the response window did not result in a reduction of the
SC-IAT error rate.
As in the previous studies, IAT and SC-IAT scores were com-
puted by using the D algorithm. A comparative race SC-IAT score
was obtained by subtracting Black SC-IAT scores from White
SC-IAT scores. For both the race SC-IAT and IAT, positive scores
indicate more positive associations with Whites (and/or negative
associations with Blacks) than negative associations with Whites
(and/or positive associations with Blacks).
To determine the reliability of the SC-IAT, we calculated a
SC-IAT score for each of the two test blocks of 24 test trials. A
reliability analysis on these scores revealed greater internal con-
sistency for the White SC-IAT (adjusted r ⫽ .70) than for the
Black SC-IAT (adjusted r ⫽ .55). For the IAT, a reliability
correlation was computed by correlating IAT scores computed
from the practice trials with IAT scores computed from the test
trials (adjusted r ⫽ .75). Similar to Study 1, the SC-IAT displayed
lower levels of internal consistency than did the IAT.
Implicit and Explicit Measures of Racial Attitudes
Explicit measures. For the comparative explicit race bias mea-
sures, White and Black participants reported an in-group bias (see
top of Table 5). On both the feeling thermometer and semantic
differential, White participants rated Whites more positively than
Blacks ( ps ⱕ .01, ds ⱖ 1.39), and Black participants rated Blacks
 SINGLE CATEGORY IMPLICIT ASSOCIATION TEST 25

Table 5
Study 3: Descriptive Statistics

Difference between
White and Black
Whites (n ⫽ 79) Blacks (n ⫽ 34) participants

Attitude measure M SD d M SD d t(111) p d

Explicit measures

Race feeling thermometer 9.19 16.22 0.57 ⫺15.21 16.26 ⫺0.94 7.33 ⬍.01 1.39
White feeling thermometer 24.35 14.19 1.72 18.91 17.36 1.09 1.75 .08 0.33
Black feeling thermometer 15.16 16.65 0.91 24.12 15.82 2.15 5.63 ⬍.01 1.07
Race semantic differential 0.40 1.02 0.39 ⫺0.47 0.75 ⫺0.62 4.45 ⬍.01 0.85
White semantic differential 0.85 0.84 1.01 0.61 0.82 0.75 1.38 .17 0.26
Black semantic differential 0.45 0.80 0.56 1.08 0.98 1.10 3.60 ⬍.01 0.68

Implicit measures

Race IAT 0.63 0.30 2.07 0.09 0.42 0.22 7.68 ⬍.01 1.46
Race SC-IAT 0.16 0.52 0.30 ⫺0.07 0.63 ⫺0.10 2.03 .04 0.39
White SC-IAT 0.17 0.37 0.46 0.16 0.44 0.36 0.18 .86 0.03
Black SC-IAT 0.01 0.33 0.04 0.22 0.43 0.52 2.80 ⬍.01 0.49

Note. For comparative race measures, higher numbers indicate more positive (or fewer negative) attitudes or
associations with Whites than with Blacks. For single category measures, higher numbers indicate more positive
attitudes. IAT ⫽ Implicit Association Test; SC-IAT ⫽ Single Category Implicit Association Test.

more positively than Whites ( ps ⱕ .01, ds ⱖ .62). Examination of
the separate ratings of Whites and Blacks revealed that these
in-group biases were due to in-group favoritism and not out-group
prejudice. On both the semantic differential and feeling thermom-
eter measures, White and Black participants reported favorable
attitudes toward both Whites and Blacks (all ps ⱕ .01, ds ⱖ 0.56).
Implicit measures. As expected, for White participants, the
implicit comparative measures of racial attitudes revealed signif-
icant race biases (see bottom of Table 5). The IAT revealed a large
race bias (d ⫽ 2.07, p ⬍ .01), whereas the SC-IAT revealed a
small- to medium-sized race bias (d ⫽ 0.30, p ⬍ .01). Echoing the
findings on the explicit measures, these findings indicate that
White participants showed an in-group bias at the implicit level,
favoring Whites over Blacks. For Black participants, both implicit
measures of race bias revealed no significant racial biases (IAT,
d ⫽ 0.22, p ⫽ .22; SC-IAT, d ⫽ ⫺0.10, p ⫽ .57). A test for the
difference between these effects revealed that White participants
displayed a significantly stronger in-group bias than did Black
participants on both the IAT (d ⫽ 1.46, p ⬍ .01) and the SC-IAT
(d ⫽ 0.39, p ⫽ .04). Because the IAT cannot be decomposed into
component attitudes of Whites and attitudes of Blacks, no further
conclusions can be made about the nature of the race bias observed
with the IAT.
However, the race SC-IAT is composed of a difference between
a White SC-IAT and a Black SC-IAT, and each of these compo-
nents can be analyzed separately. An analysis of the White SC-IAT
component revealed that both White and Black participants had
more positive than negative associations with Whites (ds ⱖ 0.34,
ps ⱕ .04), and there was no significant difference in the valence of
White associations held by Whites and Blacks, t(111) ⫽ 0.18, p ⫽
.86, d ⫽ 0.03. But for the Black SC-IAT component, Black
participants had significantly more positive associations with
Blacks than did White participants, t(111) ⫽ 2.80, p ⬍ .01, d ⫽
0.49. Follow-up analyses revealed that Black participants had
positive associations with Blacks (d ⫽ 0.52, p ⬍ .01), and White
participants had equally positive and negative associations with
Blacks (d ⫽ 0.04, p ⫽ .72). These results suggest that, on average,
the SC-IAT racial bias emerged for Whites because of implicit
in-group favoritism (favorable associations with Whites) and not
because of implicit out-group prejudice (negative associations with
Blacks).
SC-IAT scores can also be used to help interpret the contribu-
tion of evaluative White and Black associations to IAT scores. To
investigate whether IAT scores reflect meaningful variations in
both evaluative White and evaluative Black associations, a regres-
sion analysis was conducted to predict IAT scores from Black
SC-IAT and White SC-IAT scores. This analysis revealed that race
IAT scores were positively associated with White SC-IAT scores
(␤ ⫽ .24, p ⬍ .01) and negatively associated with Black SC-IAT
scores (␤ ⫽ ⫺.30, p ⬍ .01). Thus, as expected, a Black–White
IAT measured evaluative associations with both Blacks and
Whites.
Correlations between measures. As expected, race SC-IAT
scores were significantly correlated with IAT scores, r(111) ⫽ .40,
p ⬍ .01. This relationship remained significant when examining
only White participants, r(77) ⫽ .36, p ⬍ .01, and also when
examining only Black participants, r(32) ⫽ .40, p ⫽ .02. Further-
more, White SC-IAT scores were uncorrelated with Black SC-IAT
scores, r(111) ⫽ ⫺.11, p ⫽ .23, suggesting that evaluative asso-
ciations with Whites were independent from evaluative associa-
tions with Blacks.
All of the implicit measures tended to not be correlated with
explicit measures of racial bias (see Table 6). For White partici-
pants in this sample, both the IAT and a race SC-IAT failed to
correlate with explicit measures of racial attitudes (all ps ⬎ .22),
whereas for Black participants, both the IAT and a race SC-IAT
26 KARPINSKI AND STEINMAN

Table 6 Whites. However, neither the IAT nor the SC-IATs consistently
Study 3: Correlations Between Implicit and Explicit Measures correlated with the explicit attitudes measures.
One area of concern regarding the SC-IAT is the low reliability
SC-IAT measures coefficients of the SC-IAT measures, particularly the Black SC-
Race Race White Black
IAT. Consistent with the findings of Study 1, the SC-IATs dis-
Explicit attitude measure IAT SC-IAT SC-IAT SC-IAT played lower levels of internal consistency than did the IAT.
Although the White SC-IAT displayed adequate levels of reliabil-
White participants (n ⫽ 79) ity, the reliability of the Black SC-IAT was poor. In this study, the
number of critical trials in the SC-IAT procedure was reduced
Explicit race attitudes ⫺.05 .14 .17 ⫺.08
Explicit attitudes toward Whites .08 .17 .16 ⫺.08 from 72 to 48. It is possible, and indeed likely, that this modifi-
Explicit attitudes toward Blacks .14 .01 .17 ⫺.02 cation adversely affected the reliability of the SC-IAT measure. A
second modification to the SC-IAT procedure was an increase in
Black participants (n ⫽ 34) the length of the response window from 1,500 ms to 2,000 ms.
Explicit race attitudes .29† .29† .48* .10 This change was intended to reduce the error rate, but this change
Explicit attitudes toward Whites .07 .01 .25 .23 was ineffective. The reliability and validity of the SC-IAT appear
Explicit attitudes toward Blacks ⫺.23 ⫺.23 ⫺.15 .16 to be greater when the SC-IAT procedure includes at least 72
critical trials and a 1,500-ms response window, as was used in
Note. SC-IAT ⫽ Single Category Implicit Association Test; IAT ⫽
Implicit Association Test.
Study 2.
† p ⬍ .10. * p ⬍ .01. A more troubling possibility is that the low reliability of the
SC-IAT was due to participants attempting to respond to the task
in a socially desirable manner. This concern arises from the finding
that the Black SC-IAT had poor reliability and from the reduced
correlated marginally with explicit measures of racial attitudes (all effect size on the race SC-IAT in White participants, compared
ps ⱕ .10). Additionally, little evidence was found for correlations with the IAT. The SC-IAT was designed to be relatively imper-
between the White and Black SC-IATs and their respective ex- vious to controlled processing in general and especially to social
plicit attitude measures, either for White or Black participants (all desirability concerns. The moderate-sized correlation between the
ps ⬎ .15). IAT and a race SC-IAT suggests that social desirability pressures
might not have affected SC-IAT responses in a sizable way.
Discussion Nevertheless, the possibility that the SC-IAT may not be as im-
plicit as previously thought is a disturbing finding, and Study 4
Study 3 provided evidence for the validity of the SC-IAT on was designed to investigate this issue more thoroughly.
multiple levels. First, similar mean-level effects were observed on
the IAT and the race SC-IAT. On both these measures, White Study 4
participants displayed a pro-White bias, and Black participants
displayed no significant evaluative race bias. For White partici- One of the advantages of implicit measures of social cognition
pants, the race bias observed by using the IAT was substantially is that they are impervious to self-presentational motivations and
larger than the race bias observed by using the SC-IAT. However, social desirability concerns. Thus, for the SC-IAT to be useful as
because no objective measure of implicit race bias exists, it is an implicit measure of social cognition, it must be shown that
unclear whether this difference in effect sizes between the mea- participants cannot control or fake their responses on the task.
sures is meaningful. There are reasons to suspect that it may be easier for participants
Second, the White SC-IAT and Black SC-IAT facilitated the to control or fake their responses on the SC-IAT than on the IAT.
interpretation of IAT and race SC-IAT scores. For White partici- Undoubtedly, it is easier for participants to reconceptualize the
pants, the SC-IAT race bias emerged because of positive evalua- instructions in order to disrupt measurement of the associations of
tive White associations and neutral evaluative Black associations; interest on the SC-IAT than on the IAT. For example, participants
for Black participants, the lack of a SC-IAT race bias emerged may recode the SC-IAT instructions as follows: “If it is a ‘bad’
because of positive evaluative Black associations and equally target word, press the right key; for all other target words, press the
positive evaluative White associations. This pattern of results left key.” Similarly, if the target category is represented by pictures
suggests that the SC-IAT and IAT race biases observed in White and the evaluative dimension is represented by words (as in
participants were, on average, a result of in-group favoritism and Studies 1 and 3), participants may use an “all pictures to the left”
not out-group prejudice. At the very least, the current results show rule and avoid encoding the content of the target pictures. Al-
that IAT and SC-IAT race biases can be due to in-group favoritism though participants could use these strategies to control or fake
and may occur in the absence of out-group negativity. their responses on the SC-IAT, it is not clear that they do use these
Third, the race SC-IAT correlated significantly with the race strategies spontaneously.
IAT. Thus, there is some evidence that the race SC-IAT and the Study 4 was conducted to investigate participants’ ability to
race IAT are measuring similar aspects of Black–White evaluative control or fake their responses on a SC-IAT measuring evaluative
associations. These results suggest that a Black–White IAT, a associations with women. Participants were randomly assigned to
White SC-IAT, and a Black SC-IAT may all provide useful infor- display positive or negative attitudes toward women and then
mation regarding evaluative associations regarding Blacks and completed a SC-IAT measuring evaluative associations with
 SINGLE CATEGORY IMPLICIT ASSOCIATION TEST
women. Because the modifications of the SC-IAT procedure used
in Study 3 (reducing the number of critical trials in each stage to
48 and increasing the response window to 2,000 ms) were inef-
fective, we used a SC-IAT with 72 critical trials in each stage and
with a 1,500-ms response window (as was used in Study 2). For
comparison purposes, participants also completed a male–female
IAT and explicit attitude measures toward women.
We expected that participants would easily be able to present the
desired response on the explicit attitude measures. A number of
researchers have investigated the susceptibility of the IAT to
faking. Steffens (2004) reviewed the literature and concluded that
small, nonsignificant effects of faking on the IAT were typically
observed. Thus, we anticipated a small effect of instruction on the
male–female IAT scores. The primary question of interest was the
effect of instruction on the female SC-IAT scores and how this
effect compared with the effect observed for the IAT and explicit
attitude measures.
Method
Participants
Eighty-four students (20 men, 64 women) enrolled in an introductory
psychology course at Temple University participated in this study. All
participants received course credit for their participation.
Procedure
Participants were randomly assigned to respond to all questions and
tasks as if they had very positive attitudes or very negative attitudes toward
women. Participants then completed a SC-IAT measure of evaluative
associations with women, a male–female IAT, and self-report measures of
attitudes toward women.
SC-IAT measure of female associations. The female SC-IAT was
identical to the self-SC-IAT procedure used in Study 2, with appropriate
modifications to measure associations with women. The evaluative dimen-
sion was labeled good and bad, and the concept dimension was labeled
female. Five target words were used for the category female (her, woman,
girl, she, and lady).
IAT measure of male–female associations. The IAT procedure was
identical to the procedure used in Study 3 (i.e., unlike Studies 1 and 2, the
procedure included response feedback), with appropriate modifications to
measure associations with men and women. The evaluative dimension was
labeled pleasant and unpleasant, and the attitude object dimension was
labeled female and male. Five target words were used for each of the
attitude object dimensions (male: him, man, boy, he, and guy; female: her,
woman, girl, she, and lady). In Blocks 3 and 4, participants paired fe-
male ⫹ pleasant and male ⫹ unpleasant. This pairing was reversed in
Blocks 6 and 7: male ⫹ pleasant and female ⫹ unpleasant.
Explicit measures of attitudes toward females. Participants next com-
pleted three explicit measures of attitudes toward women: a female seman-
tic differential (␣ ⫽ .98), a female feeling thermometer, and the Modern
Sexism Scale (Swim, Aikin, Hall, & Hunter, 1995). The semantic differ-
ential and feeling thermometer measures were identical to those used in the
previous studies, but the target category was changed to be females.
Participants responded to the items of the Modern Sexism Scale on a scale
from 1 (strongly disagree) to 5 (strongly agree; ␣ ⫽ .88). Modern sexism
scores were recoded so that higher numbers indicated less sexism. The
three attitude measures were highly interrelated (␣ ⫽ .92) and thus were
averaged to compute a composite measure of self-reported attitudes toward
women.
27
Results
IAT and SC-IAT Data Reduction and Reliability Analysis
As in the previous studies, IAT scores were computed by using
the D algorithm, such that higher numbers indicate a profemale
(and/or antimale) bias. For the IAT, a reliability coefficient was
computed by correlating an IAT score computed from the practice
trials with an IAT score computed from the test trials (adjusted r ⫽
.78). SC-IAT scores were computed by using the modified D
algorithm, such that higher numbers indicate more positive than
negative associations with women. To determine the reliability of
the SC-IAT, a SC-IAT score was calculated for each of the three
test blocks of 24 test trials (adjusted r ⫽ .85). Similar to Study 2,
the SC-IAT displayed slightly higher levels of internal consistency
than the IAT.
Effect of the Self-Presentation Instructions
As expected, participants instructed to present positive attitudes
toward women displayed significantly more positive explicit atti-
tudes toward women than did participants instructed to present
negative attitudes toward women (present positive, M ⫽ 0.71;
present negative, M ⫽ ⫺0.78), t(82) ⫽ 11.84, p ⬍ .01, d ⫽ 2.61.
For the IAT, we first examined whether the error rate differed
across conditions. A t test of the IAT error rate revealed no
significant difference in the IAT error rate as a function of pre-
sentation instructions, t(82) ⫽ 0.40, p ⫽ .69, d ⫽ 0.09. Following
the procedure from the previous studies, 1 participant with an error
rate larger than 20% was excluded from the remaining IAT anal-
yses, and the resulting error rate was 5.30%. Somewhat surpris-
ingly, a medium instruction effect was observed for the IAT.
Participants instructed to present positive attitudes toward women
displayed significantly higher IAT scores than did participants
instructed to present negative attitudes toward women (present
positive, M ⫽ 0.60; present negative, M ⫽ 0.37), t(81) ⫽ 2.03, p ⫽
.05, d ⫽ 0.45.2
For the SC-IAT, we also examined whether the error rate
differed across conditions. A t test of the SC-IAT error rate
revealed a significant medium-sized difference in the SC-IAT
error rate as a function of presentation instructions, t(82) ⫽ 2.15,
p ⫽ .04, d ⫽ 0.48. Participants who attempted to present negative
attitudes toward women had a significantly higher error rate on the
task (M ⫽ 18.04%) than did participants who attempted to present
positive attitudes toward women (M ⫽ 12.94%). An analysis of the
SC-IAT scores of all participants revealed a large instruction effect
(present positive, M ⫽ 0.68; present negative, M ⫽ 0.28), t(82) ⫽
3.31, p ⬍ .01, d ⫽ 0.73. However, the standard practice with IAT
and SC-IAT data is to exclude participants with high error rates
from the analyses. Once the 25 participants with a SC-IAT error
rate larger than 20% were removed from the data (10 from the
positive condition and 15 from the negative condition), the overall
SC-IAT error rate no longer differed by condition (M ⫽ 9.42%),
2
This presentation effect on IAT scores was observed only when the
D-score algorithm was used to compute IAT scores; when the original
log-based IAT scoring algorithm was used, no presentation effect emerged,
t(81) ⫽ 0.10, p ⫽ .92, d ⫽ 0.02.
28 KARPINSKI AND STEINMAN
t(57) ⫽ 0.47, p ⫽ .64, d ⫽ 0.12. Furthermore, a t test revealed that
SC-IAT scores did not differ significantly as a function of condi-
tion, although a small presentation effect in the anticipated direc-
tion was observed (present positive, M ⫽ 0.60; present negative,
M ⫽ 0.50), t(57) ⫽ 0.92, p ⫽ .36, d ⫽ 0.24. These analyses
suggest that participants can fake a SC-IAT score, but to do so they
also increase their error rate.
An alternative interpretation of these results is that the fakers
(those who had large SC-IAT error rates) are different from
nonfakers, confounding the interpretation of the results. Although
we cannot entirely rule out this possibility, we think it is unlikely.
First, the fakers were not significantly different from the nonfakers
on any demographic variable we had collected (gender, ethnicity,
and native language; all ps ⬎ .25). Second, it is possible that the
nonfakers were insensitive to the presentation manipulation and/or
that the fakers tried harder than the nonfakers. However, the
nonfakers did show the predicted self-presentation effect on the
explicit attitude measures, t(57) ⫽ 7.73, p ⬍ .01, d ⫽ 2.05, and the
size of this self-presentation effect did not differ between fakers
and nonfakers, t(80) ⫽ 1.45, p ⫽ .15, d ⫽ 0.32. Furthermore, if the
nonfakers did not take the SC-IAT task seriously, then we would
expect their responses to be mostly error variance. On the contrary,
these participants displayed a strong bias in favor of women,
t(58) ⫽ 12.60, p ⬍ .01, d ⫽ 1.64, a result that is consistent with
other female SC-IAT findings (Karpinski & Lytle, 2005). Finally,
the fact that the nonfakers had low error rates also suggests that
they approached the task with sincerity.
A Closer Analysis of SC-IAT Error Rates
The high error rates observed in this study appear to be specif-
ically associated with the instruction manipulation. The SC-IAT
error rate was higher in Study 4 than in Studies 1–3, and other
studies have found normal error rates on the female SC-IAT when
no presentation instructions were provided (see Karpinski & Lytle,
2005).
We conducted several follow-up analyses of the SC-IAT error
rates to better understand why the instruction manipulation re-
sulted in such high error rates on the SC-IAT. First, we investi-
gated whether the increased error rate was due to an increase in
true errors (pressing the wrong response key) or due to an increase
in the failure to respond within the 1,500-ms response window. An
inspection of the error rates revealed that the overall error rate
(15.37%) was due to a high rate of incorrect responses (13.69%)
and a low rate of failure to respond within the response window
(1.68%).
Next, we computed overall error rates separately for each class
of target word (female vs. good vs. bad) within each block of the
SC-IAT (Block 1: female ⫹ good vs. Block 2: female ⫹ bad). A
Target Word ⫻ Block ⫻ Instruction Condition mixed ANOVA on
these error rates revealed main effects for each of the factors, but
these main effects were qualified by two significant higher order
effects: a Block ⫻ Condition interaction, F(1, 164) ⫽ 12.00, p ⬍
.01, and a Target Word ⫻ Block interaction, F(2, 164) ⫽ 6.70, p ⬍
.01 (see Figure 2).
The Block ⫻ Condition interaction indicates that participants
who were instructed to present positive attitudes toward women
had higher error rates when pairing female ⫹ bad than when
Figure 2. Single Category Implicit Association Test error rates by target,
stage, and instruction condition (N ⫽ 84).
pairing female ⫹ good (d ⫽ 0.92), whereas participants who were
instructed to present negative attitudes did not significantly differ
in their error rates across the blocks (d ⫽ 0.05). On average,
participants in both instruction conditions showed evidence of
more positive than negative associations with women, suggesting
that in the absence of faking, most participants have more positive
than negative associations with women (see also Karpinski &
Lytle, 2005). Thus, participants instructed to present positive atti-
tudes may not have had to adjust their responses for the female ⫹
good pairings as a result of the instructions— only when they
reached Block 2 (pairing female ⫹ bad) did they have to start
monitoring their responses, resulting in higher error rates. Con-
versely, participants instructed to present negative attitudes had to
adjust their responses for the female ⫹ good pairings in response
to the instructions. As a consequence, they may have adjusted their
responses from the start of the task, resulting in high error rates in
both blocks. Therefore, this Block ⫻ Condition interaction is
consistent with the claim that participants make more errors when
they are attempting to fake a SC-IAT score. Consistent with this
interpretation, when participants with high error rates were ex-
cluded from the analysis, the Block ⫻ Condition interaction
dropped to nonsignificance, F(1, 114) ⫽ 1.52, p ⫽ .22.
The Target Word ⫻ Block interaction indicates that partici-
pants’ error rates for the types of target words differed across
blocks. For Block 1 (female ⫹ good), participants had a lower
error rate for female target words than for good target words (d ⫽
0.44) or for bad target words (d ⫽ 0.61; and error rates for good
and bad target words did not significantly differ, d ⫽ .17). For
Block 2 (female ⫹ bad), participants had higher error rates for
good target words than for female or bad target words (ds ⱖ 0.43),
and error rates for female and bad words did not significantly
differ (d ⫽ 0.01). The interpretation of this interaction is unclear,
but it does not appear to be a result of the presentation instructions.
Supporting this conclusion, the Target Word ⫻ Block interaction
remained significant once individuals with high error rates were
excluded from the analysis, F(2, 114) ⫽ 7.87, p ⬍ .01.
Finally, we examined whether the D-score algorithm might
exaggerate the effect of a high error rate on SC-IAT scores. The
 SINGLE CATEGORY IMPLICIT ASSOCIATION TEST
D-score algorithm replaces errors with the block mean plus a
400-ms penalty. To determine whether this error penalty affected
the results, we computed an alternative SC-IAT score by using the
D-score algorithm, but we eliminated all error responses from the
computation. An analysis on these adjusted SC-IAT scores paral-
leled the previous analysis. When all participants were analyzed,
an instruction effect was observed, t(82) ⫽ 2.53, p ⫽ .01, d ⫽
0.55, and this effect disappeared when participants with high error
rates were excluded, t(57) ⫽ 0.54, p ⫽ .59, d ⫽ 0.14. Thus, the
observed effects are not dependent on the inclusion or exclusion of
incorrect response times in the scoring algorithm.
Discussion
The results of Study 4 suggest that participants can fake a
SC-IAT score, but to do so they also significantly increase their
error rate. Once participants with high error rates were excluded
from the analysis, only a small, nonsignificant presentation effect
remained. The bad news about these findings is that when partic-
ipants wish to fake a SC-IAT score, a significant number of them
spontaneously develop strategies that enable them to present the
desired attitude. Participants in this study were not provided with
strategies to fake the SC-IAT; they discovered strategies on their
own. The good news about this finding is that when participants
attempt to fake a SC-IAT score or to present a certain attitude on
a SC-IAT, they are likely to make many errors, and they can be
identified and excluded from subsequent analyses.
Somewhat surprisingly, participants were also able to fake IAT
scores in this study. Participants who were able to present a
particular attitude on the IAT did so without significantly increas-
ing their error rate, making it impossible to identify them and
remove them from the sample. We are somewhat skeptical of this
result, as no faking effect was found when IAT scores were
computed by using the older log-based scoring algorithm. Never-
theless, the concerns about the faking of SC-IAT scores appear to
be no worse than concerns about faking of IAT scores.
The results of Study 4 suggest that several cautions regarding
the use of the SC-IAT are in order. First, the ability of participants
to fake a SC-IAT score may decrease power (due to the discarding
of those with high error rates). Second, it may be the case that
participants who are able to fake a SC-IAT score are different from
those who were unable to fake a SC-IAT, resulting in a biased
sample, although there is no evidence for this claim in the current
study. Third, as previously highlighted, it may be easier for par-
ticipants to fake a SC-IAT score if category targets are pictures and
evaluative targets are words. For potentially sensitive domains, it
may be prudent to present category targets and evaluative targets
by using the same form of presentation (but there was no evidence
of high error rates or faking of SC-IAT scores in Study 3 when
Black and White faces were used as the target stimuli).
Finally, although we did not find presentation effects on the
SC-IAT once participants with high error rates were removed, it
does not mean that participants cannot devise other strategies to
fake a SC-IAT. Although participants did not spontaneously devise
effective strategies to fake a SC-IAT score without increasing their
error rate, if we had provided participants with a strategy, it is
likely that they could have carried out the strategy and we would
29
have found a presentation effect on the SC-IAT. Participants may
be more likely to spontaneously discover these alternative strate-
gies if they complete multiple SC-IATs (although in another study,
participants had to complete as many as four SC-IATs in a study
session and showed no evidence of faking or presentation effects;
see Karpinski & Lytle, 2005).
General Discussion
We developed the SC-IAT as a single-category measure of
social cognition to complement the IAT. Whereas the IAT mea-
sures comparative associations between two attitude objects, the
SC-IAT can assess the evaluative associations with a single atti-
tude object. Across three different attitude domains—soda brand
preferences, self-esteem, and racial attitudes—we found evidence
that the SC-IAT makes unique contributions in the ability to
measure and understand implicit social cognition. In the soda
brand domain, a Coke SC-IAT and a Pepsi SC-IAT predicted
behavioral intentions above and beyond what was predicted by the
IAT and explicit measures of soda brand attitudes. As a measure of
self-esteem, the SC-IAT demonstrated a significant medium- to
large-sized correlation with explicit measures of self-esteem. In
addition, the self-SC-IAT and the self– other IAT score were only
marginally correlated, supporting the claim that the SC-IAT’s
measure of self-associations is theoretically distinct from the
IAT’s measure of self– other associations. As a measure of racial
attitudes, similar mean-level effects were observed on the race IAT
and the race SC-IAT. However, the use of a White SC-IAT and a
Black SC-IAT allowed for a more detailed interpretation of the
implicit race bias.
We also extensively examined the reliability of the SC-IAT. In
general, the reliability of implicit measures of social cognition has
been relatively poor, with the exception of the IAT (Bosson et al.,
2000; Olson & Fazio, 2003). Across the four studies and six
different SC-IAT measures, the internal consistency of the SC-IAT
was reasonable (average r ⫽ .69; ranging from r ⫽ .55 to r ⫽ .85).
These reliability coefficients are similar to the internal consistency
observed for the IATs used in these studies (average r ⫽ .73;
ranging from r ⫽ .58 to r ⫽ .82) and similar to the internal
consistency observed for the IAT in previous research (Greenwald
et al., 2003; Nosek et al., 2005). Thus, the SC-IAT appears to have
a sufficient level of reliability to be of use as an individual
difference measure of implicit social cognition.
For the SC-IAT to be useful as a measure of implicit social
cognition, the SC-IAT must be relatively unsusceptible to faking
or self-presentational concerns. We found that when participants
attempted to self-present an attitude on the SC-IAT, many of them
had high error rates. Once participants with large error rates were
removed from the sample, there was only a small, nonsignificant
effect of self-presentation (d ⫽ .24). This result is consistent with
previous research on the IAT, suggesting that there may be a small
effect of faking or self-presentation on the IAT (Steffens, 2004).
Together, these findings provide strong evidence for the reliability
and validity of the SC-IAT as a measure of implicit social
cognition.
30 KARPINSKI AND STEINMAN
Interpreting the SC-IAT as a Measure of Evaluative
Associations
There has been some controversy surrounding the interpretation
of IAT scores (for overviews, see Arkes & Tetlock, 2004; Fazio &
Olson, 2003). Because the methodology of the SC-IAT is a mod-
ification of the IAT, it is possible that the various proposed
limitations and interpretations of the IAT also apply to the SC-
IAT. A common theme of these alternative interpretations is that
IAT effects may be due, at least in part, to factors other than
affective valence. For example, IAT effects may be observed
because of the cost of task switching (Mierke & Klauer, 2001), a
criterion shift across the different blocks of the task (Brendl,
Markman, & Messner, 2001), and the salience of the attitude
objects (Rothermund & Wentura, 2004). Additionally, the IAT has
been described as a measure of environmental associations, or
extrapersonal associations, rather than as a measure of one’s per-
sonal attitudes (Arkes & Tetlock, 2004; Karpinski & Hilton, 2001;
Olson & Fazio, 2004). Some of these alternative interpretations are
likely to apply to the SC-IAT as well. For example, the SC-IAT
may reveal more about one’s environmental associations than
one’s personal evaluative beliefs.
Additionally, we have been careful to describe the SC-IAT as a
measure of associations with a single attitude object and not as an
absolute measure of attitudes toward an attitude object. It is likely
that no attitude can be measured in absolute terms and that all
attitudes require some type of comparative judgment (Festinger,
1950). For example, to determine that one likes Pepsi, a person
must know how much he or she likes other beverages or how much
other people like Pepsi. Although a contrast category is not spec-
ified in the SC-IAT and the SC-IAT is less comparative than the
IAT (which explicitly requires a comparison category), the SC-
IAT also may not be an absolute measure of associations, in the
purest sense. Furthermore, although the SC-IAT avoids using a
contrast category for the target category of interest, it still mea-
sures the evaluation dimension comparatively (see Blanton et al.,
2006).
The presence of these possible alternative explanations for SC-
IAT effects and methodological issues regarding the SC-IAT do
not suggest that the SC-IAT should not be used as a measure of
associations, only that SC-IAT scores should be interpreted cau-
tiously and with these potential limitations in mind.
Recommendations for Using the SC-IAT
On the basis of the results of the studies reported here and other
studies conducted in our lab, we have several recommendations for
researchers interested in using the SC-IAT as a measure of implicit
social cognition. First, as previously mentioned, the SC-IAT is
conceptually identical to the ST-IAT and has only minor proce-
dural differences from the ST-IAT (Wigboldus et al., 2005). Com-
pared with the SC-IAT, the ST-IAT includes an initial practice
stage with only good and bad target words, has fewer target words
in each stage, and does not use a response window. Furthermore,
the internal consistency of the SC-IAT tends to be higher than the
internal consistency for the ST-IAT (Christian ST-IAT, adjusted
r ⫽ .39; Islamic ST-IAT, adjusted r ⫽ .68). Until studies are
conducted to investigate the effects of these procedural differences
on the reliability of single category or target measures, we recom-
mend following the SC-IAT procedure to keep the reliability of the
measure high.
Second, we recommend that the SC-IAT be used with at least 24
practice trials and 72 critical trials in each critical block. Unlike the
IAT, the SC-IAT does not have separate practice stages; within
each block, it is necessary to include practice trials and to exclude
those trials from the final calculation of SC-IAT scores. SC-IATs
with only 48 trials have been found to have lower internal consis-
tencies than SC-IATs with 72 trials. Likewise, the internal consis-
tency of the ST-IAT, with only 20 trials per block, has also been
found to be lower than the internal consistency of the SC-IAT
(Wigboldus et al., 2005). Thus, we suggest that 72 trials be the
lower bound for the number of trials that should be used for each
stage in a SC-IAT.
Third, we recommend that a 1,500-ms response window be
included in the SC-IAT procedure. The response window may
decrease the likelihood that participants engage in controlled pro-
cessing during the task. Longer response windows do not appear to
increase the reliability or to decrease the error rate, but it may be
possible to decrease the response window without adversely af-
fecting the error rate or the reliability of the measure.
Fourth, we recommend using the D-score algorithm to compute
SC-IAT scores. For the IAT, the D-score algorithm has been
shown to increase reliability, increase the correlations between the
IAT and explicit measures, reduce the correlation between the IAT
measure and speed of responding, and reduce the effect of proce-
dural variables (Greenwald et al., 2003). Although not presented
here, we also examined SC-IAT scores by using a log-based
scoring algorithm, modeled on the original IAT scoring algorithm
(Greenwald et al., 1998). We observed a very small trend for larger
effect sizes and larger SC-IAT-explicit correlations when the
D-score algorithm is used compared with the log-based algorithm.
We did not have the large samples required to detect small differ-
ences between the scoring algorithms by using tests of statistical
significance, but because of the similar methodology shared by the
SC-IAT and IAT, it is likely that many of the advantages of the
D-score algorithm will carry over to the SC-IAT.
Fifth, when the SC-IAT and the IAT are used in the same
experimental session, we recommend that the SC-IAT measures
precede any IAT measure. The IAT encourages a complementary,
dichotomous mind-set toward the categories of interest. When an
IAT is completed prior to a SC-IAT, the complementary mind-set
may carry over to the SC-IAT, possibly resulting in a task that is not
a measure of the associations with a single target category and in
lower reliability for the task. However, it is also possible that com-
pleting the SC-IAT may adversely affect the IAT, and future studies
are needed to examine the interplay between these two measures.
Finally, we have presented the SC-IAT as a measure of the
evaluative associations with a single target category. However,
like the IAT, the SC-IAT is a flexible measure that can be modified
to assess other aspects of implicit social cognition. For example,
the IAT has been used as a measure of gender identity by assessing
the strength of associations between the self and the male–female
dimension, relative to the strength of associations between others
and the male–female dimension (Aidman & Carroll, 2003; Green-
wald & Farnham, 2000; Rudman & Goodwin, 2004). The SC-IAT
can be easily modified to assess the strength of associations
 SINGLE CATEGORY IMPLICIT ASSOCIATION TEST
between the self and the male–female dimension, without also
assessing other associations. In this case, the gender dimension
(male–female) would still be assessed as a comparative dimension.
The SC-IAT can be used to eliminate one comparative dimension
but not both of the comparative dimensions. Likewise, the SC-IAT
can be modified to provide measures of implicit stereotypes and
other aspects of implicit social cognition (see Karpinski & Lytle,
2005). Additionally, for investigators who wish to reduce the
contaminating effects of environmental, or extrapersonal, associ-
ations (see Olson & Fazio, 2004), a personalized SC-IAT can be
constructed by replacing the category labels good and bad with I
like and I don’t like.
Conclusion
The IAT has been a breakthrough in the ability to measure and
understand implicit processes. The SC-IAT is a measure that can
be used alone or in combination with the IAT to further illuminate
these implicit processes. By continually developing and refining
individual difference measures of implicit social cognition, social
psychologists can improve their understanding of implicit and
explicit social cognition.
References
Aidman, E. V., & Carroll, S. M. (2003). Implicit individual differences:
Relationships between implicit self-esteem, gender identity, and gender
attitudes. European Journal of Personality, 17, 19 –36.
Arkes, H. R., & Tetlock, P. E. (2004). Attributions of implicit prejudice, or
“Would Jesse Jackson ‘Fail’ the Implicit Association Test?” Psycholog-
ical Inquiry, 15, 257–278.
Bargh, J. A., & Ferguson, M. L. (2000). Beyond behaviorism: On the
automaticity of higher mental processes. Psychological Bulletin, 126,
925–945.
Blanton, H., & Jaccard, J. (2006). Arbitrary metrics in psychology. Amer-
ican Psychologist, 61, 27– 41.
Blanton, H., Jaccard, J., Gonzales, P. M., & Christie, C. (2006). Decoding
the Implicit Association Test: Perspectives on criterion prediction. Jour-
nal of Experimental Social Psychology, 42, 192–212.
Bosson, J. K., Swann, W. B., & Pennebaker, J. W. (2000). Stalking the
perfect measure of implicit self-esteem: The blind men and the elephant
revisited? Journal of Personality and Social Psychology, 79, 631– 643.
Brendl, C. M., Markman, A. B., & Messner, C. (2001). How do indirect
measures of evaluation work? Evaluating the inference of prejudice in
the Implicit Association Test. Journal of Personality and Social Psy-
chology, 81, 760 –773.
Dasgupta, N., McGhee, D. E., Greenwald, A. G., & Banaji, M. R. (2000).
Automatic preference for White Americans: Eliminating the familiarity
explanation. Journal of Experimental Social Psychology, 33, 316 –328.
De Houwer, J. (2002). The Implicit Association Test as a tool for studying
dysfunctional associations in psychopathology: Strengths and limita-
tions. Journal of Behavior Therapy and Experimental Psychiatry, 33,
115–133.
De Houwer, J. (2003). The Extrinsic Affective Simon Task. Experimental
Psychology, 50, 77– 85.
Dovidio, J. F., Kawakami, K., Johnson, C., Johnson, B., & Howard, A.
(1997). On the nature of prejudice: Automatic and controlled processes.
Journal of Experimental Social Psychology, 33, 510 –540.
Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995).
Variability in automatic activation as an unobtrusive measure of racial
attitudes: A bona fide pipeline? Journal of Personality and Social
Psychology, 69, 1013–1027.
31
Fazio, R. H., & Olson, M. A. (2003). Implicit measures in social cognition
research: Their meaning and use. Annual Review of Psychology, 54,
297–327.
Festinger, L. (1950). A theory of social comparison. Human Relations, 7,
117–140.
Gehring, W. J., Karpinski, A., & Hilton, J. L. (2003). Cognition during
interracial interactions. Nature Neuroscience, 6, 1241–1243.
Greenwald, A. G. (1992). New Look 3: Unconscious cognition reclaimed.
American Psychologist, 47, 766 –779.
Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition:
Attitudes, self-esteem, and stereotypes. Psychological Review, 102,
4 –27.
Greenwald, A. G., & Farnham, S. D. (2000). Using the Implicit Associa-
tion Test to measure self-esteem and self-concept. Journal of Personality
and Social Psychology, 79, 1022–1038.
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring
individual differences in implicit cognition: The Implicit Association
Test. Journal of Personality and Social Psychology, 74, 1464 –1480.
Greenwald, A. G., & Nosek, B. A. (2001). Health of the Implicit Associ-
ation Test at age 3. Zeitschrift für Experimentelle Psychologie, 48,
85–93.
Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Understanding
and using the Implicit Association Test: I. An improved scoring algo-
rithm. Journal of Personality and Social Psychology, 85, 197–216.
Karpinski, A. (2004). Measuring self-esteem using the Implicit Association
Test: The role of the other. Personality and Social Psychology Bulletin,
30, 22–34.
Karpinski, A., & Hilton, J. L. (2001). Attitudes and the Implicit Associa-
tion Test. Journal of Personality and Social Psychology, 81, 774 –788.
Karpinski, A., & Lytle, J. (2005). Measuring implicit gender attitudes,
gender identity, and self-esteem using the Single Category Implicit
Association Test. Unpublished manuscript.
Kihlstrom, J. (1990). The psychological unconscious. In L. A. Pervin (Ed.),
Handbook of personality: Theory and research (pp. 445– 464). New
York: Guilford Press.
Maison, D., Greenwald, A. G., & Bruin, R. H. (2004). Predictive validity
of the Implicit Association Test in studies of brands, consumer attitudes,
and behavior. Journal of Consumer Psychology, 14, 405– 415.
Mierke, J., & Klauer, K. C. (2001). Implicit association measurement with
the IAT: Evidence for effects of executive control processes. Zeitschrift
für Experimentelle Psychologie, 48, 107–122.
Monteith, M. J., Voils, C. I., & Ashburn-Nardo, L. (2001). Taking a look
underground: Detecting, interpreting, and reacting to implicit racial
biases. Social Cognition, 19, 395– 417.
Nosek, B. A., & Banaji, M. R. (2001). The go/no-go association task.
Social Cognition, 19, 625– 664.
Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002). Harvesting
implicit group attitudes and beliefs from a demonstration Web site.
Group Dynamics: Theory, Research, and Practice, 6, 101–115.
Nosek, B. A., Greenwald, A. G., & Banaji, M. R. (2005). Understanding
and using the Implicit Association Test: II. Method variables and con-
struct validity. Personality and Social Psychology Bulletin, 31, 166 –
180.
Nunnally, J. (1978). Psychometric theory. New York: McGraw-Hill.
Olson, M. A., & Fazio, R. H. (2003). Relations between implicit measures
of prejudice: What are we measuring? Psychological Science, 14, 636 –
639.
Olson, M. A., & Fazio, R. H. (2004). Reducing the influence of extraper-
sonal associations on the Implicit Association Test: Personalizing the
IAT. Journal of Personality and Social Psychology, 86, 653– 657.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton,
NJ: Princeton University Press.
Rothermund, K., & Wentura, D. (2004). Underlying processes in the
32 KARPINSKI AND STEINMAN

Implicit Association Test (IAT): Dissociating salience from associa-
tions. Journal of Experimental Psychology: General, 133, 139 –165.
Rudman, L. A., & Goodwin, S. A. (2004). Gender differences in automatic
in-group bias: Why do women like women more than men like men?
Journal of Personality and Social Psychology, 87, 494 –509.
Steffens, M. C. (2004). Is the Implicit Association Test immune to faking?
Experimental Psychology, 51, 165–179.
Swim, J. K., Aikin, K. J., Hall, W. S., & Hunter, B. A. (1995). Sexism and
racism: Old-fashioned and modern prejudices. Journal of Personality
and Social Psychology, 68, 199 –214.
Teige, S., Schnable, K., Banse, R., & Asendorpf, J. B. (2004). Assess-
ment of multiple implicit self-concept dimensions using the Extrinsic
Affective Simon Task (EAST). European Journal of Personality, 18,
495–520.
Wentura, D., Kulfanek, M., & Greve, W. (2005). Masked affective
priming by name letters: Evidence for a correspondence of explicit
and implicit self-esteem. Journal of Experimental Social Psychology,
41, 657– 663.
Wigboldus, D. H. J., Holland, R. W., & van Knippenberg, A. (2005).
Single target implicit associations. Unpublished manuscript.
Wittenbrink, B., Judd, C. M., & Park, B. (1997). Evidence for racial
prejudice at the implicit level and its relationship with questionnaire
measures. Journal of Personality and Social Psychology, 72, 262–
274.

Appendix

Target Words Used in the SC-IAT and IAT

IAT target words SC-IAT target words

Pleasant Unpleasant Good Bad

brilliant awkward beautiful angry

diamond deatha celebrating brutal
joy hateb cheerful destroy
truth failureb excellent dirty
sunrise filtha excitement disaster
slumb fabulous disgusting
stink friendly dislike
uglya glad evil
glee gross
happy horrible
laughing humiliate
likable nasty
loving noxious
marvelous painful
pleasure revolting
smiling sickening
splendid terrible
superb tragic
paradise ugly
triumph unpleasant
wonderful yucky

Note. SC-IAT ⫽ Single Category Implicit Association Test; IAT ⫽

Implicit Association Test.
a
Unpleasant words used in IAT for Study 3 only. b Unpleasant words
used in IAT for Studies 1, 2, and 4 only.

Received May 20, 2005

Revision received August 18, 2005
Accepted August 19, 2005 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 171–187 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.171

Investigating the Dopaminergic Basis of Extraversion in Humans:

A Multilevel Approach
Jan Wacker, Mira-Lynn Chavanon, and Gerhard Stemmler
Philipps-Universität Marburg

A recent theory suggests that the agency facet of Extraversion (E) is based on brain dopamine (DA). The
paucity of human data relevant to this model is probably due to the lack of widely accessible noninvasive
psychophysiological indices and well-established behavioral measures sensitive to both E and manipu-
lations of DA activity. Aiming to identify such measures, the authors assessed the electroencephalogram
and n-back task performance in groups of introverts and extraverts after administration of either placebo
or a selective DA D2 receptor antagonist. As predicted, the antagonist’s effects on n-back reaction time
measures and frontal versus parietal electroencephalogram theta activity were strongly and specifically
modulated by E. New research avenues and theoretical extensions suggested by these results are
discussed.

Keywords: agentic Extraversion, dopamine D2 receptor antagonist, n-back task, electroencephalogram

theta activity, frontal electroencephalogram asymmetry

Thirty-five years ago, Jeffrey A. Gray (1970) proposed that
Extraversion (E) and other personality traits reflect individual
differences in motivational systems that evolved to deal with
certain classes of stimuli associated with positive and negative
reinforcement. Depue and Collins (1999) have integrated this
general idea with the extensive database generated by behavioral
neuroscience during the last decades into a novel psychobiological
theory of E. In brief, their model holds that E is based on a broad
motivational system that “facilitates” behavior motivated by pos-
itive incentives and that is neurobiologically based on the meso-
corticolimbic dopamine (DA) system.
Although Depue and Collins (1999) firmly rooted their theory of
E in basic neuroscience insights gleaned from animal research,
they have thus far not been particularly interested in using their
model to derive new hypotheses and viable empirical approaches
for personality research with humans. As also noted by others, “the
model in its present form specifies, at the human level, little more
than an expected pattern of correlations between trait measures and
indices of dopaminergic neurotransmission” (commentary by
Pickering in Depue & Collins, 1999, p. 534). However, currently,
such indices of dopaminergic neurotransmission either require
invasive measurements of blood hormone levels (e.g., Depue,
Jan Wacker, Mira-Lynn Chavanon, and Gerhard Stemmler, Faculty of
Psychology, Philipps-Universität Marburg, Germany.
This article is based on the doctoral dissertation of Jan Wacker under the
supervision of Gerhard Stemmler. The reported research was conducted
with the help of Deutsche Forschungsgemeinschaft Grant Ste 405/9-1. We
thank Thomas Scherer and Erwin Hennighausen for hardware and software
support and Christof Kemper for his assistance in conducting the
experiment.
Correspondence concerning this article should be addressed to Jan
Wacker, Faculty of Psychology, Philipps-Universität Marburg, Gutenberg-
strasse 18, D-35032, Marburg, Germany. E-mail: wackerj@staff.uni-
marburg.de
Luciana, Arbisi, Collins, & Leon, 1994) or extremely expensive
neuroimaging techniques (e.g., Farde, Gustavsson, & Jonsson,
1997). These methodological requirements considerably impede
empirical evaluation of the model because the necessary laboratory
equipment is typically either not available to personality psychol-
ogists or too expensive to be used with the large sample sizes
necessary for the investigation of individual differences. There-
fore, it is not surprising that the paucity of relevant data noted by
Depue and Collins (1999) 7 years ago is still largely uncompen-
sated. Noninvasive and more accessible psychophysiological indi-
ces of central DA functioning are needed for progress in this
exciting field of research. Furthermore, even if the hypothesized
link between E and brain DA had already been firmly established
empirically, the important question of how individual differences
in DA functioning are translated into extraverted versus intro-
verted patterns of behavior would remain largely unanswered. To
begin to address this problem at the human level, behavioral
paradigms that are sensitive to both manipulations of functional
brain DA activity and individual differences in E need to be
established.
The aims of the present study were: (a) to introduce an easily
obtainable electroencephalogram (EEG) index that can be used to
measure those individual differences in functional brain DA ac-
tivity that are relevant to E and (b) to identify a behavioral
paradigm that is sensitive to both manipulations of functional brain
DA activity and individual differences in E. These psychophysio-
logical and behavioral measures could then be used as a basis to
further extend the theory proposed by Depue and Collins (1999) to
the human level.
E and DA
Although in several influential psychobiological theories (e.g.,
Cloninger, Svrakic, & Przybeck, 1993; Pickering & Gray, 1999;
Zuckerman, 1991) the dopaminergic system is regarded as (part of)
the basis of one of several highly correlated traits from the

171
172 WACKER, CHAVANON, AND STEMMLER
impulsivity-sensation seeking spectrum (novelty seeking, impul-
sivity, impulsive antisocial sensation seeking, or impulsive sensa-
tion seeking), based on psychometric analyses and theoretical
considerations Depue and Collins (1999) argue that impulsivity-
sensation seeking is rather heterogeneous in terms of both the
underlying motivations and brain systems and that therefore con-
sistent and replicable associations with markers of DA neurotrans-
mission are not to be expected. Depue and Collins (1999) further
note that E can be subdivided into sociability on the one hand and
agency on the other. It is the agency facet of E (i.e., a motivational
disposition that comprises social dominance, enthusiasm, energy,
assertiveness, ambitiousness, and achievement striving), which
they regard as the central characteristic of E and for which they
propose a dopaminergic basis in their theory.
Depue and Collins (1999) suggest that agentic E is based on a
broad motivational system, which they term the behavioral facil-
itation system (BFS). The BFS modulates or “facilitates” behavior
motivated by positive incentives through increasing their motiva-
tional salience. Agentic extraverts possess a more reactive BFS
and therefore are generally more enduring, motivated, and vigor-
ous in their pursuits of positive incentive goals. In the neurobio-
logical part of their theory, Depue and Collins (1999) propose that
the mesocorticolimbic DA system (i.e., the dopaminergic projec-
tions of the ventral tegmental area in the midbrain to the nucleus
accumbens, medial orbital prefrontal cortex (PFC), and other cor-
tical and subcortical structures) form the basis of the BFS and that
individual differences in functional properties of this neurobiolog-
ical system are the basis of agentic E. To be more specific, they
argue in their “psychobiological threshold model” that positive
incentive stimuli induce a certain state level of postsynaptic DA
receptor activation in terminal structures of the mesocorticolimbic
DA system. This state level adds to the tonic trait level of DA
activation, which is higher in individuals high versus low in
agentic E. A certain level of trait plus state DA activity is neces-
sary to induce behavioral facilitation. As a consequence, agentic
extraverts reach this critical threshold more easily. Therefore, for
these individuals, relatively weak incentive stimuli are sufficient to
induce behavioral facilitation; in accordance, they experience and
display the affective and behavioral concomitants of positive in-
centive motivation more frequently.
To date, the strongest empirical support for Depue and Collins’
(1999) central theoretical proposals comes from studies in which a
selective DA receptor agonist was administered while blood sam-
ples were taken to measure plasma levels of the hormone prolactin.
It is known that the secretion of prolactin is modulated by central
DA activity. Thus, changes in prolactin blood levels in response to
challenges with dopaminergic drugs can be used as a peripheral
indicator of central DA activity. In their landmark study, Depue et
al. (1994) reported strong and specific correlations between agen-
tic E as measured by Tellegen’s Multidimensional Personality
Questionnaire (MPQ; scale Positive Emotionality; Tellegen &
Waller, in press) and several characteristics of the DA agonist-
induced inhibition of prolactin secretion in a sample of N ⫽ 11
female volunteers. These findings were replicated using the same
methods in a larger sample, and no significant correlations
emerged for the MPQ scales Negative Emotionality and Constraint
(Depue, 1995, 1996). Although these observations strongly sup-
port the theory, both a replication by other research groups and a
direct statistical comparison of the validities of agentic E and
impulsivity sensation seeking with respect to DA indicators remain
to be conducted.
The development of dopaminergic tracers that can be used in
neuroimaging studies to more directly measure certain aspects of
central DA activity recently has opened up new possibilities to
investigate the hypothesized association between personality traits
and DA. However, the few studies that have already employed this
novel technique do not yet provide a conclusive picture (Breier et
al., 1998; Farde et al., 1997; Gray, Pickering, & Gray, 1994;
Kestler, Malhotra, Finch, Adler, & Breier, 2000; Laakso et al.,
2000, 2003).
In summarizing the evidence, it can still be concluded with
Pickering and Gray (2001, p. 124) that “it seems premature to
conclude that measures of E are more strongly and/or more
consistently associated with DA functioning than measures of
ImpASS [impulsive antisocial sensation seeking].” We would add
to this conclusion that fast and significant progress on this question
is unlikely as long as the established approaches to measurement
of central DA activity in humans [i.e., invasive measurement of
blood hormone levels and positron emission tomography (PET)
scans with dopaminergic tracers that cost several thousand dollars
per participant] are largely inaccessible to personality psycholo-
gists. Eysenck’s (1967) famous arousal theory of E inspired more
than 1,000 empirical studies and stimulated some of the most
creative research in personality psychology for the last 4 decades.
It is conceivable that this enormous success of Eysenck’s (1967)
theoretical proposals depended significantly on easily accessible
noninvasive psychophysiological methods (e.g., EEG indices) to
measure the central construct of the theory (i.e., general cortical
arousal) in humans. Thus, easily obtainable noninvasive psycho-
physiological indices of individual differences in DA functioning
that are relevant to E could prove very useful for progress in this
field of research. In the following section, we selectively review
the current literature in search of such indices in the EEG.
E, DA, and Frontal EEG Activity
The EEG signal measured on the scalp is generated largely in
cortical areas (Davidson, Jackson, & Larson, 2000). Therefore, the
EEG cannot be used to directly measure activity in those subcor-
tical elements of the mesocorticolimbic system that constitute the
neurobiological core of the BFS. However, the PFC receives
strong dopaminergic inputs from mesolimbic structures, and it is
well known that DA plays an important modulatory role in PFC
functioning (e.g., Seamans & Yang, 2004). As a consequence,
EEG indices of PFC activity are a good place to start searching for
noninvasive psychophysiological measures of central DA
functioning.
Asymmetrical Frontal EEG Activity
During the past 25 years, research on the relationship between
frontal cortical asymmetry and emotion, motivation, and person-
ality has attracted considerable attention (for recent review, see
Coan & Allen, 2004a). Besides many other interesting associa-
tions, significant correlations between relative left frontal cortical
activity, inferred from greater right than left frontal EEG activity
in the alpha band (8 –13 Hz), and several personality traits from the
E spectrum have been reported repeatedly (e.g., Coan & Allen,
 EXTRAVERSION AND DOPAMINE
2003; Harmon-Jones & Allen, 1997; Schmidt, 1999; Sutton &
Davidson, 1997). Findings like these typically have been inter-
preted within the theoretical framework proposed by Davidson and
colleagues (e.g., Davidson, 1998), according to which the left and
the right anterior region of the brain are part of two separate neural
systems underlying approach and withdrawal motivation, respec-
tively. The approach system is activated by the perception of goals
and is thought to underlie initiation of goal-directed approach
behavior toward these goals as well as elicitation of approach-
related motivational and emotional states (i.e., desire, wanting,
enthusiasm, pregoal attainment positive affect, but also anger, see
Harmon-Jones, 2004). On the neuroanatomical level, the approach
system encompasses the left dorsolateral and medial prefrontal
cortices and the dopaminergic circuitry of the basal ganglia.
Several investigators (see, e.g., the commentary by Kline in
Depue & Collins, 1999; Harmon-Jones & Allen, 1997; Sutton &
Davidson, 1997; Wacker, Heldmann, & Stemmler, 2003) have
noted a considerable conceptual overlap between Davidson’s
(1998) approach system and Depue and Collins’ (1999) BFS or
Gray’s behavioral activation system (BAS; e.g., Pickering & Gray,
1999). However, it is still a matter of debate whether and how
Davidson’s description of a left frontal approach system can be
integrated with the latter two models that focus on dopaminergic
neurotransmission in largely subcortical brain structures and do
not make any assumptions about asymmetric cortical activation
patterns (Coan & Allen, 2004b; Hewig, Hagemann, Seifert, Nau-
mann, & Bartussek, 2004; Wacker et al.). An empirical demon-
stration that frontal cortical asymmetry is sensitive to changes in
brain DA functioning (or vice versa) would, thus, constitute an
important step toward conceptual integration. In summary, frontal
EEG asymmetry in the alpha band has not only been associated
with traits from the E spectrum in several empirical studies, but
there are also good theoretical reasons to expect that this psycho-
physiological index is related to certain aspects of central DA
activity.
Symmetrical Frontal EEG Activity
Several researchers recently have emphasized the importance of
bilateral (or symmetrical) resting frontal EEG activity in the con-
text of personality traits (Davidson, 2004; Hewig et al., 2004;
Hewig, Hagemann, Seifert, Naumann, & Bartussek, 2006) and
affective disorders (Nitschke, Heller, Etienne, & Miller, 2004). For
example, Hewig et al. proposed that low resting bilateral frontal
EEG activity in the alpha band (i.e., presumably high frontal
cortical activity) is associated with high habitual activity in Gray’s
BAS. The BAS is a DA-based motivational system closely related
to Depue and Collins’ (1999) BFS, which responds to stimuli of
reward and nonpunishment, elicits positive emotions, and moti-
vates goal-directed behavior (e.g., Pickering & Gray, 1999).
Hewig et al. (2004, 2006) repeatedly observed significant negative
correlations between scores in Carver and White’s (1994) BAS
scale and resting EEG alpha power at frontal scalp sites. Further-
more, several recent functional magnetic resonance imaging stud-
ies have found substantial correlations between traits from the E
spectrum and either increased or reduced activity in the anterior
cingulate cortex (ACC) during a number of cognitive and emo-
tional tasks (Canli, Amin, Haas, Omura, & Constable, 2004; Canli
et al., 2001; Gray & Braver, 2002; Kumari, ffytche, Williams, &
173
Gray, 2004). The ACC is an area of the PFC that receives strong
dopaminergic inputs (Holroyd & Coles, 2002; Seamans & Yang,
2004) and whose activity is influenced by genes controlling PFC
DA neurotransmission (Blasi et al., 2005). It is interesting that
there is now substantial evidence (e.g., Pizzagalli, Oakes, & Da-
vidson, 2003) that activity in this region is reflected in frontal
midline EEG activity in the theta band (4 – 8 Hz). Taken together,
the recent literature suggests that both increased activity in the
frontal cortex (as measured by reduced EEG alpha activity at
frontal sites) and either increased or reduced activity in the ACC
(as measured by increased or reduced frontal midline EEG theta
activity) are not only related to E and/or related traits but are also
modulated by brain DA.
To measure individual differences in resting frontal EEG activ-
ity, two issues need to be considered. First, irrelevant individual
differences in skull thickness affect the amplitude of the EEG
signal measured on the scalp. Second, due to volume conduction,
electrical activity generated in other areas of the cortex is also
measurable at the sites of interest and vice versa. In prior research,
two approaches have been used to control for these effects. The
first approach employs a regression procedure that corrects the
EEG power at the target site in the frequency band of interest for
global EEG power in that frequency band computed as the average
power across all electrodes. However, to use this method, the EEG
needs to be recorded from a relatively large number of electrodes
to derive valid estimates of global EEG power (a minimum of 20
electrodes is recommended by Davidson et al., 2000). The second
approach is to simply compute difference values between EEG
power measured at the site of interest and EEG power measured at
a separate site, thought to be influenced by different brain pro-
cesses. This approach has been employed with great success in
research on frontal EEG asymmetry by calculating an asymmetry
index as the difference between homologous right and left elec-
trodes. Although this method only yields measures of relative
activity at the two sites used to compute the difference values, it
bears the advantage that as few as two active electrodes are
sufficient to obtain the EEG index of interest. Application of this
approach, however, requires that the second electrode site is not
chosen randomly but based on relevant theoretical and empirical
information. Difference values should only be computed when the
relative activity at the two sites arguably captures a (theoretical)
meaningful aspect of EEG activity (e.g., EEG lateralization in
research on frontal asymmetry).
It is interesting that Hewig et al. (2004, 2006) did not only
repeatedly observe negative correlations between BAS sensitivity
and EEG alpha power at frontal scalp sites, as already noted above,
but also an (unexpected) positive correlation of similar magnitude
between BAS sensitivity and EEG alpha power at parietal and
parietooccipital sites. As a consequence, in these two studies, a
difference value computed as the difference between EEG alpha
activity at frontal sites minus activity at parietal sites would have
been maximally suited to capture the aspect of brain electrical
activity related to habitual BAS activation. Also, it has long been
known that two different manifestations of EEG theta rhythm can
be distinguished (Pizzagalli et al., 2003; Schacter, 1977). In addi-
tion to the already described frontal midline variant, which has
been linked to alert states characterized by focused attention and
mental effort, a second variant exists, which exhibits a more
posterior and widespread distribution and has been associated with
174 WACKER, CHAVANON, AND STEMMLER
decreased prestimulus alertness and less focused, more freely
floating attention (e.g., during hypnagogic states). Because the
mental processes associated with the two types of theta rhythm are
at least in part mutually exclusive, it would seem likely that the
activity in brain sources of frontal midline and posterior theta are
also reciprocally related. If this were in fact the case, the difference
between frontal midline theta activity and theta activity at posterior
(e.g., parietal) sites would be well suited to capture a meaningful
aspect of EEG theta activity in a single measure.
In summary, it can be concluded that based on prior findings,
symmetrical frontal EEG activity both in the alpha and theta bands
can be hypothesized to be associated with both E-related traits and
brain DA functioning and that difference values of EEG band
power at frontal sites minus band power at parietal sites are
well-suited to capture these aspects of brain electrical activity in
two simple measures.
E, DA, and Behavior: The N-Back Task
There is ample evidence for an involvement of DA in working
memory (WM) functions from a variety of different methodolog-
ical approaches (for recent reviews, see, e.g., Barch, 2004; Sea-
mans & Yang, 2004). It is interesting that researchers have repeat-
edly reported that DA agonists (e.g., amphetamine or the D2
receptor agonist bromocriptine) improve performance in WM
tasks (and other tasks known to strongly recruit the PFC, such as
the Wisconsin Card Sorting Task) only in individuals with poor
baseline performance but impair performance in individuals with
good baseline performance (e.g., Kimberg, D’Esposito, & Farah,
1997; Mattay et al., 2000, 2003). Using the well-known n-back
WM task with increasing levels of task difficulty (i.e., WM loads
of 1-, 2-, and 3-back), Mattay et al. (2000, 2003) recently demon-
strated that there are genetically determined individual differences
in DA neurotransmission that are associated with both perfor-
mance in the most difficult version of the n-back task (3-back)
under normal conditions and changes in performance and PFC
activity induced by dopaminergic drugs.
However, are these individual differences in DA neurotransmis-
sion also associated with E? Several recent observations indicate
that this may indeed be the case. It has been demonstrated that both
E and BAS sensitivity predict the magnitude of increases in brain
activity in dorsolateral PFC and the ACC produced by the difficult
3-back task (Gray & Braver, 2002; Kumari et al., 2004). In
addition, Lieberman and Rosenthal (2001) have shown that reac-
tion times (RTs) in the difficult (2- and 3-back) but not in the easy
(0- and 1-back) versions of the n-back task are negatively corre-
lated with E, that is, extraverts reacted faster than introverts
specifically in those versions of the task that impose strong de-
mands on WM. However, until now, the negative correlation
between E and n-back RTs has not been successfully replicated,
and it is unknown whether it is indeed (partly) based on individual
differences in DA functioning as proposed by Lieberman and
Rosenthal (2001). If a dopaminergic basis could be demonstrated
for this E effect, this would not only support Depue and Collins’
(1999) claim of a link between DA and E but also suggest that the
n-back task could be a useful behavioral paradigm to investigate
the question of how individual differences in DA functioning are
translated into extraverted versus introverted patterns of behavior.
The Present Study
Based on prior findings, agentic E is hypothesized to be asso-
ciated with: (a) asymmetrical frontal EEG activity and in the alpha
band and symmetrical frontal EEG activity in the alpha and theta
bands and (b) performance in the n-back task. One aim of the
present study was to demonstrate these associations empirically. It
is more important that the present work aimed at examining
whether these associations can be attributed to a common dopa-
minergic basis. To this end, we measured the EEG and n-back
performance in extreme groups of participants either high or low in
agentic E after administration of either placebo or a selective
dopaminergic drug. Because dopaminergic drugs have been shown
to invert preexisting group differences in measures of brain activ-
ity and task performance (e.g., Corr & Kumari, 1997; Kimberg et
al., 1997; Mattay et al., 2000, 2003; Takeshita & Ogura, 1994), we
predicted that the differences between introverts and extraverts1
expected for the placebo condition would be inverted in the drug
condition, resulting in a significant interaction between E (intro-
verted vs. extraverted) and pharmacological substance (placebo vs.
dopaminergic drug). If such a significant interaction could be
demonstrated for our behavioral and/or EEG parameters, this
would not only add to the still very limited database currently
supporting the suggested link between agentic E and DA (Depue &
Collins, 1999) but also open up new avenues for experimental
personality research on the human level by providing more acces-
sible indicators of central DA functioning.
Method
Participants
A sample of 46 participants who were either high or low in agentic E
was selected from a pool of N ⫽ 507 male student volunteers (see below).
We only used male participants because in females, the possibility of an
undetected pregnancy constitutes an additional risk in pharmacological
studies and because the menstrual cycle is known to introduce error
variance into physiological recordings. Further inclusion criteria were
German native language, age 18 to 40 years, right-handedness, and no
history of mental disorders as assessed with a standardized clinical inter-
view (Margraf, 1997). Exclusion criteria were treatment with prescription
drugs during the last 3 months, acute diseases, history of abnormal con-
ditions of the liver or the kidney, habitual smoking of more than 10
cigarettes per day, or use of illegal drugs. Six participants were excluded
because of strong artifacts in the EEG recordings (excessive blinking
and/or muscle artifacts). The average age of the N ⫽ 40 participants in the
final sample was M ⫽ 24.4 years (range, 18 –34 years), and all were
university or high school students. Participants were told that they were
selected based on self-ratings of certain aspects of their personality and that
the aim of the study was to further delineate the brain processes and
neurotransmitters underlying a RT task and to find out how these relate to
personality. They were paid 45 EUR (55 USD) for approximately 5 hours
of involvement in the study.
1
To facilitate reading, we will occasionally use the terms E or extra-
verted/introverted instead of agentic E and high/low in agentic E when
referring to the extreme groups of participants examined in the present
study. However, we do not wish to imply that agentic E and E are one and
the same: In addition to the agency facet, the super trait E also includes
sociability and, in some conceptions, impulsivity (see Depue & Collins,
1999).
 EXTRAVERSION AND DOPAMINE
Experimental Design and Extreme Group Selection
Participants were assigned to one of four experimental groups formed by
the combination of E (extraverted, introverted) and Substance (placebo, D2
antagonist).2 For the analyses of n-back performance and posttask waiting
period EEG, this 2 ⫻ 2 design was supplemented by the repeated factor
Task (0- to 3-back task with the possibility to obtain reward and 2-back
task without reward; see below) with the effect of presentation order
controlled via cyclic permutation in a Latin Square.
Depue et al. (1994) recommended the MPQ scale Positive Emotionality
(Tellegen & Waller, in press) for measuring agentic E. Therefore, we
developed a German short version of this scale to use it as an economical
screening instrument for extreme group selection. This brief questionnaire
contains three positively correlated 10-item scales that correspond to the
three MPQ primary scales most relevant to Positive Emotionality: Well-
being, Achievement, and Social Potency (see Tellegen & Waller, in press).
In prior unpublished work from our laboratory, the three newly developed
scales correlated highly with the respective MPQ primary scales (r ⬎ .85,
N ⫽ 140). The sum of the three scales constituted our aggregate measure
of agentic E. A total of N ⫽ 507 male study volunteers were contacted on
several locations across the university campus and filled out the newly
developed 30-item personality questionnaire. Internal reliabilities calcu-
lated in this pool of volunteers were satisfactory for both the three 10-item
primary scales (Cronbach’s ␣ ⱖ .81) and the 30-item aggregate scale
(Cronbach’s ␣ ⫽ .87). Participants scoring above the median in all three
primary scales constituted the extraverted group, whereas participants
scoring not higher than the median on any of the three scales constituted
the introverted group. The 40 participants in the sample of the present study
scored either above the 67th percentile (20 participants in the extraverted
extreme group) or below the 33rd percentile (20 participants in the intro-
verted extreme group) on the aggregate scale of agentic E. Ten participants
of each extreme group were randomly assigned to either the placebo or the
D2 antagonist group.
Sulpiride: A Selective DA D2 Receptor Antagonist
We used the selective D2 antagonist sulpiride (200 mg) to challenge the
DA system. The substituted benzamide sulpiride exerts its antagonistic
effects predominantly at D2 receptor sites to which it binds with high
affinity (Missale, Nash, Robinson, Jaber, & Caron, 1998). D1 receptors,
GABA receptors, and adrenergic, cholinergic, histaminergic, and seroto-
nergic receptors are largely unaffected by sulpiride (Caley & Weber, 1995;
Mauri, Bravin, Bitetto, Rudelli, & Invernizzi, 1996). The drug accumulates
more strongly in the mesolimbic than in the nigrostriatal DA system, and
it completely blocks stimulatory effects induced by the selective D2
agonist bromocriptine (Jackson & Hashizume, 1987). Sulpiride reaches its
maximum plasma concentration 1 to 6 h after oral administration (Caley &
Weber, 1995; Mauri et al., 1996). A single dose of sulpiride is well
tolerated by healthy volunteers, who are typically not able to tell whether
they received placebo or sulpiride (e.g., McClelland, Cooper, & Pilgrim,
1990; Meyer-Lindenberg, Rammsayer, Ulferts, & Gallhofer, 1997).
Setting and Apparatus
The experimental room (4 ⫻ 3.4 m) was sound-attenuated and air-
conditioned and had a largely nontechnical appearance. Participants sat
comfortably in a reclined position. Electrodes were connected to a cus-
tomized headbox (NeuroScan, Sterling, VA), where signals were pream-
plified with a gain of 30 (input impedance, 10 M⍀). N-back stimuli were
presented on a 15-inch flat screen display placed in front of the participant
(distance to the eyes, approximately 60 –70 cm). Participants reacted by
pressing buttons on a response box (XQMS, Frankfurt, Germany). Prere-
corded instructions and verbal task feedback were presented via two
loudspeakers located on the walls of the experimental room.
175
In an adjacent room were placed a 32-channel SynAmps 5083 amplifier
(NeuroScan), a Macintosh Power Mac G4/450 (Apple Computer, Cuper-
tino, CA) with a PCI 6503 SCSI card (National Instruments, Austin, TX)
that performed recording and storage of the digitized EEG data under
Labview 5.0 (National Instruments), and a personal computer that per-
formed experimental control under Presentation 0.5 (Neurobehavioral Sys-
tems, Albany, CA).
Procedure
Informed consent was obtained at the beginning of the session. The
experimenter then conducted a standardized clinical interview (Margraf,
1997) to check for lifetime absence of Diagnostic and Statistical Manual
of Mental Disorders (American Psychiatric Association, 1994) Axis I
psychiatric disorders. If the participant met this final inclusion criterion, the
experimenter administered identical capsules of either sulpiride (200 mg)
or placebo under double-blind conditions at 9:30 a.m. ⫾ 75 minutes.
Participants fasted and abstained from nicotine and caffeine from 10 p.m.
of the previous night until completion of the experiment with the exception
of a light breakfast (two rolls with butter, cheese, and/or jam) and an
8-ounce glass of 1.5% milk taken at the time of medication (or placebo) to
prevent nausea. Directly after completion of the breakfast, a brief (20 –25
minute) intelligence test was conducted; then, participants completed sev-
eral personality questionnaires (see below).
The experimenter, aided by an assistant, then positioned electrodes and
transducers and explained the n-back task. Afterward, 40 practice trials of
the 2-back task were presented. Before leaving the room, the experimenter
reminded the participant to sit quietly to help prevent artifacts in the
physiological recordings and also told him that further instructions and task
feedback were prerecorded and would be presented to him over the
loudspeakers. On average, the experiment was started 186 minutes (range,
150 –205 minutes) after administration of the medication (or placebo)
capsule.
First, participants were instructed to relax during a 10-min rest period
(with three embedded 1-min EEG recording periods) while keeping their
eyes open. Next, the n-back tasks were presented in five consecutive blocks
with the effect of block position controlled via cyclic permutation in a
Latin Square. Each block was followed by a 1-min period, during which
participants were instructed to simply wait for a moment, before either the
next block would begin (2-back task without possibility to obtain reward)
or the experimenter would briefly enter to give performance feedback and,
if the participant passed the criterion, to hand out the promised reward (0-,
1-, 2-, and 3-back task with the possibility to obtain reward). During each
of the five 1-min posttask (prereward) waiting periods, the EEG was
recorded. After completion of the final task block, the participant filled out
a task questionnaire presented on his flat screen display and answered a
brief postexperimental interview. The experiment ended approximately 4 h
and 20 min after administration of the medication (or placebo) capsule.
N-Back Task and Incentive Motivational Context
The version of the n-back task employed was similar to the one used by
Braver et al. (1997). Single white letters (Times New Roman, 60 point)
2
Initially, the factor Substance included a third group of participants
who received a low dose (1.25 mg) of the selective D2 agonist bromocrip-
tine. However, unexpectedly, several participants in this drug condition
experienced severe side effects (nausea, dizziness). Because blindness to
the experimental condition thus could not be secured for the bromocriptine
group, rendering the obtained results difficult to interpret, we decided not
to report the bromocriptine data. The few statistically reliable effects that
were observed for bromocriptine despite the error variance that was prob-
ably introduced by the aversive side effects were similar to those reported
for sulpiride, although somewhat smaller in magnitude.
176 WACKER, CHAVANON, AND STEMMLER

were presented successively on a black computer screen. For each of the 1978). As a consequence, in studies like the present one, it is mandatory to
four levels of difficulty (0- to 3-back), there was a different decision consider the possibility that potential E effects are based on a significant
criterion for making the target response of either yes or no (see Braver et speed–accuracy trade-off. To simplify the presentation of the results and to
al., 1997) with the index or middle fingers of the right hand, respectively. ensure that measures of speed and accuracy were based on the same trials,
For each version of the task, a sequence of 60 practice trials (20 targets) we decided to quantify accuracy as the percentage of correct target reac-
followed by 120 evaluated trials (40 targets) was presented. The last 50 tions (see also Gray et al.). An alternative analysis using signal detection
practice trials of each task version were used to compute the 90th percentile measures (d⬘ and C) did not yield any additional information. (c) Individual
of the individual RT distribution for correct reactions (only RTs that were distributions of RTs are typically skewed leading to a positive correlation
not slower than the individual RT mean plus 3 times the individual RT SD between average RTs and individual SD values of RTs: Individuals whose
were considered). This served as an individual RT criterion for the fol- RTs vary more strongly on a trial to trial basis (e.g., between targets and
lowing evaluated task sequence. Each letter was presented for 500 ms nontargets) also tend to have longer average RTs. Therefore, it is possible
followed by a 1,650-ms interval, during which the screen was black. The that the RT differences between introverts and extraverts observed by
participant was expected to react during this 2,150-ms interval. Each trial Lieberman and Rosenthal (2001) would have been even more pronounced
ended with the presentation of a verbal feedback (350 ms) that indicated had these authors used measures of RT variability instead of RT means.
whether the participant’s preceding reaction was “correct”, “wrong” (i.e., To prevent distortion of the results by invalid reactions, all performance
either a wrong reaction or no reaction), or “slow” (i.e., correct, but slower measures were based only on trials in which participants pressed either the
than the individual RT criterion). Directly after the feedback had ended, the yes button or the no button in a 1,950-ms time window starting 200 ms after
next letter was presented. the beginning of the presentation of the letter and ending with the begin-
As explained in more detail elsewhere (Pauls, Wacker, & Crost, 2005; ning of the verbal feedback. The mean percentage of valid trials was at
Stemmler, 1997), we conceptualize E and other personality traits as dis- least 96.7% in all possible combinations of tasks and experimental groups.
positions that are activated only in relevant situational contexts (for dis- For the calculation of all statistical tests, the two RT measures were square
cussion of related theoretical perspectives, see Mischel & Shoda, 1998). root transformed to normalize distributions. Reliability estimates for the
Even though the current study was not designed to provide evidence for three performance parameters are shown in Table 1.
this traits-as-dispositions approach, this general concept nonetheless sig-
nificantly influenced certain aspects of the present experimental setup. EEG Recording and Analysis
When traits are only activated in relevant situational contexts, their pre-
dictive power concerning behavior and the associated brain states will also In keeping with the 10-20 International System (Jasper, 1958) and using
be limited to such relevant contexts. For agentic E, as defined by Depue Easy Cap electrode caps (Falk Minow Services, Herrsching-Breitbrunn,
and Collins (1999), relevant contexts are characterized by the presence of Germany), recordings were made from midfrontal (F3/F4), frontal midline
positive incentives. As a consequence, we took special precautions to (Fz), and parietal midline (Pz) regions of the scalp. EEG data were also
obtained from central (C3/C4/Cz), anterior temporal (T3/T4), and parietal
create an experimental setting that is strongly characterized by positive
(P3/P4) sites, but these recordings are not analyzed here. All sites were
incentives. Right at the beginning of the laboratory session, the participant
referenced to resistor-linked mastoids (M1–M2; 5-K⍀ resistors). To record
was told that in each version of the task, he had the chance to obtain a
eye blinks and vertical eye movements, electrodes were placed midline
reward for good performance. The potential rewards consisted of a variety
above and below the right eye. Electrodes on the outer canthi of both eyes
of popular goodies each worth about 1 EUR (1.3 USD). Goodies were
were applied to record horizontal eye movements. Electrode impedances
displayed throughout the whole experimental session in a large cardboard
were kept under 5 K⍀ for the EEG electrodes and under 1 K⍀ for the
box placed across the room from the sitting participant. After completion
ground electrode by cleaning the skin with alcohol and treating it with a
of each task version, the participant was allowed to pick one of the goodies
mild abrasive. For all EEG sites, InVivo-Metrics (Healdsburg, CA) Ag-
if he reached the following relatively low performance criterion not known
to him: He had to react correctly and fast enough to more than 50% of the
targets and to more than 50% of the nontargets. Because, by definition, the Table 1
first three letters of the evaluated sequence could not be targets in the 3-back
Grand Means (SDs) and Reliabilities of the Three Performance
version of the task, only the last 117 trials were used to check the performance
Measures in the Five Versions of the N-Back Task
criterion and to compute performance measures for statistical analyses. Par-
ticipants were told to react as quickly and as accurately as possible. RT Correct target
To check whether the possibility to obtain a reward moderated E effects, Target RT Variability reactions (%)
the 2-back version of the task was presented twice (using different se-
quences of letters): once with and once without the possibility to obtain M M M
reward. In contrast to the task presentation with possible reward already Task (SD) Rel (SD) Rel (SD) Rel
described above, in the presentation without possible reward, the partici-
pant was told that this time, he could not obtain a reward and that the 0-Back 452 .91 87 .79 94 .66
(57) (27) (05)
experimenter would not enter the room after the task to give performance
1-Back 490 .96 116 .86 87 .62
feedback. However, the participant was explicitly asked to engage in the (91) (38) (10)
task even though a reward could not be obtained. 2-Back 600 .94 157 .89 72 .82
For each of the five task presentations, the following three performance (149) (53) (17)
measures were calculated: (a) the average RT for correct reactions to 2-Back (no reward) 595 .95 165 .90 72 .85
targets, (b) the percentage of correct reactions to targets, and (c) the SD of (147) (63) (16)
RTs for all correct reactions. These choices were based on the following 3-Back 672 .95 177 .92 55 .58
considerations: (a) Gray, Chabris, and Braver (2003) have convincingly (214) (72) (14)
argued that the performance in target trials is a more valid indicator of
Note. N ⫽ 40. Rel ⫽ reliability estimated by applying the Spearman-
frontal cortical functions than the average RT of all correct reactions as Brown prophecy formula to the correlations between individual values
used by Lieberman and Rosenthal (2001). (b) It has long been known that calculated for the first and second halves of the trials. RT values were
when given the choice, introverts compared with extraverts tend to work square root transformed before calculation of correlations. For each task, both
more slowly but also more accurately on a variety of tasks (e.g., Wilson, halves contained 20 target trials each.
 EXTRAVERSION AND DOPAMINE
AgCl electrodes (8 mm) were used. For the ground electrode and the
electrooculogram (EOG) sites, disposable VivoMed (Servoprax, Wesel,
Germany) Ag-AgCl electrodes (10 mm) were employed.
EEG and EOG were amplified (EEG, gain ⫽ 500; EOG, gain ⫽ 100),
filtered (bandpass set to 1–50 Hz for EEG, lowpass set to 1,000 Hz for
EOG, 50-Hz notch filter enabled), digitized at 2,000 Hz, and stored. In a
second step, the signal was down-sampled to 250 Hz and converted to
physical units and visually scored for artifact. Data were excluded if
muscle, eye movement, or other artifacts were present. Spectral power was
then computed for 1-s artifact-free epochs overlapping by 50% using a fast
Fourier transform with a Hamming window. The Fourier transform was
performed on 4-s segments each containing 1-s epochs of windowed data
padded with zeros on both sides. The resulting estimates of power density
(microvolts squared per hertz) for each artifact-free 0.5-s interval were
clustered into two bands, theta (4.00 –7.75 Hz) and low alpha (8.00 –10.25
Hz). We decided to focus on low instead of broad alpha (8 –13 Hz) because
low alpha power has been shown to be more strongly associated with PET
measures of cortical activity (Oakes et al., 2004) and because we have
previously found frontal asymmetry effects to be more pronounced in the
low alpha band (Wacker et al., 2003).
Power values for the two bands were transformed to natural logarithms
to normalize the distributions of scores to be used in statistical analyses
(see, e.g., Davidson et al., 2000). Because estimates of spectral power
based on short time intervals are unreliable (Davidson et al.), means of
ln-transformed power estimates per 1-min recording phase were obtained
only if more than 20% of the 0.5-s data (i.e., at least 12 s of EEG data) were
nonmissing, otherwise the phase was set missing data for a given region-
band combination. We then computed the EEG indices of interest sepa-
rately for each recording phase as follows: (a) frontal EEG asymmetry ⫽
ln-transformed power at F4 minus ln-transformed power at F3 and (b)
frontal versus parietal EEG activity ⫽ ln-transformed power at Fz minus
ln-transformed power at Pz.
An analysis of variance (ANOVA) with E (extraverted, introverted) and
Substance (placebo, sulpiride) as between factors did not reveal any sig-
nificant group differences in the mean number of artifact-free epochs used
to compute within-subject averages of frontal EEG asymmetry and sym-
metrical frontal versus parietal EEG activity for each of the eight data
recordings, F(1, 36) ⱕ 0.34, p ⬎ .57. On average, the resting EEG indices
for each 1-min recording were based on M ⫽ 62.0 (SD ⫽ 20.5) artifact-free
epochs. Finally, we averaged the three recordings of the rest period.
Intelligence Test and Personality Questionnaires
We assessed general fluid intelligence using the German short version of
Cattell’s Culture Fair Test (Scale 3, Cattell & Weiß, 1971). Besides our
short version of the MPQ scale Positive Emotionality (see above) used for
selection of extreme groups according to agentic E, we also used German
versions of the revised Eysenck Personality Questionnaire (EPQ-R; Ruch,
1999) and the Zuckerman-Kuhlman Personality Questionnaire (ZKPQ;
Zuckerman, 2002). The EPQ-R measures Eysenck’s personality traits of E
(mainly the sociability component), Neuroticism, and Psychoticism. The
ZKPQ measures Zuckerman’s “Alternative Big Five” Aggression-
Hostility, Neuroticism-Anxiety, Sociability, Activity, and Impulsive Sen-
sation Seeking. The ZKPQ was employed for two reasons: (a) It nicely
separates the sociability component from the agency component of E by
providing separate scales for Sociability and Activity. Based on Depue and
Collins’ (1999) theory, strong relationships with indicators of brain DA are
expected only for Activity. (b) It also measures Impulsive Sensation
Seeking, that is, the trait considered by some to be more closely related to
brain DA than agentic E (see, e.g., Pickering & Gray, 1999).
Statistical Data Analysis
N-back performance measures and EEG parameters from the posttask
waiting periods were analyzed with three-way analysis of covariance
177
(ANCOVA) with Substance and E as between-subjects factors, Task (5; 0-
to 3-back with reward and 2-back without reward) as repeated factor, and
Block Number (coded from ⫺2 to 2) as a covariate. Making full use of the
Latin Square design employed to control for possible effects of task
presentation order, the covariate Block Number was included to reduce the
error term by controlling irrelevant variance attributable to linear changes
across task blocks. Using PROC MIXED of SAS/STAT (SAS 6.12 for
Macintosh, SAS Institute Inc., Cary, NC), the error variance-covariance
matrix for the repeated factor was specified to be completely general
(TYPE ⫽ UN option). EEG parameters from the initial rest period were
analyzed with two-way ANOVAs with Substance (2; placebo, sulpiride)
and E (2; introverted, extraverted) as between-subjects factors.
ANOVAs and ANCOVAs were followed by a priori-specified contrasts
tested with an ␣ level of .05. Contrasts for the EEG parameters from the
initial rest period were defined on the level of the Substance ⫻ E interac-
tion and depicted E effects within substance groups and Substance effects
within E groups. The E effects for the two EEG alpha measures (frontal
asymmetry and frontal vs. parietal activity) under placebo were tested
one-tailed, and the remaining contrasts for EEG variables were tested
two-tailed. Contrasts for the behavioral variables and the EEG parameters
from the posttask waiting periods were defined on the level of the Sub-
stance ⫻ E ⫻ Task interaction and were designed to test for: (a) differences
between introverts and extraverts within substance groups for each of the
five n-back tasks; (b) differences of linear changes in performance with
increasing task difficulty (0- to 3-back) between introverts and extraverts
within substance groups; (c) differences of these E effects (in each of the
five tasks and in the linear performance changes) between substance
groups or, put differently, differences of the substance effects (sulpiride-
placebo) between introverts and extraverts; and (d) differences between the
2-back tasks with versus without possible reward within each of the four
experimental groups. For the behavioral variables with the exception of the
contrasts subsumed under (d) directed hypotheses could be derived from
prior findings (see the Results section). Thus, for the behavioral variables,
one-tailed tests were calculated for contrasts (a), (b), and (c), whereas
contrasts (d) were tested two-tailed. For the two EEG alpha measures
(frontal asymmetry and frontal vs. parietal activity) during the waiting
periods, the differences between introverts and extraverts in the placebo
group for each of the five tasks were tested one-tailed (compared with
introverts, extraverts were expected to display greater relative right vs. left
frontal EEG alpha activity and lower frontal vs. parietal EEG alpha
activity). The remaining EEG contrasts were tested two-tailed.
Results
Check for Side Effects and Blindness to Substance Group
Participants did not report any adverse side effects. The ratings
of nausea and dizziness averaged across experimental phases were
very low (⬍1) in all four experimental groups (0 ⫽ not at all
applicable and 1 ⫽ not applicable). Also, the percentage of
participants, who guessed (in a forced choice question posed after
completion of the experiment) that they had received a pharma-
cologically active substance, did not differ between the four ex-
perimental groups (extraverts-sulpiride, 2/10; introverts-sulpiride,
5/10; extraverts-placebo, 5/10; introverts-placebo, 3/10). When
asked to evaluate the confidence in their own guess concerning the
substance received in a probability rating, none of the participants
reported to be 100% sure (M ⫽ 63.8%, SD ⫽ 18.7). Thus, it can
be concluded that participants were indeed blind to the experimen-
tal condition as intended.
178 WACKER, CHAVANON, AND STEMMLER
Check for Preexisting Differences Between Experimental
Groups
A series of ANOVAs with Substance and E as between-subjects
factors and age, general fluid intelligence, and all ZKPQ and
EPQ-R scales as dependent variables revealed no significant main
effects of Substance and no significant E ⫻ Substance interactions.
Several significant main effects of E were observed; supporting
convergent validity of our brief MPQ scale of agentic E extraverts
compared with introverts scored higher in the alternative measure
of agentic E; ZKPQ Activity, F(1, 36) ⫽ 36.82, p ⬍ .0001, ␩2 ⫽
.48; as well as in ZKPQ Sociability, F(1, 36) ⫽ 16.25, p ⬍ .001,
␩2 ⫽ .30, and in broader E as assessed by EPQ-R E, F(1, 36) ⫽
39.01, p ⬍ .0001, ␩2 ⫽ .50. However, extraverts compared with
introverts also scored somewhat lower on EPQ-R Neuroticism,
F(1, 36) ⫽ 8.80, p ⬍ .01, ␩2 ⫽ .19, and ZKPQ Neuroticism-
Anxiety, F(1, 36) ⫽ 8.35, p ⬍ .01, ␩2 ⫽ .19, possibly as a result
of the moderate negative correlation between E and neuroticism
factors frequently described in the literature (e.g., John & Srivastava,
1999). Finally, extraverts compared with introverts unexpectedly also
scored lower in general fluid intelligence, F(1, 36) ⫽ 8.63, p ⬍ .01,
␩2 ⫽ .18. No further main effects of E were observed.
N-Back Task Performance
The main effect of Task was significant for all performance
measures: target RT, F(4, 36) ⫽ 27.18, p ⬍ .0001; RT variability,
F(4, 36) ⫽ 47.56, p ⬍ .0001; and percent correct target reactions,
F(4, 36) ⫽ 111.27, p ⬍ .0001. As expected, both target RT and RT
variability increased with increasing task difficulty, whereas the
percentage of correct target reactions decreased (see Ms and SDs in
Table 1). For each of the three performance measures, a significant
linear effect of task block was observed: target RT, F(1, 36) ⫽
74.08, p ⬍ .0001; RT variability, F(1, 36) ⫽ 54.33, p ⬍ .0001; and
percent correct target reactions, F(1, 36) ⫽ 6.01, p ⬍ .05. The
more task blocks participants had already completed, the shorter
were the observed RTs in the current task block and the higher
were the observed percentages of correct target reactions. These
differences between task blocks are probably due to practice
effects that are not relevant to our hypotheses.
For the percentage of correct target reactions, no main effects or
interactions besides an unpredicted interaction of Substance and
Task were significant, F(4, 36) ⫽ 2.93, p ⬍ .05; all remaining p ⱖ
.09.3 However, for both RT measures, a significant two-way
interaction, E ⫻ Substance was observed: target RT, F(1, 36) ⫽
4.26, p ⬍ .05; RT variability, F(1, 36) ⫽ 8.85, p ⬍ .01; and for RT
variability, but not for target RT, the predicted three-way interac-
tion E ⫻ Substance ⫻ Task was significant: target RT, F(4, 36) ⫽
2.08, p ⫽ .10; RT variability, F(4, 36) ⫽ 3.27, p ⬍ .05. Besides a
two-way interaction, E ⫻ Task for RT variability, F(4, 36) ⫽ 2.39,
p ⫽ .07, no further main effects or interactions approached sig-
nificance (all p ⱖ .28).
Tests of a Priori Contrasts
For none of the performance measures, a significant difference
between the 2-back tasks with versus without possible reward was
observed in any of the four experimental groups (all p ⱖ .07, two-
tailed). The tests of the central a priori contrasts are documented in
Table 2. The corresponding means are shown in Figure 1. Based on
the findings by Lieberman and Rosenthal (2001), we predicted per-
formance differences between introverts and extraverts under placebo
specifically in the more difficult n-back tasks and a reversal of these
differences under sulpiride, as suggested by prior reports of reversals
of preexisting performance differences after application of dopami-
nergic agents (e.g., Corr & Kumari, 1997; Kimberg et al., 1997).
Indeed, the pattern of results observed for the target RTs in the
placebo group closely replicates the findings reported by Lieberman
and Rosenthal (2001): Introverts had longer target RTs than extraverts
in all tasks, and this difference was statistically reliable only for the
3-back task. Also, the linear increase in target RTs from 0- to 3-back
tasks under placebo was more pronounced in introverts than in ex-
traverts. In addition, the effects of sulpiride on target RTs differed
significantly between introverts and extraverts specifically in the 2-
and 3-back tasks (significant tetrad differences in the sulpiride-
placebo column of Table 2): Sulpiride tended to slow target RTs in
extraverts and to speed up target RTs in introverts.4 Put differently,
under sulpiride, the pattern of results observed under placebo tended
to be reversed as predicted; compared with introverts, extraverts had
numerically slower target RTs in the 2- and 3-back tasks and signif-
icantly larger linear increases in RTs from 0- to 3-back tasks.
It is interesting that both the E effects observed under placebo
and the modulation of the sulpiride effects by E could not only be
documented for target RTs but also (and even more strongly) for
RT variability (see Table 2 and Figure 1). In addition, under
placebo, introverts compared with extraverts did not only react
more slowly but also more accurately in target trials of the two
difficult n-back tasks, indicating a significant speed–accuracy
trade-off, correlation between target RT and percentage of correct
target reactions for the 3-back task, r(40) ⫽ .30, p ⬍ .05, one-
tailed.5 Also, the linear decrease in accuracy from 0- to 3-back was
significantly less pronounced for introverts than for extraverts.
However, apart from a weak tendency in the expected direction
observed for the rewarded 2-back task, the predicted reversal of the
E effects under sulpiride could not be shown for the percentage of
correct target reactions (see Table 2).
3
This effect resulted from the fact that compared with placebo the percent-
age of correct target responses was significantly higher under sulpiride only in
the 2-back task without reward, t(36) ⫽ 3.19, p ⫽ .003, but not in the other
tasks, p ⬎ .40. Because all other observed sulpiride effects on performance
measures (see Table 2) were similar in the 2-back task with and without
reward, this unpredicted observation is difficult to interpret.
4
For the 3-back task this tetrad difference (i.e., difference of the effects
of sulpiride-placebo in introverts vs. extraverts) was also significant when
an analogous ANCOVA model was calculated for mean RT of all correct
answers, t(36) ⫽ ⫺2.18, p ⬍ .05, one-tailed, or for mean RT of incorrect
answers, t(36) ⫽ ⫺2.30, p ⬍ .05, one-tailed.
5
The correlations between target RT and percentage of correct target
reactions for the 3-back task within the two substance groups were of
similar magnitude, although not significant due to the smaller sample size:
placebo, r(20) ⫽ .27, p ⫽ .13, one-tailed; sulpiride, r(20) ⫽ .33, p ⫽ .08,
one-tailed.
 EXTRAVERSION AND DOPAMINE 179

Table 2
Differences in N-Back Performance Between Introverts and Extraverts Within the Two Substance Groups and in Substance Effects
(Sulp-Plac): t-Values of Contrasts

Target RT RT Variability (SD) % correct target reactions

Extraversion contrast (I-E) Plac Sulp Sulp-Plac Plac Sulp Sulp-Plac Plac Sulp Sulp-Plac

0-back 1.86 0.60 ⫺0.89 1.78 0.29 ⫺1.06 1.66 0.47 ⫺0.84
1-back 1.15 ⫺0.00 ⫺0.81 2.07* ⫺0.32 ⫺1.69 0.27 ⫺0.61 ⫺0.62
2-back 1.38 ⫺1.10 ⫺1.76† 1.96† ⫺1.83† ⫺2.68†† 2.09† ⫺0.59 ⫺1.90†
2-back (no reward) 1.40 ⫺1.54 ⫺2.08† 2.69†† ⫺2.08† ⫺3.38†† 0.93 0.45 ⫺0.34
3-back 2.54†† ⫺1.26 ⫺2.68†† 3.41†† ⫺1.33 ⫺3.36†† 1.69† 0.63 ⫺0.75
Linear trend 0- to 3-back 1.96† ⫺1.78† ⫺2.64†† 2.60†† ⫺2.06† ⫺3.29†† 1.73† 0.38 ⫺0.96

Note. RT parameters were square root transformed before calculation of statistical analyses. df ⫽ 36. Plac ⫽ placebo; Sulp ⫽ sulpiride. Contrasts for the
0-back task and for the 1-back task were tested two-tailed; the remaining contrasts were tested one-tailed.
* p ⬍ .05, two-tailed. † p ⬍ .05, one-tailed. †† p ⬍ .01, one-tailed.

EEG Indices: Posttask Waiting Periods
Frontal EEG Asymmetry
Three-way ANCOVAs with E and Substance as between-groups
factors, Task as within-group factor, and Block Number as covari-
ate did not reveal any significant main effects or interactions for
frontal EEG asymmetry in the theta or low alpha bands, for all
effects F(1, 36) ⱕ 1.65, p ⱖ .21 or F(4, 36) ⱕ 2.25, p ⱖ .08.
Frontal Versus Parietal EEG Activity
Three-way ANCOVAs computed for frontal versus parietal EEG
activity in the theta and low alpha bands revealed the following
statistically reliable effects: a main effect for Task in the theta
band, F(4, 36) ⫽ 3.21, p ⬍ .05, resulting from a linear decrease
from 0- to 3-back waiting period, t(36) ⫽ ⫺3.02, p ⬍ .01, two-tailed,
a two-way interaction E ⫻ Substance for the theta band, F(1,
36) ⫽ 19.77, p ⬍ .0001, and, less pronounced, for the low alpha
band, F(1, 36) ⫽ 4.40, p ⬍ .05, and finally a three-way interaction,
E ⫻ Substance ⫻ Task, for both the theta band, F(4, 36) ⫽ 4.99, p ⬍
.01, and the low alpha band, F(4, 36) ⫽ 4.67, p ⬍ .01.
Tests of a Priori Contrasts
Neither for the theta band nor for the low alpha band a signif-
icant difference between the 2-back tasks with versus without
possible reward was observed in any of the four experimental
groups (all p ⱖ .12, two-tailed). The tests of the central a priori
contrasts are documented in Table 3. For the theta band, the
corresponding means are shown in Figure 2.
Under placebo, compared with extraverts, introverts had greater
relative frontal versus parietal EEG theta activity during the wait-
ing periods after all tasks except the 0-back task. Under sulpiride,
an opposite difference between introverts and extraverts was ob-
served after all tasks except the 0-back task. Furthermore, under
placebo, extraverts, but not introverts, showed a linear decrease in
frontal versus parietal theta activity from 0- to 3-back task, t(36) ⫽
⫺2.02, p ⫽ .05, two-tailed, whereas under sulpiride only, intro-
verts, but not extraverts, showed this decrease, t(36) ⫽ ⫺2.62, p ⫽
.01, two-tailed. Thus, sulpiride did not only completely reverse the
overall differences in frontal versus parietal theta activity observed
under placebo, it also reversed the differences in linear decreases
from 0- to 3-back: Under placebo, extraverts showed a (numeri-
cally) larger decrease than introverts, whereas under sulpiride,
introverts showed a significantly larger decrease than extraverts.
As shown in Table 3, the pattern of results observed for relative
frontal versus parietal EEG activity in the low alpha band was similar
to the one just described for relative frontal versus parietal EEG
activity in the theta band. However, the effects were generally less
pronounced with the differences between introverts and extraverts
within substance groups only significant for the 1-back task.

Figure 1. Target RT and RT variability in the four rewarded n-back tasks (M and SEM values of untransformed variables).
180 WACKER, CHAVANON, AND STEMMLER

Table 3
Differences in Frontal vs. Parietal EEG Activity (Post Task Waiting Periods) Between Introverts
and Extraverts Within the Two Substance Groups and in Substance Effects (Sulp-Plac):
t-Values of Contrasts

Frontal vs. Parietal Theta Frontal vs. Parietal Low Alpha

Extraversion Contrast (I-E) Plac Sulp Sulp-Plac Plac Sulp Sulp-Plac

0-back 1.55 ⫺1.11 ⫺1.89 0.16 ⫺1.01 ⫺0.82

1-back 3.17** ⫺4.08** ⫺5.12** 2.02† ⫺2.91** ⫺3.48**
2-back 3.12** ⫺2.47* ⫺3.95** 0.27 ⫺1.15 ⫺1.01
2-back (no reward) 2.49* ⫺2.44* ⫺3.49** 0.73 ⫺1.26 ⫺1.40
3-back 2.42* ⫺3.67** ⫺4.29** 1.21 ⫺1.72 ⫺2.07*
Linear trend 0- to 3-back 1.29 ⫺3.51** ⫺3.35** 0.64 ⫺0.46 ⫺0.78

Note. df ⫽ 36. Plac ⫽ placebo; Sulp ⫽ sulpiride. Contrasts for low alpha in the placebo group were tested
two-tailed except for the linear trend contrast; the latter as well as the remaining contrasts were tested one-tailed.
* p ⬍ .05, two-tailed. ** p ⬍ .01, two-tailed. † p ⬍ .05, one-tailed.

EEG Indices: Initial Rest Period
As explained in the Method section, our conceptualization of
traits as dispositions that are activated only in relevant situational
context led us to expect strong correlations between agentic E and
the EEG, particularly when the recording situation is characterized
by the presence of positive incentives. In the present study, posi-
tive incentives were arguably most salient in the waiting periods
after each of the four potentially rewarded n-back tasks, during
which participants were waiting to be rewarded for their perfor-
mance. As documented above, we found some strong relationships
between agentic E and EEG parameters obtained for these prere-
ward recordings. However, the pattern of results, although some-
what weaker for the 0-back task, was quite similar across the four
task versions and, most notably, did not differ between the re-
warded and the nonrewarded version of the 2-back task. In addi-
tion, prior research on the relationship between personality traits
and asymmetrical and symmetrical frontal EEG activity focused
almost exclusively on recordings made under resting conditions
without any explicit reward (or punishment) signals (Coan &
Allen, 2004a; Hewig et al., 2004, 2006). Therefore, we also
checked whether the effects reported for the posttask waiting
periods can be documented for the initial rest period.
Frontal EEG Asymmetry
Paralleling the observations for the posttask waiting periods,
two-way ANOVAs with the factors E and Substance did not reveal
significant main effects or interactions for frontal EEG asymmetry
in the theta and low alpha bands, for all effects F(1, 36) ⱕ 1.42,
p ⱖ .24, ␩2 ⱕ .038. Also, a priori contrasts did not reveal any
significant differences between introverts and extraverts within
either the placebo group or the sulpiride group (all p ⱖ .38,
two-tailed).

Frontal Versus Parietal EEG Activity

Figure 2. Frontal versus parietal EEG theta activity during the waiting
periods after the four rewarded n-back tasks (M and SEM values).
Two-way ANOVAs with the factors E and Substance did not
show significant main effects for frontal versus parietal EEG
activity in the theta or low alpha bands, for all main effects F(1,
36) ⱕ 1.88, p ⱖ .18, ␩2 ⱕ .045. However, paralleling the obser-
vations obtained for the posttask waiting periods, the two-way
interaction E ⫻ Substance was highly significant for the theta
band, F(1, 36) ⫽ 18.62, p ⬍ .0001, ␩2 ⫽ .332, and just failed to
reach formal significance for the low alpha band, F(1, 36) ⫽ 3.21,
p ⫽ .08, ␩2 ⫽ .076. A priori contrasts revealed that introverts had
greater frontal versus parietal EEG theta activity than extraverts
under placebo, t(36) ⫽ 2.17, p ⬍ .05, two-tailed, whereas the
opposite was true under sulpiride, t(36) ⫽ ⫺3.39, p ⬍ .01, two-
tailed (M ⫽ 0.40, 0.66, 0.77, and 0.44, SD ⫽ 0.18, 0.22, 0.22, and
0.23, for Extraverts-Placebo, Introverts-Placebo, Extraverts-
Sulpiride, and Introverts-Sulpiride, respectively). The difference
between sulpiride and placebo was significant for both introverts
and extraverts, t(36) ⫽ ⫺2.25, p ⬍ .05, two-tailed, and t(36) ⫽
3.86, p ⬍ .001, two-tailed, respectively.
 EXTRAVERSION AND DOPAMINE 181

Relationship Between Resting EEG
and 3-Back Performance
As shown above, the pattern of results observed for frontal
versus parietal EEG activity in the theta band is not only highly
consistent across the different recording periods (initial rest period
and posttask waiting periods) but also highly similar to the pattern
observed for RT measures in the 3-back task. Thus, the question of
whether 3-back RT measures can be predicted from resting frontal
versus parietal EEG activity in the theta band arises. Correlation
analyses revealed that 3-back target RT and RT variability were
not only highly interrelated, r(40) ⫽ .88, p ⬍ .0001, one-tailed, but
also could be predicted from frontal versus parietal EEG theta
activity measured during the initial rest period, target RT, r(40) ⫽
.26, p ⫽ .05, one-tailed; RT variability, r(40) ⫽ .29, p ⬍ .05,
one-tailed.
Are the Observed Effects Specific to Agentic E?
To check whether the effects of sulpiride on target RT, RT
variability, and frontal versus parietal EEG theta activity during
the initial rest period were indeed only modulated by agentic E and
not by other (correlated) personality traits, age, or general fluid
intelligence, we calculated a series of ANCOVAs for each of the
three dependent variables using the statistical models described
above, but now entering an additional covariate (either an EPQ-R
scale, a ZKPQ scale, age, or intelligence scores), its two-way
interaction with Substance, and, for the RT data, its two-way
interaction with Task and its three-way interaction with Substance
and Task (see Table 4 for the intercorrelations of the questionnaire
scales, age, and intelligence scores). If the effects reported above
are indeed specific to agentic E, they should disappear when the
alternative measure of agentic E (ZKPQ Activity) is used as a
covariate but remain significant when either one of the other
personality scales, age, or general fluid intelligence is used as a
covariate. In brief, this is exactly what we found. After controlling
for ZKPQ Activity, the E ⫻ Substance interaction was no longer
statistically reliable for frontal versus parietal EEG theta activity,
F(1, 34) ⫽ 3.18, p ⬎ .05, Cohen’s d ⫽ 0.61. However, the E ⫻
Substance interaction was still highly significant when either one
of the other personality scales, age or general fluid intelligence,
was controlled for, F(1, 34) ⱖ 12.93, p ⱕ .001, Cohen’s d ⱖ 1.23.
Likewise, substance no longer modulated the E effects (cf.
sulpiride-placebo column in Table 2) for 3-back target RT, t(34) ⫽
⫺0.55, p ⫽ .29, one-tailed, and 3-back RT variability, t(34) ⫽
⫺1.38, p ⫽ .08, one-tailed, after controlling for ZKPQ Activity
(Cohen’s d ⱕ 0.47). However, these contrasts remained significant
for 3-back target RT, t(34) ⱕ ⫺2.15, p ⱕ .02, one-tailed, and
3-back RT variability, t(34) ⱕ ⫺2.72, p ⱕ .005, when either one
of the other personality scales, age or general fluid intelligence,
was controlled for (Cohen’s d ⱖ 0.74). It is notable that controlling
for general fluid intelligence even strengthened the E ⫻ Substance
interaction (i.e., the tetrad difference contrast) for both 3-back
target RT, t(34) ⫽ ⫺3.63, p ⫽ .0013, and 3-back RT variability,
t(34) ⫽ ⫺4.05, p ⫽ .0003, Cohen’s d ⱖ 1.25.
In summary, these results show that (a) the effects are indeed
specific for agentic E (⫽discriminant validity) and (b) not specific
to the measure of agentic E used to select groups of introverts and
extraverts but also observable with an alternative questionnaire
scale (⫽convergent validity; see Figure 3 for a scatterplot of the
simple correlations between ZKPQ Activity and frontal vs. parietal
EEG theta activity within the two substance groups).
Discussion
With the present study, we aimed at identifying an easily ob-
tainable EEG index and a behavioral paradigm that can be used to
investigate Depue and Collins’ (1999) suggestion of a dopaminer-
gic basis for the personality trait agentic E in humans. Based on
recent findings, we argued that performance in the difficult ver-
sions of the n-back task and both asymmetrical frontal EEG
activity in the low alpha band and symmetrical frontal versus
parietal EEG activity in both theta and (low) alpha bands can be
hypothesized to be related not only to agentic E but also to certain
aspects of central DA functioning. To put these hypotheses to an
initial test, we assessed performance in the n-back task and the
EEG variables of interest in groups of healthy young men high
versus low in agentic E after administration of either placebo or the
selective DA D2 antagonist sulpiride. We predicted that under

Table 4
Intercorrelations of Personality Scales, General Fluid Intelligence, and Age in the Whole Sample

ZKPQ ZKPQ ZKPQ ZKPQ ZKPQ EPQ-R EPQ-R EPQ-R EPQ-R CFT 3
Act Sy N-Anx ImpSS Agg-H E N P L (gf)

ZKPQ Sy .60**
ZKPQ N-Anx ⫺.23 ⫺.31
ZKPQ ImpSS .21 .38* ⫺.23
ZKPQ Agg-H .22 .34* .21 .11
EPQ-R E .60** .81** ⫺.53** .58** .22
EPQ-R N ⫺.23 ⫺.24 .77** ⫺.11 .13 ⫺.31
EPQ-R P ⫺.26 ⫺.10 .00 .30 .30 ⫺.10 .06
EPQ-R L .08 ⫺.19 .02 ⫺.19 ⫺.17 ⫺.05 ⫺.07 ⫺.35*
CFT 3 (gf) ⫺.37* ⫺.29 .04 ⫺.06 ⫺.26 ⫺.44** .20 .20 ⫺.32*
Age ⫺.26 ⫺.31 .11 ⫺.22 .02 ⫺.16 .09 ⫺.09 .28 ⫺.08

Note. N ⫽ 40. Act ⫽ activity; Agg-H ⫽ aggression-hostility; CFT 3 (gf) ⫽ general fluid intelligence as measured by the short version of the Culture Fair
Test (Scale 3); E ⫽ extraversion; ImpSS ⫽ impulsive sensation seeking; L ⫽ lie scale; N ⫽ neuroticism; N-Anx ⫽ neuroticism-anxiety; P ⫽ psychoticism;
Sy ⫽ sociability.
* p ⬍ .05, two-tailed. ** p ⬍ .01, two-tailed.
182 WACKER, CHAVANON, AND STEMMLER
Figure 3. Scatterplot of the correlation between agentic E (ZKPQ Ac-
tivity) and frontal versus parietal EEG theta activity during the initial rest
period within sulpiride (closed triangles; r ⫽ .64, p ⬍ .01, two-tailed) and
placebo (open circles; r ⫽ ⫺.50, p ⬍ .05, two-tailed). The correlations
differed significantly between sulpiride and placebo, z ⫽ 3.81, p ⬍ .0002,
two-tailed. Significant correlations within substance groups were not ob-
served for age, general fluid intelligence, the EPQ-R scales, or the remain-
ing ZKPQ scales.
placebo, participants high versus low in agentic E would show (a)
shorter target RTs and lower RT variability in the difficult n-back
tasks, (b) greater relative left frontal cortical activity (i.e., greater
relative right frontal EEG activity in the low alpha band; see Oakes
et al., 2004), and (c) lower EEG activity in the low alpha band and
either lower or higher EEG theta activity at frontal versus parietal
electrode locations. Furthermore, we predicted that all of these
differences between participants high versus low in agentic E
would be reversed under sulpiride, resulting in significant two-way
interactions of the two factors E (high vs. low) and Substance
(placebo vs. sulpiride) for each of the dependent variables.
These predictions were strongly supported by the present data
for n-back RT measures and for relative frontal versus parietal
resting EEG theta activity but not for frontal EEG asymmetry. In
addition, supplementary analyses showed that the observed effects
were indeed specific to agentic E and not due to different person-
ality traits (e.g., Sociability, Neuroticism, or Impulsive Sensation
Seeking), age, or general fluid intelligence. We will now discuss
each of these findings in more detail before sketching some im-
plications for further empirical and theoretical work on the psy-
chobiological basis of agentic E.
E, DA, and N-Back Performance
The differences in n-back performance observed between par-
ticipants high versus low in agentic E in the placebo group closely
replicate the observations of Lieberman and Rosenthal (2001):
Exactly as reported by these authors, (agentic) introverts had
longer RTs than extraverts, and this difference was maximal (and
statistically significant) in the 3-back task. Also, in our study, just
like in theirs, (agentic) introverts showed a significantly larger
linear increase in RTs from the 0- to the 3-back task. In addition,
under sulpiride, the RT differences between introverts and extra-
verts tended to be completely reversed as predicted.
Two further results of the present study indicate that the introverts’
slower responding in the difficult n-back tasks does not represent a
simple performance deficit, as suggested by Lieberman and Rosenthal
(2001), but may be more accurately interpreted as a different way of
performing the task: First, the pattern of differences between intro-
verts and extraverts under placebo and the partial reversal of these
differences under sulpiride documented for mean target RT were also
observable (even somewhat more pronouncedly) for the variability of
RTs across trials. Thus, it is unclear whether the observed RT effects
represent differences in the overall speed level (i.e., mean RTs), in the
variability of this level across trials, or both. In future studies, an
in-depth examination of differences in patterns of RT trial-to-trial
variation between introverts and extraverts could provide some inter-
esting insights concerning this issue (e.g., introverts compared with
extraverts may show more pronounced post-error slowing, resulting
in increased RT variability). Second, several observations indicate the
presence of a significant speed–accuracy trade-off at least for the
more difficult versions of the n-back task: Under placebo and relative
to extraverts, introverts reacted not only slower on target trials, but
they also made fewer mistakes. Furthermore, speed and accuracy in
the target trials of the 3-back task were negatively correlated: Partic-
ipants who reacted relatively slowly on target trials (e.g., introverts)
tended to make relatively few mistakes. The fact that the predicted
reversal of the E effect under sulpiride could be observed more clearly
for the RT measures than for the percentage of correct target trials is
difficult to interpret because it may simply be a consequence of the
lower reliability of the latter parameter (see Table 1).
Even though the n-back performance differences between intro-
verts and extraverts cannot be represented on a continuum ranging
from good to poor performance and, thus, seem to be more
complex than suggested by Lieberman and Rosenthal (2001), the
present results provide initial support for these authors’ proposal
that they are related to brain DA. Because the RT differences
between introverts and extraverts (in particular, the differences in
the linear RT increases from easy to difficult tasks) were almost
completely reversed by application of the selective D2 antagonist
sulpiride, it seems prudent to conclude that the observed individual
differences in behavior are at least partly based on individual
differences in functional brain DA activity at D2 receptor sites.
These conclusions certainly are not only limited by several caveats
that we will discuss below, but they also do not speak to the important
question of which presumably dopaminergically modulated cognitive
processes underlie the observed differences in behavior. This question
cannot be answered with the present data because it is well known that
the difficult n-back tasks involve a rather large variety of cognitive
processes (e.g., Gray et al., 2003), and it is currently unclear which of
these processes are tapped by the available performance measures.
Because there is at least some consensus in the literature that the
difficult n-back tasks tap certain aspects of cognitive control, our
findings conform with the conclusion recently drawn by Gray and
Burgess (2004, p. 36) that “some component of the cognitive control
network depends on BAS [or agentic E] in a true functional sense,
albeit the relation is complex.”
Recent models of DA functioning may serve as a starting point
to tackle this complexity in future studies: Seamans and Yang
(2004) suggested that DA acts to contract or expand the breadth of
 EXTRAVERSION AND DOPAMINE
information held in WM buffers in PFC networks. In a similar
manner, Dreisbach and Goschke (2004) proposed that DA adjusts
the balance of goal maintenance versus flexible switching between
multiple goals. It is possible that under normal conditions, extra-
verts compared with introverts tend more toward flexible switch-
ing and to holding a greater breadth of information in WM (see
Avila, Barros, Ortet, Parcet, & Ibanez, 2003). Of course, this
interpretation fits nicely with Lieberman and Rosenthal’s (2001)
suggestion that the E-related differences in n-back RTs result from
superior multitasking ability in extraverts. However, the notion of
a trade-off between flexible switching and stable maintenance of
goals also implies that life may not be quite as unfair for introverts as
surmised by these authors: The extraverts’ greater multitasking or
switching ability may come at the price of increased distractibility.
E, DA, and Frontal EEG Activity
Asymmetrical frontal EEG activity. In the present study, under
placebo, agentic E was not significantly related to resting frontal
EEG asymmetry in either the theta or low alpha bands. With regard
to the low alpha band, these observations contrast with several
studies (e.g., Coan & Allen, 2003; Harmon-Jones & Allen, 1997;
Schmidt, 1999; Sutton & Davidson, 1997), in which E and related
traits have been linked to relative left frontal resting cortical
activity (i.e., relative right frontal EEG activity in the alpha band).
However, Hagemann, Naumann, Thayer, and Bartussek (2002)
have shown that this measure is strongly influenced by unknown
occasion-specific state factors, which account for almost as much
variance as a temporally stable latent trait factor. Thus, it is
possible that in the present study, these unknown occasion-specific
influences on frontal EEG asymmetry completely masked an ex-
isting association of the latent trait component with agentic E.
An alternative explanation for the unexpected absence of cor-
relations between E and frontal asymmetry is suggested by our
conceptualization of E and other personality traits as dispositions
that predict behavior and the associated brain states only in rele-
vant situational contexts (for details, see Stemmler, 1997). In the
present study, participants knew that they were about to engage in
a challenging task, and the goodies that could be obtained for good
performance were placed directly in front of them during the
whole experiment, including the initial rest period. However, be-
cause we observed virtually no differences between the potentially
rewarded and the nonrewarded 2-back task, we simply do not
know whether these efforts to create an experimental setting that is
strongly characterized by positive incentives and therefore suitable
to activate agentic E were actually successful. It is possible that
different features of the EEG recording situation were even more
salient for the participants. For instance, recent findings suggest
that for male participants female experimenters represent a situa-
tional context that activates an affiliative form of approach moti-
vation that is not mediated by brain DA but nonetheless strongly
related to frontal EEG asymmetry (Kline, Blackhart, & Joiner,
2002; Pauls et al., 2005). In the present study, both men and
women served as experimenters. Thus, besides the possibility that
in the predicted correlations between agentic E and EEG asym-
metry were masked by a sizable contribution of irrelevant
occasion-specific variance to asymmetry scores, it could also be
that we did not succeed in providing a sufficiently “pure” positive
incentive context to specifically produce variance in EEG asym-
183
metry relevant to agentic E. Direct comparisons between these two
alternatives are needed to reach reliable conclusions concerning
the reasons for the inconsistent trait-asymmetry correlations ob-
served here and elsewhere (for review, see Hagemann et al., 2002).
Likewise, it would certainly be premature to reject the idea of a partly
dopaminergic basis of frontal EEG asymmetry based on the complete
absence of reliable effects of the selective DA D2 receptor antagonist
sulpiride observed in this study. As detailed in the Introduction, there are
well-founded reasons to expect an association between resting frontal
brain asymmetry and DA, and initial indirect evidence that such an
association indeed exists comes from studies in which significant asso-
ciations between resting frontal EEG asymmetry in the alpha band and
spontaneous eye blink rates (i.e., a peripheral indicator of DA activity, see
Karson, 1983) were observed (e.g., Pauls et al., 2005). It is interesting that
very recent work with monkeys and humans suggests that spontaneous
eye blink rate may depend more strongly on DA D1 than DA D2 receptor
signaling (Jutkiewicz & Bergman, 2004; van der Post, de Waal, de Kam,
Cohen, & van Gerven, 2004).6 If this relative D1-specificity of dopami-
nergic effects also holds for frontal EEG asymmetry, this would offer a
straightforward explanation for the absence of reliable D2 antagonist
effects in the present study. Research with different dopaminergic agents
that also tap the D1 receptor system is needed to directly test this
hypothesis.
Symmetrical frontal versus parietal EEG activity. As pre-
dicted, we observed a significant difference in EEG theta activity
at frontal relative to parietal electrode locations between partici-
pants low versus high in agentic E under placebo: High agentic E
was associated with stronger parietal (vs. frontal) theta activity.
This finding is consistent with earlier work by Stenberg (1992),
who reported an association of a BAS-E factor with posterior theta
activity and an association of a behavioral inhibition system (BIS)-
Anxiety factor with frontal theta activity.
The finding that E is related to posterior versus frontal brain
activity is also echoed by observations from a PET study of
cerebral blood flow correlates of E. In summarizing their findings,
Johnson et al. (1999) note that on the whole, introversion was
associated with activity in frontal areas, whereas E was associated
with activity in more posterior areas. In addition, as already noted
above, a number of recent functional magnetic resonance imaging
studies reported significant associations between E-related traits
and activity in the ACC (Canli et al., 2001, 2004; Gray & Braver,
2002; Kumari et al., 2004). Given the evidence that frontal midline
EEG theta activity reflects activity in the ACC (Pizzagalli et al.,
2003), it seems prudent to assume that the effects observed in the
present study for frontal versus parietal theta activity partly result
from E-related activity differences in the ACC. If this hypothesis
holds up in future studies that combine EEG source localization
algorithms and brain imaging technology or at least employ some-
what more densely spaced EEG derivations that allow to tease
apart the individual contributions of frontal midline and parietal
6
Also, note that Depue et al. (1994) did not observe a statistically
reliable main effect of the D2 agonist bromocriptine on (maximum) eye
blink rate, nor a modulation of such an effect by E. Only the time to blink
maximum after administration of bromocriptine was found to correlate
with E. However, in contrast to the link between E and the bromocriptine-
induced prolactin response, Depue and colleagues did not report a repli-
cation of this correlation in a larger sample (Depue, 1995, 1996).
184 WACKER, CHAVANON, AND STEMMLER
theta by providing a means to control for volume conduction and
individual differences in skull thickness (see Introduction section),
then the current knowledge on the functions of the ACC could be
used as a basis to develop hypotheses concerning the functional
significance of frontal versus parietal theta activity; currently, the
ACC is hypothesized to function as a discrepancy detector that
signals the organism when things are going worse than expected
(e.g., Holroyd & Coles, 2002) and that serves to recruit lateral
prefrontal cortices to exert top-down control to resolve the de-
tected discrepancies and/or bias responding in a goal-congruent
manner (e.g., MacDonald, Cohen, Stenger, & Carter, 2000). Thus,
greater frontal versus parietal theta may indicate the initiation of
conflict- or discrepancy-triggered goal shielding or, in terms of
Seamans and Yang’s (2004) model of DA functioning in the PFC, a
contraction (vs. expansion) of the breadth of information held in WM.
Although the present data thus do not allow definite conclusions
concerning both the underlying neural generators and the func-
tional significance of frontal versus parietal theta activity, they do
provide some interesting insights with respect to the implicated
neurochemistry. For this EEG measure, we observed a strong
interaction of the two factors E (high vs. low) and Substance
(placebo vs. sulpiride); under sulpiride, E effects were completely
reversed, with high agentic E associated with stronger frontal
versus parietal theta activity. The selective DA D2 receptor antag-
onist sulpiride resulted in a shift toward more frontal versus
parietal EEG theta activity in participants high in agentic E and in
a shift toward more parietal versus frontal EEG theta activity in
participants low in agentic E leading to no discernible effects for
the group as a whole. These observations suggest that frontal
versus parietal EEG theta activity contains a considerable amount
of trait variance associated both with agentic E and DA function-
ing. Therefore, the present study provides initial evidence that
frontal versus parietal EEG theta activity qualifies as an easily
obtainable EEG index that can be used as an efficient tool to
further probe the DA basis of E.
For frontal versus parietal EEG activity in the low alpha band,
we observed only weak and (with the exception of the waiting
period after the 1-back task) statistically unreliable effects that
mirrored the strong effects observed for the theta band. Why were
we not able to demonstrate a more consistent relationship between
frontal versus parietal EEG activity and E in the placebo group as
the results reported by Hewig et al. (2004, 2006) would seem to
suggest? A number of possible explanations are offered by several
significant methodological differences between the Hewig et al.
studies and the present work (e.g., sample size, number and length
of EEG recordings, parametrization, and EEG reference scheme).
Anyway, the similarity of the patterns we observed in the low
alpha and theta bands points to the intriguing possibility that both
the Hewig et al. findings and the present results are based on the
same electrocortical phenomenon: Boundaries between traditional
EEG frequency bands are somewhat arbitrary. Maybe the presum-
ably DA-related aspects of oscillatory brain activity that we ob-
served most clearly in the theta range can sometimes (or for some
individuals) also be detected in the alpha band.
(Agentic) E and the BIS?
In contrast to the theoretical perspective adopted here (see
Depue & Collins, 1999), other personality theorists (e.g., Gray,
1970; Patterson & Newman, 1993) maintain that E is not only
related to increased sensitivity or activity of a BFS-like system that
facilitates behavior in response to incentives (i.e., in Gray’s ter-
minology, the BAS) but also to reduced sensitivity or activity of a
second broad motivational system, the BIS, that serves to inhibit
behavior in response to events that are potentially incompatible
with the organism’s current goals (e.g., signals for punishment and
nonreward). Thus, for these authors, E represents the habitual
balance between behavioral activation/facilitation on the one hand
and behavioral inhibition on the other. It is interesting that several
of the present findings may also be viewed from this theoretical
perspective. According to Gray and McNaughton (2000), the core
function of the BIS lies in the detection of conflict among con-
currently activated goals. As explained above, the detection of
such goal conflicts is thought to be associated with activity in the
ACC and frontal midline EEG theta. Also, some authors (e.g.,
Stenberg, 1992) have reported a positive correlation between traits
from the BIS spectrum and frontal EEG theta. Thus, the greater
frontal versus parietal theta we observed in introverts compared
with extraverts under placebo may indicate stronger BIS sensitiv-
ity in addition to (or even rather than) weaker BAS/BFS sensitivity
in the former. In a similar manner, the introverts’ tendency to
perform more slowly but somewhat more accurately than extra-
verts in the difficult n-back tasks under placebo may indicate not
only lower BAS/BFS sensitivity, but also greater BIS sensitivity,
resulting in a greater responsiveness to the negative feedback
associated with mistakes (see also Patterson & Newman, 1993).
In brief, our suggestion that both the electrocortical and the
behavioral differences between introverts and extraverts reported
here indicate individual differences in the balance of the contrac-
tion versus expansion of the breadth of information held in WM
could potentially be mapped onto earlier suggestions that E results
from individual differences in the balance between two neurobe-
havioral systems, the BIS and the BAS/BFS. This would of course
provide a straightforward explanation for the fact that some of the
associations we observed for agentic E are about as suggestive of
individual differences in BIS activity as of individual differences
in BAS/BFS activity. However, in the present study, we did not
observe any significant effects for traits more closely aligned with
habitual BIS sensitivity (i.e., EPQ-R Neuroticism and ZKPQ
Neuroticism-Anxiety), and the current psychobiological personal-
ity literature provides no theoretical basis to predict an association
between habitual BIS sensitivity and brain DA. Therefore, it may
be more appropriate to stick to Depue and Collins’ (1999) sug-
gestion that agentic E maps only onto a single, affectively unipolar
motivational system, which is neurobiologically based on brain
DA (i.e., the BFS) and to conceptualize the involvement of the BIS
as somewhat more indirect. For example, one could simply add the
idea that high state activity in the BFS/BAS is incompatible with
high state activity in the BIS, which places partly antagonistic
constraints on behavior and the associated brain states.
Underlying Mechanisms: The Return of Eysenck’s (1967)
Inverted U?
The present findings do not speak to the important question of
why the D2 antagonist sulpiride had opposite psychophysiological
and behavioral effects in individuals high versus low in agentic E.
Opposing effects of dopaminergic drugs in different groups of
 EXTRAVERSION AND DOPAMINE
participants have been described frequently before (e.g., Corr &
Kumari, 1997; Kimberg et al., 1997; Mattay et al., 2000, 2003;
Takeshita & Ogura, 1994). It is typical that such effects are
explained by assuming that the different groups occupy separate
locations on a continuum of baseline dopaminergic functioning
and that the relationship between the dependent variable, in which
effects of the dopaminergic drug were observed, and baseline DA
follows an inverted U-shaped function (e.g., Depue & Collins,
1999; Mattay et al., 2003). Invoking such an inverted U-shaped
function to explain differences between introverts and extraverts
will surely remind many readers of Eysenck’s (1967) famous
arousal theory of E in which he postulated that by virtue of
hardwired individual differences in the brain’s ascending reticular
activating system, introverts are typically more aroused/arousable
than extraverts except for particularly arousing conditions in which
the protective mechanism of transmarginal inhibition leads to
paradoxical arousal decreases in introverts, resulting in a reversal
of the typical arousal differences. Indeed, given that the ascending
DA fibers are an important (although not exclusive) component of
the ascending reticular activating system (Robbins, 1997), the
present findings could be interpreted as evidence that, on the
whole, Eysenck (1967) was right all along (see also Corr &
Kumari, 1997). However, what exactly are the hypothesized dif-
ferences in baseline DA functioning between extraverts and intro-
verts, and where do they come from? Novel and more specific
hypotheses concerning this issue may eventually come from mo-
lecular genetic studies that have recently begun to identify func-
tionally relevant gene polymorphisms that are linked to both E and
functional differences in DA neurotransmission (see, e.g., Mattay
et al., 2003; Reuter & Hennig, 2005). Combining such molecular
genetic assessments with the methods employed in the present
study could thus be a promising approach for further research
aimed at going beyond the inverted U to specify the precise
mechanisms underlying the differences between introverts and
extraverts reported here.
Limitations and Future Directions
Even though we were able to demonstrate that the observed
effects were indeed due to agentic E rather than several other
correlated personality traits, age, or general fluid intelligence, of
course the possibility still remains that some other third variable
correlated with E actually underlies our findings. Such potential
alternative variables range from other unmeasured personality
dimensions to established correlates of E like the responsiveness to
pharmacological manipulations of general arousal (e.g., with caf-
feine) and the sensitivity of task performance to time of day
(Revelle, Humphreys, Simon, & Gilliand, 1980), just to name a
few. As one of the reviewers noted, such potential third variable
explanations represent a general problem for this kind of research
as long as we have not found a way to experimentally vary E itself,
which would probably be akin to changing the engrained hardwir-
ing of the brain. For the time being, all we can do is include
multiple methods of assessment for the targeted construct to foster
integration of different lines of investigation of E and measure as
many potentially relevant third variables as possible to rule out at
least the more obvious alternative explanations.7
Because we only used one (albeit highly selective) dopaminer-
gic drug, it remains unclear whether the observed effects are in fact
185
specific to DA or rather attributable to unknown unspecific side
effects of sulpiride. In addition, in this initial study, we only
employed one single dosage of sulpiride and made no attempt to
investigate dose-response relationships. Future studies will have to
use a greater variety of pharmacological agents and dosages to
provide converging evidence for the interpretation in terms of a
DA mechanism suggested here. Also, it should be noted that
because at present no pharmacological agents exist that exclu-
sively influence the mesocorticolimbic DA system rather than the
other DA systems of the brain (nigrostriatal, tuberoinfundibular,
and incertohypothalamic), the strong Depue and Collins (1999)
hypothesis of a specific role of the mesocorticolimbic DA system
cannot be validated with the methods employed here. Furthermore,
because we only investigated preselected groups of male partici-
pants, it remains an open question whether the present findings
generalize to randomly drawn female samples.
Finally, because we conceptualize traits as dispositions that are
only operative in relevant situational contexts (see Stemmler,
1997), the present study was conducted in an incentive motiva-
tional setting specifically designed to activate agentic E. However,
because we made only a relatively weak and absolutely unsuccess-
ful attempt to manipulate the situational context (i.e., announcing
for one of the more challenging tasks that no reward and no
performance feedback would be given afterward), we cannot tell
whether our observations are indeed specific to trait-relevant pos-
itive incentive situations. If such a situational specificity of the
observed associations could be demonstrated in future studies, this
would not only support the general concept of traits as dispositions
(Stemmler, 1997), but also Depue and Collins’ (1999) claim that
agentic E is based on positive incentive motivation, thereby forg-
ing further links between these authors’ theoretical model and the
extensions proposed here.
Conclusions
In the present study, we demonstrated that the effects of the
selective DA D2 antagonist sulpiride on both RT measures in the
difficult n-back tasks and frontal versus parietal EEG theta activity
are strongly and specifically modulated by agentic E. These ob-
servations not only offer novel support for Depue and Collins’
(1999) DA theory of agentic E, they also provide new opportuni-
ties to further extend this theory to the human level by identifying
behavioral indicators as well as an easily accessible, noninvasive
EEG measure that are sensitive to both E and manipulations of
functional brain DA activity and thus open up a range of new
research avenues for further work on the psychobiological basis of
this important dimension of human personality.
7
At least a third variable explanation based on the Revelle et al. (1980)
findings concerning E, time of day, and caffeine effects would not seem
particularly convincing because these authors observed consistent findings
only for impulsivity, which is a facet of Eysenck’s early E concept quite
distinct from agentic E as defined and measured in the present article.
There were no associations between ZKPQ Impulsive Sensation Seeking
and either one of our two alternative measures of agentic E. Also, ZKPQ
Impulsive Sensation Seeking (even though strongly correlated to EPQ-R E)
did not produce any significant effects in the present study.
186 WACKER, CHAVANON, AND STEMMLER
References
American Psychiatric Association. (1994). Diagnostic and statistical man-
ual of disorders (4th ed.). Washington, DC: Author.
Avila, C., Barros, A., Ortet, G., Parcet, M. A., & Ibanez, M. I. (2003).
Set-shifting and sensitivity to reward: A possible dopamine mechanism for
explaining disinhibitory disorders. Cognition and Emotion, 17, 951–959.
Barch, D. M. (2004). Pharmacological manipulation of human working
memory. Psychopharmacology (Berlin), 174, 126 –135.
Blasi, G., Mattay, V. S., Bertolino, A., Elvevag, B., Callicott, J. H., Das, S.,
et al. (2005). Effect of catechol-O-methyltransferase val158met geno-
type on attentional control. Journal of Neuroscience, 25, 5038 –5045.
Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides, J., Smith, E. E., &
Noll, D. C. (1997). A parametric study of prefrontal cortex involvement
in human working memory. Neuroimage, 5, 49 – 62.
Breier, A., Kestler, L., Adler, C., Elman, I., Wiesenfeld, N., Malhotra, A.,
et al. (1998). Dopamine D2 receptor density and personal detachment in
healthy subjects. American Journal of Psychiatry, 155, 1440 –1442.
Caley, C. F., & Weber, S. S. (1995). Sulpiride: An antipsychotic with
selective dopaminergic antagonist properties. Annals of Pharmacother-
apy, 29, 152–160.
Canli, T., Amin, Z., Haas, B., Omura, K., & Constable, R. T. (2004). A
double dissociation between mood states and personality traits in the
anterior cingulate. Behavioral Neuroscience, 118, 897–904.
Canli, T., Zhao, Z., Desmond, J. E., Kang, E., Gross, J., & Gabrieli, J. D.
(2001). An fMRI study of personality influences on brain reactivity to
emotional stimuli. Behavioral Neuroscience, 115, 33– 42.
Carver, C. S., & White, T. L. (1994). Behavioral inhibition, behavioral activation,
and affective responses to impending reward and punishment: The BIS/BAS
scales. Journal of Personality and Social Psychology, 67, 319–333.
Cattell, R. B., & Weiß, R. H. (1971). Grundintelligenztest Skala 3 (CFT 3).
Göttingen, Germany: Hogrefe.
Cloninger, C. R., Svrakic, D. M., & Przybeck, T. R. (1993). A psychobi-
ological model of temperament and character. Archives of General
Psychiatry, 50, 975–990.
Coan, J. A., & Allen, J. J. B. (2003). Frontal EEG asymmetry and the behavioral
activation and inhibition systems. Psychophysiology, 40, 106–114.
Coan, J. A., & Allen, J. J. B. (2004a). Frontal EEG asymmetry as a
moderator and mediator of emotion. Biological Psychology, 67,7– 49.
Coan, J. A., & Allen, J. J. B. (2004b). The state and trait nature of frontal
asymmetry in emotion. In K. Hugdahl & R. J. Davidson (Eds.), The
asymmetrical brain (pp. 565– 615). Cambridge, MA: MIT Press.
Corr, P. J., & Kumari, V. (1997). Sociability/impulsivity and attenuated
dopaminergic arousal: Critical flicker/fusion frequency and procedural
learning. Personality and Individual Differences, 22, 805– 815.
Davidson, R. J. (1998). Affective style and affective disorders: Perspec-
tives from affective neuroscience. Cognition and Emotion, 12, 307–330.
Davidson, R. J. (2004). What does the prefrontal cortex “do” in affect: Perspectives
on frontal EEG asymmetry research. Biological Psychology, 67, 219–233.
Davidson, R. J., Jackson, D. C., & Larson, C. L. (2000). Human electro-
encephalography. In J. T. Cacioppo, L. G. Tassinary & G. G. Berntson
(Eds.), Handbook of psychophysiology (2nd ed., pp. 27–52). New York:
Cambridge University Press.
Depue, R. A. (1995). Neurobiological factors in personality and depres-
sion. European Journal of Personality, 9, 413– 439.
Depue, R. A. (1996). Neurobiology and the structure of personality:
Implications for the personality disorders. In J. Clarkin & M. F. Len-
zenweger (Eds.), Major theories of personality disorders. New York:
Guilford Press.
Depue, R. A., & Collins, P. F. (1999). Neurobiology of the structure of
personality: Dopamine, facilitation of incentive motivation, and extra-
version. Behavioral and Brain Sciences, 22, 491–569.
Depue, R. A., Luciana, M., Arbisi, P., Collins, P., & Leon, A. (1994).
Dopamine and the structure of personality: Relation of agonist-induced
dopamine activity to positive emotionality. Journal of Personality and
Social Psychology, 67, 485– 498.
Dreisbach, G., & Goschke, T. (2004). How positive affect modulates
cognitive control: Reduced perseveration at the cost of increased dis-
tractibility. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 30, 343–353.
Eysenck, H. J. (1967). The biological basis of personality. Springfield, IL:
Charles C Thomas.
Farde, L., Gustavsson, J. P., & Jonsson, E. (1997). D2 dopamine receptors
and personality traits. Nature, 385, 590.
Gray, J. A. (1970). The psychophysiological basis of introversion-
extraversion. Behaviour Research and Therapy, 8, 249 –266.
Gray, J. A., & McNaughton, N. (2000). The neuropsychology of anxiety
(2nd ed.). New York: Oxford University Press.
Gray, J. R., & Braver, T. S. (2002). Personality predicts working-memory-
related activation in the caudal anterior cingulate cortex. Cognitive,
Affective and Behavioral Neuroscience, 2, 64 –75.
Gray, J. R., & Burgess, G. C. (2004). Personality differences in cognitive
control? BAS, processing efficiency, and the prefrontal cortex. Journal
of Research in Personality, 38, 35–36.
Gray, J. R., Chabris, C. F., & Braver, T. S. (2003). Neural mechanisms of
general fluid intelligence. Nature Neuroscience, 6, 316 –322.
Gray, N. S., Pickering, A. D., & Gray, J. A. (1994). Psychoticism and
dopamine D2 binding in the basal ganglia using single photon emission
tomography. Personality and Individual Differences, 17, 431– 434.
Hagemann, D., Naumann, E., Thayer, J. F., & Bartussek, D. (2002). Does resting
electroencephalograph asymmetry reflect a trait? An application of latent state-
trait theory. Journal of Personality and Social Psychology, 82, 619–641.
Harmon-Jones, E. (2004). Contributions from research on anger and cog-
nitive dissonance to understanding the motivational functions of asym-
metrical frontal brain activity. Biological Psychology, 67, 51–76.
Harmon-Jones, E., & Allen, J. J. B. (1997). Behavioral activation sensi-
tivity and resting frontal EEG asymmetry: Covariation of putative indi-
cators related to risk for mood disorders. Journal of Abnormal Psychol-
ogy, 106, 159 –163.
Hewig, J., Hagemann, D., Seifert, J., Naumann, E., & Bartussek, D. (2004). On the
Selective Relation of Frontal Cortical Asymmetry and Anger-Out Versus
Anger-Control. Journal of Personality and Social Psychology, 87, 926–939.
Hewig, J., Hagemann, D., Seifert, J., Naumann, E., & Bartussek, D. (2006).
The relation of cortical activity and BIS/BAS on the trait level. Biolog-
ical Psychology, 71, 42–53.
Holroyd, C. B., & Coles, M. G. (2002). The neural basis of human error
processing: Reinforcement learning, dopamine, and the error-related
negativity. Psychological Review, 109, 679 –709.
Jackson, D. M., & Hashizume, M. (1987). Bromocriptine-induced loco-
motor stimulation in mice is modulated by dopamine D-1 receptors.
Journal of Neural Transmission, 69, 131–145.
Jasper, H. H. (1958). The ten-twenty electrode system of the International Feder-
ation. Electroencephalography and Clinical Neurophysiology, 10, 371–375.
John, O. P., & Srivastava, S. (1999). The Big Five Trait taxonomy: History,
measurement, and theoretical perspectives. In L. A. Pervin & O. P. John
(Eds.), Handbook of personality: theory and research (2nd ed., pp.
102–138). New York: Guilford Press.
Johnson, D. L., Wiebe, J. S., Gold, S. M., Andreasen, N. C., Hichwa, R. D.,
Watkins, G. L., et al. (1999). Cerebral blood flow and personality: A
positron emission tomography study. American Journal of Psychiatry,
156, 252–257.
Jutkiewicz, E. M., & Bergman, J. (2004). Effects of dopamine D1 ligands on
eye blinking in monkeys: Efficacy, antagonism, and D1/D2 interactions.
Journal of Pharmacology and Experimental Therapeutics, 311, 1008 –1015.
Karson, C. N. (1983). Spontaneous eye-blink rates and dopaminergic
systems. Brain, 106, 643– 653.
Kestler, L. P., Malhotra, A. K., Finch, C., Adler, C., & Breier, A. (2000).
The relation between dopamine D2 receptor density and personality:
 EXTRAVERSION AND DOPAMINE
Preliminary evidence from the NEO personality inventory-revised. Neu-
ropsychiatry, Neuropsychology, and Behavioral Neurology, 13, 48 –52.
Kimberg, D. Y., D’Esposito, M., & Farah, M. J. (1997). Effects of
bromocriptine on human subjects depend on working memory capacity.
Neuroreport, 8, 3581–3585.
Kline, J. P., Blackhart, G. C., & Joiner, T. E. (2002). Sex, lie scales, and
electrode caps: An interpersonal context for defensiveness and anterior
electroencephalographic asymmetry. Personality and Individual Differ-
ences, 33, 459 – 478.
Kumari, V., ffytche, D. H., Williams, S. C., & Gray, J. A. (2004).
Personality predicts brain responses to cognitive demands. Journal of
Neuroscience, 24, 10636 –10641.
Laakso, A., Vilkman, H., Kajander, J., Bergman, J., Paranta, M., Solin, O., et al.
(2000). Prediction of detached personality in healthy subjects by low dopamine
transporter binding. American Journal of Psychiatry, 157, 290–292.
Laakso, A., Wallius, E., Kajander, J., Bergman, J., Eskola, O., Solin, O., et
al. (2003). Personality traits and striatal dopamine synthesis capacity in
healthy subjects. American Journal of Psychiatry, 160, 904 –910.
Lieberman, M. D., & Rosenthal, R. (2001). Why introverts can’t always
tell who likes them: Multitasking and nonverbal decoding. Journal of
Personality and Social Psychology, 80, 294 –310.
MacDonald, A. W., Cohen, J. D., Stenger, V. A., & Carter, C. S. (2000).
Dissociating the role of the dorsolateral prefrontal and anterior cingulate
cortex in cognitive control. Science, 288, 1835–1838.
Margraf, J. (1997). Mini-DIPS (Kurz-)Interview bei psychischen Störun-
gen. Berlin: Springer.
Mattay, V. S., Callicott, J. H., Bertolino, A., Heaton, I., Frank, J. A.,
Coppola, R., et al. (2000). Effects of dextroamphetamine on cognitive
performance and cortical activation. Neuroimage, 12, 268 –275.
Mattay, V. S., Goldberg, T. E., Fera, F., Hariri, A. R., Tessitore, A., Egan,
M. F., et al. (2003). Catechol O-methyltransferase val158-met genotype
and individual variation in the brain response to amphetamine. Proceed-
ings of the National Academy of Sciences, USA, 100, 6186 – 6191.
Mauri, M. C., Bravin, S., Bitetto, A., Rudelli, R., & Invernizzi, G. (1996).
A risk-benefit assessment of sulpiride in the treatment of schizophrenia.
Drug Safety, 14, 288 –298.
McClelland, G. R., Cooper, S. M., & Pilgrim, A. J. (1990). A comparison
of the central nervous system effects of haloperidol, chlorpromazine and
sulpiride in normal volunteers. British Journal of Clinical Pharmacol-
ogy, 30, 795– 803.
Meyer-Lindenberg, A., Rammsayer, T., Ulferts, J., & Gallhofer, B. (1997).
The effects of sulpiride on psychomotor performance and subjective
tolerance. European Neuropsychopharmacology, 7, 219 –223.
Mischel, W., & Shoda, Y. (1998). Reconciling processing dynamics and
personality dispositions. Annual Review of Psychology, 49, 229 –258.
Missale, C., Nash, S. R., Robinson, S. W., Jaber, M., & Caron, M. G.
(1998). Dopamine receptors: From structure to function. Physiological
Review, 78, 189 –225.
Nitschke, J. B., Heller, W., Etienne, M. A., & Miller, G. A. (2004).
Prefrontal cortex activity differentiates processes affecting memory in
depression. Biological Psychology, 67, 125–143.
Oakes, T. R., Pizzagalli, D. A., Hendrick, A. M., Horras, K. A., Larson,
C. L., Abercrombie, H. C., et al. (2004). Functional coupling of simul-
taneous electrical and metabolic activity in the human brain. Human
Brain Mapping, 21, 257–270.
Patterson, C. M., & Newman, J. P. (1993). Reflectivity and learning from
aversive events: Toward a psychological mechanism for the syndromes
of disinhibition. Psychological Review, 100, 716 –736.
Pauls, C. A., Wacker, J., & Crost, N. W. (2005). The two components of
social desirability and their relations to resting frontal brain asymmetry.
Journal of Individual Differences, 26, 29 – 42.
Pickering, A. D., & Gray, J. A. (1999). The neuroscience of personality. In
187
L. A. Pervin & O. P. John (Eds.), Handbook of personality: Theory and
research (2nd ed., pp. 277–299). New York: Guilford Press.
Pickering, A. D., & Gray, J. A. (2001). Dopamine, appetitive reinforce-
ment, and the neuropsychology of human learning: An individual dif-
ferences approach. In A. Eliasz & A. Angleitner (Eds.), Advances in
research on temperament (pp. 113–149). Lengerich, Germany: Pabst.
Pizzagalli, D. A., Oakes, T. R., & Davidson, R. J. (2003). Coupling of theta
activity and glucose metabolism in the human rostral anterior cingulate
cortex: An EEG/PET study of normal and depressed subjects. Psycho-
physiology, 40, 939 –949.
Reuter, M., & Hennig, J. (2005). Association of the functional catechol-
O-methyltransferase VAL158MET polymorphism with the personality
trait of extraversion. Neuroreport, 16, 1135–1138.
Revelle, W., Humphreys, M. S., Simon, L., & Gilliand, K. (1980). The
interactive effect of personality, time of day, and caffeine: A test of the
arousal model. Journal of Experimental Psychology General, 109, 1–31.
Robbins, T. W. (1997). Arousal systems and attentional processes. Biolog-
ical Psychology, 45, 57–71.
Ruch, W. (1999). Die revidierte Fassung des Eysenck Personality Ques-
tionnaire und die Konstruktion des deutschen EPQ-R bzw. EPQ-RK.
The Eysenck Personality Questionnaire-Revised and the construction of
the German standard and short versions (EPQ-R and EPQ-RK).
Zeitschrift für Differentielle und Diagnostische Psychologie, 20, 1–24.
Schacter, D. L. (1977). EEG theta waves and psychological phenomena: A
review and analysis. Biological Psychology, 5, 47– 82.
Schmidt, L. A. (1999). Frontal brain electrical activity in shyness and
sociability. Psychological Science, 10, 316 –320.
Seamans, J. K., & Yang, C. R. (2004). The principal features and mech-
anisms of dopamine modulation in the prefrontal cortex. Progress in
Neurobiology, 74, 1–58.
Stemmler, G. (1997). Selective activation of traits: Boundary conditions for the
activation of anger. Personality and Individual Differences, 22, 213–233.
Stenberg, G. (1992). Personality and the EEG: Arousal and emotional
arousability. Personality and Individual Differences, 13, 1097–1113.
Sutton, S. K., & Davidson, R. J. (1997). Prefrontal brain asymmetry: A
biological substrate of the behavioral approach and inhibition systems.
Psychological Science, 8, 204 –210.
Takeshita, S., & Ogura, C. (1994). Effect of the dopamine D2 antagonist
sulpiride on event-related potentials and its relation to the law of initial
value. International Journal of Psychophysiology, 16, 99 –106.
Tellegen, A., & Waller, N. G. (in press). Exploring personality through test
construction: Development of the Multidimensional Personality Ques-
tionnaire (MPQ). Minneapolis: University of Minnesota Press.
van der Post, J., de Waal, P. P., de Kam, M. L., Cohen, A. F., & van
Gerven, J. M. (2004). No evidence of the usefulness of eye blinking as
a marker for central dopaminergic activity. Journal of Psychopharma-
cology, 18, 109 –114.
Wacker, J., Heldmann, M., & Stemmler, G. (2003). Separating emotion
and motivational direction in fear and anger: Effects on frontal asym-
metry. Emotion, 3, 167–193.
Wilson, G. (1978). Introversion/extroversion. In H. London & J. E. Exner
(Eds.), Dimensions of personality (pp. 217–262). New York: Wiley.
Zuckerman, M. (1991). The psychobiology of personality. New York:
Cambridge University Press.
Zuckerman, M. (2002). Zuckerman-Kuhlman personality questionnaire
(ZKPQ): An alternative five-factorial model. In B. d. Raad & M.
Perugini (Eds.), Big Five assessment (pp. 377–396). Göttingen, Ger-
many: Hogrefe & Huber.
Received August 12, 2005
Revision received January 25, 2006
Accepted February 3, 2006 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 188 –204 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.188

Possible Selves and Academic Outcomes: How and When Possible Selves
Impel Action
Daphna Oyserman, Deborah Bybee, and Kathy Terry
The University of Michigan

Puzzled by the gap between academic attainment and academic possible selves (APSs) among low-
income and minority teens, the authors hypothesized that APSs alone are not enough unless linked with
plausible strategies, made to feel like “true” selves and connected with social identity. A brief interven-
tion to link APSs with strategies, create a context in which social and personal identities felt congruent,
and change the meaning associated with difficulty in pursuing APSs (n ⫽ 141 experimental, n ⫽ 123
control low-income 8th graders) increased success in moving toward APS goals: academic initiative,
standardized test scores, and grades improved; and depression, absences, and in-school misbehavior
declined. Effects were sustained over a 2-year follow-up and were mediated by change in possible selves.

Keywords: possible selves, African American, Hispanic, prevention, self-regulation

From early adolescence, the future is an important component of
self-concept (McGuire & Padawe-Singer, 1976), and doing well in
school is a common element of youths’ future-oriented selves
(Oyserman, Johnson, & Bybee, 2006). These future or possible
selves (PSs) are positive and negative images of the self already in
a future state—the “clever” self who passed the algebra test, the
“fat” self who failed to lose weight, the “fast” self who fell in with
the “wrong” crowd (Oyserman & Markus, 1990). A number of
self-regulatory models have posited positive and negative conse-
quences of PS–PS motivate current action (Baumeister, 1998;
Carver & Scheier, 1990; Eccles & Wigfield, 2002; Higgins, 1996a;
Oyserman & Markus, 1990; Weinstein, 1993). Failure to attain PSs
(Oyserman & Fryberg, 2006), hoped for selves, or ideal selves
(Strauman, 2002; Strauman & Higgins, 1988) may increase risk of
depression. There is some evidence that imagining successful PSs
improves well-being (King, 2001) and performance (Ruvolo &
Markus, 1992). However, PSs do not always sustain self-
regulatory action. Youth fail algebra, fall off weight maintenance,
and engage in risky behavior, perhaps also increasing their risk of
feeling depressed at their inability to move toward their self-goals.
Why might PSs fail to sustain self-regulatory action? Given that
the self is multidimensional and includes multiple potentially
competing goals (e.g., Abrams, 1994; Burke, 2003; King & Smith,
2004; Oyserman, 2001; Settles, 2004), a particular PS may fail to
sustain self-regulatory action because it conflicts with other parts
Daphna Oyserman, Deborah Bybee, and Kathy Terry, The Institute for
Social Research, The University of Michigan.
This work was supported by National Institute of Mental Health Grant
R01 MH 58299. We thank the students, parents, trainers, teachers, school
staff, and community members involved in the intervention, data collec-
tion, and tracking; Jim Klein for help in obtaining school data; and our
coordinating and database managers, Carol Carlin, Johnessa Dimicks,
Tami Hart-Johnson, and Angelique Lange.
Correspondence concerning this article should be addressed to Daphna
Oyserman, The Institute for Social Research, The University of Michigan,
Ann Arbor, MI 48106-1248. E-mail: daphna.oyserman@umich.edu
of self-concept. Another possibility is that PSs fail to sustain
self-regulatory action because sustaining self-regulatory effort
over time is difficult, and youth may misinterpret difficulty as
evidence that the PS is not a reasonable goal and should be
abandoned. To predict when a particular PS is likely to motivate
action, is it necessary to have a model that predicts which PSs are
likely to be online in working memory and, of these PSs, which
will be invested in over time.
In the current paper, we focus on school-related PSs and outline
and test a predictive model. We propose that youth have difficulty
creating and sustaining school-focused PSs when they perceive
these PSs to be incongruent with important social identities (e.g.,
racial-ethnic identities), misinterpret difficulties in working on
these PSs as evidence that academic goals are unrealistic PSs, and
live in social contexts that fail to cue strategies for attaining their
PS goal. Youth will commit sustained self-regulatory effort to a PS
when the PS itself is effective and contains behavioral strategies
and social context supports working on the PS, when the PS feels
congruent with important social identities, and when difficulty
working on the PS is construed as normative. In the following
sections, we outline each of these links and the intervention we
developed based on this model with a goal of enhancing the
self-regulatory impact of PS.
PSs and Self-Regulatory Behavior
Effective PSs
A number of studies suggest that PSs differ in self-regulatory
effectiveness. Self-regulatory effort improves when youth have
both positive PSs (goals) and negative PSs (fears) in the same
domain (“balanced” PSs) (Oyserman & Markus, 1990) and when
youth have incorporated detailed strategies into their PSs (“plau-
sible” PSs) (Oyserman, Bybee, Terry, & Hart-Johnson, 2004).
When PSs are balanced, individuals select strategies that both
increase the likelihood of becoming like the positive PSs and
decrease the likelihood of becoming like the negative PSs, thereby
focusing self-regulation and broadening effort (Oyserman &

188
 POSSIBLE SELVES AND ACADEMIC OUTCOMES
Markus, 1990). Plausibility provides automatic cuing of predevel-
oped strategies (e.g., “set my alarm,” “go to class even if my
friends skip”); indeed, youth with plausible academic PSs (APSs)
are able to improve their grades over the course of the school year,
whereas those who simply had APSs are not (Oyserman et al.,
2004). Despite the fact that most low-income youth have at least
one PS focused on school, few of these PSs include strategies
(Oyserman et al., 2006). Many youths fail to attain even the basic
academic PS goal of graduating high school. Thus, the national
average for on-time graduation is 75% overall, 50% for African
Americans, and 53% for Hispanics) (Orfield, Losen, Wald, &
Swanson, 2004). In urban centers like Detroit, graduation esti-
mates are even lower (between 40% and 44% on-time graduation)
(Detroit News, 2005).
Contextual Cuing of PSs
Not only is the potential self-regulatory impact of having a PS
goal undermined when PSs are not balanced or plausible, self-
regulatory effectiveness is also undermined when social contexts
do not cue the PS. Because information that is cued (chronically or
situationally made salient) is likely to be used in judgments and
decision-making (Higgins, 1996b), contextually cued PSs should
influence self-regulatory behavior more than those that are not
cued. Moreover, to sustain ongoing engagement in school, PSs
must be linked with behavioral strategies; positive expected APSs
need to be linked with strategies to attain them, and feared APSs
need to be linked with strategies to avoid them.
Although resource-rich contexts such as a middle-class neigh-
borhood and school provide models of success and a developed
structure to guide the process of attaining APSs, this is unlikely to
be the case in resource-limited contexts. In middle-class contexts,
strategies may be automatically cued; parents, teachers, and par-
ents of friends all converge to emphasize homework, persistence in
the face of difficulty, tutoring, or staying after school if needed. In
low-socioeconomic status (SES) contexts, strategies are unlikely to
be automatically cued because these contexts are less likely to
present easily accessible models or to guide success (e.g., Rod-
erick, 2003). In low-SES contexts, youth encounter adults who are
likely to be unemployed, have low academic attainment, and hold
nonprofessional jobs. Given lack of easily accessible models or
automatically cued strategies, youth may maintain an abstract
commitment to education without connecting these PSs to every-
day behavior, expressing high aspirations even as their behavior
reflects avoidance or even flight from school (for qualitative
description, see Roderick, 2003).
APSs and Social Identities
Self-concepts include both personal and social identities. Social
identities are aspects of self-concept based not in individual traits
and goals but on group-based traits and goals (Oyserman, in press).
It seems reasonable that social identities will incorporate commu-
nity expectations about the occupations and academic attainment
of in-group members. Minority youth living in low-SES contexts are
exposed to images of the in-group as low achieving (Thomas,
Townsend, & Belgrave, 2003), raising the possibility that the selves
possible for in-group members may not feel congruent with APSs.
189
The negative link youths perceive between minority status and
academic attainment and eventual occupation can be seen both in
ethnographic descriptions and in experimental paradigms. Ethno-
graphic research suggests that high school students perceive Lati-
nos as more likely to become manual laborers, Asians to do well
in school, and African Americans to do poorly in school (Kao,
2000). The same results emerge from scenario-based experimental
research, whether focused on the link between minority status and
low academic attainment or on the link between low social class
and low academic attainment. Thus, when a failing student is
described, Latino and African American students are more likely
to predict that the target is Latino or African American than White
(Graham, 2001). When asked to predict academic performance of
a target student, low-income students infer worse performance
from low (vs. middle) social class peers (Régner, Huguet, &
Monteil, 2002; Weinger, 2000).
When imagining what future is possible for one’s self, such
negative preformed group images are likely to be highly accessi-
ble, making social group membership feel like it conflicts with
APSs. Working toward one’s APS is likely to feel harder in the
presence of accessible images of in-group members engaging in
behaviors that undermine chances of attaining APSs and failing to
engage in behaviors that would help attain APSs. Although there
is debate as to the degree of evidence that school success is viewed
as a White middle-class goal (Cook & Ludwig, 1997; Ferguson,
1998), there is consistent evidence of the stereotyped link between
minority status and low achievement (e.g., Steele, 1997).
The stereotype threat literature documents that simply bringing
to mind category membership as minority or working class damp-
ens academic performance (e.g., Croizet & Claire, 1998; Steele,
1997). Individuation or separation of self from in-group alleviates
this effect (Ambady, Paik, Steele, Owen-Smith, & Mitchell, 2004),
but the idea of intervening to disconnect youth from their racial-
ethnic in-group (e.g., creating “racelessness”; see Fordham &
Ogbu, 1986) is unappealing and likely to have other negative
consequences (Arroyo & Zigler, 1995). Thus, rather than attempt-
ing to dampen the centrality of important social identities as a way
to improve success in attaining APSs, increasing felt congruence
between APSs and social identity, is a more reasonable strategy.
APSs and Inoculation From Overinterpretation of
Difficulty
In addition to the linkage of the APS to self-regulatory behavior
and integration with important social identities, taking into account
research on how salient information influences judgment and
behavior may close what appears to be a puzzling gap between the
high value placed on education and the actual attainment of low-
income and minority youth. Working on one’s APS is likely to be
difficult. To judge what this difficulty means, whether APSs are
“true” PSs or contradict other important social identities, to judge
whether attaining these PSs is plausible or not worth the effort,
teens must answer the implied questions “Why is engaging in this
APS so hard for me; is this really the true me? Do we have PSs like
this?” Likewise, they must judge whether particular behavioral
patterns (e.g., asking for help) are likely to work and if they contradict
in-group identity (e.g., “Will asking the teacher for help actually help
me succeed in school or is it just a ‘White’ thing to do?”).
190 OYSERMAN, BYBEE, AND TERRY
To better understand how the feelings of difficulty that are
experienced as adolescents imagine and pursue their PS influence
commitment to that PS, it is fruitful to reconsider James’ (1890)
original formulation of the self. James viewed the self as com-
prised both of content—what one thinks about when one thinks
about one’s self and the accompanying metacognitive process—
the feeling of thinking about one’s self. This implies that self-
judgments about who one is or may become are based on both
content (what comes to mind) and process. Considering the content
of thoughts about the self separately from the feelings associated
with these thoughts parallels work in social cognition (Schwarz,
1998, 2002). This work proposes that human reasoning is accom-
panied by metacognitive experiences of relative ease (difficulty)
and fluency (disfluency) (Schwarz, 1998, 2002).
Following from this research in social cognition, when imagin-
ing a PS is accompanied by a metacognitive experience of diffi-
culty, the feeling of difficulty is interpreted with a naı̈ve theory—
things that are hard to think of are less likely to be true (Higgins,
1998; Schwarz & Bless, 1992; Schwarz & Clore, 1996). The
experience of ease or difficulty when bringing to mind a PS can
provide the basis for inferring whether a PS is a “true” self that is
worth pursuing and investing effort in or a “false” self, conflicting
with social identities. Metacognitive experience of ease also pro-
vides feedback as to whether the gap between the current and PS
is manageable or unmanageable and therefore whether the PS
should be expanded or abandoned.
Although the experience of metacognitive difficulty is generally
interpreted as meaning “not true for me,” a number of studies have
documented that other interpretations are possible (Rothman &
Schwarz, 1998). Sports stories abound with reinterpretation of the
meaning of experienced difficulty (e.g., “no pain, no gain”) and the
need to keep trying (e.g., “you miss 100% of the shots you don’t
take”). In the case of attempting to attain APS, although the
metacognitive experience of difficulty is generally interpreted as
“not the true me,” the experience of difficulty could be reinter-
preted to mean other things. Difficulty can be viewed as a norma-
Social
Identity
STJ Possible
Intervention Selves
Meta-cognitive
Experience
Figure 1. Process model. Theorized effects of the STJ intervention on PSs and of PSs on academic engage-
ment, academic outcomes, and depression.
tive part of the process (e.g., “success is 1% inspiration and 99%
perspiration”). Difficulty can also provide evidence of progress
(e.g., “the important things in life are the ones you really have to
work for”); if difficulty and failures along the way are viewed as
critical to eventual success, then difficulty is evidence of striving.
Successful movement toward positive APSs and away from
feared APSs requires ongoing behavior; it is not enough to com-
plete one homework assignment or stay after class one day. If
one’s metacognitive experience is that working on a PS is difficult
and if this difficulty is interpreted with a naı̈ve theory that ease is
associated with truth, then difficulties associated with working
toward the PS will undermine it. As we have outlined, low-income
and minority youth are likely to experience at least three sources of
difficulty— difficulty bringing to mind APSs and linking them to
strategies, difficulty sustaining the behavioral self-regulation PS
strategies entail, and difficulty integrating APSs and social
identities.
Youth growing up in low-SES contexts have multiple models of
adults who failed to attain their PSs, making it unlikely that they
will recognize the normativeness of difficulties and instead likely
that they will misinterpret feelings of difficulty as a sign of
inevitable failure. This misinterpretation is crucial because it is
likely to undermine behavioral persistence in pursuit of PS goals.
Thus, rather than assuming that youth are able to make sense of
difficulty as normative, low-SES youth are likely to need specific
inoculation from overinterpreting current difficulty and failure as
predictive of future possibilities. Taken together, their metacogni-
tive experience of difficulty is likely to provide feedback that
APSs are false rather than true selves, cuing disengagement from
these PSs and the goal pursuit they imply.
PSs: A Process Model and Proposed Intervention
Figure 1 presents our process model of the connections among
PSs, self-regulatory behaviors, academic outcomes, and mental
health. The process model links social identities and metacognitive
Academic
Outcomes
Self-Regulatory
Behaviors
Well-being
Mental Health
 POSSIBLE SELVES AND ACADEMIC OUTCOMES
experience to PSs and links PSs to persistent engagement in
self-regulatory behavior. To test this model, we developed an
intervention (labeled STJ in Figure 1). The goals of the interven-
tion were to evoke PSs and strategies to attain them, forge links
between PSs and strategies that are not otherwise automatic, in-
oculate youth from misinterpreting failure and setbacks in attain-
ing these PSs, and create a link between social identity and PSs.
Ease is typically the basis for metacognitive judgment, so we
took care to ensure that sessions felt easy, so that students would
not immediately develop a metacognitive sense that thinking of the
future is hard and therefore “not for me.” We based the interven-
tion in school so that strategies articulated in the intervention
would be cued in school. We structured intervention activities to
make APSs salient, create linking connections between APSs and
strategies, make salient naı̈ve theories of the meaning of difficulty
that link difficulty with progress toward meaningful goals, and
create a space in which APSs and social identity are congruent.
Together, these activities were designed to “inoculate” youth
from withdrawing effort to attain positive APSs and avoid feared
APSs. In the long term, this change in PSs was expected to evoke
persistent change in self-regulatory behavior. That is, we expected
that behavior relevant to APSs (e.g., doing homework) would
increase and that behavior undermining APSs (e.g., misbehaving
in class) would decrease. Over time, sustained self-regulatory
behavior should produce better academic outcomes, that is, grade
point average (GPA). Moreover, because sustained self-regulatory
behavior should reduce discrepancy between current and future
selves, intervention youth should have the added benefit of re-
duced risk of depression (Higgins, 1997) and other mental health
problems.
Session 1 focused on making APSs salient and relevant as well
as relevant to social identity: Each group member introduced a
partner in terms of the skills or ability the partner possessed that
would help him or her complete the school year successfully (e.g.,
“well-organized,” “positive attitude”). This provided an initial
example of APSs and social identity as congruent; because all
youth engaged in the task, the metamessage was “we all care about
school.” Session 2 focused on adult PSs: Youth picked photo-
graphs that fit their adult “visions” (PSs). Because most adult PSs
youth describe are images of material success, the metamessage
was “we all want a good future.” Because all youth choose
photographs of adult PSs, successful PSs and social identity are
presented as congruent.
In Session 3, students drew role models and negative forces—
people or things that provide energy to work toward PSs and those
that are draining or nay-saying. The metamessage was “everyone
faces obstacles and difficulties; this does not make the PSs less
part of the ‘true’ self.” In Session 4, students drew timelines into
the future, including forks in the road and obstacles. The metames-
sage was “everyone has difficulties, and failures and setbacks are
a normal part of timelines and do not mean that APSs are not true
selves.”
Session 5 introduced action goals, allowing students to practice
articulating specific strategies to attain their APSs, further high-
lighting the normativeness of difficulty in attaining APSs. Sessions
6 and 7 focused on PSs and strategies to attain them, building on
previous sessions, with a different concrete medium, poster board,
stickers, and markers. Students chose next year feared and to-be-
expected PSs and linked them with current and possible strategies.
191
Sessions 8 to 10 focused on decoupling difficulty and genuine-
ness explicitly through work in smaller groups on everyday prob-
lems, social problems, academic problems, and the process of
getting to high school graduation, and the metatheme was that all
students care about these issues, that difficulties are normative and
not self-defining. Session 11 cemented the new metacognitive
interpretations by having participants review and critique the ses-
sions. Two follow-up sessions included parent/guardians and com-
munity members, with the goal of helping youth broker their APSs
in the community outside of school in ways that link APSs with
social identities.
Hypotheses
Two key hypotheses follow from our model. (a) The interven-
tion will directly influence PSs, self-regulatory behaviors, aca-
demic outcomes, and risk of depression. (b) Effects of the inter-
vention on self-regulatory behaviors, academic outcomes, and
depression will be mediated by intervention effects on PSs. That is,
shift in PSs will improve engagement in self-regulatory behaviors,
and engagement in self-regulatory behaviors will increase aca-
demic success and reduce feelings of depression. We also hypoth-
esized (c) that the intervention would influence the relationship
between APSs and social identity, creating conditions for APSs
and social identity to feel congruent. To test these hypotheses, we
examined the direct effects of the intervention on change in PSs
and the longitudinal effects of the intervention, both direct and as
mediated by PSs. Finally, we examined the long-term effects of the
intervention on the relationship between APSs and social identity
as operationalized by racial-ethnic identity.
Method
Sample
Data were collected at three Detroit middle schools; 71.6% of students
were African American, 17.4% were Latino, and 11.0% were White; total
N for analyses ⫽ 264 as detailed below. Students were low-income, as
evidenced by school lunch program (two-thirds received free or reduced
lunch), neighborhood (in the students’ census tracts, n ⫽ 105, 54.1% of
households were below the poverty line and only 43.4% of adults were
employed; U.S. Bureau of Census, 2000), housing stability (45.2% moved
at least once over the 2 years we tracked them; those who moved averaged
1.54 moves), and school stability (20% changed schools during the 1st
year).
Procedure
Obtaining permission. School district and school principals approved
random assignment and provision of the intervention during the elective
hour. School-based data collection in low-income and minority populations
typically is hampered by low consent rates (Ellickson & Hawes, 1989).
Parents who are lower income and have more academically at-risk students
are less likely to return consent forms, reducing generalizability of findings
(Kearney, Hopkins, Mauss, & Weisheit, 1983; Pokorny, Jason, Schoeny,
Townsend, & Curie, 2001). Because not returning a consent form mailed
home (the standard protocol) is not the same as refusing participation when
asked directly, these authors argue for passive rather than signed consent to
protect generalizability of findings. We also obtained the typical low
response rate with the standard mailing procedure (15% return rate). Rather
than limit generalizability, we followed up the 85% nonresponders with
personal contact (by phone if they had one or by going to the home) to
192 OYSERMAN, BYBEE, AND TERRY
explain the study and what we would do with the data and to ask for
permission or refusal. Together, the initial mailing and direct communica-
tion were well received, resulting in a total signed consent rate of 94.3%
and no differences by race/ethnicity or condition.
Data collection. Core subject teachers (blind to experimental assign-
ment) assessed student in-class behavior. Test scores, GPA, and attendance
were obtained from school records. Student surveys were administered
in-class. Teacher and student baseline data collection occurred prior to the
intervention. Teacher and student postintervention data were collected at
three additional times, at the end of the 8th grade school year and again in
the fall and spring of the following year. School records were obtained each
time report cards were issued over the 2-year data collection period (1st
quarter of 8th grade through the end of 9th grade). Because the intervention
occurred prior to 1st quarter 8th grade grades, the youth baseline survey
included self-report of grades to provide a prior academic attainment
control variable. Teachers were not present for student surveys to assure
confidentiality. Instructions were read aloud, and research staff circulated
to assist students. In-home interviews were completed in year 1 for the 56
students suspended (n ⫽ 11), expelled (n ⫽ 9), transferred (n ⫽ 10), or
otherwise not in school (n ⫽ 26). By year 2, students were enrolled in 80
schools; the at-home interview procedure was followed when in-school
survey completion was not feasible. Teachers were reimbursed $5 for each
assessment ($10 in year 2); students were reimbursed $5 (year 2 only).
Missing data. Extensive tracking efforts minimized data loss. We
obtained data from at least one source across all four semesters of data
collection for a full 96% of our sample, 89.8% to 97.3% of student, 92.4%
to 95.8% of school records, and 83.3% to 96.6% of teacher ratings. All but
4 youth had data from at least one source at semester 4. To preserve
generalizability, expectation maximization methods were used to estimate
the approximately 7.9% of the data matrix that were missing due to skipped
items or unavailable information. Although missingness was not com-
pletely random, Little’s missing completely at random ␹2(df ⫽ 16,115) ⫽
30,685.46, p ⬍ .001, there was no evidence that it was not ignorable (Little
& Rubin, 2002). All results reported use estimated data.
Experimental manipulation. In each homeroom, approximately one-
half of students were randomly assigned to attend the regular elective
period (control group), and one-half were assigned to attend the interven-
tion as the elective period (experimental group). We called the intervention
School-to-Jobs (STJ) and the overall project Pathways for Youth to prevent
any potential stigmatization. Students who enrolled in school late (n ⫽ 3)
or had an elective period broken by lunch (n ⫽ 34) were not randomized,
resulting in an effective cohort of 280. Analyses involve the 264 students
with signed parental consent (n ⫽ 141 experimental, n ⫽ 123 control; n ⫽
140 female; n ⫽ 191 African American, n ⫽ 45 Latino, n ⫽ 28 White). The
STJ intervention thumbnail sketched in the introduction is fully manualized
and includes a detailed fidelity assessment protocol (both available from
the first author). STJ was provided twice weekly over a 7-week period (due
to elective periods missed for school half days) with the two supplemental
parent-youth STJ sessions provided the following evenings or weekend.
Groups averaged 12 participants. Of youth in school, attendance was 80%
to 90% for the youth in-school sessions, and 40% attended at least one
parent-youth supplemental session. Two trainers ran each group. Fidelity to
protocol was maintained via in vivo rating and weekly staff meetings.
Trainers and observers were female Detroiters with previous experience
who received intensive structured training in the protocol (for trainers, 39.5
hours; for observers, 69.5 hours).
Verification of random assignment. Randomization was successful.
Comparison of the experimental and control groups on the 13 baseline
measures showed no significant differences on the multivariate, F(13,
250) ⫽ 1.07, p ⫽ .39, or on any of the univariate Fs.
Measures
Demographics. Schools provided gender and birthdates; youths pro-
vided their race/ethnic description.
PSs. PSs were assessed in 8th grade and spring of 9th grade using the
standard open-ended format (available at http://www.sitemaker.umich.edu/
culture.self/files/possible_selves_measure.doc, 07–22-2003, Oyserman &
Saltz, 1993; Oyserman, Terry, & Bybee, 2002). Youth generated expected
PSs, noted strategies to attain each PS, and repeated this process for feared
PSs. PSs were content coded following Oyserman and Markus (1990) by
five coders with access only to PS responses. All PSs were double coded,
and interrater reliability was 94% (disagreements discussed to agreement).
Analyses focused on the two most commonly generated PSs, academic PSs
and feared off-track PSs. At baseline, 96.6% of youth generated an ex-
pected APS, 63.6% generated a feared APS, and 51.1% generated a feared
off-track PSs.
Feared off-track PSs. These PSs focused on involvement in gangs or
violence (30% of responses), drugs (30%), delinquency or involvement
with the police (30%), and becoming pregnant and other “status” offenses
(10%), fall 8th grade, M ⫽ 0.74, SD ⫽ .90; spring 8th grade, M ⫽ .85,
SD ⫽ .83; spring 9th grade, M ⫽ 1.41, SD ⫽ 1.04. Raw counts were
skewed so they were transformed into change scores (spring-fall), improv-
ing distribution and focusing on the impact of the intervention on change
in PSs during 8th grade; change scores are appropriate when groups are
randomly assigned and variances similar across observations (Cribbie &
Jamieson, 2000; Maris, 1998). Change in feared off-track PSs ranged from
⫺3 to ⫹ 3, M ⫽ 0.11, SD ⫽ 1.04.
APS balance. Balance (the number of pairs of expected and feared
APSs, e.g., expecting to “pass the 8th grade” while wanting to avoid
“failing and having to be an 8th grader again”) was coded following
Oyserman and Markus (1990). About one-third of youth had at least one
balanced APS pair, fall 8th grade, M ⫽ 0.39, SD ⫽ .56; spring 8th grade,
M ⫽ .40, SD ⫽ .57; spring 9th grade, M ⫽ .30, SD ⫽ .59. Raw counts were
skewed so they were transformed into change scores (spring-fall), improv-
ing distribution, and focusing on the impact of the intervention on change
in PSs during 8th grade; change in APS balance ranged from ⫺2 to ⫹ 3,
M ⫽ 0.01, SD ⫽ 0.69. For moderated regression analysis (Hypothesis 3),
9th grade APS balance scores were log-transformed (raw spring 9th grade
scores had skew of 2.5 and kurtosis of 8.47), reducing skew and kurtosis
to ⬍1.5.
APS plausibility. The set of expected and feared APSs and strategies
was coded using a 6-point scale (0 ⫽ no APSs to 5 ⫽ at least 4 APSs and
APS strategies) (Oyserman et al., 2004). Youth averaged about two APSs
and strategies, fall 8th grade, M ⫽ 2.26, SD ⫽ 1.30; spring 8th grade, M ⫽
2.08, SD ⫽ 1.25; spring 9th grade, M ⫽ 2.18, SD ⫽ 1.26. APS plausibility
was not skewed, so log-transformation was not necessary, and intervention
effects were assessed by examining plausibility in spring 8th grade. The
coding manual and detailed instructions are found on http://sitemaker
.umich.edu/daphna.oyserman/files/plausibility_instructions_only012903.doc.
Social Identity
Social identity was operationalized with a 4-item, 5-point Likert re-
sponse (1 ⫽ strongly disagree, 5 ⫽ strongly agree) in-group connectedness
scale, spring 9th grade, M ⫽ 4.18, SD ⫽ .59, ␣ ⫽ .76. Students filled in
their main racial-ethnic group and then responded. For the group “African
American,” example items are “I feel close to African Americans,” “It is
important to me to think of myself as an African American,” and “I feel a
part of the African American community.” Analysis of connectedness to
racial-ethnic in-group excluded the n ⫽ 28 White students because “White
identity” is unlikely to have the same meaning as minority racial-ethnic
identity.
Self-Regulatory Behavior
Time spent doing homework. In 8th grade, we asked “How many hours
a week do you usually spend doing homework?” Open ended responses
ranged from 0 to 20 hours and were log-transformed to reduce skew. On an
 POSSIBLE SELVES AND ACADEMIC OUTCOMES
8-point closed ended follow-up question (0 ⫽ no time to 7 ⫽ more than 10
hours), fall M ⫽ 3.89 (SD ⫽ 1.77) and spring M ⫽ 3.89 (SD ⫽ 1.95), mean
responses reflected about 2 hours a week. In 9th grade, we used a 1-week
event history calendar (Belli, 1998); again, students reported about 2 hours
of weekly homework, fall M ⫽ 160.20 minutes, SD ⫽ 151.89; spring M ⫽
112.15 minutes, SD ⫽ 111.28. Homework time was low; as a benchmark,
national daily diary report data for 12 to 14 year olds are 4.40 hours weekly
(Juster, Ono, & Stafford, 2004).
Disruptive behavior. Eighth grade youth reported “How often does the
teacher make you leave the classroom because of your behavior?” on a
7-point scale (1 ⫽ never, 2 ⫽ once or twice a year, 3 ⫽ less than once a
month, 4 ⫽ once a month or so, 5 ⫽ once every few weeks, 6 ⫽ once a
week or so, 7 ⫽ more than once a week), fall M ⫽ 2.18, SD ⫽ 1.55; spring
M ⫽ 2.64, SD ⫽ 1.94. Teachers reported on youth behavior with a 4-item
5-point response (1 ⫽ never, 5 ⫽ always) (Finn Disruptive Behavior Scale,
Finn, Pannozzo, & Voelkl, 1995; 8th grade revision, J. Finn, personal
communication, October 14, 1998) scale each semester, ␣ ⫽ 79, .81, .78,
and .80, respectively. Beginning with the stem “This student . . . ,” items
were “annoys peers or interferes with peers’ work,” “is critical of peers
who do well in school,” “needs to be reprimanded or sent to the office,”
and “is verbally or physically abusive to the teacher.”
Initiative-taking behavior. Teachers reported on youth behavior with a
4-item 5-point response (1 ⫽ never, 5 ⫽ always) (Finn Initiative Scale,
Finn et al., 1995; 8th grade revision, J. Finn, personal communication,
October 14, 1998) scale each semester, ␣ ⫽ .75, .82, .82, and .85,
respectively. Beginning with the stem “How often does this student . . . ,”
items were “do more than the work assigned?,” “persist when confronted
with difficult problems?,” “actively participate in class discussions?,” and
“engage me in conversation about subject matter before or after school or
outside of class?”
Absences. Eighth grade youth reported “How often are you absent
from school or miss a class during the day?” on a 7-point scale (1 ⫽ never,
2 ⫽ once or twice a year, 3 ⫽ less than once a month, 4 ⫽ once a month
or so, 5 ⫽ once every few weeks, 6 ⫽ once a week or so, 7 ⫽ more than
once a week), fall M ⫽ 3.17, SD ⫽ 1.67; spring M ⫽ 3.82, SD ⫽ 1.79. As
a second measure of absences, a count of unexcused absences was obtained
from school records in 8th and 9th grades (log transformed to reduce
skew). Each day not enrolled in any school was counted as an absence.
Academic Outcomes
GPA. School record of core (math, history, science, English) grades
(0.0 ⫽ F to 4.0 ⫽ A) was obtained; 0.0 was assigned as GPA for students
not enrolled in any school during the semester. As a baseline measure of
grades prior to the intervention, we obtained youth report of prior grades.
Three additional academic outcome measures were obtained at the end
of 8th grade: (a) proportion of subject tests passed on the 8th grade
Essential Skills Attainment Test (based on the Michigan Educational As-
sessment Program (MEAP) test), M ⫽ .80, SD ⫽ .25, range ⫽ 0.0 to 1.0),
(b) referral to remedial summer school as a prerequisite to high school
promotion (47% of youth), and (c) retained in 8th grade according to
school records by the end of the summer (9.5% of students). These
measures were not available in 9th grade (standardized tests are not
administered each year); therefore, 9th grade academic outcomes analyses
focus on GPA only.
Depression
Youth report of depression, the standard 20-item, 4-point Center for
Epidemiological Studies Depression Scale (Radloff, 1977), anchored at
0 ⫽ not at all or less than one day in the past week and 3 ⫽ 5 to 7 days
in the past week, was obtained spring of 9th grade (M ⫽ 11.35, SD ⫽ 8.33).
Items include affective (e.g., “I felt depressed”) and somatic (e.g., “I did
not feel like eating; my appetite was poor”) aspects of depression. Propor-
193
tion of youth meeting the clinical threshold for depression (sum scores of
at least 16) (Radloff & Locke, 2000) was 22.7%, comparable with the
proportion of youth reaching the threshold for depression in national
surveys (Costello, Mustillo, & Erkanli, 2003; Roberts, Attkisson, & Rosen-
blatt, 1998).
Results
We first examined the direct effect of the STJ intervention on
PSs, self-regulatory behaviors, academic outcomes, and depres-
sion. Then, we examined PSs as mediators of intervention effects
on self-regulatory behaviors and, via their impact on self-
regulatory behavior, as mediators of intervention effects on aca-
demic outcomes and depression. We also examined whether the
STJ intervention influenced the association between APSs and
social identity. Prior to describing these analyses and results, we
outline our decision to present analyses of both the “intention-to-
treat” (ITT) and compliant sample analyses.
ITT and Compliant Sample Analyses
Following standard procedure, direct effects analyses (Hypoth-
esis 1) were conducted twice, once with the ITT sample and a
second time with the compliant sample. Analyses of meditating
effects of PSs (Hypothesis 2) and of the impact of the intervention
on the relationship between APSs and social identity (Hypothesis
3) utilized the compliant sample. The ITT sample included all 264
youth randomized into experimental or control conditions, whether
they were in school to receive the intervention or not.
Although analysis based on an ITT sample preserves random
assignment, it is not a good test of the intervention effects to the
extent that some people who did not receive the intervention are
included in the treatment group for analyses. The alternative is to
compare those who received a meaningful dose of the intervention
with those who did not. This is termed compliance analyses. Use
of compliance analyses to estimate effects for individuals who
received a meaningful dose of an intervention is appropriate if
assessed variables provide a useful and valid estimate of those
likely to participate and if nonparticipators among those assigned
to the experimental condition do not differ in outcomes from
control condition youth (Jo, 2002a, 2002b). In the current study,
both requirements were met, as outlined in more detail at http://
www.sitemaker.umich.edu/culture.self/files/ appendix_a_on_web_
site.doc, 26.0 kb, the compliant sample (n ⫽ 228) included the
“participating” youth (n ⫽ 116) and a comparable sample of
control youth (n ⫽ 112). Participating youth were assigned to the
experimental condition and received a reasonable dose of the
intervention (i.e., attended at least five sessions). The comparable
sample of control youth (n ⫽ 112) were youth who had a similar
unobserved likelihood of “participating” had they been random-
ized to intervention according to the expectation maximization
compliance model incorporating the full set of preintervention
variables (Little & Yau, 1998). Most of the difference in sample
size between the ITT and compliant analyses was due to youth
being suspended or expelled within the 1st month of school; some
youth simply were not in school to receive the intervention. Effects
presented are from the compliant (participating) sample unless
otherwise specified.
194 OYSERMAN, BYBEE, AND TERRY

Hypothesis 1: STJ Intervention Has Direct Effects on PSs,
Self-Regulatory Behavior, and Outcomes
Analysis Plan
We used multilevel modeling (MLM; Raudenbush & Bryk,
2002; Snijders & Bosker, 1999) to appropriately analyze variance
at three levels—time nested within students and students nested
within homerooms. We used standard methods for MLM, first
building baseline models containing random effects for homeroom
and time and then adding condition assignment and baseline con-
trol variables to optimize statistical power by accounting for pre-
intervention variability in sex, race, age, baseline GPA (student
report), baseline behavioral problems (teacher report), and baseline
assessment of each dependent measure. For ease of interpretation,
control variables (other than baseline assessment of the dependent
variable) were constant across analyses. We show direct effects at
the end of 8th grade and over time (through the end of 9th grade).
For the longitudinal models predicting change in self-regulatory
behavior and academic outcomes, we set the intercept for time at
spring of 9th grade to obtain estimates of between-condition ef-
fects at the final measurement point. We used Raudenbush and
Liu’s (2000, 2001) methods to compute standardized effect sizes;
these effects are calibrated to the residual level 1 variance of the
unconditional model (i.e., the model containing no predictors other
than homeroom) for 8th Grade MLM analyses and to the random
slope and intercept variances of the unconditional model (i.e., the
model with no predictors other than time and homeroom) for the
longitudinal MLM. Standardized effect sizes are interpreted like
Cohen’s d.
STJ Effects by End of 8th Grade
Effects of STJ for the total ITT sample and the compliant
(participating) sample are presented in Table 1. For each depen-
dent variable, the raw coefficient is interpreted as a regression
weight and indicates, in the metric of the dependent variable, the
average within-homeroom difference between students assigned to
STJ rather than the control group. The standardized effect size is
interpreted as Cohen’s d and expresses the same difference in SD
units. To further facilitate interpretation of intervention effects, esti-
mated means for spring 8th grade are in the first columns of Table 2.
We summarize findings by dependent variable in the sections below.
PSs. Following Hypothesis 1, both ITT and compliant sample
analyses showed significant intervention effects on PSs in the
spring of 8th grade. Experimental youth generated more balanced
APSs, more plausible APSs, and more feared off-track PSs, setting
the stage for self-regulation.
Self-regulatory behavior. As can be seen in the negative co-
efficients connoting fewer absences, participation in the STJ in-
tervention significantly reduced absences, whether assessed via

Table 1
Multilevel Models: Effects of the STJ Intervention in Spring 8th Grade

Total ITT sample Participating samplea

(N ⫽ 264) (N ⫽ 228)

Raw coefficients Raw coefficients

Standardized
Dependent variables B p B p effect sizeb

Change in APS balance .157 .023 .186 .014 .269

Change in feared off-track PS .286 .004 .308 .003 .302
APS plausibility .340 .015 .400 .008 .329
Absences (student report) ⫺.417 .043 ⫺.457 .036 ⫺.267
Unexcused absences (school records)c ⫺.183 .049 ⫺.202 .022 ⫺.726
Homework time (hrs/week; open ended)c .081 .090 .099 .135 .155
Homework time (hrs/week; closed ended) .263 .229 .406 .079 .240
Disruptive behavior (student report) ⫺.307 .165 ⫺.380 .094 ⫺.209
Disruptive behavior (teacher report) ⫺.099 .261 ⫺.139 .149 ⫺.172
In-class initiative (teacher report) .179 .043 .142 .142 .170
Core academic GPA (school records) .229 .021 .218 .039 .252
Standardized tests (school records; proportion passed) .068 .017 .086 .006 .360
Referral to remedial summer schoold ⫺.194 .457 ⫺.419 .140
Retention in 8th graded ⫺.598 .164 ⫺.979 .076

Note. The cross-sectional multilevel models analyzed variance at two levels: students (Level 1), nested within
homerooms (Level 2). All Level 1 intercepts were random, and all modeled fixed effects were at Level 1
(student). Models controlled for the following Level 1 (student) covariates: race, sex, baseline age, preinter-
vention GPA (student-report), preintervention behavioral problems (teacher report), and preintervention assess-
ment of the dependent variable, when available.
a
The participating sample were 116 students who attended at least five in-school STJ sessions plus 112 students
in the control group who had similar likelihood of attending, had they been randomly assigned to the
intervention, according to maximum likelihood (EM) modeling of intervention compliance. b Interpretation is
similar to d for STJ intervention effects on individuals within homerooms. ES ⫽ B/V⫺1 p , where B ⫽ raw
coefficient and Vp ⫽ residual variance in the unconditional intercept model, following Raudenbush & Liu
c
(2001). These positively skewed dependent variables were log-transformed to reduce the influence of
extreme scores. d Multilevel models for these dichotomous dependent variables used a logit link function.
Standardized effect sizes could not be computed for logit models.
 POSSIBLE SELVES AND ACADEMIC OUTCOMES 195

Table 2
Multilevel Models: Estimated Means in Experimental and Control Conditions: Spring 8th and
9th Grades

Estimated condition means/percentages

Spring of 8th grade Spring of 9th grade

STJ Control STJ Control

Dependent variables (n ⫽ 116) (n ⫽ 112) (n ⫽ 116) (n ⫽ 112)

Change in APS balance 0.07 ⫺0.12

Change in feared off-track PS 0.31 0.00
APS plausibility 2.36 1.96
Absences (student report) 3.49 3.95
Unexcused absences (school records)b 12.28 14.53 22.52 24.77
Homework time (hours/week; open ended)b 3.49 3.28 2.51 1.57
Homework time (hours/week; closed ended) 4.15 3.74
Disruptive behavior (student report) 2.37 2.63
Disruptive behavior (teacher report) 1.74 1.83 1.55 1.73
In-class initiative (teacher report) 2.51 2.44 2.48 2.25
Core academic GPA (school records) 1.98 1.83 1.64 1.36
Standardized tests (school records; proportion passed) 0.83 0.77
Referral to remedial summer school 36.2% 48.2%
Retention in 8th grade 4.3% 10.7%
Depression (CESD—spring of 9th grade)c 10.35 12.29

Note. N ⫽ 228 youth in the participating sample.

a
For variables with estimated means for both 8th and 9th grades, estimates were from the intercept term of a
three-level longitudinal multilevel model (time, students, homerooms); for variables with estimated means for
only one grade, estimates were from cross-sectional two level multilevel models (students in homerooms).
b
These positively skewed variables were log transformed for analysis; however, to facilitate interpretation,
means are presented here in the original metric (i.e., in exponentiated form). c Depression was measured only
in the spring of 9th grade; means were estimated by a two-level cross-sectional multilevel model.

school or youth report, in both ITT and compliant samples. For
school-reported absences, the effect of the intervention is in Co-
hen’s large range, with d near .80. Each semester intervention
youth attended more school, over 2 additional days, than control
youth. With regard to time spent in homework and behavior in
class, results are suggestive of STJ impact (for homework, student
and teacher rated behavior, all p ⬍ .15).
Academic outcomes. Intervention youth showed significantly
better GPA and better standardized test scores using both ITT and
compliance analyses. With regard to 8th grade retention, more than
twice as many control youths (10.7%) as intervention youths
(4.3%) were retained, p ⬍ .10).
STJ Effects by End of 9th Grade
Estimated means for the spring of 9th grade are in Table 2.
Table 3 presents the direct effects of the STJ intervention on the
trajectory of change over time and on the difference between
intervention and control groups in the spring of 9th grade. For each
dependent variable, the raw slope coefficient indicates the average
within-homeroom difference in linear trajectory between students
assigned to STJ and the control group. The raw intercept coeffi-
cient indicates the average spring 9th grade within-homeroom
difference between students assigned to STJ and control group.
Standardized effect sizes are interpreted as Cohen’s d and express
the same differences in SD units. The trajectories of change for each
dependent variable are presented graphically in Figures 2 and 3.
Self-regulatory behavior. STJ youth spent significantly more
time doing homework, were more likely to take initiative in class,
less disruptive, and less likely to skip class than control group
youth. As displayed in Figure 2a, time spent on homework de-
clined for all youth, but the decline was significantly less for STJ
youth. By spring of 9th grade, STJ youth were spending on
average 2.51 hours a week on homework, nearly an hour more per
week than control youth who averaged 1.57 hours per week. Both
ITT and compliant sample analyses showed a large and significant
effect of STJ on both level and trajectory of change over time in
homework.
Trajectories of in-class behavior are displayed in Figure 2b
(initiative taking) and 2c (disruptive behavior). As can be seen,
initiative-taking declined for control youth, not for STJ youth;
disruptive behavior declined for all youth, but the decline was
significantly steeper for STJ youth. Size of intervention effects is
moderate for the trajectory of change in initiative-taking and large
for the trajectory of change in disruption over time.
With regard to absences (skipping), as displayed in Figure 3a,
the large and significant effect of the intervention at the end of 8th
grade remained stable and significant through 9th grade. Each
semester of 9th grade, STJ youth averaged 2.25 more days in
school than control youth.
Academic outcomes. STJ and control youth differed signifi-
cantly in GPA by the end of 9th grade, with the size of gap
between groups increasing over time, as displayed graphically in
Figure 3b; intervention effect sizes for both end of 9th grade level
and trajectory over time were “small-to-moderate” using Cohen’s
(1988) rules of thumb. Within-time models comparing the STJ and
control groups at each grading period showed that a significant
196 OYSERMAN, BYBEE, AND TERRY

Table 3
Longitudinal Multilevel Models: Effects of STJ Intervention on Change (Slope) and Level
(Spring of 9th Grade) of Self-Regulatory Behavior, Academic Outcomes, and Depression

Total ITT sample Participating samplea

(N ⫽ 264) (N ⫽ 228)

Raw coefficients Raw coefficients

Standardized
Dependent variables B p B p effect sizeb

Homework time (open ended)c

Spring 9th grade (intercept) 0.166 .043 0.183 .035 0.742
Change trajectory (linear slope) 0.011 .050 0.012 .041 1.044
Teacher-report of in-class initiative
Spring 9th grade (intercept) 0.166 .115 0.234 .037 0.326
Change trajectory (linear slope) 0.012 .073 0.015 .036 0.430
Teacher-report disruptive behavior
Spring 9th grade (intercept) 0.085 .283 ⫺0.178 .030 ⫺0.326
Change trajectory (linear slope) ⫺0.005 .366 ⫺0.010 .040 ⫺0.776
School record of unexcused absencesc
Spring 9th grade (intercept) ⫺0.144 .051 ⫺0.175 .027 ⫺0.296
Change trajectory (linear slope) 0.000 .994 0.001 .847 0.000
School record of core academic GPA
Spring 9th grade (intercept) 0.211 .074 0.277 .031 0.300
Change trajectory (linear slope) 0.164 .266 0.030 .055 0.354
Depression (CESD)d
Spring 9th grade (intercept) ⫺1.355 .187 ⫺1.943 .049 ⫺0.257

Note. The longitudinal multilevel models analyzed variance at three levels: time (Level 1) nested within
students (Level 2), nested within homerooms (Level 3). All Level 2 intercepts and slopes were random. Models
controlled for the following Level 2 (student) covariates: race, sex, baseline age, and preintervention GPA
(student-report).
a
The participating sample was 116 students who attended at least five in-school STJ sessions plus 112 students
in the control group who had similar likelihood of attending, had they been randomly assigned to the
intervention, according to maximum likelihood (EM) modeling of intervention compliance. b Interpretation is
similar to d. For STJ intervention effects on the intercept, ES ⫽ B/T⫺1
00 , where B ⫽ raw coefficient and T00 ⫽
intercept variance in the unconditional change model. For STJ effects on change in linear slopes, ES ⫽ B/T⫺111 ,
where B ⫽ raw coefficient and T11 ⫽ variance of the individual linear change trajectories in the unconditional
change model. For STJ effects on individuals within homerooms (for depression in spring of 9th grade), ES ⫽
B/V⫺1p , where B ⫽ raw coefficient and Vp ⫽ residual variance in the unconditional intercept model (Raudenbush
& Liu, 2001). c These positively skewed dependent variables were log-transformed to reduce the influence of
extreme scores. d Depression was measured only in spring of 9th grade; coefficients are from a cross-sectional,
two-level MLM.

difference in GPA emerged at the 3rd quarter, 2 quarters after the
intervention (B ⫽ .22, p ⬍ .02). GPA between groups continued to
diverge over time. By the end of the 9th grade, estimated average
GPA was 1.64 for STJ youth, compared with 1.36 for control
group youth.
Depression. Intervention effects on depression were signifi-
cant; estimated mean Center for Epidemiological Studies Depres-
sion Scale scores were nearly 2 points lower for STJ (M ⫽ 10.35)
compared with control youth (M ⫽ 12.29) at the end of 9th grade.
Hypothesis 2: PSs Mediate STJ Effects on Self-Regulatory
Behavior, Academic Outcomes, and Depression
Analysis Plan
Having found support for hypothesis 1 (significant direct effects
of STJ on PSs, self-regulatory behavior, academic outcomes, and
depression) allowed for examination of hypothesis 2 (mediating
effects of PSs). To test for mediation, we used latent structural
equation modeling, employing standard methods to develop the
measurement model defining the latent constructs, to assess overall
model fit, and to calculate and test indirect effects. All SEM
analyses used AMOS 5 software (Arbuckle & Wothke, 2003).
Figure 4 displays the structural model of the hypothesized direct
and indirect effects including four observed constructs (8th grade
APS balance, APS plausibility, feared off-track PSs, and 9th Grade
GPA) and four latent constructs (attendance, homework, class-
room behavioral problems, and depression). The measurement
model was developed to test associations between latent constructs
and the observed indicators that measure them. As presented in
Table 4, all loadings were above .46 (all p ⬍ .01). For the
multiindicator latent constructs (all but depression), measurement
error was reflected in the loadings linking observed indicators to
latent constructs. For the single-indicator latent construct (the
multiitem depression scale), measurement error was reflected by
weighting the scale error variance by 1 ⫺ Cronbach’s alpha for the
depression scale (1 ⫺ .84 ⫽ .16). Following standard practice, we
allowed correlated errors among those indicator variables that
shared measurement variance due to common source and item
format. These were closed-ended student-report of attendance,
homework, and classroom behavioral problems (error correlations
 POSSIBLE SELVES AND ACADEMIC OUTCOMES 197

Figure 2. Effect of the STJ intervention over time on hours spent on homework each week (a), teacher-rated
initiative (b), and teacher-rated disruptive behavior (c).

ranged from r ⫽ .12 to r ⫽ .26). Correlations among measurement
errors of other indicators were fixed at zero.
The structural model of hypothesized direct and indirect effects
was applied to the latent constructs defined in the measurement
model. Single-headed arrows in Figure 4 indicate the directional
conceptualization of the structural model: The influence of the
intervention on PSs, self-regulatory behavior, academic outcomes,
and depression. Associations among 8th grade measures of self-
regulatory behavior (homework effort, classroom behavior prob-
lems, and absences) were specified as nondirectional correlations,
indicated by curved, double-headed arrows. Multivariate normality
of the full model was acceptable: Mardia’s (1985) test of multi-
variate kurtosis ⫽ 10.57, with univariate skew and kurtosis less
than 1.2 for all endogenous variables. Fit of the full structural
model to the data was assessed by ␹2, supplemented by tests and
critical thresholds recommended by Hu and Bentler (1999): stan-
dardized root mean residual ⱕ .08, root mean square error of
approximation (RMSEA) ⱕ .06, and Bollen’s incremental fit
index ⱖ .95. With the sample size of 228 and df exceeding 100,
power exceeds .80 for a test of overall model fit using RMSEA
(MacCallum, Browne, & Sugawara, 1996). By current criteria, our
sample size of 228 is adequate for stable and unbiased estimation
of indirect effects and SEs (e.g., MacKinnon, Warsi, & Dwyer,
1995). Bootstrapped, bias-corrected SEs were used to test the
significance of indirect effects (MacKinnon, Lockwood, Hoffman,
West, & Sheets, 2002; MacKinnon, Lockwood, & Williams,
2004). Power to test the significance of mediated effects generally
exceeded .80 given sample size, measure reliability, extent of
collinearity between independent and mediating variables, and use
of latent constructs (see Hoyle & Kenny, 1999).
Testing the Full Process Model
The final structural model, shown in Figure 4, was an excellent
fit to the data: ␹2(df ⫽ 146) ⫽ 156.36, p ⫽ .26; standardized root
mean residual ⫽ .05, RMSEA ⫽ .02, incremental fit index ⫽ .99.
198 OYSERMAN, BYBEE, AND TERRY

Figure 3. Effect of the STJ intervention on days absent per semester (a) and core academic grades (b).

The intervention had direct positive effects on PSs as well as a
direct negative effect on school absences (reducing absences).
Each PS variable had significant effects on 8th and 9th grade
outcomes; all mediation paths were significant at p ⬍ .05. Non-
directional correlations among constructs were moderate and an-
ticipated: 8th grade classroom behavioral problems were corre-
lated positively with absences (r ⫽ .50) and negatively with
homework (r ⫽ ⫺.33). Table 5 lists the total and specific medi-
ation paths linking the intervention with 8th and 9th grade effects.
Interpreting the Mediating Effects of PSs
APSs. The STJ intervention had positive direct effects on
APSs (APS balance B ⫽ .14 and APS plausibility, B ⫽ .12), as
well as an indirect effect on APS plausibility through APS balance
(indirect B ⫽ .02). APS balance and APS plausibility were com-
plete mediators of intervention effects on two of the self-regulatory
behaviors in the 8th grade— behavioral problems (indirect B ⫽
⫺.02) and time spent in homework (indirect B ⫽ .03). On 9th

Figure 4. Effects of the STJ intervention over time as mediated by PSs.

 POSSIBLE SELVES AND ACADEMIC OUTCOMES 199

Table 4 sences (indirect B ⫽ ⫺.02). For 9th grade outcomes, off-track PS

Measurement Coefficients for Latent Structural Equation Model partially mediated intervention effects on academic outcomes ⫺
GPA (indirect B ⫽ .01), as well as 9th grade depression (indirect
Standardized B ⫽ ⫺.01).
Latent constructs and observed indicators coefficient

Classroom behavior problems (8th grade) Hypothesis 3: STJ Influences the Relationship Between
Teacher report of disruptive behaviora 0.817
Student report of disruptive behavior 0.465
APS and Social Identity
Teacher report of initiative ⫺0.468
Homework (8th grade)
Analysis Plan
Closed-ended rating of time spent on homework each We used hierarchical multiple regression to examine the mod-
weeka 0.888
Open-ended report of time spent on homework each erating effect of the STJ intervention on the Spring 9th Grade
week 0.764 APS-social identity relationship. Fall 8th Grade APS was entered
Absences (8th grade) as a control variable at Block 1, experimental assignment at Block
School record of unexcused absencesa 0.736 2, Social Identity Connectedness at Block 3, and the Connected-
Student report of absences 0.568
Depression (9th grade)b
ness ⫻ Intervention interaction at Block 4. Methods recommended
CESD 0.910 by Aiken and West (1991) were used to probe interaction effects
and test significance of simple slopes.
Note. N ⫽ 228 youth in the participating sample. All coefficients signif-
icant at p ⬍ .001.
a
These indicators were set to 1 to define the scale of the construct. Effects on APS
b
Measurement error in this latent construct was modeled by weighting the
error variance of the single indicator (the CESD scale score) by APS plausibility. The full model containing main effects and
1-Cronbach’s alpha (⫺.84 ⫽ .16). the STJ ⫻ Social Identity Connectedness interaction was signifi-
cant, F(4, 223) ⫽ 5.02, p ⬍ .001, with added variance explained
by the STJ ⫻ Social Identity Connectedness interaction, B ⫽ .57,
grade academic outcomes, APS balance and APS plausibility ⌬F(1, 223) ⫽ 3.78, p ⫽ .05. For intervention youth, Connected-
partially mediated intervention effects on GPA (indirect B ⫽ .03). ness and APS plausibility were positively associated, simple slope
APS balance and plausibility also partially mediated intervention B ⫽ .80, p ⬍ .001, but for control youth, there was no significant
effects on 9th grade depression (indirect B ⫽ ⫺.01). relationship, simple slope B ⫽ .22, p ⫽ .31.
Feared off-track PSs. The intervention had positive direct APS balance. The full model containing main effects and the
effects on feared off-track PSs (B ⫽ .13). Off-track PS was a STJ ⫻ Social Identity Connectedness interaction was marginally
partial mediator of intervention effects on 8th grade school ab- significant, F(4, 223) ⫽ 1 .95, p ⫽ .10, and the STJ ⫻ Social

Table 5
Mediational Paths Linking the STJ Intervention to Spring 8th Grade and Spring 9th Grade Effects

Standardized
General and specific mediational paths effect p

STJ effects on spring 8th grade outcomes

Total STJ effect on classroom behavior problems (8th grade) ⫺0.022 0.023
STJ 3 APS balance 3 APS plausibility 3 classroom behavior problems ⫺0.003 0.034
STJ 3 APS plausibility 3 classroom behavior problems ⫺0.019 0.044
Total STJ effect on homework (8th grade) 0.026 0.015
STJ 3 APS balance 3 APS plausibility 3 homework 0.003 0.036
STJ 3 APS plausibility 3 homework 0.023 0.025
Total STJ effect on absences (8th grade) ⫺0.208 0.002
STJ 3 feared off-track PS 3 absences ⫺0.021 0.027
STJ 3 absences ⫺0.186 0.003
STJ effects on spring 9th grade outcomes
Total STJ effect on GPA (spring 9th grade) 0.090 0.002
STJ 3 APS balance 3 GPA 0.019 0.018
STJ 3 APS balance 3 APS plausibility 3 classroom behavior 3 GPA 0.002 0.028
STJ 3 APS plausibility 3 classroom behavior problems 3 GPA 0.009 0.024
STJ 3 absences 3 GPA 0.054 0.006
STJ 3 feared off-track PS 3 absences 3 GPA 0.006 0.024
Total STJ effect on depression (spring 9th grade) ⫺0.091 0.001
STJ 3 APS balance 3 APS plausibility 3 homework 3 depression ⫺0.001 0.023
STJ 3 APS plausibility 3 homework 3 depression ⫺0.005 0.014
STJ 3 feared off-track PS 3 absences 3 depression ⫺0.009 0.017
STJ 3 absences 3 depression ⫺0.076 0.008

Note. N ⫽ 228 youth in the participating sample. Italicized entries are for general indirect effects; nonitalicized
entries are for specific indirect effect paths.
200 OYSERMAN, BYBEE, AND TERRY
Identity Connectedness interaction, B ⫽ .21, ⌬F(1, 223) ⫽ 6.39,
p ⫽ .01, added significantly to variance explained. For interven-
tion youth, Connectedness and APS balance were positively asso-
ciated, simple slope B ⫽ .13, p ⬍ .03, but for control youth, there
was no significant relationship, simple slope B ⫽ ⫺.08, p ⫽ .20.
Discussion
Although PSs focused on school success are common, so is
failure in school. We asked why PSs fail to produce sustained
self-regulation, outlining a process model in which self-regulation
and, therefore behavior change, is likely when PSs are linked with
contextually salient strategies, when they are congruent with social
identity, and when difficulty is understood as a normative part of
the process of attaining PSs. Focusing explicitly on low-income
and minority youth, we developed an intervention to link PSs and
strategies, to incorporate difficulty as a normative part of pursuing
PSs, and to facilitate a positive link between APSs and important
social identities. Our basic premises were that self-concept is not
monolithic, that PSs are differentially accessible, and that they are
likely to influence behaviors only when linked to strategies, when
experienced as compatible with social identity, and when difficulty
working toward the PS is viewed as normative.
To ensure that the intervention effects would be sustained over
time, the intervention occurred in school, with peers, during the
school day and targeted each aspect of our process model. Struc-
tured group activities evoked academically focused PSs, made
clear that APSs were held by peers (and therefore something that
“we” aspire to), and highlighted the normativeness of difficulties
and failures along the way to attaining PS goals. Thus, the inter-
vention operationalized our theory of how PSs might influence
self-regulatory behaviors (and through these behaviors, academic
outcomes and well-being).
Although attaining and sustaining an intervention effect is no-
toriously difficult, we documented effects that were stable and
even increasing over time. This sustained effect over 2 years is
particularly impressive, given that the high-poverty neighborhoods
the youth were embedded in and the difficulty of improving
academic outcomes when prior academic attainment accounts for
much of the variance to be explained. We were able to directly
assess each aspect of the process model except the positive self-
regulatory consequences of changing metacognitive experience.
Moreover, we documented effects using two very different ana-
lytic strategies (structural equation modeling and longitudinal
MLM) with different strengths and assumptions.
Structural equation modeling provided explicit test of hypothe-
sized mediation paths via PSs to self-regulatory behavior to be-
havioral outcomes, using multiple sources of information to ac-
count for measurement error. MLM, by allowing us to look
directly at intervention effects on trajectories of individual change
as well as at specific points, provided an explicit test of the
hypothesized effects over time. MLM documented that interven-
tion effects on self-regulatory behavior (attendance, homework
time, and behavior problems) and academic outcomes were stable
or increased over our 2-year follow-up, while accounting for the
nesting of students within homerooms.
Our focus on PSs of low-income and minority youths and their
fit with social identity as working class, African American, or
Latino was intentional. We chose PSs and social identity because
these social identities are often assumed to be at odds with aca-
demic self-goals (Steinitz & Solomon, 1986), and school failure
has important consequences, increasing risk of depression and
delinquency (Joseph, 1996; Kasen, Cohen, & Brook, 1998), future
unemployment, problems in parenting, and mental health problems
(e.g., Stoep, Weiss, Kuo, Cheney, & Cohen, 2003) and making
successful completion of other developmental tasks less likely
(Sandler & Chassin, 2002). Low SES is a clear risk factor for
school failure (Blair, Blair, & Madamba, 1999; Orfield et al.,
2004); the combination of minority race/ethnicity and low social
class together account for about 19% of variance in academic
attainment (McDermott, 1995). Improving outcomes in this high-
risk group is both critical and difficult given that poverty and
minority status are “fixed-risk” factors, unlikely to change dramat-
ically over the student’s middle and high school years.
As a test of our model, we chose to intervene in the fall of the
final year of middle school. We chose this point both because the
transition from middle to high school is associated with increased
salience of important social class and racial ethnic social identities
and because school failure rates begin to rise dramatically from the
1st year of high school (e.g., Seidman, LaRue, Aber, Lawrence, &
Feinman, 1994). The transition to high school is stressing, as can
be seen in the outcomes for control group youth; without the
intervention, self-regulatory behavior declines, academic out-
comes erode, and risk of depression is higher. An intervention that
bolsters APSs and behavioral engagement will improve outcomes
in the last year of middle school, making the transition to high
school less risky. In addition, middle schools are on average
smaller and more homogeneous than high schools, making inter-
vention focused on linking PSs and social identity simpler.
Youth who are not equipped with specific PSs that make en-
gagement in school (attending, putting effort into homework,
engaging teachers) a necessity are more likely to fall behind in
high school simply because high schools are more impersonal and
less likely to routinely fulfill their educational needs. Indeed, the
transition to high school has been described as a consciousness-
raising experience due to the influx of students of diverse racial-
ethnic and economic backgrounds (Seidman, Aber, Allen, &
French, 1996). Ethnographic accounts suggest that low-SES teens
do not realize that they are poor in homogeneously low-SES
schools and neighborhoods where they appear average (Roderick,
2003; Steinitz & Solomon, 1986). Low-SES students report learn-
ing that they are part of a low-SES group when they enter high
school., making the transition to high school very stressing for
these youth. Not surprisingly, racial-ethnic identity also becomes
more salient in the transition (Altschul, Oyserman, & Bybee, in
press). Not only does the transition to high school raise conscious-
ness of social identities and their relative standing, it also high-
lights ambiguities about the one’s probable future. Schools no
longer provide vocational or technical training, so youths not in
advanced classes are in “general studies,” implying that the point
of school is either college or marking time. For all of these reasons,
transition to high school is characterized by dropping grades and
school involvement for low-SES and minority children (Seidman
et al., 1994).
We documented that our intervention does change PSs, increas-
ing both feared off-track and academically focused PSs and strat-
egies to attain them. As hypothesized, intervention youth both had
more of these PSs and were better able to use them to improve
 POSSIBLE SELVES AND ACADEMIC OUTCOMES
behavioral self-regulation. The intervention produced measurable
change in PSs, and change in PSs predicted change in behavioral
self-regulation— going to school rather than skipping, behaving
and participating in class, and spending time on homework; self-
regulation not only improved academic outcomes but, equally
importantly, reduced risk of depression.
The impact of academically focused PSs on self-regulatory
behaviors was distinct from the impact of feared off-track PSs on
self-regulatory behaviors. Youth with balanced and plausible ac-
ademically focused PSs spent more time doing homework, were
less disruptive, and more behaviorally engaged in class-room
activities. Youth with feared off-track PSs attended school more
(had fewer school absences). The distinct role of feared off-track
PSs is congruent with a number of self-regulatory models: Carver
(2004) describes self-regulation to avoid feared PSs or antigoals as
discrepancy-enlarging self-regulation. Larsen (2004) describes the
self-regulatory system as vigilant to environmental dangers; when
danger is cued, individuals are more cautious about engaging in
behaviors that may increase risk. Higgins (Higgins & Spiegel,
2004) describes prevention-focused self-regulation as risk averse.
Following these perspectives, youth with feared off-track PSs can
be expected to be cautious about risk-increasing behaviors and
likely to engage in action (e.g., attending school) they perceive as
antithetical to their off-track selves of becoming pregnant, in-
volved in drugs, crime, or the police. Indeed, increased feared
off-track PSs reduced risk of school absences. Although vigilant
focus on antigoals or prevention is likely to reduce risk of harm,
active engagement in goal attainment (discrepancy-reducing self-
regulation) is likely to increase chances of success.
Although self-regulatory systems theories posit roles for both
discrepancy-reducing (promotion) and discrepancy-increasing
(prevention) systems, prevention focus, engaging in self-
regulatory behavior to avoid feared PSs, is not particularly prom-
inent in the academic goal literature. Perhaps this is because in
middle-class contexts, pursuit of academic goals may more com-
monly involve the discrepancy reducing feedback system— engag-
ing in self-regulatory behavior to attain positive expected PSs.
There is evidence that college students are more likely to use
promotion- than prevention-focused self-regulation (e.g., Lock-
wood, Sadler, Fyman, & Tuck, 2004). However, prevention fo-
cused self-regulation may become more salient in a number of
circumstances. First, prevention focused self-regulation may be
more likely when social contextual risk is high, such as in circum-
stances of poverty. For example, first generation college students
are more likely to engaging in strategies to avoid feared PSs than
strategies to attain positive PSs (Oyserman, Gant, & Ager, 1995,
Study 1). Ethnographic evidence from low-income high school
students also highlights the salience of feared off-track PSs—
becoming unemployed, homeless, and destitute (Steinitz & So-
lomon, 1986; see also Kaiser Foundation, 2002). Second, it is
possible that culture influences choice of self-regulatory system;
Euro-Canadian college students find promotion-focused strategies
compelling, whereas Asian Canadian college students find
prevention-focused strategies compelling (Lockwood, Marshall, &
Sadler, 2005). Similarly, Hong Kong Chinese college students
found prevention-focused reasons for action more convincing than
did American college students (Lee, Aaker, & Gardner, 2000,
Study 4).
201
Whether focused on positive expected or feared PSs, self-
regulation fails when individuals do not realize that a particular
action is antithetical to goal achievement, when PSs do not provide
clear standards of what to attain or avoid, or when PSs are not
linked with self-regulatory behaviors (e.g., Carver, 2004; Higgins,
1987). For low-income teens, lapses in self-regulation may be
difficult to repair—when risk of failure is high, any misstep can
spell disaster. This contrasts with the situation of middle-class
teens whose self-regulatory lapses can be compensated for by
contextual regulation set in place by neighborhood, school, and
parents. Middle-class students are more likely to be provided
mentoring, tutoring, monitoring, and enrichment activities whether
they seek them out or not (e.g., Sampson, Morenoff, & Earls,
1999). Low-income students are more likely to live in contexts
lacking such collective efficacy resources (Sampson et al., 1999).
Thus, for low-income youths, self-initiated engagement in self-
regulation that focuses on attaining positive PSs and avoiding
negative PSs is likely necessary.
Our emphasis on the situation of low-income and minority
youth is congruent with research using the stereotype threat model
(Steele, 1997) that demonstrates the undermining effect on aca-
demic outcomes of making stigmatized in-group social identity
salient. To date, stereotype threat research has not documented the
mechanism by which stereotype threat produces negative conse-
quences over time. Our results suggest that stereotype threat may
undermine academic attainment by making APSs less salient and
school-focused strategies less accessible as ways of avoiding off-
track PSs. This process model is congruent with a number of
studies documenting that when stereotype threat is activated, pre-
vention (discrepancy enhancing) focus increases (Oyserman,
Uskul, Yoder, Nesse, & Williams, 2005; Seibt & Förster, 2004), as
do negative thoughts about one’s math capacity (Cadinu, Maass,
Rosabianca, & Kiesner, 2005).
Our results demonstrate the real-world power of a social psy-
chological conceptualization of the self as a motivational resource.
By integrating PSs, social identity and metacognitive perspectives,
we developed a process model that, when operationalized, pro-
duced lasting change on PSs, self-regulation, academic outcomes,
and depression. These results are particularly promising given the
difficulty in producing sustained improvement in the educational
outcomes of low-income youths and the great need to develop
integrative social science models that can be applied to ameliorat-
ing this large-scale social problem.
References
Abrams, D. (1994). Social self-regulation. Personality and Social Psychol-
ogy Bulletin, 20, 473– 483.
Aiken, L., & West, S. (1991). Multiple regression: Testing and interpreting
interactions. Newbury Park, CA: Sage.
Altschul, I., Oyserman, D., & Bybee, D. (in press). Racial-ethnic identity
in mid-adolescence: Content and change as predictors of academic
attainment. Child Development.
Ambady, N., Paik, S., Steele, J., Owen-Smith, A., & Mitchell, J. (2004).
Deflecting negative self-relevant stereotype activation: The effects of
individuation. Journal of Experimental Social Psychology, 40, 401– 408.
Arbuckle, J., & Wothke, W. (2003). AMOS 5.0 user’s guide. Chicago:
SmallWaters.
Arroyo, C., & Zigler, E. (1995). Racial identity, academic achievement,
and the psychological well-being of economically disadvantaged ado-
lescents. Journal of Personality and Social Psychology, 69, 903–914.
202 OYSERMAN, BYBEE, AND TERRY
Baumeister, R. (1998). The self. In D. Gilbert, S. Fiske, & G. Lindzey
(Eds.), Handbook of social psychology (4th ed., pp. 680 –740). New
York: McGraw-Hill.
Belli, R. (1998). The structure of autobiographical memory and the event
history calendar: Potential improvements in the quality of retrospective
reports in surveys, Memory, 6, 383– 406.
Blair, S., Blair, M., & Madamba, A. (1999). Racial/ethnic differences in
high school students’ academic performance: Understanding the inter-
weave of social class and ethnicity in the family context. Journal of
Comparative Family Studies, 30, 539 –555.
Burke, P. (2003). Relationships among multiple identities. In P. Burke &
T. Owens (Eds.), Advances in identity theory and research (pp. 195–
214). New York: Kluwer Academic/Plenum Press.
Cadinu, M., Maass, A., Rosabianca, A., & Kiesner, J. (2005). Why do
women underperform under stereotype threat? Evidence for the role of
negative thinking. Psychological Science, 16, 572–578.
Carver, C. (2004). Self-regulation of action and affect. In R. Baumeister &
K. Vohs (Eds.), Handbook of self-regulation: Research, theory and
applications (pp. 13–39). New York: Guilford Press.
Carver, C., & Scheier, M. (1990). Principles of self-regulation: Action and
emotion. In T. Higgins & R. Sorrentinoeds (Eds.), Handbook of moti-
vation and cognition (Vol. 2, pp. 3–52). New York: Guilford Press.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences
(2nd ed.). Hillsdale, NJ: Erlbaum.
Cook, P., & Ludwig, J. (1997). Weighing the “burden of acting White”:
Are there race differences in attitudes toward education. Journal of
Policy Analysis and Management, 16, 256 –278.
Costello, E., Mustillo, S., & Erkanli, A. (2003). Prevalence and develop-
ment of psychiatric disorders in childhood and adolescence. Archives of
General Psychiatry, 60, 837– 844.
Cribbie, R., & Jamieson, J. (2000). Structural equation models and the
regression bias for measuring correlates of change. Educational and
Psychological Measurement, 60, 893–907.
Croizet, J., & Claire, T. (1998). Extending the concept of stereotype and
threat to social class: The intellectual underperformance of students from
low socioeconomic backgrounds. Personality and Social Psychology
Bulletin, 24, 588 –594.
Detroit News (2005, May 29). Graduation rates. Retrieved September 15,
2005, from http://www.detnews.com/2005/specialreport/0505/30/
A01-196697.htm.
Eccles, J., & Wigfield, A. (2002). Motivational beliefs, values, and goals.
Annual Review of Psychology, 53, 109 –132.
Ellickson, P., & Hawes, J. (1989). An assessment of active versus passive
methods for obtaining parental consent. Evaluation Review, 13, 45–55.
Ferguson, R. (1998). Comment on the burden of acting White: Do Black
adolescents disparage academic achievement? by P. J. Cook and J.
Ludwig. In C. Jencks & M. Phillips (Eds.), The Black-White test score
gap (pp. 394 –397). Washington, DC: Brookings Institution.
Finn, J., Pannozzo, G., & Voelkl, K. (1995). Disruptive and inattentive-
withdrawn behavior and achievement among fourth graders. The Ele-
mentary School Journal, 95, 421– 434.
Fordham, S., & Ogbu, J. (1986). Black students’ school success: Coping
with the “burden of acting white.” Urban Review, 18, 176 –206.
Graham, S. (2001). Inferences about responsibility and values: Implication
for academic motivation. In F. Salili, & C. Chiu (Eds.), Student moti-
vation: The culture and context of learning (pp. 31–59). Dordrecht,
Netherlands: Kluwer Academic.
Higgins, T. (1987). Self-discrepancy: A theory relating self and affect.
Psychological Review, 94, 319 –340.
Higgins, T. (1996a). The “self digest”: Self-knowledge serving self-
regulatory functions. Journal of Personality and Social Psychology, 71,
1062–1083.
Higgins, T. (1996b). Knowledge activation: Accessibility, applicability,
and salience. In T. Higgins & A. Kruglanski (Eds.), Social psychology,
handbook of basic principles (pp. 133–168). New York: Guilford Press.
Higgins, T. (1997). Beyond pleasure and pain. American Psychologist, 52,
1280 –1300.
Higgins, T. (1998). The aboutness principle: A pervasive influence on
human inference. Social Cognition, 16, 173–198.
Higgins, T., & Spiegel, S. (2004). Promotion and prevention strategies for
self-regulation: A motivated cognition perspective. In R. Baumeister &
K. Vohs (Eds.), Handbook of self-regulation: Research, theory and
applications (pp. 171–187). New York: Guilford Press.
Hoyle, R., & Kenny, D. (1999). Sample size, reliability, and tests of
statistical mediation. In R. Hoyle (Ed.), Statistical strategies for small
sample research (pp. 195–222). Thousand Oaks, CA: Sage.
Hu, L., & Bentler, P. (1999). Cutoff criteria for fit indexes in covariance
structure analysis: Conventional criteria versus new alternatives. Struc-
tural Equation Modeling, 6, 1–55.
James, W. (1890). Principles of psychology. New York: Holt.
Jo, B. (2002a). Estimation of intervention effects with noncompliance:
Alternative model specifications. Journal of Educational and Behav-
ioral Statistics, 27, 385– 409.
Jo, B. (2002b). Model misspecification sensitivity analysis in estimating
causal effects of interventions with noncompliance. Statistics in Medi-
cine, 25, 305–327.
Joseph, J. (1996). School factors and delinquency: A study of African
American youths. Journal of Black Studies, 26, 340 –355.
Juster, T., Ono, H., & Stafford, F. (2004). Changing times of American
youth, 1981–2003. Retrieved September 14, 2005, from the University
of Michigan Institute for Social Research Child Development Supple-
ment to the Panel Study for Income Dynamics: http://www.umich.edu/
news/Releases/2004/Nov04/teen_time_report.pdf
Kaiser Foundation (2002). National Survey of Latinos. Summary of find-
ings. Retrieved September 14, 2005, from http://www.kff.org/
kaiserpolls/20021217a-index.cfm
Kao, G. (2000). Group images and possible selves among adolescents:
Linking stereotypes to expectations by race and ethnicity. Sociological
Forum, 15, 407– 430.
Kasen, S., Cohen, P., & Brook, J. (1998). Adolescent school experiences
and dropout, adolescent pregnancy, and young adult deviant behavior.
Journal of Adolescent Research, 13, 49 –72.
Kearney, K., Hopkins, R., Mauss, A., & Weisheit, R. (1983). Sample bias
resulting from a requirement for written parental consent. Public Opin-
ion Quarterly, 47, 96 –102.
King, L. (2001). The health benefits of writing about life goals. Personality
and Social Psychology Bulletin, 27, 798 – 807.
King, L., & Smith, N. (2004). Gay and straight possible selves: Goals,
identity, subjective well-being, and personality development. Journal of
Personality, 72, 967–994.
Larsen, R. (2004, November). Emotion and cognition: The case of auto-
matic vigilance. APA online. Retrieved September 14, 2005, from http://
www.apa.org/science/psa/sb-larsen.html
Lee, A., Aaker, J., & Gardner, W. (2000). The pleasures and pains of
distinct self-construals: The role of interdependence in regulatory focus.
Journal of Personality and Social Psychology, 78, 1122–1134.
Little, R., & Rubin, D. (2002). Statistical analysis with missing data (2nd
ed). New York: Wiley.
Little, R., & Yau, L. (1998). Statistical techniques for analyzing data from
prevention trials: Treatment of no-shows using Rubin’s causal model.
Psychological Methods, 3, 147–159.
Lockwood, P., Marshall, T., & Sadler, P. (2005). Promoting success or
preventing failure: Cultural differences in motivation by positive and
negative role models. Personality and Social Psychology Bulletin 31,
379 –392.
Lockwood, P., Sadler, P., Fyman, K., & Tuck, S. (2004). To do or not to
 POSSIBLE SELVES AND ACADEMIC OUTCOMES
do: Using positive and negative role models to harness motivation.
Social Cognition, 22, 422– 450.
MacCallum, R., Browne, M., & Sugawara, H. (1996). Power analysis and
determination of sample size for covariance structure modeling. Psycho-
logical Methods, 1, 130 –149.
MacKinnon, D., Lockwood, C., Hoffman, J., West, S., & Sheets, V. (2002).
A comparison of methods to test mediation and other intervening vari-
able effects. Psychological Methods, 7, 83–104.
MacKinnon, D., Lockwood, C., & Williams, J. (2004). Confidence limits
for the indirect effect: Distribution of the product and resampling meth-
ods. Multivariate Behavioral Research, 39, 99 –128.
MacKinnon, D., Warsi, G., & Dwyer, J. (1995). A simulation study of
mediated effect measures. Multivariate Behavioral Research, 30, 41– 62.
Mardia, K. (1985). Mardia’s test of multinormality. In S. Kotz & N.
Johnson (Eds.), Encyclopedia of statistical sciences (Vol. 5, pp. 217–
221). New York: Wiley.
Maris, E. (1998). Covariance adjustment versus gain scores: revisited.
Psychological Methods, 3, 309 –327.
McDermott, P. (1995). Sex, race, class, and other demographics as expla-
nations for children’s ability and adjustment: A national appraisal.
Journal of School Psychology, 33, 75–91.
McGuire, W., & Padawe-Singer, A. (1976). Trait salience in the sponta-
neous self-concept. Journal of Personality and Social Psychology, 33,
743–754.
Orfield, G., Losen, D., Wald, J., & Swanson, C. (2004). Losing our future:
How minority youth are being left behind by the graduation rate crisis.
Cambridge, MA: The Civil Rights Project at Harvard University. Re-
trieved September 14, 2005, from http://www.urban.org/UploadedPDF/
410936_LosingOurFuture.pdf
Oyserman, D. (2001). Self-concept and identity. In A. Tesser & N.
Schwarz (Eds.), Blackwell handbook of social psychology (pp. 499 –
517). Malden, MA: Blackwell Press.
Oyserman, D. (in press.) Social identity and self-regulation. In A. Kruglan-
ski & T. Higgins, Social psychology: Handbook of basic principles. New
York: Guilford Press.
Oyserman, D., Bybee, D., Terry, K., & Hart-Johnson, T. (2004). Possible
selves as roadmaps. Journal of Research in Personality, 38, 130 –149.
Oyserman, D., & Fryberg, S. A. (2006). The possible selves of diverse
adolescents: Content and function across gender, race, and national
origin. In C. Dunkel & J. Kerpelman (Eds.), Possible selves: Theory,
research, and application (pp. 17–39). Huntington, NY: Nova.
Oyserman, D., Gant, L., & Ager, J. (1995). A socially contextualized
model of African American identity: Possible selves and school persis-
tence. Journal of Personality and Social Psychology, 69, 1216 –1232.
Oyserman, D., Johnson, E., & Bybee, D. (2006). Effects of low socioeco-
nomic status on early adolescent possible selves. Manuscript under
editorial review.
Oyserman, D., & Markus, H. (1990). Possible selves and delinquency.
Journal of Personality and Social Psychology, 59, 112–125.
Oyserman, D., & Saltz, E. (1993). Competence, delinquency, and attempts
to attain possible selves. Journal of Personality and Social Psychology,
65, 360 –374.
Oyserman, D., Terry, K., & Bybee, D. (2002). A possible selves interven-
tion to enhance school involvement. Journal of Adolescence, 24, 313–
326.
Oyserman, D., Uskul, A., Yoder, N., Nesse, R., & Williams, D. (2005).
Unfair treatment and self-regulatory focus. Manuscript submitted for
editorial review.
Pokorny, S., Jason, L., Schoeny, M., Townsend, S., & Curie, C. (2001). Do
participation rates change when active consent procedures replace pas-
sive consent. Evaluation Review, 25, 567–580.
Radloff, L. (1977). The CES-D scale: A self-report depression scale for
research in the general population. Applied Psychological Measurement,
1, 385– 401.
203
Radloff, L., & Locke, B. (2000). Center for Epidemiologic Studies De-
pression Scale (CESD). In Taskforce for the Handbook of Psychiatric
Measures (Eds.), Handbook of psychiatric measures: American Psychi-
atric Association (pp. 523–526). Washington, DC: American Psychiatric
Association.
Raudenbush, S., & Bryk, A. (2002). Hierarchical linear models: Applica-
tions and data analysis methods (2nd ed.). Thousand Oaks, CA: Sage.
Raudenbush, S., & Liu, X. (2000). Statistical power and optimal design for
multisite randomized trials. Psychological Methods, 5, 199 –213.
Raudenbush, S., & Liu, X. (2001). Effects of study duration, frequency of
observation, and sample size on power in studies of group differences in
polynomial change. Psychological Methods, 6, 387– 401.
Régner, I., Huguet, P., & Monteil, J. (2002). Effects of socioeconomic
status (SES) information on cognitive ability inferences: When low-SES
students make use of a self-threatening stereotype. Social Psychology of
Education, 5, 253–269.
Roberts, R., Attkisson, C., & Rosenblatt, A. (1998). Prevalence of psycho-
pathology among children and adolescents. American Journal of Psy-
chiatry, 155, 715–725.
Roderick, M. (2003). What’s happening to the boys? Early high school
experiences and school outcomes among African American male ado-
lescents in Chicago. Urban Education, 38, 538 – 607.
Rothman, A., & Schwarz, N. (1998). Constructing perceptions of vulner-
ability: Personal relevance and the use of experiential information in
health judgments. Personality and Social Psychology Bulletin, 24,
1053–1064.
Ruvolo, A., & Markus, H. (1992). Possible selves and performance: The
power of self-relevant imagery. Social Cognition, 10, 95–124.
Sampson, R., Morenoff, J., & Earls, F. (1999). Beyond social capital:
Spatial dynamics of collective efficacy for children. American Sociolog-
ical Review, 64, 633– 660.
Sandler, I., & Chassin, L. (2002). Training of prevention researchers:
Perspectives from the Arizona State University Prevention Research
Training Program. Prevention & Treatment, 5, Item 5. Retrieved April 10,
2006, from http://journals.apa.org/prevention/volume5/pre0050006a.html
Schwarz, N. (1998). Accessible content and accessibility experiences: The
interplay of declarative and experiential information in judgment. Per-
sonality and Social Psychology Review, 2, 87–99.
Schwarz, N. (2002). Situated cognition and the wisdom of feelings: Cog-
nitive tuning. In L. Feldman Barrett & P. Salovey (Eds.), The wisdom in
feelings (pp. 144 –166). New York: Guilford Press.
Schwarz, N., & Bless, H. (1992). Constructing reality and its alternatives:
An inclusion/exclusion model of assimilation and contrast effects in
social judgment. In: L. Martin & A. Tesser (Eds.), The construction of
social judgments (pp. 217–245). Hillsdale, NJ: Erlbaum.
Schwarz, N., & Clore, G. (1996). Feelings and phenomenal experiences.
In: T. Higgins & A. Kruglanski (Eds.), Social psychology: A handbook
of basic principles. New York: Guilford Press.
Seibt, B., & Förster, J. (2004). Stereotype threat and performance: How
self-stereotypes influence processing by inducing regulatory foci. Jour-
nal of Personality and Social Psychology, 87, 38 –56.
Seidman, E., Aber, J., Allen, L., & French, S. (1996). The impact of the
transition to high school on the self-esteem and perceived social context
of poor urban youth. American Journal of Community Psychology, 24,
489 –515.
Seidman, E., Allen, L., Aber, J. L., Mitchell, C., & Feinman, J. (1994). The
impact of school transitions in early adolescence on the self-system and
perceived social context of poor urban youth. Child Development, 65,
507–522.
Settles, I. H. (2004). When multiple identities interfere: The role of identity
centrality. Personality and Social Psychology Bulletin, 30, 487–500.
Snijders, T., & Bosker, R. (1999). Multilevel analysis: An introduction to
basic and advanced multilevel modeling. London: Sage.
204 OYSERMAN, BYBEE, AND TERRY
Steele, C. (1997). A threat in the air: How stereotypes shape intellectual
identity and performance. American Psychologist, 52, 613– 629.
Steinitz, V., & Solomon, E. (1986). Starting out: Class and community
in the lives of working-class youth. Philadelphia: Temple University
Press.
Stoep, A., Weiss, N., Kuo, E., Cheney, D., & Cohen, P. (2003). What
proportion of failure to complete secondary school in the U.S. popula-
tion is attributable to adolescent psychiatric disorder? Journal of Behav-
ioral Health Services and Research, 30, 119 –124.
Strauman, T. (2002). Self-regulation and depression. Self & Identity, 1,
151–157.
Strauman, T., & Higgins, T. (1988). Self-discrepancies as predictors of
vulnerability to distinct syndromes of chronic emotional distress. Jour-
nal of Personality, 56, 685–707.
Thomas, D., Townsend, T., & Belgrave, F. (2003). The influence of
cultural and racial identification on the psychosocial adjustment of
inner-city African American children in school. American Journal of
Community Psychology, 32, 217–228.
U.S. Bureau of the Census (2000). Census 2000. Retrieved April 10, 2006,
from http://www.census.gov
Weinger, S. (2000). Opportunities for career success: Views of poor and
middle-class children. Children & Youth Services Review, 22, 13–35.
Weinstein, N. (1993). Testing four competing theories of health-protective
behavior. Health Psychology, 12, 324 –333.
Received December 22, 2004
Revision received January 4, 2006
Accepted January 24, 2006 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 33– 48 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.33

Thinking Within the Box: The Relational Processing Style Elicited by

Counterfactual Mind-Sets

Laura J. Kray Adam D. Galinsky

University of California, Berkeley Northwestern University

Elaine M. Wong
University of California, Berkeley

By comparing reality to what might have been, counterfactuals promote a relational processing style
characterized by a tendency to consider relationships and associations among a set of stimuli. As such,
counterfactual mind-sets were expected to improve performance on tasks involving the consideration of
relationships and associations but to impair performance on tasks requiring novel ideas that are
uninfluenced by salient associations. The authors conducted several experiments to test this hypothesis.
In Experiments 1a and 1b, the authors determined that counterfactual mind-sets increase mental states
and preferences for thinking styles consistent with relational thought. Experiment 2 demonstrated a
facilitative effect of counterfactual mind-sets on an analytic task involving logical relationships; Exper-
iments 3 and 4 demonstrated that counterfactual mind-sets structure thought and imagination around
salient associations and therefore impaired performance on creative generation tasks. In Experiment 5,
the authors demonstrated that the detrimental effect of counterfactual mind-sets is limited to creative
tasks involving novel idea generation; in a creative association task involving the consideration of
relationships between task stimuli, counterfactual mind-sets improved performance.

Keywords: counterfactuals, mind-sets, relational processing, analytical, creativity

Whenever individuals consider how the past might have turned
out differently, they are engaging in counterfactual thinking. For
example, people who ponder, “What would life be like if I had
married that other person?” or “Would I be better off if I had
selected that other job?” are implicitly comparing reality with what
might have been. Thoughts of “if only” and “what if” are signposts
for counterfactual musings, and their presence in mental life is
both pervasive and predictable. A growing body of literature
suggests that counterfactuals are not merely fodder for daydream-
ers stuck in the past but rather serve important functions for
directing future behavior (Roese & Olson, 1995).
Laura J. Kray and Elaine M. Wong, Haas School of Business, University
of California, Berkeley; Adam D. Galinsky, Kellogg Graduate School of
Management, Northwestern University.
Elaine M. Wong is now at the School of Communication, Northwestern
University.
This research was supported in part by National Science Foundation
Grants SES-0233294 and SES-0136931. We gratefully acknowledge in-
sightful comments provided by Phil Tetlock on an earlier version of this
article and the valuable feedback provided by participants in colloquia at
the Haas School of Business, Carnegie Mellon University, and the Uni-
versity of Chicago. We are indebted to the many research assistants who
helped out in every stage of the research process. In particular, we are
grateful to Joyce Chen, Joe Gacula, Linda Pham, and Alissa Roberts.
Correspondence concerning this article should be addressed to Laura J.
Kray, Haas School of Business, 545 Student Services Building, #1900,
Berkeley, CA 94720. E-mail: kray@haas.berkeley.edu
In the present article, we examined the impact of counterfactual
thinking on subsequent thinking styles and problem solving.
Broadly speaking, we explored how reflecting back on events in
which an outcome almost turned out differently impacts future
problem solving. More precisely, we explored the nature of the
mind-set that results from constructing counterfactual thoughts.
Because counterfactuals involve a consideration of both reality and
what might have been, they are inherently relational in nature. Our
central thesis is that constructing counterfactual thoughts in one
context produces a counterfactual mind-set characterized by a
tendency to process information relationally in subsequent con-
texts. Thus, the mental structure of logical relationships created
through counterfactual thinking increases the ability to understand
and perceive relationships in subsequent contexts, structuring
thought around salient associations and the pursuit of connections.
Because the counterfactual mind-set occurs regardless of the con-
tent or valence of the counterfactual thoughts (Galinsky & Kray,
2004; Galinsky & Moskowitz, 2000; Kray & Galinsky, 2003),
what lingers is the form and structure of the counterfactual (i.e., the
tendency to consider relationships and connections between a set
of stimuli).
Although previous research has explored the consequences of
the counterfactual mind-set on subsequent performance across a
variety of tasks, the present research is the first to identify the
mental states and thinking styles that counterfactuals activate and
the mechanisms by which counterfactual mind-sets have their
effects. We characterize the counterfactual mind-set as a structured
form of thought involving a consideration of relationships and

33
34 KRAY, GALINSKY, AND WONG
associations between a set of stimuli. We expect this processing
style to have three consequences. First, it should create phenom-
enological experiences and preferences for structured thought
(Sternberg, 1988). Second, it should promote lay conceptions of
analytic thought, defined as “an examination of a complex, its
elements and their relations” (Merriam-Webster, 2005). Third, it
should promote structured imagination, or the tendency to build on
existing knowledge structures (Ward, 1994). We have chosen the
term relational processing style to capture the essence of the
counterfactual mind-set because it both describes the nature of
counterfactual thoughts and because it is broad enough to encom-
pass the full range of effects produced by the activation of a
counterfactual mind-set.
The Conceptual Link Between Counterfactuals and
Relational Processing
Although the ability to undo events and construct possible
worlds is theoretically unlimited, in reality, when and how coun-
terfactuals are constructed is fairly predictable. The commence-
ment of counterfactual thinking is often initiated when an event
nearly occurred (Kahneman & Miller, 1986; Roese & Olson,
1997). For example, missing a plane by 5 min tends to evoke more
counterfactuals than missing a plane by 1 hr, presumably because
the former case more readily conjures up elements of the day
leading up to the flight that could have been undone. In addition to
near misses, “abnormal” events tend to produce counterfactual
thoughts. For example, it is easier to undo missing one’s flight
when a new, atypical route to the airport was taken than it is after
taking one’s usual route. It is often the mere presence of an
obvious mutable component to an event that leads to the sponta-
neous generation of counterfactuals.
Just like the rules that govern when counterfactuals are gener-
ated, how counterfactual thoughts are constructed is predictable.
Counterfactuals involve a comparison of the relationship between
reality and what might have been. Constructing a counterfactual
thought implicitly involves laying out a causal chain of events in
an action sequence and mutating one step in the process to con-
struct an alternate reality. As such, running a counterfactual sim-
ulation in one’s head is the mental equivalent of conducting an
experiment. Like the experimental process, counterfactual thinking
involves a logical consideration of relationships and causal asso-
ciations between events (Einhorn & Hogarth, 1986; Mandel, 2003;
Mandel & Lehman, 1996; Wells & Gavanski, 1989). For example,
Wells and Gavanski showed that an initial event was judged to be
causally connected to a subsequent event to the extent that a
mutation to the initial event would have undone the occurrence of
the subsequent event.
As described above, a logical structure underlies when and how
counterfactuals are created. Counterfactuals are most likely when
potential mutations to a sequence of events are salient or when
potential alternative worlds are close in time and space. Implicit
comparisons assessing similarities and differences between the
alternate world and reality are then made that facilitate the iden-
tification of causal connections (Markman & McMullen, 2003).
Broadly speaking, relational processing is used in counterfactual
thinking in that it involves the consideration of relationships and
connections between events. We expect that once the counterfac-
tual mind-set is activated, it persists because the relational pro-
cessing style that accompanies it is a well-learned and useful
strategy for assessing causality, a critical tool for comprehending
the world (cf. Kelley, 1973).
Past research has demonstrated that situations with an obvious
mutable component tend to elicit counterfactual thoughts, which
then orient cognition in subsequent contexts. This cognitive ori-
entation has been called a “counterfactual mind-set” (Galinsky &
Kray, 2004; Kray & Galinsky, 2003). The idea that the activation
of a counterfactual mind-set impacts subsequent cognition and
performance has received strong support across a variety of indi-
vidual and group tasks. More specifically, counterfactual mind-
sets appear to be an asset when the consideration of alternatives
facilitates performance (Galinsky & Kray, 2004; Galinsky & Mos-
kowitz, 2000; Kray & Galinsky, 2003). Counterfactual mind-sets
also appear to encourage skepticism about the dominant hypoth-
esis. For example, having just considered an alternate reality can
reduce the confirmation bias or the tendency to seek information
that is consistent with an existing hypothesis to the relative detri-
ment of information that could potentially disconfirm it (Galinsky
& Moskowitz, 2000; Kray & Galinsky, 2003). In fact, the number
of counterfactual thoughts generated in response to a mutable
scenario has been shown to be an important predictor of subse-
quent disconfirmatory information search (Kray & Galinsky,
2003). Counterfactual mind-sets also improve decision accuracy
by increasing the discussion of unique information critical for
group decision making and promoting synergistic coordination, or
the tendency of group members to build on and develop relation-
ships between each other’s ideas (Galinsky & Kray, 2004; Liljen-
quist, Galinsky, & Kray, 2004). Finally, counterfactual thinking
has been shown to increase the scrutiny of persuasive message
content (Krishnamurthy & Sivaraman, 2002), enabling decision
makers to distinguish between strong and weak arguments. Despite
the diversity of tasks explored in this growing body of literature, a
common facilitative effect has emerged after the activation of a
counterfactual mind-set. Overall, it appears that by considering
alternative realities in one context, greater clarity regarding task
associations is gained in later contexts. Although the counterfac-
tual mind-set has been described generally as promoting “a con-
sideration of alternatives” (Galinsky & Moskowitz, 2000), in the
present investigation, we seek to better understand the content of
and processing style associated with this counterfactual mind-set
and the mechanisms by which it impacts performance.
Research Overview
Despite numerous demonstrations that counterfactual mind-sets
affect subsequent problem solving, to date, our understanding of
the exact nature of a counterfactual mind-set has been limited
because little effort has been made to explore the underlying
process by which counterfactual mind-sets influence performance.
Although previous research has examined the content and structure
of counterfactual thoughts (Roese, 1994; Roese & Olson, 1995),
no research to date has explored the phenomenological experience
produced by counterfactual thinking. We argue that the mind-set
promotes a relational processing style, characterized by a tendency
to ponder associations and make connections between a set of
stimuli. As such, we suggest that counterfactual mind-sets will
 COUNTERFACTUAL MIND-SETS
improve performance on analytic tasks, which typically require
that one identify and understand the relationships among a set of
stimuli.
In seeking to understand the unique characteristics of the coun-
terfactual mind-set, we also explored its impact on two types of
creative tasks. Our hypothesis that counterfactual mind-sets pro-
mote a relational processing style suggests that they might lead to
imaginative processes that build from existing knowledge struc-
tures. Ward (1994) termed the tendency to rely on existing knowl-
edge in the creative process as “structured imagination.” Although
Ward demonstrated that structured imagination is fairly character-
istic of how creative tasks are approached in general, we argue that
this tendency is intensified by the activation of a counterfactual
mind-set. Just as the generation of counterfactual thoughts are
structured and conform to certain rules and logic, we contend that
imagination following the activation of a counterfactual mind-set
is structured around salient knowledge structures. Therefore, coun-
terfactual mind-sets should have a positive effect on creative tasks
that require the identification of associations within and between a
set of stimuli. However, if a counterfactual mind-set consists of a
relational processing style that structures imagination, then it
might hinder the generation of novel ideas. This increased atten-
tion to associations among task stimuli may actually decrease
one’s ability to “think outside the box.”
Each of the findings described above is consistent with the basic
hypothesis that counterfactual mind-sets foster not just a consid-
eration of alternatives, as previous research has suggested, but
more precisely a consideration of relationships and connections
between or among a set of stimuli, structuring thought and imag-
ination around these associations. The present set of experiments
was conducted to test this hypothesis. In Experiment 1a and 1b, we
examined whether counterfactual mind-sets promote a phenome-
nological experience consistent with a relational processing style.
In Experiment 2, we examined whether counterfactual mind-sets
improve performance on an analytic task requiring the consider-
ation of relationships and associations. Specifically, we examined
performance on a standardized test designed to assess the analytic
reasoning skills of potential graduate school applicants. In Exper-
iments 3 and 4, we explored the impact of counterfactual mind-sets
on structured imagination. In Experiment 5, we examined the
relative impact of counterfactual mind-sets on creative generation
versus creative association tasks, which generally differ in the
optimal degree of conceptual attention devoted to relationships and
associations between task stimuli. Overall, the studies presented
here clarify the nature and phenomenology of the counterfactual
mind-set by demonstrating that activating a counterfactual mind-
set promotes a relational processing style and therefore has pre-
dictable effects, both beneficial and detrimental, on a range of
problem-solving tasks.
Experiment 1a: Counterfactual Mind-Sets and Mental
States
The purpose of Experiment 1a was twofold. First, we sought to
better understand how a counterfactual mind-set is experienced at
a phenomenological level. As described above, we expected that
the mental structure of logical relationships created through coun-
terfactual thought would promote a relational processing style. In
35
considering how this processing style might translate into a mental
state, we reasoned that an individual who is primed to consider
logical relationships should feel poised for analytic and critical
thinking.
A second goal was to rule out the possibility that a third variable
is responsible for the effect of counterfactual mind-sets on perfor-
mance. Because analytic thinking is affected by mood (Schwarz &
Bless, 1991), it is possible that counterfactual mind-sets simply
depress moods. Previous research has addressed this possibility by
manipulating whether participants generate downward versus up-
ward counterfactuals (Roese, 1994). Downward counterfactuals, or
thoughts about how events could have been worse, tend to evoke
positive feelings such as joy, relief, and surprise. Upward coun-
terfactuals, or thoughts about how events could have been better,
tend to evoke negative feelings such as regret, remorse, and
disappointment. Because upward and downward counterfactuals
tend to evoke different emotional experiences, the fact that the
valence of the counterfactual thoughts has not moderated any of
the findings to date on counterfactual mind-sets and problem
solving (Galinsky & Moskowitz, 2000; Kray & Galinsky, 2003)
suggests that affect and counterfactual mind-sets have independent
effects on problem solving. In the present experiment, in addition
to manipulating the valence of the scenario, we also measured
mood to determine whether it operates as a mediator. Recognizing
that common psychological measures of affect rely on self-
assessments (cf. Watson, Clark, & Tellegen, 1988), we compared
self-reported cognitive and affective states after engaging in coun-
terfactual thinking.
Method
Design. The experiment had a 2 (type of prime: counterfactual vs.
noncounterfactual) ⫻ 2 (valence of prime: positive vs. negative) between-
subjects factorial design.
Participants. Participants were 65 students on a large western univer-
sity campus. Participants were approached in several cafes on campus and
asked to complete a short questionnaire. In return for their compliance,
they were given a pencil bearing the university’s logo.
Procedure. Participants read a scenario that described the actions of
Jane, a woman who was attending a rock concert (see Galinsky & Mos-
kowitz, 2000). In each scenario, an individual at the rock concert wins a
valuable prize, a trip to Hawaii. In half of the scenarios, Jane is the winner
of the prize (positive valence); in the other half of the scenarios, Jane is not
the winner of the prize (negative valence). In addition, half of the scenarios
describe a sequence of events designed to elicit counterfactual thoughts,
and half of the scenarios describe a sequence of events that is not expected
to elicit counterfactual thoughts. In the downward counterfactual scenario,
Jane wins the trip to Hawaii when the new seat she had just switched to (in
order to get a better view of the stage) was chosen. In the upward
counterfactual scenario, Jane loses the trip to Hawaii when the seat that she
had just switched from wins the trip. In the noncounterfactual conditions,
Jane does not switch seats. After reading one of the four scenarios,
participants were asked to consider some thoughts likely to be running
through Jane’s mind.
After considering the scenario, participants were asked to “Please indi-
cate the extent to which your current mental state is characterized by the
following attributes.” This measure allowed us to evaluate the degree to
which participants’ current cognitive state was consistent with a sense of
being analytic. Specifically, participants’ cognitive states were assessed
36 KRAY, GALINSKY, AND WONG

along the following dimensions (␣ ⫽ .85): analytic, critical, focused, and
smart.1
We also sought to distinguish the impact of counterfactual thinking on
cognitive states from its impact on affective states. As affective experience
can be distinguished along valence-based and arousal-based dimensions
(Feldman, 1995; Russell, 1980), we measured both of these affective
responses separately. To measure valence-based affect, we included a
two-item mood measure (␣ ⫽ .80): mood and happy. To measure arousal-
based affect, we simply asked participants to assess their perceived arousal.
Mood was measured on a 9-point scale ranging from 1 (very negative) to
9 (very positive). All other assessments were done on 9-point scales
ranging from 1 (not at all characteristic) to 9 (very characteristic).
Results and Discussion
To test our hypotheses regarding the cognitive and affective
states elicited by a counterfactual mind-set, we conducted 2 (type
of prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence of
prime: positive vs. negative) between-subjects analyses of variance
(ANOVAs). Table 1 provides a correlation matrix of all experi-
mental variables.
Cognitive state. We hypothesized that the process of thinking
counterfactually would elicit a sense of being analytic. Consistent
with our hypothesis, participants in a counterfactual mind-set
(M ⫽ 5.8, SD ⫽ 1.6) rated their cognitive state as being more
consistent with a sense of being analytic than did participants
exposed to a noncounterfactual scenario (M ⫽ 4.3, SD ⫽ 1.9), F(1,
61) ⫽ 14.12, p ⬍ .001.2 No other effects emerged as statistically
significant.
Valence-based affective experience. Although we did not ex-
pect our counterfactual manipulation to affect participants’ affec-
tive experience, we explored whether the protagonist winning or
losing the valuable prize would affect participants’ emotional
reaction. Participants exposed to a positive outcome scenario in-
deed reported a more positive emotional experience (M ⫽ 6.9,
SD ⫽ 1.5) than did participants exposed to a negative outcome
scenario (M ⫽ 5.39, SD ⫽ 2.1), F(1, 61) ⫽ 8.43, p ⫽ .005. Neither
the effect for type of prime, F(1, 61) ⫽ 1.2, p ⫽ .28, nor the
interaction between type and valence of prime, F(1, 61) ⫽ 0.08,
p ⫽ .78, were statistically significant.
Arousal-based affective experience. Like affect, we expected
that only the valence of the outcome in the rock concert scenario
would impact participants’ perceived level of arousal. Consistent
with this hypothesis, a main effect for valence of prime emerged as
statistically significant, F(1, 61) ⫽ 5.57, p ⫽ .02. Participants who
read the scenario in which a positive outcome occurred (M ⫽ 5.37,
SD ⫽ 2.97) reported feeling more aroused than participants who
Table 1
Experiment 1a: Correlations Between Variables
Variable 1 2 3 4 5 We included several other exploratory, distractor variables that were
1. Type of prime — these additional variables, they will not be discussed further.
2. Valence of prime .04 — 2
3. Cognitive state ⴚ.41 .08 — In a separate sample of 51 participants, we replicated this effect. That
4. Valence-based affective state .14 ⴚ.34 ⫺.20 — is, participants exposed to a counterfactual prime rated themselves higher
5. Arousal-based affective state ⫺.10 ⴚ.28 ⫺.20 .12 — on terms characteristic of an analytic style of thinking (M ⫽ 6.74, SD ⫽
Note. Significant correlations ( p ⬍ .05) are in boldface.
read a scenario in which a negative outcome occurred (M ⫽ 3.63,
SD ⫽ 2.7). An unanticipated two-way interaction between valence
of prime and type of prime also emerged, F(1, 61) ⫽ 4.17, p ⫽ .05.
In the noncounterfactual condition, participants reported more
arousal after the positive outcome (M ⫽ 6.22, SD ⫽ 2.86) relative
to the negative outcome (M ⫽ 2.96, SD ⫽ 2.54), t(31) ⫽ 3.17, p ⫽
.01; however, in the counterfactual condition, the difference be-
tween the two outcomes (M ⫽ 4.6, SD ⫽ 2.99 vs. M ⫽ 4.4, SD ⫽
2.72) was not statistically significant, t(30) ⫽ 0.22, ns. No other
effects emerged as statistically significant.
The main purpose of this study was to provide support for our
assertion that counterfactual mind-sets create a phenomenological
experience consistent with a relational processing style. Consistent
with this hypothesis, participants in the counterfactual condition
reported feeling more poised for critical thinking than did baseline
participants. After having engaged in relational processing through
the construction of counterfactual thoughts, participants in the
resulting mind-set reported more affinity for analytic thinking.
Another goal of this study was to bolster support for the asser-
tion that the impact of counterfactual mind-sets on cognitive
processes operates independently of emotional experiences. The
fact that the counterfactual manipulation had no effect on partic-
ipants’ reported affective state casts serious doubt on the possibil-
ity that mood accounts for the relationship between counterfactual
mind-sets and relational processing. Whereas the counterfactual
mind-set had no effect on the mood of participants, it had a clear
effect on their cognitions. Given the reasonable assumption that
participants are similarly adept at assessing their cognitive and
affective experiences, this finding suggests that the effects of
counterfactual thinking on cognitive processing occur indepen-
dently of moods.
Experiment 1b: Counterfactual Mind-Sets and Thinking
Styles
In Experiment 1a, we observed that counterfactual mind-sets led
to self-assessed mental states consistent with a relational process-
ing style. Because the measures we used to assess internal states
had not been independently validated, we thought it important to
replicate the effect with established measures of cognitive process-
ing styles. To this end, we examined the impact of our counter-
factual manipulation on Sternberg’s (1988) thinking style
preferences.
Sternberg proposes that abilities may be used for three distinct
functions in mental self-government, including legislative, execu-
tive, and judicial thinking styles (O’Hara & Sternberg, 2001).
Individuals with executive styles prefer tasks that have clearly
defined structure and guidelines from which to solve problems and
build. Individuals with legislative thinking styles prefer tasks that
1
not theoretically relevant. As no statistically significant effects emerged for
1.06) than those exposed to the noncounterfactual prime (M ⫽ 5.88, SD ⫽
1.06), F(1, 47) ⫽ 7.26, p ⬍ .01.
 COUNTERFACTUAL MIND-SETS
have little assigned structure and that allow them to invent new
ideas and tend to enjoy creating original works. Individuals with
judicial styles like to evaluate and critique others’ ideas and enjoy
giving feedback and advice.
We hypothesized that the executive thinking style preference
would be stronger for individuals in a counterfactual mind-set. We
expected that the mental structure of logical relationships created
through counterfactual thinking would increase preferences for the
structured, rule-based, logical nature of the executive thinking
style.
Another goal of the present experiment was to explore the
connection between the construction of counterfactual thoughts
and the strength of the resulting mind-set. Past research has dem-
onstrated that the counterfactual mind-set’s impact occurs regard-
less of the content or valence of the counterfactual thoughts
(Galinsky & Kray, 2004; Kray & Galinsky, 2003), which suggests
that what lingers is the implicit mental structure of the counter-
factual. An established methodology for assessing the strength of
the counterfactual mind-set involves counting the number of coun-
terfactual thoughts generated in the pretask scenario. Because no
counterfactual is explicitly stated in the scenario, the tendency to
infer counterfactual thoughts following the mutable versus non-
mutable scenario can be assessed through this process. Kray and
Galinsky (2003) observed that the strength of counterfactual acti-
vation mediated the relationship between the experimental manip-
ulation and disconfirmatory information search. In the present
experiment, we examined whether counterfactual activation (the
number of counterfactual thoughts) mediated the relationship be-
tween type of prime and thinking styles. Finally, because we were
interested in determining whether writing out the counterfactual
thoughts (vs. simply pondering them) intensifies the impact of the
prime on thinking styles, we included this as a factor in our design.
Method
Design. The experiment had a 2 (type of prime: counterfactual vs.
noncounterfactual) ⫻ 2 (valence: positive vs. negative) ⫻ 2 (thought
listing: yes vs. no) between-subjects factorial design. We also included an
additional no-valence control condition to establish that our valence-based
control conditions serve as appropriate baseline comparisons.
Participants. Participants were 139 students from a large midwestern
university campus. Participants were paid $10 an hour for their
participation.
Procedure. Participants were greeted in the laboratory by an experi-
menter who explained that they would complete several questionnaires
Table 2
Experiment 1b: Correlations Between Variables
Variable 1 2
1. Type of prime —
2. Valence of prime .03
3. Number of counterfactual thoughts .69
4. Executive style .21
5. Judicial style ⫺.01
6. Legislative style ⫺.03
7. Mood .08
Note. Significant correlations ( p ⬍ .05) are in boldface.
37
related to decision making. The experimental manipulations and the de-
pendent variables were embedded in a single packet of questionnaires.
Participants read the same rock concert scenario used in Experiment 1a,
considered thoughts going through the main character’s mind (for approx-
imately half the participants), and then completed both thinking style and
mood measures.
Experimental manipulation. The counterfactual prime manipulations
were identical to those used in Experiment 1a. We also include an addi-
tional control condition, with no counterfactual prime and no valence. In
this scenario, participants read the following: “Three weeks ago, Jane
bought a general admission ticket to a rock concert of her favorite band.
Jane is now at the concert, which is about to begin.” Approximately half
the participants were asked simply to read the scenario as in Experiment 1a.
The other participants were asked to “List some thoughts going through
Jane’s mind.” The sheet on which they listed their thoughts was numbered
from 1 to 10, but participants were told to list only as many thoughts as
came to their mind. Because none of the scenarios contain actual counter-
factual statements, but rather only possess the potential to generate coun-
terfactual thoughts, the number of counterfactual thoughts that participants
subsequently listed was our measure of the strength of the counterfactual
mind-set.
Thinking style. Participants’ preferred thinking style was assessed us-
ing a 24-item subset of Sternberg and Wagner’s (1991) Thinking Styles
Inventory. The Thinking Styles Inventory comprises three subscales, in-
cluding Executive Style, Legislative Style, and Judicial Style. Participants
were asked to indicate how well each statement characterized their current
preferred approach to solving problems and making decisions using a
7-point Likert scale ranging from 1 (extremely uncharacteristic) to 7
(extremely characteristic). The subscales had a high degree of reliability
ranging from .72 to .76.
Mood. Mood was assessed using a one-item 9-point scale ranging from
1 (very negative) to 9 (very positive).
Results and Discussion
Table 2 provides a correlation matrix for all experimental
variables.
Counterfactual activation. Two independent coders identified
the number of counterfactual thoughts listed by participants. The
reliability for counterfactual thoughts was high (␣ ⫽ .95), and,
therefore, the ratings of the two coders were averaged. We sub-
mitted the number of counterfactual thoughts to a 2 (type of prime:
counterfactual vs. noncounterfactual) ⫻ 2 (valence: positive vs.
negative) between-subjects factorial design. As expected, counter-
factual prime participants (M ⫽ 1.78, SD ⫽ 1.1) listed signifi-
cantly more counterfactual thoughts than noncounterfactual prime
participants (M ⫽ 0.19, SD ⫽ 0.42), F(1, 67) ⫽ 69.48, p ⬍ .001.
3 4 5 6 7
—
ⴚ.24 —
.10 .28 —
.04 .03 .10 —
ⴚ.23 .00 ⫺.29 .34 —
⫺.09 .08 .10 .24 .25 —
38 KRAY, GALINSKY, AND WONG
Consistent with previous research (Galinsky & Moskowitz, 2000;
Roese & Hur, 1997), participants exposed to a negatively valenced
event (M ⫽ 1.25, SD ⫽ 1.2) generated more counterfactual
thoughts than participants exposed to a positively valenced event
(M ⫽ 0.70, SD ⫽ 1.0), F(1, 67) ⫽ 7.85, p ⫽ .007. More important
for establishing the independence of counterfactual thinking and
mood, however, the two-way interaction was not significant, F(1,
67) ⫽ 1.79, ns.
Thinking styles. To test our hypotheses, we conducted separate
2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (va-
lence of prime: positive vs. negative) ⫻ 2 (thought listing: yes vs.
no) between-subjects ANOVAs for each thinking style. We hy-
pothesized that exposure to a counterfactual prime would elicit an
executive thinking style. In support of our hypothesis, participants
in a counterfactual mind-set (M ⫽ 5.28, SD ⫽ 0.68) rated them-
selves higher on executive thinking style than participants exposed
to a noncounterfactual scenario (M ⫽ 4.97, SD ⫽ 0.82), F(1,
114) ⫽ 4.85, p ⫽ .03. The only other effect to emerge was a
tendency for participants exposed to a scenario in which the
protagonist did not win the trip to Hawaii to indicate a greater
preference for the legislative thinking style (M ⫽ 5.38, SD ⫽ 0.81)
than participants exposed to the positive emotional experience
(M ⫽ 5.04, SD ⫽ 0.66), F(1, 114) ⫽ 5.74, p ⫽ .02. No other
effects emerged as statistically significant.
Relationship between counterfactual activation and executive
thinking style. To better understand the relationship between
counterfactual mind-sets and the executive thinking style, we
examined whether the amount of counterfactual activation medi-
ated the relationship between the counterfactual primes and an
increased executive thinking style. When we regressed executive
thinking style on both type of counterfactual prime and number of
counterfactual thoughts listed, the effect of number of counterfac-
tual thoughts on the outcome approached significance (␤ ⫽ .15),
t(68) ⫽ 1.77, p ⫽ .08, but the partial effect of the type of prime on
the outcome was significantly reduced in magnitude once the
number of thoughts was controlled (␤ ⫽ ⫺.00), t(68) ⫽ ⫺0.01, ns.
A Sobel test determined the reduction in the significance level was
statistically significant (Z ⫽ 2.35, p ⬍ .05). This finding suggests
that the generation of counterfactual thoughts accounted for the
relationship between the mutable scenario and thinking style
preferences.
Supplementary control condition analyses. We also conducted
analyses that included the additional no-valence control condition
(M ⫽ 5.00, SD ⫽ 0.83) to build confidence that our counterfactual
manipulation was responsible for the increased preference for an
executive thinking style. To do so, we conducted planned contrasts
comparing the two counterfactual conditions with the three control
conditions (no-valence, negative outcome, positive outcome non-
counterfactuals), which revealed that participants in the counter-
factual conditions scored higher on executive thinking style than
those in the control conditions, t(134) ⫽ 2.24, p ⬍ .05. Addition-
ally, we tested whether the additional control differed from the two
original control conditions and found no significant differences for
their effects on executive thinking style, t(134) ⫽ 0.16, ns.
Supplementary mood analyses. To determine whether mood
was affected by our experimental manipulation, we first conducted
an ANOVA, with mood as the dependent variable and type of
prime, valence of prime, and thought listing as the independent
variables. No effects were statistically significant. We also con-
ducted an ANOVA, with executive thinking style as the dependent
variable, including our independent variables and the covariate
mood. The main effect of type of counterfactual prime remained
reliable, F(1, 113) ⫽ 4.47, p ⫽ .04. These findings suggest that
mood did not mediate the effect of counterfactual prime on exec-
utive thinking style.
Three noteworthy findings emerged from this set of experi-
ments. First, the heightened sense that one was poised for analytic
thinking (see Experiment 1a) and the greater preference for an
executive thinking style (see Experiment 1b) following the acti-
vation of a counterfactual mind-set is consistent with the idea that
the mind-set promotes a relational processing style. Second, the
number of counterfactual thoughts was implicated as the mediating
mechanism through which our experimental manipulation elicited
a preference for an executive style of thinking (see Experiment
1b). And finally, across both experiments, we failed to observe any
indication that mood is a driving force behind counterfactual
mind-sets. This conclusion is backed by a lack of a significant
effect of the counterfactual manipulation on mood and by the fact
that mood did not reduce the effect of counterfactual primes on
executive thinking style. In addition, the observation that the
no-valence control was identical to the valence-control conditions
gives us confidence that the remaining experiments in which
valence-based controls were used establish valid baseline
conditions.
Experiment 2: Counterfactual Mind-Sets and Analytical
Reasoning
The demonstration that counterfactual mind-sets affect mental
states in a manner consistent with a relational processing style was
our first step in understanding the process through which counter-
factual mind-sets have their impact. The second step is to explore
how counterfactual mind-sets impact analytical reasoning abilities.
Because the analytical process involves the examination and iden-
tification of logical relationships between task variables, we ex-
pected the relational processing style activated by counterfactual
thoughts to facilitate analytical reasoning abilities.
To test this hypothesis, we had participants complete a version
of the Law School Admission Test (LSAT) analytical reasoning
section. The LSAT analytical reasoning section assesses one’s
ability to understand and apply rules, determine relationships be-
tween concepts, analyze situations and draw conclusions, and
apply logic to ambiguous or complex situations (Princeton Re-
view, 2005). Specifically, this LSAT section consists of problems
designed to “simulate the kinds of detailed analyses of relation-
ships that a law student must perform in solving legal problems”
(Law School Admission Council, 2005, p. 10). For example, one
might be told to arrange guests at a dinner party and be given rules,
including who may sit next to whom. These rules are followed by
several questions that examine one’s understanding of the relation-
ships between dinner guests. If counterfactual mind-sets promote a
relational processing style, then invoking a counterfactual mind-set
should be positively associated with LSAT analytical reasoning
performance.
 COUNTERFACTUAL MIND-SETS
Method
Design. The experiment had a 2 (type of prime: counterfactual vs.
noncounterfactual) ⫻ 2 (valence: positive vs. negative) between-subjects
factorial design. We also included an additional control condition in which
participants simply took the LSAT exam without first reading a scenario.
Participants. Participants were 135 students from a large midwestern
university campus. Participants were paid $10 an hour for their
participation.
Procedure. Participants were greeted in the laboratory by an experi-
menter who explained that they would complete several questionnaires
related to decision making. The experimental manipulations and the de-
pendent variables were embedded in one large packet of questionnaires.
Participants were randomly assigned to one of the five experimental
conditions described in Experiment 1b. After reading one of the scenarios,
all participants were asked to consider some thoughts likely to be running
through Jane’s mind and then to indicate their current mood on a 9-point
scale ranging from 1 (very negative) to 9 (very positive). Finally, partici-
pants completed one section of the LSAT analytical reasoning test.
LSAT analytical reasoning problems. Each analytical reasoning sec-
tion of the LSAT has four games with 24 questions that must be accurately
diagrammed in order to be answered correctly. Our main dependent vari-
able was task performance. We followed LSAT scoring guidelines within
which the overall score consists of the number of questions answered
correctly adjusted for the number of guesses. We also measured task effort,
as gauged by the number of attempted answers.
Results and Discussion
Counterfactual activation. Two independent coders identified
the number of counterfactual thoughts in relation to the rock
concert scenario. The reliability for counterfactual thoughts was
high (␣ ⫽ .95), and, therefore, the ratings of the two coders were
averaged. We submitted the number of counterfactual thoughts to
a 2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2
(valence: positive vs. negative) between-subjects factorial design.
As expected, counterfactual prime participants (M ⫽ 1.32, SD ⫽
0.66) listed significantly more counterfactual thoughts than non-
counterfactual prime participants (M ⫽ 0.29, SD ⫽ 0.54), F(1,
108) ⫽ 88.08, p ⬍ .001. We also observed a main effect for
valence of prime, with the negative outcome (M ⫽ 0.99, SD ⫽
0.79) generating more counterfactuals than the positive outcome
(M ⫽ 0.62, SD ⫽ 0.75), F(1, 108) ⫽ 10.61, p ⫽ .002. The
interaction between type and valence of prime was not significant,
F(1, 108) ⫽ 0.71, p ⫽ .40.
Task performance. To measure analytical performance, we
assessed the LSAT score using an ANOVA, which included the
type of prime and valence as factors. We hypothesized that coun-
terfactual mind-sets would improve analytical reasoning perfor-
mance. Consistent with this hypothesis, counterfactual mind-set
participants (M ⫽ 7.92, SD ⫽ 2.74) outperformed noncounterfac-
tual mind-set participants (M ⫽ 6.09, SD ⫽ 3.45), F(1, 108) ⫽
9.51, p ⫽ .003.3 No other effects emerged as statistically signifi-
cant for this analysis.
Task effort. To determine whether our manipulations affected
task effort, we conducted an ANOVA in which the dependent
variable was the number of LSAT items attempted, and the inde-
pendent variables were the type of prime and valence. Results
indicated an effect for the valence of the prime such that partici-
pants exposed to the positive emotional experience attempted more
LSAT questions (M ⫽ 15.35, SD ⫽ 4.50) than those exposed to the
39
negative emotional experience (M ⫽ 13.30, SD ⫽ 4.65), F(1,
108) ⫽ 5.53, p ⫽ .02. No other results emerged as statistically
significant.
Relationship between counterfactual activation and LSAT per-
formance. We next examined whether the amount of counterfac-
tual activation mediated the relationship between the mutable
primes and LSAT performance. When we regressed LSAT score
on both type of counterfactual prime and number of counterfactual
thoughts listed, number of counterfactual thoughts was statistically
significant (␤ ⫽ .33), t(109) ⫽ 2.82, p ⫽ .006, but not type of
prime (␤ ⫽ .07), t(109) ⫽ 0.58, ns. A Sobel test determined that
the reduction in the significance level was statistically significant
(Z ⫽ 3.85, p ⬍ .001). This finding suggests that the generation of
counterfactual thoughts accounted for the relationship between the
mutable scenario and performance.
Supplementary control condition analyses. We also conducted
analyses that included the additional no-valence control condition
(M ⫽ 5.95, SD ⫽ 3.84) to build confidence that our counterfactual
manipulation was responsible for the better LSAT performance.
To do so, we conducted planned contrasts comparing the two
counterfactual conditions with the three control conditions, which
revealed that participants in the counterfactual conditions per-
formed better than those in the control conditions, t(130) ⫽ 3.28,
p ⫽ .001. Additionally, we tested whether the additional control
differed from the two original control conditions and found no
significant differences for their effects on LSAT performance,
t(130) ⫽ ⫺0.18, ns.
Supplementary mood analyses. We sought to determine
whether mood played a role in the relationship between counter-
factual mind-sets and LSAT task effort and performance. To do so,
we first conducted an ANOVA, with mood as the dependent
variable and type of prime and valence of prime as between-
subjects factors. No effects were statistically significant. We also
conducted analyses of covariance using LSAT task effort (LSAT
attempted) and task performance (overall LSAT score) as the
dependent variables, type of prime and valence as our independent
variables, and mood as a covariate. For the LSAT number at-
tempted, the main effect for valence remained reliable, F(1,
107) ⫽ 5.49 p ⬍ .05. For the overall LSAT score, the main effect
of type of prime remained reliable, F(1, 107) ⫽ 7.16, p ⫽ .009.
These findings suggest that mood did not mediate the effect of
counterfactual prime or valence on LSAT task effort or task
performance.
One goal of the present study was to provide support for the
assertion that counterfactual mind-sets promote a relational pro-
cessing style, thereby facilitating the recognition and better under-
standing of ambiguous and complex relationships. In support of
this hypothesis, we found that participants in a counterfactual
mind-set outperformed participants who were not in a counterfac-
tual mind-set on the LSAT analytic reasoning test. In combination
with the results of Experiment 1, which demonstrated that the
3
We also tested our hypothesis by examining only the number of correct
LSAT items (unadjusted for guesses). We found the same pattern of
results, with participants in a counterfactual mind-set (M ⫽ 9.21, SD ⫽
2.61) outperforming noncounterfactual mind-set participants (M ⫽ 7.71,
SD ⫽ 3.20), F(1, 108) ⫽ 7.42, p ⫽ .008.
40 KRAY, GALINSKY, AND WONG
activation of a counterfactual mind-set was predictive of phenom-
enologies and preferences consistent with a relational processing
style, Experiment 2 demonstrated that performance on a task
measuring one’s ability to understand logical relationships and
make connections between task stimuli is improved by counter-
factual mind-sets.
Experiment 3: Counterfactual Mind-Sets and Structured
Imagination
Thus far, we have provided evidence consistent with the idea
that counterfactual mind-sets promote a relational processing style
in terms of both mental states and analytical reasoning perfor-
mance. Because we expect the effects of a relational processing
style to be quite broad, in Experiment 3, we explore whether and
how counterfactual mind-sets influence the tendency to build on
existing knowledge structures during the creative generation pro-
cess, or structured imagination.
Intuitively it might seem that thoughts of “if only” associated
with counterfactual thinking would facilitate creative generation.
The construction of alternative possible worlds would seem to be
the epitome of creative generation, and it is easy to speculate that
counterfactual thinking would encourage imagination to roam un-
fettered and unencumbered by mental constraints. To the contrary,
we contend that the counterfactual mind-sets may not be the
springboard to freewheeling generation. A key insight underlying
this counterintuitive hypothesis concerns the fact that logical rules
and clear structure govern when and how counterfactuals are
constructed. As we have repeatedly demonstrated, counterfactual
mind-sets have their impact regardless of the content or direction
of the preceding counterfactual thoughts. This observation sug-
gests that what lingers following the construction of counterfactual
thoughts is their underlying logical form. We argue that the mental
structure of logical relationships created through counterfactual
thought increases the tendency to structure thought around salient
associations and the pursuit of connections.
In addition to the conceptualization described above, the coun-
terfactual mind-sets increase a preference for structured thinking,
as was empirically demonstrated in Experiment 1b. Structured
thinking differs from the mental states that encourage creative
generation, which generally requires an expansion of conceptual
attention that goes beyond the bounds of what is presently known
or salient (Guilford, 1950). Some have even characterized the
mind-set that encourages creative generation as a “risky” process-
ing style (Friedman & Forster, 2001; quotations in original). Not-
ing the connections and relationships between stimuli and making
structured associations, as counterfactual participants did with
relative ease in Experiment 2, may inhibit the tendency to go
beyond what is already known or salient, and thereby impair
creative generation.
Creative generation tasks typically elicit responses that are
loosely defined, which creates the potential for an infinite number
of unique responses (Guilford, 1950). However, performance on
creative generation tasks tends to be impaired because participants
borrow from and rely too heavily on existing knowledge struc-
tures. For example, participants instructed to create creatures “be-
yond their wildest imagination” tended to produce results that
conformed to the attributes of realistic earth creatures, including
bilateral symmetry, and ordinary sensory receptors and append-
ages (Ward, 1994). On the basis of Ward’s concept of structured
imagination, we expected participants in a counterfactual mind-set
to adopt a more structured approach to generating creative output
than under baseline conditions. Given the relational processing
style characteristic of counterfactual thinking, the resulting mind-
set should increase the tendency to structure one’s imagination
around existing knowledge structures. That is, in contrast to the
“thinking outside the box” characteristic of creative generation
processes, we expected counterfactual mind-sets to promote
“thinking within the box.”
Method
Design. The experiment had a 2 (type of prime: counterfactual vs.
noncounterfactual) ⫻ 2 (valence: positive vs. negative) ⫻ 2 (type of
scenario: rock concert vs. spelling bee) between-subjects factorial design.
Participants. Participants were 93 students from a large western uni-
versity campus enrolled in an introductory organizational behavior course.
By participating in the experiment, participants received partial credit
toward a class requirement.
Procedure. Participants were greeted in the laboratory by an experi-
menter who told them that they would complete several tasks predictive of
future job performance. Participants first read one version of the two prime
scenarios. As in previous experiments, approximately half the participants
read a version of the rock concert scenario. To ensure that our results
generalize to different instantiations of the counterfactual mind-set, the
other half of participants read a version of a spelling bee scenario, in which
a young boy named Paul competes to advance in the National Junior
Spelling Bee. In each scenario, an individual at the spelling bee advances
to the next round. To manipulate valence, in half of the spelling bee
scenarios, Paul correctly spells the assigned word and advances to the next
round, whereas in the other half, he misspells the word and is eliminated
from the competition. Additionally, half of the scenarios describe a se-
quence of events designed to elicit counterfactual thoughts, whereas the
other half of the scenarios describe a sequence of events not expected to
elicit counterfactual thoughts. In the downward counterfactual scenario,
Paul advances to the next round of the competition after his place in line
is altered (because he had to use the restroom), and he is asked to spell a
word he knows (but had he stayed in his original place in line, he would
have been given a word he did not know how to spell). In the upward
counterfactual scenario, Paul is eliminated from the competition after his
place in line is altered, and he is asked to spell a word he does not know
(but had he stayed in his original place in line, he would have been given
a word he knew how to spell). In the noncounterfactual conditions, he
either spells or misspells a word but does not alter his place in line. After
reading the scenario and listing thoughts in the protagonists’ mind, partic-
ipants began the creative task.
Creative task. Following Ward (1994), we asked participants to “imag-
ine going to another galaxy in the universe and visiting a planet very
different from earth” and to spend 7 min drawing a picture of an animal
that is local to this planet. Immediately upon completion of the drawing
task, participants completed a questionnaire in which they provided a short
written description of how they went about approaching this creative task
and then evaluated their approach along several domains. Participants’
evaluations of their approach to the task were used to determine the extent
to which they engaged in structured imagination. Specifically, they indi-
cated the extent to which they considered the following five items of
knowledge: known science fiction creatures, general attributes of science
fiction creatures, known earth animals, general attributes of earth animals,
and consideration for the local environment. Each item was rated on a
7-point scale ranging from 1 (not at all) to 7 (completely).
 COUNTERFACTUAL MIND-SETS
Creative coding. Because Ward’s (1994) study suggests potential ceil-
ing effects for some characteristics of the imagined creatures (e.g., bilateral
symmetry) and demonstrated greater variability for atypical sensory or-
gans, our dependent variable for the drawings was the atypicality of the
sensory organs.4 Three trained coders who were blind to the study hypoth-
eses coded the drawings and descriptions for atypical sensory organs.
Following Ward’s coding, sensory organs were considered atypical if they
(a) lacked a major sensory organ (i.e., eyes, ears, nose), (b) had atypical
numbers of a sensory organ (e.g., three eyes), (c) demonstrated an unusual
configuration of the senses (e.g., eyes located below the nose), (d) had an
exaggerated or unusual ability (e.g., eyes that had laser beams), or (e)
served an atypical function (e.g., ears for protection). The total number of
atypicalities were tallied for each participant. The codings for the drawing
were highly reliable (␣ ⫽ .93), as were the codings for the descriptions
(␣ ⫽ .89), and discrepancies were resolved through discussion. The coders’
data were averaged to create one measure for sensory atypicality in the
drawing and one measure for sensory atypicality in the description.
Results and Discussion
Counterfactual activation. Three independent coders identi-
fied the number of counterfactual thoughts after the rock concert
and spelling bee scenarios. The reliability for both was high (␣s ⫽
.88 and .87, respectively), so the ratings were averaged. We
submitted the number of counterfactual thoughts to a 2 (type of
prime: counterfactual vs. noncounterfactual) ⫻ 2 (valence: posi-
tive vs. negative) ⫻ 2 (type of scenario: rock concert vs. spelling
bee) between-subjects ANOVA. As expected, counterfactual
prime participants (M ⫽ 1.32, SD ⫽ 0.96) listed significantly more
counterfactual thoughts than noncounterfactual prime participants
(M ⫽ 0.94, SD ⫽ 0.71), F(1, 85) ⫽ 4.00, p ⬍ .05. Participants
exposed to a negatively valenced event (M ⫽ 1.33, SD ⫽ 0.77)
generated more counterfactual thoughts than participants exposed
to a positively valenced event (M ⫽ 0.93, SD ⫽ 0.90), F(1, 85) ⫽
6.30, p ⫽ .01. No other effects were statistically significant.
Self-reported structured imagination. We expected partici-
pants in the counterfactual mind-set condition to have a more
structured imagination than noncounterfactual mind-set partici-
pants. To test our hypothesis, we conducted a multivariate
ANOVA, including each of the five statements described above, as
well as type of prime, valence of prime, and type of scenario as
between-subjects factors. Consistent with our hypothesis, the only
statistically significant effect to emerge was a tendency for coun-
terfactual mind-set participants (M ⫽ 4.49, SD ⫽ 0.77) to report
more structured imagination than noncounterfactual mind-set par-
ticipants (M ⫽ 4.04, SD ⫽ 0.85), F(5, 80) ⫽ 3.18, p ⫽ .01. An
examination of the univariate effects revealed that counterfactual
mind-sets promoted a reliance on specific science fiction creatures,
F(1, 84) ⫽ 4.34, p ⬍ .04; general attributes of science fiction
creatures, F(1, 84) ⫽ 4.28, p ⬍ .05; and local environment
considerations, F(1, 88) ⫽ 8.18, p ⬍ .01. No other effects were
statistically significant.
Creative coding. Although the self-report results support our
hypothesis, we further tested it by examining the actual drawings
and drawing descriptions. To do so, we used a mixed-model
ANOVA, with type of prime, valence of prime, and type of
scenario as between-subjects factors and type of coding (drawings
vs. descriptions) as a repeated measure. In support of our hypoth-
esis, counterfactual mind-set participants (M ⫽ 1.11, SD ⫽ 1.00)
were rated as incorporating fewer sensory atypicalities into their
41
drawings and descriptions than noncounterfactual mind-set partic-
ipants (M ⫽ 1.69, SD ⫽ 1.61), F(1, 85) ⫽ 4.92, p ⬍ .03. No other
effects were statistically significant.5
The primary goal of this study was to demonstrate that the
relational processing style associated with counterfactual mind-
sets can lead to “thinking within the box.” The results of this
experiment are consistent with our hypothesis. Participants in a
counterfactual mind-set reported structuring their imaginative pro-
cess around existing knowledge structures to a larger degree than
control participants. In addition, they showed evidence of consid-
ering the local environment in constructing their drawings. Finally,
participants in a counterfactual mind-set incorporated fewer atypi-
calities into their drawings and descriptions than participants in the
noncounterfactual mind-set. Together, these findings reinforce the
notion that counterfactual mind-sets promote a form of structured
imagination.
A secondary goal of the present study was to demonstrate that
the results found in the previous studies were not driven by the
nature of the rock concert scenario. To do so, we included a
different scenario that involved a mutable event at a spelling bee.
Regardless of which mutable scenario participants read, counter-
factual mind-sets promoted a form of structured imagination. This
observation gives us confidence that our results thus far generalize
to different instantiations of the counterfactual mind-set.
Experiment 4: Counterfactual Mind-Sets and Creative
Generation
In Experiment 4, we further explore whether counterfactual
mind-sets impact the creative generation process. Because the
drawings created in the previous experiment were highly variable,
and it was difficult to determine what aspects of science fiction
creatures counterfactual mind-sets might have been borrowing
from, in the present experiment, we used a creative generation task
that can be more easily coded for structured imagination.
In the previous experiments, participants reported relying on
existing knowledge and examples in constructing their alien crea-
tures. The presence of salient examples in a creative generation
task leads to less novel solutions than when examples are not
provided a priori (Smith, Ward, & Schumacher, 1993). If coun-
terfactual mind-sets facilitate a relational processing style that
seeks connections with readily available cognitive representations,
then participants in this mind-set should be more attentive to and
influenced by a set of provided examples. As a result, the output of
participants in a counterfactual mind-set should be less novel than
baseline conditions. To test this hypothesis, we examined perfor-
mance on a creative generation task involving the creation of novel
product labels. Prior to starting the task, participants were provided
with a set of examples. In the present experiment, we measured the
degree to which individuals were able to ignore these examples in
4
Our data replicate this ceiling effect. For instance, 92% of the drawings
demonstrated bilateral symmetry, 1% did not, and the remaining 7% were
excluded from the data analysis because the drawings were profiles, and
bilateral symmetry could not be determined.
5
We also examined whether participants differed in terms of appendage
atypicalities in their drawings and descriptions, but no effects emerged as
statistically significant.
42 KRAY, GALINSKY, AND WONG
their creation of labels for the new products. That is, we measured
the extent to which participants’ own product labels resembled the
examples we provided them.
In addition to coding for resemblance to examples, we also
coded the product labels on two dimensions: overall creativity and
descriptiveness of the labels in representing the product category.
In the previous experiment, participants in a counterfactual mind-
set reported that they were more likely to consider the local
environment in drawing their space aliens. This suggests that
counterfactual mind-sets gear participants toward structuring
imagination to reveal something about the essence of the object. If
this is true, then counterfactual mind-sets should produce product
labels that are representative of the product. We predicted the
counterfactual mind-set participants would both draw on the given
examples and be more likely to produce labels that were more
descriptive of the product. However, to the extent that the labels
were not judged to be particularly novel, they should suffer in
terms of their perceived creativity.
Method
Overview and design. The experiment had a 2 (type of prime: coun-
terfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs.
negative) between-subjects factorial design.
Participants. Participants were 29 undergraduate business students
from a large western university campus enrolled in an introductory orga-
nizational behavior course. The experiment was conducted outside of the
classroom setting. By participating in the experiment, partial fulfillment of
a course requirement was granted.
Procedure. Participants were greeted in the laboratory by an experi-
menter who explained that they would complete several questionnaires
related to decision making. The experimental manipulations and our de-
pendent variables were embedded in the packet of materials that partici-
pants received. Participants were given up to 30 min to complete the packet
of materials.
Experimental manipulations. The manipulations were identical to
those used in Experiment 1a. Because Experiment 1b revealed that the
strength of the counterfactual mind-set is not dependent on whether
thoughts are listed, we chose not to have participants list their thoughts for
this experiment. Instead, participants were simply asked to ponder thoughts
running through the protagonist’s mind.
Creative generation task. We used a modified version of the creative
generation task, described in Rubin, Stoltzfus, and Wall (1991). Partici-
pants were asked to imagine that they were interviewing with a top
marketing firm, and part of the interview involved testing their aptitude for
business. To do so, participants were tasked with creating new labels for
new products. Specifically, they were instructed to create at least one (and
up to three) new label for each of three categories of products (pasta,
nuclear element, pain reliever). Six examples were provided for each
category, as described in Appendix A. To encourage creative output,
participants were advised not to use or copy aspects of the examples
provided.
For each of the categories, the examples provided had two common
endings, which are defined as a letter or cluster of letters that ended at least
one multisyllabic word. For example, all of the examples provided of
nuclear elements ended in _on or _ium (e.g., radon, plutonium, argon,
carbon, radium, uranium). Creative output was operationalized in terms of
the number of product names created for each category that did not share
the word endings of the examples. We also examined the sheer number of
product names created for each category to determine whether motivational
differences were evident.
Creative generation coding. Three independent coders evaluated each
idea generated by participants on 9-point scales ranging from 1 (not at all)
to 9 (extremely) for creativeness and descriptiveness. Specifically, coders
were asked to consider “How creative is this response?” and “How de-
scriptive is this response in revealing the type of product?”
Results and Discussion
Creative generation. To examine the novelty of name gener-
ation, we submitted the total number of names with endings
deviating from the supplied examples to a 2 (type of prime:
counterfactual vs. noncounterfactual) ⫻ 2 (valence of prime: pos-
itive vs. negative) ⫻ 3 (product category: pasta vs. nuclear element
vs. pain reliever) mixed model ANOVA, with repeated measures
on the third factor. The only significant effect to emerge from this
analysis was a main effect for counterfactual prime, F(1, 25) ⫽
4.92, p ⬍ .05. Across the three categories, counterfactual mind-set
participants (M ⫽ 1.26, SD ⫽ 1.44) generated significantly fewer
novel names than noncounterfactual mind-set participants (M ⫽
2.70, SD ⫽ 2.20). We also conducted a similar analysis examining
the raw number of ideas generated and observed no statistically
significant effects, suggesting our experimental manipulation did
not impact effort.
Creative versus descriptive ratings. Three independent coders
who were blind to condition and our hypotheses evaluated each
idea generated by participants for creativeness and descriptiveness.
Reliabilities were good (␣ ⫽ .72 and .73, respectively), so we
combined the ratings of each coder. We submitted the ratings to a
2 (type of prime: counterfactual vs. noncounterfactual) ⫻ 2 (va-
lence of prime: positive vs. negative) ⫻ 2 (rating: creativity vs.
descriptiveness) mixed model ANOVA, with repeated measures
on the third factor. A main effect for type of coding emerged,
indicating that labels were judged to be more descriptive (M ⫽
5.75, SD ⫽ 0.69) than they were creative (M ⫽ 5.09, SD ⫽ 0.76),
F(1, 25) ⫽ 8.61, p ⬍ .01. More important for testing our hypoth-
esis was the statistically significant two-way interaction between
type of prime and type of coding, F(1, 25) ⫽ 4.66, p ⬍ .05.
Counterfactual prime labels (M ⫽ 4.86, SD ⫽ 0.54) were judged
to be less creative than noncounterfactual prime labels (M ⫽ 5.34,
SD ⫽ 0.90), t(27) ⫽ 1.76, p ⫽ .09; yet, counterfactual prime labels
(M ⫽ 5.97, SD ⫽ 0.59) were judged to be more descriptive than
noncounterfactual prime labels (M ⫽ 5.51, SD ⫽ 0.73), t(27) ⫽
1.86, p ⫽ .07. Counterfactual mind-sets led to more descriptive
labels that were nonetheless lacking in creativity. No other effects
emerged as statistically significant for this analysis.
The results of this experiment support our hypothesis that coun-
terfactual mind-sets can impair creative generation. Individuals
who had previously pondered counterfactual thoughts generated
new product labels that were less novel than individuals who had
not previously pondered counterfactual thoughts. This pattern
emerged regardless of whether novelty was judged in terms of the
similarity to the provided examples or a global evaluation of
creativity by independent coders. The fact that counterfactual
thinking did not impact the number of names generated suggests
sheer effort was not responsible for this effect.
Although participants were instructed to be as creative as pos-
sible and to refrain from borrowing from the examples provided,
counterfactual mind-set participants were less effective at breaking
out of the mold set by the examples. In essence, the relational
 COUNTERFACTUAL MIND-SETS
processing style characteristic of the counterfactual mind-set pro-
moted structured imagination, or thinking within the box. How-
ever, within that box, counterfactual mind-sets led to more de-
scriptive and potentially useful labels. Like participants in the
previous experiment who reported considering the local environ-
ment when drawing space aliens, counterfactual mind-sets led
participants to consider the essence of what the product was about.
The finding that labels generated by counterfactual mind-set par-
ticipants were actually judged to be more descriptive of the prod-
uct category than labels generated by control participants is con-
sistent with the hypothesis that counterfactual mind-sets promote a
relational processing style. That is, a heightened attention to the
attributes generally associated with the categories for which labels
were created (i.e., pasta) led to the generation of labels that were
deemed to be representative of the category by lay judges.
Experiment 5: Creative Generation Versus Creative
Association
We demonstrated in the previous experiment that creative gen-
eration is impaired following the invocation of a counterfactual
mind-set. Measuring creativity in terms of idea generation is but
one approach, as creativity is generally regarded to be a multidi-
mensional construct (Amabile, 1983). The purpose of Experiment
5 was to determine whether counterfactual mind-sets might im-
prove performance on creative association tasks that involve the
consideration of relationships and making connections between
disparate knowledge structures.
If counterfactual mind-sets promote a relational processing style
involving the consideration of relationships between task stimuli,
then the mind-set should improve performance on creative tasks
involving the identification of unusual associations between stim-
uli, associations that are adaptive and responsive to the present
context. Galinsky and Moskowitz (2000) provided some evidence
consistent with this hypothesis with respect to the Duncker candle
problem. This task gauges the ability of individuals to overcome
functional fixedness, characterized by a failure to recognize a use
for a particular object in a given context as a result of a fixation on
its typical use (Duncker, 1945). An individual is given a candle, a
box of tacks, and a book of matches and challenged with affixing
the candle to the wall so that it can be lit without dripping wax onto
the floor or wall. Because the box initially functions as a container
for tacks, the problem solver often fails to recognize its potential
as a solution to the problem: The tacks are dumped out of the box,
which is affixed to the wall with a couple of tacks, and the candle
is placed on top of or in the box and lit. The solution to the task
requires the problem solver to see potential relationships other than
the obvious ones—the box is not just a repository for tacks but can
also be used as stand (Glucksberg & Weisberg, 1966). In addition,
the solution involves recognizing a relationship between the candle
and the box. Galinsky and Moskowitz (2000) showed that, by
invoking a counterfactual mind-set prior to engaging in the task,
problem solvers were more likely to recognize the potential use of
the box of tacks as a platform for the candle relative to control
participants.
We argue that performance on the Duncker candle problem was
facilitated by the counterfactual mind-set because it promoted a
tendency to explore the possible relationships between the critical
43
objects. In the present experiment, we sought further evidence in
favor of this interpretation by examining performance on another
creative association task, the remote associates task (RAT; M. T.
Mednick, Mednick, & Mednick, 1964). The RAT requires an
individual to form “mutually distant associative elements into new
combinations which are useful and meet specified as well as
unforeseen requirements” (S. A. Mednick, 1962). Specifically, the
test requires identifying a unique association among three distinct
words. For example, the common link for the words sore—shoul-
der—sweat is cold. Similar to the Duncker candle problem, by
considering the relationships between task stimuli, performance on
the RAT improves.
In the present experiment, we used a within-subject design so
that we could explore the relative effect of counterfactual mind-
sets on a creative generation task versus a creative association task.
We aimed to replicate the facilitative effect of the counterfactual
mind-set observed previously with the Duncker candle for the
RAT and also demonstrate a replication of the impairing effect for
the creative generation task used in Experiment 4. By showing
both facilitation and debilitation across tasks but within individu-
als, we aimed to provide strong evidence that the counterfactual
mind-set promotes a relational processing style, which has wide-
ranging effects.
Method
Overview and design. The experiment had a 2 (type of prime: coun-
terfactual vs. noncounterfactual) ⫻ 2 (valence of prime: positive vs.
negative) ⫻ 2 (type of creative task: association vs. generation) mixed
design, with repeated measures on the third factor.
Participants. Participants were 50 undergraduate students at a mid-
western university who were recruited through e-mail solicitations; they
participated in the experiment, along with several other unrelated tasks.
They were compensated $15 for their time.
Procedure. Participants were greeted in the laboratory by an experi-
menter who explained that they would complete several tasks assessing
ability in business contexts. Participants were given a packet of materials
that contained the experimental manipulations, a modified version of the
RAT (M. T. Mednick et al., 1964), and the identical creative generation
task used in Experiment 4. All participants read the prime scenarios and
then completed the RAT first before completing the creative generative
task.6 Participants were instructed to complete each task before proceeding
to the next one. Participants were allowed to work on each task until they
had finished or could not answer any more questions. Finally, participants
were debriefed and then proceeded to work on several unrelated tasks.
Experimental primes. The primes were the four rock concert scenarios
used in the previous experiments. As in Experiment 4, participants simply
pondered thoughts running through the protagonist’s mind without listing
any thoughts.
The RAT (M. T. Mednick et al., 1964). The RAT is designed to
measure the creative ingenuity of individuals and requires identifying a
unique common denominator among three distinct words. We shortened
the original task designed by Mednick et al. to include only 10 items (see
Appendix B for the items). Consistent with our analysis of the generative
6
Although we did not counterbalance the order of tasks in the present
experiment, the findings of Experiment 3 established that performance on
the creative generation task was negatively impacted when the task imme-
diately followed the counterfactual manipulation.
44 KRAY, GALINSKY, AND WONG

cognitive task, the dependent variables for the RAT included the number of
attempted items and the number of correct items.

Results
First, because the effect of our experimental manipulations did
not vary depending on product type in the generative cognitive
task (replicating the results of Experiment 4), we collapsed across
product type in the present set of analyses. To facilitate a compar-
ison of performance across the two creative tasks, we first com-
puted z scores separately for each task and then submitted each z
score to a 2 (type of prime: counterfactual vs. noncounterfac-
tual) ⫻ 2 (valence of prime: positive vs. negative) ⫻ 2 (type of
task: creative association vs. creative generation task) mixed
model ANOVA, with repeated measures on the third factor.
Number of correct items and novel names. To measure cre-
ative performance on the two tasks, we assessed the number of
correct associations on the RAT and the number of novel names
generated on the new products task. Consistent with our hypoth-
esis, the only effect to emerge as statistically significant was the
expected two-way interaction between type of task and counter-
factual prime, F(1, 45) ⫽ 22.37, p ⬍ .001 (see Figure 1). Coun-
terfactual mind-set participants (M ⫽ 0.43, SD ⫽ 0.78) outper-
formed noncounterfactual mind-set participants (M ⫽ ⫺0.48,
SD ⫽ 1.02) on the RAT, F(1, 49) ⫽ 12.51, p ⬍ .01, yet counter-
factual mind-set participants (M ⫽ ⫺0.34, SD ⫽ 0.68) performed
worse than noncounterfactual mind-set participants (M ⫽ 0.40,
SD ⫽ 1.18) on the creative generative task, F(1, 50) ⫽ 7.97, p ⬍
.01.7 Counterfactual mind-sets improved performance on a task
requiring the identification of a common association for a set of
words but impaired performance on a task requiring the generation
of novel ideas. No other effects emerged as statistically significant
for this analysis.
Number of items attempted and labels generated. We also
looked at task effort in terms of items attempted on the RAT and
labels generated on the new products task. The only effect to
emerge was a two-way interaction between task and counterfactual
prime that approached significance, F(1, 45) ⫽ 3.72, p ⫽ .06 (see Figure 1. Mean performance as a function of type of task and type of
prime in Experiment 5. CF ⫽ counterfactual condition; NCF ⫽ non-
Figure 1). Whereas the number of items attempted on the RAT was
counterfactual condition; RAT ⫽ remote associates task.
greater in the counterfactual condition (M ⫽ 0.36) than in the
noncounterfactual condition (M ⫽ ⫺0.40), F(1, 49) ⫽ 8.16, p ⬍
.01, the difference between the two counterfactual conditions was
not statistically significant in the creative generation task (Ms ⫽ the new products labels task. The fact that both tasks measure
⫺0.05 and 0.02, F ⬍ 1, ns). No other effects emerged as statisti- aspects of creativity suggests that the debilitating effect of coun-
cally significant for this analysis. terfactual mind-sets is limited to thinking “outside the box.”

Discussion General Discussion

The question of whether thoughts about alternate worlds borne
The results of this study provide strong support for the assertion
out of mutated pasts impact the future has important theoretical
that counterfactual mind-sets promote a relational processing style,
and applied implications. Although the adage “What’s done is
which has differential effects on creative tasks measuring the
done” may suggest that pondering the past is an unproductive use
generation of novel ideas versus the identification of associations.
of time, the present research provides strong evidence to suggest
Rather than uniformly impairing creative performance, the effect
that imagining alternatives to past realities by considering a dif-
of counterfactual mind-sets appears to depend on the underlying
creative process being assessed. Specifically, counterfactual mind-
sets improved the ability of participants to identify unusual and 7
The relationship between performance on the creative generation task
useful associations between sets of words in the RAT, yet caused and the creative association task was not statistically significant, r(49) ⫽
participants to borrow too heavily from the provided examples in ⫺.12, ns; this relationship did not depend on experimental condition.
 COUNTERFACTUAL MIND-SETS
ferent path, choice, or action has a powerful impact on how future
analytic and creative problems are solved. In particular, generating
counterfactuals in one context appears to alter thought processes to
be more relational in subsequent contexts, despite the new con-
text’s irrelevance to the imagined world. Independent of the con-
tent or valence of the imagined world, the act of generating
counterfactuals produces a lingering tendency to consider relation-
ships and associations and to problem solve from within existing
frameworks.
The present article provides a range of evidence to support the
hypothesis that counterfactual mind-sets promote a relational pro-
cessing style. We demonstrated in Experiment 1 that counterfac-
tual mind-sets increase a sense of being poised for analytic and
critical thinking and preferences for a structured style of thinking.
Evidence that the counterfactual mind-set improves performance
on an analytic task involving the assessment of relationships
between task variables was provided in Experiment 2. We dem-
onstrated in Experiments 3 and 4 that counterfactual mind-sets
increase the tendency to structure imagination around existing
knowledge structures, leading to more descriptiveness but less
novelty. Finally, using a within-subject design, we demonstrated in
Experiment 5 that the heightened tendency to build on existing
knowledge structures following the activation of the counterfactual
mind-set leads to better performance on a creative association task
involving the consideration of associations between task stimuli
but to worse performance on a creative generation task.
The counterfactual thought-listing methodology used in Exper-
iments 1b and 2 also shed light on the process through which
counterfactual mind-sets promote a relational processing style. In
Experiment 1b, we demonstrated that the effect of counterfactual
mind-sets on executive thinking style preferences was mediated by
counterfactual activation. In Experiment 2, we demonstrated that
the listing of counterfactual thoughts mediated the relationship
between the mutable prime and LSAT performance. In combina-
tion with previous research in which a similar methodology was
used (Galinsky & Kray, 2004; Kray & Galinsky, 2003), the
amount of counterfactual activation has proved to be a robust
mediating mechanism in the relationship between mutable primes
and performance. The greater the number of counterfactual
thoughts generated following exposure to a scenario in which an
event almost turned out differently, the more ingrained the mental
structure of logical relationships becomes, thus increasing prefer-
ences for structure and performance on analytical tasks.
Counterfactual primes influence not just what we think but how
we think. The present experiments consistently demonstrated that
what we think, as determined by the direction of the counterfactual
elicited (upward or downward), did not moderate any of the effects
of how we think, as determined by the mutable nature of the prime.
Across two experiments (1a and 2), we also demonstrated that the
valence-based conditions produced results identical to both a base-
line condition without a pretask scenario and a pretask scenario in
which no valence-based event occurs. This finding suggests that
the valence of the outcome did not impact the results. Finally, the
fact that measurements of mood did not differ across our counter-
factual prime manipulation across multiple experiments gives us
more confidence that it is the process of thinking counterfactually,
and not the content of the counterfactuals, that is responsible for
the observed effects. In total, counterfactual mind-sets and the
45
differential emotional experiences associated with upward versus
downward counterfactuals appear to operate independently on
subsequent thinking styles and cognitive processing.
The present article is important because it is the first to our
knowledge that delves directly into the phenomenological experi-
ence of a counterfactual mind-set. The findings of Experiment 1a
and 1b suggest that counterfactuals exert a powerful impact on
how individuals perceive their own cognitive state but do not
appear to affect perceptions of affective states. Previously, re-
searchers have argued that counterfactual mind-sets involve a state
of heightened awareness of multiple possible worlds, thereby
promoting mental simulations. But mental simulations brought
about by a salient counterfactual tend not to be free-form. Instead,
they follow systematic laws of mutability that involve tweaking
particular aspects of the counterfactual context to undo a known
outcome, thereby promoting a consideration of cause– effect rela-
tionships. The present research suggests that counterfactual think-
ing primes a relational processing style in subsequent contexts that
facilitate the examination of the relationship between clues, cues,
examples, and props embedded within a problem-solving task.
The observation that the counterfactual mind-set promotes a
relational processing style sheds light on one detrimental, yet
seemingly anomalous, effect resulting from its activation. Specif-
ically, counterfactual mind-sets have been shown to impair per-
formance on the Wason (1966) card selection task involving four
cards, each bearing a symbol: “E, K, 4, 7” (Byrne & Tasso, 1994;
Galinsky & Moskowitz, 2000). The challenge is to determine what
cards must be turned over to determine whether the following
conditional statement is true: “If a card has a vowel on one side,
then it has an even number on the other side.” Both sets of
researchers independently found that counterfactual thinking im-
paired performance by leading participants to incorrectly select the
“4” card. Because the conditional statement to be tested is not
bidirectional, the selection of this card is an error of commission.
The conditional statement central to the task can be misinterpreted
to imply a bidirectional hypothesis (Byrne & Tasso, 1994). We
contend that it is the consideration of the relationship and connec-
tions between the antecedent and consequent characteristic of a
relational processing style that promoted this misguided tendency
to entertain two hypotheses at once. The fact that counterfactual
mind-set participants did not incorrectly solve the problem because
of a failure to select the potentially falsifying “E” and “7” cards or
incorrectly selecting the irrelevant “K” card further supports this
explanation.
We have claimed that counterfactual thinking elicits a particular
cognitive style characterized by a consideration of relationships
and associations. The idea that cognitive styles can differentially
affect performance on tasks that require focus from those that
require flexibility is supported by the theorizing and data of
Peterson and Nemeth (1996). These researchers found that minor-
ity influence encourages flexible thinking and improves perfor-
mance on tasks in which flexibility is an asset but that majority
influence can aid performance on tasks that require focusing on
one dimension of a two-dimension task (e.g., the Stroop task). We
are not claiming that counterfactual thinking is akin to majority
influence but rather suggesting that the work by Peterson and
Nemeth is a useful demonstration that cognitive processing styles
can have a differential influence, depending on whether a task
46 KRAY, GALINSKY, AND WONG
requires flexibility, focus, or consideration of task-embedded
alternatives.
Limitations and Future Directions
The research presented here shows a uniformly negative impact
of counterfactual mind-sets on creative generation tasks. However,
one limitation of the studies presented here is that they only
explore one aspect of the creative generation process. Creative
generation can be broken down into fluency, or the number of
ideas generated, flexibility, or the number of different categories of
ideas represented, and novelty, or the uniqueness of the idea
generated (Amabile, 1983). Perhaps counterfactual mind-sets dif-
ferentially affect these components of creativity. The experiments
presented here provide some evidence that counterfactual mind-
sets decrease novelty, as name generation was more constrained by
the suffixes of the examples after counterfactual primes. Although
no differences were observed in fluency, or the quantity of names
generated overall in Experiments 4 and 5, this null finding may be
a reflection of the fact that the number of names generated had a
limited range (from 1 to 3). The new product labels task also did
not allow for a clear gauge of flexibility, or the number of defin-
able categories of names generated. Given that counterfactual
mind-sets promote a relational processing style that involves the
consideration of relationships and associations, fluency and flex-
ibility may actually be facilitated following the mind-set’s
activation.
The observation that new product names generated by counter-
factual mind-set individuals were judged to be more descriptive of
the product category than the names generated by noncounterfac-
tual mind-set individuals suggests a practical approach to idea
generation when in a counterfactual mind-set. By broadening the
examination to consider the impact of counterfactual thinking on
innovation, which involves both the generation of novel ideas and
their successful implementation (Amabile, Conti, Coon, Lazenby,
& Herron, 1996), researchers may find that the counterfactual
mind-set facilitates the pragmatic process of turning a novel idea
into reality by modeling the execution process after previously
successful ventures.
Future research that explores different instantiations of counter-
factual mind-sets is needed. In all of the experiments in this article,
our manipulation involved the presentation of a scenario that
resulted in the spontaneous generation of counterfactual thoughts
resulting from the presence of a salient mutable component. Indi-
viduals constructed counterfactual thoughts without any direction
or guidance from the experimenter. Whereas the results of this set
of experiments show consistent effects, it is important to consider
their boundaries. For example, other research has explored the
effect of counterfactual thinking after encouraging individuals to
imagine the implications of various “what-if” scenarios of how the
past could have played out differently (cf. Tetlock & Lebow, 2001)
or prefactual considerations of what may be (Gleicher, Boninger,
Strathman, Armor, Hetts, & Ahn, 1995). Mind-sets resulting from
these approaches, which more directly focus thought in an explor-
atory, imaginative direction, may have a beneficial effect on the
subsequent generation of novel ideas. Instead of promoting think-
ing within the box, they may actually facilitate out-of-the-box
thinking. Whether the counterfactual is spontaneously generated or
brought about by “what-if” scenarios may moderate the relation-
ship between counterfactual mind-sets and a relational processing
style.
Another possible direction for future research is to explore
whether different types of counterfactual thoughts have differential
effects on relational processing. A particularly important distinc-
tion may be the additive versus subtractive nature of the counter-
factual. Additive counterfactuals refer to an action that may have
been taken to create an alternate world, whereas subtractive coun-
terfactuals refer to an action that may not have been taken to create
an alternate world (Roese, Hur, & Pennington, 1999). Although
counterfactuals generally aid in making causal judgments, Roese et
al. demonstrated that additive counterfactuals more often express
causal sufficiency, whereas subtractive counterfactuals more often
express causal necessity (McGill, 1998; McGill & Klein, 1993). In
the present research, the most common reaction to the rock concert
scenario was “If only she had not switched seats. . .,” which
expresses a subtractive counterfactual. If the causal necessity as-
sociated with subtractive counterfactuals creates a closer associa-
tion between the antecedent and consequent events than the causal
sufficiency associated with additive counterfactuals, then the re-
sulting relational processing style may be more strongly activated
for the former type of counterfactual than the latter. One conse-
quence of this possibility is that additive counterfactuals may be
more beneficial for creative generation than subtractive counter-
factuals have proved to be.
Given the powerful effect counterfactual thinking has been
shown to have on a wide variety of problem-solving tasks, it is
important to consider the implications of these findings. For ex-
ample, an important practical consideration is how the use of a
counterfactual mind-set as a debiasing technique stacks up against
other procedures for promoting analytic thinking. Procedures that
encourage decision makers to “consider the opposite” (Lord, Lep-
per, & Preston, 1984; Mussweiler, Strack, & Pfeiffer, 2000) or
assign group members the role of a “devil’s advocate” (Cosier,
1978; Janis & Mann, 1977) work by explicitly directing decision
makers to become more critical. However, no explicit training or
assignment of roles is required when a counterfactual mind-set is
subtly activated. Perhaps the activation of a counterfactual mind-
set to promote analytic thinking may be advantageous in delicate
situations in which it is particularly important to avoid the appear-
ance of a heavy hand guiding the decision-making process. Like-
wise, because blatant attempts to restrict an individual’s freedom
often provoke reactance (Brehm, 1966), the counterfactual mind-
set approach to guiding decision-making processes may be more
readily embraced by decision makers than the more directive
approaches described above.
Conclusion
The present set of experiments provides evidence across multi-
ple domains that counterfactual mind-sets promote a relational
processing style, which is characterized by a tendency to consider
relationships and associations among a class of stimuli and to
structure thought and imagination around those associations. As a
result, performance on analytic and creative tasks requiring the
identification of logical relationships and associations is aided, but
performance on creative tasks requiring the generation of novel
 COUNTERFACTUAL MIND-SETS
ideas matters is hindered. On a general level, the present research
suggests that reflecting back on events in which an outcome almost
turned out differently and mentally constructing an alternate world
can impact how future problems are approached. More specifi-
cally, thinking about what may have been can prevent one from
creating novel ideas but can lead that same person to notice hidden
connections. Simply put, counterfactual mind-sets promote think-
ing within the box.
References
Amabile, T. M. (1983). The social psychology of creativity: A componen-
tial conceptualization. Journal of Personality and Social Psychology, 45,
357–376.
Amabile, T. M., Conti, R., Coon, H., Lazenby, J., & Herron, M. (1996).
Assessing the work environment for creativity. Academy of Management
Journal, 39, 1154 –1184.
Brehm, J. W. (1966). A theory of psychological reactance. New York:
Academic Press.
Byrne, R. M. J., & Tasso, A. (1994, August). Counterfactual reasoning:
Inferences from hypothetical conditional. In A. Ram & K. Eiselt (Eds.),
Proceedings of the 16th Annual Conference of the Cognitive Science
Society. Hillsdale, NJ: Erlbaum.
Cosier, R. A. (1978). The effects of three potential aids for making
strategic decisions on prediction accuracy. Organizational Behavior and
Human Decision Processes, 22, 295–306.
Duncker, K. (1945). On problem solving. Psychological Monographs, 58,
270.
Einhorn, H. J., & Hogarth, R. M. (1986). Judging probable cause. Psycho-
logical Bulletin, 99, 3–19.
Feldman, L. A. (1995). Valence focus and arousal focus: Individual dif-
ferences in the structure of affective experience. Journal of Personality
and Social Psychology, 69, 153–166.
Friedman, R. S., & Forster, J. (2001). The effects of promotion and
prevention cues on creativity. Journal of Personality and Social Psy-
chology, 81, 1001–1013.
Galinsky, A. D., & Kray, L. J. (2004). From thinking about what might
have been to sharing what we know: The effects of counterfactual
mind-sets on information sharing in groups. Journal of Experimental
Social Psychology, 40, 606 – 618.
Galinsky, A. D., & Moskowitz, G. B. (2000). Counterfactuals as behavioral
primes: Priming the simulation heuristic and consideration of alterna-
tives. Journal of Experimental Social Psychology, 36, 257–383.
Gleicher, F., Boninger, D. S., Strathman, A., Armor, D., Hetts, J., & Ahn,
M. (1995). With an eye toward the future: The impact of counterfactual
thinking on affect, attitudes, and behavior. In N. J. Roese & J. M. Olson
(Eds.), What might have been: The social psychology of counterfactual
thinking (pp. 283–304). Hillsdale, NJ: Erlbaum.
Glucksberg, S., & Weisberg, W. R. (1966). Verbal behavior and problem
solving: Effects of labeling in a functional fixedness problem. The
Journal of Experimental Psychology, 71, 659 – 664.
Guilford, J. P. (1950). Creativity. American Psychologist, 5, 444 – 454.
Janis, I. L., & Mann, L. (1977). Emergency decision making: A theoretical
analysis of responses to disaster warnings. Journal of Human Stress, 3,
35– 48.
Kahneman, D., & Miller, D. T. (1986). Norm theory: Comparing reality to
its alternatives. Psychological Review, 93, 136 –153.
Kelley, H. H. (1973). The processes of causal attribution. American Psy-
chologist, 28, 107–128.
Kray, L. J., & Galinsky, A. D. (2003). The debiasing effect of counterfac-
tual mind-sets: Increasing the search for disconfirmatory information in
group decisions. Organizational Behavior and Human Decision Pro-
cesses, 91, 69 – 81.
47
Krishnamurthy, P., & Sivaraman, A. (2002). Counterfactual thinking and
advertising responses. Journal of Consumer Research, 28, 650 – 658.
Law School Admission Council. (2005).Preparing for the LSAT. Retrieved
April 11, 2006, from http://www.lsac.org/pdfs/2006 –2007/TestPrep06.pdf
Liljenquist, K. A., Galinsky, A. D., & Kray, L. J. (2004). Exploring the
rabbit hole or possibilities by myself or with my group: The benefits and
liabilities of activating counterfactual mind-sets for information sharing
and group coordination. Journal of Behavioral Decision Making, 17,
263–279.
Lord, C. G., Lepper, M. R., & Preston, E. (1984). Considering the opposite:
A corrective strategy for social judgment. Journal of Personality and
Social Psychology, 47, 1231–1243.
Mandel, D. R. (2003). Judgment dissociation theory: An analysis of dif-
ferences in causal, counterfactual, and covariational reasoning. Journal
of Experimental Psychology: General, 132, 419 – 434.
Mandel, D. R., & Lehman, D. R. (1996). Counterfactual thinking and
ascriptions of cause and preventability. Journal of Personality and
Social Psychology, 71, 450 – 463.
Markman, K. D., & McMullen, M. N. (2003). A reflection and evaluation
model of comparative thinking. Personality and Social Psychology
Review, 7, 244 –267.
McGill, A. L. (1998). Relative use of necessity and sufficiency information
in causal judgments about natural categories. Journal of Personality and
Social Psychology, 75, 70 – 81.
McGill, A. L., & Klein, J. G. (1993). Contrastive and counterfactual
reasoning in causal judgment. Journal of Personality and Social Psy-
chology, 64, 897–905.
Mednick, M. T., Mednick, S. A., & Mednick, E. V. (1964). Incubation of
creative performance and specific associative priming. Journal of Ab-
normal and Social Psychology, 69, 84 – 88.
Mednick, S. A. (1962). The associative basis of the creative process.
Psychological Review, 69, 220 –232.
Merriam-Webster. (2006). Merriam-Webster online dictionary. Retrieved
April 11, 2006, from http://www.m-w.com/dictionary/analysis.
Mussweiler, T., Strack, F., & Pfeiffer, T. (2000). Overcoming the inevita-
ble anchoring effect: Considering the opposite compensates for selective
accessibility. Personality and Social Psychology Bulletin, 26, 1142–
1150.
O’Hara, L. A., & Sternberg, R. J. (2001). It doesn’t hurt to ask: Effects of
instructions to be creative, practical, or analytical on essay-writing
performance and their interaction with students’ thinking styles. Cre-
ativity Research Journal, 13, 197–210.
Peterson, R. S., & Nemeth, C. J. (1996). Focus versus flexibility: Majority
and minority influence can both improve performance. Personality and
Social Psychology Bulletin, 22, 14 –23.
Princeton Review. (2005). The LSAT in detail. Retrieved May 3,
2006, from http://www.princetonreview.com/law/testprep/testprep.asp?
TPRPAGE⫽87&TYPE⫽LSAT-SECTIONS
Roese, N. J. (1994). The functional basis of counterfactual thinking.
Journal of Personality and Social Psychology, 66, 805– 818.
Roese, N. J., & Hur, T. (1997). Affective determinants of counterfactual
thinking. Social Cognition, 15, 274 –290.
Roese, N. J., Hur, T., & Pennington, G. L. (1999). Counterfactual thinking
and regulatory focus: Implications for action versus inaction and suffi-
ciency versus necessity. Journal of Personality and Social Psychology,
77, 1109 –1120.
Roese, N. J., & Olson, J. M. (1995). Functions of counterfactual thinking.
In N. J. Roese & J. M. Olson (Eds.), The social psychology of counter-
factual thinking (pp. 169 –197). Hillsdale, NJ: Erlbaum.
Roese, N. J., & Olson, J. M. (1997). Counterfactual thinking: The inter-
section of affect and function. In M. P. Zanna (Ed.), Advances in
experimental social psychology (Vol. 29, pp. 1–59). New York: Aca-
demic Press.
48 KRAY, GALINSKY, AND WONG

Rubin, D. C., Stoltzfus, E. R., & Wall, K. L. (1991). The abstraction of
form in semantic categories. Memory & Cognition, 19, 1–7.
Russell, J. A. (1980). A circumplex model of affect. Journal of Personality
and Social Psychology, 39, 1161–1178.
Schwarz, N., & Bless, H. (1991). Happy and mindless, but sad and
smart? The impact of affective states on analytic reasoning. In J. P.
Forgas (Ed.), Emotion and social judgments: International series in
experimental social psychology (pp. 55–71). Elmsford, NY: Perga-
mon Press.
Smith, S. M., Ward, T. B., & Schumacher, J. S. (1993). Constraining
effects of examples in a creative generation task. Memory & Cognition,
21, 837– 845.
Sternberg, R. J. (1988). Mental self-government: A theory of intellectual
styles and their development. Human Development, 31, 197–224.
Sternberg, R. J., & Wagner, R. K. (1991). MSG thinking styles inventory
manual. Unpublished manuscript.
Tetlock, P. E., & Lebow, R. N. (2001). Poking counterfactual holes in
covering laws: Cognitive styles and historical reasoning. American Po-
litical Science Review, 95, 829 – 843.
Ward, T. B. (1994). Structured imagination: The role of category structure
in exemplar generation. Cognitive Psychology, 27, 1– 40.
Wason, P. C. (1966). Reasoning. In B. Foss (Ed.), New horizons in
psychology (pp. 135–151). Middlesex, England: Penguin.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and vali-
dation of brief measures of positive and negative affect: The PANAS
scales. Journal of Personality and Social Psychology, 54, 1063–1070.
Wells, G. L., & Gavanski, I. (1989). Mental simulation of causality.
Journal of Personality and Social Psychology, 56, 161–169.

Appendix A

Examples Provided in Creative Generation Task Used in Experiments 4 and 5

(1) Please generate a name for: a new pasta Examples: radon, plutonium, argon, carbon, radium, uranium

Examples: spaghetti, lasagna, fettuccini, rotini, pastina, rigatoni (3) Please generate a new name for: an analgesic (pain reliever)

(2) Please generate a name for: a nuclear element Examples: Tylenol, Anacin, aspirin, bufferin, panadol, Midol

Appendix B

Modified Version of the Remote Associates Task

Please identify a common word that links each set of 3 words together. (6) chocolate—fortune—tin (cookie)
You should answer as many questions as you can.
Example: sore—shoulder—sweat Answer: cold (7) barrel—root— belly (beer)

(1) blank—white—lines ( page) (8) broken— clear— eye (glass)

(2) magic—plush—floor (carpet) (9) pure— blue—fall (water)

(3) thread—pine—pain (needle) (10) widow— bite—monkey (spider)

(4) stop—petty—sneak (thief) (Note: Answers appear in parentheses.)

(5) envy— golf— beans (green)

Received June 17, 2005

Revision received September 28, 2005
Accepted September 29, 2005 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 49 – 62 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.49

Self-Regulatory Processes Defend Against the Threat of Death:

Effects of Self-Control Depletion and Trait Self-Control
on Thoughts and Fears of Dying

Matthew T. Gailliot Brandon J. Schmeichel

Florida State University Texas A&M University

Roy F. Baumeister
Florida State University

Nine studies (N ⫽ 979) demonstrated that managing the threat of death requires self-regulation. Both trait
and state self-control ability moderated the degree to which people experienced death-related thought and
anxiety. Participants high (vs. low) in self-control generated fewer death-related thoughts after being
primed with death, reported less death anxiety, were less likely to perceive death-related themes in
ambiguous scenes, and reacted with less worldview defense when mortality was made salient. Further,
coping with thoughts of death led to self-regulatory fatigue. After writing about death versus a control
topic, participants performed worse on several measures of self-regulation that were irrelevant to death.
These results suggest that self-regulation is a key intrapsychic mechanism for alleviating troublesome
thoughts and feelings about mortality.

Keywords: self-regulation, mortality salience, death anxiety, thought suppression, terror management

The thought of death can be frightening. As a consequence, demand and consume some of an individual’s limited resources for
people prefer to avoid thinking about death (e.g., Aries, 1981; self-regulation (Muraven & Baumeister, 2000).
Becker, 1973). By what means are people capable of avoiding
thoughts of death? The current work assesses the role of self-
regulation in minimizing thoughts and anxiety surrounding death. Self-Regulation: Trait and State Differences
Self-regulation (or self-control) is the capacity to override one’s
Some individuals are highly adept at self-regulation whereas
thoughts, feelings, and habitual patterns of behavior. Self-
others are not, and these individual differences in trait self-control
regulation is a highly adaptive capacity that facilitates success in
have been associated with a diverse range of behavior. For exam-
myriad domains of life, including interpersonal relationships, ac-
ple, compared with those lower in trait self-control, people higher
ademic achievement, and coping with and adjusting to stress (e.g.,
in trait self-control are better at coping with anxiety and other
Baumeister, Heatherton, & Tice, 1994; Mischel, Shoda, & Peake,
negative moods, avoiding addictive behaviors, and responding to
1988; Shoda, Mischel, & Peake, 1990; Tangney, Baumeister, &
other people in prosocial, constructive ways (e.g., Finkel & Camp-
Boone, 2004). Awareness of mortality creates the potential for
bell, 2001; Mischel et al., 1988; Tangney et al., 2004). The diverse
stress (e.g., see Greenberg et al., 2003; Pollak, 1979), and so it
and substantial benefits of high trait self-control underscore the
seems plausible that one benefit of self-regulation may be to
idea that self-regulation is a valuable tool in many aspects of life.
minimize death-related thoughts and anxiety. Hence our central
Self-regulation can vary not just as trait but also as state. For any
hypotheses in this investigation were (a) that low capacity for
individual, self-regulatory success is more likely at some times
self-regulation, either as state or trait, would increase vulnerability
than at others. Recent work has suggested that self-regulation
to disturbing thoughts and feelings about death and (b) that coping
relies upon a limited resource or strength (for reviews, see Gailliot
with thoughts of death, like other acts of self-control, should
& Baumeister, in press; Muraven & Baumeister, 2000). Effortful
acts of self-regulation appear to consume or deplete this limited
resource, thereby impairing later attempts at self-regulation. For
Matthew T. Gailliot and Roy F. Baumeister, Department of Psychology, instance, participants in one study who performed a task that
Florida State University; Brandon J. Schmeichel, Department of Psychol- required self-control (i.e., suppressing or exaggerating responses to
ogy, Texas A&M University. an emotional film) were less able to exert self-control on a sub-
This research was supported by National Institutes of Health Grant MH
sequent task (i.e., squeezing a handgrip) than were participants
65559. Thanks to Lauren Brewer, Shenika Thomas, Savana Carroll, and
Daniel Donaldson for helping to conduct the studies reported here.
who first completed a task that did not require self-control (i.e.,
Correspondence concerning this article should be addressed to Matthew watching the film without regulating emotional responses; Mu-
T. Gailliot or Roy F. Baumeister, Department of Psychology, Florida State raven, Tice, & Baumeister, 1998). Presumably, the initial act of
University, One University Way, Tallahassee, FL 32306-1270. E-mail: self-control temporarily depleted self-regulatory strength, thereby
gailliot@psy.fsu.edu or baumeister@psy.fsu.edu impairing subsequent self-regulated performance.

49
50 GAILLIOT, SCHMEICHEL, AND BAUMEISTER
Thus, the ability to self-regulate may differ both across individ-
uals (trait self-control) and within individuals across time (self-
control depletion). The current work examined whether poor trait
and state self-regulation, respectively, would hamper the manage-
ment of concerns about death and whether the management of
concerns about death would hamper self-regulation.
Self-Regulation and the Management of Mortality
Concerns
The present work is based on the assumption that the idea of
death is threatening and can evoke aversive thoughts and feelings.
There are at least two ways in which self-regulation might facili-
tate the defense against these unwanted states. The first is thought
control. Research indicates that people cope with thoughts of death
and minimize preoccupation with death in part by suppressing
such thoughts or redirecting their thoughts away from death (e.g.,
Greenberg, Pyszczynski, Solomon, Simon, & Breus, 1994;
Harmon-Jones et al., 1997; Pollak, 1979). For instance, on explicit
measures of mortality concerns (e.g., measures of death anxiety),
most people score relatively low. On more implicit measures (e.g.,
word-association tests), however, people score significantly higher
(e.g., Feifel & Branscomb, 1973; Pollak, 1979). This suggests that
people possess an implicit awareness of death and that such
thoughts are suppressed from explicit awareness.
Moreover, suppressing thoughts about death requires effortful,
controlled processing (Greenberg et al., 1994; Harmon-Jones et al.,
1997; Pyszczynski, Greenberg, & Solomon, 1999; see also Weg-
ner, 1994). For example, immediately after thinking about mortal-
ity, one’s death thoughts are suppressed and are less accessible to
awareness (e.g., Arndt, Greenberg, Solomon, Pyszczynski, & Si-
mon, 1997; Greenberg et al., 1994; Harmon-Jones et al., 1997). If
cognitive resources are diverted by a concurrent task, however,
death thoughts remain highly accessible to awareness (Arndt et al.,
1997; Greenberg, Arndt, Schimel, Pyszczynski, & Solomon,
2001). Apparently the cognitive load diverts the resources needed
to suppress death thoughts, so death thoughts remain highly ac-
cessible (see Smart & Wegner, 1999; Wegner & Zanakos, 1994;
Wenzlaff & Wegner, 2000).
Low trait self-control and depleted self-control strength might
also undermine the suppression of death thoughts. Thought sup-
pression requires self-control (e.g., Baumeister et al., 1994; Weg-
ner, 1994), and so dispositionally low or temporarily depleted
self-control should be linked to poorer suppression, leading to
increased accessibility of death-related thoughts and anxiety. Ef-
fortfully suppressing a thought reduces both its explicit and im-
plicit accessibility (Anderson & Green, 2001; MacLeod, 1989;
McBride & Dosher, 1997), and disrupting controlled efforts to-
ward suppressing death thoughts increases the implicit accessibil-
ity of those thoughts (e.g., Arndt et al., 1997; Pyszczynski et al.,
1999). Low self-control should therefore be associated with in-
creased accessibility of both implicit and explicit death thoughts.
Likewise, thought suppression depletes self-control strength
(e.g., Gordijn, Hindriks, Koomen, Dijksterhuis, & Van Knippen-
berg, 2004; Muraven et al., 1998). After thinking about death,
people suppress thoughts about death. Self-regulatory resources
should be depleted as a result. Self-regulation should therefore be
impaired shortly after people think about and suppress thoughts of
death.
A second way that self-regulation might facilitate the manage-
ment of concerns about death is the regulation of emotion. An
individual who capably controls his or her emotions is likely to
experience less of the fear and anxiety associated with death.
Emotion regulation requires self-control, and therefore both low
trait self-control and self-control depletion undermine emotion
regulation (e.g., Baumeister, Bratslavsky, Muraven, & Tice, 1998;
Finkel & Campbell, 2001; Muraven et al., 1998; Tangney et al.,
2004). Consequently, low trait self-control and self-control deple-
tion should be associated with increased emotional preoccupation
with death. In sum, we posited that self-regulation should facilitate
suppressing thoughts about death and, to some extent, regulating
emotional responses to concerns about death.
Another potential basis for predicting a link between self-
regulation and death-thought accessibility, however, involves self-
esteem. Terror management theory has suggested that self-esteem
is primarily a defense against death, so chronic low self-esteem
could leave people more vulnerable to thoughts and anxieties
about death (e.g., see Pyszczynski, Greenberg, Solomon, Arndt, &
Schimel, 2004). The present work is not concerned with self-
esteem per se but with the possibility that self-esteem could
account for the effects of self-regulation. Trait self-esteem and trait
self-control are positively correlated (Tangney et al., 2004), hence
it is theoretically possible that any effects of self-control are due to
self-esteem. It is also plausible that high self-esteem results from
having more effective self-control, so people with high self-esteem
might be less vulnerable to self-control depletion. To test these
possibilities, we assessed the role of self-esteem.
Research Overview
In nine studies, using both correlational and experimental meth-
ods, we tested whether self-regulation facilitates the management
of mortality concerns. We first examined whether poor self-
regulation would hamper the management of mortality concerns.
Our first studies tested the prediction that people with low (vs.
high) trait self-control would be more prone to come up with
death-related thoughts in response to ambiguously evocative stim-
uli (Studies 1A and 1B) and would have higher death anxiety
(Study 1C). Next, we manipulated temporary self-regulatory ca-
pacity by having participants complete either an initial task that
required self-regulation or one that did not require self-regulation.
To the extent that exercising self-control depletes the capacity for
self-regulation (e.g., see Muraven & Baumeister, 2000), we ex-
pected to find greater death-thought accessibility (Study 2) and a
greater proclivity to perceive death-related themes in an ambigu-
ous stimulus (Study 3) among participants who had (vs. had not)
previously expended their self-regulatory resources. Subsequently,
we assessed (rather than manipulated) self-control depletion and
death-thought accessibility and death anxiety (Study 4). In the final
study in this group (Study 5), we sought to show that low trait
self-control would leave people more vulnerable to the threat of
mortality salience and would result in a heightened tendency to
respond to that threat by bolstering support for a political leader
(Landau et al., 2004). Such an externalizing response would pre-
sumably indicate that the inner defense mechanisms against death
were inadequate.
Having thus shown the effects of poor self-regulation on mor-
tality defenses, we turned this around to investigate the effects of
 THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION 51

mortality defense on self-regulation. If (as argued above) self-

regulation consumes a limited resource and defending against the
threat of death consumes some of that resource then people’s
capacity for self-regulation should be diminished after they defend
against such a threat. Participants in the last four studies wrote
about either death or a control topic. We predicted that writing
about death would activate defensive processes that would deplete
self-regulatory resources, as indicated by poorer subsequent per-
formance on the Stroop task (Study 6), impaired logical reasoning
abilities (Study 7), reduced success at solving anagrams (Study 8),
and less effortful persistence on word puzzles (Study 9).

Studies 1A–1C
In Studies 1A–1C we relied on measurement of trait self-control
to test the hypothesis that weaker self-control would be associated
with greater susceptibility to thoughts about death. We measured
death-thought accessibility in different ways in each of these
studies. In Study 1A we used ambiguous word puzzles that could
be solved with either death-related or death-irrelevant words (e.g.,
grave, grape). We believed that greater accessibility of death
thoughts among persons low in self-control would be reflected in
higher rates of death-related word solutions. In a similar vein, in
Study 1B we showed participants an ambiguous visual stimulus
that could be construed as either related or unrelated to death. We
then assessed spontaneous thoughts to determine the frequency of
death-related thoughts the image evoked. In Study 1C we relied on
a straightforward measure of death anxiety (Templer, 1970) to
ascertain whether low self-control would be associated with
greater fear related to death.

Method
Participants. Participants in these and all subsequent studies were
undergraduate students enrolled in introductory psychology courses who
received either extra credit or credit toward fulfilling a course requirement.
The size and gender makeup of the samples in Studies 1A–1C were as
follows: Study 1A (N ⫽ 279; 166 women, 113 men), Study 1B (N ⫽ 13;
12 women, 1 man), and Study 1C (N ⫽ 163; 118 women, 45 men).
Materials and procedures. At the start of the academic semester,
participants completed measures of trait self-control (the Self-Control
Scale; Tangney et al., 2004) and self-esteem (the Rosenberg Self-Esteem
Scale; Rosenberg, 1965) during a mass-testing session. We used the brief
version of the Self-Control Scale, which contains 13 items (e.g., “I have a
hard time breaking bad habits” [reverse scored]; “I am good at resisting
temptation”) answered on a scale from 1 (not at all like me) to 5 (very much
like me). The Rosenberg Self-Esteem Scale contains 10 items (e.g., “On the
whole, I am satisfied with myself.”) answered on a scale from 1 (strongly
disagree) to 5 (strongly agree). Higher scores on these measures indicate
higher self-control and self-esteem, respectively.
Participants in Study 1C also completed the Social Interaction Anxiety
Scale (Mattick & Clarke, 1998), as a measure of social anxiety, and the
State–Trait Anxiety Index (Spielberger, Gorsuch, & Lushene, 1970), as a
measure of general trait anxiety. These measures were used to control for
general negative affect.
Studies 1A–1C were conducted 4 –12 weeks after the initial mass-testing
session. In each study, participants completed questionnaire packets that
contained the focal materials along with other, non-death-related measures.
In Study 1A, the packet included a list of 20 word fragments, some of
which could be completed with death-related thoughts (e.g., sk_ll and
gra_e could be solved with the words skull and grave or skill and grape,
respectively). We counted and then standardized the number of death-
Figure 1. “All Is Vanity” by Charles Allan Gilbert. This picture, shown
to participants in Study 1B, could be interpreted as either a woman or a
skull.
related words used to complete the relevant word fragments to serve as a
measure of implicit death-thought accessibility (see Greenberg et al.,
1994).1
The questionnaire packet completed by participants in Study 1B in-
cluded a measure of explicit death-thought accessibility. Specifically, these
participants were shown a picture that could be interpreted as depicting
either (a) a woman sitting in front of a mirror while putting on makeup or
(b) a skull (see Figure 1). Participants were instructed to list the first 10
words that came to mind as they viewed the image. A judge who was blind
to trait self-control scores determined which thoughts on the thought-listing
task were related to death (e.g., skull and skeleton). The number of thoughts
related to death constituted the measure of explicit death-thought
accessibility.
The questionnaire packet completed by participants in Study 1C con-
tained the Death Anxiety Scale (Templer, 1970), which includes 15 items
(e.g., “I am very much afraid to die” and “I fear dying a painful death”)
1
Participants completed a list that contained either 5 or 7 word frag-
ments that could be solved with death-related thoughts. Accordingly, the
number of death-related thoughts was standardized among participants
who completed each list.
52 GAILLIOT, SCHMEICHEL, AND BAUMEISTER
answered as either “true” or “false.” Higher scores indicate greater death
anxiety.
In these and all subsequent studies, participants were last thanked and
fully debriefed. Because of the threatening nature of death, we were careful
to explain the importance of and need for the research and to assuage any
fears or concerns participants may have had. In this fashion, we tried to
ensure that no participant left feeling distressed.
Results
In each study, higher trait self-control was associated with less
preoccupation with death. Trait self-control correlated negatively
with the number of word fragments completed with death-related
words, r(279) ⫽ ⫺.13, p ⬍ .05 (Study 1A), the number of death
thoughts listed to the ambiguous image, r(13) ⫽ ⫺.59, p ⬍ .05
(Study 1B), and death anxiety, r(163) ⫽ ⫺.36, p ⬍ .01 (Study 1C).
Further analyses indicated that each relationship remained sig-
nificant when controlling for self-esteem (all ps ⱕ .05) and that
controlling for self-esteem did not change the strength of any
relationship (all ps ⬎ .11). Self-esteem thus did not appear to
account for the relationship between self-control and mortal
concern.
Moreover, the relationship between self-control and death anx-
iety in Study 1C remained significant when controlling for social
anxiety and trait anxiety (scores on the Social Interaction Anxiety
Scale and State–Trait Anxiety Index, respectively; ps ⬍ .05). This
suggests that the association between high trait self-control and
low death anxiety was not simply the result of trait self-control
buffering against negative affect in general.
Discussion
Studies 1A–1C provided initial evidence for the view that self-
regulation protects people from the aversive awareness of death
and mortality. High self-control predicted less explicit (death anx-
iety and responses to an image of a skull) and implicit (responses
to word fragments) preoccupation with death.
We assume that people periodically confront cues or thoughts
that remind them of death and that they therefore exercise self-
regulation to prevent these thoughts from lingering in conscious
awareness and escalating into greater anxiety. These results sug-
gest that low self-control leaves people less effectively defended
against the threatening idea of death and hence more prone to
suffer both disturbing thoughts and anxious emotions in connec-
tion with death.
Study 2
In Studies 2 and 3 we used laboratory manipulations of self-
regulation (rather than correlational evidence) to test the causal
hypothesis that low self-control weakens defenses against death. In
Study 2 we examined whether temporarily depleted self-regulatory
strength would increase the cognitive accessibility of death-related
thought. People typically try to minimize or suppress thoughts
associated with death. We hypothesized that if self-regulation is
needed to suppress death thoughts, then self-regulatory depletion
should undermine this suppression and lead to increased accessi-
bility of death-related thoughts.
Participants in Study 2 first completed either a task that required
self-regulation (thought suppression) or a control task. Thought
suppression requires self-control and has been shown to deplete
self-regulatory strength (Muraven et al., 1998).
Following this initial task, participants completed the same
measure of implicit death thoughts used in Study 1A (word frag-
ment completion). We predicted that participants in the depletion
condition would complete the word fragments with death-related
words more often than would participants in the no-depletion
condition, indicating greater accessibility of death-related thoughts
when self-regulatory capacity had been reduced.
Method
Participants. Participants were 19 undergraduates (13 women, 6 men)
who completed the Rosenberg Self-Esteem Scale during a mass-testing
session at the start of the semester. The present study occurred approxi-
mately 3 months later. Participants were run individually and told the study
was to investigate impression formation and thought patterns.
Procedure. The first task served as the manipulation of self-regulatory
resources. Specifically, participants completed a thought-recording exer-
cise (borrowed from Wegner, Schneider, Carter, & White, 1987). Partici-
pants were given a sheet of paper and asked to write down all of their
thoughts for 5 min. Participants randomly assigned to the no-depletion
condition received no additional instructions. Participants assigned to the
depletion condition were instructed further not to think about a white bear.
Each time they did happen to think about a white bear, they were to place
a mark on the page and attempt to stop thinking about a white bear.
After the thought-listing task, participants rated the difficulty of the task
(as a manipulation check), and they completed the Brief Mood Introspec-
tion Scale (BMIS; Mayer & Gaschke, 1988). The BMIS contains 20 items
indicative of mood (e.g., happy, sad) and arousal (e.g., peppy, drowsy).
Participants rated each item to indicate how they were feeling at the present
moment, using a scale from 1 (definitely do not feel) to 7 (definitely feel).
Participants also indicated to what extent they tried to follow the task
instructions (effort) and how well they felt that they followed those
instructions (perceptions of task performance).
Next, participants completed an unrelated filler task (i.e., making per-
sonality judgments of children in a video) that bolstered the cover story.
The personality-judgment task took approximately 4 min. Afterward, par-
ticipants completed the same series of word fragments used in Study 1A.
Results and Discussion
Manipulation check. Participants in the depletion condition
rated their suppression task as being somewhat more difficult
(M ⫽ 3.78, SD ⫽ 1.92) than did the participants in the no-
depletion condition (M ⫽ 2.40, SD ⫽ 1.26), t(17) ⫽ 2.53, p ⫽ .08.
This pattern suggests that the depleting task required more effort-
ful self-regulation than did the nondepleting task.
Implicit death thoughts. We predicted and confirmed that
death-related thoughts would be more accessible to the minds of
depleted than to those of nondepleted participants. Depleted par-
ticipants (M ⫽ 2.22, SD ⫽ 1.48) solved more word fragments with
death-related words than did nondepleted participants (M ⫽ 0.90,
SD ⫽ 1.20), t(17) ⫽ ⫺2.15, p ⬍ .05. This pattern is consistent
with the view that self-regulatory resources help reduce death-
thought accessibility. Participants who possessed greater self-
regulatory strength (i.e., nondepleted participants) exhibited less
death-thought accessibility than did participants who had previ-
ously expended their self-regulatory strength. Presumably, de-
pleted participants were less able to suppress thoughts of death,
and thus death thoughts increased in accessibility. The word-
fragment completion task required participants to execute a mental
 THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION 53

search for possible solutions, and solutions related to death ap-
peared to be more accessible when self-regulatory resources were
depleted.
Self-esteem, mood, arousal, task performance, and effort. Ad-
ditional analyses suggested that the difference in accessibility of
implicit death thoughts between depleted and nondepleted partic-
ipants was not attributable to differences in self-esteem, mood,
arousal, effort, or perceptions of task performance. Regression
analysis indicated that self-esteem did not moderate the effect of
depletion on the accessibility of death thoughts (t ⬍ 1, ns). Further,
depleted and nondepleted participants did not differ in arousal, task
performance, or effort (all ps ⬎ .24), nor did any of these factors
significantly predict the number of implicit death thoughts (all
ps ⬎ .24). Depleted and nondepleted participants did differ in
self-reported mood valence ( p ⬍ .05) such that depleted partici-
pants were in a more negative mood, but mood did not signifi-
cantly predict the number of implicit death thoughts ( p ⬎ .32), and
the difference in death thoughts between the two conditions was
significant even when controlling for mood, F(1, 16) ⫽ 3.19, p ⬍
.05 (one-tailed).
Study 3
The purpose of Study 3 was to provide converging evidence that
self-regulatory depletion increases death-thought accessibility. The
design of the study mirrored that of Study 2. Participants first
completed either a task (controlling attention while watching a
video) that has been shown to deplete self-regulatory strength (e.g.,
Schmeichel, Vohs, & Baumeister, 2003) or a control task. Partic-
ipants then completed a measure of death-thought accessibility for
which they listed their thoughts about an ambiguous image. We
predicted that participants who first performed the self-regulatory
(vs. the control) task would respond to the ambiguous image with
more death-related thoughts because they had depleted self-
regulatory strength.
BMIS) 4 –5 min after the initial video-watching task. Participants subse-
quently completed a measure of death-thought accessibility (listing
thoughts to an ambiguous drawing) that was identical to the one used in
Study 1B except that a different drawing was used (see Figure 2).
Results and Discussion
Explicit death thoughts. Depleted participants (M ⫽ 1.34,
SD ⫽ 1.00) listed more death-related thoughts while viewing the
drawing than did nondepleted participants (M ⫽ .97, SD ⫽ .45),
t(65) ⫽ 1.99, p ⫽ .05. Presumably, participants who possessed full
self-regulatory strength were able to suppress thoughts of death
while viewing the drawing. Depleted participants, lacking self-
regulatory strength, were less able to suppress such thoughts. The
results of Study 3 thus replicated those of Study 2, using different
methods. In both studies, weakening self-control by depleting its
resources led participants to have more thoughts about death.
Moreover, the finding that depletion increased death thoughts
and anxiety after a short delay indicates that the effects of deple-
tion are somewhat enduring. One may very well be susceptible to
greater mortal concern long after having exerted self-control.
Mood and arousal. Analyses indicated that the effect of de-
pletion on the accessibility of death-related thoughts was not
mediated by mood or arousal. Depleted and nondepleted partici-
pants did not differ in arousal ( p ⬎ .19), but depleted participants

Method
Participants. Sixty-seven undergraduate students (49 women, 18 men)
participated in a classroom setting. They were told that the study was to
investigate people’s attitudes and opinions.
Procedure. The first task served as the manipulation of self-regulatory
resources. Specifically, participants watched a 6-min video (without sound)
of a woman talking (modified from Gilbert, Krull, & Pelham’s, 1988,
study). In the bottom corner of the screen, common one-syllable words
(e.g., hair, hat) appeared individually for 10 s. Participants randomly
assigned to the depletion condition were instructed to focus their attention
on the woman’s face and to refrain from looking at the words. If they
happened to look at the words, they were to refocus their attention on the
woman as quickly as possible. Attention automatically orients toward
novel stimuli appearing in the environment (e.g., see Shiffrin & Schneider,
1977), and so the task required these participants to exert self-control by
overriding prepotent orienting of attention to the words and maintain
attention instead on the woman’s face only. Participants assigned to the
no-depletion condition were instructed to watch the video as they would
normally (i.e., as if they were sitting at home watching TV) and hence were
not required to exert self-control.
After the video-watching task, participants completed the target materi-
als (described below) that were embedded among other measures (not Figure 2. Artwork (Gillam) from the May 1894 cover of Judge maga-
related to death) that helped bolster the cover story about attitudes and zine. This picture, shown in Study 3, could be interpreted as either two men
opinions. Participants completed measures of mood and arousal (the or a skull.
54 GAILLIOT, SCHMEICHEL, AND BAUMEISTER
were in a somewhat more negative mood ( p ⫽ .06). However,
neither mood nor arousal significantly predicted the number of
death thoughts ( ps ⫽ .22 and .38, respectively), and the difference
in death thoughts between the two conditions was significant, even
when controlling for mood, F(1, 63) ⫽ 2.93, p ⬍ .05 (one-tailed).
Study 4
The purpose of Study 4 was to provide additional evidence that
self-regulatory depletion increases death thoughts and anxiety.
Specifically, we assessed (rather than manipulated) temporary
self-regulation abilities using a recently developed scale that has
been shown to measure self-regulatory depletion (Twenge, Mu-
raven, & Tice, 2004; see also Finkel & Campbell, 2001, for a
similar measure). Participants then completed measures of implicit
death-thought accessibility (word fragments) and death anxiety.
We hypothesized the following: If self-regulation helps to mini-
mize death-thought accessibility and death anxiety, then partici-
pants who report more self-regulatory depletion should exhibit
more death-thought accessibility and death anxiety than should
participants who report relatively less self-regulatory depletion.
Method
Participants. One hundred fifty-nine undergraduate students (110
women, 49 men) participated. The study was conducted in a classroom
setting, and participants were told that it was to investigate attitudes and
vocabulary. They completed questionnaire packets that contained all in-
structions and materials (i.e., the focal materials along with other, non-
death-related personality measures.)
Procedure. Specifically, participants completed a measure that as-
sessed perceptions of the momentary availability of self-regulatory re-
sources (the State Depletion Scale; Twenge et al., 2004). The State De-
pletion Scale contains 25 items (e.g., “I feel like my willpower is gone.”
“My mental energy is running low.”) answered on a scale from 1 (not true)
to 7 (very true; ␣ ⫽ .91). Higher scores indicate greater self-regulatory
depletion.
Subsequent measures in the packet assessed implicit death-thought ac-
cessibility and death anxiety. Specifically, participants completed the
Death Anxiety Scale (Templer, 1970) and the same word fragments as used
in Study 1A and Study 2.
Results and Discussion
State self-control depletion correlated positively with implicit
death thoughts, r(159) ⫽ .20, and death anxiety, r(159) ⫽ .43
(both ps ⬍ .05). Participants who indicated that they were rela-
tively more depleted solved more word fragments with death-
related words and reported more death anxiety than did partici-
pants whose self-regulatory resources were less depleted.
These findings converge with those of the previous studies.
Measured self-control depletion revealed the same association
between depletion and death-thought accessibility as did manipu-
lated self-control depletion in Studies 2 and 3. Presumably, par-
ticipants who were relatively less depleted were more capable of
managing concerns about death and thus experienced relatively
less preoccupation with death. The results thus far are highly
consistent with the hypothesis that self-regulation minimizes
death-thought accessibility and death anxiety.
Study 5
The purpose of Study 5 was to test the hypothesis that good
self-regulation reduces concerns about death by examining the
consequences of increased death concerns rather than the extent of
them. Specifically, Study 5 examined the role of trait self-control
in one well-documented consequence of mortality salience: in-
creased worldview defense.
When mortality is made salient, people bolster faith in their own
culture and react more positively toward others who support their
cultural norms and values, and they react more negatively toward
others who disagree with their cultural norms and values (e.g.,
Florian & Mikulincer, 1997; Greenberg et al., 1990; Heine, Hari-
hara, & Niiya, 2002; Ochsmann & Mathey, 1994). Likewise,
mortality salience has been shown to increase support for political
leaders. For instance, after writing about death or the September
11, 2001, terrorist attacks on the United States, participants (from
the United States) showed increased support for current President
George W. Bush compared with participants who wrote about
merely aversive topics such as dental pain (Landau et al., 2004).
Study 5 of the current investigation was a partial replication of the
Landau et al. (2004) study that assessed the relationships among
mortality salience, trait self-control, and support for the President.
We made a few hypotheses about the results of Study 5: If
self-control is used to reduce death-related thought and anxiety,
then participants higher in trait self-control should be more capable
of managing concerns about death following a mortality-salience
induction than should participants lower in trait self-control. The
more effective suppression abilities of individuals high (vs. low) in
trait self-control should reduce both implicit and explicit death-
thought accessibility (Anderson & Green, 2001; MacLeod, 1989;
McBride & Dosher, 1997). To the extent that support for President
Bush is exacerbated by death concerns, participants high in trait
self-control should not show increased support for the President
after thinking about death.
Participants first did or did not write about their own death.
Then they read a short passage in support of President Bush and
indicated how much they agreed with the passage. Our prediction
was that mortality salience would boost support for the President
but only among participants with low trait self-control.
Method
Participants. Participants were 88 undergraduates (41 women, 47
men) who completed a brief measure of trait self-control (Tangney et al.,
2004) and self-esteem (Rosenberg, 1965) during a mass-testing session
earlier in the semester. The present study was conducted in classrooms in
groups of 10 –25 participants approximately 2 weeks prior to the 2004 U.S.
presidential election (some 7 weeks after the mass-testing session). Partic-
ipants were told that the study was to investigate personality characteristics
and political attitudes. They received a packet that contained all materials
for the experiment.
Procedure. Participants first completed a demographic questionnaire
and two filler personality-related questionnaires. Participants randomly
assigned to the mortality-salience condition then completed a questionnaire
that asked them to describe the emotions that the thought of their own death
aroused in them and to write about what will happen to their bodies as they
physically die (see Rosenblatt, Greenberg, Solomon, Pyszczynski, & Lyon,
1989). Participants in the control condition did not complete this question-
naire but instead proceeded immediately to the next items in the packet,
two personality-relevant measures intended to bolster the cover story.
 THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION
These measures were included immediately after the mortality-salience
manipulation but before the target dependent variable (described below),
because the effects of the mortality-salience manipulation used typically
emerge only after a short delay or distraction (see Pyszczynski et al., 1999).
To test the effect of mortality salience on worldview defense, we asked
participants next to read and evaluate a statement (used in Landau et al.’s,
2004, study) regarding President Bush and his antiterrorism policies. The
passage praised the United States and President Bush and his actions
against terrorism.
After reading the passage, participants responded to the following items
on a scale from 1 (strongly disagree) to 5 (strongly agree): “To what extent
do you endorse this statement?”; “I share many of the attitudes expressed
in the above statement”; and “Personally I feel secure knowing that the
President is doing everything possible to guard against any further attacks
against the United States.” Participants also indicated their political orien-
tation by placing a mark on a continuous spectrum that ranged from Liberal
to Conservative. The responses to the Likert items and spectrum (measured
in centimeters across the spectrum) were standardized (z scored) and
averaged (␣ ⫽ .91), and the single score served as our dependent measure
of support for President Bush (higher scores indicated greater support).
Results
A regression analysis was conducted to predict support for the
President from trait self-control scores, mortality-salience condi-
tion, and their interaction, while controlling for self-esteem.2 The
analysis indicated a significant effect of mortality-salience condi-
tion (␤ ⫽ 1.35, p ⬍ .05). Participants in the mortality-salience
condition indicated supporting President Bush to a greater extent
than did participants in the control condition. This replicates the
findings of Landau et al. (2004), in which mortality salience
increased support for President Bush. The effect of self-control
approached significance (␤ ⫽ .29, p ⫽ .06) as did its interaction
with mortality-salience condition (␤ ⫽ ⫺1.89, p ⫽ .06).
To interpret the interaction between mortality salience and self-
control, and also to test directly our specific hypothesis, we as-
sessed the simple effect of mortality-salience condition among
participants who were relatively high versus relatively low in trait
self-control (one standard deviation above and below the mean on
the self-control scale, respectively; Aiken & West, 1991), control-
ling for trait self-esteem. Results indicate that the effect of mor-
tality salience on support for the President was significant and
robust for participants low in self-control ( p ⬍ .05) but was
nonsignificant and negligible for those high in self-control ( p ⬎
.67; see Figure 3). Thus, although mortality salience increased
support for the President, this effect occurred primarily among
participants lower in trait self-control. Participants higher in self-
control did not show increased support for the President after
thinking about death.
Discussion
These findings converge upon the hypothesis that self-
regulation facilitates the management of concerns about death.
Whereas the results of the previous studies indicate that good trait
and state self-control reduces death-thought accessibility and death
anxiety, the results of Study 5 suggest that high trait self-control
helps to reduce defensive reactions to mortality salience. Mortality
salience increases liking for charismatic and decisive leaders (Co-
hen, Solomon, Maxfield, Pyszczynski, & Greenberg, 2004;
Landau et al., 2004). People with high trait self-control were
55
apparently less cognizant of death after mortality salience, how-
ever, and so after thinking about death they did not increase
support for President Bush.
Study 6
The central hypothesis of this investigation is that defending
against the threatening idea of death requires self-regulation and
consumes self-regulatory resources. Thus far we have presented
evidence that when the capacity for self-regulation is low, thoughts
and fears of death are more common. We turn now to a comple-
mentary way of testing the hypothesis, which is to show that when
people defend themselves against thoughts and feelings about
death, they deplete their self-regulatory resources. Past work
has shown that following mortality salience, people actively
and effortfully suppress thoughts of death (Arndt et al., 1997;
Greenberg et al., 1994; Harmon-Jones et al., 1997). The result
should be impaired performance on subsequent tasks requiring
self-regulation.
In Study 6, participants wrote about death or a control topic. We
assumed that writing about death but not the control topic would
activate disturbing thoughts and feelings that would require self-
regulatory exertion to defend the self against them. Participants
were then provided a delay that allowed participants sufficient
time to suppress thoughts of death (if they had them). After the
delay, participants completed the Stroop color–word interference
task as a measure of self-regulation. For the Stroop task, partici-
pants saw words (i.e., red, blue, green) on a computer screen, and
they were to respond by indicating the font color of the word. On
some trials (incongruent trials), the meaning of the word differed
from its font color (e.g., red appeared in a blue font), so on these
trials participants had to exert self-control by overriding the ten-
dency to read the word and respond instead according to the font
color. Performance on the Stroop task was operationalized as the
number of errors made during incongruent trials. We predicted that
participants who wrote about death would make more Stroop
errors than would those who wrote about a neutral topic, thereby
suggesting that their self-regulatory resources had been depleted
by the mortality-salience induction.
We also assessed performance on Stroop trials that did not
require self-control (i.e., congruent trials or trials for which the
font color and meaning of the word were the same). This allowed
us to determine whether mortality salience would impair automatic
cognitive processing, as might be expected if the self-regulatory
impairments following mortality salience are caused by another
factor aside from self-control depletion (e.g., mood or arousal). We
predicted that mortality salience would not influence performance
on congruent trials.
2
Including self-esteem and its higher order interactions with self-control
and mortality-salience condition yielded no significant effects as a function
of self-esteem (all ps ⬎ .21). In addition, our primary results remained
relatively unchanged when excluding self-esteem from the model. We
included self-esteem in the model to demonstrate that the relationship
between trait self-control and worldview defense (support for President
Bush) was not attributable to self-esteem.
56 GAILLIOT, SCHMEICHEL, AND BAUMEISTER
Figure 3. Support for President George W. Bush as a function of mortality-salience condition and trait
self-control (Study 5).
Method
Participants. Fifty-seven undergraduate students (45 women, 12 men)
participated. Five participants were excluded from all analyses because of
technical difficulties (e.g., being interrupted while completing the Stroop
task) or for not following instructions, leaving a final sample of 52 (42
women, 10 men). Participants were randomly assigned to condition.
Procedure. Participants were run individually and were told that the
study was to investigate personality and perception. Participants first
completed a computer task that was similar to the Stroop task so they could
become familiarized with how to respond on the keyboard. Specifically,
participants completed 60 trials in which a string of Xs (XXXXX) appeared
on the computer screen in either a red, green, or blue font. Participants were
to indicate the color of the Xs by pressing one of three computer keys (the
R, G, or B key) as quickly as possible. Following each response, the next
string of Xs appeared immediately.
After practicing the computer task, participants wrote about either death
or dental pain. As in Study 5, mortality-salience participants described the
emotions that the thought of their own death aroused in them and what will
happen to them as they physically die, whereas the dental-pain participants
described the emotions that the thought of their own dental pain aroused in
them and what would happen to them if they were to experience dental pain
(see Rosenblatt et al., 1989). Participants then completed the BMIS, as a
measure of mood and arousal, and a filler task (a crossword puzzle) for 4
min to provide sufficient time for them to suppress thoughts of death
(Pyszczynski et al., 1999).
Participants then completed the final Stroop task. For this task,
participants completed 3 blocks of trials in which the word red, blue, or
green appeared on the computer screen in either a red, blue, or green
font. Participants were to indicate the font color by pressing one of three
computer keys (the R, G, or B key) and were asked to respond as
quickly and accurately as possible. The first and third blocks consisted
of 30 congruent trials in which the meaning and font color of the word
were the same. The second block consisted of 60 incongruent trials in
which the meaning and font color of the word were different. At the end
of the task, participants completed demographic information, indicated
whether they had completed the Stroop task before (“yes” or “no”), and
were probed for suspicion. No participant indicated that he or she
thought that writing about death or dental pain might influence perfor-
mance on the Stroop task.
Results
Stroop task performance. In all of our analyses involving Stroop
task performance, we controlled for whether participants had ever
performed the Stroop task before, because there is a large practice
effect on Stroop task performance (MacLeod, 1991). An analysis of
covariance indicated that mortality-salience participants made more
errors (M ⫽ 2.50, SD ⫽ 1.86) on the incongruent trials on the Stroop
task than did dental-pain participants (M ⫽ 1.40, SD ⫽ 1.85), F(1,
49) ⫽ 4.26, p ⬍ .05. This result suggests that mortality salience
impaired participants’ ability to exert self-control.
Additional analyses indicated that the two groups did not differ
on the number of errors on either block of congruent trials (both
Fs ⬍ 1, ns). This pattern of results suggests that mortality salience
did not impair automatic processing (congruent trials) but did
impair controlled processing (incongruent trials). Further, partici-
pants did make some errors on the congruent trials on the first and
third blocks (in the mortality-salience condition, M ⫽ .32 and .63
[excluding one outlier on this measure], respectively; in the dental-
pain condition, M ⫽ .25 and .63, respectively). This suggests that
the congruent trials were somewhat cognitively demanding, yet
mortality salience did not impair performance on these trials.
In addition, mortality-salience and dental-pain participants did
not differ in their reaction times on either the incongruent or
 THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION
congruent trials (Fs ⬍ 1, ns). This suggests that the increased
number of errors among mortality-salience participants was not the
result of a speed–accuracy trade-off, such as if they had made more
errors because they responded faster.
Mood and arousal. Mortality-salience and dental-pain partic-
ipants did not differ in mood valence or arousal (as assessed by the
BMIS; ts ⬍ 1, ns). This suggests that mortality-salience partici-
pants’ performing worse on the Stroop task than dental-pain par-
ticipants was probably not attributable to mood or arousal.
Discussion
Participants who wrote about death performed worse (i.e., made
more errors) on the Stroop task than did control participants but
only on trials that required self-control (i.e., incongruent trials).
This suggests that mortality salience impaired the ability to exert
self-control. Presumably, after thinking about death, participants
suppressed thoughts of death, and the act of thought suppression
depleted their self-regulatory strength. Participants who wrote
about death rather than dental pain did not perform worse on trials
that did not require self-control (i.e., congruent trials), which
suggests that mortality salience did not have an adverse effect on
automatic cognitive processes.
Study 7
The results of Study 6 suggest that mortality salience subse-
quently impairs self-control but not cognitive processing in gen-
eral. It is plausible, however, that the congruent Stroop trials were
relatively easy and therefore a poor measure of general cognitive
abilities. In Study 7 we therefore provided another test of the
hypothesis that mortality salience impairs self-control but not other
cognitive processes by using a more difficult task to assess general
cognitive abilities.
Participants first wrote about either death or uncertainty. Un-
certainty salience has been shown to be a poignant self-threat (e.g.,
van den Bos, Poortvliet, Maas, Miedema, & van den Ham, 2005)
and therefore seemed an adequate control topic. Participants then
completed problems requiring either analytical reasoning or rote
memory.
Research has indicated that analytical reasoning suffers more
from self-regulatory depletion than does rote memory (Schmeichel
et al., 2003). Consistent with the idea that coping with thoughts of
death depletes self-regulatory resources, we therefore predicted
that mortality salience would cause participants to perform worse
on problems requiring analytical reasoning but not rote memory.
Method
Participants. Participants were 38 undergraduates (24 women, 14
men). Data from 1 participant who failed to follow instructions were
excluded from all analyses. Participants were run together in a large group
and were told that the study was to investigate personality and cognitive
abilities.
Procedure. Participants randomly assigned to the mortality-salience
condition first completed the essay about death used in the previous
studies. Participants assigned to the uncertainty-salience condition de-
scribed the emotions uncertainty arouses in them and what would happen
to them if they were to experience uncertainty. Afterward, participants
57
completed a filler questionnaire for approximately 5 min to allow them to
suppress any thoughts about death they may have had.
Participants then completed either 6 analytical reasoning problems or 20
verbal definition problems requiring rote memory (e.g., defining morato-
rium, allude, augment). These problems were borrowed from a Graduate
Record Examinations preparation book and were classified as easy. Each
problem had 4 or 5 multiple-choice answers. Participants were instructed to
choose the most appropriate answer.
Results and Discussion
A 2 (mortality vs. uncertainty salience) ⫻ 2 (analytical vs. rote
memory problems) between-subjects analysis of variance indi-
cated a significant interaction between mortality-salience condi-
tion and problem type, F(1, 33) ⫽ 4.76, p ⬍ .05. Simple contrasts
indicated that, among participants who completed the analytical
reasoning problems, mortality-salience participants (M ⫽ 1.38,
SD ⫽ 1.19) solved fewer problems correctly than did uncertainty-
salience participants (M ⫽ 2.56, SD ⫽ 1.13), F(1, 33) ⫽ 6.34, p ⬍
.05. Among participants who completed the rote memory problems
(verbal definitions), mortality-salience (M ⫽ 16.20, SD ⫽ 1.55)
and uncertainty-salience (M ⫽ 15.50, SD ⫽ 3.27) participants did
not differ in the number of problems solved correctly (F ⬍ 1, ns).
Mortality salience thus impaired performance on problems re-
quiring self-control (analytical reasoning skills) but not problems
requiring only more basic cognitive processing (rote memory for
verbal definitions). This underscores the notion that mortality
salience depletes self-regulatory strength and that the deficits in
self-regulation following mortality salience are particular to self-
regulation and not cognitive processing in general.
An alternative interpretation of why mortality salience might
impair self-regulation, however, might be that thoughts of death
rebound after mortality salience (e.g., Arndt et al., 1997; Wegner,
1994) and thereby distract participants from self-regulatory tasks.
Similarly, it is plausible that other aftereffects of mortality salience
(e.g., worldview activation; Arndt, Greenberg, & Cook, 2002)
could somehow interfere with self-regulation. The results of Stud-
ies 6 and 7 directly contradict these possibilities, however. Mor-
tality salience did not impair performance on congruent Stroop
trials (Study 6) or on rote memory (Study 7), as one might have
expected if the hyperaccessibility of death thoughts or worldview
constructs impaired cognitive performance. If participants’ minds
were full of rebounding thoughts of death, for instance, they
probably would not have been able to perform very well at rote
memory.
To provide another test of the alternative explanation that mor-
tality salience impairs self-regulation via distraction, in Study 8 we
tested for moderation by trait self-esteem. There is some evidence
that mortality salience might not increase the accessibility of death
thoughts or worldview defense among individuals with high self-
esteem (see Pyszczynski et al., 2004). We made the following
hypothesis: If the hyperaccessibility of death thoughts and not the
depletion of self-regulatory resources impairs self-regulation fol-
lowing mortality salience, then self-esteem might moderate the
effect of mortality salience on self-regulation.
Study 8
Participants in Study 8 first wrote about either death or dental
pain and then solved anagrams. Solving anagrams has been used
58 GAILLIOT, SCHMEICHEL, AND BAUMEISTER
frequently in prior research on self-regulation (e.g., Baumeister et
al., 1998; Gordijn et al., 2004). The task requires self-control
insofar as one must combine letters into different groupings and
then break them apart (overriding) to try a different combination.
Persistence and attention control are also required, insofar as one
must work diligently on the task and avoid being distracted, and so
self-regulation is required to remain focused and succeed. We
predicted that participants would solve fewer anagrams after writ-
ing about death than after writing about dental pain, consistent
with the idea that suppressing thoughts of death requires and
therefore depletes self-regulatory strength.
Method
Participants. Participants were 46 undergraduates (32 women, 14
men) who completed the Rosenberg Self-Esteem Scale during a mass-
testing session at the start of the semester. The present study occurred
approximately 3 months later. The study was conducted in a classroom
setting, and participants were told that the study was to investigate per-
sonality and cognition.
Procedure. At the beginning of the session, participants were given a
list of 80 five-letter anagrams to solve to serve as a baseline measure of
anagram-solving ability. They were given 5 min to complete as many
anagrams as they could. Next, participants were randomly assigned to
respond to the questions about death or dental pain that were used in Study
6. To provide a delay that would allow time for participants to suppress
thoughts of death, we then asked participants to complete two filler
questionnaires for approximately 5 min. For the final task, participants
were given another list of anagrams to solve for 5 min.
Last, participants were probed for suspicion. No participant indicated
that he or she thought that writing about death or dental pain might
influence performance at solving anagrams.
Results and Discussion
Anagram performance. An analysis of covariance that used
the number of anagrams solved at the start of the session as a
covariate indicated that mortality-salience participants solved
fewer anagrams at the end of the session (M ⫽ 18.27, SD ⫽ 5.68)
than did dental-pain participants (M ⫽ 21.80, SD ⫽ 5.68), F(1,
43) ⫽ 4.41, p ⬍ .05. Mortality-salience and dental-pain partici-
pants did not differ in the number of anagrams solved at the start
of the session (t ⬍ 1, ns). These results are consistent with the idea
that coping with thoughts of death requires and therefore depletes
self-control strength.
Self-esteem. A regression analysis indicated that the effect of
mortality salience on anagram performance was not moderated by
self-esteem (t ⬍ 1, ns). This suggests that the effect of mortality
salience was not related to self-esteem, as might have been ex-
pected if the relationship between mortality salience and anagram
performance was attributable to hyperaccessibility of death
thoughts or worldview constructs.
Study 9
The purpose of Study 9 was to provide additional evidence that
mortality salience depletes self-regulatory strength and that the
self-regulatory impairments following mortality salience are not
caused by worldview activation or a rebound in thoughts of death.
Specifically, after writing about death or dental pain, participants
completed a sheet filled with word fragments (e.g., so_a). We told
participants to solve all of the word fragments but that they could
stop when they wanted. The word fragments were relatively easy
and with enough persistence (self-control), one could eventually
solve each one by substituting into the word fragment different
letters in the alphabet. Still, the task is lengthy and discouraging,
and so persisting requires one to override the impulse (strength-
ened by the experimenter’s permission) to quit. In this sense,
quitting early and leaving more word fragments unsolved was an
indicator of impaired self-regulation.
Further, as in the previous studies, some of the word fragments
could be solved with words related to death. Responses to these
items served as a measure of death-thought accessibility that
allowed us to test whether impaired self-regulation following
mortality salience might be attributable to the hyperaccessibility of
death thoughts.
As another approach to examine whether the hyperaccessibility
of death thoughts (rather than self-control depletion) after mortal-
ity salience impairs subsequent self-regulatory performance, we
provided an opportunity for participants to engage in worldview
defense before solving the word fragments. There is some evi-
dence that engaging in worldview defense after thinking about
death reduces the accessibility of death thoughts (Arndt et al.,
1997; Greenberg et al., 2001). Therefore, if mortality salience
leads to worldview defense, then any subsequent impairments in
self-regulation would likely not be due to heightened accessibility
of death thoughts.
Method
Participants. Participants were 55 undergraduates (24 women, 18
men, 13 unreported). The study was conducted in a classroom setting, and
participants were told that it was to investigate the relationship between
people’s attitudes and their lexicon. Participants received a packet that
contained all materials for the experiment and were allowed to work
through the packet at their own pace.
Procedure. Participants first completed a filler questionnaire and then
the same mortality-salience manipulation used in Study 8. Specifically,
participants were randomly assigned to write about either death or dental
pain. After this questionnaire, participants completed two more filler
questionnaires for approximately 5 min to provide time for participants to
suppress any thoughts about death.
The next questionnaire constituted the measure of worldview defense.
Specifically, participants read two handwritten essays about the United
States that were ostensibly written by two foreigners (borrowed from
Greenberg, Simon, Pyszczynski, Solomon, & Chatel’s, 1992, study). The
order of the two essays was counterbalanced across participants. One essay
was in support of the United States and praised Americans, whereas the
other essay was not in support of the United States and criticized Ameri-
cans. Participants evaluated the truth and validity of the essay and the
likability, intelligence, and knowledgeability of each essay’s author on
9-point scales. The summed evaluations of each essay served as the
measures of favorability toward worldview-consistent and worldview-
inconsistent opinions, respectively. In accord with past research (e.g.,
Greenberg et al., 1994), worldview defense was defined as the difference
between these two measures. Larger differences indicate more pronounced
worldview defense.
For the final task, participants completed the same list of 20 word
fragments used in Study 1A. Specifically, 15 word fragments could be
solved with only neutral words not related to death (e.g., soda), and 5 word
fragments could be solved with either neutral or death-related words.
Participants were asked to do their best to solve all of the word fragments
 THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION
and were given as much time as they needed to complete the task. It was
left to their discretion as to when they decided to stop.
Results and Discussion
Worldview defense. Mortality-salience participants engaged in
worldview defense (M ⫽ 13.71, SD ⫽ 12.93) to a greater extent
than did dental-pain participants (M ⫽ 7.69, SD ⫽ 13.06), t(51) ⫽
⫺1.65, p ⫽ .05 (one-tailed). This finding is consistent with pre-
vious research showing that mortality salience increases world-
view defense. Given past findings that worldview defense reduces
the accessibility of death thoughts (Arndt et al., 1997; Greenberg
et al., 2001), this suggests that the accessibility of death thoughts
may have been attenuated among mortality-salience participants.
Unsolved word fragments. Mortality-salience participants left
more neutral word fragments unsolved (M ⫽ 3.52, SD ⫽ 3.86)
than did dental-pain participants (M ⫽ 1.43, SD ⫽ 1.58),
t(27.37) ⫽ ⫺2.45, p ⬍ .05 (degrees of freedom have been adjusted
to correct for a violation of the homogeneity of variance assump-
tion). This is consistent with the idea that mortality salience, or at
least the typical response to it, is depleting. Mortality-salience
participants persisted less on the subsequent task than did dental-
pain participants.
Further, the difference between the two conditions in the num-
ber of word fragments left unsolved remained significant even
when controlling for the number of target items completed with
words related to death (implicit death thoughts) and worldview
defense scores (both Fs ⱖ 6.87, ps ⬍ .05). This suggests that any
difference between mortality-salience and dental-pain participants
in the extent to which death thoughts were cognitively accessible
or the extent to which they engaged in worldview defense did not
account for the finding that mortality-salience participants left
more word fragments unsolved. The impairments in self-regulation
among mortality-salience participants therefore appeared indepen-
dent of death-thought accessibility and worldview defense.
In addition, mortality-salience and dental-pain participants did
not differ in the number of target items related to death unsolved
(t ⬍ 1.05, ns). This suggests that mortality-salience participants
did not leave more neutral word fragments unsolved simply be-
cause they solved instead more death-related word fragments.
Further analysis indicated that greater worldview defense was
associated with leaving more neutral word fragments unsolved,
r(53) ⫽ .27, p ⫽ .05. This suggests that being more threatened by
the thought of death (and hence being more defensive) was asso-
ciated with greater depletion. The fact that worldview defense did
not account for the effect of mortality salience on depletion,
however, indicates that engaging in worldview defense was not the
cause of depletion. Rather, it was likely the self-regulatory effort
of having to cope with the threat of death that depleted self-
regulatory resources.
In sum, participants who wrote about death solved fewer neutral
word fragments than did participants who wrote about dental pain.
Results also indicated that the accessibility of death thoughts
probably did not account for the differences in self-control.
Mortality-salience participants persisted less even though they
were given the opportunity to engage in worldview defense, which
should have attenuated the accessibility of death thoughts (Arndt et
al., 1997; Greenberg et al., 2001). Likewise, the number of implicit
death thoughts and the extent of worldview defense did not seem
59
to account for their reduced persistence. The most parsimonious
explanation for these results is that participants’ self-regulatory
resources were depleted following mortality salience.
General Discussion
The current work shows that self-regulation plays a key role in
managing thoughts of death and death-related anxiety. Using a
variety of measures and manipulations, in nine studies we showed
that poor self-regulation undermines the management of mortality
concerns and that the management of mortality concerns under-
mines subsequent self-regulation. Specifically, people with low
trait self-control had more thoughts about death and higher death
anxiety than did people with high trait self-control (Studies 1A–
1C). Low state self-control, created by prior and seemingly irrel-
evant exercises that depleted self-regulatory strength, likewise
made people more susceptible to intrusive thoughts about death
(Studies 2–3). Perceiving oneself to be low in self-regulatory
strength was associated with a similar upsurge in thoughts of death
and death anxiety (Study 4). In addition, one defensive reaction to
the threat of death is to increase support for an ostensibly strong
and patriotic authority figure (Cohen et al., 2004; Landau et al.,
2004), and such defensive shifts in political preferences in re-
sponse to thoughts of death were found mainly among people low
in trait self-control (Study 5).
Further, mortality salience led to impairments in self-regulation.
Participants who had to think and write about death later per-
formed worse on the Stroop task (Study 6) and analytical reason-
ing problems (Study 7), solved fewer anagrams (Study 8), and quit
sooner on a task of persistence (Study 9), as compared with
participants who wrote about a control topic. People effortfully
suppress thoughts of death following their activation (Arndt et al.,
1997; Greenberg et al., 1994; Harmon-Jones et al., 1997; Pyszc-
zynski et al., 1999), and so it seems likely that the act of suppress-
ing thoughts of death after writing about death depleted partici-
pants’ self-regulatory strength.
These findings are consistent with the view that thoughts about
death and mortality are disturbing and are widely treated as threats
that must be kept at bay. Specifically, we propose that the world
contains many cues that could evoke thoughts and feelings about
death, and that people use self-regulation to prevent these cues
from flooding the conscious mind with troublesome thoughts and
aversive emotions such as anxiety (Greenberg, Pyszczynski, &
Solomon, 1986). But because the capacity for self-regulation is
limited, and because some people have more of it than others, the
regulatory process is not consistently effective. Hence some stim-
uli will evoke conscious thoughts about death, at least for some of
the people some of the time. Trait and state capacity for self-
regulation appears to offer an effective basis for predicting and
understanding who will suffer most from such thoughts and fears.
Implications, Limitations, and Alternative Explanations
These results are consistent with previous work showing that
people actively and effortfully suppress thoughts of death (e.g.,
Greenberg et al., 1994; Harmon-Jones et al., 1997). Whereas
previous work showed that a concurrent cognitive load undermines
the suppression of death thoughts (Arndt et al., 1997), the current
work indicates that depleted self-regulatory strength and low trait
60 GAILLIOT, SCHMEICHEL, AND BAUMEISTER
self-control also undermine the suppression of death thoughts. And
conversely, the suppression of death thoughts depletes self-control
strength. Taken together, the picture that emerges is that people
actively seek to suppress or minimize awareness of death and that
concurrent cognitive activity, previous self-regulatory exertion, or
habitually poor self-regulatory abilities undermine such efforts.
Though a few alternative explanations may be proposed for the
link between self-control and mortal concern, they seem largely
unable to account for the bulk of the current findings. Specifically,
the results appeared independent from any effects of self-esteem,
arousal, mood, or general negative affect (i.e., general anxiety and
social anxiety in Study 1C). Some effects may be explainable in
terms of social desirability, such as if people giving socially
desirable responses reported having both high self-control and low
death anxiety, yet this view cannot adequately account for all of
the findings. It is unclear, for example, how social desirability
would have been related to support for President Bush (Study 5).
The social desirability explanation also does not easily account for
the fact that mortality salience impaired performance on tasks that
required self-control but not other cognitive tasks (Studies 6 and
7). We also controlled for self-esteem in most of the studies.
Insofar as reporting having high self-esteem indicates giving so-
cially desirable responses, then the results do not appear caused by
differences in social desirability.
Further, the effect of depletion following mortality salience did
not appear attributable to distraction by thoughts about death
(Study 9), active attempts toward defending one’s worldviews
(Study 9), or deficits to general cognitive abilities (Studies 6 and
7). The most parsimonious conclusion is that mortality salience
impaired self-control by depleting self-regulatory resources. In-
deed, mortality salience increased worldview defense (Study 9), a
defensive reaction aimed at coping with the threat of death, and
greater worldview defense predicted (but did not account for)
greater depletion. Participants who seemingly attempted to cope
the most with death and therefore found it the most frightening
were the most depleted.
Studies 1A–1C were correlational, which makes it difficult to
draw causal conclusions. Still, they do indicate a personality
pattern that seems to make people more vulnerable to unwelcome
thoughts and emotions. Preliminary evidence has linked low self-
control to a variety of maladaptive behavior patterns (e.g., Gott-
fredson & Hirschi, 1990; Tangney et al., 2004). It is plausible that
one reason people with low versus high self-control engage in such
maladaptive behaviors is that they have substantial death anxiety.
For example, both low self-control and death anxiety are associ-
ated with increased aggression (McGregor et al., 1998; Stucke &
Baumeister, in press; Tangney et al., 2004) and risky sexual
behavior (Taubman Ben-Ari, 2004; Wills, Gibbons, Gerrard,
Murry, & Brody, 2003). Perhaps people with low self-control
engage in such behaviors in part as a response to preoccupation
with death (see Tice, Bratslavsky, & Baumeister, 2001).
If we could generalize the present results to other sources of
stress, threat, and trauma, this work would be even more impor-
tant, but we hesitate to generalize without further evidence. It is
plausible that self-regulation might be more closely tied to coping
with the threat of death than with other threats or aversive states.
The thought of death can be especially threatening and trigger
active efforts toward suppression (e.g., Arndt et al., 1997), perhaps
more so than most other thoughts. Our use of dental pain and
uncertainty as control conditions was intended to differentiate
death from other bad thoughts, and clearly in the present studies
death had effects that were not duplicated by dental pain or
uncertainty. Death could represent a particularly important prob-
lem because it threatens to nullify one’s life and its meaning, and
so people may seek to defend against the thought of death more
extensively than they would defend against the thought of other
misfortunes. Future research may profitably examine whether self-
regulation and self-control depletion are relevant to psychological
defenses against more than death.
Concluding Remarks
Coping with unwanted thoughts and feelings can be an effortful
struggle. Death is inevitable and inescapable, and so people must
protect themselves from thoughts of death if they are to reduce the
anxiety that the awareness of death might evoke. Rational attempts
to deny death ultimately fail, however, and so people must engage
in other defenses to ward off the threat of death (Pyszczynski et al.,
1999). The current work indicates that self-regulation is a useful
tool in the psychological defense against death. Through effective
self-regulation, people can minimize their awareness of death and
reduce anxiety associated with death. But this peace of mind
comes at a cost. When thoughts of death arise, it takes effortful
self-control to suppress them, and these efforts consume resources
that leave people with less self-control afterward— causing people
to perform less effectively on many tasks. Conversely, expending
resources on any sort of self-control task reduces one’s defenses,
thereby opening the door for disturbing thoughts of death.
References
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and
interpreting interaction. Newbury Park, CA: Sage.
Anderson, M. C., & Green, C. (2001, March 15). Suppressing unwanted
memories by executive control. Nature, 6826, 366 –369.
Aries, P. (1981). The hour of our death. New York: Oxford University
Press.
Arndt, J., Greenberg, J., & Cook, A. (2002). Mortality salience and the
spreading activation of worldview-relevant constructs: Exploring the
cognitive architecture of terror management. Journal of Experimental
Psychology: General, 131, 307–324.
Arndt, J., Greenberg, J., Solomon, S., Pyszczynski, T., & Simon, L. (1997).
Suppression, accessibility of death-related thoughts, and cultural world-
view defense: Exploring the psychodynamics of terror management.
Journal of Personality and Social Psychology, 73, 5–18.
Baumeister, R. F., Bratslavsky, E., Muraven, M., & Tice, D. M. (1998).
Ego depletion: Is the active self a limited resource? Journal of Person-
ality and Social Psychology, 74, 1252–1265.
Baumeister, R. F., Heatherton, T. F., & Tice, D. M. (1994). Losing control:
How and why people fail at self-regulation. San Diego, CA: Academic
Press.
Becker, E. (1973). The denial of death. New York: Academic Press.
Cohen, F., Solomon, S., Maxfield, M., Pyszczynski, T., & Greenberg, J.
(2004). Fatal attraction: The effects of mortality salience on evaluations
of charismatic, task-oriented, and relationship-oriented leaders. Psycho-
logical Science, 15, 846 – 851.
Feifel, H., & Branscomb, A. B. (1973). Who’s afraid of death? Journal of
Abnormal Psychology, 81, 282–288.
Finkel, E. J., & Campbell, W. K. (2001). Self-control and accommodation
in close relationships: An interdependence analysis. Journal of Person-
ality and Social Psychology, 81, 263–277.
 THOUGHTS AND FEARS OF DEATH AND SELF-REGULATION
Florian, V., & Mikulincer, M. (1997). Fear of death and the judgment of
social transgressions: A multidimensional test of terror management
theory. Journal of Personality and Social Psychology, 73, 369 –380.
Gailliot, M. T., & Baumeister, R. F. (in press). Self-control and business
ethics: How strengthening the self benefits the corporation and the
individual. In R. A. Giacalone, C. L. Jurkiewicz, & C. Dunn (Eds.),
Positive psychology, ethics, and social responsibility in organizational
life. Greenwich, CT: Information Age.
Gilbert, D., T., Krull, D. S., & Pelham, B. W. (1988). Of thoughts
unspoken: Social inference and the self-regulation of behavior. Journal
of Personality and Social Psychology, 55, 685– 694.
Gordijn, E. H., Hindriks, I., Koomen, W., Dijksterhuis, A., & Van Knip-
penberg, A. (2004). Consequences of stereotype suppression and internal
suppression motivation: A self-regulation approach. Personality and
Social Psychology Bulletin, 30, 212–224.
Gottfredson, M. R., & Hirschi, T. (1990). A general theory of crime.
Stanford, CA: Stanford University Press.
Greenberg, J., Arndt, J., Schimel, J., Pyszczynski, T., & Solomon, S.
(2001). Clarifying the function of mortality salience-induced worldview
defense: Renewed suppression or reduced accessibility of death-related
thoughts? Journal of Experimental Social Psychology, 37, 70 –76.
Greenberg, J., Martens, A., Jonas, E., Eisenstadt, D., Pyszczynski, T., &
Solomon, S. (2003). Psychological defense in anticipation of anxiety:
Eliminating the potential for anxiety eliminates the effect of mortality
salience on worldview defense. Psychological Science, 14, 516 –519.
Greenberg, J., Pyszczynski, T., & Solomon, S. (1986). The causes and
consequences of a need for self-esteem: A terror management theory. In
R. F. Baumeister (Ed.), Public self and private self (pp. 189 –212). New
York: Springer-Verlag.
Greenberg, J., Pyszczynski, T., Solomon, S., Rosenblatt, A., Veeder, M.,
Kirkland, S., & Lyon, D. (1990). Evidence for terror management theory
II: The effects of mortality salience on reactions to those who threaten or
bolster the cultural worldview. Journal of Personality and Social Psy-
chology, 58, 308 –318.
Greenberg, J., Pyszczynski, T., Solomon, S., Simon, L., & Breus, M.
(1994). Role of consciousness and accessibility of death-related thoughts
in mortality salience effects. Journal of Personality and Social Psychol-
ogy, 67, 627– 637.
Greenberg, J., Simon, L., Pyszczynski, T., Solomon, S., & Chatel, D.
(1992). Terror management and tolerance: Does mortality salience al-
ways intensify negative reactions to others who threaten one’s world-
view? Journal of Personality and Social Psychology, 63, 212–220.
Harmon-Jones, E., Simon, L., Greenberg, J., Pyszczynski, T., Solomon, S.,
& McGregor, H. (1997). Terror management theory and self-esteem:
Evidence that increased self-esteem reduces mortality salience effects.
Journal of Personality and Social Psychology, 72, 24 –36.
Heine, S. J., Harihara, M., & Niiya, Y. (2002). Terror management in
Japan. Asian Journal of Social Psychology, 5, 187–196.
Landau, M. J., Solomon, S., Greenberg, J., Cohen, F., Pyszczynski, T.,
Arndt, J., et al. (2004). Deliver us from evil: The effects of mortality
salience and reminders of 9/11 on support for President George W.
Bush. Personality and Social Psychology Bulletin, 30, 1136 –1150.
MacLeod, C. M. (1989). Directed forgetting affects both direct and indirect
tests of memory. Journal of Experimental Psychology: Learning, Mem-
ory, and Cognition, 15, 13–21.
MacLeod, C. M. (1991). Half a century of research on the Stroop effect: An
integrative review. Psychological Bulletin, 109, 163–203.
Mattick, R. P., & Clarke, J. C. (1998). Development and validation of
measures of social phobia scrutiny fear and social interaction anxiety.
Behaviour Research and Therapy, 36, 455– 470.
Mayer, J. D., & Gaschke, Y. N. (1988). The experience and meta-
experience of mood. Journal of Personality and Social Psychology, 55,
102–111.
McBride, D. M., & Dosher, B. A. (1997). A comparison of forgetting in an
61
implicit and explicit memory task. Journal of Experimental Psychology:
General, 126, 371–392.
McGregor, H. A., Lieberman, J. D., Greenberg, J., Solomon, S., Arndt, J.,
Simon., L., & Pyszczynski, T. (1998). Terror management and aggres-
sion: Evidence that mortality salience motivates aggression against
worldview-threatening others. Journal of Personality and Social Psy-
chology, 74, 590 – 605.
Mischel, W., Shoda, Y., & Peake, P. K. (1988). The nature of adolescent
competencies predicted by preschool delay of gratification. Journal of
Personality and Social Psychology, 54, 687– 696.
Muraven, M., & Baumeister, R. F. (2000). Self-regulation and depletion of
limited resources: Does self-control resemble a muscle? Psychological
Bulletin, 126, 247–259.
Muraven, M., Tice, D. M., & Baumeister, R. F. (1998). Self-control as a
limited resource: Regulatory depletion patterns. Journal of Personality
and Social Psychology, 74, 774 –789.
Ochsmann, R., & Mathey, M. (1994). Depreciating of and distancing from
foreigners: Effects of mortality salience. Unpublished manuscript, Uni-
versitat Mainz, Mainz, Germany.
Pollak, J. M. (1979). Correlates of death anxiety: A review of empirical
studies. Omega: Journal of Death & Dying, 10, 97–121.
Pyszczynski, T., Greenberg, J., & Solomon, S. (1999). A dual process
model of defense against conscious and unconscious death-related
thoughts: An extension of terror management theory. Psychological
Review, 106, 835– 845.
Pyszczynski, T., Greenberg, J., Solomon, S., Arndt, J., & Schimel, J.
(2004). Why do people need self-esteem? A theoretical and empirical
review. Psychological Bulletin, 130, 435– 468.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton,
NJ: Princeton University Press.
Rosenblatt, A., Greenberg, J., Solomon, S., Pyszczynski, T., & Lyon, D.
(1989). Evidence for terror management theory: I. The effects of mor-
tality salience on reactions to those who violate or uphold cultural
values. Journal of Personality and Social Psychology, 57, 681– 690.
Schmeichel, B. J., Vohs, K. D., & Baumeister, R. F. (2003). Intellectual
performance and ego depletion: Role of the self in logical reasoning and
other information processing. Journal of Personality and Social Psy-
chology, 85, 33– 46.
Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human
information processing: I. Detection, search, and attention. Psychologi-
cal Review, 84, 1– 66.
Shoda, Y., Mischel, W., & Peake, P. K. (1990). Predicting adolescent
cognitive and self-regulatory competencies from preschool delay of
gratification: Identifying diagnostic conditions. Developmental Psychol-
ogy, 26, 978 –986.
Smart, L., & Wegner, D. M. (1999). Covering up what can’t be seen:
Concealable stigma and mental control. Journal of Personality and
Social Psychology, 77, 474 – 486.
Spielberger, C. D., Gorsuch, R. L., & Lushene, R. E. (1970). Manual
for the State–Trait Anxiety Inventory. Palo Alto, CA: Consulting
Psychologists.
Stucke, T. S., & Baumeister, R. F. (in press). Ego depletion and aggressive
behavior: Is the inhibition of aggression a limited resource? European
Journal of Social Psychology.
Tangney, J. P., Baumeister, R. F., & Boone, A. L. (2004). High self-control
predicts good adjustment, less pathology, better grades, and interper-
sonal success. Journal of Personality, 72, 271–322.
Taubman Ben-Ari, O. (2004). Intimacy and risky sexual behavior: What
does it have to do with death? Death Studies, 28, 865– 887.
Templer, D. I. (1970). The construction and validation of a death anxiety
scale. Journal of General Psychology, 82, 165–177.
Tice, D. M., Bratslavsky, E., & Baumeister, R. F. (2001). Emotional
distress regulation takes precedence over impulse control: If you feel
bad, do it! Journal of Personality and Social Psychology, 80, 53– 67.
62 GAILLIOT, SCHMEICHEL, AND BAUMEISTER
Twenge, J. M., Muraven, M., & Tice, D. M. (2004). Measuring state
self-control: Reliability, validity, and correlations with physical and
psychological stress. Unpublished manuscript, San Diego State Univer-
sity.
van den Bos, K., Poortvliet, P. M., Maas, M., Miedema, J., & van den Ham,
E. J. (2005). An enquiry concerning the principles of cultural norms and
values: The impact of uncertainty and mortality salience on reactions to
violations of bolstering of cultural worldviews. Journal of Experimental
Social Psychology, 41, 91–113.
Wegner, D. M. (1994). Ironic processes of mental control. Psychological
Review, 101, 34 –52.
Wegner, D. M., Schneider, D., Carter, S. R., & White, T. L. (1987).
Paradoxical effects of thought suppression. Journal of Personality and
Social Psychology, 53, 5–13.
Wegner, D. M., & Zanakos, S. (1994). Chronic thought suppression.
Journal of Personality, 62, 615– 640.
Wenzlaff, R. M., & Wegner, D. M. (2000). Thought suppression. Annual
Review of Psychology, 51, 59 –91.
Wills, T. A., Gibbons, F. X., Gerrard, M., Murry, V. M., & Brody, G. H.
(2003). Family communication and religiosity related to substance use
and sexual behavior in early adolescence: A test for pathways through
self-control and prototype perceptions. Psychology of Addictive Behav-
iors, 17, 312–323.
Received May 1, 2005
Revision received November 1, 2005
Accepted November 3, 2005 䡲
62 GAILLIOT, SCHMEICHEL, AND BAUMEISTER
Twenge, J. M., Muraven, M., & Tice, D. M. (2004). Measuring state
self-control: Reliability, validity, and correlations with physical and
psychological stress. Unpublished manuscript, San Diego State Univer-
sity.
van den Bos, K., Poortvliet, P. M., Maas, M., Miedema, J., & van den Ham,
E. J. (2005). An enquiry concerning the principles of cultural norms and
values: The impact of uncertainty and mortality salience on reactions to
violations of bolstering of cultural worldviews. Journal of Experimental
Social Psychology, 41, 91–113.
Wegner, D. M. (1994). Ironic processes of mental control. Psychological
Review, 101, 34 –52.
Wegner, D. M., Schneider, D., Carter, S. R., & White, T. L. (1987).
Paradoxical effects of thought suppression. Journal of Personality and
Social Psychology, 53, 5–13.
Wegner, D. M., & Zanakos, S. (1994). Chronic thought suppression.
Journal of Personality, 62, 615– 640.
Wenzlaff, R. M., & Wegner, D. M. (2000). Thought suppression. Annual
Review of Psychology, 51, 59 –91.
Wills, T. A., Gibbons, F. X., Gerrard, M., Murry, V. M., & Brody, G. H.
(2003). Family communication and religiosity related to substance use
and sexual behavior in early adolescence: A test for pathways through
self-control and prototype perceptions. Psychology of Addictive Behav-
iors, 17, 312–323.
Received May 1, 2005
Revision received November 1, 2005
Accepted November 3, 2005 䡲
 INTERPERSONAL RELATIONS AND GROUP PROCESSES

Peacocks, Picasso, and Parental Investment: The Effects of Romantic

Motives on Creativity
Vladas Griskevicius, Robert B. Cialdini, and Douglas T. Kenrick
Arizona State University

Four experiments explored the effects of mating motivation on creativity. Even without other incentives
to be creative, romantic motives enhanced creativity on subjective and objective measures. For men, any
cue designed to activate a short-term or a long-term mating goal increased creative displays; however,
women displayed more creativity only when primed to attract a high-quality long-term mate. These
creative boosts were unrelated to increased effort on creative tasks or to changes in mood or arousal.
Furthermore, results were unaffected by the application of monetary incentives for creativity. These
findings align with the view that creative displays in both sexes may be linked to sexual selection,
qualified by unique exigencies of human parental investment.

Keywords: creativity, sexual selection, parental investment, self-presentation, mating goals

“In order to create there must be a dynamic force—and what force is mythology—are universally female. Yet if “there is no biological
more potent than love?” reason why a man can’t provide the elements of inspiration” (p. 9,
—Igor Stravinsky Prose, 2002), how could it be that the elixir of inspiration seems to
be primarily concocted by women and predominantly imbibed by
The Guinness Book of World Records lists Pablo Picasso as the
men?
most prolific artist in history with an astounding 147,800 works of
The current research presents an evolutionary cognitive frame-
art. Picasso’s career is often depicted as a tortuous series of
work designed to shed light on the mystery of the muses and also
profoundly inspired artistic periods— blue, rose, cubist, surreal-
to explain and predict a wider range of behavioral phenomena.
ist—in which his subjects underwent extravagant visual transfor-
This framework is grounded in two underlying premises: First, that
mations at the hands of a creative genius performing at the apogee
a number of human mental traits—including the capacity for and
of his ability. But a closer look at Picasso’s generative periods
display of creativity—may in part be linked to evolutionary pro-
reveals an intriguing constant: Each new epoch blossoms with
cesses of sexual selection and differential parental investment
paintings of a new woman—not a sitter or a model, but a mis-
(Miller, 1999, 2000; Trivers, 1972); and second, that evolutionar-
tress— each of whom is touted to have served Picasso as an
ily relevant contextual cues can serve to activate goals (e.g., the
incandescent, albeit temporary, muse (Crespelle, 1969;
goal to attract a mate) that facilitate behaviors historically associ-
MacGregor-Hastie, 1988). Picasso’s artistic history, however, is
ated with success for the attainment of such goals (e.g., Maner et
not unique: Creative juggernauts such as Salvador Dalı́, Friedrich
al., 2005; Roney, 2003; Wilson & Daly, 2004). From this foun-
Nietzsche, and Dante were also acutely inspired by their own
dation, four experiments examine whether cues designed to tem-
muses (Prose, 2002). The enigmatic notion of a muse is rooted in
porarily activate a goal to attract a mate can increase people’s
Greek mythology, in which nine godly muses traversed the land,
creativity. Drawing on the theory of differential parental invest-
stirring the creative spirits of mortal artists and scientists. And
ment, the research also explores which specific cues might stim-
according to historian Francine Prose (2002), all muses share one
ulate a creative boost for men versus women. Finally, the studies
striking and inextricable feature: Muses— both in history and in
gather evidence regarding the nature of the psychological mecha-
nism that may underlie how mating motives serve to foster
creativity.
Vladas Griskevicius, Robert B. Cialdini, and Douglas T. Kenrick, De-
partment of Psychology, Arizona State University.
This research was facilitated by a National Science Foundation Graduate Natural and Sexual Selection of Creative Displays
Research Fellowship awarded to Vladas Griskevicius and by National
The construct of creativity is composed of multiple dimensions
Institutes of Health Grant 5R01MH64734 to Douglas T. Kenrick. We thank
Steve Neuberg, Cathy Cottrell, Noah Goldstein, and Josh Tybur for their
that share a common thread: Creativity is the ability to produce
helpful comments on a draft of this article. work that is both novel (i.e., original, unexpected) and appropriate
Correspondence concerning this article should be addressed to Vladas (i.e., useful, valuable; Sternberg & Lubart, 1999). Although clas-
Griskevicius, Department of Psychology, Arizona State University, sical creativity research has not been concerned with the ultimate
Tempe, AZ 85287-1104. E-mail: vladasg@asu.edu origins of human creativity (e.g., see Simonton, 2000), those who
Journal of Personality and Social Psychology, 2006, Vol. 91, No. 1, 63–76
Copyright 2006 by the American Psychological Association 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.63

63
64 GRISKEVICIUS, CIALDINI, AND KENRICK
have offered evolutionary explanations have generally presumed
that creative abilities evolved because they somehow enhanced the
likelihood of our ancestors’ survival without adding unnecessary
metabolic costs (e.g., Byrne, 1995; Gibson & Ingold, 1993; King-
don, 1993). For example, a creative way to spear a fish or build a
hut could have enhanced survival for their creator and his or her
kin. Other evolutionary theorists have speculated that creative
capacities may be a byproduct of various perceptual and cognitive
mechanisms (e.g., Pinker, 1997). However, Miller (2000; Haselton
& Miller, in press) has recently argued that such explanations are
inadequate for several reasons. For instance, not only have other
large-brained animals not evolved similar capacities, but many
human displays of creativity are highly valued socially yet are
difficult to explain in terms of survival value. For example, a
farmer produces more tangible survival benefits in a week than a
team of musicians, poets, and sculptors will produce in a lifetime.
Yet a provocative melody, poem, or sculpture is likely to elicit
greater appreciation than an absolutely perfect melon, potato, or
zucchini.
Instead of producing direct survival benefits, Miller (2000) and
others have proposed that several human mental traits, including
creativity, are likely to have at least in part evolved via sexual
selection (e.g., Eysenck, 1995; Kanazawa, 2000). Unlike natural
selection, whereby traits evolve solely because they enhance the
probability of an individual’s survival, Darwin (1871) suggested
that some traits, such as the elaborate plumage of peacocks, evolve
via sexual selection—they evolve because they enhance an indi-
vidual’s ability to attract a mate, which may or may not be
independent of whether the trait enhances survival (see Andersson,
1994).
Supporting this viewpoint, human creativity has several features
in common with sexually selected traits across different species.
Just as members of various species prefer partners with prominent
sexually selected traits, such as brilliant tails, humans show a
desire for creativity in a romantic partner (Buss & Barnes, 1986;
Li, Bailey, Kenrick, & Linsenmeier, 2002). Sexually selected traits
across species also tend to function as markers of “good genes”
(Møller & Petrie, 2002; Zahavi & Zahavi, 1997). Correspondingly,
Haselton and Miller (in press) have found suggestive empirical
evidence indicating that creativity may partly serve a similar good
genes function in humans.1 Moreover, because one of the distinct
markers of sexually selected traits across species is the conspicu-
ous display of such traits in courtship (Andersson, 1994), the
current research explores the extent to which people may be more
likely to display creativity when they are primed with cues related
to courtship.
Cueing Creativity Through Mating Motivation
Mental mechanisms that evolved to solve specific adaptive
problems are often highly sensitive to ecological cues indicating a
particular adaptive problem or opportunity, such as a potential
threat or mating opportunity (Cosmides & Tooby, 1992; Öhman &
Mineka, 2001; Schaller, Faulkner, Park, Neuberg, & Kenrick,
2004; Todd & Gigerenzer, 2000). Moreover, much research has
shown that various cues can automatically activate certain goal and
need states (Chartrand & Bargh, 1996; Schaller, 2003), and that
such states can influence perception and behavior without explicit
conscious awareness (Bargh, 1990; Bargh & Chartrand, 1999).
Given the central role of reproduction in evolutionary processes,
a functional perspective suggests that mating goals are likely to be
closely linked to adaptive outcomes (Bugental, 2000; Kenrick, Li,
& Butner, 2003). Cues related to mating can serve to both activate
a mating goal and its affective responses (Scott, 1980) and trigger
specific mating-related cognitive mechanisms (Gutierres, Kenrick,
& Partch, 1999; Haselton & Buss, 2000; Kenrick, Sadalla, &
Keefe, 1998). Furthermore, mating-related motives appear to fa-
cilitate particular perceptions, cognitions, and behaviors associated
with reproductive success (Griskevicius, Goldstein, Mortensen,
Cialdini, & Kenrick, in press; Maner et al., 2005; Roney, 2003). If
displays of creativity have evolved in part because of their benefit
in courtship, cues designed to activate mating motives may also
trigger displays of creativity.
Courtship Displays and Parental Investment
But should mating motives produce creative displays in both
sexes? Across most species—and in 95%–97% of all mammals—it
is exclusively males who display sexually selected traits during
courtship (Cronin, 1993). This sex-differentiated pattern is consis-
tent with the notion of predominantly female muses eliciting
creative courtship displays in male artists (Prose, 2002). Given this
evidence, it is reasonable to hypothesize that mating cues should
elicit creativity only for males—the male-only creative display
hypothesis.
However, it is also plausible that mating cues could stimulate
creative displays for both men and women. Sex differences in
mammalian sexually selected traits are primarily linked to a spe-
cies’ levels of maternal and paternal parental investment—the time
and energy devoted to producing viable offspring (Trivers, 1972).
In those species in which courtship displays are an exclusively
male sport, such as peacocks, males tend to invest the absolute
minimum in offspring—sperm. However, in some species these
sex roles are reversed such that the males contribute the majority
of parental investment (e.g., Mormon cricket, pipefish seahorse,
Panamanian poison arrow frog). Consequently, in such species it is
the females and not the males who display their elaborate sexually
selected traits in courtship (Andersson, 1994; Buss, 2005).
Humans are somewhat different from most mammals in that
both men and women tend to contribute significant parental in-
vestment to their offspring (Geary, 2000). Thus, it is reasonable to
predict that courtship might stimulate creative displays for both
men and women. Consistent with this reasoning are findings that
men and women express relatively equal preferences for creativity
in a mate (Li et al., 2002). Thus, a consideration of the particulars
of human parental investment suggests the alternative hypothesis
that mating cues will elicit creativity for both men and women—
the unisex creative display hypothesis.
Study 1
The initial experiment tested whether priming people with cues
related to mating would cause an increase in their creative displays
relative to people primed with neutral cues unrelated to mating. To
1
More specifically, Haselton & Miller (in press) found that ovulating
women tend to unconsciously shift their preferences toward preferring
creativity-related traits compared with other desirable “good dad” traits in
potential sex partners.
 ROMANTIC MOTIVES AND CREATIVITY
test this idea, we used photographs of desirable and available
targets to prime mating. Qualitative differences in people’s dis-
plays of creativity were assessed via ratings of short stories written
by participants in each condition. Comparative cross-species find-
ings, data on human mate preferences, and parental investment
theory lent themselves to two competing hypotheses: the male-
only creative display hypothesis (that a mating prime will elicit
creativity only for males) and the unisex creative display hypoth-
esis (that a mating prime will increase creative displays by both
men and women).
Method
Participants
Ninety-one participants (Ps; 35 men and 56 women) were recruited from
introductory psychology classes as partial fulfillment of their class require-
ment. All Ps came to the lab in groups of 2 to 6 and were each seated
between partitions at a computer. The mean age for women was 19.2
(SD ⫽ 1.7), and the mean age for men was 19.7 (SD ⫽ 2.1).
Design and Procedure
The overall design of the experiment was a between-participants 2
(P sex) ⫻ 2 (Prime: mating vs. control) design. At the beginning of the
study, all Ps wrote two short stories that were used to assess their dispo-
sitional creative ability and their general motivation to display it. After the
first two writing tasks, half of the participants were primed with cues
related to mating. Following the prime, all Ps wrote an additional two
stories that were later rated for their levels of creativity. These ratings
served as the main dependent measure in the study.
Creativity tasks. Although there are multiple psychological measures
designed to evaluate various forms of creativity (see Kerr & Gagliardi,
2004), these measures are generally not designed to allow participants to
display creativity in ways akin to a courtship situation. To allow Ps the
freedom to display their creativity in multiple ways—to allow them to
produce something that could be perceived as both novel and appropri-
ate—we developed a methodology in which Ps were asked to write a short
story about an ambiguous image. Ps viewed and wrote stories about two
types of images: cartoon drawings and abstract paintings. Each of the
cartoon drawings showed a pair of individuals in an ambiguous situation
(in a prison cell and in a café), and both abstract paintings were composed
of multiple abstract colorful shapes. Each image was sequentially presented
on a computer screen, and Ps were asked “What do you think is happening
in this image?” All Ps had up to 5 min to write a story for each image,
although if they finished earlier they could advance to the next part of the
study.
Because there is considerable variation in people’s creative abilities, a
baseline level was established for each person. Thus, each P saw and wrote
about two of the four ambiguous images (one cartoon and one painting)
before the prime manipulation, and they wrote about the two other images
(the counterpart cartoon and the counterpart painting) after the prime
manipulation. The premanipulation stories served as a baseline measure of
each P’s creative ability; the postmanipulation stories served as the main
dependent measures. The order of the two types of images that Ps saw was
always the same: The cartoon drawing was first and the abstract painting
was second both before and after the manipulation. However, the order of
which version of the cartoon and painting that Ps saw was counterbalanced,
and a P never saw the same image twice. Thus, the sequence of the study
for all Ps was: Cartoon A (or B), Painting A (or B), prime manipulation,
Cartoon B (or A), and Painting B (or A).
Mating prime. After establishing a baseline level of creativity, half of
the participants were primed with mating cues similar to those that have
been successfully used to activate mating goals in similar procedures
(Roney, 2003; Wilson & Daly, 2004). To induce a romantic mindset, Ps
65
viewed an array of six total photos of three attractive opposite-sex indi-
viduals—two photos of each person. (The photos were collected from
Match.com—a dating website—and were prerated by students as being
highly attractive). All Ps were then asked to select one person from the
array whom they thought was the most desirable romantic partner. After
making their selection, the photo of the selected person remained on the
screen, and Ps were asked to imagine that they were preparing to go on a
first date with this individual. In an effort to make the mating prime more
powerful, Ps were asked to write about their idea of the perfect first date
with this person. All Ps had up to 3 min to write their descriptions,
although if they finished before time was up they had the option of
advancing to the next part of the study.
Control participants underwent a similar procedure devoid of any ro-
mantic connotations. They saw a photo of a street with several buildings
and were asked to imagine being on that street. They were then given up
to 3 min to write about their idea of the most pleasant weather conditions
in which to walk around and look at the buildings.
Priming booster shots. After the prime manipulation, all Ps wrote a
story about one of the cartoon images. Before going on to write the fourth
and last story (about the abstract painting), all Ps underwent a prime
“booster shot” to ensure that they were still in a romantic or a control frame
of mind. The booster shots were procedurally identical to the original prime
manipulations except that they consisted of a different array of photos of
attractive individuals and a photo of another building.
Creativity measures. All four stories written by each P were rated for
creativity by four student judges (two men and two women) who were
blind to experimental conditions. The judges were not experts on creativity;
instead, they were fellow students intended to resemble the type of person
who might find him or herself on a date with one of the participants. The
judges rated each story on eight attributes: the extent to which they thought
the story was creative, original, clever, imaginative, captivating, funny,
entertaining, and charming. Each attribute was rated on a 1 (not at all) to
9 (very) scale. Before beginning the rating process, judges read a selection
of the stories for each image to familiarize themselves with what to expect.
Thus, the creativity ratings of the stories were relative to the other stories
written in the same situation.
Measures of effort. The study also included two measures designed to
ascertain how much effort Ps put into writing each story. First, the amount
of time Ps took to work on each story (maximum 5 min) was recorded.
Second, the number of words Ps used to compose each story was also
counted.
Results
Overview of Stories
The stories written by participants showed quite a bit of vari-
ability in creativity. Because Ps were not specifically instructed to
be creative, a sizable portion of the stories were somewhat bland.
For instance, when writing about the cartoons, some Ps merely
wrote a sentence or two describing the situation with little addi-
tional insight (e.g., for the prison cartoon: “These are two men in
prison. They are there because they were suspected of terrorism.”
For the café cartoon: “These two people work together and are on
a break from work at a coffee shop.”). However, many Ps were
also inspired to spontaneously include more information, which
usually resulted in more creative answers, as in the following
description of what could be going on in the café:
Nigel is trying to decide whether or not to a get a nose job. He just
can’t decide. However, his friend Reginald had one and his nose was
simply stunning. Reginald is a very particular sort of fellow you
know. That latte he’s drinking had to be just so. Soy milk with a
dollop of foam and merely a whisper of cinnamon. Too much of one
ingredient might completely throw off the balance of his day. When
66 GRISKEVICIUS, CIALDINI, AND KENRICK

one is so particular about cinnamon, you could only imagine how he’d
prefer his nose. All of these things Nigel noted to himself as Reginald
went on and on.

Despite the fact that Ps wrote about what could be interpreted as

a comedic cartoon, the majority of the answers lacked any element
of humor. Only a few participants responded to the cartoon ques-
tion by writing a witty one-sentence tagline common in comic
strips (e.g., for the two imprisoned men in shackles: “I am badly in
need of a pedicure.”).
The two abstract images had fewer standard responses. Al-
though something about a psychedelic experience was periodically
mentioned, a good portion of Ps made up a more interesting short
description (e.g., “I think in this painting there’s a butterfly break-
ing out of its cocoon. However, I think this is a metaphor for
someone breaking free of their past” or “I see a basketball game in
a poor neighborhood with a lot of graffiti. The people are playing
because they hope that basketball will help them get out of this
neighborhood.”). Multiple Ps wrote responses that were more
interesting, such as the following:

The setting is a seedy, underground jazz club, where bands have to
compete with drug dealers for the patrons’ attention. A good quintet
is performing, with a tenor saxophone, two trumpets, a trombone and
a drummer. The instruments are old and worn, but the music that they
make is enough to turn the attention of the crack dealers and the
junkies. The music is haphazard and at times seems arrhythmic and
amelodic, but it fits the scene like a velvet glove.
Story Creativity
The eight measures of creativity showed high cohesiveness for
each of the four judges (alphas for the eight ratings for each judge
ranged from .88 –.94).2 Thus, the eight ratings were combined into
a creativity index for each of the four coders. The ratings of all four
coders also showed high interrater reliability (␣ ⫽ .86 for the
cartoon; ␣ ⫽ .88 for the painting), and the ratings for the four
judges were combined into a creativity index for the cartoon and a
creativity index for the painting.
We next tested whether the postmanipulation creativity ratings
of the stories about the cartoon and the painting were differentially
influenced by P sex and the prime manipulation. This was done via
a repeated-measures mixed-model 3-factor analysis of covariance
(ANCOVA) with the premanipulation creativity ratings as a co-
variate; P sex and prime were entered as between-participants
factors and type of image was entered as a within-participants
factor. As expected, type of image did not interact with P sex or
prime (all ps ⬎ .35), meaning that creativity for both types of
images was similarly affected by the prime and by P sex. Thus, the
creativity indices for the two types of stories were combined in the
remainder of the analysis.
The mean rating of creativity for the premanipulation (covari-
ate) stories was 3.83 (SD ⫽ 1.41). Men tended to be more creative
than women (M ⫽ 4.30, SD ⫽ 1.65 for men vs. M ⫽ 3.54, SD ⫽
1.15 for women; p ⫽ .011). To test the specific hypotheses of the
study, we performed two planned comparisons— one for men and
one for women— using the premanipulation creativity scores as a
covariate, which took into account the differences at baseline. For
men, the mating prime significantly elevated creativity compared
with men in the control, t(86) ⫽ 2.81, p ⫽ .006, ␩2 ⫽ .088 (see
Figure 1). However, for women the same romantic prime had no
Figure 1. Men’s and women’s story creativity depending on type of
prime in Study 1 with standard error bars (adjusted means).
effect on their creativity compared with the women in the control,
t(86) ⫽ .17, p ⫽ .88. Thus, results supported the male-only
creative display hypothesis, indicating that mating cues increased
creativity for men but not for women.
Expended Effort
The amount of effort Ps expended on each story was assessed by
the amount of time they took to write the stories and the number
of words used. We first assessed whether the prime manipulation
had a similar impact on the expended effort measures for the two
types of postmanipulation images. One ANCOVA was performed
on time taken to write the story using the premanipulation time as
a covariate, and another ANCOVA was performed on number of
words in a story using the premanipulation number of words as a
covariate. Both analyses indicated that type of image did not
interact with P sex or with prime (all ps ⬎ .3), meaning that the
primes had a similar effect on both men’s and women’s effort on
stories about the cartoon and the painting. Thus, the effort mea-
sures for the two types of stories were collapsed.
A two-factor ANCOVA on time spent writing the stories with
the premanipulation time as a covariate revealed no significant
interaction or main effects of P sex and prime (all ps ⬎ .35). A
two-factor ANCOVA for the number of words used in the stories
with the premanipulation words as a covariate also showed no
significant main effects of P sex and prime (all ps ⬎ .13) or
interaction ( p ⬎ .70). Overall, despite a difference for men in the
creative quality of the stories in the mating prime condition, there
was no indication that this creative jump was related to exerting
more effort—spending more time or words— on the stories.
2
Of the eight measures, humor had the lowest correlation with the other
seven measures. This was primarily because very few Ps (less than 20%)
received a rating above 1 on humor. However, when Ps did display humor
it was generally considered very creative, and humor still had a correlation
above .50 with each of the other seven measures.
 ROMANTIC MOTIVES AND CREATIVITY
Discussion
The results of Study 1 supported the male-only creative display
hypothesis, whereby a mating prime stimulated only men to write
stories that were judged as more creative. Notably, this creative
boost occurred despite there being no actual incentive for men to
be more creative; that is, the stimuli were photographs and the
romantic situations were imaginary. This creative increase was
also not accompanied by a parallel increase in either the time spent
or the number of words used to write the stories. Thus, it does not
seem that a mating prime generated greater creativity because it
induced participants to exert more effort into writing the story.
Instead, mating cues appear to inspire an increase in the quality of
men’s displays.
Study 2
Although the results of Study 1 were supportive of the male-
only creative display hypothesis, the findings were limited. For
instance, research indicates that men are more sexually aroused by
visual images than women (Hamann, Herman, Nolan, & Wallen,
2004). Because the key part of the mating prime procedure con-
sisted of viewing photographs of attractive individuals, it is pos-
sible that this procedure may have been stronger at inducing a
romantic mindset for males than for females.
However, if the priming procedure was successful for both men
and women, the lack of a creative spike for females is both slightly
puzzling and highly instructive. It is puzzling because prior re-
search indicates that both men and women indicate a relatively
equal preference for creativity in a mate (Li et al., 2002). More-
over, because women’s creative displays are likely to be attractive
to potential mates, it seems odd that courtship would not spur
creativity. However, the lack of a creative increase for women in
Study 1 is also instructive: Perhaps it reflects a tendency for
women to not display their creativity when encountering a partic-
ular type of potential mate but not when encountering other types
of potential mates.
Short-Term and Long-Term Mating Strategies
A closer inspection of the research on human mating indicates
abundant evidence that women and men behave differently in their
pursuit of long-term and short-term mates (Buss & Schmitt, 1993;
Gangestad & Simpson, 2000; Kenrick, Sadalla, Groth, & Trost,
1990). Whether a relationship is likely to be short-term (e.g., a
one-night stand) or long-term (e.g., a marriage) has vastly different
implications for the expected parental investment of men and
women. If offspring result from a short-term mateship, the female
is likely to contribute the majority of parental investment, whereas
the male contributes practically nothing. Within a long-term rela-
tionship, however, both parents expect to contribute significantly
to offspring.
A short-term relationship is more akin to courtship in the vast
majority of other mammalian species, in which fathers contribute
little or nothing to the offspring. Under such circumstances, fe-
males tend to be highly selective about their mating partners,
choosing only those mates who manifest characteristics associated
with good genes (Gangestad & Simpson, 2000). Consequently, in
most mammalian species and in many other vertebrate species
such as the peacock, it is the male who does the displaying, and the
67
female who does the evaluating. When people are pursuing a
short-term strategy, therefore, it would be expected that only the
males would increase their displays of creativity. It seems likely
that just this kind of a romantic relationship was made salient in
Study 1, in which participants saw photos of multiple attractive
opposite sex strangers and had no information indicating whether
these strangers were potential long-term mates. Moreover, attrac-
tive males may be more likely to be perceived to be inclined
toward unrestricted mating strategies (Gangestad, Haselton, &
Buss, in press).
Unlike most other mammals, however, human mating often
involves substantial investment by fathers— up to and including a
lifetime monogamous commitment of the male’s effort and re-
sources. High male parental investment, though rare in mammals,
tends to coevolve with slow-developing high-cost offspring that
are helpless at birth (Geary, 2000). In species with high male
parental investment, fathers and mothers invest a great deal in a
long-term mate, and both sexes tend to be choosy about the
characteristics of a long-term partner. It would follow that when
people are pursuing a long-term mate, both men and women would
be under selective pressure to display desirable sexually selected
characteristics, including the ability to be creative.
Study 2 tested whether priming participants with cues for an
explicitly short-term versus a long-term mating situation would
produce a different pattern in creative displays for males versus
females. Given cross-species findings and differences in expected
parental investment for the two types of mating strategies, two
predictions were generated. For men, it was predicted that both
short-term and long-term mating primes would elicit creativity.
However, for women it was predicted that only the long-term
prime would produce a creative boost. To prime the two different
mating situations, all participants read a short scenario in which
they imagined themselves desiring a short-term or a long-term
mate. To avoid the potentially problematic sex difference in
arousal to visual cues, we did not use photographs.
Method
Participants
Two hundred participants (64 men and 136 women) were recruited from
introductory psychology classes as partial fulfillment of their class
requirement.
Design and Procedure
The study design was a between-participants 2 (P sex) ⫻ 3 (Prime:
short-term mating vs. long-term mating vs. control) design. The general
procedure of the study was very similar to that of Study 1, including using
the identical method to measure creativity. However, the method used to
prime mating was different.
Mating primes. All Ps read and imagined themselves in one of three
scenarios (short-term mating, long-term mating, and control) that were of
similar length (about 850 words). In the short-term scenario, Ps imagined
themselves during the last day of their vacation on an exotic island. On this
last day they met a desirable person and spent a romantic afternoon and had
dinner with the new romantic interest. The scenario ends as the two lovers
are passionately kissing on a moonlit beach. The short-term scenario
repeatedly emphasized that the two people would likely never see each
other again.
In the long-term scenario, Ps imagined meeting someone desirable on
the university campus. Ps imagined spending a wonderful afternoon and a
68 GRISKEVICIUS, CIALDINI, AND KENRICK

romantic evening with this person, including a candlelight dinner and a

sweet kiss goodnight. Throughout the scenario, the reader ponders that this
person may be a good long-term partner, and the scenario ends as the
reader is anticipating going out on an “official” first date with this person.
In the control scenario, Ps imagined getting ready to go to a much-
anticipated concert with a same-sex friend. During the night of the show,
Ps imagined that they could not find the tickets. However, the scenario has
a happy ending as the friend shows up with the tickets, and they both head
off in a great mood anticipating a delightful musical experience.
To test whether the scenarios were successful at eliciting romantic
feelings, self-report measures of romantic and sexual affect were collected
from a total of 58 male and female Ps. After reading one of the three
scenarios, Ps indicated to what extent they were experiencing sexual and
romantic arousal on a 7-point scale with the endpoints at 1 (not at all) and
7 (very much). Results indicated that there were no differences in the
measures across the two romantic scenarios for men or women (all ps ⬎
.35). However, compared with Ps in the control scenario, Ps who read the
mating scenarios reported significantly more romantic arousal (mating
M ⫽ 5.88, SD ⫽ .61; control M ⫽ 1.85, SD ⫽ .48; p ⬍ .01) and
significantly more sexual arousal (mating M ⫽ 4.93, SD ⫽ .78; control
M ⫽ 1.18, SD ⫽ .29; p ⬍ .01).
Priming booster shots. As in Study 1, the current experiment had a
booster of the prime manipulation after Ps wrote the first postprime
creative story. In the booster for both romantic prime conditions, Ps were
Figure 2. Men’s and women’s story creativity depending on type of
asked to imagine themselves back in the scenario that they read earlier.
prime in Study 2 with standard error bars (adjusted means).
Then, Ps were given up to 3 min to write in detail about what physical
characteristics they would desire in this person. In the control condition, Ps
were also asked to go back to the scenario they read earlier. However, they
wrote about the physical characteristics of the anticipated concert venue.
term condition did indeed write significantly more creative stories
Dependent measures. As in Study 1, four student judges (two men and
two women) who were blind to experimental condition rated the stories on
than men in the control condition, t(193) ⫽ 3.77, p ⫽ .003, ␩2 ⫽
the same creativity dimensions. The amount of time Ps spent writing each .047. Although men in the short-term mating condition also dis-
story and the number of words used were also recorded. played more creativity than men in the control condition, this
difference was only marginally significant, t(193) ⫽ 1.69, p ⫽
.096, ␩2 ⫽ .016.4 Nevertheless, men in the long-term and the
Results
short-term conditions did not differ from one another, t(193) ⫽
Creativity 1.41, p ⫽ .16, and the combination of the two mating conditions
was significantly different from the male control condition,
The eight creativity measures again showed high cohesiveness t(193) ⫽ 2.71, p ⫽ .009, ␩2 ⫽ .037. Thus, both a short-term and
for each of the four judges (alphas for the eight ratings for each a long-term mating prime boosted men’s creative displays to some
judge ranged from .90 –.94), so the eight measures were combined extent.
into a creativity index for each judge. The ratings of all four judges As expected, women in the short-term mating condition showed
also showed high interrater reliability (␣ ⫽ .87 for the cartoon; no significant difference in creativity from women in the control,
␣ ⫽ .88 for the painting). Thus, the ratings for the four judges were t(193) ⫽ .34, p ⫽ .73. This predicted lack of difference concep-
combined into a cartoon creativity index and a painting creativity tually replicated the findings for women from Study 1. However,
index. a planned comparison of the long-term mating prime and the
To test whether the prime had a similar effect on men’s and control conditions did not indicate the predicted increase in cre-
women’s creativity on each type of postmanipulation image, we ativity for women, t(193) ⫽ .98, p ⫽ .32. Thus, neither the
used the same analysis strategy from Study 1. A repeated-measures
mixed-model 3-factor ANCOVA with type of image as a within-
participants factor and the premanipulation ratings of creativity as 3
The lack of interactions also indicates that imagining the physical
a covariate again indicated that type of image did not interact with characteristics of the potential mate in the booster shot did not have a
the other two factors (all ps ⬎ .60). Thus, the creativity indices for different impact on creativity than just reading the original scenario.
4
the two types of stories were combined into one measure.3 The slightly lower creativity for men in the short-term versus the
The mean rating of creativity for the premanipulation (covari- long-term condition can be primarily attributed to 2 Ps, who reported
ate) stories was 4.18 (SD ⫽ 1.44), and males again tended to be currently being in a committed romantic relationship and showed a sizable
more creative than females at baseline (M ⫽ 4.52, SD ⫽ 1.48 for creative drop after imagining pursuing a short-term fling. However, the
identical short-term scenario did produce significant increases in male
males vs. M ⫽ 4.02, SD ⫽ 1.40 for females; p ⫽ .021). To test the
creativity in Studies 3 and 4, and Ps’ current relationship status did not
specific hypotheses of the study, we performed a series of planned significantly affect the results in any of the studies. In addition, analyses of
contrasts using the premanipulation stories as covariate, which Ps’ sociosexual orientation (Simpson & Gangestad, 1991, 1992) and cre-
took into account the differences at baseline. ativity in the different conditions also did not indicate any significant
It was predicted that both of the romantic scenarios would effects, perhaps because the scenarios primed specific mating strategies
increase men’s creativity. As seen in Figure 2, men in the long- irrespective of a person’s sociosexual inclination.
 ROMANTIC MOTIVES AND CREATIVITY
short-term nor the long-term mating primes boosted women’s
creativity.5
Expended Effort
Two three-factor ANCOVAs (one for time and one for number
of words) again indicated that type of image (cartoon or painting)
did not interact with either P sex or prime (all ps ⬎ .60). Thus, the
effort measures for the two types of stories were collapsed. An
ANCOVA for time spent writing the stories again indicated no
significant interaction or main effects of P sex or prime (all ps ⬎
.40). An ANCOVA for the number of words used in the stories
also indicated no significant interaction or main effects (all ps ⬎
.30).
Discussion
The results of Study 2 provided further qualified evidence that
cues related to mating can lead to an increase in displayed cre-
ativity, at least in males. As predicted, men showed an increase in
creativity when primed with thoughts of pursuing either a short-
term or a long-term mate. Also as predicted, women did not show
an increase in creativity in the short-term prime condition. How-
ever, contrary to prediction, women failed to show an increase in
creativity in the long-term condition. The overall pattern for men
and women was consistent with the results of Study 1. Also
following the results of Study 1, none of the increases in creativity
produced by the mating primes was accompanied by any indica-
tion of increased effort to produce the display.
Study 3
Although the results of Study 2 supported a mating-inspired
increase in creativity for men, the supposed long-term mating
prime did not lead to the predicted increase in women’s creative
displays. Why not? First, it is possible that as in many species,
women may simply not display this sexually selected characteristic
in courtship. Instead, women may have evolved creative abilities
primarily to judge the quality of men’s creative displays (Miller,
2000). However, it is possible that our long-term mating prime
manipulation, which involved preparing for a first date with a man
who had long-term potential, may have been insufficient to reach
the threshold for women’s mating-linked displays of creativity.
Because women incur significantly higher reproductive costs if
they are abandoned by their mates, women selecting a long-term
mate should be especially focused on and sensitive to the trust-
worthiness and commitment levels of the man (Haselton & Buss,
2000; Hrdy, 1999; Hurtado, Hill, Kaplan, & Hurtado, 1992).
Following this logic, women would need to be assured of the
trustworthiness and the commitment level of a potential mate
before committing themselves strongly. This reasoning is consis-
tent with findings that women are, compared with men, slower to
report feelings of love during courtship and have a higher desire
for commitment from their romantic partner before consenting to
sex (Peplau, 2003). Women are also, compared with men, rela-
tively more conservative about believing professions of love by
members of the opposite sex (Haselton & Buss, 2000). Although
the long-term scenario used in Study 2 specified that the imagined
mate was a potential long-term partner, the overall perceived mate
quality of the person— especially as it relates to trustworthiness
69
and commitment—may have been too ambiguous to ensure that
women would perceive this person to be a truly high-quality
mating prospect. Study 3 was undertaken to examine this
possibility.
Study 3 was conceptually similar to Study 2, except that a new
key condition was added in which men and women read about a
long-term mate who was a clearly committed long-term partner.
The results of Study 2, along with male and female differences in
parental investment, led to two separate predictions— one for men
and one for women. For men, it was predicted that creativity would
increase after all of the three mating primes. However, the pattern
for women was predicted to be distinctly different: The committed
long-term mating prime should be the only one to produce a higher
level of creativity when compared with the other conditions, which
should again not differ from one another.
In an attempt to explore whether specific mating cues can
produce creative boosts in other domains of creativity besides
story-writing, Study 3 also used a different measure of creativity—
the Remote Associates Test (RAT; Mednick, 1962). The RAT was
originally developed as an objective test of creativity, whereby
creativity is defined in a much narrower scope: Creativity is the
ability to make rapid appropriate associations between various
concepts. Each RAT question consists of providing people three
words (e.g., “dress, dial, flower”) and giving them a limited
amount of time (15 seconds in the current study) to come up with
the one correct word linked to all three of the original words
(“sun”). Success on the RAT has been shown to correlate reliably
with success on classic insight problems, and the RAT is often
used in the study of creative problem solving (Bowden & Beeman,
1998; Schooler & Melcher, 1995).
Finally, to assure that the results of Studies 1 and 2 were not
caused by some peculiar effect produced by the content of the
control conditions, Study 3 used a no-prime control condition.
That is, participants in the control did not read a scenario.
Method
Participants
One hundred and fifty-seven participants (85 men and 72 women) were
recruited from introductory psychology classes to participate as partial
fulfillment of their class requirement.
Design and Procedure
The design of the study was a between-participants 2 (P sex) ⫻ 4
(Prime: short-term vs. potential long-term vs. committed long-term vs.
control) design. The conceptual procedure was similar to that of Study 2,
except for the addition of a new romantic scenario for the committed
long-term condition and the use of the RAT to assess creativity.
RAT. The RAT can be administered using any number of questions
while giving Ps any amount of time to answer them. In the current study,
Ps saw 40 RAT-like questions and had 15 s to answer each question. The
5
In addition to judging creativity, the stories were also judged on the
perceived intelligence of the participant. The measures of intelligence
showed high concordance with the measures of creativity, whereby the
mating primes led males to be perceived as more intelligent. However, it
may have been difficult to meaningfully separate creativity and intelligence
within the judging context, making it uncertain whether mating primes
indeed boosted a purer form of intelligence display.
70 GRISKEVICIUS, CIALDINI, AND KENRICK
first 20 questions were used to establish a baseline score for each P
regarding their general RAT performance. Then, after the priming manip-
ulation, Ps worked on a different set of 20 RAT questions; performance on
this last set of 20 constituted the dependent measure in the study. All of the
specific RAT questions were adapted from Bowden and Beeman (2003),
who provide normative statistics regarding the percentage of people who
tend to solve specific RAT questions within various time limits. The 40
questions used in the current study were shown to have been solved
40%– 60% of the time within a 15-s time limit by university students. Ps
in all conditions responded to the same first set of 20 questions (baseline)
and to a different set of 20 subsequent questions (dependent measure). The
two sets were matched to be of equal difficulty, and the order of the
questions within each set of 20 was randomized.
Mating primes. The same procedure from Study 2 was used to prime
mating. Two of the romantic scenarios (short-term and potential long-term)
were identical to those used in Study 2, except that the original long-term
scenario has now been labeled potential long-term. To prime cues of a
committed high-quality long-term mate, a brief paragraph was added to the
original long-term scenario. The paragraph contained three new pieces of
information: (a) the couple had already been dating for a while, which was
intended to signify commitment; (b) the P had met and approved of the
target’s friends, which was intended to signify trustworthiness; and (c) the
target had met the P’s friends who had given their approval, which was
intended to signify that the target was good relationship material. Unlike in
Studies 1 or 2, Ps in the control condition did not read a story or see any
photos. Instead, they were given a short break and saw a blank screen
before continuing to the next set of RAT questions.
Priming booster shot. As in Studies 1 and 2, all Ps received a booster
shot to refresh the mating prime after completing the first 10 of the second
set of the dependent-measure RAT questions. The specific nature of the
booster shot was identical to that in Study 2, in which people were asked
to write about the physical characteristics of the imagined person from the
romantic scenario. In the control condition, participants were merely given
another short break and saw a blank screen before continuing with the
remaining 10 RAT questions.
Figure 3. Men’s and women’s performance on the Remote Associates Test (RAT) depending on type of prime
in Study 3 with standard error bars (adjusted means).
Results
Creativity
For the 20 baseline RAT questions, participants solved 8.72
(SD ⫽ 3.44) questions, and there were no sex differences in
performance. In addition to testing the predictions of the study via
planned contrasts, we also tested whether the predicted contrasts
explained most of the between-cell variance by testing the residual
contrast, which was predicted to be nonsignificant (see Levin &
Neumann, 1999).
Male creativity. For men, it was predicted that the three mat-
ing primes would produce a higher level of creativity than in the
control condition. As seen in Figure 3, a planned contrast using the
baseline RAT measures as covariates indicated that this was in-
deed the case, F(1, 80) ⫽ 2.58, p ⫽ .012, ␩2 ⫽ .077. Creativity in
the three mating conditions did not differ from one another ( p ⫽
.99). The test of the residual contrast was also not significant,
Fresidual (2, 80) ⫽ 0.55, ns, meaning that the predicted contrast
accounted for most of the between-cell variance.
Female creativity. For women, it was predicted that only the
prime for the committed long-term mate would produce a creative
boost above that of the other three conditions. As seen in Figure 3,
a planned contrast with baseline RAT scores as the covariate
indicated that this was indeed the case, F(1, 72) ⫽ 2.03, p ⫽ .048,
␩2 ⫽ .056. As would be predicted from Studies 1 and 2, the
remaining three conditions—short-term, potential long-term, and
control— did not differ from one another ( p ⫽ .70). The test of the
residual contrast was also not significant, Fresidual (2, 72) ⫽ 2.96,
ns, meaning that the predicted contrast explained most of the
between-cell variance in creativity. Thus, women showed a cre-
ative increase only when they were primed with thoughts of a
committed long-term mate who was clearly high-quality relation-
ship material.
 ROMANTIC MOTIVES AND CREATIVITY
Differences in Imagined Mate Across Scenarios
To examine whether participants indeed perceived the key dif-
ferences across the romantic scenarios, we collected additional
data to ascertain what Ps actually perceived to be the differences
and similarities of the imagined mate in each of the three scenarios.
A total of 92 separate male and female Ps underwent the same
priming procedure used in the current study, whereby each person
only read one of the three scenarios. Afterward, Ps indicated the
quality of various characteristics regarding the person whom they
imagined desiring in the scenario. More specifically, Ps were
asked 13 questions in random order that were generally phrased in
the following way: “To what extent is this person __________ ,”
and Ps provided their responses on a scale of 1 (not at all) to 9
(very).
The questions asked about the imagined mate’s perceived trust-
worthiness (trustworthy, truthful, honest; ␣ ⫽ .93), level of com-
mitment (committed, faithful, likely to cheat [reverse scored]; ␣ ⫽
.80), and desirability as a long-term partner (good relationship
material, the right person for me, confidence that they’re a good
boyfriend or girlfriend; ␣ ⫽ .91). In addition, Ps were also asked
to report the extent to which the potential mate was seen as being
creative, intelligent, funny, and physically attractive. To test
whether men and women differed in their perceptions of the mate
on the characteristics across the scenarios, 13 two-factor analyses
of variance (P sex ⫻ Prime) were performed. Results indicated no
significant interactions (all ps ⬎ .30). Thus, male and female
ratings were combined for the remainder of the analysis.
As seen in Table 1, Ps did indeed perceive different levels of
trustworthiness, commitment, and long-term mate value across the
three conditions: The short-term mate was always perceived as
having the least of the three qualities, and the committed long-term
mate was always perceived as having the most of the three qual-
ities. When compared with the potential long-term scenario, the
person imagined in the committed long-term scenario was per-
ceived as significantly more trustworthy, t(89) ⫽ 2.48, p ⫽ .015,
␩2 ⫽ .064, more committed, t(89) ⫽ 2.10, p ⫽ .038, ␩2 ⫽ .047,
and of higher long-term mate quality, t(89) ⫽ 2.08, p ⫽ .040, ␩2 ⫽
.047. Thus, men and women did perceive the key differences
between the two long-term primes. Analyses of Ps’ perceptions of
physical attractiveness, creativity, intelligence, and humor indi-
cated no significant differences among the three conditions (all
ps ⬎ .30), with all of the ratings being relatively high (see Table
1 for ratings of attractiveness and creativity as an example).
Table 1
Perceived Attributes in Imagined Mate Depending on Scenario
Type of romantic scenario
Potential Committed
Perceived attribute Short term long term long term
Trustworthy 6.00 (1.58) 6.69 (1.18) 7.53 (1.25)
Committed 5.32 (1.48) 6.43 (1.16) 7.15 (1.55)
Relationship material 5.83 (1.82) 7.01 (1.46) 7.77 (1.14)
Intelligent 8.84 (0.76) 8.00 (1.02) 8.16 (1.32)
Creative 6.85 (1.46) 6.75 (1.27) 6.97 (1.14)
Note. Means are on a 1–9 scale in which higher numbers indicate more
of the perceived characteristic. Numbers in parentheses denote standard
deviations.
71
Mood and Arousal
One possible mechanism for how the romantic primes may have
produced the specific patterns of creativity was by influencing
men’s and women’s mood or arousal. To investigate this possibil-
ity, a separate group of a total of 63 male and female Ps underwent
the priming procedure from Study 3. Afterward, Ps rated to what
extent they felt positive arousal (energetic, excited, passionate; ␣⫽
.75), negative arousal (upset, tense, nervous; ␣ ⫽ .83), and positive
mood (happy, upbeat, joyful; ␣ ⫽ .83). Ratings were provided on
a 1 (not at all) to 7 (very much) scale.
Results indicated that none of the romantic primes produced
much negative arousal for men or women (Ms between 1.98 and
2.40). However, the patterns for positive mood and positive
arousal were instructively different across the different conditions
(see Table 2). For men, the two long-term primes produced similar
levels of mood and arousal, and ratings for both primes were
slightly (nonsignificantly) higher than those for the short-term
prime. For women, mood and arousal for the two long-term mating
primes also did not differ from each other, and both were signif-
icantly higher than the mood and arousal produced by the short-
term prime ( p ⬍ .05). This overall pattern for women is quite
different from the pattern of women’s performance on the RAT.
Specifically, although the two long-term conditions produced
highly similar patterns of positive mood and positive arousal,
women’s RAT performance in these two conditions differed.
Moreover, despite men’s and women’s different performance on
the RAT in the short-term mating condition, the short-term sce-
nario produced no sex differences in positive arousal or mood (all
ps ⬎ .60). Taken as a whole, the RAT performance patterns
produced by the romantic scenarios appear to be different from the
mood and arousal patterns produced by these primes.
Discussion
Results from Study 3 provided a clearer picture of how mating
cues influence creativity. Study 3 replicated both the short-term
and the (potential) long-term findings from Study 2. It is important
to note that these results were replicated with a different measure
of creativity—the RAT—and a no-prime control condition. As
predicted, Study 3 also showed that when participants were primed
with the desire for a committed long-term mate—a mate who was
perceived to be more trustworthy, committed, and generally better
relationship material— both men and women showed higher levels
of creativity. This finding supports the notion that when males or
females are pursuing a mating strategy in which they each expect
to contribute significant parental investment, both sexes are likely
to display a sexually selected trait such as creativity. However,
men and women appear to have different mate-quality thresholds
for displaying creativity: For men, the requirements are relatively
low—the desire to attract any desirable woman will do; for
women, the requirements are high— only a desire for a long-term
mate who is clearly a high-quality mate produces the display.
Additional data also indicated that the precise pattern of creative
boosts for men and women could not be explained solely by the
arousal or positive mood that were produced by each of the
romantic scenarios.
Study 4
Thus far, findings from all three studies show that specific cues
related to mating can produce an increase in creativity for men and
72 GRISKEVICIUS, CIALDINI, AND KENRICK

Table 2 RAT problems. This question served as a manipulation check of

Men’s and Women’s Mood and Arousal Depending on Scenario the incentive instructions. Analyses indicated that motivation for
men and women did not interact with condition ( p ⬎ .70), and that
Type of romantic scenario men’s and women’s motivation did not differ from each other
Potential Committed
across the three conditions (for men, M ⫽ 4.79, SE ⫽ .16; for
Short term long term long term women, M ⫽ 4.80, SE ⫽ .15), F(2, 161) ⫽ .085, p ⫽ .92. Thus,
the motivation scores for men and women were combined. Results
Positive mood also indicated that Ps in the mating and control conditions also did
Males 4.42 (1.65) 5.00 (1.27) 5.07 (1.16)
not differ from each other in motivation (for mating, M ⫽ 4.59,
Females 4.27 (1.65) 5.70 (0.85) 5.73 (0.77)
Positive arousal SE ⫽ .18; for control, M ⫽ 4.71, SE ⫽ .19; p ⫽ .63). Thus, these
Males 4.18 (1.94) 4.54 (1.61) 4.57 (1.41) two groups were combined into a no-incentive condition.
Females 3.67 (1.90) 5.03 (0.96) 5.20 (0.92) A comparison of the monetary incentive condition and the
combined no-incentive group revealed that participants in the
Note. Means are on a 1–7 scale in which higher numbers indicate a more
intense state. Numbers in parentheses denote standard deviations. monetary incentive condition reported being significantly more
motivated to do well on the RAT items (for monetary incentive,
M ⫽ 5.09, SE ⫽ .19; for no incentive, M ⫽ 4.65, SE ⫽ .13), F(1,
women. However, would mating cues continue to produce a cre- 165) ⫽ 3.90, p ⫽ .051, ␩2 ⫽ .023. Thus, the manipulation to
ative boost when compared with people who are motivated to do induce motivation to do well on the task had the desired effect.
well on the creative task? To test this question, the current study
used a methodology almost identical to that of some conditions of
Study 3, except that it added a key new condition in which
Creativity
participants were provided with a monetary incentive to do well on For the 20 baseline RAT questions, participants solved 8.01
the RAT but in which no romantic prime was used—the monetary (SD ⫽ 3.12) questions and there were no sex differences in
incentive condition. The study had two competing hypotheses with performance. To test the specific hypotheses of the study, we
different implications for the process by which mating cues may performed a series of planned comparisons using the baseline RAT
stimulate creativity. First, it was plausible that Ps in the monetary scores as a covariate. First, examining the men, a planned contrast
incentive and the men in the (short-term) mating condition would revealed the predicted increase in RAT performance in the mating
perform significantly better on the RAT than the Ps in the control condition when compared with the control condition, t(160) ⫽
condition. Such a result would indicate that mating cues might 2.83, p ⫽ .005, ␩2 ⫽ .048. However, as seen in Figure 4, men in
facilitate creativity by motivating people to somehow work harder the mating condition also performed significantly better than men
at the task. However, a different outcome was also deemed plau- in the monetary incentive condition, t(160) ⫽ 2.21, p ⫽ .029, ␩2 ⫽
sible. The second possibility was that men in the mating condition .030. Although men in the monetary incentive condition performed
would perform significantly better than men in any of the other slightly better than men in the control condition, this difference
conditions. Such a result would indicate that mating cues do not
increase creativity simply by leading people to try harder at the
task.

Method
Participants
One hundred and sixty-seven participants (78 men and 89 women) were
recruited from introductory psychology classes to participate as partial
fulfillment of their class requirement.

Design and Procedure

The design of the study was a between-participants 2 (P sex) ⫻ 3
(Condition: mating vs. control vs. monetary incentive) design. The proce-
dures in the (short-term) mating and the control prime conditions were
identical to those of Study 3. In the monetary incentive condition, the
procedure was similar to that of the control, except that before Ps started
the last 20 RAT problems, they received the following instructions: “On
the remaining problems, we would like you to try your hardest to do the
best you can. If you are in the top 30% of everyone’s scores, you will be
entered in a raffle to win $60!”

Results
Motivation Check
Figure 4. Men’s and women’s performance on the Remote Associates
After Ps finished the RAT questions, they were asked to what Test (RAT) depending on condition in Study 4 with standard error bars
extent they were motivated to do well on the 20 postmanipulation (adjusted means).
 ROMANTIC MOTIVES AND CREATIVITY
was not significant ( p ⫽ .31). For women, there were no signifi-
cant differences across conditions, F(2, 85) ⫽ 2.24, p ⫽ .44.
Discussion
The results of Study 4 indicated that men primed with mating
cues performed better on the creativity task even when compared
with people who were trying harder at the task. Previous research
(Amabile, 1996) has indicated that extrinsic motivation, such as a
monetary incentive, does not always result in higher creative
quality, so perhaps the finding that an incentive to do well on the
RAT did not significantly improve performance is not surprising.
However, the fact that a mating prime led to a creative boost above
that produced by merely trying hard at the task suggests that the
process by which mating cues stimulate creativity is not related to
external motivation. Consistent with the findings from Studies 1
and 2, the mating-inspired creative boost appears to be unrelated to
expending more effort on the creative task.
General Discussion
The current research set out to explore whether priming men and
women with a variety of theoretically relevant mating cues would
lead to an increase in creativity. This question was tested in four
studies, in which men and women first either looked at photo-
graphs of attractive opposite sex individuals (Study 1) or imagined
being in a particular romantic scenario (Studies 2– 4), and then
performed tasks assessing creativity on subjective (Studies 1 & 2)
or objective (Studies 3 & 4) measures. For men, cues designed to
stimulate a motive to attract either a desirable short-term or a
long-term mate produced an increase in creativity. That is, after
just thinking of attracting a desirable woman as any kind of a
romantic partner, men showed an increase in creativity— even if
they themselves could not actually benefit romantically from this
creative burst in the current setting. In contrast, women only
increased their creative output after imagining wanting to attract a
clearly high-quality (i.e., trustworthy and committed) long-term
mate. Women did not show a creative increase when primed to
think about attracting a short-term mate or a potential long-term
mate who had yet to prove his worth as good relationship material.
These findings are consistent with the theories of sexual selec-
tion and differential parental investment. When pursuing a short-
term mating strategy, women are expected to provide the vast
majority of parental investment. In other species, selectively high
female investment is also associated with courtship displays pri-
marily by males. For example, male bowerbirds, who do not
contribute direct resources to offspring, have evolved to construct
elaborate bowers during courtship; they then artfully decorate the
bowers with a colorful assortment of flower petals, berries, and
snail shells, or even “paint” it with regurgitated bluish residue
(Borgia, 1986). Females inspect the bowers and preferentially
mate with males who have the largest, most symmetric, and best
decorated displays because, like the peacock’s tail, the creative
bower displays seem to serve as a reliable indicator of fitness
(Borgia, 1995).
When pursuing a long-term mate, however, both men and
women expect to invest significantly in offspring. In other species,
high male parental investment is associated with both sexes being
choosier when selecting a mate and with both sexes tending to
display desirable qualities in courtship. Such a pattern can be seen
73
in gibbons, in which each sex invests substantially in offspring.
During courtship, both male and female gibbons have evolved to
sing complex and elaborate musical patterns to each other—a
behavior called dueting (Geissmann, 2000; Raemaekers & Rae-
maekers, 1984). Despite the fact that human males often invest
substantially in shared offspring, women always stand a reasonable
danger of substantially higher costs if they choose a long-term
mate who does not intend to commit (Kenrick et al., 1990). In this
light, it makes some sense that women require assurance that a
prospective mate is really going to invest in offspring before
investing the energy in creative displays.
The Psychological Mechanism
The current research examined several plausible psychological
processes by which mating cues could stimulate creativity. Results
from three of the studies indicated that mating cues are unlikely to
elicit creativity by stimulating one to try harder to be creative. In
Studies 1 and 2, participants neither spent more time nor used more
words to write creative stories. In addition, participants given a
monetary incentive in Study 4 still performed worse than people
primed with mating cues. Although it is possible that a chance to win
$60 may not have been a strong incentive, participants in this condi-
tion nevertheless reported being more motivated to try harder.
Study 3 indicated that the mood and arousal states produced by
each of the specific romantic scenarios showed a distinctly differ-
ent pattern from that of creativity. For instance, although women
were equally positively aroused by the potential and the committed
long-term mate scenarios, only the committed scenario produced a
creative boost for women. Moreover, men and women reported
relatively equal mood and arousal after reading the short-term
scenario, although only men showed a creative increase in that
situation. Although the mood and arousal ratings were self-
reported, these findings suggest that the particular pattern of in-
creases in creativity cannot be explained solely by changes in
mood and arousal— or at least the kind of general arousal typically
measured in psychological research.
A third process, which we believe is most likely to underlie these
effects, can be inferred from Studies 3 and 4, in which the RAT was
used to assess creativity. Because the purpose of each RAT question
is to ascertain whether a person can rapidly locate a particular concept
in their mind, the RAT is an appropriate and rather specific test of the
accessibility of various concepts. The results of Study 4 indicated that
merely trying harder did not improve performance on the RAT;
however, a mating prime did improve performance, most likely be-
cause it enabled superior accessibility to relevant but remote infor-
mational links. The results of Studies 1 and 2 are also consistent with
the possibility that mating cues enabled better accessibility to various
concepts: By having access to a richer set of creative avenues brought
about by the romantic primes, people could write stories that were
more creative.
Alternative Explanations
The current research has adopted a framework inspired by existing
theory regarding sexual and natural selection accounts of creativity
(Miller, 2000). It would surely be possible to derive predictions
regarding creativity and mating from several other theoretical per-
spectives, though none seem to offer as straightforward an account of
the pattern of results obtained in this series of studies. For example, it
74 GRISKEVICIUS, CIALDINI, AND KENRICK
is possible that any vivid fantasy could prime a creative process and
lead to an increase in creativity. Thus, for women, perhaps only
imagining a high-quality long-term mate may have led to a fantasy
that was vivid. However, because both the potential and committed
long-term mating primes were very similar to each other but only one
of them produced an increase in creativity, this explanation seems
unlikely. Not only did both primes produce highly similar mood and
arousal for women, but both primes are likely to have produced
relatively equally vivid fantasies.
Perhaps an association between creativity and mating arises
because of simple mechanisms of associative priming (Srull &
Wyer, 1979; see Higgins, 1996 for a review). For example, Char-
trand, Fitzsimons, and Fitzsimons (2004) have shown that partic-
ipants primed with thoughts about Apple computers—a brand
marketed as the computer for creative individuals— became more
creative compared with people primed with other computer brands.
Although priming participants with photos of attractive individuals
or romantic scenarios may very well activate concepts of creativ-
ity, it is difficult to see how an associative model framework could
account for the very specific pattern of sex differences and sex
similarities found across the different conditions.
The functional framework used in this research is by no means
an alternative to the associative network model of cognition. Both
models imply that there are certain links between motivation,
cognition, and behavior. However, the functional model does more
than just assert that priming specific ideas will lead to the activa-
tion of associatively linked semantic and affective categories.
Rather, the functional model leads to more finely articulated pre-
dictions about how activating specific functional goals should lead
to specific goal-consistent—and sex-consistent— cognitive and
behavioral responses (Maner et al., 2005).
A social learning model might also suggest that men and women
have been differentially rewarded for producing creativity, al-
though it is again difficult to predict the precise prime-specific
pattern of sex differences and similarities that we found from this
perspective. Further, the final study suggested that tangible re-
wards actually failed to increase creativity. A social role theory
might also posit that it is part of the male role to be creative.
However, without a consideration of sex differences in parental
investment, it would be difficult for that theory to have predicted
a priori why creativity is part of the male role to be creative in all
mating contexts but part of the female role to be creative only in
committed long-term mating contexts.
Neither social role theories nor social learning theories are
mutually exclusive with evolutionary accounts, as evolutionary
theorists presume that social roles across societies are to some
extent a function of evolved adaptations in men and women and
that many behaviors involve an adaptive interplay of learning and
evolved predispositions (Kenrick, Trost, & Sundie, 2004; Öhman
& Mineka, 2001). We are not aware, however, of social role or
social learning theorists who have offered predictions for the
pattern of results obtained here—a pattern which follows directly
from considerations of sexual selection and parental investment.
Limitations and Future Directions
One of the limitations of the current research is that it did not
test whether people become more creative during actual courtship
(as suggested by anecdotal evidence). Although future research
needs to explore this question, the current framework would pre-
dict that the presence of a desirable mate is likely to produce the
strongest creative boost, as long as other forces, such as social
anxiety, are not working against the display. The current studies
also explored only two specific forms of creativity in one culture.
However, the proposed evolutionary framework would predict that
mating-related cues should spur creativity across cultures, al-
though the specific forms of creativity will surely depend on
cultural or local norms (Norenzayan & Heine, 2005; Norenzayan,
Schaller, & Heine, in press). Moreover, there may also be multiple
person variables that could be used to better predict the display.
For instance, mating cues might lead to a boost in people’s most
favored or best-practiced form of creativity: a musician might
become more musically creative, a comedian might be seen as
funnier, a poet might be inspired to display verbal fluidity, and a
dancer may become more expressive through movement. Future
research might also examine why creativity isn’t always “turned
on” in men as it appears to confer significant benefits. One
possibility is that there are costs associated with allocating one’s
energies to permanent creative displays, such as decreased capac-
ity to attend to other matters or decreased functioning of short-term
memory. Another possibility is that there are limitations in most
people’s abilities to be creative, and that creativity is in a sense a
form of “truth in advertising” for a highly functioning brain (see
Miller, 2000).
Although the current findings are consistent with a sexual se-
lection account of creativity, these studies were not designed to
ascertain whether creativity is an exclusively sexually selected
versus naturally selected trait. Indeed, it is often difficult to draw
a line between sexual selection and natural selection, partly be-
cause mating choices are often based on traits that are themselves
naturally selected adaptations (such as symmetrical and healthy
feather displays or the ability to fly well in birds). Thus, it is
plausible that some more rudimentary form of creativity in humans
may have originally evolved because of its association with sur-
vival ability (e.g., enhancing ability to create new tools or verbally
negotiate one’s way out of a conflict). Given that such abilities
were adaptive for whatever reason, the members of one sex would
have been well served to use them as cues to infer mate value in
the other sex. At that point, it would have become useful to display
creativity as a courtship tactic. Thus, creativity would be a function
of both natural and sexual selection. That is, creativity might
provide a survival advantage and function as a heritable fitness
indicator (Haselton & Miller, in press). The fact that our research
indicates that creativity is likely to be displayed by males across
mating contexts, but only selectively by females in contexts asso-
ciated with quality long-term mates, does suggest, however, that
sexual selection plays some role in human creative displays (what-
ever their ultimate origin).
A larger question pertains to how creativity fits in with other
possibly related constructs that may also be in part sexually se-
lected. Although the kinds of creativity measured in the first two
studies (story-writing) may be similar to artistic displays, the
creativity displayed via the RAT in the latter two studies appears
to be more closely related to problem solving rather than artistic
ability. From the perspective adopted here, creativity is likely to be
related to intelligence (Miller, 2000). In this light, future research
might address whether mating motives provide a boost in many
types of creative intelligence or perhaps whether only specific
forms of creative intelligence are associated with courtship dis-
plays. If such displays evolved as fitness indicators, one interesting
 ROMANTIC MOTIVES AND CREATIVITY
possibility is that those individuals who show greater creative
boosts in response to mating primes might also be more likely to
possess characteristics associated with heritable fitness.
Conclusion
This set of studies began with a puzzling question: Why are muses
predominantly women who inspire men? Four experiments indicated
that when men merely thought of pursuing a desirable romantic
partner, either for a long-term relationship or a short-term fling, such
thoughts consistently produced a “muse effect,” whereby they
boosted male creativity. These findings are consistent with the expe-
riences of Picasso and other male artists who were inspired by female
muses. The findings also generally align with the view that creative
displays may be linked to sexual selection, which tends to select for
more intensive male displays, as witnessed in the courtship tail dis-
plays of peacocks. Unlike in peacocks, however, because men often
invest heavily in their offspring, women are also subject to sexual
selection pressures when it comes to long-term mating. Consistent
with parental investment theory, when women were primed with cues
designed to instill a desire to attract a high-quality long-term mate,
these cues also served to produce a muse effect for women by
spurring their creativity.6
6 Gangestad, S. W., Haselton, M. G., & Buss, D. M. (in press). Evolutionary
The renowned Victorian poet Elizabeth Barrett Browning may provide
one such example (Winwar, 1950). Mired in artistic mediocrity in her 20s,
Elizabeth received a letter from an adoring fan, Robert Browning, who
professed his love for her. Not surprisingly, Elizabeth was unconvinced of
his intentions until a year—and many of Robert’s letters—later, at which
point the two lovers agreed to elope in the near-distant future. It was
precisely during this phase of her relationship when Elizabeth was inspired
to pen Sonnets from the Portuguese, which would become her greatest and
most critically acclaimed creative work.
References
Amabile, T. M. (1996). Creativity in context: Update to the social psy-
chology of creativity. Boulder, CO: Westview.
Andersson, M. (1994). Sexual selection. Princeton, NJ: Princeton Univer-
sity Press.
Bargh, J. A. (1990). Auto-motives: Preconscious determinant of social
interaction. In E. T. Higgins & R. M. Sorrentino (Eds.), Handbook of
motivation and social cognition: Foundations of social behavior (Vol. 2,
pp. 93–130). New York: Guilford Press.
Bargh, J. A., & Chartrand, T. L. (1999). The unbearable automaticity of
being. American Psychologist, 54, 464 – 479.
Borgia, G. (1986). Sexual selection in bowerbirds. Scientific American,
254, 92–100.
Borgia, G. (1995). Complex male display and females choice in the spotted
bowerbird: Specialized functions for different bower decorations. Ani-
mal Behavior, 49, 1291–1301.
Bowden, E. M., & Beeman, M. J. (1998). Getting the right idea: Semantic
activation in the right hemisphere may help solve insight problems.
Psychological Science, 9, 435– 440.
Bowden, E. M., & Beeman, M. J. (2003). Normative data for 144 com-
pound remote associate problems. Behavior Research Methods, Instru-
ments, and Computers, 35, 634 – 639.
Bugental, D. B. (2000). Acquisition of the algorithms of social life: A
domain-based approach. Psychological Bulletin, 126, 187–219.
Buss, D. M. (2005). Evolutionary psychology: The new science of the mind
(2nd ed.). Boston: Allyn & Bacon.
75
Buss, D. M., & Barnes, M. (1986). Preferences in human mate selection.
Journal of Personality and Social Psychology, 50, 559 –570.
Buss, D. M., & Schmitt, D. P. (1993). Sexual strategies theory: An
evolutionary perspective on human mating. Psychological Review, 100,
204 –232.
Byrne, R. (1995). The thinking ape: Evolutionary origins of intelligence.
Oxford, United Kingdom: Oxford University Press.
Chartrand, T. L., & Bargh, J. A. (1996). Automatic activation of impres-
sion formation and memorization goals: Nonconscious goal priming
reproduces effects of explicit task instructions. Journal of Personality
and Social Psychology, 71, 464 – 478.
Chartrand, T. L., Fitzsimons, G. M., & Fitzsimons, G. J. (2004, February).
The effects of priming anthropomorphized objects on behavior. Paper
presented at the meeting for the Society for Personality and Social
Psychology, Austin, TX.
Cosmides, L., & Tooby, J. (1992). Cognitive adaptations for social ex-
change. In J. H. Barkow, L. Cosmides, & J. Tooby (Eds.), The adapted
mind (pp. 163–228). Oxford, United Kingdom: Oxford University Press.
Crespelle, J. P. (1969). Picasso and his women. New York: Hodder &
Stoughton.
Cronin, H. (1993). The ant and the peacock: Altruism and sexual selection
from Darwin to today. New York: Cambridge University Press.
Darwin, C. (1981). The descent of man, and sexual selection in relation to
sex (Vols. 1 & 2). Princeton, NJ: Princeton University Press (Original
work published 1871)
Eysenck, H. J. (1995). Genius: The natural history of creativity. Cam-
bridge, United Kingdom: Cambridge University Press.
foundations of cultural variation: Evoked culture and mate preferences.
Psychological Inquiry.
Gangestad, S. W., & Simpson, J. A. (2000). On the evolutionary psychol-
ogy of human mating: Trade-offs and strategic pluralism. Behavioral
and Brain Sciences, 23, 573–587.
Geary, D. C. (2000). Evolution and proximate expression of human pater-
nal investment. Psychological Review, 126, 55–77.
Geissmann, T. (2000). Duet songs of the siamang, Hylobates syndactylus:
I. Structure and organisation. Primate Report, 56, 33– 60.
Gibson, K. R., & Ingold, T. (Eds.). (1993). Tools, language, and cognition
in human evolution. Cambridge, United Kingdom: Cambridge Univer-
sity Press.
Griskevicius, V., Goldstein, N. J., Mortensen, C. R., Cialdini, R. B., &
Kenrick, D. T. (in press). Going along versus going alone: When
fundamental motives facilitate strategic (non)conformity. Journal of
Personality and Social Psychology.
Gutierres, S. E., Kenrick, D. T., & Partch, J. J. (1999). Beauty, dominance,
and the mating game: Contrast effects in self-assessment reflect gender
differences in mate selection. Personality & Social Psychology Bulletin,
25, 1126 –1134.
Hamann, S., Herman, R. A., Nolan, C. L., & Wallen, K. (2004). Men and
women differ in amygdala response to visual sexual stimuli. Nature
Neuroscience, 7, 411– 416.
Haselton, M. G., & Buss, D. M. (2000). Error management theory: A new
perspective on biases in cross-sex mind reading. Journal of Personality
and Social Psychology, 78, 81–91.
Haselton, M. G., & Miller, G. F. (in press). Women’s fertility across the
cycle increases the short-term attractiveness of creative intelligence
compared to wealth. Human Nature.
Higgins, E. T. (1996). Knowledge activation: Accessibility, applicability,
and salience. In E. T. Higgins & A. W. Kruglanski (Eds.), Social
psychology: Handbook of basic principles (pp. 133–168). New York:
Guilford Press.
Hrdy, S. B. (1999). Mother nature: Maternal instincts and how they shape
the human species. New York: Ballantine Books.
Hurtado, A. M., Hill, K., Kaplan, H., & Hurtado, I. (1992). Trade-offs
76 GRISKEVICIUS, CIALDINI, AND KENRICK
between female food acquisition and child care among Hiwi and Ache
foragers. Human Nature, 3, 185–216.
Kanazawa, S. (2000). Scientific discoveries as cultural displays: A further
test of Miller’s courtship model. Evolution and Human Behavior, 21,
317–321.
Kenrick, D. T., Li, N. P., & Butner, J. (2003). Dynamical evolutionary
psychology: Individual decision-rules and emergent social norms. Psy-
chological Review, 110, 3–28.
Kenrick, D. T., Sadalla, E. K., Groth, G., & Trost, M. R. (1990). Evolution,
traits, and the stages of human courtship: Qualifying the parental invest-
ment model. Journal of Personality, 58, 97–117.
Kenrick, D. T., Sadalla, E. K., & Keefe, R. C. (1998). Evolutionary
cognitive psychology: The missing heart of modern cognitive science. In
C. Crawford & D. L. Krebs (Eds.), Handbook of evolutionary psychol-
ogy (pp. 485–514). Hillsdale, NJ: Erlbaum.
Kenrick, D. T., Trost, M. R., & Sundie, J. M. (2004). Sex-roles as
adaptations: An evolutionary perspective on gender differences and
similarities. In A. H. Eagly, A. Beall, & R. Sternberg (Eds.), Psychology
of gender (pp. 65–91). New York: Guilford Press.
Kerr, G., & Gagliardi, C. (2004). Measuring creativity in research and
practice. Unpublished manuscript, Arizona State University, Tempe.
Kingdon, J. (1993). Self-made man and his undoing. New York: Simon &
Schuster.
Levin, J. R., & Neumann, E. (1999). Testing for predicted patterns: When
interest in the whole is greater than the sum of its parts. Psychological
Methods, 4, 44 –57.
Li, N. P., Bailey, J. M., Kenrick, D. T., & Linsenmeier, J. A. (2002). The
necessities and luxuries of mate preferences: Testing the trade-offs.
Journal of Personality and Social Psychology, 82, 947–955.
MacGregor-Hastie, R. (1988). Picasso’s women. London: Lennard
Publishing.
Maner, J., Kenrick, D., Becker, D., Robertson, T., Hofer, B., Neuberg, S.,
et al. (2005). Functional projection: How fundamental social motives
can bias interpersonal perception. Journal of Personality and Social
Psychology, 88, 63–78.
Mednick, S. A. (1962). The associative basis of the creative process.
Psychological Review, 69, 220 –232.
Miller, G. F. (1999). Sexual selection for cultural displays. In R. Dunbar,
C. Knight, & C. Power (Eds.), The evolution of culture (pp. 71–91).
Edinburgh, United Kingdom: Edinburgh University Press.
Miller, G. F. (2000). The mating mind: How sexual choice shaped the
evolution of human nature. New York: Doubleday.
Møller, A. P., & Petrie, M. (2002). Condition dependence, multiple sexual
signals, and immunocompetence in peacocks. Behavioral Ecology, 13,
248 –253.
Norenzayan, A., & Heine, S. J. (2005). Psychological universals across
cultures: What are they and how do we know? Psychological Bulletin,
131, 763–784.
Norenzayan, A., Schaller, M., & Heine, S. J. (in press). Evolution and
culture. In M. Schaller, J. Simpson, & D. Kenrick (Eds.), Evolution and
social psychology. New York: Psychology Press.
Öhman, A., & Mineka, S. (2001). Fears, phobias, and preparedness:
Toward an evolved module of fear and fear learning. Psychological
Review, 108, 483–522.
Peplau, L. A. (2003). Human sexuality: How do men and women differ?
Current Directions in Psychological Science, 12, 37– 40.
Pinker, S. (1997). How the mind works. New York: Norton.
Prose, F. (2002). The lives of the muses: Nine women and the artists they
inspired. New York: Harper Collins Publishers.
Raemaekers, J. J., & Raemaekers, P. M. (1984). Vocal interactions be-
tween two male gibbons, Hylobates lar. Natural History Bulletin of the
Siam Society, 32, 95–106.
Roney, J. R. (2003). Effects of visual exposure to the opposite sex:
Cognitive aspects of mate attraction in human males. Personality and
Social Psychology Bulletin, 3, 393– 404.
Schaller, M. (2003). Ancestral environments and motivated social percep-
tion: Goal-like blasts from the evolutionary past. In S. J. Spencer, S.
Fein., M. P. Zanna, & J. M. Olson (Eds.), Motivated social perception
(pp. 215–231). Mahwah, NJ: Erlbaum.
Schaller, M., Faulkner, J., Park, J. H., Neuberg, S. L., & Kenrick, D. T.
(2004). Impressions of danger influence impressions of people: An
evolutionary perspective on individual and collective cognition. Journal
of Cultural and Evolutionary Psychology, 2, 231–247.
Schooler, J. W., & Melcher, J. (1995). The ineffability of insight. In S. M.
Smith, T. B. Ward, & R. A. Finke (Eds.), The creative cognition
approach (pp. 249 –268). Cambridge, MA: MIT Press.
Scott, J. P. (1980). The function of emotions in behavioral systems: A
systems theory analysis. In R. Plutchik & H. Kellerman (Eds.), Emotion:
Theory, research, and experience (Vol. 1). New York: Academic Press.
Simonton, D. K. (2000). Creativity: Cognitive, personal, developmental,
and social aspects. American Psychologist, 55, 1.
Simpson, J. A., & Gangestad, S. W. (1991). Individual differences in
sociosexuality: Evidence for converging and discriminant validity. Jour-
nal of Personality and Social Psychology, 67, 870 – 883.
Simpson, J. A., & Gangestad, S. W. (1992). Sociosexuality and romantic
partner choice. Journal of Personality, 60, 31–51.
Srull, T. K., & Wyer, R. S. (1979). The role of category accessibility in the
interpretation of information about persons: Some determinants and
implications. Journal of Personality and Social Psychology, 37, 1660 –
1672.
Sternberg, R. J., & Lubart, T. I. (1999). The concept of creativity: Pros-
pects and paradigms. In R. J. Sternberg (Ed.), Handbook of creativity
(pp. 3–15). Cambridge, United Kingdom: Cambridge University Press.
Todd, P. M., & Gigerenzer, G. (2000). Precis of “simple heuristics that
make us smart.” Behavioral & Brain Sciences, 1, 727–780.
Trivers, R. L. (1972). Parental investment and sexual selection. In B.
Campbell (Ed.), Sexual selection and the descent of man: 1871–1971
(pp. 136 –179). Chicago: Aldine.
Wilson, M., & Daly, M. (2004). Do pretty women inspire men to discount
the future? Proceedings of the Royal Society, Series B, 271(Suppl. 4),
177–179.
Winwar, F. (1950). The immortal lovers: Elizabeth Barrett & Robert
Browning; a Biography. New York: Harper.
Zahavi, A., & Zahavi, A. (1997). The handicap principle: A missing piece
of Darwin’s puzzle. New York: Oxford University Press.
Received June 30, 2005
Revision received October 24, 2005
Accepted October 31, 2005 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 77–96 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.77

Navigating the Interdependence Dilemma: Attachment Goals and the Use

of Communal Norms With Potential Close Others
Jennifer A. Bartz and John E. Lydon
McGill University

Four studies investigated attachment in the context of new relationship development. Anxiously attached
individuals overwhelmingly used communal norms and avoided using exchange norms when interacting
with a potential close other; however, when a potential close other used communal norms, anxious
individuals experienced increased interpersonal anxiety. Anxious individuals also used discrete commu-
nal behaviors to diagnose relationship potential. By contrast, secure individuals were more comfortable
in potential communal situations. Moreover, implicit thoughts about closeness were associated with
improved performance on a mental concentration task for secure individuals, whereas implicit closeness
thoughts were associated with poorer performance for anxious individuals. Finally, avoidant individuals
disliked the potential close other when the other used communal norms and downplayed relational
motives for the other’s communal behavior.

Keywords: attachment, communal, exchange, relationship development, interdependence dilemma

The need to establish close bonds is a basic human characteristic
(Baumeister & Leary, 1995; Bowlby, 1969, 1973; Maslow, 1962;
Sullivan, 1953). The achievement of close bonds has been found to
be associated with improved mental and physical well-being
(House, Landis, & Umberson, 1988; Kiecolt-Glaser & Newton,
2001), whereas the failure to achieve closeness has been linked to
mental (Bowlby, 1969; Davilia, Burge, & Hammen, 1997; Leary,
1990) and physical illness (Lynch, 1979) and suicide (Trout,
1980). Because of the central role that close relationships play in
Jennifer A. Bartz and John E. Lydon, Department of Psychology,
McGill University, Montreal, Quebec, Canada.
Portions of this research were presented in symposia presentations at the
Fourth Annual Meeting of the Society for Personality and Social Psychol-
ogy, Los Angeles, California, February 2003, and at the Fifth Annual
Meeting of the Society for Personality and Social Psychology, Austin,
Texas, January 2004. In addition, Study 3 was presented at the Seventh
Annual Meeting of the Society for Personality and Social Psychology,
Palm Springs, California, January 2006. This research is based on Jennifer
A. Bartz’s doctoral dissertation.
This research was supported in part by a Social Sciences and Humanities
Research Council of Canada doctoral fellowship and the Mary Louise
Taylor McGill Major fellowship to Jennifer A. Bartz and by grants from
the Social Sciences and Humanities Research Council of Canada and the
Fonds Pour la Formation de Chercheurs et l’Aide a la Recherche (Quebec,
Canada) to John E. Lydon. We thank Mark Baldwin and Jeff Simpson for
their helpful comments and suggestions, as well as Niall Bolger and Ringo
Moon-Ho Ho for advising about statistical analyses. We also gratefully
acknowledge Hugo Gagnon, Sarah Kerner, Dominique Pipher, Kristin
Tallman, Laura Monner, Rachel Firmer, Clara Wagner, and Miriam Ro-
zenek for their help in conducting this research.
Correspondence concerning this article should be addressed to Jennifer
A. Bartz, who is now at the Department of Psychiatry, Mount Sinai School
of Medicine, One Gustave L. Levy Place, Box 1230, New York, NY
10029. E-mail: jennifer.bartz@mssm.edu
the human experience, understanding the factors that are involved
in the development of these bonds is an important endeavor.
At the outset of a relationship, there is a great deal of uncer-
tainty: “Is the other person equally interested in developing a
relationship?” “How do I communicate interest?” “Will I be re-
jected?” People want to demonstrate interest and commitment but
are reluctant because trust is not yet established. Ironically, as
Holmes (1991) noted, although feelings of trust influence one’s
level of involvement, trust cannot be assessed unless one is at least
somewhat involved. Thus, one is presented with an interdepen-
dence dilemma in which the desire to express interest and com-
mitment must be weighed against the risk of rejection. Impor-
tantly, though the ability to navigate the interdependence dilemma
has consequences for relationship progress: If one is not willing to
take a leap of faith, the relationship is unlikely to get off the
ground. Although some find it relatively easy to deal with this
uncertainty, others find it more difficult. What factors are associ-
ated with the ability to navigate the interdependence dilemma?
Attachment theory has made substantial contributions to the
understanding of close relationships (Bowlby, 1969, 1973; Fraley
& Shaver, 2000; Hazan & Shaver, 1987, 1994; Mikulincer &
Shaver, 2003; Simpson & Rholes, 1998). Although considerable
research has been conducted to investigate attachment in the
context of established adult relationships, less research has been
conducted to probe the role of attachment in the context of new
relationship development. Yet the notion that attachment models
play a part in new relationships, guiding interpersonal perceptions,
expectations, and behaviors with new partners, is a basic assump-
tion of adult attachment theory (Collins & Read, 1994; Fraley &
Shaver, 2000; Hazan & Shaver, 1987; Mikulincer & Shaver, 2003;
Pietromonaco & Feldman Barrett, 2000). The goal of this research
was to explore attachment differences in the development of
closeness, specifically with regard to the interdependence dilemma
people face at the outset of a close relationship.

77
78 BARTZ AND LYDON
Close Relationships and the Use of Communal and
Exchange Norms
What is a close relationship? One feature distinguishing close
relationships from more casual relationships is the set of norms
that govern the giving and receiving of benefits. According to
Clark and colleagues (Clark, 1984a, 1984b; Clark & Mills, 1979;
Clark, Mills, & Powell, 1986), close relationships, such as those
between family members, friends, and romantic partners, are as-
sociated with communal norms, whereas more casual relation-
ships, such as those between business partners, acquaintances, and
strangers, are associated with exchange norms. Communal norms
reflect a genuine concern for the welfare of the other. Benefits are
given on the basis of need or to please, and people do not keep
track of individual contributions. Giving help to or doing a favor
for the other person does not necessitate that the other person
reciprocate. Likewise, receiving help or a favor does not require
one to respond in kind. By comparison, exchange norms reflect the
idea that no obligation is felt toward the needs of the other person.
Benefits are given in return for benefits received or with the
expectation of repayment, and people tend to keep track of indi-
vidual contributions. In essence, the use of communal norms
means that benefits and aid are given freely and that if there is
reciprocation, it is performed to meet the needs of the other person,
whereas the use of exchange norms means that benefits and aid are
not given freely and that receiving a benefit or aid calls for prompt
reciprocation, preferably in kind, to eradicate the outstanding debt.
Clark and colleagues have found considerable support for this
theory. In one study (Clark, 1984b), participants worked with a
confederate on a task for a shared reward. On completion of the
task, participants were responsible for dividing the reward between
the two group members. The dependent variable was whether
participants chose to work with the same color pen as or a different
color pen from their partner. Choosing a different color pen more
often than chance was thought to indicate exchange norms, as
individual contributions would be clear, whereas choosing the
same color pen more often than chance was thought to indicate the
avoidance of exchange norms and the use of communal norms,1 as
individual contributions to the task would be obscured. Supporting
Clark’s theory, participants avoided using exchange norms when
they believed friendship with their partner was possible—presum-
ably to signal their interest in a relationship with the other per-
son— but participants used exchange norms when they thought
friendship was unlikely.
In a second study investigating the behavior of existing friends
(communal condition) and strangers (exchange condition), Clark
(1984b) found that whereas strangers generally chose to work with
a different color pen from their partner (i.e., used exchange
norms), pen choice for existing friends was random. It was theo-
rized that the existing friends did not go out of their way to actively
avoid using exchange norms because they had an established
relationship and, therefore, should not need to send a message
signaling interest in friendship (Clark, 1984b). It is important to
note, however, that random pen choice is consistent with the use of
communal norms because it suggests participants were not trying
to keep track of individual contributions. In sum, these studies
suggest that people avoid using exchange norms when the poten-
tial (and desire) exists for friendship with another person, presum-
ably to signal interest in a communal relationship, but people use
exchange norms when the potential for friendship does not exist.
New Relationship Development and the Interdependence
Dilemma
Lydon, Jamieson, and Holmes (1997) conducted a series of studies
focusing on the awkward position people face when they hope to
establish a communal relationship with an acquaintance. These re-
searchers found that when people desire a communal relationship with
someone, consistent with Clark’s (1984b) theory, they appear to know
the communal script (i.e., give freely without concern for reciproca-
tion), and they try to behave communally, for example, by going out
of their way to do a favor for or help the potential friend. However,
putting themselves on the line in this way creates anxiety, and to
reduce their anxiety, people look for signs of interest and commitment
(“Does he or she care about me?”), especially in patterns of social
exchange. As Holmes (1991) noted, patterns of social exchange are
often used to diagnose signs of caring because they yield information
about the other’s willingness to make personal sacrifices and to
respond to one’s needs. However, Holmes (1981, 1991) warned, this
microlevel perspective on social exchange patterns and on the other’s
behavior can have the ironic effect of increasing feelings of vulnera-
bility. Specifically, he argued, when people look to discrete behaviors
to diagnose relationship potential, they tend to attach more meaning to
those behaviors than is appropriate and, as a result, the importance of
the behavior for the relationship becomes exaggerated; moreover, he
stated, when discrete behaviors are analyzed in isolation, it tends to
lead to perceived imbalances in what was given relative to what has
been received (Holmes, 1981, 1991).
Thus, at the outset of a relationship, there is a strong desire for
closeness but uncertainty about the other’s motives, and this cre-
ates a great deal of anxiety. The hope is that the relationship will
move forward, but for this to happen, feelings of anxiety and
uncertainty need to be regulated. This leads to a vigilant monitor-
ing of the other’s behavior in an effort to confirm the other’s
interest and commitment. Ironically, though, this vigilant monitor-
ing and microlevel perspective can perpetuate the very feelings of
anxiety and uncertainty they are performed to control.
Adult Attachment Theory
According to adult attachment theory, over the course of re-
peated interactions with significant others, individuals develop
1
The distinction between the avoidance of the use of exchange norms
and the use of communal norms is important. Simply not keeping track of
individual contributions (indicated by random pen choice) is thought to
reflect the use of communal norms, as it suggests participants were not
anticipating using task contribution information when it came time to
distribute the reward. By comparison, the active avoidance of the use of
exchange norms (indicated by participants’ choosing the same color pen
significantly more often than chance) is thought to reflect an effort to avoid
looking as though one would prefer an exchange relationship (Clark,
1984b). As Clark (1984b) stated, “when people are trying to form a
communal relationship, they are not only concerned with following com-
munal norms but also with avoiding any perception on others’ parts . . . that
they might prefer an exchange relationship” (p. 553).
 NAVIGATING THE INTERDEPENDENCE DILEMMA
mental models for close relationships that contain beliefs about
whether the self is worthy of love and affection and whether others
are trustworthy and reliable (Hazan & Shaver, 1987, 1994; Miku-
lincer & Shaver, 2003). Moreover, in addition to the importance of
attachment models in guiding cognition, theorists have begun to
stress the emotion and behavioral regulation properties of the
attachment system when describing differences in attachment (Fra-
ley & Shaver, 2000; Mikulincer & Shaver, 2003). Whereas adult
attachment models have been conceptualized in different ways,
current theory emphasizes a dimensional approach in which dif-
ferences in attachment lie along the two dimensions of anxiety and
avoidance (Brennan, Clark, & Shaver, 1998; Fraley & Waller,
1998).
Attachment anxiety is associated with a heightened desire for
closeness and intimacy combined with concerns about attachment-
figure (un)availability. Furthermore, attachment anxiety is theo-
rized to be associated with the use of hyperactivation strategies
that aim to secure attention from an unresponsive or inconsistently
responsive attachment figure (Mikulincer & Shaver, 2003). Spe-
cifically, as Mikulincer and Shaver (2003) described, attachment
anxiety is associated with hypervigilance and increased sensitivity
to cues of acceptance and rejection, attachment figures are in-
tensely monitored, and efforts are made to maintain contact with
the attachment figure. Furthermore, as these researchers noted,
attachment anxiety is associated with a preoccupation with self-
worth, sensitivity to one’s internal distress, and emotion-focused
coping (e.g., ruminating over worries and concerns). Finally, be-
cause the attachment system is chronically engaged in the pursuit
of attachment-related goals, individuals high in attachment anxiety
tend to have few resources left for the pursuit of such nonattach-
ment endeavors as exploration and affiliation (Mikulincer &
Shaver, 2003).
Attachment avoidance, on the other hand, is associated with a
strong need for independence and the tendency to minimize the
importance of closeness. As Mikulincer and Shaver (2003) ex-
plained, attachment figures are seen as unreliable and unable to
provide protection, and so, to deal with ensuing feelings of vul-
nerability, attachment needs are denied, and self-reliance is pur-
sued. Thus, in contrast to attachment anxiety, attachment avoid-
ance is associated with the use of deactivation strategies that
function to prevent distress from the failure to attain closeness by
shutting down the attachment system (Mikulincer & Shaver,
2003). In this case, monitoring of the attachment figure is avoided,
and efforts are made to prevent confrontations with threatening
information that could activate the attachment system and cause
distress (Mikulincer & Shaver, 2003). In contrast to attachment
anxiety, attachment avoidance is associated with distance coping
in which efforts are made to suppress worries and concerns (Miku-
lincer & Shaver, 2003).
Finally, attachment security (reflecting low anxiety and low
avoidance) is associated with comfort with closeness and auton-
omy and the use of primary attachment strategies (e.g., seeking
closeness and support) in times of need, rather than the secondary
strategies of hyperactivation or deactivation. Basically, a history of
attachment-figure availability reinforces the beliefs that people are
generally well intentioned and that the world is a safe place, thus
allowing more securely attached individuals to develop new rela-
79
tionships and to engage in such non-attachment-related activities
as affiliation and exploration (Mikulincer & Shaver, 2003).
The Present Investigations
We believe the attachment system should be critical in guiding
perceptions, expectations, and behaviors with potential relation-
ship partners, especially with respect to navigating the interdepen-
dence dilemma. Although attachment models arise over the course
of repeated interactions with significant others and primarily re-
flect expectations about close others, we believe they nevertheless
should come into play with potential partners—that is, individuals
with whom one would like to be close. Collins and Read (1994)
argued that chronic attachment models should be especially influ-
ential in interpersonal situations in which little is known about the
other person (also see Pietromonaco & Feldman Barrett, 2000).
Interpersonal situations occurring at the outset of a relationship are
ambiguous, and little is known about the other person; thus, people
should be especially likely to draw on their working models of
attachment to guide expectations and to gauge behavior. Collins
and Read (1994) also noted that attachment models should come
into play when the situation is relevant to chronic attachment goals
(e.g., seeking acceptance or establishing independence). To the
extent that people see these situations as opportunities to satisfy
chronic attachment goals, attachment models and associated be-
liefs, plans, and strategies should come into play (Mikulincer &
Shaver, 2003).
Hypotheses
Situations involving the desire and opportunity for closeness
should activate the chronic goals (for closeness and acceptance)
and the hyperactivation strategies associated with attachment anx-
iety and should magnify the approach–avoidance conflict associ-
ated with this attachment orientation. Thus, we predicted that in
potential communal situations (i.e., when interacting with an at-
tractive, friendly, available partner), individuals high in anxiety
and low in avoidance ( preoccupied) would be especially likely to
signal their interest in a communal relationship by avoiding the use
of exchange norms. At the same time, given their uncertainty about
the reliability of others and their chronic concerns about self-
worth, anxious individuals should be most susceptible to interde-
pendence dilemma concerns inherent in these situations. Interde-
pendence dilemma concerns should also be particularly salient
because, in contrast to their more secure counterparts, anxiously
attached individuals’ strong desire for closeness means they have
more to lose if the relationship does not materialize. Thus, we
predicted that, ironically, when the potential close other used
communal norms, anxious individuals would feel especially anx-
ious and uncertain (“Is this person really communicating interest in
me?”) and concerned about self-worth (“How am I coming across?
Is this person attracted to me?”). Furthermore, given their uncer-
tainty about others’ interpersonal motives, anxious individuals
should require greater reassurance and, thus, should be especially
likely to focus on discrete events in an effort to diagnose relation-
ship potential. Finally, anxious individuals’ interpersonal uncer-
tainty and anxiety should have consequences: In situations involv-
ing the possibility of closeness, anxiously attached individuals’
80 BARTZ AND LYDON
interpersonal concerns should interfere with their ability to engage
in non-attachment-related activities.
Potential communal situations should also arouse the chronic
goals and deactivation strategies associated with attachment avoid-
ance. Highly avoidant individuals, especially those who are low in
anxiety (dismissive), are distrustful of close others, desire indepen-
dence and self-reliance, and prefer to maintain a distance between
themselves and close others. Thus, it was predicted that avoidant
individuals would use exchange norms with a potential close other
to signal their aversion to intimacy and to establish boundaries.
Similarly, when a potential close other communicated interest in
closeness, it was theorized that although avoidant individuals
would likely be concerned about the other encroaching on their
independence, they might nevertheless regulate their distress by
avoiding confrontations with these unpleasant thoughts. Thus, we
predicted that when a potential close other signaled interest,
avoidant individuals would disparage the other so as to avoid
dealing with the possibility of closeness (if the other is seen as
unattractive or unlikable, one does not have to think about what
might happen in a relationship with that person). Finally, given
their desire to downplay the importance of closeness, avoidant
individuals should be less likely to make relational attributions for
a potential close other’s communal behavior.
In contrast to their more insecure counterparts, secure individ-
uals should demonstrate greater interpersonal confidence in poten-
tial communal situations because of their positive beliefs about
themselves and others. In contrast to their more anxious counter-
parts, secure individuals should be less desperate to communicate
interest in closeness. That is, they might not feel compelled to go
out of their way to signal interest in a close relationship because of
their trust that the relationship would develop. Moreover, their
interpersonal confidence should translate into greater tolerance for
uncertainty, reducing the need to microscopically analyze the
other’s behavior for signs of interest. In contrast to their more
avoidant counterparts, secure individuals should not feel threat-
ened by another’s interest in closeness because they are less
concerned with preserving their independence. Thus, we predicted
that in contrast to their more insecure counterparts, secure indi-
viduals would feel relatively comfortable when a potential close
other expressed interest in closeness.
Study 1
This study was designed to investigate the association between
attachment and the use of communal and exchange norms with a
potential close other. The procedure for this study was based on
Clark’s (1984b) paradigm. The participant and the confederate
worked on a group task for a shared reward; the dependent variable
was whether the participant chose to work with the same color pen
as the confederate. Choosing a different color pen was considered
indicative of exchange norms because individual contributions
would be clear, whereas choosing the same color pen was consid-
ered indicative of the avoidance of exchange norms because indi-
vidual contributions would be obscured. It was predicted that
anxious individuals would go out of their way to avoid appearing
exchange oriented (i.e., they would use the same color pen as their
partner) to signal interest in closeness, whereas avoidant individ-
uals would adopt an exchange orientation (i.e., they would use a
different color pen from their partner) to signal their aversion to
closeness. Finally, it was predicted that although secure individu-
als might not actively avoid using exchange norms, nevertheless,
they would not adopt an exchange orientation (i.e., their pen choice
would be random).
Method
Participants
Seventy university students volunteered to participate. Single partici-
pants were targeted to ensure that they would be available and interested in
a communal relationship with the confederate (Clark, 1986). Three partic-
ipants were excluded from the analyses because they used their own pen to
work on the group task. Thus, there were 67 participants in the final sample
(32 men and 35 women, M age ⫽ 19.5 years). Fifty-six participants
described themselves as single, and 11 participants described themselves as
dating.2 Participants received either extra credit or $10 (Canadian) for their
participation.
Procedure
University students were recruited to participate in a study supposedly
investigating group performance and monetary incentives. In trying to
create the potential for a communal relationship, we followed Clark’s
(1984b, 1986) induction and had the participant interact with an attractive,
opposite-sex partner (a confederate) who, the participant was subtly in-
formed (see discussion of the group information sheet, below), was single
and a recent transfer student. In addition, the participant and the confed-
erate had a 2-min interaction in the hallway prior to entering the testing
room, purportedly because the experimenter was running late. This inter-
action was designed to encourage a sense of connection between the
participant and the confederate. During the interaction, the confederate
attempted to engage the participant in a conversation first by asking
whether the participant had ever participated in a psychology experiment
and then continuing with casual conversation (e.g., asking, “Where are you
from?”, “What is your major?”, etc.).
After the hallway interaction, the participant and the confederate were
brought into the lab. In accordance with Clark (1984b), the participant and
the confederate were told that the study was designed to investigate the
effects of monetary incentives on group performance and attitudes and that
they would be working on two group tasks for which they would be
rewarded. They were further told that one participant would be in charge
of dividing the reward from the first task and that the other participant
would be in charge of dividing the reward from the second task. It was
emphasized that they were free to divide the reward however they wanted
(in actuality, they never divided the reward). The first group task was then
explained. This task consisted of finding a series of number sequences
imbedded in a matrix. Fifty cents would be given for each number se-
quence found. The participant was always in charge of dividing the reward
on the first group task, and the confederate was always in charge of
dividing the reward on the second group task (which never occurred).
The participant and the confederate were then informed that, to save
time, they would be working separately so that one person could complete
the personality questionnaires while the other began the first group task.
2
The 11 dating participants were included in the analyses because
participants were sometimes confused about what dating meant (i.e., the
question did not specify dating one person exclusively). Including the
dating participants did not alter the results. Although we targeted single
participants in Studies 2, 3, and 4, those who described themselves as
dating were included in the analyses.
 NAVIGATING THE INTERDEPENDENCE DILEMMA
The experimenter then escorted the confederate to a separate testing room
and returned with an informed-consent form, the personality measures, and
a group information sheet for the participant to complete. The group
information sheet, which requested demographic information including
relationship status and length of attendance at the university, was used to
make salient the potential for a communal relationship with the confederate
(see Clark, 1984b, 1986). Specifically, the group information sheet was
given to the participant after the confederate had completed the top portion,
indicating that she or he was single and a recently arrived transfer student.
In this way, the participant was made aware that the confederate was
available for a communal relationship.
After completing the personality measures, which included the Relation-
ship Questionnaire (RQ; Bartholomew & Horowitz, 1991), a measure of
chronic attachment, and the group information sheet, the participant was
given the first task, already partially completed by the confederate, to
finish. Three red pens and three black pens were placed in a pen-holder on
the participant’s desk. The participant thus had the choice to work with
either the same color pen as or a different color pen from the confederate
(the confederate alternated working with either a red or black pen). The
participant was allotted 4 min to work on the task, after which he or she
was informed that the study was over, probed for suspicions, debriefed, and
compensated.
Measures
The RQ (Bartholomew & Horowitz, 1991) consists of four short para-
graphs describing the secure, preoccupied, avoidant-dismissive, and
avoidant-fearful attachment styles. Participants rated the extent to which
they resembled each of the four styles in their close relationships (i.e.,
relationships with parents, siblings, close friends, relatives, or romantic
partners) based on a 5-point scale ranging from 0 (not at all) to 4
(completely). Participants also selected the one attachment style that best
described how they felt in their close relationships. Following Fraley and
Shaver (1997), we created composites of anxiety and avoidance by sub-
tracting participants’ dismissive ratings from their preoccupied ratings and
by subtracting their secure ratings from their fearful ratings. We then
standardized these composite scores.
Results
To investigate whether attachment was associated with partici-
pants’ pen choice, we adopted two data-analysis strategies. First,
participants were grouped according to their self-categorized at-
tachment style, and the Pearson chi-square test was conducted to
investigate whether pen choice was associated with chronic attach-
ment. Following Clark (1984b), we also conducted difference of
proportion tests to determine whether group differences in pen
choice differed significantly from chance. Second, participants
were grouped according to whether they chose the same color pen
as or a different color pen from the confederate, and independent-
samples t tests were conducted to investigate group differences on
the continuous attachment scores (i.e., the anxious and avoidant
composites and the individual attachment items). Finally, although
we did not have specific predictions related to sex, we investigated
whether sex was associated with the use of communal and ex-
change norms and whether sex qualified any of the findings,
recognizing that sex was confounded with confederate.
Categorical Analyses
In this study, 39% of participants categorized themselves as
secure, and 20%, 21%, and 20% categorized themselves as
81
avoidant-dismissive, preoccupied, and avoidant-fearful, respec-
tively (one participant did not answer the attachment categoriza-
tion item). The Pearson chi-square test revealed that pen choice
was associated with attachment, ␹2(3, N ⫽ 66) ⫽ 13.67, p ⬍ .005
(see Table 1). Moreover, as predicted, the proportion of preoccu-
pied participants who chose the same color pen (93%) was signif-
icantly greater than the proportion expected by chance (50%; Z ⫽
3.22, p ⬍ .001). With respect to the other three groups, pen choice
did not differ significantly from chance. The proportion of secure
participants who chose the same color pen (39%) was not signif-
icantly less than the proportion expected by chance (Z ⫽ 1.17, ns),
the proportion of dismissive participants who chose the same color
pen (31%) was not significantly less than the proportion expected
by chance (Z ⫽ 1.37, ns), and the proportion of fearful participants
(46%) who chose the same color pen was not significantly less
than the proportion expected by chance (Z ⬍ 1, ns). Finally, the
Pearson chi-square test unexpectedly revealed a marginal effect for
sex, ␹2(1, N ⫽ 67) ⫽ 3.39, p ⫽ .066. Overall, men were slightly
more likely to use communal norms (63%) than were women
(40%).
Continuous Analyses
Inspection of the anxious and avoidant composites showed that
those who chose the same color pen were more anxious, M(34) ⫽
.31, than those who chose a different color pen, M(33) ⫽ ⫺.32,
t(65) ⫽ 2.67, p ⫽ .01, but there was no difference between the
same and different pen choice groups on the avoidant dimension
(t ⬍ 1.5). Analyses of the individual attachment items similarly
found that those who chose the same color pen were more preoc-
cupied (M ⫽ .41) than those who chose a different color pen (M ⫽
⫺.42), t(60.79) ⫽ 3.73, p ⬍ .001, whereas there were no differ-
ences between the same and different pen choice groups on at-
tachment security, dismissive, or fearful ratings (all ts ⬍ 1.7). It is
interesting to note, however, that when we investigated whether
sex qualified any of these findings, results revealed a significant
interaction between sex and pen choice on dismissive ratings, F(1,
63) ⫽ 7.44, p ⬍ .01. Specifically, men who chose a different color
pen were significantly more dismissive, M(12) ⫽ .84, than men
who chose the same color pen, M(20) ⫽ ⫺.10, t(63) ⫽ ⫺2.76, p ⬍
.05, whereas dismissive scores did not differ for women as a
function of pen choice (t ⬍ 1.2).
Discussion
In sum, our predictions for the anxiously attached were sup-
ported: Preoccupied individuals were more likely to choose the
Table 1
Pen Color Choice as a Function of Attachment (Study 1)
Pen color choice
Attachment style Same Different
Secure 10 16
Preoccupied 13 1
Avoidant-dismissive 4 9
Avoidant-fearful 6 7
Total 33 33
82 BARTZ AND LYDON
same color pen as their partner than one would expect by chance,
and those who chose the same color pen as their partner were more
anxious and more preoccupied. Thus, in accordance with Clark’s
(1984b) theorizing, these findings suggest that when interacting
with an attractive, available, opposite-sex partner, anxiously at-
tached individuals go out of their way to avoid appearing exchange
oriented, presumably to signal interest in their partner.
The findings for the other three groups were less clear-cut. As
predicted, pen choice for secure individuals was random; however,
pen choice for fearful individuals was also random. Finally, al-
though dismissive individuals were more likely to choose a dif-
ferent color pen than preoccupied individuals—suggesting their
preference for an exchange orientation— contrary to our predic-
tions, their pen choice was not significantly different from chance.
Moreover, those who chose a different color pen were not more
dismissive. That said, additional analyses revealed that men who
chose a different color pen were more dismissive. However, very
few women (n ⫽ 2) categorized themselves as dismissive, making
it difficult to investigate this effect for women.
On a related note, unexpectedly, women overall were slightly
less likely than their male counterparts to use the same color pen
as their partner; this may also have made it difficult to detect an
association between women’s avoidant attachment and the use of
exchange norms (i.e., if women were less likely to use communal
norms, dismissive women would have had to have been that much
less communal). Why were women slightly less likely to use
communal norms? If anything, one would expect women to be
more communal. One possible factor may have been the confed-
erate. Male and female participants interacted with different con-
federates; although we tried to select comparable confederates,
slight differences (e.g., one confederate may have been liked more
than the other) could have influenced participants’ desire for
closeness and their use of communal norms.
Study 2
In Study 1, when interacting with an available, desirable partner,
anxious individuals, compared with their more secure and avoidant
counterparts, went out of their way to express interest in closeness,
but how do anxious individuals respond when an available, desir-
able partner expresses interest in them? The goal of this study was
to investigate the relationship between attachment and partici-
pants’ response to the use of communal or exchange norms by a
potential close other. In this study, we created a new psychological
situation by modifying the procedures from Study 1. Specifically,
the confederate worked on the group task after the participant, and
the completed group task was returned to the participant, who was
in charge of tabulating the results. In this way, the participant was
able to see whether the confederate used the same or different
color pen (experimental manipulation of communal vs. exchange
norms). State feelings of interpersonal anxiety, partner liking, and
partner perceptions were assessed.
It was theorized that whereas secure individuals would feel
comfortable with a potential close other’s use of communal norms,
insecure (anxious and avoidant) individuals would feel more dis-
tressed. Anxious individuals have a strong desire for closeness but
are uncertain about others’ interest and tend to doubt their self-
worth. As we have argued, anxious individuals should be espe-
cially susceptible to interdependence dilemma concerns in situa-
tions involving possible closeness because of their chronic
interpersonal insecurities. Thus, it was predicted that, compared
with the exchange condition, anxious individuals in the communal
condition would experience more interpersonal anxiety because of
the potential for closeness in this condition. Avoidant individuals,
on the other hand, strive for independence and do not like it when
others try to get too close. Thus, it was predicted that, compared
with the exchange condition, avoidant individuals in the communal
condition would respond to their partner’s overtures with de-
creased liking and would try to cast their partner’s communal
behavior in a negative light. Finally, although situations involving
possible closeness should be distressing to avoidant individuals
given their preference for independence, avoidant individuals tend
to suppress worries and concerns. Thus, we did not have specific
predictions regarding avoidant individuals’ affective response in
this study.
Method
Participants
Sixty-three university students volunteered to participate. Single partic-
ipants were targeted; 2 participants were excluded because they were in a
serious relationship, and 2 participants were excluded because they sus-
pected that their partner was a confederate. There were 59 participants in
the final sample (26 men and 33 women, 50 single and 8 dating [1
participant did not answer this question], M age ⫽ 21.05 years). Partici-
pants were randomly assigned to the communal (n ⫽ 28) or exchange
conditions (n ⫽ 31). Participants were given either extra credit or $10
(Canadian) for their participation.
Procedure
University students were recruited to participate in a study supposedly
investigating personality, group performance, and incentives. As in Study
1, the participant had a 2-min interaction with an attractive, opposite-sex
confederate on arrival at the testing session. After the interaction, the
participant and the confederate were brought into the lab and were in-
formed that they would be working on two group tasks for which they
would be rewarded; when the first group task was completed, one person
would be given the reward to divide between the members of the group,
and when the second group task was completed, the other person would be
given the opportunity to divide the reward. Different from Study 1, it was
explained that even though participants would be working together on both
tasks, because the study was interested in the effects of two different
working conditions, they would be working in two separate rooms during
the first task, so they would be unable to communicate, and they would be
working in the same room during the second task. The group task, which
was similar to the number matrix task used in Study 1, was then explained.
Participants were told that they would be given 25¢ for each number
sequence found. Again, the participant was responsible for dividing the
reward on the first group task (the second group task never took place).
The participant was then escorted to a room next door, was given an
informed-consent form to complete, and then began the first group task.
The experimenter left and returned after 6 minutes with the group infor-
mation sheet (similar to that used in Study 1), partially completed by the
confederate, and the personality measures, which again included a measure
of chronic attachment. The experimenter then brought the group task to the
confederate, who used either the same color pen as (communal condition)
or a different color pen from (exchange condition) the participant. (Al-
though participants were not specifically told that their partner had an
 NAVIGATING THE INTERDEPENDENCE DILEMMA
option with regard to pen choice—primarily because it was difficult to
convey this information without drawing attention to the importance of
pens in the study— care was taken to display the pens on the participant’s
desk in an artificial manner so that the participant would assume that the
other’s desk space was similarly laid out.) The completed group task was
then returned to the participant, who was instructed to tabulate how many
number sequences the team had found. In this way, the participant was able
to see whether the confederate used a pen of the same color or one of a
different color. At this point, the participant was told that before beginning
the second group task in which he or she would be working with the other
participant in the same room, it was necessary to complete a social
interaction prestudy questionnaire to control for individual differences in
expectations about working together (Clark & Mills, 1979). The question-
naire assessed situational feelings of interpersonal anxiety (i.e., state affect
and self-esteem), partner perceptions, and partner liking. After completing
these measures, the participant was probed for suspicions, debriefed, and
compensated.
Measures
Experience in Close Relationships scale (ECR; Brennan et al., 1998).
This 36-item questionnaire is designed to assess the two dimensions of
attachment avoidance and anxiety. Avoidant items reflect comfort with
closeness and dependency, and anxious items reflect anxiety about being
abandoned. Participants indicate on a 7-point scale how much they agree or
disagree with each item, in terms of how they experience romantic rela-
tionships. We added the ECR to this study because of the limitations
associated with the RQ (i.e., assessment of discrete dimensions and cate-
gories). After reverse scoring items for which lower numbers reflected
greater attachment avoidance or anxiety, we computed attachment avoid-
ance and anxiety scores for each participant by taking the mean response
on the 18 avoidance (␣ ⫽ .90) and 18 anxiety (␣ ⫽ .90) items. Avoidance
and anxiety scores were not correlated, r(57) ⫽ .19, ns.
Situational feelings of interpersonal anxiety. To assess situational feel-
ings of interpersonal anxiety, we created a composite that included state
affect and aspects of state self-esteem. Specifically, the composite included
the Anxious and Uncertain subscales from the Profile of Mood States
(POMS; McNair, Lorr, & Droppleman, 1971). (These items were inter-
spersed among items from the Depressed, Hostile, Happy, and Confident
subscales of the POMS.) The composite also included the Appearance
subscale from the State Self-Esteem Scale (SSES; Heatherton & Polivy,
1991). Examples from this subscale include “I feel that others respect and
admire me,” “I feel good about myself,” “I am pleased with my appearance
right now,” and “I feel unattractive” (this last items was reverse scored).3
The Cronbach’s alpha coefficient for the Anxious, Uncertain, and Appear-
ance Self-Esteem subscales (reverse scored) was .73. To calculate the
index, we summed participants’ ratings from the Anxious and Uncertain
subscales and subtracted their Appearance Self-Esteem rating (standard-
ized scores were used in the calculation to adjust for scale differences in
these measures).
Partner perceptions and partner liking. Partner perceptions and part-
ner liking were assessed in the Social Interaction Prestudy Questionnaire.
In this questionnaire, participants rated how well certain traits applied to
their partner (the confederate). Because we were particularly interested in
how avoidant participants would construe their partner’s communal behav-
ior, we selected traits from the Extended Personality Attributes Question-
naire (Spence, Helmreich, & Holahan, 1979), which taps positive com-
munion (warm, helpful, and understanding), negative communion
(spineless, fussy, and servile), positive agency (independent, competitive,
and self-confident), and negative agency (arrogant, greedy, and hostile).
Participants indicated their response by placing a slash through a line,
anchored from 0 (not at all) to 50 (completely). Cronbach’s alpha coeffi-
cients for communal-positive, communal-negative, agency-positive, and
agency-negative composites were .78, .58, .36, and .66, respectively.
83
Participants also indicated the extent to which (a) they liked the person
with whom they were working, (b) they would like to continue a conver-
sation with that person on another occasion, (c) their partner was the kind
of person they would want to have as a friend, and (d) their partner was the
kind of person they would want to work with on another project by placing
a slash through a line, anchored from 0 (not at all) to 50 (completely). We
created a partner-liking score by taking participants’ mean response to the
four partner-liking items (␣ ⫽ .87).
Results
This study was designed to investigate attachment differences in
participants’ response to the use of communal or exchange norms
by a potential close other. The dependent variables were situational
interpersonal anxiety, partner liking, and partner perceptions. It
was predicted that attachment anxiety would be more strongly
associated with interpersonal anxiety in the communal condition
compared with the exchange condition, whereas attachment avoid-
ance would be more strongly associated with partner liking (neg-
ative correlation) and negative partner perceptions in the commu-
nal condition compared with the exchange condition. To
investigate these predictions, we used the Fisher z-transformation
test (Cohen, Cohen, West, & Aiken, 2003) to compare the within-
cell correlations between attachment and interpersonal anxiety,
partner perceptions, and partner liking between experimental
conditions.
Attachment Anxiety and Interpersonal Anxiety
Overall, attachment anxiety was associated with situational feel-
ings of interpersonal anxiety across experimental conditions,
r(57) ⫽ .58, p ⬍ .001. However, as predicted, the association
between attachment anxiety and interpersonal anxiety was signif-
icantly greater in the communal condition, r(25) ⫽ .80, than in the
exchange condition, r(28) ⫽ .37, Z ⫽ 2.59, p ⫽ .01.4 Moreover,
although avoidant attachment was also associated with interper-
sonal anxiety, r(57) ⫽ .37, p ⬍ .005, this association did not differ
between experimental conditions, Z ⫽ 1.12, ns. It is important to
note that the difference in the association between attachment
anxiety and interpersonal anxiety in the communal and exchange
3
Although the SSES taps performance, social competence, and appear-
ance, it is thought that different components of state self-esteem should be
affected by different experimental situations (Heatherton & Polivy, 1991).
We focused on the appearance component because it was theorized that
appearance concerns should be especially salient in the communal condi-
tion. That is, in both the communal and exchange conditions, participants
expected to work on a task with their partner, and thus, performance and
social interaction concerns should be salient. However, appearance con-
cerns (e.g., “I feel unattractive”) should be more salient in the communal
condition than in the exchange condition given the potential for closeness
and increased likelihood of being evaluated by another person on the basis
of one’s appearance in the communal condition.
4
This effect was moderated somewhat by sex ( p ⬍ .10). Anxiously
attached women experienced more interpersonal anxiety than their secure
counterparts across conditions, r(31) ⫽ .53, p ⫽ .001, whereas for anx-
iously attached men, this relationship was weak in the exchange condition,
r(11) ⫽ .35, ns, but very strong in the communal condition, r(11) ⫽ .92,
p ⬍ .001.
84 BARTZ AND LYDON
conditions remained when controlling for attachment avoidance.
That is, a comparison of the partial correlations between attach-
ment anxiety and interpersonal anxiety (controlling for attachment
avoidance) in the communal, r(24) ⫽ .77, and exchange condi-
tions, r(27) ⫽ .39, remained significant (Z ⫽ 2.17, p ⬍ .05).
Attachment Avoidance and Partner Liking and Partner
Perceptions
Partner liking. Overall, avoidant attachment was negatively
associated with partner liking, r(57) ⫽ ⫺.28, p ⬍ .05; however, as
predicted, this association was stronger in the communal condi-
tion, r(25) ⫽ ⫺.54, than in the exchange condition, r(28) ⫽ ⫺.02,
Z ⫽ 2.12, p ⬍ .05. By contrast, attachment anxiety was not
associated with partner liking, r(57) ⫽ ⫺.03, ns. Moreover, at-
tachment anxiety did not qualify the differential association be-
tween avoidant attachment and partner liking in the communal and
exchange conditions. That is, a comparison of the partial correla-
tions between attachment avoidance and partner liking (controlling
for attachment anxiety) in the communal, r(24) ⫽ ⫺.56, and
exchange conditions, r(27) ⫽ ⫺.02, remained significant (Z ⫽
2.18, p ⬍ .05). Finally, although this effect was not qualified by
sex, men in general liked the female confederate more than women
liked the male confederate, t(57) ⫽ 2.28, p ⬍ .05. (This finding
supports our speculation in Study 1 that any sex differences may
have been due to differences in male and female participants’
liking of the confederates rather then sex per se.)
Partner perceptions. Overall, avoidant attachment was asso-
ciated with ascribing more negative communal traits (e.g., servile)
to the partner, r(56) ⫽ .31, p ⬍ .05 (one participant did not
respond to this question); however, as predicted, the association
between avoidant attachment and perceptions of the partner’s
negative communal traits was stronger in the communal condition,
r(25) ⫽ .56, than in the exchange condition, r(29) ⫽ .09, Z ⫽ 1.95,
p ⫽ .051. Unexpectedly, attachment anxiety was also associated
with negative communal partner perceptions, r(56) ⫽ .28, p ⬍ .05.
Although this association did not differ between experimental
conditions (Z ⫽ .70, ns), controlling for attachment anxiety qual-
ified the aforementioned association between attachment avoid-
ance and perceptions of the partner’s negative communal traits
(Z ⫽ 1.58, ns). No sex differences were found in the association
between attachment avoidance and perceptions of the partner’s
negative communal traits. Finally, with respect to agency traits,
avoidant attachment was associated with seeing the partner as
more negatively agentic (e.g., greedy), r(57) ⫽ .28, p ⬍ .05, and
less positively agentic (e.g., less independent), r(57) ⫽ ⫺.31, p ⬍
.05, but these associations did not differ between experimental
conditions.
Discussion
As predicted, attachment anxiety was strongly associated with
interpersonal anxiety when a potential close other used communal
norms. Why were anxiously attached individuals not more dis-
tressed when their partner used exchange norms, as this might
suggest their partner’s disinterest in a communal relationship? We
believe it is unlikely that the anxious participants interpreted their
partner’s exchange behavior as a rejection. The norm when inter-
acting with a stranger is exchange, so anxious individuals may
have thought their partner’s behavior in the exchange conditions
(i.e., choosing to work with a different color pen) was consistent
with the situation and thus did not take it as a personal slight. It
was when the confederate deviated from the norm, expressing
interest in closeness, that they felt especially anxious, uncertain,
and concerned about their appearance. Thus, it appears that when
opportunities arise for anxiously attached individuals to pursue
their goals for closeness and acceptance, they experience more
interpersonal anxiety.
Avoidant individuals had a very different response to the po-
tential close other’s use of communal norms. Attachment avoid-
ance was associated with decreased partner liking in the communal
condition, whereas there was no association between attachment
avoidance and partner liking in the exchange condition. Thus, it
was not simply that avoidant individuals disliked their partner;
rather, they disliked their partner when their partner attempted
closeness. Avoidant individuals value independence and self-
sufficiency and prefer to keep a distance between themselves and
close others; they also tend to avoid confrontation with worries or
concerns. We believe their partner’s communal overtures likely
threatened avoidant individuals’ need for self-reliance, and they
responded by disparaging their partner.
Rather than response to possible closeness, an alternate expla-
nation of the findings from this study is that the confederate’s
obscuring of contributions in the communal condition made it
difficult for participants to divide the reward because each per-
son’s contribution to the task was ambiguous. We believe it is
unlikely that this would explain anxious individuals’ heightened
interpersonal anxiety. Referring back to Study 1, the large majority
of anxious individuals used the same color pen as their partner, and
given that they were also responsible for dividing the reward in
Study 1, they would have been presented with the same quandary.
Avoidant individuals, on the other hand, may indeed have been
perturbed when their partner used the same color pen because it
made it difficult for them to divide the reward. However, given
that distributing benefits on the basis of task contribution is an
exchange norm, this would seem to support the notion that
avoidant individuals prefer to operate from an exchange rather
than a communal perspective.
These findings also shed light on Study 1. In Study 1, anxiously
attached individuals chose to work with the same color pen as their
partner more often than chance. In line with Clark’s (1984b)
theorizing, we interpret this behavior as an effort to signal interest
in closeness by avoiding the use of exchange norms. An alternate
explanation, however, is that anxiously attached individuals ob-
scured contributions because they were concerned they would be
evaluated negatively if they performed poorly. Study 2 suggests
that this was not the case: Anxiously attached individuals had less
interpersonal anxiety when their performance was distinguishable
from their partner’s performance (i.e., in the exchange condition).
In sum, Studies 1 and 2 suggest that situations involving the
opportunity for closeness elicit an approach–avoidance conflict in
anxiously attached individuals. In Study 1, anxiously attached
individuals were more likely that their secure and avoidant coun-
terparts to symbolically approach their partner, whereas in Study 2,
attachment anxiety was associated with greater feelings of inter-
personal anxiety following a potential close other’s communal
 NAVIGATING THE INTERDEPENDENCE DILEMMA
overtures. We believe the confederate’s communal behavior likely
aroused anxious individuals’ interdependence dilemma concerns
(“Is this person interested in me?”) and their deep-seated ambiv-
alence about close others. On the one hand, given their desire for
closeness, anxiously attached individuals were probably highly
motivated to detect signs of interest in the other’s behavior, but on
the other hand, given their chronic uncertainty about others, their
hopes may have been tempered by concerns that they would be
found unworthy (Holmes, 1991). Avoidant individuals, by con-
trast, disparaged their partner when their partner used communal
norms presumably because their partner’s communal overtures
threatened their desire for independence. Studies 3 and 4 were
designed to explore these ideas.
Study 3
A core prediction of the interdependence dilemma is that in
potential communal situations, there is the desire for closeness but
uncertainty about the other’s motives and this uncertainty needs to
be resolved for the relationship to move forward. Consequently, as
Holmes (1991) noted, people seek confirmation about the other’s
motives, often using patterns of social exchange as a testing
ground—that is, they look for evidence that the other genuinely
cares as indicated by, for example, the other’s use of communal
norms (“Is the other responsive to my needs?”). As we have
argued, because of their strong desire for closeness and their
chronic uncertainty, anxiously attached individuals should be most
susceptible to interdependence dilemma concerns. Indeed, Study 2
found attachment anxiety to be more strongly associated with
feelings of interpersonal anxiety in situations involving the possi-
bility of closeness. If anxious individuals are more susceptible to
interdependence dilemma concerns, then they should be especially
likely to engage in this process of uncertainty reduction, evaluating
even relatively trivial behaviors for their diagnostic information
about relationship potential. Indeed, the notion that anxiously
attached individuals are especially alert to environmental cues
conveying information about acceptance and rejection is also a
central prediction of attachment theory (Fraley & Shaver, 2000).
The goal of Study 3 was to investigate the idea that potential
communal situations arouse anxiety and uncertainty in anxiously
attached individuals, which then sets in motion a need to micro-
scopically analyze and diagnose the other’s behavior— even trivial
behaviors—for signs of interest and commitment. Moreover, we
wanted to probe whether anxious individuals’ hypervigilance and
need to reduce uncertainty were associated with assigning greater
relational significance to more trivial situations that are less clearly
communal or whether this process would be more apparent in
situations clearly suggestive of communal potential. Although
everyone should respond with interdependence dilemma concerns
when the behavior of the other is clearly suggestive of communal
potential, we believe that even trivial behaviors should elicit in-
terdependence dilemma concerns for the anxiously attached.
In this study, we departed from the laboratory methodology and
instead used a guided-visualization, self-report methodology to
assess preexisting potential communal relationships from partici-
pants’ own social world. Moreover, although we sought to create
the potential for a romantic relationship in Studies 1 and 2, some
participants may have thought of the confederate as a potential
85
friend and not as a romantic prospect. Thus, in this study, partic-
ipants were instructed to nominate someone with whom they
hoped to establish a deeper relationship (a potential close other);
the nature of that relationship (friendship vs. romantic) was left
open, which allowed us to compare romantic versus nonromantic
potentially communal relationships. Participants then visualized
themselves in one of two scenarios. In one scenario, participants
needed to borrow notes for a class they have missed, and the
potential close other offered to loan them the notes; in the second
scenario, the potential close other suggested being study partners
and exchanging phone numbers. The offer to lend someone notes
is a relatively trivial, mundane event that can be attributed to a
variety of nonrelationship factors, whereas the suggestion to ex-
change phone numbers is less common and the motives for the
behavior, we believe, are more suggestive of relationship potential.
Affect, significance of the event for the future of the relationship,
and relational attributions for the other’s communal behavior were
assessed.
It was hypothesized that overall, participants would believe the
event had greater significance for the future of the relationship and
would make more relational attributions when the potential close
other suggested being study partners than when the potential close
other offered to lend them notes. Similarly, with respect to affect,
it was hypothesized that the study partner scenario would elicit less
indifference and greater unease compared with the notes scenario
because the prospect of closeness would arouse interdependence
dilemma concerns.
Attachment, however, was hypothesized to moderate these ef-
fects. Secure individuals (low anxious and low avoidant) were
predicted to be the baseline group—that is, they should infer
greater significance and make more relational attributions in the
less ambiguous study partner scenario than in the more trivial
notes scenario. They should also experience less indifference and
greater unease in the study partner scenario, in line with interde-
pendence dilemma predictions. Anxious individuals, by compari-
son, should infer greater significance, make more relational attri-
butions for the potential close other’s behavior, and feel more
uneasy regardless of scenario; that is, even the discrete communal
behavior scenario should set in motion interdependence dilemma
concerns. Finally, avoidant individuals should be less likely to
infer significance and should make fewer relational attributions for
the potential close other’s communal behavior because of their
tendency to downplay the importance of closeness; for the same
reason, they should also be more indifferent to the potential close
other’s communal behavior. Lastly, it was predicted that avoidant
individuals would be less happy in the notes scenario, in which
they must rely on the other for help, because of their aversion to
dependency.
Method
Participants
Two hundred seventy-nine participants were recruited on a volunteer
basis to complete one of two relationship surveys. Single participants were
targeted to ensure participants were adequately interested in pursuing a
relationship with their chosen person. We excluded 2 participants who
were unable to nominate a potential close other, 23 participants who were
insufficiently interested in developing a relationship with the potential
86 BARTZ AND LYDON
close other (i.e., their interest ratings were below the midpoint; these
participants did not differ from the other participants in their attachment
avoidance or anxiety scores), 1 participant who did not complete the
attachment questionnaire, and 1 participant who was married, had known
the potential close other for 30 years, and was more than 10 standard
deviations above the mean participant age. Thus, 252 participants com-
pleted one of two relationship surveys (126 men and 125 women, 219
single and 32 dating, M age ⫽ 20.94 years). (There were missing data on
some of the questions because of nonresponse; however, there were never
more than 5 [2%] missing data points for the relational indices, which were
the main dependent variables).
Procedure
Prospective participants were asked if they would be interested in
completing a survey about new relationship development. Participants
were told that it was important to complete the survey in one sitting, alone,
and in a quiet place. Agreeing participants completed an informed-consent
form, were given the survey and an envelope, and were instructed to return
the survey in the sealed envelope to ensure anonymity. Surveyors were
unaware of the hypotheses when administering the survey, and participants
were debriefed on returning the survey.
The survey began with a brief introduction, which stated that the goal of
the research was to learn more about the development of new relationships.
Participants were informed that the survey involved nominating someone
with whom they were not currently friends but with whom they could
imagine being close friends, visualizing themselves and this person in a
scenario, and answering some questions about the scenario as well as some
demographic questions and some personality measures. Participants were
encouraged to fully immerse themselves in the scenario (e.g., by visualiz-
ing themselves and their chosen person in the situation, imagining the
surroundings, etc.). Moreover, to help participants mentally simulate the
experience, blank spaces were inserted throughout the scenario, and par-
ticipants were instructed to write their chosen person’s first name in the
spaces provided.
After the introduction, participants were asked to think of a same-sex or
opposite-sex acquaintance. Specifically, they were instructed to think of
Someone with whom you are not currently friends, but someone with
whom you could imagine being close friends if you got to know each
other better. This person could be a casual friend with whom you
would like to establish a deeper friendship, or, possibly, a romantic
relationship. . . . Although you are not close friends right now, you
think you might really enjoy spending time together in the future. It
remains to be seen how your relationship will develop.
Because of the nature of the scenarios (borrowing class notes and exchang-
ing phone numbers to be study partners), participants were directed to
select a peer, that is, someone relatively close in age, so that the scenario
would be appropriate and realistic.
After selecting their chosen person but before going on to read the social
interaction scenario, participants were instructed to visualize their chosen
person. To aid in the visualization process, participants were presented
with a series of questions about their chosen person (e.g., “What is it like
being with this person?” and “How do you feel when you are with this
person?”). The purpose of the guided visualization was to help participants
immerse themselves so as to discourage top-of-the-head responses (Lydon
et al., 1997). After the visualization, participants read one of the following
two scenarios.
Notes (relatively trivial communal behavior). “Imagine that you and
__________ are taking a class together. You have missed a few classes
since the last midterm and __________ has the notes for the classes you
missed.” After rating how comfortable they would feel in that situation,
participants were asked to imagine that their chosen person had offered to
lend them the notes.
Study partner (less ambiguous gesture of friendship). “Imagine that
you and __________ are taking a class together. One day after class
__________ gives you his or her phone number and suggests being study
partners for the upcoming exam.”
After reading the scenario, participants answered questions about their
affective response, the significance of the event for the future of the
relationship, and attributions for the potential close other’s behavior (see
Measures, below). At the end of the questionnaire, participants indicated
(a) whether they were able to think of someone with whom they desired a
deeper friendship, (b) how long they had known the person, (c) how
interested they were in developing a relationship with the person, (d) what
type (romantic or friendship) of relationship they desired, and (e) how
realistic the scenario was to them. We also assessed chronic attachment
(the measure was presented as a relationship style questionnaire) and
demographic information at this point.
Measures
Significance of the event for the future of the relationship. Participants
rated, on a scale from 1 (not at all) to 7 (a lot), the extent to which they
thought the event would say something about the progress of the relation-
ship and how much closer they thought they would feel to their chosen
person after the event compared with prior to the event. Using the same
scale, participants also rated the extent to which they agreed with the
following statements: “We will probably get closer,” “This event cements
our relationship,” and “This event has no particular implications or signif-
icance for the relationship” (this item was reverse scored). An index of the
significance of the event for the development of the relationship was
created by averaging participants’ responses to these five items (␣ ⫽ .74).
Relational attributions for the potential close other’s behavior. Partic-
ipants rated, on a scale from 1 (disagree strongly) to 7 (agree strongly), the
extent to which they agreed with seven statements reflecting different
attributions for their chosen person’s behavior. A relational attribution
composite was created by taking the mean of participants’ relational
attributions (i.e., “because he or she likes me [i.e., this event occurred
because of the way this person feels about me],” “because he or she cares
about me,” “to become closer,” and “as a gesture of friendship”) and
subtracting their situational attribution (i.e., “I don’t think this event had
much to do with who the other person is, or the way he or she feels about
me; outside circumstances were probably the primary cause of this
event”).5
Affect. Participants rated how indifferent, uneasy, and happy they
would feel in the situation described in the scenario, using a scale from 1
(not at all) to 7 (extremely).
ECR (Brennan et al., 1998). The 36-item questionnaire used in Study
2 was used to assess attachment avoidance (␣ ⫽ .91) and anxiety (␣ ⫽ .91).
5
Participants also responded to a dispositional attribution statement
(“because he or she is a nice person”) and a self-interested attribution
statement (“out of self-interest [i.e., to improve his or her standing in the
class]”). However, these items were not included in the relational attribu-
tion composite because we believe they likely tapped participants’ chronic
beliefs about others and not simply beliefs about whether the behavior had
meaning for relationship development. Indeed, there was no correlation
between making relational attributions and making dispositional or
self-interested attributions, r(245) ⫽ ⫺.01, ns, and r(245) ⫽ .04, ns,
respectively.
 NAVIGATING THE INTERDEPENDENCE DILEMMA
Results
A series of hierarchical multiple regressions were conducted to
investigate the influence of attachment anxiety and avoidance on
participants’ affective response to the potential close other’s com-
munal behavior, their beliefs about the significance of that behav-
ior for relationship development, and the kind of attributions they
made about the potential close other’s communal behavior. At-
tachment anxiety and avoidance were standardized and entered in
the first step of the regression along with scenario (notes vs. study
partner; contrast coded as 1 and ⫺1, respectively); the two-way
interactions of attachment anxiety and avoidance, scenario and
attachment anxiety, and scenario and attachment avoidance were
entered in the second step; and the three-way interaction of sce-
nario, attachment anxiety, and attachment avoidance was entered
in the third step of the regression.
We also investigated whether the type of relationship partici-
pants thought about (friendship or romantic) or participants’ sex
influenced any of the dependent variables and whether relationship
type or participants’ sex qualified any of the effects of attachment
and/or scenario. There were no significant main or interaction
effects of relationship type on any of the dependent variables.
Similarly, with one exception, there were no main or interaction
effects of sex on any of the dependent variables. Specifically,
attachment avoidance and attachment anxiety were negatively
correlated with feeling uneasy for male participants, whereas
women’s uneasiness ratings were not correlated with avoidance or
anxiety. Basically, highly avoidant or anxious men were less likely
to report feeling uneasy compared with their less avoidant and less
anxious counterparts.
Relationship Characteristics
One hundred two participants thought of a potential friend, and
149 participants thought of a potential romantic partner (if partic-
ipants selected both, they were categorized as seeking a romantic
relationship because it was assumed that if participants indicated
any interest in a romantic relationship, then that was their true
preference). The mean interest in developing a relationship was
5.27 (range ⫽ 4.00 –7.00). The mean relationship length was
slightly less than a year and a half (497 days), and the median was
approximately 6 and a half months (196 days; range ⫽ 1 day–11
years). The mean scenario realism score was 4.82 (range ⫽ 1.00 –
7.00). Attachment was not associated with the type of relationship
participants were interested in developing (both ts ⬍ 1), nor was
attachment associated with interest in developing a relationship,
perceived scenario realism, or relationship length (all rs ⬍ .11).
Relationship type, however, was associated with interest in devel-
oping a relationship, t(230.167) ⫽ 6.37, p ⬍ .001. Those seeking
a romantic relationship were significantly more interested in de-
veloping a relationship with their chosen person (M ⫽ 5.56) than
were those seeking a friendship (M ⫽ 4.83).
Significance of Event for Relationship Development
Analyses investigating the composite measure of beliefs about
the significance of the event for relationship development revealed
an effect for scenario (B ⫽ ⫺.43, sr ⫽ ⫺.42, p ⬍ .001). Partici-
87
pants inferred greater significance from the suggestion to be study
partners than from the notes offer, providing construct validation
for the two scenarios—that is, as intended, participants viewed the
study partner scenario, which involved a less ambiguous instance
of the potential close other’s communal behavior, as having a
greater impact on relationship development than the notes sce-
nario. As predicted, however, analyses also revealed a significant
effect for attachment anxiety (B ⫽ .17, sr ⫽ .17, p ⬍ .005).
Anxiously attached individuals inferred greater significance from
the potential close other’s communal behavior regardless of sce-
nario. In addition, there was a marginal interaction between sce-
nario and attachment anxiety (B ⫽ .10, sr ⫽ .09, p ⫽ .095). Simple
effects tests found that attachment anxiety was positively associ-
ated with beliefs about the significance of the event in the notes
scenario, r(127) ⫽ .27, p ⬍ .005, but was not associated with
significance beliefs in the study partner scenario, r(121) ⫽ .11, ns,
primarily because everyone had more optimistic beliefs for the
development of the relationship in the study partner scenario.
Results also revealed an unexpected three-way interaction between
scenario, attachment anxiety, and attachment avoidance (B ⫽ .12,
sr ⫽ .11, p ⬍ .05). This interaction is explained by the fact that for
the notes scenario, high avoidant, low anxious (dismissive) partic-
ipants were considerably less likely than their peers to believe that
the event had significance for relationship development.
Relational Attributions for the Potential Close Other’s
Behavior
Analyses investigating the relational attribution composite also
revealed an effect for scenario (B ⫽ ⫺.29, sr ⫽ ⫺.29, p ⬍ .001),
again providing construct validation and suggesting that partici-
pants perceived their potential close other’s suggestion to be study
partners as more relationally motivated than the offer to lend them
notes. There was also an interaction between scenario and attach-
ment anxiety (B ⫽ .16, sr ⫽ .16, p ⬍ .01; see Figure 1). Simple
effects tests revealed that attachment anxiety was positively asso-
ciated with relational attributions in the notes scenario, r(125) ⫽
.24, p ⬍ .01, but was not associated with relational attributions in
Figure 1. Relationship attributions (for potential close other’s behavior)
as a function of scenario and attachment anxiety (Study 3).
88 BARTZ AND LYDON
the study partner scenario, r(118) ⫽ ⫺.10, ns, again, primarily
because everyone made more relational attributions in the study
partner scenario. Thus, similar to the significance of the event for
relationship development analyses, anxiously attached individuals
made more relational attributions regardless of the quality of the
other’s communal behavior. Finally, as predicted, analyses also
revealed a negative association between attachment avoidance and
making relational attributions (B ⫽ ⫺.22, sr ⫽ ⫺.22, p ⬍ .001),
suggesting avoidant individuals’ reluctance to make relational
attributions for the other’s communal behavior.
Affective Response to Potential Close Other’s Communal
Behavior
Overall, participants felt less indifferent (B ⫽ .20, sr ⫽ .20, p ⬍
.005) and more uneasy (B ⫽ ⫺.15, sr ⫽ ⫺.15, p ⬍ .05) in the
study partner scenario than in the notes scenario. Again, this
finding suggests that the study partner scenario had more emo-
tional relevance for participants than did the notes scenario. The
fact that participants were more uneasy in the study partner sce-
nario also reinforces the notion that interdependence dilemma
situations are associated with increased anxiety. As predicted,
attachment anxiety was associated with feeling uneasy across
scenarios (B ⫽ .13, sr ⫽ .12, p ⬍ .05). Finally, avoidant attach-
ment was associated with feeling more indifferent across scenarios
(B ⫽ .13, sr ⫽ .13, p ⬍ .05) and, interestingly, was also associated
with feeling more uneasy across scenarios (B ⫽ .22, sr ⫽ .22, p ⫽
.001). Moreover, results revealed a significant Scenario ⫻ Avoid-
ance interaction on happiness ratings (B ⫽ ⫺.14, sr ⫽ ⫺.14, p ⬍
.05). As predicted, avoidant attachment was negatively associated
with happiness ratings in the notes scenario, r(125) ⫽ ⫺.21, p ⬍
.05, in which participants received help from the other, but was not
associated with happiness ratings in the study partner scenario,
r(121) ⫽ .04, ns.
Discussion
As predicted, overall, participants felt the suggestion to be study
partners and exchange phone numbers was more significant for
relationship development than the potential close other’s offer to
lend class notes; however, attachment moderated these effects.
Anxious individuals, to a greater extent than their more secure
counterparts, believed the event had significance for relationship
development and were more likely to make relational attributions
for the potential close other’s communal behavior regardless of
scenario. Even in the notes scenario, which involved a relatively
mundane, trivial communal behavior, anxious individuals were
more likely to detect evidence of caring and to use that behavior to
diagnose relationship potential. Finally, avoidant individuals were
less likely to make relationship attributions for the potential close
other’s behavior, reflecting their desire to minimize the importance
of closeness and their more pessimistic expectations about others.
Interestingly enough, when we decomposed the relational attri-
bution composite and investigated relational and situational attri-
butions separately, although the pattern of results was the same,
the Scenario ⫻ Attachment Anxiety interaction only attained sta-
tistical significance for the situational attribution. This suggests
that anxious individuals were somewhat more willing than others
to make positive relational attributions for the other’s offer to lend
them notes but that they were especially unwilling to discount the
behavior as purely situational. That is, they appear to have been
clinging to the notion that there must be some meaning in the
gesture but at the same time were hesitant to conclude that the
other was definitely interested in a relationship. These findings are
consistent with the ambivalence that characterizes attachment anx-
iety and point to a possible source for the uncertainty anxious
individuals evidenced in Study 2 in response to possible closeness.
These findings are also consistent with Lydon et al.’s (1997)
research, which found that when participants were faced with an
interdependence dilemma, they were especially likely to believe
that communal gestures had some implication for relationship
development.
With respect to affect, overall, participants felt less indifferent in
the study partner scenario suggesting that they perceived the offer
to be study partners as having greater significance for the relation-
ship. It is interesting to note that participants also felt more uneasy
when the potential close other suggested being study partners. This
finding is in agreement with Lydon et al. (1997), and we believe it
is indicative of the feelings of uncertainty and anxiety people
experience in association with the interdependence dilemma.
Avoidant individuals, characteristically, reported feeling more in-
different to the potential close other’s communal behavior across
scenarios; they also felt more uneasy about the potential close
other’s communal behavior across scenarios. This indifference is
consistent with avoidant individuals’ tendency to downplay the
importance of closeness, but their unease suggests that avoidant
individuals are nonetheless uncomfortable with situations promot-
ing closeness. Finally, avoidance was negatively associated with
happiness in the notes scenario. This decreased happiness in re-
sponse to a potential close other’s helping behavior is consistent
with the avoidant preference for self-reliance.
How do these findings illuminate Study 2? The anxiously at-
tached assigned greater meaning and significance to possible com-
munal overtures. They interpreted even a relatively trivial com-
munal behavior as a sign of potential closeness. This is consistent
with Holmes’s (1991) predictions about chronic uncertainty in the
context of the interdependence dilemma. Although everyone is
thought to be susceptible to interdependence dilemma concerns—
indeed, secure individuals also read meaning into the potential
close other’s behavior in this study—it appears that anxiously
attached individuals, who are chronically uncertain, were more
susceptible to interdependence concerns as evidenced by their
tendency to hopefully latch onto even relatively trivial communal
gestures. Unfortunately, as noted earlier, this microanalytic pro-
cess may actually fuel the interpersonal anxiety anxious individ-
uals are trying to regulate. The findings from Study 3 also shed
light on the avoidant individuals in Study 2. The findings from
Study 3 suggest that avoidant individuals have a strong desire to
downplay the importance of closeness and that this may have
contributed to their negative partner evaluations in Study 2. Fi-
nally, it is important to note that—at least in Study 3—avoidant
individuals were not significantly less interested in developing a
relationship with their chosen person than were their more secure
or anxious counterparts, thus reducing the likelihood that the
avoidants’ response to the potential close other’s communal be-
 NAVIGATING THE INTERDEPENDENCE DILEMMA
havior was simply due to insufficient interest in developing a
relationship.
Study 4
In Study 1, we found that when interacting with a potential close
other, anxious individuals went out of their way to signal interest
in closeness. In Study 2, we found that ironically, when a potential
close other signaled interest in closeness, attachment anxiety was
associated with more interpersonal anxiety than when the potential
close other acted more neutrally. The findings from Study 3
suggest that situations involving possible closeness elicit a great
deal of uncertainty and that anxiously attached individuals may go
into a mode of uncertainty reduction, looking for evidence of the
other’s interest and caring, even in more trivial communal behav-
iors. Taken together, Studies 1, 2, and 3 suggest that in potential
relationship development situations, anxiously attached individu-
als are especially susceptible to interdependence dilemma con-
cerns. What, however, are the implications of these interdepen-
dence dilemma concerns? A central prediction of attachment
theory is the notion of a secure base. When attachment needs are
met, people are free to engage in non-attachment-related activities
(e.g., exploration); however, when the attachment system is chron-
ically engaged in the pursuit of attachment goals (as is often the
case with those who feel anxiously attached), fewer resources are
available for non-attachment-related activities (Mikulincer &
Shaver, 2003). The goal of Study 4 was to investigate the conse-
quences of regulating concerns about closeness for anxious indi-
viduals’ personal functioning.
In addition, although Study 3 investigated the kind of inferences
and attributions people make about a potential close other’s com-
munal behavior, it was limited because of the self-report method-
ology used and because participants did not interact with a real
person. Thus, in Study 4, we again explored how participants
construed another’s communal behavior; however, rather than
asking participants explicitly about their perceptions, we decided
to use an implicit measure. Not only do implicit measures avoid
drawing attention to specific research questions but also they
circumvent participants’ self-presentation motives. Thus, in this
study, we used a lexical decision task to assess the activation of
attachment themes related to proximity– closeness and distance–
rejection. Because the experimental manipulation concerned the
possibility of closeness, it was hypothesized that closeness themes
would primarily be activated; however, we also wanted to explore
the possible activation of rejection concerns. A final goal of this
study was to provide further evidence— using a different opera-
tionalization of the confederate’s use of communal norms—that
situations affording the possibility of closeness, compared with
situations in which closeness is less likely, are distressing to
anxiously attached individuals.
To meet these goals, we returned to the laboratory methodology
and had participants again interact with an opposite-sex confeder-
ate in a study supposedly investigating cognitive abilities and
distributive reasoning. Participants were informed that they would
be performing some cognitive tasks (which included a task of
mental concentration and a lexical decision task) and then answer-
ing some general knowledge (Trivial Pursuit; Hasbro, Inc., Paw-
tucket, RI) questions. In the communal condition, participants
89
were given the option to work as a team, and the confederate
reached out to the participant and suggested working together for
the Trivial Pursuit portion of the study. In the baseline perfor-
mance-anxiety condition, participants were not given the option to
work as a team but were simply informed that they would have to
answer the Trivial Pursuit questions individually (hence, perfor-
mance anxiety). We hypothesized that anxiously attached individ-
uals would respond to the prospect of closeness in the communal
condition with increased rumination and, consequently, poorer
performance on the mental concentration task. By contrast, we
predicted that secure individuals would not be distressed about the
prospect of closeness in the communal condition—in fact, we
thought that they might benefit from contact with a supportive
other, resulting in enhanced performance. Finally, although we
thought that avoidant individuals would likely be distressed about
the prospect of closeness in the communal condition, we did not
predict impaired performance on the mental concentration task
because their typical coping response is to distance themselves
from unpleasant thoughts.
In sum, all participants were expected to exhibit performance
deficits on the mental concentration task in the performance-
anxiety condition, but secure individuals’ performance was ex-
pected to be facilitated in the communal condition, whereas anx-
iously attached individuals’ performance was expected to be
impaired in the communal condition. Moreover, it was predicted
that secure and anxious individuals’ performance on the mental
concentration task in the communal condition would be differen-
tially associated with proximity accessibility, with proximity ac-
cessibility being associated with better performance for the secure
participants but worse performance for the anxious participants.
By comparison, avoidant individuals’ performance was not pre-
dicted to be affected by the communal condition because it was
thought they would likely keep in check or suppress concerns
about closeness.
Method
Participants
Sixty-nine male university students volunteered to participate in the
study (M age ⫽ 19.7 years). Again, single participants were targeted (54
single and 15 dating). Participants were randomly assigned to either the
communal (n ⫽ 32) or the performance-anxiety condition (n ⫽ 37) and
received $10 (Canadian) for their participation.
Procedure
As in the previous studies, the participant had a 2-min interaction with
an attractive, opposite-sex confederate, who was also supposedly a partic-
ipant. After the interaction period, the participant and confederate were
brought into the testing room. The experimenter explained that the study
was investigating cognitive abilities and distributive reasoning and that
first they would be working on some cognitive tasks individually to assess
mental concentration and then would be working on a Trivial Pursuit task.
After being given a brief description of the Trivial Pursuit task (see
Experimental Manipulation, below), the participant was escorted to a
separate room and was left alone to complete an informed-consent form
and a thought-listing task (unrelated to this study).
The experimenter returned after a few minutes with some personality
measures (including the ECR) and the group information sheet for the
90 BARTZ AND LYDON
participant to complete. Once the participant had finished these items, the
experimenter explained the d2 test (Brickenkamp, 1981), which was pre-
sented as a test of mental concentration. This test consists of a matrix of
rows containing random sequences of the letters d and p with one, two, or
no apostrophes above and/or below each letter; the objective is to go
through each row and cross out all the ds with two apostrophes. The
experimenter timed the participant as he worked on the d2 test. After
completing a short evaluation form (see d2 evaluation form, below), the
participant began the lexical decision task, which was also presented as a
test of mental concentration. Under the pretext of checking on the other
participant, the experimenter left so that the participant could work on the
lexical decision task alone (participants completed the d2 test and lexical
decision task in the absence of the confederate). The experimenter returned
shortly after to probe for suspicions, debrief, and compensate the
participant.
Experimental Manipulation
Communal condition. In this condition, the participant and the confed-
erate were informed that the second part of the study was investigating
distributive reasoning and that they would be working on a task in which
they would each be asked Trivial Pursuit questions. Moreover, they were
told that they had the option to work with their partner on this task.
Specifically, the experimenter said, “For this task, you have the option to
work with your partner; so, if you choose to work with your partner,
although each question will be directed at one of you, you will be able to
help each other.” The experimenter then paused, and the confederate
looked at the participant, smiled, and suggested working together as a team,
in this way expressing the desire to be communal. The experimenter
confirmed that they would be working as a team on the Trivial Pursuit task
(all participants agreed to work with the confederate) and asked if there
were any questions. Finally, as in Studies 1 and 2, in this condition, the
participant interacted with an attractive, opposite-sex partner (a confeder-
ate) who was presented as single and a recent transfer student, in this way
increasing the desirability and perceived likelihood of a communal rela-
tionship with the confederate.
Performance-anxiety control condition. Our objective in designing
this condition was to create a noncommunal condition—that is, an inter-
action in which the possibility of closeness was unlikely— however, we did
not want participants to feel rejected by their partner. Thus, the participant
and the confederate were informed that the second part of the study was
investigating general knowledge and would involve a Trivial Pursuit game.
In this condition, they were not given the option to work together; rather,
they were told that the experimenter would be asking them the Trivial
Pursuit questions individually and that each in turn would have the oppor-
tunity to answer the questions. Although the primary goal was to create a
noncommunal condition in which the confederate did not express the desire
to be communal by suggesting to work as a team, this meant having the
participant work individually on the Trivial Pursuit task and answer ques-
tions in front of the confederate and experimenter. As a result, this
condition was conceptualized as a performance-anxiety condition. Finally,
in this condition, we followed Clark’s (1984b) manipulation of an ex-
change relationship and presented the confederate as dating and not a
transfer student to decrease the likelihood that the participant would see her
as available for a communal relationship. The confederate was also slightly
less talkative during the hallway interaction (i.e., she did not initiate a
conversation, but if the participant asked her questions, she would respond,
so as to not appear rude).
Materials
ECR (Brennan et al., 1998). The 36-item questionnaire used in Studies
2 and 3 was used to assess attachment avoidance (␣ ⫽ .91) and anxiety
(␣ ⫽ .89).
The d2 mental concentration test (Brickenkamp, 1981). This test con-
sists of a matrix of 14 rows of random sequences of the letters d and p with
one, two, or no apostrophes above and/or below each letter. Participants
were instructed to cross out all the ds with two apostrophes and were given
15 s to go through each row (the experimenter timed participants with a
stopwatch). Participants were instructed to work as quickly and accurately
as possible. The d2 test is theorized to be an index of resistance against
interference and has been used to assess rumination (Kuhl, 1981). Perfor-
mance on the d2 test was calculated by summing the total number of hits
(i.e., the total number of items crossed out) and subtracting the total
number of mistakes (i.e., the number of non-d2s erroneously marked), in
this way taking into account both errors of omission (i.e., the number of d2s
missed) and errors of commission.
The d2 evaluation form. This questionnaire assessed participants’ ex-
perience during the d2 test. Of interest, participants rated on a scale from
0 (not at all) to 4 (extremely) how much attention they devoted to the task
and how much time they spent thinking about how well they were doing on
the task by placing a slash through a line from 0% to 100%. An attention–
distraction score was calculated by subtracting how much participants
reported they were thinking about how well they were doing from how
much attention they said they devoted to the d2 (higher numbers reflecting
greater attention).
Lexical decision task. This task was used to assess the cognitive
accessibility of proximity and distance–rejection themes. Participants were
presented with a series of letter strings on a PC and were instructed to judge
as quickly and as accurately as possible whether each letter string was a
word or nonword by pressing the appropriate key on the keyboard. Par-
ticipants were given 12 practice trials to familiarize themselves with the
task, and then they began the 87 experimental trials.
Each trial began with a star presented in the middle of the screen,
followed by the target stimulus after a pause of 500 ms. The target stimulus
lasted on the screen for 1,000 ms during which participants indicated
whether the target was a word or nonword. After they had indicated their
response, a blank screen followed for 500 ms, and then, the next trial
began. The target letter strings were taken from Mikulincer, Birnbaum,
Woddis, and Nachmias (2000) and consisted of 3 proximity words (close-
ness, love, and affection)6 and 6 distance–rejection words (separation,
rejection, abandonment, distance, loneliness, and alone). For exploratory
purposes, we also included 5 coping strategy words (intimacy, escape,
worry, security, and control), taken from Mikulincer (1998), which were
theorized to reflect the coping strategies associated with different attach-
ment orientations.7 In addition, 9 communion and 9 agency words were
included for pilot-testing purposes, along with 10 neutral words and 45
nonwords. The 10 neutral words had no positive or negative connotations
and no link to proximity or rejection themes (e.g., elephant, book, and
picture). The 45 nonwords were created by taking common English words
and changing one letter. All words and nonwords were matched for number
of letters, and trials were randomly ordered across participants. This task
was programmed using E-prime and run on a Dell Pentium 3 XPS T700r
PC.
6
Mikulincer et al. (2000) used six proximity words; however, during
pilot testing, three of these words (kiss, hug, and caress) were repeatedly
mentioned as curious and out of place given the context of the study; thus,
these words were dropped from the lexical decision task.
7
It was thought that worry and security might be more accessible for the
anxiously attached, whereas escape and control might be more accessible
for the avoidantly attached, and that this might interact with experimental
condition. However, no mean differences in reaction times (RTs) to these
words were found, nor were any correlations between these RTs and
mental concentration found as a function of attachment.
 NAVIGATING THE INTERDEPENDENCE DILEMMA 91

Manipulation check. After participants were debriefed, they rated the

extent to which they (a) liked their partner, (b) were attracted to their
partner, and (c) would like to be friends with their partner by placing a
slash through a line ranging from 0 (not at all) to 100 (very much).
Participants were given privacy to answer the questionnaire and were
assured that their responses would remain anonymous and confidential. A
composite manipulation check measure was created by averaging these
three items (␣ ⫽ .71).

Results
Manipulation Check
The independent samples t test investigating the manipulation
check composite revealed a significant difference between exper-
imental conditions, t(67) ⫽ 3.48, p ⫽ .001. Supporting the exper-
imental manipulation, those in the communal condition liked their
partner more, were more attracted to their partner, and were more
interested in a relationship with their partner, M(32) ⫽ 69.55, than
those in the performance-anxiety condition, M(37) ⫽ 58.62. The
interactive effects of experimental condition and attachment on
attraction to partner were also investigated in an analysis of vari-
ance (ANOVA) after splitting the attachment dimension scores at
the median. The ANOVA revealed a three-way interaction be-
tween condition, attachment anxiety, and attachment avoidance,
F(1, 61) ⫽ 4.63, p ⬍ .05. Focused comparisons showed that low
anxious–low avoidant participants and high anxious– high
avoidant participants were more attracted to the confederate in the
communal condition, M(7) ⫽ 77.19 and M(9) ⫽ 70.30, respec-
tively, than in the performance-anxiety condition, M(12) ⫽ 55.08
and M(11) ⫽ 57.52, respectively, t(61) ⫽ 3.58, p ⬍ .001, and
t(61) ⫽ 2.19, p ⬍ .05, whereas there were no between-condition
differences in partner attraction for low anxious– high avoidant
participants or for high anxious–low avoidant participants. There
were no attachment group differences within the communal and
performance-anxiety conditions.
Mental Concentration–Rumination
To investigate the influence of condition and attachment on
mental concentration–rumination a 2 (condition: communal vs.
performance anxiety) ⫻ 2 (attachment anxiety: low vs. high) ⫻ 2
(attachment avoidance: low vs. high) ANOVA was conducted on
participants’ d2 scores (i.e., the sum of d2 hits minus mistakes).8
The ANOVA investigating the effects of experimental condition
and attachment anxiety and avoidance on d2 performance yielded
a main effect for condition, F(1, 58) ⫽ 4.70, p ⬍ .05. Overall,
performance was worse in the performance-anxiety condition,
M(37) ⫽ 183.36, than in the communal condition, M(29) ⫽
200.66; however, this effect was qualified by an interaction be-
tween condition and attachment anxiety, F(1, 58) ⫽ 4.49, p ⬍ .05,
but not by an interaction with avoidance (F ⬍ 1.5).
As depicted in Figure 2, participants low in attachment anxiety
in the communal condition performed significantly better,
M(13) ⫽ 212.67, than their low anxious counterparts in the per-
formance-anxiety condition, M(20) ⫽ 178.48, t(58) ⫽ 3.05, p ⬍
.01, and, importantly, also performed significantly better than their
high anxious counterparts in the communal condition, M(16) ⫽
188.64, t(58) ⫽ 2.05, p ⬍ .05. There was no difference between
Figure 2. Mental concentration (d2 test performance) as a function of
experimental condition and attachment anxiety (Study 4).
the low and high anxious participants in the performance-anxiety
condition, nor was there a difference between conditions for the
high anxious participants (both ts ⬍ 1.5). Thus, as predicted,
overall, participants were less able to concentrate in the perfor-
mance-anxiety condition, suggesting their preoccupation with an-
swering the Trivial Pursuit questions in front of their partner.
However, whereas the communal condition facilitated perfor-
mance for the more secure participants, it impaired performance
for the more anxiously attached.
Self-Reported Attention–Distraction
The ANOVA investigating the effects of condition and attach-
ment on self-reported attention– distraction yielded a marginally
significant interaction between condition and attachment anxiety,
F(1, 61) ⫽ 3.32, p ⫽ .073. In the communal condition, high
anxious participants reported being less attentive and more dis-
tracted, M(18) ⫽ 29.39, than their less anxious counterparts,
M(14) ⫽ 62.36, t(61) ⫽ 2.04, p ⬍ .05. There were no other group
differences (ts ⬍ 1.5).
Lexical Decision Analyses9
Drawing on Greenwald, McGhee, and Schwartz (1998), we
recoded reaction times (RTs) on the lexical decision task less than
300 ms as 300 ms and RTs greater than 3,000 ms as 3,000 ms.
8
Three participants with errors-of-commission scores (mistakes) greater
than 60 (i.e., more than 4 standard deviations above the group mean of
5.75) were excluded from the d2 analyses because they most likely mis-
understood the instructions to the d2 test. The fourth highest participant had
an error-of-commission score of 17.
9
Although native English speakers were targeted, 13 participants re-
ported that English was their second language. One participant who rated
his English proficiency as moderate (3 on a scale from 1 to 5) was dropped
from the lexical decision analyses. The 12 remaining participants who
92 BARTZ AND LYDON
Overall, the error rates were low (the mean was 2.7%).10 RTs for
words with incorrect responses were replaced with the mean RT
for that word. Error rates were not related to experimental condi-
tion or attachment. The mean RT across all conditions was 740.96
ms (SD ⫽ 146.84).
Within-cell correlations were calculated to investigate the asso-
ciation between the accessibility of proximity or distance–rejection
and d2 performance. Specifically, standardized residuals of the
proximity and distance–rejection RTs were created using control
word RT as the predictor (negative numbers reflecting quicker RTs
and increased accessibility). Participants’ d2 scores were also
standardized (negative numbers reflecting poorer performance).
The cross-products of these standardized scores were taken (e.g.,
the product of proximity accessibility and d2 score). We then
conducted 2 (condition: communal vs. performance anxiety) ⫻ 2
(attachment anxiety: low vs. high) ⫻ 2 (attachment avoidance: low
vs. high) ANOVAs on the within-cell correlations between (a)
Proximity RT ⫻ d2 Performance and (b) Distance–Rejection
RT ⫻ d2 Performance.11
The ANOVA investigating the correlation between proximity
accessibility and d2 performance yielded a significant interaction
between condition and attachment anxiety, F(1, 57) ⫽ 4.49, p ⬍
.05. In the communal condition, proximity accessibility was asso-
ciated with better d2 performance for low anxious individuals,
M(13) ⫽ ⫺.69, but proximity accessibility was associated with
poorer d2 performance for high anxious individuals, M(15) ⫽ .47,
t(57) ⫽ 2.37, p ⬍ .05. Thus, in the communal condition, proximity
accessibility facilitated performance for the less anxiously attached
but impaired performance for the more anxiously attached.
In addition, a second interaction was found between attachment
anxiety and avoidance independent of experimental condition, F(1,
57) ⫽ 4.29, p ⬍ .05. Among those high in anxiety, proximity
accessibility was associated with increased performance for those
low in avoidance, M(15) ⫽ ⫺.46, but was associated with decreased
performance for those high in avoidance, M(17) ⫽ .76, t(57) ⫽ 2.68,
p ⬍ .01. Finally, there were no significant main or interaction effects
of condition and/or attachment on the cross-product of distance–
rejection accessibility and d2 performance (all Fs ⬍ 2.1).
Discussion
As predicted, performance on the mental concentration task was
impaired for everyone in the performance-anxiety condition, but
anxiously attached individuals also performed poorly in the com-
munal condition. Secure participants, on the other hand, appeared
to benefit from the communal condition—performing better than
their anxious counterparts in the communal condition and better
than their secure counterparts in the performance-anxiety condi-
tion. What was it about the communal condition that helped the
secures but hurt the anxiously attached participants? Analyses
looking at the association between d2 performance and proximity
accessibility suggest that whereas thoughts about closeness facil-
itated the secure individuals’ performance, these thoughts under-
mined the anxious individuals’ performance. These findings high-
reported that English was not their native language did not differ from the
native English speakers on error rates (t ⬍ 1) or RT (t ⬍ 1).
light the importance of a secure base: When secure participants
interacted with someone who expressed liking and acceptance, as
opposed to someone who was simply neutral, they went on to excel
at the mental concentration task. Unfortunately, the potential for
closeness did not have the same effect on their anxious counterparts.
Anxious individuals ruminated even in response to a relatively posi-
tive interpersonal event—that is, an attractive, available other express-
ing interest in them. Finally, situations involving possible closeness
did not trigger concerns about rejection for the anxiously attached, a
point to which we return in the General Discussion.
Avoidant individuals did not exhibit performance deficits in the
communal condition. Given that a core feature of avoidant attachment
is discomfort when others try to get too close and that avoidant
individuals liked their partner less when their partner worked as a unit
in Study 2, it is likely that the confederate’s suggestion to work as a
team would have been somewhat distressing to the avoidant partici-
pants. However, avoidant attachment is associated with a deactivation
of the attachment system, and avoidant individuals tend to cope with
distress by suppressing their concerns, thus, their distress in the
communal condition may not have compromised their performance
on the mental concentration task.
General Discussion
Our goal in conducting this research was to investigate whether
attachment is related to people’s ability to tolerate the uncertainty
inherent in the early phases of relationship development. People
who hope to establish a relationship try to follow the communal
script to signal interest, but the desire to express interest is tem-
pered by their uncertainty about the prospect for a relationship and
the other’s motives. Because differences in attachment reflect
beliefs about whether the self is worthy of affection and whether
others are trustworthy and reliable, it was theorized that attachment
would be related to the ability to tolerate the interdependence
dilemma. Overall, our predictions have been supported.
Summary and Discussion of Findings
When anxiously attached individuals interacted with an attrac-
tive, desirable, opposite-sex partner, they obscured individual con-
tributions to a group task. Drawing on Clark’s (1984b) theorizing,
we believe they did this to signal their interest in closeness (i.e., by
obscuring individual contributions, they conveyed their preference
for a communal relationship over an exchange relationship). The
notion that anxiously attached individuals would go out of their
way to signal interest in a communal relationship with a potential
close other is consistent with their chronic desire for closeness and
acceptance. These results are also in accord with Mikulincer and
Nachshon’s (1991) research findings that anxiously attached indi-
viduals are especially prone to self-disclosure and tend to self-
10
Three real words and seven nonwords had exceptionally high error
rates (greater than 15%) and consequently were dropped from the analyses.
The real words that were dropped were not target words.
11
We also looked at the effects of experimental condition and attach-
ment on the accessibility of the proximity– closeness, distance–rejection,
and control strategy words. These analyses yielded one marginal effect.
There were no significant main effects or interactions.
 NAVIGATING THE INTERDEPENDENCE DILEMMA
disclose indiscriminatingly to strangers. Indeed, these researchers
proposed that anxiously attached individuals use self-disclosure
opportunities to merge with others. In the same way, anxiously
attached individuals may have obscured contributions in the cur-
rent investigation to symbolically merge with their partner. Yet
were anxiously attached individuals obscuring contributions to
avoid their partner’s negative evaluation should they perform
poorly on the group task? The findings from Study 2 suggest this
explanation is unlikely. If anxious individuals were obscuring
contributions to avoid negative evaluation, they should have been
less anxious in the communal condition in Study 2 because their
performance was indistinguishable from their partner’s, but this
was not the case.
So, when interacting with an attractive, desirable, opposite-sex
partner, anxious individuals went out of their way to signal inter-
est. However, when such a partner signaled interest by following
the communal script, anxious individuals were not reassured. In
Study 2, attachment anxiety was associated with greater interper-
sonal anxiety (i.e., state feelings of anxiety, uncertainty, and con-
cerns about appearance) when a potential close other used com-
munal norms (behaved as a unit) compared with when the other
used exchange norms. In Study 4, anxiously attached individuals
performed poorly on an unrelated mental concentration task when
a potential close other expressed interest in closeness by suggest-
ing to work as a team; moreover, results from the lexical decision
task in Study 4 suggest that it was thoughts about closeness in
particular that undermined anxious individuals’ performance in the
communal condition. Given that securing closeness is a chronic
goal for anxiously attached individuals, why would they experi-
ence more interpersonal anxiety when a potential close other
expressed interest in a communal relationship?
We believe that anxious individuals felt especially vulnerable in
these potential communal situations and that this vulnerability led
to increased feelings of anxiety, uncertainty, and concerns about
how their partner would evaluate them. In Study 1, anxious indi-
viduals made greater overtures than their secure and avoidant
counterparts; as a result, they may have felt more anxious about
whether those overtures would be met with equal interest. That is,
by putting themselves on the line this way, they had more to lose.
On a related note, Vorauer, Cameron, Holmes, and Pearce (2003)
found that anxious individuals tend to imbue their overtures to a
potential romantic partner with increased significance and mean-
ing. Even when they do not objectively make greater overtures,
anxious individuals think they have communicated more interest
because they believe others will take into account their inhibitions
when judging their behavioral intentions. Thus, in the present
investigation, when anxious individuals did make greater over-
tures, it is likely they felt especially vulnerable.
At the heart of the matter, we believe that situations involving
possible closeness amplify anxious individuals’ interpersonal am-
bivalence and elicit an approach–avoidance conflict leading to
heightened feelings of interpersonal anxiety. Study 1 suggests that
anxious individuals were looking for closeness in these encounters,
and this may have lead to an internal dialectic between their
deepest hopes and fears (Holmes, 1991). On the one hand, their
desire for acceptance should have led them to look for evidence of
caring and to infer meaning from their partner’s behavior—indeed,
Study 3 suggests this was the case. In Study 3, anxiously attached
93
individuals inferred more meaning and were less likely to discount
even relatively trivial communal behaviors as purely situational.
On the other hand, it is likely that these judgments were guarded
by feelings of vulnerability and the acute possibility that they may
ultimately be found unworthy. As Holmes (1991) stated, “the
specter of things going well reminds [uncertain] people of the costs
of their being wrong” (p. 84). This idea is echoed in the attachment
literature as well. For example, Mikulincer and Shaver (2003)
speculated that for anxiously attached individuals, positive states
may quickly bring to mind other instances in which things began
well but ended badly.
Although Study 2 suggests that anxious individuals were con-
cerned about evaluation, Study 4 did not find evidence that anx-
iously attached individuals were specifically concerned about re-
jection, so how and when might concerns about rejection come
into play? Holmes (1991) described the interdependence dilemma
as a dialectic that unfolds over time. People signal their interest
and then look for signs of interest and commitment in the other’s
behavior; this back and forth continues as the relationship unfolds
(or not, as the case may be). This research focused on the very
early stages of relationship development, a point at which anxious
individuals may have been quite hopeful. Perhaps research that
focuses on interactions occurring later in relationship development
would yield increased concerns about rejection because there is
greater investment and, consequently, greater risk. Anxiously at-
tached individuals’ mounting interpersonal anxiety may eventually
bias their perceptions in a negative direction. It may also be that in
the same way that anxious individuals magnify relatively trivial
positive behaviors in this process of uncertainty reduction, they
also magnify relatively trivial negative behaviors. This notion is
consistent with Fraley and Shaver’s (2000) conceptualization of
the anxiety dimension of attachment as being critically involved in
the appraisal and monitoring of attachment related cues and the
idea that individuals who are high in attachment anxiety have a
low threshold for detecting cues of rejection and acceptance.
Indeed, Campbell, Simpson, Boldry, and Kashy (2005) found
support for this idea in a recent diary study in which it was found
that anxious individuals tended to magnify both positive and
negative relationship-relevant behaviors in their established rela-
tionships. Our research adds to this literature by showing that
anxious individuals’ vigilant monitoring of attachment-related be-
haviors occurs even in the context of potential new relationships.
In contrast to their more anxious counterparts, secure partici-
pants did not actively avoid using exchange norms when interact-
ing with an attractive, available, opposite-sex partner, but impor-
tantly, they also did not use exchange norms (i.e., their pen choice
was random). Secure participants also felt comfortable with the
potential close other’s use of communal norms. They were not
distressed when their partner expressed the desire to be a unit
(Studies 2 and 4), and in fact, they actually benefited from the
communal condition in Study 4. Were secure participants simply
not interested in a relationship with their partner? Because they
were the most likely group to have a strong network of friends,
secure participants may not have been especially motivated to
form a new relationship. However, we believe this explanation is
unlikely for a few reasons. First, in the current studies, we tried to
create situations in which the potential for a close relationship was
desirable and feasible. Drawing on Clark (1984b, 1986), we tar-
94 BARTZ AND LYDON

geted single participants and paired them with an attractive,
opposite-sex confederate who was single and a recent transfer
student. So, even if secure participants had several close friends,
they were still not involved in a romantic relationship and thus
should have been interested in the confederate. Second, secure
participants did not differ from their anxious counterparts on
partner liking (Studies 2 and 4) or interest in developing a rela-
tionship (Study 3). Finally, thoughts about closeness, love, and
affection were associated with a dampening of rumination in Study
4. If secure participants were not at least somewhat interested, they
should not have benefited from the potential for closeness in this
study.
Rather than secure individuals being disinterested in their part-
ner, we believe secure individuals’ self-esteem, trust in others, and
previous successful close relationship experiences account for
their behavior in the present investigations. In contrast to their
more anxious counterparts, in Study 1, secure individuals felt
comfortable letting the situation unfold without a great deal of
effort on their part. Indeed, secure individuals may have been more
like the existing friends (whose pen choice was also random) in
Clark’s (1984b) original study. Clark (1984b) theorized that pen
choice was random for existing friends because existing friends
know they are friends, so there is no need to send a message
conveying interest in closeness. In the current investigation, secure
participants may have felt confident that the relationship would
develop and, as a result, did not feel the need to go out of their way
to make it happen. The findings from Study 3 also support this
idea. Secure participants were less likely than their anxious coun-
terparts to invest their potential close other’s discrete communal
behavior with increased significance for relationship development,
suggesting that although they may have been interested in devel-
oping a relationship, they were not preoccupied with evaluating the
meaning of discrete events and using that information to infer the
other’s motives and to diagnose relationship potential.
Another possible factor distinguishing the secure participants
from their more anxious counterparts may have been the goals that
were activated in our experimental situations. In a study looking at
attachment and the pursuit of attachment and affiliation goals,
Mikulincer and Selinger (2001) found that secure adolescents
pursued attachment goals in attachment-appropriate contexts and
affiliation goals in affiliation-appropriate contexts, whereas anx-
ious adolescents focused exclusively on attachment goals and
tended to pursue attachment goals in attachment and affiliation
contexts. In the current studies, secure participants may have
viewed these situations as opportunities to pursue affiliation goals,
whereas anxious participants may have viewed them as opportu-
nities to pursue attachment goals.
Finally, contrary to our predictions, when avoidant individuals
interacted with a potential close other, overall, they did not use
exchange norms. Perhaps avoidant individuals only feel the need
to establish boundaries when they perceive that others are trying to
get too close. Indeed, the findings from Study 2 suggest this may
be the case. Compared with when their partner used exchange
norms, when their partner tried to work as a unit (i.e., used
communal norms), avoidant individuals liked their partner less.
Perhaps avoidant individuals felt exploited when their partner used
the same color pen; by obscuring how much each person contrib-
uted to the task, it would be easier for the confederate to freeload.
This hypothesis would be consistent with avoidant individuals’
pessimistic beliefs about others; however, if this were the case, one
would think avoidant individuals would have ascribed more neg-
ative agency traits (e.g., greedy) and not more negative communal
traits (e.g., servile) to their partner in the communal condition.
Avoidant individuals strive for independence, tend to downplay
the importance of closeness, and use avoidance strategies to cope
with concerns. It may be that when avoidant individuals enter
situations involving possible closeness (situations that, in theory,
should be distressing), they attempt to suppress the possibility of
closeness. Consistent with this idea, in Study 3, avoidant individ-
uals downplayed relational motives for the potential close other’s
communal behavior. Moreover, although they professed to be
more indifferent to the potential close other’s communal behavior
in Study 3, avoidant attachment was associated with feeling uneasy
in Studies 2 and 3, suggesting that their professed indifference may
have been a façade. Indeed, this notion is in line with research
suggesting that avoidant individuals’ defensive strategies may
actually be quite fragile (Mikulincer, Dolev, & Shaver, 2004).
Research Strengths and Limitations
Three of the four studies presented had participants interact with
a real person (a confederate), and behavioral responses were as-
sessed; nevertheless, there were limitations to this approach. Stud-
ies 1 and 2 did not assess how participants were construing their
behavior or the behavior of the confederate, mainly because of
concerns about drawing attention to the research questions and
possibly undermining participants’ behavioral responses. Were
participants intending to convey interest by obscuring contribu-
tions? Did they perceive their partner’s communal behavior as an
expression of interest? Study 3 shed some light on the kind of
attributions participants were making and what meaning they saw
in the potential close other’s communal behavior. That said, we
believe that although people use communal and exchange behav-
iors to convey how they feel about closeness, this may often occur
at an implicit level; thus, explicitly asking participants about their
intentions may provide limited information. The lexical decision
task used in Study 4 is one way to examine the implicit cognitions
and motivations that guide these interactions.
In addition, in our attempt to create a noncommunal condition
that paralleled the communal condition in Study 4, concerns about
performance were aroused, and anxious participants performed
poorly in both conditions. Would they have performed poorly if
they had simply been asked to do the d2 test without interacting
with another participant? Although anxious participants did not
perform more poorly than the secures in the performance-anxiety
condition, it is possible that anxious individuals are generally poor
at mental concentration tasks, and future research should be con-
ducted to explore this possibility. Nonetheless, we believe the
important finding from Study 4 is that unlike their more secure
counterparts, anxious participants did not benefit from the com-
munal condition—for them, thoughts about closeness, love, and
affection did not dampen performance anxiety and rumination.
Theoretical Contributions and Future Directions
A major proposition of attachment theory is that attachment
models are carried forward into new relationships, guiding percep-
 NAVIGATING THE INTERDEPENDENCE DILEMMA
tions, expectations, and behaviors with new partners (Collins &
Read, 1994); however, little is known about how the attachment
system operates during the early phases of relationship develop-
ment. This research establishes the importance of attachment at the
outset of a relationship and points to some circumstances under
which attachment models should come into play. Attachment
models appear to be influential when little is known about what
can be expected from the other person (Collins & Read, 1994;
Pierce & Lydon, 2001; Pietromonaco & Feldman Barrett, 2000).
The match between the situation and chronic attachment goals
related to security attainment also appears to be important. Finally,
behaviors related to the exchange of social commodities that are
used to define the nature of the relationship appear to be associated
with attachment.
One question arising from the current research is whether these
findings apply to all close relationships, friendships, or romantic
relationships in particular. To increase the chances that participants
would desire a relationship with the confederate, the possibility of
a romantic relationship was made salient. That said, as noted
earlier, some participants may have thought of the confederate as
a potential romantic partner, whereas others may have thought of
the confederate as a friend. Study 3 spoke to this issue. Participants
were allowed to nominate either a potential friend or potential
romantic partner, and relationship type did not qualify the results,
suggesting that the present findings would apply to close relation-
ships in general. It is worth noting that participants were more
interested in developing a romantic relationship in Study 3. More-
over, adult attachment relationships tend to be romantic (Fraley &
Shaver, 2000; Hazan & Zeifman, 1999). Given this, we believe the
present findings should be especially relevant to potential romantic
relationships, but may also apply to close friendships, depending
on the level of interest and the extent to which the other is
perceived as a potential attachment figure.
Is attachment associated with the use of communal and ex-
change norms in established relationships? The early stages of a
relationship are fraught with uncertainty, but in existing relation-
ships, people presumably have established a certain level of trust
and are more knowledgeable about their partner’s intentions and
level of commitment. That said, we believe attachment would
similarly be associated with the use of communal and exchange
norms in established relationships to the extent that satisfying
chronic attachment goals continues to be important. Uncertainty
may also be a critical factor. Perhaps if anxiously attached indi-
viduals felt particularly uncertain—for example, when making a
major personal or relational decision—they may be especially
likely to monitor their partner’s discrete behaviors for reassurance.
Concluding Comments
In conclusion, the early stages of a relationship are marked by
uncertainty and an interdependence dilemma in which people must
wager on the motives of the other person and decide whether being
communal is worth the risk. They must also tailor their own
behavior to communicate interest but not appear overly concerned
with social transactions. Although people share the same script for
close relationships, the belief structures and goals associated with
chronic attachment orientations provide direction on how that
script will be played out as the relationship unfolds.
95
References
Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles among
young adults: A test of a four-category model. Journal of Personality
and Social Psychology, 61, 226 –244.
Baumeister, R. F., & Leary, M. R. (1995). The need to belong: Desire for
interpersonal attachments as a fundamental human motivation. Psycho-
logical Bulletin, 117, 497–529.
Bowlby, J. (1969). Attachment and loss: Vol. 1. Attachment. New York:
Basic Books.
Bowlby, J. (1973). Attachment and loss: Vol. 2. Separation: Anxiety and
anger. New York: Basic Books.
Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report measure-
ment of adult attachment: An integrative overview. In J. A. Simpson &
W. S. Rholes (Eds.), Attachment theory and close relationships (pp.
46 –76). New York: Guilford Press.
Brickenkamp, R. (1981). Test d2 (7th ed.). Göttingen, Germany: Hogrefe.
Campbell, L., Simpson, J. A., Boldry, J., & Kashy, D. A. (2005). Percep-
tions of conflict and support in romantic relationships: The role of
attachment anxiety. Journal of Personality and Social Psychology, 88,
510 –531.
Clark, M. S. (1984a). A distinction between two types of relationships and
its implications for development. In J. C. Masters & K. Yarkin-Levin
(Eds.), Boundary areas in social and developmental psychology (pp.
241–270). Orlando, FL: Academic Press.
Clark, M. S. (1984b). Record keeping in two types of relationships. Journal
of Personality and Social Psychology, 47, 549 –557.
Clark, M. S. (1986). Evidence for the effectiveness of manipulations of
communal and exchange relationships. Personality and Social Psychol-
ogy Bulletin, 12, 414 – 425.
Clark, M. S., & Mills, J. (1979). Interpersonal attraction in exchange and
communal relationships. Journal of Personality and Social Psychology,
37, 12–24.
Clark, M. S., Mills, J., & Powell, M. C. (1986). Keeping track of needs in
communal and exchange relationships. Journal of Personality and So-
cial Psychology, 51, 333–338.
Cohen, J., Cohen, P., West, S., & Aiken, L. (2003). Applied multiple
regression/correlation analysis for the behavioural sciences (3rd ed.).
Hillsdale, NJ: Erlbaum.
Collins, N. L., & Read, S. J. (1994). Cognitive representations of attach-
ment: The structure and functions of working models. In K. Bar-
tholomew & D. Perlman (Eds.), Advances in personal relationships
(Vol. 5, pp. 53–90). London: Jessica Kingsley.
Davilia, J., Burge, D., & Hammen, C. (1997). Why does attachment style
change? Journal of Personality and Social Psychology, 73, 826 – 838.
Fraley, R. C., & Shaver, P. R. (1997). Adult attachment and the suppres-
sion of unwanted thoughts. Journal of Personality and Social Psychol-
ogy, 73, 1080 –1091.
Fraley, R. C., & Shaver, P. R. (2000). Adult romantic attachment: Theo-
retical developments, emerging controversies, and unanswered ques-
tions. Review of General Psychology, 4, 132–154.
Fraley, R. C., & Waller, N. G. (1998). Adult attachment patterns: A test of
the typographical model. In J. A. Simpson & W. S. Rholes (Eds.),
Attachment theory and close relationships (pp. 77–114). New York:
Guilford Press.
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring
individual differences in implicit cognition: The implicit association test.
Journal of Personality and Social Psychology, 74, 1464 –1480.
Hazan, C., & Shaver, P. R. (1987). Romantic love conceptualized as an
attachment process. Journal of Personality and Social Psychology, 52,
511–524.
Hazan, C., & Shaver, P. R. (1994). Attachment as an organizational
framework for research on close relationships. Psychological Inquiry, 5,
1–22.
96 BARTZ AND LYDON
Hazan, C., & Zeifman, D. (1999). Pair bonds as attachments: Evaluating
the evidence. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attach-
ment: Theory, research, and clinical applications (pp. 336 –354). New
York: Guilford Press.
Heatherton, T. F., & Polivy, J. (1991). Development and validation of a
scale for measuring state self-esteem. Journal of Personality and Social
Psychology, 60, 895–910.
Holmes, J. G. (1981). The exchange process in close relationships: Mi-
crobehavior and macromotives. In M. J. Lerner & S. C. Lerner (Eds.),
The justice motive in social behavior (pp. 261–284). New York: Plenum
Press.
Holmes, J. G. (1991). Trust and the appraisal process in close relationships.
In W. H. Jones & D. Perlman (Eds.), Advances in personal relationships
(Vol. 2, pp. 57–104). London: Jessica Kingsley.
House, J. S., Landis, K. R., & Umberson, D. (1988, July 29). Social
relationships and health. Science, 241, 540 –545.
Kiecolt-Glaser, J. K., & Newton, T. L. (2001). Marriage and health: His
and hers. Psychological Bulletin, 127, 472–503.
Kuhl, J. (1981). Motivational and functional helplessness: The moderating
effects of state versus action orientation. Journal of Personality and
Social Psychology, 40, 155–170.
Leary, M. (1990). Responses to social exclusion: Social anxiety, jealousy,
loneliness, depression, and low self-esteem. Journal of Social and Clin-
ical Psychology, 9, 221–229.
Lydon, J. E., Jamieson, D. W., & Holmes, J. G. (1997). The meaning of
social interactions in the transition from acquaintanceship to friendship.
Journal of Personality and Social Psychology, 73, 536 –548.
Lynch, J. J. (1979). The broken heart: The medical consequences of
loneliness. New York: Basic Books.
Maslow, A. H. (1962). Toward a psychology of being. New York: Van
Nostrand.
McNair, D. M., Lorr, M., & Droppleman, L. F. (1971). Profile of Mood
States. San Diego, CA: Educational and Industrial Testing Services.
Mikulincer, M. (1998). Attachment working models and the sense of trust:
An exploration of interaction goals and affect regulation. Journal of
Personality and Social Psychology, 74, 1209 –1224.
Mikulincer, M., Birnbaum, G., Woddis, D., & Nachmias, O. (2000). Stress
and accessibility of proximity-related thoughts: Exploring the normative
and intraindividual components of attachment theory. Journal of Per-
sonality and Social Psychology, 78, 509 –523.
Instructions to Authors
For Instructions to Authors, please visit www.apa.org/journals/psp and click on the “Instructions to
Authors” link in the Journal Info box on the right.
Mikulincer, M., Dolev, T., & Shaver, P. R. (2004). Attachment-related
strategies during thought suppression: Ironic rebounds and vulnerable
self-representations. Journal of Personality and Social Psychology, 87,
940 –956.
Mikulincer, M., & Nachshon, O. (1991). Attachment styles and patterns of
self-disclosure. Journal of Personality and Social Psychology, 61, 321–
331.
Mikulincer, M., & Selinger, M. (2001). The interplay between attachment
and affiliation systems in adolescents’ same-sex friendships: The role of
attachment style. Journal of Social and Personal Relationships, 18,
81–106.
Mikulincer, M., & Shaver, P. R. (2003). The attachment behavioral system
in adulthood: Activation, psychodynamics, and interpersonal processes.
In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol.
35, pp. 53–152). New York: Academic Press.
Pierce, T., & Lydon, J. E. (2001). Global and specific relational models in
the experience of social interactions. Journal of Personality and Social
Psychology, 80, 613– 631.
Pietromonaco, P. R., & Feldman Barrett, L. (2000). The internal working
models concept: What do we really know about the self in relation to
others? Review of General Psychology, 4, 155–175.
Simpson, J. A., & Rholes, W. S. (1998). Attachment in adulthood. In J. A.
Simpson & W. S. Rholes (Eds.), Attachment theory and close relation-
ships (pp. 3–21). New York: Guilford Press.
Spence, J. T., Helmreich, R. L., & Holahan, C. K. (1979). Negative and
positive components of psychological masculinity and femininity and
their relationships to self-reports of neurotic and acting out behaviors.
Journal of Personality and Social Psychology, 37, 1673–1682.
Sullivan, H. S. (1953). The interpersonal theory of psychiatry. New York:
Norton.
Trout, D. L. (1980). The role of social isolation in suicide. Suicide and
Life-Threatening Behavior, 10, 10 –23.
Vorauer, J. D., Cameron, J. J., Holmes, J. G., & Pearce, D. G. (2003).
Invisible overtures: Fears of rejection and the signal amplification bias.
Journal of Personality and Social Psychology, 84, 793– 812.
Received September 21, 2004
Revision received October 26, 2005
Accepted October 28, 2005 䡲
96 BARTZ AND LYDON
Hazan, C., & Zeifman, D. (1999). Pair bonds as attachments: Evaluating
the evidence. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attach-
ment: Theory, research, and clinical applications (pp. 336 –354). New
York: Guilford Press.
Heatherton, T. F., & Polivy, J. (1991). Development and validation of a
scale for measuring state self-esteem. Journal of Personality and Social
Psychology, 60, 895–910.
Holmes, J. G. (1981). The exchange process in close relationships: Mi-
crobehavior and macromotives. In M. J. Lerner & S. C. Lerner (Eds.),
The justice motive in social behavior (pp. 261–284). New York: Plenum
Press.
Holmes, J. G. (1991). Trust and the appraisal process in close relationships.
In W. H. Jones & D. Perlman (Eds.), Advances in personal relationships
(Vol. 2, pp. 57–104). London: Jessica Kingsley.
House, J. S., Landis, K. R., & Umberson, D. (1988, July 29). Social
relationships and health. Science, 241, 540 –545.
Kiecolt-Glaser, J. K., & Newton, T. L. (2001). Marriage and health: His
and hers. Psychological Bulletin, 127, 472–503.
Kuhl, J. (1981). Motivational and functional helplessness: The moderating
effects of state versus action orientation. Journal of Personality and
Social Psychology, 40, 155–170.
Leary, M. (1990). Responses to social exclusion: Social anxiety, jealousy,
loneliness, depression, and low self-esteem. Journal of Social and Clin-
ical Psychology, 9, 221–229.
Lydon, J. E., Jamieson, D. W., & Holmes, J. G. (1997). The meaning of
social interactions in the transition from acquaintanceship to friendship.
Journal of Personality and Social Psychology, 73, 536 –548.
Lynch, J. J. (1979). The broken heart: The medical consequences of
loneliness. New York: Basic Books.
Maslow, A. H. (1962). Toward a psychology of being. New York: Van
Nostrand.
McNair, D. M., Lorr, M., & Droppleman, L. F. (1971). Profile of Mood
States. San Diego, CA: Educational and Industrial Testing Services.
Mikulincer, M. (1998). Attachment working models and the sense of trust:
An exploration of interaction goals and affect regulation. Journal of
Personality and Social Psychology, 74, 1209 –1224.
Mikulincer, M., Birnbaum, G., Woddis, D., & Nachmias, O. (2000). Stress
and accessibility of proximity-related thoughts: Exploring the normative
and intraindividual components of attachment theory. Journal of Per-
sonality and Social Psychology, 78, 509 –523.
Instructions to Authors
For Instructions to Authors, please visit www.apa.org/journals/psp and click on the “Instructions to
Authors” link in the Journal Info box on the right.
Mikulincer, M., Dolev, T., & Shaver, P. R. (2004). Attachment-related
strategies during thought suppression: Ironic rebounds and vulnerable
self-representations. Journal of Personality and Social Psychology, 87,
940 –956.
Mikulincer, M., & Nachshon, O. (1991). Attachment styles and patterns of
self-disclosure. Journal of Personality and Social Psychology, 61, 321–
331.
Mikulincer, M., & Selinger, M. (2001). The interplay between attachment
and affiliation systems in adolescents’ same-sex friendships: The role of
attachment style. Journal of Social and Personal Relationships, 18,
81–106.
Mikulincer, M., & Shaver, P. R. (2003). The attachment behavioral system
in adulthood: Activation, psychodynamics, and interpersonal processes.
In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol.
35, pp. 53–152). New York: Academic Press.
Pierce, T., & Lydon, J. E. (2001). Global and specific relational models in
the experience of social interactions. Journal of Personality and Social
Psychology, 80, 613– 631.
Pietromonaco, P. R., & Feldman Barrett, L. (2000). The internal working
models concept: What do we really know about the self in relation to
others? Review of General Psychology, 4, 155–175.
Simpson, J. A., & Rholes, W. S. (1998). Attachment in adulthood. In J. A.
Simpson & W. S. Rholes (Eds.), Attachment theory and close relation-
ships (pp. 3–21). New York: Guilford Press.
Spence, J. T., Helmreich, R. L., & Holahan, C. K. (1979). Negative and
positive components of psychological masculinity and femininity and
their relationships to self-reports of neurotic and acting out behaviors.
Journal of Personality and Social Psychology, 37, 1673–1682.
Sullivan, H. S. (1953). The interpersonal theory of psychiatry. New York:
Norton.
Trout, D. L. (1980). The role of social isolation in suicide. Suicide and
Life-Threatening Behavior, 10, 10 –23.
Vorauer, J. D., Cameron, J. J., Holmes, J. G., & Pearce, D. G. (2003).
Invisible overtures: Fears of rejection and the signal amplification bias.
Journal of Personality and Social Psychology, 84, 793– 812.
Received September 21, 2004
Revision received October 26, 2005
Accepted October 28, 2005 䡲
Journal of Personality and Social Psychology Copyright 2006 by the American Psychological Association
2006, Vol. 91, No. 1, 97–110 0022-3514/06/$12.00 DOI: 10.1037/0022-3514.91.1.97

Intergroup Helping as Status Relations: Effects of Status Stability,

Identification, and Type of Help on Receptivity to High-Status
Group’s Help

Arie Nadler Samer Halabi

Tel Aviv University University of Haifa

Integrating research on social identity processes and helping relations, the authors proposed that
low-status group members who are high identifiers will be unwilling to receive help from the high-status
group when status relations are perceived as unstable and help is dependency-oriented. The first
experiment, a minimal group experiment, found negative reactions to help from a high-status outgroup
when status relations were unstable. The 2nd and 3rd experiments, which used real groups of Israeli
Arabs and Israeli Jews, replicated this finding and showed that high identifiers were less receptive to help
from the high-status outgroup than low identifiers. The 4th experiment, a help-seeking experiment with
real groups of competing high schools, found that the least amount of help was sought from a high-status
group by high identifiers when status relations were perceived as unstable and help was dependency-
oriented. Theoretical and applied implications are discussed.

Keywords: reactions to help, help-seeking, perceived status stability, ingroup identification, dependency-/
autonomy-oriented help

In his classic essay The Gift, anthropologist Marcel Mauss
(1907/1957) described the tribal custom of potlatch, in which a
clan leader engages in lavish displays of gift-giving to other clan
leaders. Mauss noted that the “motives for such excessive gifts . . .
are in no way disinterested. . . . To give is to show one’s superi-
ority” (p. 72). This is group-based behavior in that tribal leaders
establish their hierarchical position “to the ultimate benefit . . . of
their own clan” (italics added, p. 4). For recipients of potlatch, “to
accept without returning . . . [is] to become client and subservient”
(p. 72). Phrased in the language of modern social psychology, this
suggests that helping relations can be mechanisms through which
groups create, maintain, and change status relations.
The present research examined the willingness of low-status
group members to seek and receive help from a high-status out-
group as affected by the perceived stability of intergroup status
relations, the ingroup identification of the recipient, and type of
help (i.e., autonomy- or dependency-oriented help). The research
hypotheses follow the model of intergroup helping as status rela-
tions (Nadler, 2002), which draws on social identity research
(Tajfel & Turner, 1986) and helping relations (Bierhoff, 2002;
Penner, Dovidio, Piliavin, & Schroeder, 2005; Schroeder, Penner,
Dovidio, & Piliavin, 1995). The present research represents the
first empirical effort to assess the major tenets of this model.
Arie Nadler, Department of Psychology, Tel Aviv University, Tel Aviv,
Israel; Samer Halabi, Department of Psychology, University of Haifa,
Haifa, Israel.
This research was supported by the Argentina Chair for Research on the
Social Psychology of Conflict and Cooperation, Tel-Aviv University. We
thank Sonia Roccas for helpful comments on earlier versions of this article.
Correspondence concerning this article should be addressed to Arie
Nadler, Department of Psychology, Tel Aviv University, Ramat Aviv
69978, Israel. E-mail: arie@freud.tau.ac.il
SOCIAL IDENTITY PERSPECTIVE
Over the last three decades, the social identity perspective on
intergroup relations (Turner & Reynolds, 2001) has progressed in
two complementary lines of research and theory: social identity
theory (Tajfel & Turner, 1979, 1986) and self-categorization the-
ory (Tajfel, 1978; Turner, Hogg, Oakes, Reicher, & Wetherell,
1987). Self-categorization theory holds that people’s identity fluc-
tuates across a continuum ranging from individual (e.g., “I am a
generous person”) to social (e.g., “I am a fan of a particular soccer
team”) identity, and research has explored variables that affect the
movement between these two poles and its consequences (e.g.,
Onorato & Turner, 2004; Roccas, 2003). Social identity theory
originated with the pioneering work of Tajfel and his colleagues,
who argued that in their quest for positive identity, group members
positively distinguish themselves from outgroups by discriminat-
ing against them (i.e., Tajfel & Turner, 1979). In support of this,
research has demonstrated that the division of people into groups,
even on a relatively insignificant dimension (e.g., as specific vs.
global perceivers), is sufficient to produce ingroup favoritism (i.e.,
discrimination against that outgroup) and outgroup devaluation
(e.g., Jetten, Spears, & Manstead, 1999).
More recent research indicates that the phenomena of ingroup
favoritism and outgroup devaluation are affected by the status of
the ingroup and the outgroup (Bourhis & Gagnon, 2003). Whereas
members of high-status groups show greater discrimination toward
low- than equal-status groups (Sachdev & Bourhis, 1991), low-
status groups sometimes exhibit outgroup favoritism: When they
have internalized their low status, they favor the high-status out-
group on dimensions of comparison that are related to their status
inferiority (Sachdev & Bourhis, 1987). The inconsistency that is
created between outgroup favoritism and people’s general need for
positive identity can be resolved by individual mobility (joining the
high-status group), social creativity (e.g., reframing the intergroup

97
98 NADLER AND HALABI
comparison in a way that favors the ingroup), or social competition
(working to elevate the status of the ingroup through social
change; Tajfel & Turner, 1986). Social competition is more likely
when members of the low-status group perceive the existing status
hierarchy as relatively unstable and illegitimate and intergroup
boundaries as impermeable (Tajfel & Turner, 1986). When status
relations are perceived as illegitimate and unstable, members of
low-status groups are likely to view the existing social hierarchy as
changeable, and because the impermeability of intergroup bound-
aries prevents individual mobility, they are expected to channel
their motivation for social equality toward elevating ingroup sta-
tus. Regarding the differential effects of perceived legitimacy and
perceived stability, research has indicated that variations in levels
of perceived stability have stronger effects on the behavior and
perceptions of low-status group members toward the high-status
outgroup (e.g., Ellemers, Wilke, & van Knippenberg, 1993; Mum-
mendey, Klink, Mielke, Wenzel, & Blanz, 1999). Because the
present focus is on behavior and perceptions directed at the high-
status outgroup helper, we centered our attention on the effects of
perceived stability on receptivity to help from the high-status
group.
HELPING RELATIONS AS POWER RELATIONS
Helping relations are inherently unequal social relations. The
helper has sufficient resources to confer on a recipient, who is
dependent on the helper’s goodwill. This inequality makes receiv-
ing help a potentially self-threatening experience for the benefi-
ciary (Nadler & Fisher, 1986). In this line, empirical research has
found that when help is self-threatening, people respond negatively
to its receipt (e.g., Nadler, 1987; Nadler & Fisher, 1986) and prefer
to endure hardships rather than seek it (Nadler, 1991). Yet, this
research has converged almost exclusively on interpersonal help-
ing encounters. In an exception to this interpersonal focus, Schnei-
der, Major, Luhtanen, and Crocker (1996) studied reactions to
interracial help and found that African American participants who
received assumptive help (i.e., unsolicited help) from a European
American experienced lower self-esteem than African Americans
who received assumptive help from a fellow African American.
Similar findings were reported in a study that examined the reac-
tions of Arab-Israelis to the receipt of help from a Jewish-Israeli as
opposed to an Arab-Israeli helper (Halabi, 2003). The study found
that Arab-Israeli recipients reported lower self-evaluations when
they were helped by Jewish-Israelis (the dominant group in Israeli
society) than when they were helped by Arab-Israelis (the less
dominant group). Taken together, these studies suggest two impli-
cations. First, it seems that when there is a salient distinction
between social groups, as is the case for African Americans and
European Americans in the United States and Arabs and Jews in
Israel, interpersonal helping encounters between members of these
groups tend to be experienced as an intergroup interaction. In
support of this, Suleiman (2004) noted that in Israeli society,
interactions between Israeli Arabs and Israeli Jews are perceived
by participants as intergroup encounters even when they take the
form of interpersonal dialogue (Suleiman, 2004). Second, these
studies may be interpreted to indicate that because dependency on
the high-status outgroup reinforces its dominant position, help
from a member of the dominant outgroup is threatening to recip-
ients from the low-status group. This conclusion is echoed in
discussions that note that being on the receiving end of affirmative
action programs can be a stigmatizing experience for its benefi-
ciaries (Pratkanis & Turner, 1996; Steele, 1992).
The autonomy or dependency orientation of help may also
influence whether dependence on a more privileged outgroup will
threaten the recipient’s social identity (Nadler, 1997, 1998).
Dependency-oriented help consists of providing a full solution to
the problem at hand and reflects the helper’s view that the needy
cannot help themselves. When recipients agree that they cannot
make it on their own, dependency-oriented help is consistent with
their view of themselves and they may readily seek and accept it.
However, when recipients believe they can succeed independently,
dependency-oriented help is inconsistent with their view of them-
selves as capable actors. In this case, potential recipients of help
are likely to reject offers of dependency-oriented help and refrain
from seeking it. Autonomy-oriented help is partial and temporary
(e.g., taking the form of instructions or hints) and reflects the
helper’s view that given the appropriate tools, recipients can help
themselves (Brickman et al., 1982). Autonomy-oriented help al-
lows recipients to retain their independence despite their reliance
on the more resourceful helper (Nadler, 1997, 1998). Therefore
this type of help is not likely to clash with recipients’ view of
themselves as capable and equal actors. Applied to the present
research context, this suggests that dependency-oriented help, but
not autonomy-oriented help, will be inconsistent with the low-
status group’s motivation for equality. Because this motivation
grows higher when status relations are perceived as unstable,
members of low-status groups are expected to be reluctant to seek
or receive dependency-oriented help under these conditions.
Intergroup Helping Relations as Status Relations
Figure 1 is a schematic representation of intergroup helping as
status relations (Nadler, 2002). It suggests two clusters of inter-
group helping relations: (a) when status relations are perceived as
stable and legitimate and (b) when they are viewed as unstable and
illegitimate. In the first case, the high-status group is expected to
try to maintain its social advantage by providing dependency-
oriented help to the low-status group. Under these conditions, the
low-status group is expected to be receptive to dependency-
oriented help. When status relations are perceived as unstable
and illegitimate, members of the high-status group view their
privileged position as being threatened and are expected to try
to reaffirm their social advantage through increased efforts to
provide dependency-oriented help to the low-status group. Un-
der these conditions, members of low-status groups—which
should be motivated to gain equal status—are expected to react
negatively to the receipt of dependency-oriented help from the
high-status group and be willing to seek such help only when it
is autonomy-oriented.
The Role of Ingroup Identification
The model of intergroup helping does not account for the effects
of individual variation between group members. Yet, research on
social identity shows that not all ingroup members respond to
threats to ingroup identity the same way. These responses are
affected by individuals’ identification with their group. Mem-
bers who identify strongly with the ingroup are more defensive
when ingroup identity is threatened. They express stronger
identification with the “threatened” ingroup, more ingroup fa-
 INTERGROUP HELPING RELATIONS 99

Figure 1. Intergroup helping relations as affected by perceived legitimacy and stability of power relations
between groups. From “Inter-group helping relations as power relations: Maintaining or challenging social
dominance between groups through helping,” by A. Nadler, 2002, Journal of Social Issues, 58, 487–502.
Copyright 2002 by Blackwell Publishing. Reprinted with permission.

voritism, and increased stereotyping against the outgroup,
whereas low identifiers lower their identification with the in-
group (Ellemers, Spears, & Doosje, 1999). Also, threat to social
identity increases the perceptions, particularly among those
who identify highly with their group (Doosje, Spears, Ellemers,
& Koomen, 1999), of the homogeneity of members of the
ingroup and of the outgroup (Rothgerber, 1997). Applied to the
present context of intergroup helping, we expect the reluctance
of low-status group members to receive or seek dependency-
oriented help from the high-status group to be more character-
istic of high ingroup identifiers than low identifiers. We do not
expect similar differences when help is autonomy-oriented.
PRESENT RESEARCH AND HYPOTHESES
The present studies focused on the reactions and behaviors of
low-status group members to offers of help from the high-status
group as affected by the perceived stability of intergroup status
relations, their degree of ingroup identification, and the
dependency- or autonomy-oriented nature of help. The first
study was a minimal group experiment and assessed the link
between the perceived stability of status relations and reactions
to receiving help from a high-status group. The second exper-
iment examined the same link with real groups (i.e., Israeli Jews
and Israeli Arabs). The third experiment used the same Arab-
Israeli and Jewish-Israeli intergroup context to examine the
effects of ingroup identification on the reactions of low-status
group members to help from the high-status outgroup. The
fourth experiment used a different real group context (two
competing high schools) to examine help-seeking behavior as
affected by the perceived stability of intergroup status relations,
degree of ingroup identification, and the autonomy- versus
dependency-oriented nature of help.
Our main predictions rest on our theorizing that when status
relations are perceived as relatively unstable, dependence on the
high-status outgroup is inconsistent with group members’ quest
for equality and results in a threat to social identity. This threat
should be expressed in relatively low affect, drive group mem-
bers to positively distinguish the ingroup by discriminating
against and devaluing the outgroup, and perceive the ingroup
and the outgroup as more homogeneous (Studies 1 and 2).
These reactions to help will be more characteristic of high than
low ingroup identifiers (Study 3). Finally, high identifiers are
expected to be the least willing to seek needed help from the
high-status outgroup when status relations are perceived as
unstable and dependency-oriented help is offered (Study 4).
STUDY 1
Method
Participants and Design
The first experiment used the minimal group paradigm and consisted of
a 2 (help vs. no help) ⫻ 2 (stable vs. unstable status relations) between-
100 NADLER AND HALABI
participants experimental design. Sixty-seven Israeli undergraduate stu-
dents ages 19 to 24 years (44 female and 23 male, equally distributed
across the four experimental cells) participated in the experiment.
Procedure
Participants were seated in individual booths and were told that they
were participating in a study of the links between perceptual style, inte-
grative thinking abilities, and social decision making. We informed par-
ticipants that people usually fall into either of two perceptual style cate-
gories, global perceivers or specific perceivers, and that global perceivers
have higher integrative abilities. We then had participants take the dot
estimation task (Jetten, Spears, & Manstead, 1996), which would ostensi-
bly determine which perceptual style they belonged to. On completing the
task, all participants learned that they were specific perceivers (i.e., that
they belonged to the group with lower integrative abilities).
Independent Variable Manipulation
Stability of status relations. In the stable status condition, participants
were told that the designation of individuals as specific or global perceivers
tends to remain consistent across multiple administrations of the dot
estimation task. In the unstable status condition, participants were in-
formed that the distinction between the two groups is somewhat inconsis-
tent and may change from test to test (for a similar manipulation of
perceived status stability, see Boen & Vanbeselaere, 2000; Turner &
Brown, 1978).
Help versus no help. Participants were told that in the next part of the
experiment they would work in pairs to complete a task requiring integra-
tive ability and that each pair would include one specific and one global
perceiver; the other pair member was said to be working in another room.
Each participant would be asked to solve 20 anagrams, and participants
were informed that students like themselves could usually solve 12 of these
20 anagrams in the allotted time. However, they had to solve at least 10
anagrams to proceed to the next phase of the experiment. To make the
second phase of the experiment seem more appealing, we told participants
that those who made it to the next phase would have the chance of winning
a large monetary prize.
Following these instructions, the 20 anagrams appeared consecutively on
the computer screen. Each anagram was presented for 5 s. On the basis of
a pilot test, we predetermined that 8 of the anagrams were easily soluble
and the other 12 extremely difficult. At the end of the anagram test,
participants learned that they had solved 8 anagrams correctly, whereas the
outgroup member (i.e., a global perceiver) had solved 14 anagrams cor-
rectly. At this point a message appeared on the screen stating that to
compensate for decrements in performance due to time pressure, partici-
pants would be given 2 more minutes to work on the anagrams they had not
been able to solve. We told the participants that because students in past
sessions had asked to communicate with their pair member, they could now
do so by writing a message that the experimenter would deliver to the other
participant. The experimenter then left the room and returned, 2 min later,
with an envelope containing a message from the other participant (the help
condition). The message stated: “It seems that you’re having some diffi-
culty—let me help.” The note included the solution to 4 of the difficult
anagrams. In the no-help condition, participants were exposed to the same
information but did not receive a message from the other participant.
At this point, before they could write a message of their own, partici-
pants were asked to respond to a number of questions. These included the
dependent measures (measure of affect, ingroup favoritism, evaluation of
the outgroup, and perceived homogeneity of the ingroup and outgroup).
Participants were asked “how you experienced participation in the exper-
iment” by choosing one of three possibilities: “The experiment feels like an
interaction between (1) groups, (2) representatives of two groups, or (3)
two individual students.” This allowed us to ascertain that participants
perceived the social interaction within the experiment as an intergroup
interaction. Following this, participants were fully debriefed and the ex-
periment ended.1
Manipulation Checks
Perceived stability of status relations. Participants were asked to name
the group they belonged to (i.e., global or specific perceivers) and rate the
degree to which they thought that the differences between global and
specific perceivers would remain constant throughout the experiment, on a
7-point scale ranging from 1 (not at all constant) to 7 (very constant).
Help versus no-help manipulation. At the beginning of the debriefing,
participants were asked to report if they had received any information from
their partner (i.e., the outgroup member) and to describe the nature of this
communication.
Dependent Measures
We assessed recipients’ reactions to receiving help in terms of affect,
behavior (ingroup favoritism), and attitudes (evaluation of the outgroup
and perceived homogeneity).
Affect. Participants were asked to rate “How I feel now” on nine
bipolar adjectives: good/bad, happy/sad, negative/positive, strong/weak,
calm/nervous, angry/not angry, satisfied/dissatisfied, secure/insecure, and
successful/unsuccessful. Because of the high internal consistency between
items (Cronbach’s ␣ ⫽ .89), ratings were summed to obtain a single
measure of affect, with higher scores indicating higher affect.
Ingroup favoritism. As noted previously, the experiment was presented
as assessing the links among perceptual style, integrative thinking abilities,
and social decision making. It was indicated to participants that as part of
a social decision-making task, they were now asked to divide a hypothet-
ical sum of 3,000 New Israeli Shekels (NIS; equivalent to roughly $700)
between a member of “your group—that is, a specific perceiver—and a
person who belongs to the group of global perceivers.” Participants were to
choose one of seven alternatives: three representing ingroup favoritism
(1,800/1,200 NIS, 1,700/1,300 NIS, and 1,600/1,400 NIS for the ingroup
and outgroup, respectively), one representing equal division (1,500 NIS to
each group), and three representing outgroup favoritism (1,200/1,800 NIS,
1,300/1,700 NIS, and 1,400/1,600 NIS to the ingroup and outgroup, re-
spectively; see Federico, 1998, for a similar assessment of ingroup
favoritism).
Evaluation of the outgroup. This was measured on six 7-point bipolar
adjectives. Although the internal consistency for these items was relatively
high (Cronbach’s ␣ ⫽ .82), it increased when ratings for the aggressive/
nonaggressive item were deleted (to ␣ ⫽ .89). This, together with previous
findings that reactions to assumptive help from a high-status outgroup are
characterized by feelings of hostility toward the helper (Schneider et al.,
1996), suggested that the ratings of the outgroup on aggressiveness repre-
sented a relatively different dimension than the overall evaluation. This
was borne out in a factor analysis, using varimax rotation, which yielded
two distinct factors: The first, accounting for 58% of the variance, was a
general evaluation factor (i.e., ratings on honest/dishonest, high abilities/
low abilities, successful/unsuccessful, trustworthy/untrustworthy, and de-
pendable/not dependable), and the second, accounting for 18% of the
1
Of the participants, 21% (13 out of 62) wrote back a note to the other
participant (i.e., the outgroup member). Six were in the stable status/help
cell, 3 in the unstable status/help cell, and 4 in the stable status/no-help cell.
This small number does not allow for statistical analyses, yet the contents
of the messages are revealing and are consistent with our theoretical logic.
All 6 messages in the stable status/help cell consisted of hopeful expecta-
tions for future cooperation, words of thanks, and the like. The 4 messages
in the stable status/no-help condition were all informative requests (e.g.,
“Did you do well?” and “Was it hard?”). All 3 messages in the unstable
status/help condition included an element of resentment and were a vari-
ation on “Thanks, but I can handle it myself.”
 INTERGROUP HELPING RELATIONS
variance, consisted of ratings of the outgroup ␮s aggressiveness (i.e.,
aggressive/nonaggressive). Higher scores on the first factor denote more
favorable evaluation of the outgroup, and higher scores on the aggressive-
ness item denote more perceived aggressiveness.
Perceived homogeneity of ingroup and outgroup. Perceived homoge-
neity was measured by asking participants to indicate the degree to which
specific perceivers were similar to each other (ingroup homogeneity) and
the degree to which global perceivers were similar to each other (outgroup
homogeneity). Both scales ranged from 1 (not at all similar to each other)
to 7 (very similar to each other; see Ellemers, Spears, & Doosje, 1997, for
a similar assessment).
Results
All of the participants correctly identified themselves as mem-
bers of the specific-perceivers group. Five of the 67 participants
perceived the interaction as an interpersonal interaction. They were
evenly distributed across the four experimental cells (2 in one cell
and 1 in each of the other three cells). The analyses reported here
are based on the results of all 67 participants.2 Analyses revealed
no significant main effects or interactions involving gender, and it
was therefore not included as a factor in subsequent analyses in
this research.
Manipulation Checks
Status Manipulation
Participants in the stable status condition, more than participants
in the unstable status condition, thought that the differences be-
tween the two groups would remain the same throughout the
experiment; means were 4.60 and 3.50, respectively, F(1, 65) ⫽
8.07, p ⬍ .01.
Help Manipulation
All participants in the help condition correctly remembered that
they had received a communication from their partner and that this
communication contained help.
Dependent Measures
Measure of Affect
A 2 (help vs. no help) ⫻ 2 (stable vs. unstable status) analysis
of variance (ANOVA) yielded a main effect for status stability,
F(1, 63) ⫽ 20.08, p ⬍ .001, which is qualified by a significant
Stability ⫻ Help interaction, F(1, 63) ⫽ 3.80, p ⬍ .05. This
interaction reflected the finding that participants who had received
help in the unstable status condition had lower ratings of affect
than those who had received help in the stable status condition;
means were 3.20 and 4.58, respectively, t(63) ⫽ 4.12, p ⬍ .001.
The comparable difference in the no-help condition was not sig-
nificant; means were 4.10 and 4.80, respectively.
Ingroup Favoritism
A 2 (help vs. no help) ⫻ 2 (stable vs. unstable status) ANOVA
on the measure of ingroup favoritism revealed no significant
effects. Although the predicted Help ⫻ Stability interaction was
not significant, F(1, 63) ⬍ 1, we tested the a priori hypotheses by
comparing the mean score of ingroup favoritism in the unstable
status/help cell to the stable status/help cell. This comparison
101
revealed that in line with our hypothesis, participants in the un-
stable status/help cell tended to be more discriminatory toward the
outgroup than those in the stable status/help cell; means were 3.52
and 2.81, respectively, F(1, 31) ⫽ 3.70, p ⬍ .07. The comparable
difference in the no-help condition was not significant; means
were 3.10 and 2.70, respectively, F(1, 32) ⬍ 1.
Another approach to assessing the experimental hypothesis was
to compare the percentages of ingroup favoritism choices. We
collapsed the ingroup favoritism index, which assessed the mag-
nitude of discrimination on a 7-point scale, into the three catego-
ries of ingroup favoritism, equal division, and outgroup favoritism.
Because no participant made an outgroup favoritism choice, par-
ticipants’ allocation decisions were compared between the two
categories of ingroup favoritism and equal division. As predicted,
participants in the help/unstable status cell made significantly
more ingroup favoritism choices than those in the help/stable
status cell; average percentages were 62.5% and 23.5%, respec-
tively, ␹2(1, N ⫽ 33) ⫽ 5.10, p ⬍ .05. The equivalent comparison
in the no-help condition was not significant; average percentages
were 47% and 30%, respectively, ␹2(1, N ⫽ 34) ⫽ 1.12, ns.
Evaluation of the Outgroup
A 2 (help vs. no help) ⫻ 2 (stable vs. unstable status) ANOVA
on the general evaluation score indicated that when status relations
were perceived as unstable, the outgroup was evaluated less pos-
itively than when they were perceived as stable; means were 4.30
and 5.00, respectively, F(1, 63) ⫽ 11.54, p ⬍ .001. Also, partic-
ipants who had received help rated the outgroup less favorably
than those who had not received help; means were 4.40 and 4.80,
respectively, F(1, 63) ⫽ 4.70, p ⬍ .05. The Stability ⫻ Help
interaction was not significant, F(1, 63) ⬍ 1.
A 2 (help vs. no help) ⫻ 2 (stable vs. unstable status) ANOVA
on perceived aggressiveness of the outgroup revealed a significant
interaction, F(1, 63) ⫽ 3.70, p ⬍ .05. Participants who had
received help viewed the outgroup as more aggressive when status
relations were perceived as unstable than when they were per-
ceived as stable; means were 3.60 and 4.50, respectively, t(63) ⫽
2.12, p ⬍ .05. The difference between the stable and unstable
status cells in the no-help condition was nonsignificant (means
were 4.60 and 4.40, respectively).
Perceived Ingroup and Outgroup Homogeneity
A 2 (stable vs. unstable status) ⫻ 2 (help vs. no help) ANOVA
on the item assessing perceived ingroup homogeneity revealed no
significant effects. A similar ANOVA on the perceived homoge-
neity of the outgroup revealed a significant Status Stability ⫻ Help
interaction, F(1, 63) ⫽ 8.27, p ⬍ .005. This interaction derived
from the finding that participants in the unstable status conditions
who had received help from the outgroup perceived the outgroup
as more homogeneous than those who had received help in the
stable status conditions; means were 4.70 and 3.00, respectively,
t(63) ⫽ 3.77, p ⬍ .001. The comparable difference was not
significant in the no-help condition (means were 3.50 and 3.60,
respectively; see Table 1).
2
The main analyses with and without these 5 participants revealed
similar patterns of findings.
102 NADLER AND HALABI

Table 1 the empirical and theoretical tradition of the social identity per-
Means (and Standard Deviations) of Affect Scores, Percentage spective, it limits the generalizability of our findings. Several
of Ingroup Favoritism Choices, Perceived Aggressiveness of authors suggest that there are differences between empirical rela-
Outgroup, and Perceived Outgroup Homogeneity as a Function tionships observed with ad hoc groups in the laboratory and similar
of Perceived Status Stability and Help (N ⫽ 62): Study 1 relationships in studies that examine real groups (Jetten et al.,
1996; Mullen, Brown, & Smith, 1992).
Experimental condition To address this possibility and replicate the findings, the second
Help No help
experiment assessed the same hypothesis with real groups. It
examined the reactions of Arab-Israeli participants to the receipt of
Dependent measure Unstable Stable Unstable Stable help from an Arab-Israeli or a Jewish-Israeli helper (i.e., lower and
higher status groups in Israeli society, respectively) under varying
Affect 3.20 4.58 4.10 4.80
degrees of perceived stability of intergroup status relations. To
(0.55) (1.01) (0.81) (1.20)
Percentage of extend generalizability, we used the same theoretical logic but
discriminatory choices 62.5% 23.5% 47.0% 30.0% changed the focus of the empirical comparison. Whereas the first
Perceived aggressiveness of 3.60 4.50 4.60 4.40 experiment focused on a comparison between the help and no-help
outgroup (1.20) (0.99) (0.99) (1.50) conditions, the second experiment focuses on comparing reactions
Perceived outgroup 4.70 3.00 3.50 3.60
homogeneity (1.06) (1.50) (1.06) (1.41) to receiving help from a high-status outgroup versus help from an
ingroup helper under varying levels of perceived stability of status
Note. Lower scores on the measure of perceived aggressiveness denote relations. We predict that consistent with the theoretical model
greater level of perceived aggressiveness. (Nadler, 2002) and the findings of the first experiment, Arab-
Israelis’ social identity will be most threatened when they receive
help from a Jewish-Israeli helper and status relations are perceived
Discussion
as unstable. We expect this high threat to be reflected in depressed
The results of the first experiment support the central hypothe- affect and to result in high motivation for positive ingroup distinc-
sis. When the status hierarchy was perceived as relatively stable, tiveness, which will be expressed by increased ingroup favoritism
the receipt of help from the high-status outgroup did not influence and devaluation of the outgroup.
recipients’ affect, ingroup favoritism, and perceptions of the out-
group. Yet when the status hierarchy was perceived as unstable, STUDY 2
being helped by a member of the high-status outgroup led recip-
ients to feel worse. This relatively negative affect may reflect the Method
greater threat to social identity under these conditions, leading to
efforts for positive ingroup distinctiveness (expressed in more Participants and Design
discrimination toward the outgroup in the unstable/help cell than Participants were 71 Arab-Israeli high school students in a school in
the stable/help cell). The parallel comparison in the no-help con- northern Israel, ages 16 and 17. The experiment was conducted on school
dition was not significant. The findings for perceived aggressive- premises and consisted of a 2 (Arab-Israeli vs. Jewish-Israeli helper) ⫻ 2
ness indicate that the outgroup was perceived as most aggressive (stable vs. unstable status hierarchy) between-participants design. The
by participants who had perceived status relations as unstable and proportion of male to female participants in each of the experimental cells
had received help from the high-status outgroup. This finding is was the same.
conceptually important and suggests that when help thwarted the
low-status group members’ motivation for equality (i.e., status Procedure
relations were perceived as unstable), recipients viewed the helper
as “forcing” his or her generosity on them. Finally, consistent with Participants took part in the experiment in small groups of 5– 6 individ-
research that threat to social identity leads to viewing the source of uals who were seated in individual booths. The experiment was portrayed
as an assessment of the validity of the Israeli psychometric tests, which are
threat as more homogeneous (Doosje et al., 1999), participants in
used as entrance examinations for institutions of higher learning in Israel.
the unstable status condition who had received help viewed the The test was said to include an assessment of verbal, quantitative, and
outgroup as more homogeneous than those who had not received social skills. The research was said to be sponsored by the National Testing
such help. Center in Israel and conducted nationwide. After making these short
The conclusions from the first experiment are limited by the introductory comments, the experimenter excused himself, saying that he
nature of the help that was studied and the type of groups that were must attend a prescheduled meeting with the school’s principal and that his
used. Regarding the nature of help, the participants received help coworker would administer the study.
without having asked for it (i.e., assumptive help; Schneider et al.,
1996). This raises the possibility that help from a high-status group Independent Variable Manipulations
when status relations are perceived as unstable poses a threat to the
recipient’s social identity only when it is imposed by the high- Stability of status relations. Participants were asked to read an infor-
mation page on psychometric testing, which was presented as part of the
status helper. We address this issue in the fourth experiment and
measurement of verbal skills. The information page ended with a statement
discuss it more extensively in the General Discussion section. on the achievements of different subgroups in Israeli society on psycho-
Concerning the type of groups used, this experiment used ad metric tests and noted that the average scores of Arab-Israeli students were
hoc, experimentally created groups. While this reliance on the lower than the average scores of Jewish-Israelis. In the stable status
minimal group paradigm increases confidence in the internal va- condition, the text stated that this gap had remained constant over the years,
lidity of the observed phenomena and places this research within and in the unstable status condition, the text stated that the gap was
 INTERGROUP HELPING RELATIONS
consistently narrowing. Because it is commonly known that members of
disadvantaged groups within Israeli society, including many Israeli Arabs,
score lower on these tests than members of advantaged groups, this
information did not constitute a deception.
Participants were then asked to answer a few questions ostensibly
assessing their understanding of the material they had just read. One of
these questions comprised the manipulation check on the stability of status
relations.
Group affiliation of helper and help-giving. The experimenter intro-
duced himself, by using a different name, to half of the participants as an
Arab-Israeli and to the other half as a Jewish-Israeli. Participants were
asked to work on a sample of 12 psychometric problems, 6 of which were
extremely difficult and 6 of which were readily soluble. Participants were
given 15 min to solve all 12 problems and had been told that students can
usually do so in about 10 min. Six minutes into their work, the experi-
menter, walking between the booths, discreetly pointed to the correct
answer on 4 difficult problems (i.e., provided assumptive help). Subse-
quently, participants were asked to fill out the dependent measures. The
experimenter was blind to the stability condition to which participants
belonged.
Manipulation Checks
Stability of status relations. Following the status stability manipula-
tion, participants were asked to answer a number of questions on the
material they had just read. One of the questions asked them to rate their
agreement, on a 7-point scale ranging from 1 (strongly disagree) to 7
(strongly agree), with the statement that “I think that the gaps between
Arab-Israelis and Jewish-Israelis in Israeli society have remained constant
over the years.”
Group affiliation of helper. Toward the end of the experiment, partic-
ipants were asked to report their impressions of the session they had just
participated in. These included questions asking them for the group affil-
iation of the test administrator and whether or not he had helped them.
Dependent Variables
As in Study 1, here we also assessed affective, behavioral, and attitudinal
reactions (i.e., affect, ingroup favoritism, and outgroup evaluation, respec-
tively) to the receipt of help from the high-status outgroup, as influenced
by different levels of perceived stability of status relations.
Affect. Participants were asked to rate their affect on the same nine
7-point bipolar adjectives that were used in Study 1, and the average of
these items was used as an affect score (Cronbach’s ␣ ⫽ .88).
Ingroup favoritism. This measure was presented as assessing social
decision making. Participants were asked to imagine that they had just
learned that high schools in their area (i.e., the Haifa–Nazareth region in
northern Israel) needed funds to operate a special educational program. The
schools were arranged in four pairs of schools, each consisting of an Arab
and a Jewish school (identified by name). Participants were asked to
allocate funds by dividing 1,000 NIS between the two schools in each pair
by choosing one out of seven possible allocation possibilities. Three
represented ingroup favoritism (50/950 NIS, 200/800 NIS, or 350/650 NIS
to the Jewish-Israeli and the Arab-Israeli school, respectively), one repre-
sented equal division (500 NIS to each school), and three represented
outgroup favoritism (950/50 NIS, 800/200 NIS, or 650/350 NIS to the
Jewish-Israeli and Arab-Israeli school, respectively). The correlations be-
tween choices made on each of these four pairs of Arab-Israeli/Jewish-
Israeli schools were significant and high (ranging from .57 to .77). Ingroup
favoritism scores were computed by subtracting the average amount allo-
cated to the ingroup (an Arab-Israeli school) from the average amount
allocated to the outgroup (a Jewish-Israeli school). Thus, positive scores
reflect ingroup favoritism and negative scores reflect outgroup favoritism.
The higher the score, the greater the amount of ingroup favoritism.
Outgroup evaluation. At the end of the experiment, we asked partic-
ipants if they would be willing to fill out a scale for a separate study by a
103
sociology student who was investigating how different groups in Israel
perceived each other. They all agreed and were asked to evaluate Israeli
Jews on seven 7-point bipolar adjective scales: wise/stupid, honest/dishon-
est, good/bad, devious/trustworthy, truthful/deceitful, egoistic/altruist, and
materialistic/nonmaterialistic. The average of these items was used as an
index of outgroup evaluation (Cronbach’s ␣ ⫽ .91).
Results
Manipulation Checks
Perceived Stability of Status Relations
Participants in the stable status condition perceived the status
gap between Arabs and Jews in Israel as being more constant over
the years than did participants in the unstable status condition;
means were 6.05 and 3.30, respectively, t(69) ⫽ 2.26, p ⬍ .001.
Group Affiliation of Helper
All of the participants correctly remembered the group affilia-
tion of the experimenter (i.e., the helper), and 97% (69 partici-
pants) also indicated that he had helped them.3
Dependent Measures
Affect
A 2 (Arab-Israeli vs. Jewish-Israeli helper) ⫻ 2 (stable vs.
unstable status) ANOVA on affect scores revealed a main effect
for helper’s group affiliation, F(1, 67) ⫽ 5.15, p ⬍ .05, which was
qualified by a two-way interaction, F(1, 67) ⫽ 3.97, p ⬍ .05. The
interaction was due to the finding that affect scores in the Jewish-
Israeli helper/unstable status condition were significantly lower
than affect scores in the Jewish-Israeli helper/stable status condi-
tion; means were 4.26 and 5.48, respectively, t(67) ⫽ 3.70, p ⬍
.001. The comparable difference in the Arab-Israeli helper condi-
tion was not significant; means were 5.24 and 5.54, respectively,
t(67) ⬍ 1.
Ingroup Favoritism
A 2 (Arab-Israeli vs. Jewish-Israeli helper) ⫻ 2 (stable vs.
unstable status) ANOVA revealed a significant two-way interac-
tion, F(1, 67) ⫽ 7.27, p ⬍ .01. This interaction is due to the finding
that participants who had received help from a Jewish-Israeli under
conditions of an unstable status hierarchy displayed greater in-
group favoritism than those who had received help from a Jewish-
Israeli under conditions of a stable status hierarchy; means were
554.2 and 252.8, respectively, t(67) ⫽ 2.92, p ⬍ .01. When the
helper was an Arab-Israeli, the ingroup favoritism scores were not
significantly different when status relations were perceived as
3
The analyses on the ratings of all 71 participants yielded the same
effects as analyses using only the 69 participants who correctly remem-
bered having been helped.
104 NADLER AND HALABI

Table 2
Means (and Standard Deviations) of Affect Scores, Amount of Funds Allocated to the Ingroup,
and Evaluation of the High-Status Outgroup as a Function of Helper’s Group Affiliation and
Status Stability (N ⫽ 71): Study 2

Experimental condition

Jewish-Israeli helper Arab-Israeli helper

Dependent measure Unstable Stable Unstable Stable

Affect 4.26 5.48 5.24 5.54

(0.67) (0.99) (1.30) (0.78)
Amount given to the ingroup 554.20 252.80 295.60 200.00
(334.50) (281.00) (368.30) (234.90)
Evaluation of the outgroup (Jewish-Israelis) 2.75 4.33 4.32 4.55
(0.98) (1.38) (1.51) (0.93)

unstable versus stable; means were 295.6 and 200.0, respectively,
t(67) ⬍ 1.4
Outgroup Evaluation
A 2 (Arab-Israeli vs. Jewish-Israeli helper) ⫻ 2 (stable vs.
unstable status) ANOVA revealed a main effect for the helper’s
group affiliation, F(1, 67) ⫽ 9.47, p ⬍ .01, which is qualified by
an interaction between status stability and the helper’s group
affiliation, F(1, 67) ⫽ 5.24, p ⬍ .05. The interaction is due to the
finding that the evaluation of the outgroup (i.e., Israeli Jews) was
lower when status relations were perceived as unstable than when
they were perceived as stable; means were 2.75 and 4.33, respec-
tively, t(67) ⫽ 3.80, p ⬍ .001. The parallel difference between the
stable and unstable status conditions when the helper was an
Arab-Israeli was not statistically significant; means were 4.32 and
4.55, respectively, t(67) ⬍ 1 (see Table 2).
Discussion
The main findings of the second study are consistent with our
predictions and the findings of the first study. The greater threat to
social identity was evident on affective, behavioral, and attitudinal
reactions to receiving help from the high-status outgroup. Partic-
ipants who had been helped by the high-status outgroup helper
(i.e., a Jewish-Israeli) felt worse under conditions of unstable
status relations than participants who had received help from the
high-status outgroup helper under conditions of stable status rela-
tions, and worse also than those who had been helped by an
ingroup helper (i.e., an Arab-Israeli). The depressed affect in the
unstable status/outgroup helper cell is consistent with the sugges-
tion that these conditions present the greatest level of threat to
social identity. Congruent with the idea that this threat will result
in increased efforts to attain positive ingroup distinctiveness, par-
ticipants in this cell displayed a higher level of discrimination
against the outgroup and evaluated it less positively than partici-
pants who had received help from the high-status outgroup under
the stable status conditions. Variations in the perceived stability of
status relations did not affect the discrimination and evaluation of
the outgroup by participants who had been helped by an ingroup
member (i.e., an Arab-Israeli).
There are two important differences between Studies 1 and 2:
(a) The first study was a minimal group experiment and the second
was conducted with real groups, and (b) the first study tested the
hypothesis by comparing responses between the stable and unsta-
ble status cells in the help and no-help conditions, whereas the
second study tested the hypothesis by comparing between the
stable and unstable cells in the ingroup and high-status outgroup
helper conditions. The support for the major experimental hypoth-
esis, given these differences, raises our confidence in the theoret-
ical logic that underlies the phenomena under study. The findings
of the second experiment, by themselves, are open to the alterna-
tive interpretation that these discrepancies reflect the difference
between receiving help from an ingroup versus an outgroup helper,
irrespective of the outgroup’s status. Yet, the agreement with the
findings of the first experiment, and the conceptually consistent
effects of the perceived stability of status relations, makes this
alternative interpretation less plausible.
Reactions to threat to social identity are determined by level of
ingroup identification. In the face of threat to social identity, high
identifiers attempt to positively distinguish their group from the
source of threat (i.e., the outgroup) more so than low identifiers
(Ellemers et al., 1999). Applied to the context of intergroup help-
ing, this suggests that the threat posed by help from the high-status
outgroup will be higher for high than low identifiers. High iden-
tifiers are expected to report worse affect when receiving help
from a high-status outgroup and to discriminate against and de-
value the source of threat (i.e., the outgroup helper) more so than
those whose ingroup identification is relatively low. The third
study examined this hypothesis.
4
Another index of ingroup favoritism is the value that is derived from
the choice that participants make on the 7-point scale, where 1 represents
extreme ingroup favoritism (i.e., 950 and 50 NIS to the ingroup and
outgroup, respectively), 4 represents equality (i.e., 500 NIS to each), and 7
represents an extreme outgroup allocation choice (i.e., 50 and 950 NIS to
the ingroup and outgroup, respectively). Decisions on the four allocation
items were internally consistent (Cronbach’s ␣ ⫽ .87) and summed to
obtain a single index of ingroup favoritism. A 2 (stable vs. unstable
status) ⫻ 2 (Arab vs. Jewish helper) ANOVA revealed a significant
interaction, F(1, 67) ⫽ 6.29, p ⬍ .01. The pattern of findings was similar
to those obtained on the measure of average funds allocated to the ingroup,
as reported in the Results section (i.e., the Jewish helper/unstable status cell
had the highest ingroup favoritism score).
 INTERGROUP HELPING RELATIONS
STUDY 3
Method
Participants and Design
Sixty-four high school students from an Arab-Israeli high school in
northern Israel, ages 16 and 17, participated in a 2 (high identifiers vs.
control) ⫻ 2 (ingroup vs. outgroup helper) between-participants experi-
mental design. The proportion of male to female participants in each of the
four experimental cells was the same.
Procedure
The procedure was identical to the one used in Study 2. The experiment
was presented as an investigation of three dimensions in psychometric
tests: verbal, quantitative, and social skills. The first part was described as
an assessment of verbal skills and consisted of an ingroup identification
manipulation.
Independent Variable Manipulations
Ingroup identification. In the high identification condition, participants
were asked to read half a page on the contributions of Arab culture to
humankind in medicine, mathematics, and the arts. In the control condition,
participants read a neutral section. Following the reading, participants were
asked to answer a few questions on the section they had just read. One of
these questions was a check on the ingroup identification manipulation (see
Jetten, Spears, & Manstead, 1997, for a similar manipulation).
Group affiliation of helper and help-giving. The experimenter pre-
sented himself to half of the participants as an Israeli Jew and to the other
half as an Israeli Arab. Subsequently, and in an identical manner to the
procedures in the second experiment, participants began working on the
quantitative part of the assessment. The experimenter provided assumptive
help on 4 out of the 20 sequences of geometric problems that the partici-
pants were working on.
Manipulation Checks
For the manipulation check on ingroup identification, participants were
asked to rate their agreement with the statement “I am proud to be an Arab”
on a 7-point scale. The manipulation check for group affiliation of helper
was identical to Study 2.
Dependent Measures
As in the first two experiments, we again assessed the affective, behav-
ioral, and attitudinal reactions to receiving help from the high-status
outgroup. Measures of affect, ingroup favoritism, and outgroup evaluation
were identical to those used in Study 2.
Results
Manipulation Checks
Ingroup Identification
The difference between the high identification and control con-
ditions was highly significant, t(62) ⫽ 4.97, p ⬍ .0001 (means
were 6.62 and 4.18, respectively).
Helper’s Group Affiliation
All of the participants correctly remembered the group affilia-
tion of the helper, and 98% reported that he had helped them.
105
Dependent Measures
Affect
A 2 ⫻ 2 ANOVA on the affect measure revealed a main effect
for the group affiliation of the helper, F(1, 60) ⫽ 5.66, p ⬍ .05,
due to the lower affect scores of participants who had been helped
by a member of the outgroup than by an ingroup helper (means
were 4.74 and 5.38, respectively). The two-way interaction failed
to exceed an acceptable level of significance, F(1, 60) ⫽ 1.92, p ⬍
.17; however, to assess the a priori hypothesis that only in the high
identification condition would recipients of help from the high-
status outgroup experience lower affect than recipients of help
from an ingroup helper, we conducted planned comparisons be-
tween relevant cell means. Consistent with the hypothesis, partic-
ipants in the high identification/Jewish-Israeli helper group had
lower affect than participants in the high identification/Arab-
Israeli helper cell; means were 4.34 and 5.39, respectively, t(60) ⫽
2.70, p ⬍ .01. There was no significant difference between the
affect scores in the Arab-Israeli and Jewish-Israeli helper cells in
the control condition; means were 5.36 and 5.14, respectively,
t(60) ⬍ 1.
Ingroup Favoritism
A 2 (high identification vs. control) ⫻ 2 (Arab-Israeli vs.
Jewish-Israeli helper) between-participants ANOVA revealed a
significant two-way interaction, F(1, 60) ⫽ 4.09, p ⬍ .05. Planned
comparisons between relevant cell means indicate that this inter-
action is due to the outcome that high identifiers who had received
help from the high-status outgroup (i.e., from a Jewish-Israeli
helper) discriminated against the outgroup more than high identi-
fiers who had received help from an ingroup helper; means were
436.0 and 291.0, respectively, t(60) ⫽ 2.06, p ⬍ .05. The compa-
rable difference for participants in the control condition was not
significant; means were 230.0 and 318.0, respectively, t(60) ⬍ 1.5
Outgroup Evaluation
A 2 ⫻ 2 ANOVA revealed a similar two-way interaction, F(1,
60) ⫽ 4.41, p ⬍ .05. Planned comparisons indicate that high
identifiers who received help from the Jewish-Israeli helper rated
the outgroup lower than did high identifiers who had received help
from an Arab-Israeli helper; means were 3.68 and 4.62, respec-
tively, t(60) ⫽ 2.02, p ⬍ .05. The parallel difference in the control
condition was not significant; means were 4.55 and 4.20, respec-
tively, t(60) ⬍ 1 (see Table 3).
Discussion
Study 3 indicates that the threat to social identity inherent in
receiving help from the high-status outgroup depends on ingroup
5
Here, too, an index of ingroup favoritism was created by summing the
choices made on the four allocation items (Cronbach’s ␣ ⫽ .79; i.e., scores
ranging from 1 to 7, where 1 represents the highest level of ingroup
favoritism). A 2 (high vs. low identification) ⫻ 2 (Jewish vs. Arab helper)
ANOVA on these scores revealed a significant interaction, F(1, 59) ⫽
3.97, p ⬍ .05. This interaction reflects a similar pattern of findings as those
obtained on the average allocation of funds, as reported in the Results
section (i.e., the highest ingroup favoritism score was found in the high
identification/Jewish helper cell).
106 NADLER AND HALABI
Table 3
Means (and Standard Deviations) of Affect Scores, Amount of Funds Allocated to the Ingroup,
and Evaluation of the High-Status Outgroup as a Function of Helper’s Group Affiliation and
Ingroup Identification (N ⫽ 64): Study 3
Jewish-Israeli helper
High ingroup
Dependent measure identification
Affect 4.34
(0.85)
Amount given to the ingroup 435.90
(387.70)
Evaluation of the outgroup 3.68
(Jewish-Israelis) (1.50)
identification. High Arab-Israeli identifiers had lower affect after
receiving help from the dominant outgroup than after receiving
help from an ingroup helper, but affect for participants in the
control condition did not differ in these two cells. This lower affect
suggests that receiving help from a high-status outgroup is more
threatening for high identifiers than for participants in the control
condition who had not undergone a manipulation to increase
ingroup identification. Similar to the findings in the first two
experiments, this greater threat to social identity led to greater
efforts for positive ingroup distinctiveness, which found expres-
sion in more discrimination and devaluation of the source of threat
(i.e., the high-status outgroup). It may be noted that the manipu-
lation of ingroup identification (i.e., praising Arab contributions to
global art and culture) might be construed as having led to per-
ceptions of higher ingroup status. Yet, we maintain that the effects
are attributable instead to increased identification. This interpreta-
tion is supported by the consistency of the findings with the results
of Study 4, which are based on a different manipulation of iden-
tification (i.e., informing high school students that graduates of
their school are committed to their school).
The theoretical model guiding the present research program also
predicts that low-status group members will avoid seeking help
from the dominant outgroup when dependency on the high-status
group poses a threat to social identity (Nadler, 2002). Taken
together with the findings of the previous experiments, this model
suggests that high identifiers will be most reluctant to seek help
from the high-status outgroup when status relations are perceived
as unstable. This reluctance is likely to be evident only when the
assistance is dependency-oriented (i.e., a full solution is provided).
When assistance is autonomy-oriented (i.e., consisting of a partial
solution or hints), seeking help from the high-status outgroup is not
likely to pose a threat for social identity, and high-identifying
members of low-status groups are likely to seek such help even
when perceived status relations are unstable. These predictions are
the focus of the fourth study.
STUDY 4
The fourth experiment examined the willingness of low-status
group members to seek autonomy- or dependency-oriented help
from a high-status outgroup as a function of the interaction be-
tween (a) perceived stability of status relations and (b) ingroup
Experimental condition
Arab-Israeli helper
Low ingroup High ingroup Low ingroup
identification identification identification
5.14 5.39 5.36
(1.32) (1.04) (0.98)
229.70 290.60 318.70
(273.10) (202.80) (248.70)
4.55 4.20 4.62
(1.38) (1.11) (1.19)
identification. Thus, it represents a full examination of the predic-
tions of the intergroup helping model (Nadler, 2002), together with
extensions suggested by the findings of Study 3 (i.e., the role of
ingroup identification). This examination was carried out using a
behavioral measure of willingness to seek help.
Method
Participants and Design
Fifty-six students from a high school in a midsize town in northern Israel
participated in the study. The design was a 2 (high ingroup identification
vs. control) ⫻ 2 (stable vs. unstable status relations) between-participants
design. Each of the four experimental cells included a similar number of
male and female students. The dependent measures consisted of the fre-
quency that participants chose to (a) not seek help, (b) seek dependency-
oriented help, and (c) seek autonomy-oriented help. As in Studies 2 and 3,
this experiment also used real groups. However, unlike the previous
experiments, which used national group affiliation, the present experiment
used high school affiliation as the intergroup context.
Procedure
During the first part of the experiment the experimenter, who had been
introduced as an employee of the Israeli Ministry of Education, described
the study as an assessment of alternative forms of psychometric testing.
Participants were told that they would be working on tasks assessing verbal
abilities and interactive analytical abilities. Following this general intro-
duction, participants were told that the study was being carried out in
different geographical areas in the country and that their school and another
school had been chosen to represent northern Israel. The other school was
said to be a prestigious and reputable school in the city of Haifa (i.e., a
high-status outgroup).
Independent Variables Manipulation
Stability of status relations. In the stable status condition, participants
learned that a comparative analysis had revealed that over the last 5 years
their school’s overall performance on various criteria was consistently
lower than that of the higher status school (e.g., as measured by admission
to selective university programs). In the unstable status condition, partic-
ipants were told that over the last 5 years the gap between their school and
the high-status school had grown consistently narrower.
Ingroup identification. Following the manipulation of status stability,
the manipulation of ingroup identification was introduced. In the guise of
 INTERGROUP HELPING RELATIONS
taking the verbal part of the psychometric test, participants were asked to
read a short article that they would later be questioned on as part of the
“verbal abilities” section of the psychometric test. In the high ingroup
identification condition, the short article recounted the history of their high
school, praising the school and noting the commitment that students and
graduates felt for it. The article was said to have been taken from the local
community newspaper. In the control condition, participants read an article
of similar length that focused on environmental issues. After they had
finished reading the articles, participants were asked to answer a few
questions on what they had just read. This short questionnaire included
checks on the status stability and ingroup identification manipulations.
Manipulation Checks
Stability of status relations. Participants were asked to respond to two
questions on a 7-point scale: the chances that a graduate of their high
school and a graduate of the high-status school would be admitted to
prestigious university departments, and how many years it would be before
the scholastic achievements of the two schools would become relatively
equal. Responses to these two items were significantly correlated (r ⫽
⫺.55, p ⬍ .001), and ratings were summed to obtain a single index of
perceived status stability.
Ingroup identification. Participants answered two questions on a
7-point scale: the degree to which they had a sense of belonging to their
school and whether they would recommend enrolling at their school to
others. Answers to these two questions were highly correlated (r ⫽ .80,
p ⬍ .001), and ratings were summed to obtain a single index of ingroup
identification.
Dependent Measures: Avoidance of Help-Seeking,
Dependency-Oriented Help-Seeking, and Autonomy-
Oriented Help-Seeking
Participants were told that the next part of the study would assess
interactive analytic thinking. Participants were given five index cards, each
of which displayed an arithmetic problem that had to be solved within 60 s.
Two of the problems were readily soluble and were solved by all partici-
pants. The other three cards contained insoluble problems. After it became
clear that participants could not solve the three problems, and in prepara-
tion for the upcoming “interactive” part of the session, the students were
asked to indicate one of three choices for each problem they had been
unable to solve: (a) not receiving any assistance from a student from the
other high school, who was said to be working on the same task (i.e.,
avoidance of seeking help); (b) receiving the solution to the unsolved
problem from the other high school student (i.e., seeking dependency-
oriented help); or (c) receiving a hint from the other high school student
that might help them find the solution on their own (i.e., seeking autonomy-
oriented help). The indices for avoidance of help-seeking, dependent
help-seeking, and autonomous help-seeking consisted of the average num-
ber of times that participant chose either of these three alternatives. Thus,
scores for either of these options could range from 0 (i.e., never choosing
this option) to 3 (i.e., choosing that option on all three problems).
Results
Manipulation Checks
Perceived Stability of Status Relations
Participants in the stable status group perceived the gap between
the ingroup and the outgroup as more stable than did participants
in the unstable status condition, F(1, 52) ⫽ 9.29, p ⬍ .01 (means
were 3.11 and 2.04, respectively).
107
Ingroup Identification
Participants in the high identification condition identified with
their group more than participants in the control condition, F(1,
52) ⫽ 14.90, p ⬍ .001 (means were 6.01 and 4.77, respectively).
Dependent Measures
Avoidance of Help-Seeking
A 2 (stable vs. unstable status) ⫻ 2 (high identification vs.
control) ANOVA on the avoidance of help-seeking scores revealed
a significant Stability ⫻ Identification interaction, F(1, 52) ⫽ 6.15,
p ⬍ .01. This indicates that for participants in the control condi-
tion, scores for avoidance of help-seeking did not differ between
the stable and unstable status conditions; means were 0.67 and
0.50, respectively, t(52) ⫽ 0.80, ns, whereas high identifiers ex-
hibited significantly higher avoidance when status relations were
perceived as unstable than when they were perceived as stable;
means were 1.31 and 0.36, respectively, t(52) ⫽ 2.40, p ⬍ .05.
Dependent Help-Seeking
Consistent with our hypothesis, the incidence of help-seeking in
the unstable status/high identification condition was zero, render-
ing the use of ANOVA impossible. To counter this problem, we
conducted two t tests for independent samples in which equal
variance is not assumed; these compared the amount of
dependency-oriented help sought in the stable versus unstable
status cells, within the high identification and control conditions
separately. In the control condition, the amount of dependency-
oriented help did not differ between the stable and unstable status
cells; means were 1.20 and 1.43, respectively, t(26.51) ⫽ 0.52, ns.
In the high identification condition, participants in the stable status
cell sought more dependency-oriented help than those in the un-
stable status cell; means were 1.21 and 0.00, respectively,
t(13.00) ⫽ 3.82, p ⬍ .01.
Because students had to choose among three alternatives, there
were two degrees of freedom on this measure. We expected
significant Stability ⫻ Identification interaction on the avoidance
index, and a similar interaction, but in an opposite direction, on the
dependent help-seeking index. We expected, and found, that high
identifiers will have the highest avoidance scores when status
relations were unstable and lowest dependency help-seeking
scores when relations were unstable. Under these conditions, the
ANOVA on the third measure (i.e., autonomy-oriented help-
seeking) could not have produced any significant effects. We
therefore did not conduct an ANOVA on this index.
Discussion
The findings of Study 4 support the hypotheses. Participants
were most reluctant to seek help from a member of the high-status
outgroup when they had been induced to identify with the ingroup
and viewed intergroup status relations as unstable. It is important
to note that this reluctance was evident only in the case of
dependency-oriented help. In fact, not a single participant in the
unstable status/high identification condition sought dependency-
oriented help. We interpret this to reflect that under these condi-
tions, seeking help conflicted with the motivation of high identi-
fiers to elevate the status of their ingroup and attain equality with
108 NADLER AND HALABI
the outgroup. When assistance was framed as autonomy-oriented,
high identifiers sought similar amounts of help from the outgroup
as participants in the control condition, regardless of the stability
of status relations.
While the first three studies examined segments of the theoret-
ical links suggested by the present analysis (i.e., the stability of
power relations in Studies 1 and 2, and ingroup identification in
Study 3), the fourth experiment assessed all three conceptual
elements in tandem. It found support for the joint effects of (a)
structural characteristics of intergroup power relations (i.e., per-
ceived stability), (b) characteristics of the person in need of help
(i.e., ingroup identification), and (c) the nature of the help (i.e.,
autonomy- or dependency-oriented) on help-seeking. Also, the
study extends the generalizability of the earlier findings in two
major aspects. First, it allows for the generalization of the theo-
retical principles from the realm of reactions to help (as in the first
three experiments) to actual help-seeking behavior. Second, while
Study 1 focused on experimentally created ad hoc groups and
Studies 2 and 3 on relations between Israeli Arabs and Jews, Study
4 examined status differences between different high schools. The
overall similarity in the pattern of findings across these different
kind of groups and operationalizations reinforces our confidence
that the phenomenon under study represents a basic phenomenon
of intergroup relations.
GENERAL DISCUSSION
The above four experiments support the hypotheses of the
intergroup helping model (Nadler, 2002). Taken together, these
studies indicate that members of low-status groups are least recep-
tive to help from a high-status outgroup when status relations are
perceived as unstable and help is dependency-oriented. Further,
this unwillingness to seek or receive help from the high-status
group is particularly characteristic of high ingroup identifiers. We
interpret this to indicate that under the abovementioned conditions,
dependency on a high-status helper is inconsistent with the moti-
vation of the low-status group for social equality and that such help
poses a threat to social identity.
Although the overall empirical picture is consistent with this
interpretation, the studies contain some methodological ambigu-
ities that require discussion. In the second and fourth experiments,
the perceived instability of status relations was induced by inform-
ing participants that the gap between their group and the high-
status group was narrowing over time. It is possible that such a
manipulation could have resulted in the perception of higher
ingroup status or may have led to greater identification with the
ingroup. Our interpretation that these effects are attributable to the
different perceptions of the stability of status relations is reinforced
on a number of grounds. First, this manipulation is consistent with
previous manipulations of perceived status stability and instability
in this context (Boen & Vanbeselaere, 2000; Federico, 1998;
Turner & Brown, 1978). Second, our manipulation checks showed
that participants perceived the manipulation as intended. Third, the
effects of status stability in Studies 2 and 4 are consistent with
those in the first experiment, which was a minimal group experi-
ment and in which a different manipulation of stability was used
(i.e., the likelihood that participants would be assigned to the
low-status group in the future).
In social systems in which the power hierarchy is perceived as
stable and legitimate, it is the privileged group’s duty to cater to
the needs of the low-status group. The low-status group’s recep-
tiveness to such help serves as a behavioral acknowledgment of its
inferior position. This is captured in Mauss’s (1907/1957) state-
ment that by accepting gifts without returning them in kind, the
recipients “become client and subservient” (p. 72). Such social
situations are likely to have characterized relations between racial
and gender groups in past centuries (e.g., African Americans and
European Americans, males and females) and in societies in which
the power structure is relatively rigid (such as tribal societies). It is
interesting to note that such a unidirectional flow of assistance
from the strong to the weak is also evident in nature, in which a
less dominant member in the community of birds that was studied
rarely gives food to the more dominant member of its group
(Zahavi & Zahavi, 1997). This idea is consistent with the argument
made by Worchel (1984) that helping may be a way for dominant
groups to maintain their supremacy and with analyses of paternal-
ism in gender and race relations (e.g., Jackman, 1994; Pratto &
Walker, 2001; van der Berghe, 1967). In the same line, Leach,
Snider, and Iyer (2002) argued that one of the consequences of
feeling secure in one’s privileged social position is benevolence
toward the disadvantaged. They wrote: “The security of the for-
tunates’ advantage allows a somewhat benevolent reaction to the
disadvantaged in the form of pity” (p. 7). In the same way that
receptivity to help from the high-status outgroup is a behavioral
affirmation of the disadvantaged position of the low-status group,
refusing help from the advantaged group may signify a challenge
to the existing status quo. As our data indicate, this is more likely
when status relations are perceived as unstable.
The empirical support for our hypotheses is underscored by the
differences that existed between the four experiments, as reflected
in the types of groups studied, the status dimensions used, the
comparisons used to assess the main hypotheses, and the depen-
dent measures that were assessed. Regarding the types of groups
studied, the first study was a minimal group experiment, whereas
the other three experiments studied real groups. Regarding the
dimension of status, the second and third experiments studied
low-status groups within a given society (i.e., Arab citizens in
Israel), whereas the fourth experiment focused on status differ-
ences between two high schools. The central comparisons across
the four experiments also varied (i.e., receiving vs. not receiving
help in Study 1, and reactions to receiving help from a high-status
outgroup and ingroup helper in Studies 2 and 3). Furthermore, both
help-seeking behavior and reactions to help were predicted by the
same theoretical logic. In closing, it may be noted that future
research should analyze the mediating role of relevant psycholog-
ical motivations (e.g., positive ingroup distinctiveness) and emo-
tions (e.g., feeling humiliated by dependency) on the observed
relationships.
Direct or Indirect Means to Maintain or Challenge
Intergroup Status Relations
The logic guiding the present research has implications for both
high-status groups and help-giving behavior. It suggests that high-
status groups may try to retain a jeopardized position of dominance
by providing dependency-oriented help to the source of threat (i.e.,
the low-status group). This leads to an important question: When
will a high-status group try to maintain dominance by providing
help, and when will it do so by the more direct means of asserting
its superiority? The same question also applies to the low-status
 INTERGROUP HELPING RELATIONS
group, which at times may directly challenge existing inequality by
confronting the high-status group with a demand for equal power,
and at other times less directly by refusing dependency on the
more dominant group. Although the direct and less-direct forms of
maintaining intergroup power relations are not mutually exclusive,
the question of when groups are more likely to use one or the other
method remains open. One possible determinant is the social and
normative context within which intergroup relations take place. If,
for example, the social context is confrontational (e.g., groups are
in open conflict), they may be likely to use more explicit and direct
methods to maintain or challenge existing status relations. If,
however, the social context discourages open confrontation,
groups may be likely to use less explicit and confrontational means
to maintain or challenge power relations. Helping relations are
such means.
Intergroup and Interpersonal Helping Relations
Although the present analysis focused on intergroup helping, we
theorize that the same processes apply to the analysis of helping
relations between differentially powerful individuals (e.g., manag-
ers and employees). Consistent with self-categorization theory, the
analysis of the helping interaction will be informed by variables
that are relevant to this level of analysis (e.g., ingroup identifica-
tion) when interactants’ social identity is salient. However, when
the interactants’ individual self is salient, the helping interaction
may be construed as an interpersonal interaction, requiring con-
sideration of variables that are relevant to this level of analysis
(e.g., self-esteem). In both cases, however, the pattern of helping
relations depends on the perceived stability and legitimacy of
existing power relations between the parties and the dependency or
autonomy orientation of the help.
Social Change and Helping Relations
Times of social change give rise to motivational conflicts be-
tween more and less socially dominant groups, and this may be
expressed in the framework of helping relations as well. Whereas
a stronger party may seek to defend its social advantage by
providing dependency-oriented help, the less dominant party may
likely be motivated to refuse such help. This may fuel tension
between the two parties. In this scenario, one can imagine the more
dominant group feeling baffled and angry at the low-status group’s
resistance to its generosity, and the less dominant party viewing
the generosity of its counterpart as a manipulative effort to retain
social advantage. These processes are described as operating in
affirmative action programs (Pratkanis & Turner, 1996), and sim-
ilar barriers have been underscored on the road to peace-building,
wherein the stronger party’s efforts to assist its former adversary
may be spurned out of hand as mechanisms to retain dominance
(Lederach, 1997; Nadler & Saguy, 2004). The fact that helping is
both an expression of caring and a demonstration of superiority
makes it an especially effective instrument of dominance in the
hands of a more advantaged group. As our data show, members of
low-status groups are sensitive to this danger and under certain
conditions resent such help and are unwilling to seek it.
References
Bierhoff, H. (2002). Prosocial behavior. New York: Psychology Press.
Boen, F., & Vanbeselaere, N. (2000). Responding to membership of a
109
low-status group: The effects of stability, permeability and individual
ability. Group Processes and Intergroup Relations, 3, 41– 63.
Bourhis, R. Y., & Gagnon, A. (2003). Social orientations in the minimal
group paradigm. In R. Brown & S. Gaertner (Eds.), Intergroup processes
(pp. 89 –112). Oxford, England: Blackwell.
Brickman, P., Rabinowitz, V. C., Karuza, J., Coates, D., Cohn, E., &
Kidder, L. (1982). Models of helping and coping. American Psycholo-
gist, 37, 368 –384.
Doosje, B., Spears, R., Ellemers, N., & Koomen, W. (1999). Perceived
group variability in intergroup relations: The distinctive role of social
identity. European Review of Social Psychology, 10, 41–73.
Ellemers, N., Spears, R., & Doosje, B. (1997). Sticking together or falling
apart: In-group identification as a psychological determinant of group
commitment versus individual mobility. Journal of Personality and
Social Psychology, 72, 617– 626.
Ellemers, N., Spears, R., & Doosje, B. (Eds.). (1999). Social identity:
Context, commitment, content. Oxford, England: Blackwell.
Ellemers, N., Wilke, H., & van Knippenberg, A. (1993). Effects of the
legitimacy of low group or individual status on individual and collective
identity enhancement strategies. Journal of Personality and Social Psy-
chology, 64, 766 –778.
Federico, C. (1998). The interactive effects of social dominance orienta-
tion, group status, and perceived stability on favoritism for high-status
groups. Group Processes and Intergroup Relations, 2, 119 –143.
Halabi, S., (2003). Reactions to receiving assumptive help in or out-group
member: The moderating effects of task difficulty. Unpublished master’s
thesis, Tel Aviv University, Israel.
Jackman, M. R. (1994). The velvet glove: Paternalism and conflict in
gender, class, and race relations. Berkeley: University of California
Press.
Jetten, J., Spears, R., & Manstead, A. S. R. (1996). Intergroup norms and
intergroup discrimination: Distinctive self-categorization and social
identity effects. Journal of Personality and Social Psychology, 71,
1222–1233.
Jetten, J., Spears, R., & Manstead, A. S. R. (1997). Strength of identifica-
tion and intergroup differentiation: The influence of group norms. Eu-
ropean Journal of Social Psychology, 27, 603– 609.
Jetten, J., Spears, R., & Manstead, A. S. R. (1999). Group distinctiveness
and in-group discrimination. In N. Ellemers, R. Spears, & B. Doosje
(Eds.), Social identity (pp. 107–126). Oxford, England: Blackwell.
Leach, C. W., Snider, S. L., & Iyer, A. (2002). “Poisoning the conscience
of the fortunate”: The experience of relative advantage and support for
social equality. In I. Walker & H. J. Smith (Eds.), Relative deprivation:
Specification, development and integration (pp. 1–14). New York: Cam-
bridge University Press.
Lederach, J. P. (1997). Building peace: Sustainable reconciliation in
divided societies. Washington, DC: U.S. Institute of Peace.
Mauss, M. (1954). The gift: Forms and functions of exchange in archaic
societies. Glencoe, IL: Free Press. (Original work published 1907)
Mullen, B., Brown, R., & Smith, C. (1992). Ingroup bias as a function of
salience, relevance, and status: An integration. European Journal of
Social Psychology, 22, 103–122.
Mummendey, A., Klink, A., Mielke, R., Wenzel, M., & Blanz, M. (1999).
Socio-structural characteristics of intergroup relations and identity man-
agement strategies: Results from a field study in East Germany. Euro-
pean Journal of Social Psychology, 29, 259 –287.
Nadler, A. (1987). Determinants of help seeking behavior: The effects of
helper’s similarity, task centrality and recipient’s self esteem. European
Journal of Social Psychology, 17, 57– 67.
Nadler, A. (1991). Help-seeking behavior: Psychological costs and instru-
mental benefits. In M. S. Clark (Ed.), Review of personality and social
psychology (Vol. 12, pp. 290 –312). New York: Sage.
Nadler, A. (1997). Autonomous and dependent help seeking: Personality
characteristics and the seeking of help. In B. Sarason, I. Sarason, & R. G.
110 NADLER AND HALABI
Pierce (Eds.), Handbook of personality and social support (pp. 258 –
302). New York: Plenum Press.
Nadler, A. (1998). Relationship, esteem and achievement perspectives on
autonomous and dependent help seeking. In S. A. Karabenick (Ed.),
Strategic help seeking: Implications for learning and teaching (pp.
61–95). Mahwah, NJ: Erlbaum.
Nadler, A. (2002). Inter-group helping relations as power relations: Main-
taining or challenging social dominance between groups through help-
ing. Journal of Social Issues, 58, 487–502.
Nadler, A., & Fisher, J. D. (1986). The role of threat to self-esteem and
perceived control in recipient reactions to aid: Theory development and
empirical validation. In L. Berkowitz (Ed.), Advances in experimental
social psychology (Vol. 19, pp. 81–123). New York: Academic Press.
Nadler, A., & Saguy, T. (2004). Trust building and reconciliation between
adversarial groups: A social psychological perspective. In H. Langholtz
& C. E. Stout (Eds.), The psychology of diplomacy (pp. 29 – 46). New
York: Praeger.
Onorato, R. S., & Turner, J. C. (2004). Fluidity in the self-concept: The
shift from personal to social identity. European Journal of Social Psy-
chology, 34, 257–278.
Penner, L. A., Dovidio, J. F., Piliavin, J. A., & Schroeder, D. A. (2005).
Prosocial behavior: Multilevel perspective. Annual Review of Psychol-
ogy, 56, 365–392.
Pratkanis, A. R., & Turner, M. E. (1996). The proactive removal of
discriminatory barriers: Affirmative action as effective help. Journal of
Social Issues, 52, 111–132.
Pratto, F., & Walker, A. (2001). Dominance in disguise: Power, benefi-
cence and exploitation in personal relationships. In A. Y. Lee-Chai &
J. A. Bargh (Eds.), The use and abuse of power: Multiple perspectives on
the causes of corruption (pp. 93–112). Washington, DC: Psychology
Press.
Roccas, S. (2003). Identification and status revisited: The moderating role
of self-enhancement and self-transcendence values. Personality and
Social Psychology Bulletin, 29, 726 –736.
Rothgerber, H. (1997). External intergroup threat as an antecedent to
perceptions in ingroup and outgroup homogeneity. Journal of Person-
ality and Social Psychology, 73, 1206 –1212.
Sachdev, I., & Bourhis, R. (1987). Status differentials and intergroup
behavior. European Journal of Social Psychology, 17, 277–293.
Sachdev, I., & Bourhis, R. (1991). Power and status differentials in
minority and majority group relations. European Journal of Social
Psychology, 21, 1–24.
Schneider, M. E., Major, B., Luhtanen, R., & Crocker, J. (1996). Social
stigma and the potential costs of assumptive help. Personality and Social
Psychology Bulletin, 22, 201–209.
Schroeder, D. A., Penner, L. A., Dovidio, J. F., & Piliavin, J. A. (1995).
The psychology of helping and altruism: Problems and puzzles. New
York: McGraw-Hill.
Steele, C. (1992, April). Race and the schooling of Black Americans.
Atlantic Monthly, 68 –78.
Suleiman, R. (2004). The planned meeting as a microcosmos. In R. Halabi
(Ed.), Israeli and Palestinian identities in dialogue (pp. 47– 67). Pisca-
taway, NJ: Rutgers University Press.
Tajfel, H. (1978). Interindividual behavior and intergroup behavior. In H.
Tajfel (Ed.), Differentiations between social groups: Studies in the
social psychology of intergroup relations (pp. 27– 45). New York:
Academic Press.
Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup
conflict. In W. G. Austin & S. Worchel (Eds.), The social psychology of
intergroup relations (pp. 33– 47). Monterey, CA: Brooks/Cole.
Tajfel, H., & Turner, J. C. (1986). The social identity theory of intergroup
behavior. In W. G. Austin & S. Worchel (Eds.), The social psychology
of intergroup relations (pp. 7–24). Chicago: Nelson-Hall.
Turner, J. C., & Brown, R. (1978). Social status, cognitive alternatives and
intergroup relations. In H. Tajfel (Ed.), Differentiation between social
groups: Studies in the social psychology of intergroup relations (pp.
201–234). London: Academic Press.
Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell, M. S.
(1987). Rediscovering the social group: A self-categorization theory.
Oxford, England: Blackwell.
Turner, J. C., & Reynolds, K. J. (2001). The social identity perspective in
inter-group relations: Theories, themes and controversies. In R. Brown
& S. Gaertner (Eds.), Intergroup processes (pp. 133–153). Oxford,
England: Blackwell
Van der Berghe, P. L. (1967). Race and racism. New York: Wiley.
Worchel, S. (1984). The darker side of helping: The social dynamics of
helping and cooperation. In D. Bar-Tal, J. Karylowski, & J. Reykowski
(Eds.), The development and maintenance of prosocial behavior (pp.
379 –397). New York: Plenum.
Zahavi, A., & Zahavi, A. (1997). The handicap principle: A missing piece
of Darwin’s puzzle. Oxford, England: Oxford University Press.
Received April 26, 2005
Revision received October 29, 2005
Accepted November 4, 2005 䡲