Professional Documents
Culture Documents
University of Utah, 1645 E. Campus Center Dr., Salt Lake City, UT 84112, United States
Idaho State University, 921 South 8th Ave, Pocatello, ID 83209, United States
a r t i c l e
i n f o
Article history:
Received 4 October 2011
Accepted 23 July 2013
Accepted by Richard Moreland
Keywords:
Condence
Expertise
Extroversion
Group decision making
Group performance
Shared cognition
Shared mental model
Transfer of learning
a b s t r a c t
Groups often struggle to distinguish expert members from others who stand out for various reasons but
may not be particularly knowledgeable (Littlepage & Mueller, 1997). We examined an intervention
designed to improve group decision making and performance through instructing group members to
search for information they already possessed that was relevant to a problem. Participants estimated values and expressed their condence in their estimates individually and then a second time either individually or in a group. This was done with or without the intervention. Results indicated that: (1) groups
were more condent than, and out-performed, individuals, (2) group decision making was best captured
by models predicting more inuence for more accurate members when the intervention was used and
more inuence for more condent members in its absence, and (3) groups that received the intervention
out-performed groups that did not.
2013 Elsevier Inc. All rights reserved.
Introduction
In a group, each member brings different expertise, and condence in that expertise, to the task. However, to be successful, this
expertise must be properly utilized (Faraj & Sproull, 2000; Littlepage, Hollingshead, Drake, & Littlepage, 2008; Stasser, Stewart, &
Wittenbaum, 1995). So, the challenge in many problem-solving situations is how to leverage the knowledge of group members to
solve a given problem (Lightle, Kagel, & Arkes, 2009; McGrath,
1984), or how problem solvers can bridge the gap between what
they already know and what they are trying to gure out (Hummel
& Holyoak, 1997).
To effectively leverage their knowledge, group members must
be capable of distinguishing member inputs in terms of their accuracy. This may be difcult because some of the more salient factors
that seem to signal an individuals competency do not consistently
correlate with actual accuracy (Bonner, Sillito, & Baumann, 2007;
Littlepage, Schmidt, Whistler, & Frost, 1995; Trotman, Yetton, &
Zimmer, 1983). Importantly, research has shown that member
condence, irrespective of its merit, can serve as a compelling
proxy for member accuracy (Sniezek, 1989), as can member extroversion (Bonner, 2000). The problem, then, is how group members
Corresponding author.
E-mail addresses: mgtblb@business.utah.edu (B.L. Bonner), boliale2@isu.edu
(A.R. Bolinger).
0749-5978/$ - see front matter 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.obhdp.2013.07.005
can demonstrate the actual merits of their preferences to one another in a context lled with potentially misleading cues (Laughlin
& Ellis, 1986).
We examined how a knowledge transfer intervention aimed at
increasing the ability of group members to demonstrate the quality
of their preferences affects the decision-making process of cooperative groups. This intervention involves accessing knowledge already held by group members; it does not require external
sources of information (e.g., performance feedback) that may be
difcult or costly to acquire (e.g., Ashford, Blatt, & VandeWalle,
2003; Bonner, Baumann, & Dalal, 2002) or intensive coaching
(e.g., Hackman & Wageman, 2005a). It could thus be implemented
across a wide range of real world problem-solving tasks. We tested
the impact of member accuracy, condence, extroversion, and
other factors on decision making with and without this intervention and examined how the intervention affects performance and
if social context moderates its effectiveness.
Enhancing demonstrability through knowledge transfer
People solve unfamiliar problems by transferring knowledge
they possess from past learning to the new situation (Blanchette
& Dunbar, 2001; Ellis, 1965; Haslered, 1972; Holyoak & Thagard,
1997; Kolodner, 1997; Nokes-Malach, 2009). The dening feature
of knowledge transfer interventions is that they help people recognize the potential usefulness of knowledge in the context of an
B.L. Bonner, A.R. Bolinger / Organizational Behavior and Human Decision Processes 122 (2013) 214221
215
216
B.L. Bonner, A.R. Bolinger / Organizational Behavior and Human Decision Processes 122 (2013) 214221
Hypothesis 5. Groups receiving the knowledge transfer intervention will perform better than groups not receiving it.
Participants
Hypothesis 6. Groups receiving the knowledge transfer intervention in an interactive group context will perform better than
groups receiving it prior to the group interaction.
Methods
B.L. Bonner, A.R. Bolinger / Organizational Behavior and Human Decision Processes 122 (2013) 214221
the estimate is made). To this end, we selected items that we believed would promote a decision-making process with some level
of uncertainty, as opposed to using items that participants were
likely to know off-hand or that seemed impossibly difcult or futile
(Davis, 1982). To satisfy these criteria, we used estimates grounded
in knowledge accessible to participants in our sample, including
world knowledge questions and topics of local interest (Bonner &
Baumann, 2012). The items we used were: (1) What is the minimum freeway driving distance from Salt Lake City to New York
City? (answer: 2,174.41 miles), (2) How many students are enrolled at the University of Utah? (29,012 students), (3) How much
did the Guinness record holder for heaviest person of all time
weigh? (answer: 1400 lbs.), (4) What is the population of Utah?
(answer: 2,389,039 people), and (5) What is the elevation of Kings
Peak (i.e., the highest point in Utah, answer: 13,528 ft.)?
Design
This two-phase study used an Individual-Individual vs. Individual-Group (II vs. IG) design. All participants completed Time 1
estimations individually after which the experimenter collected
their materials. Participants could allocate their time among items
as they saw t. Then half of the participants repeated the estimations at Time 2 as part of a 3-person group, whereas the remaining
half repeated the estimates individually. This allowed for an appropriate comparison between group and individual problem solvers
at Time 2. This was critical because comparing group performance
to the performance of constituent members or to individuals seeing the problem for the rst time would have confounded the comparison. This method also facilitated the modeling of Time 2 group
decisions from Time 1 member preferences. Crossed with this
manipulation, one third of participants generated associated
knowledge items (i.e., received our intervention) at Time 1, one
third did so at Time 2, and the remaining third did not generate
associated knowledge items. Thus, the overall design of the study
was a 2 (problem solver: II vs. IG) 3 (knowledge intervention:
no intervention, Time 1, Time 2) factorial design. Participants took
part in the study in clusters of up to 16 people and were assigned
to condition and to a specic group (if applicable) at random.
Groups consisted of members from the same experimental cell.
Data were collected over the course of one academic year.
Time 1
We manipulated whether participants were encouraged to consider their associated knowledge relevant to the estimates at Time
1, at Time 2, or not at all. All participants were given the following
instructions at Time 1:
In this study you will be working to solve a series of ve knowledge problems. Most people will not know the exact answers to
these questions off the top of their heads. It is your task to provide
the best answers you can, that is, answers that are as close as possible to the correct response.
Participants in the no-intervention condition and the Time 2
knowledge transfer condition then proceeded to answer the ve
estimation items. In addition to receiving the information above,
those in the Time 1 knowledge transfer condition also received
the following instructions:
You may not know the answer to a given question, but you may
know other information that is related to the question. For each
question, generate two items of associated knowledge to help
you answer as accurately as possible. Your associated knowledge
can take any form you choose. Come up with your associated
knowledge before you answer each question.
Time 2
After generating their Time 1 estimates (and, when applicable,
associated knowledge) group members were assigned to small
217
break-out rooms where they could interact freely. Individuals remained in a large central room. Participants in the group conditions were not required to make their Time 1 estimates public
during the Time 2 phase of the experiment, although they could
elect to do so. Both individuals and groups were instructed to provide their best answers to the estimates and to respond with a
quantitative value for each of the ve items. Again, participants
could allocate their time among items at their discretion. Prior to
providing estimates, those in the Time 2 knowledge intervention
condition were given the same additional instructions as those given at Time 1 to participants in the Time 1 knowledge intervention
condition. Those assigned to groups were also informed that they
should discuss the ideas until they reached consensus on each one.
After completing their Time 2 estimates, participants lled out a
short questionnaire (Goldberg, 1992) and were debriefed. The
questionnaire was used to provide the basis for testing the effects
of member extroversion. Because extroversion has been shown to
be a stable trait (e.g., Costa & McCrae, 2006; Magnus, Diener, Fujita,
& Pavot, 1993) this variable was measured after the problem solving task (Bonner, 2000; Bonner et al., 2007). Our debrieng involved informing participants of the purpose of the study,
providing them with contact information for both the experimenter and the local institutional review board, and giving them
the opportunity to ask questions. The entire study lasted 1 h.
Results
Associated knowledge
Two hundred and sixteen of the participants in this study (twothirds) were assigned to conditions requiring that they generate
associated knowledge prior to providing estimates. Those in the
individual cells (N = 108) generated two units of associated knowledge for each of the ve items, for a total of 1080 units. Those in
the Time 1 group cell (N = 54) also generated two units of associated knowledge per group member for each item, for a total of
540 units. Those in the Time 2 group cell (N = 54) generated a total
of two units of associated knowledge per group for each item, for a
total of 180 units.
We predicted (Hypothesis 1) that a higher proportion of the
associated knowledge generated in an interactive group context
would be of the strategy/process variety compared to the associated knowledge generated by comparison individuals. These associated knowledge statements typically provided background to
support the validity of an estimate (e.g., I hiked Kings Peak last
summer so I know how high it is), or what Henry (1993) refers
to as credibility statements. Some participants also identied a
problem-solving strategy (e.g., record-holding statistics are usually higher than youd expect so add more to the heaviest person
of all time estimate). All statements were independently sorted
by two coders who were blind to the hypotheses of this research.
The coders were trained and provided with numerous examples
(from a separate data set) of statements that did and did not fall
into the strategy/process category. They sorted these practice
statements and were provided with feedback. Each coder then
independently coded all associated knowledge from the current research. The agreement between coders was acceptable, unweighted Cohens j = .79 (Cohen, 1968).
An examination of the differences between the proportions of
strategy/process statements in different cells of the design revealed that the associated knowledge generated in a group interaction (i.e., by Time 2 groups) included a higher proportion of
process/strategy statements than did the associated knowledge
generated by comparison individuals (i.e., Time 2 individuals).
Among participants who generated associated knowledge in the
218
B.L. Bonner, A.R. Bolinger / Organizational Behavior and Human Decision Processes 122 (2013) 214221
Social combination
As we have discussed, group members may be differentially
inuential within groups as a function of their relative levels of
accuracy, condence, or other factors including centrality, extremity, or extroversion. To assess the impact of these potential sources
of inuence, we represented each as a distinct quantitative social
decision scheme (Davis, 1973; Hinsz, 1999). The approach we used
to build and test our models is detailed in Bonner et al. (2004) and
is based largely on the social judgment scheme model (Davis,
1996). For the accuracy, centrality, and extremity models, we calculated individual member intra-group inuence weights for each
model on the assumption that as the distance between a members
proposal and the criterion of interest (i.e., the correct response or
the intra-group mean) increased, that members inuence in the
group decision would adjust. For the condence model, we assumed that as member condence increased (as reected by narrower credibility intervals), their inuence within the group
increased (c.f., Stephenson et al., 1986). Finally, for the extroversion model, more extroverted members were assumed to be inuential relative to less extroverted members (Bonner et al., 2007).
All models are presented in Appendix A. Each model was used to
generate predictions as to what the groups would decide at Time
2, based on the preferences of group members at Time 1. The t
of these point predictions to the actual obtained group estimates
was assessed using KolmogorovSmirnov one-sample tests (Kolmogorov, 1941). An alpha level of .20 is the standard for KolmogorovSmirnov tests involving group decision making and
constitutes a conservative test of t (e.g., Davis, 1996).
The results for the t tests for the no-intervention, Time 1
knowledge transfer intervention, and Time 2 knowledge transfer
intervention conditions are summarized in Table 1. As the table
shows, in both intervention conditions all of the models were rejected, except for the accuracy model. This nding supported
Hypothesis 2. In the no-intervention condition, the condence
model was the only model not rejected. This nding supported
Hypothesis 3.
Performance
4.5
No Intervention
Table 1
Model tests.
Condition
Estimation models
4.0
Time 1 Intervention
Time 2 Intervention
3.5
One challenge in group research is ensuring an appropriate performance standard against which group performance can be compared. To this end, we gathered data from an equivalent number of
individual problem solvers (compared to the members of threeperson groups) for each of the three group conditions. Thus, we
could directly compare group performance to the performance of
the best, median, and third-best of equivalent members of randomly assembled pseudo-groups.
3.0
2.5
2.0
1.5
1.0
0.089*
0.171
0.156
0.133
0.271
0.167
0.211
0.211
0.167
0.156
0.305
0.167
0.5
0.0
Group
1st Ind.
2nd Ind.
3rd Ind.
Fig. 1. Group and individual performance. Lower scores indicate greater accuracy.
B.L. Bonner, A.R. Bolinger / Organizational Behavior and Human Decision Processes 122 (2013) 214221
Discussion
In keeping with the results of a number of published estimation
studies (e.g., Bonner et al., 2007; Laughlin et al., 1999, 2003; Sniezek & Henry, 1989), groups in the current study outperformed
average (in this case, median) comparison individuals, but not
the best comparison individuals. This was the expected level of
performance given the nature of quantity estimation tasks in general (Hastie, 1986). Also in line with past research, groups in our
study demonstrated more condence in their estimates than did
individuals (Sniezek, 1992). Thus, our ndings on the general effects of individual versus group performance and condence
yielded results consistent with past research. This serves to embed
our study in the extant literature.
Many group problem-solving interventions involve either
externally provided data (e.g., cues or explicit frames of reference,
Laughlin et al., 1999), externally provided information about the
group members (e.g., performance feedback, Bonner et al., 2002),
or instructions to focus on the group in a certain way (e.g., to nd
the best member, Henry, 1995). Our study took a different and theoretically distinct approach by focusing members either on their
own internal knowledge or on the knowledge of their group as a
whole in an attempt to improve the accuracy of estimates. The
knowledge transfer intervention we used is light on resources
and easy to implement. compared to providing performance feedback (Bonner et al.) or extensive coaching (Hackman & Wageman,
2005a), making it widely applicable.
We focused primarily on two forms of inuence in groups,
namely accuracy-based inuence and condence-based inuence, but we also examined other potentially inuential factors,
such as extroversion. Because accuracy and condence were only
modestly related to one another in this study, the predictions of
these two models diverged, allowing the t of the models to be
distinguished in the different conditions of the study. The decision patterns of both group conditions receiving instructions to
generate associated knowledge (i.e., either at Time 1 or at Time
2) were t by models predicting that more accurate proposals
would carry more weight in the group estimate. All other models
were rejected for these conditions. However, for groups in the nointervention condition, group decisions were t by a model
assuming member inuence as a function of condence, with
all other models being rejected. The t of the condence model
in the absence of the intervention implies that when demonstrability is relatively low group members may use proxies for
accuracy.
Participants produced their own cues or frames of reference as
part of the task. We found that when group members generated
knowledge interactively, as opposed to generating it individually
before entering the group, they were more likely to use the associated knowledge to provide strategies for the group or to facilitate
processes. This is an important nding, because it shows that the
interactive groups were, at least to some extent, actively managing
the complexities of their task. Our distinction between strategy/
process and other statements is similar to Henrys (1993) distinction between task-relevant statements (e.g., I know that Beethoven was born after Bach) and credibility-relevant statements
(e.g., I visited Beethovens birthplace when I was in Germany).
Hastie (1986) also makes a similar distinction between evidence
information (e.g., facts of a case going to trial) and opinion infor-
219
mation (e.g., how other members of the jury are planning to vote).
Our research shares the assumption of this prior work that information need not be purely factual to affect group decision making.
Process or strategy statements may facilitate shared expectations
and understanding about the task (Thompson & Fine, 1999; Weingart, 1992).
We found that better performance was achieved when the
consideration of member knowledge occurred in an interactive
group context (i.e., at Time 2 rather than Time 1). A partial explanation for this may be that the groups use of more process-oriented associated knowledge led to better coordination of their
inputs. An alternative explanation might be that by generating
associated knowledge collectively group members further increased their joint understanding of the task and its associated
conceptual systems, resulting in greater task demonstrability
(Laughlin & Ellis, 1986; Orasanu, 1994). Groups may also have
benetted from starting the interactive task unencumbered by
earlier judgments (Reimer, Reimer, & Hinsz, 2010). Future research seeking to unpack these potential causal agents could
use self-report measures of group coordination and shared
understanding from participants (Lewis, 2003; Weingart, 1992),
along with manipulating whether or not prior judgments are generated (to help account for potential anchoring effects; see Tversky & Kahneman, 1981). There is also evidence that interventions
in general may have greater efcacy once the group begins its
work together rather than before (Hackman & Wageman,
2005a). Previous research suggests that groups are more open
to interventions during, rather than before, the interactive
phases, because group members gain a better appreciation of
the task and discuss task strategies as well as relevant facts and
information (Hackman & Wageman, 2005b; Woolley, 1998).
This research has important implications for the way that we
manage knowledge workers. Our ndings indicate that encouraging workers to access the knowledge that they will use in solving
problems as part of the interactive group process leads to more
accurate outcomes. We argue that this benet comes from increasing the ease with which group members can communicate effectively about the merits of different member inputs, which
follows from a better shared understanding of the problem. This
implies that management approaches that put all the information
on the table and then ask groups to solve a problem may be preventing groups from reaching their full potential. Our research suggests that a better approach might be to allow groups to conduct
their information search (through whatever means are available)
as the rst part of the problem-solving process. This would allow
them to gain a deeper understanding of the problem space, as well
as the competencies of various group members. Our recommendations are consistent with ndings in literatures on shared mental
models and transactive memory that suggest multiple functions
(e.g., development of shared understandings about who knows
what in a group, agreement on member expertise, and learning
that may transfer to related tasks) that occur during the interactive
phases of group activity (Lewis, Lange, & Gillis, 2005; Liang, Moreland, & Argote, 1995; Littlepage, Robison, & Reddington, 1997).
Appendix A.
Assume that group decisions (Gi) are weighted sums of r individual preferences on i problems, denoted xij. Further, let cij denote
the weight of the jth member on the ith problem, with the sum of
all cij totaling 1. Or, [Gi = ci1xi1 + ci2xi2 + cirxir]. The accuracy inuence equation (Bonner, Gonzalez, & Sommer, 2004) is dened as,
ejdijj
cij Pr jdijj
j1 e
220
B.L. Bonner, A.R. Bolinger / Organizational Behavior and Human Decision Processes 122 (2013) 214221
where dij is a discrepancy score, ((xij the correct answer)/the correct answer) and e represents the natural log e. The condence
inuence equation (c.f., Stephenson et al., 1986) is dened as,
ejpijj
cij Pr jpijj
j1 e
ejZijj
cij Pr jZijj
j1 e
where Zij is a standardized value that reects distance from the intra-group average, ((xij intra-group mean)/intra-group standard
deviation) and e represents the natural log e. The extremity inuence equation (Bonner & Baumann, 2012) is dened as,
ejZijj
cij Pr jZijj
j1 e
where Zij is a standardized value that reects distance from the intra-group average, ((xij intra-group mean)/ intra-group standard
deviation) and e represents the natural log e. Finally, the extroversion equation (Bonner et al., 2007) is dened as,
eEj
cj Pr Ej
j1 e
B.L. Bonner, A.R. Bolinger / Organizational Behavior and Human Decision Processes 122 (2013) 214221
Laughlin, P. R., Bonner, B. L., Miner, A. G., & Carnevale, P. J. (1999). Frames of
reference in quantity estimations by groups and individuals. Organizational
Behavior and Human Decision Processes, 80, 103117.
Laughlin, P. R. (1980). Social combination process of cooperative, problem-solving
groups at verbal intellective tasks. In M. Fishbein (Ed.), Progress in social
psychology. Hillsdale, NJ: Erlbaum. Vol. 1, pp. 127-155.
Laughlin, P. R., & Ellis, A. L. (1986). Demonstrability and social combination
processes on mathematical intellective tasks. Journal of Experimental Social
Psychology, 22, 177189.
Laughlin, P. R., Gonzalez, C. M., & Sommer, D. (2003). Quantity estimations by
groups and individuals: Effects of known domain boundaries. Group Dynamics:
Theory, Research., and Practice, 7, 5563.
Laughlin, P. R., Hatch, E. C., Silver, J. S., & Boh, L. (2006). Groups perform better than
the best individuals on letters-to-numbers problems: Effects of group size.
Journal of Personality and Social Psychology, 90, 644651.
Lewis, K. (2003). Measuring transactive memory systems in the eld: Scale
development and validation. Journal of Applied Psychology, 88, 587604.
Lewis, K., Lange, D., & Gillis, L. (2005). Transactive memory systems, learning, and
learning transfer. Organization Science, 16, 581598.
Liang, D. W., Moreland, R. L., & Argote, L. (1995). Group versus individual training
and group performance: The mediating role of transactive memory. Personality
and Social Psychology Bulletin, 21, 384393.
Lightle, J. P., Kagel, J. H., & Arkes, H. R. (2009). Information exchange in group
decision making: The hidden prole problem reconsidered. Management
Science, 55, 568581.
Lim, R. G. (1994). Eliciting condence intervals within the context of the revision
and weighting model of group consensus judgment. Organizational Behavior and
Human Decision Processes, 59, 348370.
Littlepage, G. E., Hollingshead, A. B., Drake, L. R., & Littlepage, A. M. (2008).
Transactive memory and performance in work groups: Specicity,
communication, ability differences, and work allocation. Group Dynamics:
Theory, Research, and Practice, 12, 223241.
Littlepage, G. E., & Mueller, A. L. (1997). Recognition and utilization of expertise in
problem-solving groups: Expert characteristics and behavior. Group Dynamics,
Theory, Research, and Practice, 1, 324328.
Littlepage, G. E., Robison, W., & Reddington, K. (1997). Effects of task experience and
group experience on group performance, member ability, and recognition of
expertise. Organizational Behavior and Human Decision Processes, 69, 133147.
Littlepage, G. E., Schmidt, G. W., Whistler, E. W., & Frost, A. G. (1995). An inputprocess output analysis of inuence and performance in problem-solving
groups. Journal of Personality and Social Psychology, 69, 877889.
Lord, R., & Maher, K. (1993). Leadership and information processing. New York:
Routledge.
Lorge, I., & Solomon, H. (1955). Two models of group behavior in the solution of
Eureka-type problems. Psychometrika, 20, 139148.
Magnus, K., Diener, E., Fujita, F., & Pavot, W. (1993). Extraversion and neuroticism as
predictors of objective life events: A longitudinal analysis. Journal of Personality
and Social Psychology, 65, 10461053.
Masui, C., & De Corte, E. (1999). Enhancing learning and problem solving skills:
Orienting and self-judging, two powerful and trainable learning tools. Learning
Instruction, 9, 517542.
McGeoch, J. A. (1942). The psychology of human learning: An introduction. New York:
Longmans.
McGrath, J. E. (1984). Groups: Interaction and performance. Englewood Cliffs, NJ:
Prentice-Hall.
McKeough, A., Lupart, J., & Marini, A. (1995). Teaching for transfer: Fostering
generalization in learning. Mahwah, NJ: Lawrence Erlbaum Associates.
Nokes-Malach, T. J. (2009). Mechanisms of knowledge transfer. Thinking &
Reasoning, 15, 136.
Nokes-Malach, T. J., VanLehn, K., Belenky, D. M., Lichtenstein, M., & Cox, G. (2012).
Coordinating principles and examples through analogy and self-explanation.
European Journal of Psychology of Education, 27, 127.
Orasanu, J. M. (1994). Shared problem models and ight crew performance. In N.
Johnston, N. McDonald, & R. Fuller (Eds.), Aviation psychology in practice
(pp. 255285). Hants: Avebury Technical.
221