You are on page 1of 9

Organizational Behavior and Human Decision Processes 116 (2011) 163171

Contents lists available at ScienceDirect

Organizational Behavior and Human Decision Processes


journal homepage: www.elsevier.com/locate/obhdp

How do you fake a personality test? An investigation of cognitive models


of impression-managed responding
Mindy K. Shoss a,, Michael J Strube b
a
b

Department of Psychology, University of Houston, TX, USA


Department of Psychology, Washington University in St. Louis, MO, USA

a r t i c l e

i n f o

Article history:
Received 3 February 2010
Accepted 14 May 2011
Available online 28 June 2011
Accepted by Paul Levy
Keywords:
Cognition
Faking
Personality questionnaire

a b s t r a c t
Because faking poses a threat to the validity of personality measures, research has focused on ways of
detecting faking, including the use of response times. However, the applicability and validity of these
approaches are dependent upon the actual cognitive process underlying faking. This study tested three
competing cognitive models in order to identify the process underlying faking and to determine whether
response time patterns are a viable method of detecting faking. Specically, we used a within-subjects
manipulation of instructions (respond honestly, make a good impression, make a specic impression)
to examine whether the distribution of response times across response scale options (e.g., disagree,
agree) could be used to identify faking on the NEO PI-R. Our results suggest that individuals reference
a schema of an ideal respondent when faking. As a result, response time patterns such as the well-known
inverted-U cannot be used to identify faking.
2011 Elsevier Inc. All rights reserved.

Introduction
Personality measures have re-emerged in personnel selection
because research has supported their validity in such contexts
(Ones, Dilchert, Viswesvaran, & Judge, 2007). The ease and low cost
of administering these assessments, as well as the lower adverse
impact compared to other selection tools (Hough, Oswald, &
Ployhart, 2001), adds to their attraction. However, researchers
have questioned the validity of these measures in light of research
suggesting that individuals can (Grifth & McDaniel, 2006;
Viswesvaran & Ones, 1999) and do (Birkeland, Manson, Kisamore,
Brannick, & Smith, 2006) effectively distort their responses in
situations where it is advantageous to do so. Recent research suggests that distortions can come at a considerable cost to predictive
validity (Komar, Brown, Komar, & Robie, 2008), especially for those
who score at the high end of the distribution (Mueller-Hanson,
Heggestad, & Thornton, 2003).
Thus, faking undermines the use of personality scales as research instruments and in applied settings (Mesmer-Magnus &
Viswesvaran, 2006). Faking subverts the purpose of content-driven
personnel selection and instead may lead to the selection of applicants who are especially skilled at impression management rather
than especially noteworthy on the intended attribute. At an even
more basic level, faking on selection instruments raises questions
about the basic honesty of applicants and potential employees.
Corresponding author. Address: Department of Psychology, University of
Houston, Houston, TX 77024, USA.
E-mail address: mmkrischer@uh.edu (M.K. Shoss).
0749-5978/$ - see front matter 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.obhdp.2011.05.003

Particularly troublesome, 74% of applicants believe that other


applicants fake on selection assessments (English et al., 2005 as cited in Grifth & McDaniel, 2006), suggesting considerable pressure
for applicants to fake in order to not be at a signicant disadvantage. Not surprisingly, the self-help literature now has many
how to guides for manipulating responses to personality assessments to increase the chances of being hired (Grifth & McDaniel,
2006).
In response to these issues, researchers have sought to identify
ways to detect fakers, and have suggested that response times
might be useful towards this end (e.g., Holden & Kroner, 1992;
Holtgraves, 2004). However, research ndings on the differences
in response times between those faking and those responding honestly have largely been inconsistent. Not only does this call into
question the idea that response times may be useful for identifying
fakers, but also it raises a broader issue of identifying the cognitive
process that underlies faking on personality questionnaires that
might lead to differences in response times.
In the current study, we examine the fundamental assumption
underlying the suggestion to use response times to detect faking
that the cognitive process engaged while faking is different enough
from honest responding that response times will be consistently
different when an individual is faking than when an individual is
responding honestly. Specically, we test three competing models
of fakingthe self-schema, adopted-schema, and semantic evaluation modelsby examining patterns of response times for different
response options. Our approach to testing these models is a novel
one. In contrast to prior studies that have examined average response times across items, our approach allows us to explicitly test

164

M.K. Shoss, M.J Strube / Organizational Behavior and Human Decision Processes 116 (2011) 163171

the nuanced predictions these models make for how response time
patterns associated with faking compare to the well-known inverted-U response time pattern for honest responding (Kuiper,
1981). As a result, we are not only able to test competing models
of faking, but are also able to examine whether response time patterns can be used to identify faking.
Our study, therefore, advances the literature on faking by shedding light on the cognitive processes underlying faking and, in
doing so, indicating whether response time patterns are a viable
way of detecting faking. As these models of faking take as their
starting point the cognitive processes presumed to underlie honest
responding, we begin by discussing cognitive models of honest
responding. We then describe the models of faking and the predictions that each makes for response time patterns across response
options.
Cognitive models of honest responding
Despite the fact that reading a 20-item scale takes about
1.5 min, the average administration time for the same length scale
is about 1015 min (Grifth & McDaniel, 2006). Therefore, other
cognitive processes are likely engaged when an individual responds to a personality assessment. For example, Holtgraves
(2004) argued that these cognitive processes include a retrieval
stage, during which one gathers the information necessary to formulate a response, and a judgment stage, during which one maps
the retrieved information onto the response options and selects the
appropriate response.
More specically, others have argued that honest responding
requires accessing a self-schema in the retrieval phase and comparing item responses to the exemplars that dene the individual
when making judgments about the response option to select (Kuiper, 1981). The endpoints of a response scale (e.g., strongly agree,
strongly disagree) are more likely to correspond to simple, clear,
and easily retrieved exemplars, leading to fast responses because
comparison to self-relevant information is accomplished with little
effort (see also Popham & Holden, 1990). In contrast, middle response options (e.g., agree somewhat, slightly disagree) correspond to more complex, conicting, or context-dependent
exemplars, requiring more thought (e.g., extensive retrieval) and
resulting in longer response times. The resulting inverted-U effect has now been conrmed by numerous studies (e.g., Casey &
Tyron, 2001; Mueller, Thompson, & Dugan, 1986). Recently,
Akrami, Hedlund, and Ekehammar (2007) reported its presence
for responses to the NEO Personality Inventory, the measure used
in the current study.
When individuals fabricate responses to questionnaire items,
additional processes are likely to be engaged as attempts are made
to manage or project an impression that deviates from the true or
honest self. Several models of impression-managed responding
have been proposed, each with implications for how response
times might deviate from the inverted-U pattern.
Cognitive models of impression-managed responding
Vasilopoulos, Reilly, and Leaman (2000) discussed three distinct, mutually exclusive models of the cognitive processes that
might guide impression-managed responding. In the self-schema
model, individuals access their self-schemas but decide to make responses inconsistent with those schemas, which requires some
deliberation while the social desirability of responses is considered. The adopted-schema model is similar to the self-schema model except that impression-managed responses involve referencing
a schema of an ideal respondent, not ones own self-schema. The
semantic-exercise model suggests that impression-managed responses require a simple semantic evaluation of how the trait

being assessed relates to either a favorable or stereotypic image.


Responses are selected solely on the basis of social desirability,
rather than by the extent to which they are self-referent.
Research examining these models has primarily examined average response times across items, rather than examining average response times for particular response options, and has resulted in
inconsistent and conicting ndings. Supporters of the self-schema
model cite research indicating that the social desirability implications of questionnaire items have been found to correlate with response times, reecting the delayed judgment stage implied by
this model (McDaniel & Timm, 1990). For example, Holtgraves
(2004) varied the social desirability signals of instructions for a
personality assessment and found that response times were positively associated with social desirability concerns. In contrast, Holden and colleagues (Holden & Hibbs, 1995; Holden & Kroner, 1992)
found that impression-managed response times for responses consistent with an ideal applicant were identical to honest-responding
response times, which they note is consistent with the use of an
adopted-schema. Finally, Hsu, Santelli, and Hsu (1989) found that
impression-management was associated with faster responses
than honest responding, consistent with the semantic-exercise
model, which suggests a judgment stage without a retrieval stage.
Contributing to the confusion, Vasilopoulos et al. (2000) suggested that the variation in ndings may be a result of differences
in the manipulation of the impression management task. They predicted that when the ideal job characteristics are not familiar or an
individual is merely trying to make a global positive impression,
response times would be slower than honest-responding response
times because no clear impression is being constructed. Consistent
with these predictions, they found that impression-management
was associated with slower response times than honest responding
for a low job familiarity (fake-good) condition, whereas a high job
familiarity (fake-specic) condition was associated with slightly
quicker responses. These results indicate that schema strength,
characterized as familiarity with the job, may inuence the average
speed with which individuals respond to personality inventories.
These ndings, however, do not lend clear support for a given model and make the interpretation of prior research on the response
times associated with faking difcult to interpret.
We suggest that past research contains a number of problems
that limit the ability to test model validity. One major problem is
the use of between-subjects designs. These designs require the
use of a standardization technique to control for individual differences in reading speed and simple response time that might obscure comparisons across conditions. For example, Holden (1998;
Holden, Fekken, & Cotton, 1991) used a double-standardization
technique to analyze the response latency data for a number of
studies that assigned participants to either a fake or honest
responding condition. In this approach, response times are rst
standardized across items within each subject, in order to control
for reading speed, and then response times are standardized across
subjects within each item, to control for item-specic factors such
as item length. Casey and Tryon (2001) attempted to improve on
Holdens technique by having participants push a key after they
had nished reading the question, then a key for an answer. The response time for the individual was the time difference between
both key pushes, assumed to just measure psychological response
time. However, individuals may begin the retrieval process as they
are reading or may not remember or be able to separate when
reading the question ends and thinking about a response begins.
In other words, the standardization process may remove variability
due to response distortionthe variability of primary interest in
response fabrication research. Indeed, empirical results indicate
higher response-distortion estimates for within-subjects designs
(that require no standardization) than for between-subjects designs (Mesmer-Magnus & Viswesvaran, 2006).

M.K. Shoss, M.J Strube / Organizational Behavior and Human Decision Processes 116 (2011) 163171

165

Table 1
Predicted response time patterns for models of impression-managed responding.
Model

Schema

Response process

Predicted response time distribution for


fake good condition

Predicted difference between fake-good and fake-specic


conditions

Self-schema

Self
Ideal
Respondent
None

Flatter and slower compared to honest


responding
Indistinguishable from honest
responding
Flatter and faster than honest
responding

No difference

Adoptedschema
Semanticexercise

Retrieval and prolonged


judgment stages
Retrieval and judgment
stages
Judgment stage only

Fake-specic will show a faster and more pronounced


inverted-U distribution than fake-good
No difference

Note. The honest responding condition is expected to produce a signicant inverted-U pattern, consistent with past research.

A second, related problem is that these studies generally compare average response times between items (i.e., averaging together response times regardless of the response option
selected), eliminating the possibility of using more subtle patterns
such as the inverted-U effect to detect honest versus faked responses. Moreover, using response time distributions allows competing tests of the previously described cognitive models, as each
has distinct implications for the way impression-managed
responding should emerge with regard to the inverted-U pattern.
If response time distributions across response options are different
when individuals fabricate their responses compared to responding honestly, then response time distributions could be a viable
method to detect faking in applied settings. We describe implications of the three models for response latency distributions in
the following discussion. This information is also summarized in
Table 1.
Self-schema model
Recall that the self-schema model of impression-managed
responding suggests that individuals access their self-schemas
but decide to make responses inconsistent with these schemas in
an attempt to appear more socially desirable. This deliberation
about social desirability prolongs the judgment stage in Holtgraves (2004) conceptualization. Moreover, the self-schema model
suggests that extreme dissimulated responses should engage relatively more cognitive processes than less extreme responses. Responses on the end-points of the scales (e.g., strongly agree),
which ordinarily are retrieved and selected easily when the response is honest, require additional thought, especially if an individuals self-schema indicates that a moderate response would
be honest. On average, this process produces a atter response distribution than occurs for honest responding and one that is slower
overall due to the added deliberation about social desirability of responses relative to the self. Additionally, this model suggests no
differences between the distributions when individuals are trying
to make a general good impression versus a very specic impressionboth involve additional deliberation that takes additional
time, especially for extreme responses.
Adopted-schema model
According to the adopted-schema model, individuals access a
schema of an ideal respondent. Provided that individuals have a
schema of what a good applicant might look like, the adopted schema model predicts a pattern of response times that should be also
be inverted-U shaped because the responses are honest with respect to the referenced schemathe referenced schema is adopted
as if it were ones own. The adopted-schema model can be taken to
make a more extreme prediction, however, in line with the results
reported by Vasilopoulos et al. (2000) suggesting that schema
strength may affect response times. To the extent that a clear schema is available and is simpler than the self-schema or a general
good impression schema, the adopted schema model predicts an
even more pronounced inverted-U pattern for impression-managed responding and overall responses times that are faster.

Semantic-exercise model
The semantic-exercise model involves a social desirability
assessment of each item. In Holtgraves (2004) conceptualization,
this model corresponds to the elimination of the retrieval stage.
Responses are selected solely on the basis of social desirability,
rather than by the extent to which they are self-referent or reect
an ideal schema. Response times may show an inverted-U distribution to the extent that extreme responses more readily convey a
good impression. However, given that a semantic evaluation engages relatively few cognitive processes, moderate responses
(e.g., somewhat agree) should also be made relatively quickly.
Thus, although an inverted-U may emerge, this will be less pronounced than in the honest-responding condition, and response
times will be generally faster than is true for honest responding.
Because selection merely requires a semantic evaluation, this
model suggests there will no differences between the response
time distributions when individuals are trying to make a general
good impression versus a specic impression.
The current study
The present study employed a within-subjects design in which
participants responded to the NEO PI-R under each of three conditions: (a) honestly, (b) to make a good impression, and (c) to make
a specic impression. Both general and specic impression-managed conditions were included because the three models make
different predictions for the response time distributions under
fake-general and fake-specic instructions. Thus, including a
fake-specic condition is necessary to provide a complete test of
the models.
Within the specic impression condition, participants were given one of two sets of instructionsto respond as if applying for
a job in accounting or to respond as if applying for a job in public
relations. Given that popular stereotypes of these positions involve
greater differences in levels of extraversion than in levels of other
dimensions (e.g., openness), we expected this condition would primarily involve a manipulation of the extraversion dimension. Prior
research has found increased extraversion scores as a result of faking (Bagby & Marshall, 2003), but this may depend on the impression one is trying to make. Although faking to make a generally
favorable impression and specic impressions appropriate for
many jobs likely involve portraying oneself as extraverted, some
jobs may call for a more introverted nature. This study, in addition
to testing the aforementioned models, determined if respondents
could dissimulate in this way when the circumstances called for it.
As noted previously, the cognitive models have different implications for response time distributions, both their shape and their
overall level. Prior research does not make a strong case for a particular model, so we note here the predicted pattern for each of the
models (see Table 1). The self-schema model predicts a atter response time pattern for the fake-good and fake-specic conditions
compared to the honest response condition and overall response
times that are signicantly slower. The adopted-schema model
predicts an inverted-U pattern for the fake-good condition that

166

M.K. Shoss, M.J Strube / Organizational Behavior and Human Decision Processes 116 (2011) 163171

should be indistinguishable from that found in the honestresponding condition. But, the adopted-schema model also makes
a more specic prediction with regard to the specic instruction
conditions (especially for the extraversion dimension that underlies these instructions) that allows it to be differentiated from honest responding. For the extraversion dimension, the inverted-U for
the fake-specic public relations condition should be more pronounced than for the other conditions. The instruction to fake
appropriateness for a public relations job represents a very specic
and clear ideal that, if adopted, should produce a particularly pronounced inverted-U pattern. Moreover, because both the public
relations and the accountant jobs likely represent clearer and more
specic ideals than faking to make a general positive impression,
the overall response times for them should be signicantly faster
than for honest responding or general fake-good responding. The
semantic-exercise model predicts generally faster response times
for the faking conditions than for honest responding, and a less
pronounced inverted-U effect.

Instructions for non-specic good impression


Please answer the questions as if your responses were going to
be read by potential employers for a job you really want or graduate school admissions committees for a school you really want to
go to. You should try to present yourself in the best light possible
so, for each item, select the response that you feel will give you the
highest score and make you look like the best applicant.
Instructions for specic impressionpublic relations
Please answer the questions as if you were the ideal candidate
for a position in public relations. According to the top public relations rms, an ideal candidate has the following qualities: outgoing, condent, credible, creative, takes initiative, and exercises
good judgment.

Method

Instructions for specic impressionaccounting


Please answer the questions as if you were the ideal candidate
for a position in accounting. An ideal candidate in accounting is detailed-oriented, punctual, has good time management, and is
disciplined.

Participants

Measures

Sixty-two participants were recruited from the Psychology


Department Volunteer Participant Pool. Twenty-ve participants
identied themselves as European Americans, 6 as African Americans, 4 Hispanic, 11 Asian Americans, and 4 as other ethnicities.
The average participant age was 19.38 (SD = 1.19). Two participants neglected to return for the second session, resulting in their
rst session data not being used in the analyses. Of the 60 participants included in the analyses, 22 were men and 38 were women.

Personality
The Revised NEO Personality Inventory (Costa & McCrae, 1992)
is a 240-question assessment measuring the Five Factor Model
dimensions of personality: Neuroticism (Cronbachs a = .90), Extraversion (a = .93), Openness to Experience (a = .85), Agreeableness
(a = .91), and Conscientiousness (a = .91). Each dimension is measured using 48 items, each rated on a 5-point Likert scale ranging
from strongly disagree (1) to strongly agree (5). This measure was
developed for use with adults without overt psychopathology
and has been found to have a high level of internal consistency
and testretest reliability over long periods of time. For this reason,
it has been translated into many different languages and used in
many different cultures.

Design
This experiment used a within-subjects manipulation of response instructions: honest, fake-good, and fake-specic. The
fake-specic response instructions were manipulated between
subjects, resulting in two fake-specic conditions that instructed
participants to dissimulate so as to obtain a position in either
accounting or public relations. Although we expected the fake-specic conditions to be primarily a manipulation of the extraversion
dimension, we used job positions in order to increase realism and
generalizability of our ndings to selection contexts. In addition,
individual difference assessments (NEO-PI-R dimension scores)
from the honest responding condition were incorporated into statistical analyses to determine if response-time patterns were moderated by particular personality characteristics.
Procedure
The experiment was administered by computer and required
approximately one hour for the rst session and half an hour for
the second session. In the rst session, participants were asked
to respond honestly to the NEO PI-R (Costa & McCrae, 1992) and
response times were recorded without participants knowledge.
At the end of the session, participants responded to a brief demographic questionnaire.
During the second session, participants completed the NEO PI-R
twice: (a) under non-specic instructions to give a good impression, and (b) under specic instructions to respond as an ideal candidate for one of two jobs. All participants responded to the general
instructions to fake good. Participants were randomly assigned to
just one of the specic instruction conditions. The computer program counterbalanced the instructions and measured response
times.

Results
Faking estimates and manipulation check
The effectiveness of the instructions was assessed by comparing
scores from the NEO scale administered under the three instruction conditions in a 2 (Specic Instructions: Extravert versus Introvert)  3 (Type of Instruction: Honest, Good Impression, Specic
Impression) analysis of variance with the latter factor treated as
a repeated measure. Results and specic comparisons among conditions for all scales are presented in Table 2. Based on previous research demonstrating that applicants responding to the NEO PI-R
exhibited higher mean conscientiousness than non-applicants
(Grifn, Hesketh, & Grayson, 2004), we expected a signicant increase in conscientiousness scores when participants attempted
to make a good impression compared to their honest responses.
Also, based on prior faking research (McFarland, Ryan, & Ellis,
2002; Vasilopoulos, Reilly, & Leaman, 2000), we expected elevated
extraversion and agreeableness scores and lower neuroticism
scores in the fake-good condition compared to the honest responding condition. However, we did not expect a signicant difference
in openness scores across conditions based on evidence suggesting
that openness to experience is considerably difcult to fake (Grifn
et al., 2004).
The most relevant NEO scale in our experiment was extraversion and the analysis of variance revealed a signicant specic
instruction by type of instruction interaction, F(2, 116) = 20.77,
p < .001. As expected, extraversion scores were signicantly higher

167

M.K. Shoss, M.J Strube / Organizational Behavior and Human Decision Processes 116 (2011) 163171
Table 2
NEO dimension scores as a function of instruction condition.
Scale

Honest (H)

Fake-good (FG)

Fake-specic
Extravert (FSE)

Neuroticism1
Extraversion2
Openness3
Agreeableness4
Conscientiousness5
1
2
3
4
5

Introvert (FSI)

SD

SD

SD

SD

148.47
170.12
175.50
162.28
165.02

20.65
24.48
15.99
20.43
20.20

94.35
192.65
178.83
212.50
187.30

17.56
13.02
14.50
17.05
15.31

96.44
202.80
185.36
203.12
174.44

22.60
14.60
17.75
21.55
23.40

93.57
164.30
145.00
220.29
163.37

18.27
23.19
24.09
14.50
21.10

F(2, 116) = 3.24, p < .01; t-tests revealed signicant differences (p < .01) between H&FG, H&FSE, H&FSI.
F(2, 116) = 20.77, p < .01; t-tests revealed signicant differences (p < .01) between H&FG, H&FSE, FSE&FSI.
F(2, 116) = 36.94, p < .01; t-tests revealed signicant differences (p < .01) between H&FSI, FSE&FSI.
F(2, 116) = 9.37, p < .01; t-tests revealed signicant differences (p < .01) between H&FG, H&FSE, H&FSI, FSE&FSI.
F(2, 116) = 4.90, p < .01; t-tests revealed signicant differences (p < .01) between H&FG.

in the fake-good condition than in the honest responding condition, in the fake-specic-extraversion condition than in the honest
responding condition, and in the fake-specic-extraversion condition than the fake-specic-introvert condition. Additionally, neuroticism scores were lower in all of the faking conditions than in
the honest responding condition, F(2, 116) = 3.24, p < .01. Similarly,
agreeableness scores were higher in all of the faking conditions
than in the honest responding condition, F(2, 116) = 9.37, p < .01.
Conscientious scores were signicantly higher in the fake-good
condition than in the honest responding condition, F(2, 116) =
4.90, p < .01. The pattern of results clearly shows that participants
successfully implemented both the general instructions to make a
favorable impression and the more specic instructions targeted to
particular job characteristics.
HLM analysis
We used hierarchical linear modeling (HLM) to analyze the response time data because observations (i.e., response times) were
nested within individuals (Osborne, 2000). Specically, each individual had response times for each response option that he or
she selected while responding to the NEO questionnaire. After reverse coding appropriate items, we averaged the response times
for each response option (e.g., agree) for a given scale (e.g., extraversion) for each individual participant (and separately by condition). We examined the NEO scales individually so that the
honest and fake-good conditions could be easily compared to the
fake-specic condition, which we expected manipulated primarily
the extraversion dimension. Thus, each person had 25 dependent
variables (response time averages; 5 per each of the 5 NEO scales)
for each experimental condition. HLM uses the data available in the
case of individuals who did not select a given response option
when responding to items in a particular scale. The inverted-U effect was examined by testing for a quadratic trend in average response times across the ve response options. Then, we used
moderator variables to determine if the relationship between response times and response options was general to all participants
or depended on the experimental manipulations and individual
differences.
Honest responding model specication and analysis
The initial HLM analysis tested the presence of the inverted-U
pattern under the honest responding instructions. The Level 1
model represented the average response times for each participant
as predicted by the selected response options, modeled to include
the linear and quadratic relations between response option and response time:

RT ri p0i p1i ROri p2i RO2ri eri

RTri is the average response time given by participant i for response


option r. ROri refers to the response option (15) chosen for questions from one dimension of the NEO PI-R. The response option variable was centered so that RO could take on values of 2, 1, 0, 1,
and 2 for each individual. Homogeneity of error variance was assumed for each individual and conrmed by analysis of residuals.
HLM assumes that residuals are randomly and normally distributed.
To achieve these assumptions, response time data were trimmed
such that responses over 8 s were excluded from the analysis.
The Level 2 model for analysis of honest responding treats the
Level 1 regression parameters (i.e., the intercept, linear, and quadratic effects of response options on response times) as outcomes
to be predicted. Note that the Level 2 parameters are at the individual level and reect the extent to which individual variability exists in the Level 1 parameters.

p0i b00 r0i


p1i b10 r1i
p2i b20 r2i
Each parameter can be construed as xed, randomly varying, or
nonrandomly varying (i.e., moderators are present). Tests were
conducted to determine the best model and in all cases either a
xed or randomly varying model (shown above) was found to be
optimal. The latter implies the presence of individual difference
moderators.
Estimates and standard errors of the xed coefcients are presented in Table 3, and results for the extraversion scale are also displayed graphically in Fig. 1. The analyses show a robust replication
of the inverted-U response time effect for each of the NEO PI-R subscales with negative and signicant quadratic coefcients in each
case. There were several signicant linear effects as well; however,
these effects were weak when compared with the quadratic inverted-U response time effect. Additionally, as can be seen in the
intercept parameter (p0i), the openness subscale was responded
to more slowly overall than other subscales, supporting previous
research suggesting that these items are more difcult to respond
to than the other Big Five scales (Grifn & Hesketh, 2004). The robust replication of the inverted-U response time effect in the honest responding condition provides the baseline against which to
assess response patterns in the faking conditions.
The r terms corresponding to error were either xed at zero or
estimated based on chi-square tests in a preliminary analysis.
These tests of the variance components were signicant (p < .05)
for the intercept for each NEO subscale, the linear component for
the agreeableness subscale, and the quadratic component for
extraversion, openness, agreeableness, and conscientiousness subscales. The signicant r terms suggest that there is variability
across individuals in terms of the size of these parameters (e.g.,

168

M.K. Shoss, M.J Strube / Organizational Behavior and Human Decision Processes 116 (2011) 163171

Table 3
Honest-responding xed effects estimation (with robust standard errors).
Scale

Fixed effect

Coefcient

Standard error

df

Neuroticism

Intercept, p0i
Linear, p1i
Quadratic, p2i

3624.36
62.16
137.41

92.51
27.15
22.98

39.18
2.29
5.98

59
271
271

.001
.023
.001

Extraversion

Intercept, p0i
Linear, p1i
Quadratic, p2i

3484.36
44.28
148.06

98.69
26.11
28.26

35.31
1.70
5.24

59
261
59

.001
.091
.001

Openness

Intercept, p0i
Linear, p1i
Quadratic, p2i

4081.31
37.79
181.07

98.93
40.18
34.38

41.25
0.94
5.27

59
59
59

.001
.351
.001

Agreeableness

Intercept, p0i
Linear, p1i
Quadratic, p2i

3516.84
25.57
87.75

92.19
36.38
30.09

38.15
0.70
2.92

59
59
59

.001
.485
.006

Conscientiousness

Intercept, p0i
Linear, p1i
Quadratic, p2i

3687.11
94.24
89.55

97.21
32.41
33.30

37.93
2.91
2.69

59
264
59

.001
.004
.010

4400
Honest Responding

4200

Good Impression

4000

Specific Impression (Extravert)

3800

Specific Impression (Introvert)

Response Time (ms)

3600
3400
3200
3000
2800
2600
2400
2200
2000
1800
1

Response Option
Fig. 1. Response times as a function of response option for the extraversion dimension of the NEO PI-R under honest response, fake-good response, and fake-specic response
instructions.

in the overall response times or in the shape of the response time


distributions). Attempts to nd individual difference moderators
(NEO-PI-R dimension scores) were largely unsuccessful, however.1
This suggests that there may be individual differences not measured
in this study (e.g., alertness) that may cause systematic variability in
response times.
Fake-good response time analysis
In the previous analysis, only the response times for the honestresponding condition were examined. We next included both the
honest and fake-good conditions in order to assess whether the
distribution of response times across response options varied
1
In an attempt to account for individual variability in the honest responding, fakegood, and fake-specic analyses, follow-up analyses were conducted in which
individual difference measures (NEO-PI-R dimension scores from the honest response
condition) were entered as predictors in the highest level model. Very few signicant
effects emerged and no consistent patterns were found. Thus, it did not appear that
the personality traits included in this study were associated with differential abilities
to fake.

signicantly as a function of the desire to make a generally good


impression. Examination of response times under the fake good
instructions required a three-level model. As in the previous analysis, the Level 1 model corresponded to response times per response options, but unlike the previous analysis, required an
additional subscript, s, indicating the nesting of response times under the set of instructions (honest or good impression) under
which they were collected.

RT rsi p0si p1si ROrsi p2si RO2rsi ersi


The Level 2 model examined the variability in the Level 1
parameters and included a variable to account for the instructions
under which responses were made (honest versus good impression). Therefore, the p coefcients now include variability across
conditions.

p0si b00s b01s Instructionsi r0si


p1si b10s b11s Instructionsi r1si
p2si b20s b21s Instructionsi r2si

M.K. Shoss, M.J Strube / Organizational Behavior and Human Decision Processes 116 (2011) 163171

169

The estimates of the xed coefcients are presented in Table 4.


Because the instruction condition was dummy coded with 0 assigned to the honest responding condition, the signicance of
the b20 coefcient corresponds to the curvature of the distribution
of response times for the honest response condition. As indicated
by the non-signicant b21 coefcients, response distributions for
the neuroticism, extraversion, and openness scales displayed a
signicant inverted-U shape in the fake-good condition that was
not signicantly different from the honest-responding invertedU. Furthermore, the nonsignicant b01 coefcients for all but
the openness scale indicate that overall response times in the
fake-good condition were not distinguishable from the honest
responding condition. Taken together, these results are most consistent with the adopted schema model. Agreeableness and conscientiousness were the only two scales for which the curvature
was different between the honest-responding and fake-good conditions, and in both cases the inverted-U pattern was more pronounced in the fake-good condition. Likewise, the only overall
response time difference occurred for openness and was faster
for the fake-good condition. These ndings are consistent with
the adopted schema model under the assumption that the fakegood instruction called to mind a rather specic image to be fabricated with regard to these dimensions (at least more specic
than the participants own self-concepts).2 These results are certainly inconsistent with the self-schema model and the semanticexercise model.

response instruction condition is compared to the honest response


condition (c211) and when it is compared to the fake-good condition (c221). Both parameters were signicant [c211, t(177) = 2.80,
p = .006; c221, t(177) = 2.11, p = .036]. These effects are displayed
graphically in Fig. 1. The fake-specic public relations instruction
resulted in a signicantly more distinct inverted-U distribution
than the fake-specic accountant instruction, t(118) = 2.56,
p = .012. This between-subjects difference only occurred in the
specic instruction conditions. The c010 coefcient tests whether
the honest responding condition had overall response times that
were different from the public relations condition. This coefcient
was signicant [t(58) = 2.16, p = .035] and indicated that honest
responding was signicantly slower. The c001 coefcient tests
whether the accountant condition had overall response times that
were different from the public relations condition. This coefcient
was signicant [t(58) = 2.04, p = .046] and indicated that the
accountant condition produced the fastest response times overall.
By implication, this condition was also faster than honest responding. These results support the predictions made by the adopted
schema model that there would be no difference in the quadratic
effect between the honest and fake-good conditions, that the
fake-specic public relations condition would exhibit the most
profound inverted-U relative to the other conditions, and that the
specic impression conditions would produce faster responding
than the honest response condition.3

Fake-specic response time analysis

Discussion

Extraversion response times for the specic instruction conditions were also examined in a three-level model that required a
modication at Levels 2 and 3. At Level 2, an additional parameter
was necessary to account for all three instruction conditions which
were dummy coded (specic: 0, 0; honest: 1, 0; fake-good: 0, 1):

The key parameter tests are for c010, c001, c211 and c221. The latter two indicate whether the inverted-U pattern is different between the two specic instruction conditions when the specic

The purpose of this study was to evaluate the validity of three


competing models of impression-managed responding to a personality testthe self-schema, adopted-schema, and semantic-exercise models. Distinguishing between and testing these different
models is crucial for determining whether response times, specically the inverted-U response time pattern, could be used to identify fakers. We assessed response time patterns for responses to the
NEO PI-R when participants were asked to respond honestly, make
a good impression, and make a specic impression. The latter condition was included to test nuanced predictions made by these previously hypothesized models of impression-managed responding.
The inverted-U response time pattern found for each of the
scales in the honest response condition replicates previous studies
(Akrami et al., 2007; Kuiper, 1981). In addition, it provides the
baseline for the analysis of response time differences between conditions. Our results support the predictions made by the adoptedschema modelthat all conditions would produce an inverted-U
response time pattern, the inverted-U would be more pronounced
for the fake-specic public relations condition, and response times
would be faster overall for the fake-specic conditions. Thus, it appears that individuals, when faking, access a schema of an ideal
respondent. This schema may be more or less clear depending on
whether one is trying to portray a particular impression or a more
general good impression.
The differences in the scale scores between conditions suggest
what the schemas used in faking might entail. The individual scale
scores for neuroticism, extraversion, agreeableness, and conscientiousness differed signicantly when individuals responded
honestly compared to when they responded to make a good
impression, as expected. The differences in extraversion scores
between the conditions were particularly interesting. Supporting
prior research ndings (Bagby & Marshall, 2003), extraversion
scores were higher when participants were responding to make a
good impression than when they were responding honestly. Those

2
The Level 3 model estimated individual variability in the Level 2 parameters. No
interpretable effects were found.

3
The Level 3 model also estimated individual variability in the Level 2 parameters.
No interpretable effects were found.

p0si b00s b01s Instruction 1si b02s Instruction 2si r0si


p1si b10s b11s Instruction 1si b12s Instruction 2si r1si
p2si b20s b21s Instruction 1si b22s Instruction 2si r2si
These b coefcients were modeled in a Level 3 model where the
fake-specic condition was dummy coded (public relations = 0;
accountant = 1). Recall that the fake-specic condition was a between-subjects manipulation where half of the participants were
instructed to respond as if applying for a job in public relations
and the other half were instructed to respond as if applying for a
job in accounting:

b00s c000 c001 Specifici u00i


b01s c010 c011 Specifici u01i
b02s c020 c021 Specifici u02i
b10s c100 c101 Specifici u10i
b11s c110 c111 Specifici u11i
b12s c120 c121 Specifici u12i
b20s c200 c201 Specifici u20i
b21s c210 c211 Specifici u21i
b22s c220 c221 Specifici u22i

170

M.K. Shoss, M.J Strube / Organizational Behavior and Human Decision Processes 116 (2011) 163171

Table 4
Analysis of Level-2 model (instruction as a predictor) xed effects estimation (with robust standard errors).
Fixed effect for:
Neuroticism

Intercept, p0i
Linear, p1i
Quadratic, p2i

Extraversion

Intercept, p0i
Linear, p1i
Quadratic, p2i

Openness

Intercept, p0i
Linear, p1i
Quadratic, p2i

Agreeableness

Intercept, p0i
Linear, p1i
Quadratic, p2i

Conscientiousness

Intercept, p0i
Linear, p1i
Quadratic, p2i

Coefcient

SE

df

b00
b01
b10
b11
b20
b21

(intercept)
(condition)
(intercept)
(condition)
(intercept)
(condition)

3625.38
141.09
61.09
181.34
134.92
48.89

92.87
109.27
27.13
65.10
22.86
48.90

39.04
1.29
2.25
2.79
5.90
1.00

59
118
118
118
118
118

.001
.199
.026
.007
.001
.320

b00
b01
b10
b11
b20
b21

(intercept)
(condition)
(intercept)
(condition)
(intercept)
(condition)

3484.35
119.94
42.37
112.47
148.06
33.79

98.52
116.75
26.74
47.89
27.96
41.62

35.37
1.03
1.59
2.35
5.30
0.81

59
118
512
512
118
118

.001
.307
.113
.019
.001
.419

b00
b01
b10
b11
b20
b21

(intercept)
(condition)
(intercept)
(condition)
(intercept)
(condition)

4080.90
406.43
40.68
71.66
179.03
20.04

99.07
91.01
40.43
42.07
34.54
42.19

41.19
4.47
1.01
1.70
5.18
0.48

59
118
118
118
118
118

.001
.001
.317
.091
.001
.635

b00
b01
b10
b11
b20
b21

(intercept)
(condition)
(intercept)
(condition)
(intercept)
(condition)

3513.86
143.10
20.12
85.51
86.85
90.84

91.76
95.41
36.18
50.15
30.05
42.08

38.29
1.50
0.56
1.71
2.89
2.16

59
118
59
118
59
118

.001
.136
.580
.090
.006
.033

b00
b01
b10
b11
b20
b21

(intercept)
(condition)
(intercept)
(condition)
(intercept)
(condition)

3688.12
153.91
85.45
31.97
88.71
109.91

96.96
111.30
32.49
64.66
33.62
53.79

38.04
1.38
2.63
0.49
2.64
2.04

59
118
118
118
118
118

.001
.169
.010
.621
.010
.043

in the fake-specic-public relations condition inated their extraversion scores above their fake-good scores. Those in the fake-specic accountant condition, however, reported essentially the same
level of extraversion as they did when they were responding honestly. This mirrors the results from the response-time analysis, and
suggests it may be more difcult to fake introversion than extraversion; that is, relatively more thought may need to go into selecting extreme responses. Perhaps it is difcult to fake introversion
because the general desire to make a good impression in any job
interview calls for a more outgoing and friendly demeanor. Therefore, the pressure to make a good impression and to appear to be
an ideal applicant for an accountant position may cancel each
other, resulting in baseline (honest responding) level scores and
a atter response time pattern than that for the fake-specic extraversion dimension. However, because this schema is likely less
complex than individuals self-schemas and schemas for a general
good impression, response times overall were faster. These results,
however, suggest that individuals are sensitive to stereotypes associated with different positions and may reference these stereotypes when engaging in impression-managed responding (Mahar
et al., 2006).
The analysis of the response time distributions for the impression-managed responses demonstrates that the inverted-U response time effect cannot be used to discern those who are
faking from those who are responding honestly. In fact, using an
individuals response time distribution could lead to false predictions on both the agreeableness and conscientiousness scales,
where the inverted-U effect was more pronounced in the fakegood condition than in the honest response condition. This suggests that a good-impression, in general, acts like a category and
individuals can access this category readily when responding. This
also suggests that the self-category is not privileged in any way, an
implication consistent with the adopted-schema model of impression-managed responding.

The prior literature on faking has suggested that a way to minimize the extent of faking would be to have individuals complete
the personality assessment with as few specics details of the
job made known as possible (Vasilopoulos et al., 2000). This study
suggests that doing so would not result in lower levels of faking. In
fact, for the conscientiousness scale, it could result in a higher level
of response distortion. It is also interesting to note that the standard deviations for the scale scores decreased in the fake good condition. This suggests that individuals shared a consolidated view of
not only what attributes are considered positive, but also at what
levels.
Practically, in addition to suggesting that response time distributions, such as the inverted-U, would be ineffective for detecting
faking, this study also casts doubt on the validity of response times
for detecting faking in general. As previously noted, response times
can depend on a number of different factors including attention
paid to each item, reading speed, time to search for information,
schema clarity, familiarity with questions, test anxiety, and so
forth. Moreover, in practical settings it is nearly impossible to acquire a baseline average response rate with which to compare, given the wide range of factors that can inuence an individuals
average response rate. Thus, it seems that researchers and practitioners interested in detecting and reducing faking would do well
to focus on other strategies.
Limitations
This study has a number of limitations that warrant additional
investigation. First, the fake-specic manipulation was not a direct
manipulation of extraversion. The manipulation used depended on
individuals stereotypes of these two positions, which are
inuenced by prior experience interacting with individuals in
these professions. Despite the fact that the strength of the manipulation depended on the participants schemas, the results suggest

M.K. Shoss, M.J Strube / Organizational Behavior and Human Decision Processes 116 (2011) 163171

that, on average, participants associated public relations with


extraversion and accounting with introversion. Therefore, this
study does appear to provide a manipulation, albeit imperfect, of
the extraversion dimension. In addition, the participants in our
sample were not actual applicants. It would be interesting to
examine response time distributions in samples for which the relevant schemas are especially salient and the application outcome
more consequential.
Second, due to the within-subjects design, participants in this
study responded to the personality items three times. All participants responded to the items in the honest responding condition
rst and to the other two conditions in random order. Therefore,
participants responses to the later conditions may have been
biased by their previous responses. However, if responses were
inuenced by prior responses then information should be easily
accessible because it was just retrieved. Therefore, we would expect that the distribution of response times would be at. The fact
that we found an inverted-U distribution despite any potential
practice effects suggests that our ndings are robust.
Additionally, given that the number of participants in this study
is somewhat small, albeit similar to that typically used in other
within-subjects designs (e.g., Browne, Curley, & Benson, 1999),
the power to detect quadratic effects may be reduced. Thus, this
made the statistical analysis relatively conservative. The fact that
the inverted-U was found for all conditions again suggests that this
effect is particularly robust.
In conclusion, this study suggests that individuals access a schema of an ideal respondent when responding to give a specic
impression as well as to give general good impression. As clearly
dened exemplars are easier to retrieve, response times across
response options have an inverted-U shape. This indicates that
response time distributions, such as the inverted-U, are not effective in detecting faking. We encourage more research on the cognitive processes involved in responding to questionnaires as well as
ways to mitigate or detect faking.

References
Akrami, N., Hedlund, L. E., & Ekehammar, B. (2007). Personality scale response
latencies as self-schema indicators: The inverted-U effect revisited. Personality
and Individual Differences, 43, 611618.
Bagby, R. M., & Marshall, M. B. (2003). Positive impression management and its
inuence on the revised NEO personality inventory: A comparison of analog and
differential prevalence group designs. Psychological Assessment, 15, 333339.
Birkeland, S. A., Manson, T. M., Kisamore, J. L., Brannick, M. T., & Smith, M. A. (2006).
A meta-analytic investigation of job applicant faking on personality measures.
International Journal of Selection and Assessment, 14, 317335.
Browne, G., Curley, S., & Benson, P. (1999). The effects of subject-dened categories
on judgmental accuracy in condence assessment tasks. Organizational Behavior
and Human Decision Processes, 80, 134154.
Casey, M. M., & Tryon, W. W. (2001). Validating a double-press method for
computer administration of personality inventory items. Psychological
Assessment, 13, 521530.
Costa, P. T., Jr., & McCrae, R. R. (1992). NEO PI-R professional manual. Odessa, FL:
Psychological Assessment Resources, Inc.

171

Grifn, B., & Hesketh, B. (2004). Why openness to experience is not a good predictor
of job performance. International Journal of Selection and Assessment, 12,
243251.
Grifn, B., Hesketh, B., & Grayson, D. (2004). Applicants faking good: Evidence of
item bias in the NEO PI-R. Personality and Individual Differences, 36, 15451558.
Grifth, R., & McDaniel, M. (2006). The nature of deception and applicant faking
behavior. In R. L. Grifth & M. H. Peterson (Eds.), A closer examination of
applicant faking behavior (pp. 120). Greenwich, Conn.: Information Age Publ.
Holden, R. R. (1998). Detecting fakers on a personnel test: Response latencies versus
a standard validity scale. Journal of Social Behavior & Personality, 13, 387398.
Holden, R. R., Fekken, G. C., & Cotton, D. H. (1991). Assessing psychopathology using
structured test-item response latencies. Psychological Assessment, 3, 111118.
Holden, R., & Hibbs, N. (1995). Incremental validity of response latencies for
detecting fakers on a personality test. Journal of Research in Personality, 29,
362372.
Holden, R. R., & Kroner, D. G. (1992). Relative efcacy of differential response
latencies for detecting faking on a self-report measure of psychopathology.
Psychological Assessment, 4, 170173.
Holtgraves, T. (2004). Social desirability and self-reports: Testing models of socially
desirable responding. Personality and Social Psychology Bulletin, 30, 161172.
Hough, L., Oswald, F., & Ployhart, R. (2001). Determinants, detection and amelioration of adverse impact in personnel selection procedures: Issues, evidence
and lessons learned. International Journal of Selection and Assessment, 9,
152194.
Hsu, L. M., Santelli, J., & Hsu, J. (1989). Faking detection validity and incremental
validity of response latencies to MMPI subtle and obvious items. Journal of
Personality Assessment, 53, 278295.
Komar, S., Brown, D., Komar, J., & Robie, C. (2008). Faking and the validity of
conscientiousness: A Monte Carlo investigation. Journal of Applied Psychology,
93, 140154.
Kuiper, N. A. (1981). Convergent evidence for the self as a prototype: The invertedU RT effect for self and other judgments. Personality and Social Psychology
Bulletin, 7, 438443.
Mahar, D., Coburn, B., Grifn, N., Hemeter, F., Potappel, C., Turton, M., et al. (2006).
Stereotyping as a response strategy when faking personality questionnaires.
Personality and Individual Differences, 40, 13751386.
McDaniel, M. A., & Timm, T. (1990). Lying takes time: Predicting deception in biodata
using response latency. In H. Wing (Chair), Alternative predictors. Symposium
presented at the 98th Annual conference of the American Psychological
Association, Boston, MA.
McFarland, L., Ryan, A., & Ellis, A. (2002). Item placement on a personality measure:
Effects on faking behavior and test measurement properties. Journal of
Personality Assessment, 78, 348369.
Mesmer-Magnus, J., & Viswesvaran, C. (2006). Assessing response distortion in
personality tests: A review of research designs and analytic strategies. In R. L.
Grifth & M. H. Peterson (Eds.), A closer examination of applicant faking behavior
(pp. 2142). Greenwich, Conn.: Information Age Publ.
Mueller, J. H., Thompson, W. B., & Dugan, K. (1986). Trait distinctiveness and
accessibility in the self-schema. Personality and Social Psychology Bulletin, 12,
8189.
Mueller-Hanson, R., Heggestad, E. D., & Thornton, G. C. (2003). Faking and selection:
Considering the use of personality from select-in and select-out perspectives.
Journal of Applied Psychology, 88, 348355.
Ones, D. S., Dilchert, S., Viswesvaran, C., & Judge, T. A. (2007). In support of
personality assessment in organizational settings. Personnel Psychology, 60,
9951027.
Osborne, Jason. W. (2000). Advantages of hierarchical linear modeling. Practical
Assessment, Research & Evaluation, 7, 13.
Popham, S. M., & Holden, R. R. (1990). Assessing MMPI constructs through the
measurement of response latencies. Journal of Personality Assessment, 54,
469478.
Vasilopoulos, N. L., Reilly, R. R., & Leaman, J. A. (2000). The inuence of job
familiarity and impression management on self-report measure scale scores
and response latencies. Journal of Applied Psychology, 85, 5064.
Viswesvaran, C., & Ones, D. S. (1999). Meta-analyses of fakability estimates:
Implications for personality measurement. Educational and Psychological
Measurement, 59, 197210.

You might also like