Professional Documents
Culture Documents
Jurnal 2
Jurnal 2
Received December 30, 1998; revised November 1, 1999; accepted December 13, 1999
This study investigated the effects of individual experience versus mixed individual and
group experience on individual and group performance in rule induction. Random series
of eight rule-induction tasks were performed by sets of three participants randomly
assigned to either individual training (performing all tasks in a nominal group) or to mixed
training (alternating nominal and collaborative group task performance). Subsequent
individual and group performance were measured. A general positive group-to-individual
transfer was shown. In the individual posttest, mixed-training participants performed error
checking more promptly and as a result generated fewer nonplausible final solutions than
did individual-training participants. In the group posttest, mixed-training groups were
superior in collective error checking and more effective in collective truth detection than
were individual-training groups. Discussion focuses on how group experience in rule
induction influences group member learning and subsequent group performance. © 2000
Academic Press
Groups are often employed for problem solving and decision making in
organizations because it is assumed that group task performance typically sur-
passes that of an individual working alone. This assumption has received em-
pirical support for a wide range of problem-solving tasks (e.g., Hill, 1982). The
group superiority over the average individual can be explained by the greater
number of resources in a group that allows for combination of individual
contributions. When group performance is evaluated against normative standards
The research reported in this article was made possible by grants from the German Research
Community (DFG, Fr 472/16-3/4) to the first author. We gratefully acknowledge the comments from
Elisabeth Brauner, Caroline Bush, Dieter Frey, John Levine, Stefan Schulz-Hardt, Michael West, and
three anonymous reviewers on an earlier version of this article and thank those who participated in
the study.
Address correspondence and reprint requests to Felix C. Brodbeck, Department of Psychology,
Ludwig-Maximilians Universitaet, Leopoldstrasse 13, 80802 Munich, Germany. E-mail:
Brodbeck@psy.uni-muenchen.de.
621
0022-1031/00 $35.00
Copyright © 2000 by Academic Press
All rights of reproduction in any form reserved.
622 BRODBECK AND GREITEMEYER
with demonstrably correct solutions. They reasoned that the ease with which
effective rules or strategies can be made evident and adopted by other group
members might determine the occurrence of general positive G-I transfer. They
further argued that the reason why general positive G-I transfer cannot be
demonstrated for rule induction is that in these tasks effective strategies cannot
be made evident easily and thus are not adopted by other group members.
In our view the failure to demonstrate general positive G-I transfer in rule
induction may have two other causes. First, in the previous studies of rule
induction, only one or two group tasks were induced. As shown in the work of
McGlynn and Schick (1973), it may take more than one or two tasks for learning
effects to occur. Second, previous studies mainly focused on strategies for
detecting correct solutions rather than on strategies for rejecting nonplausible
solutions (e.g., Crott, Giesel, & Hoffmann, 1998; Hollingshead, 1998; Laughlin,
Bonner, & Altermatt, 1999; Laughlin, Magley, & Shupe, 1997). As is described
in more detail below, strategies for rejecting nonplausible solutions in rule
induction should be easy to demonstrate, and thus, according to Stasson et al.’s
(1991) reasoning, they should lead to general G-I transfer.
Rule induction is described by Laughlin and Hollingshead (1995) as the search
for descriptive, predictive, and explanatory generalizations, rules, and principles.
Individuals or members of a group observe patterns and regularities in a partic-
ular domain and propose hypotheses to account for them. They then evaluate the
hypotheses by observation and experiment and revise them accordingly. Re-
search teams, medical diagnostic teams, or audit teams conducting analytical
reviews use collective rule induction in various applied fields (cf. Laughlin,
VanderStoep, & Hollingshead, 1991). For rule-induction tasks as described by
Laughlin et al. there is an unlimited set of plausible hypotheses to start with. This
needs to be reduced via systematic testing of evidence during multiple task steps.
As the constraints imposed by the accumulation of evidence reduce the number
of possible hypotheses, the correct hypothesis (arbitrarily selected by the re-
searcher) acquires distinctiveness because of its relative simplicity or elegance
compared to other hypotheses that are also consistent with the evidence (Mc-
Glynn, Tubbs, & Holzhausen, 1995).
Demonstrability of hypothesis quality in rule induction. Inductive tasks are
both intellective and judgmental. For inductive tasks, any proposed hypothesis is
either plausible or nonplausible. Plausible hypotheses are consistent with avail-
able evidence, whereas nonplausible hypotheses are inconsistent with available
evidence. A proposed nonplausible hypothesis can be demonstrated to be incor-
rect by failing to correspond to the available evidence. As Laughlin and Holl-
ingshead (1995) describe it, this aspect of an induction task is intellective because
incorrectness can be made evident to other persons in a group context (high
demonstrability). One of the plausible hypotheses may be arbitrarily designated
as correct by a researcher in an experiment or by social consensus outside of an
experiment. A proposed plausible hypothesis cannot be demonstrated to be
uniquely correct relative to any other plausible hypothesis that is also consistent
624 BRODBECK AND GREITEMEYER
with the available evidence. This aspect of rule induction is judgmental because
correctness cannot be unequivocally demonstrated (low demonstrability).
Demonstrability of strategies for optimal rule induction. Inductive reasoning
involves learning general systematic approaches or algorithms for hypothesis
generation and selection of instances to test the hypotheses generated, called
strategies (Bruner, Goodnow, & Austin, 1956; Johnson, 1971; Laughlin, 1973).
For the subtask of generating plausible and rejecting nonplausible hypotheses
in multistep rule-induction tasks, we assumed that rather simple strategies of
error checking suffice to ensure that the likelihood of a nonplausible final
solution is reduced. Error checking is defined as the monitoring of interim
hypotheses for their correspondence with the available evidence. Error checking
can be performed in two ways, either by anticipative error avoidance or by post
hoc error management (cf. Frese & Zapf, 1994; Zapf, Brodbeck, Frese, Peters,
& Prümper, 1992). In the context of rule induction, error avoidance means
thoroughly checking an imagined hypothesis for potential errors before expli-
cating it; that is, before actually testing it (in the case of individual rule induction)
or before suggesting it to other group members (in the case of collective rule
induction). Error management means checking for potential errors when the
consequences of hypothesis explication are apparent, as when, for example,
actual hypothesis testing resulted in unexpected evidence (in the case of indi-
vidual rule induction) or when the communication of one’s individual hypothesis
to other group members resulted in unexpected disagreement (e.g., when group
members say, “The hypothesis you suggested can’t be plausible because of the
evidence obtained so far!”). In both instances an unexpected consequence casts
some doubt about the plausibility of the hypothesis under consideration and post
hoc checking for its correspondence with the available evidence seems to be a
sensible strategy for ensuring that a large number of plausible hypotheses are
tested in the course of the overall rule-induction task. Both types of error
checking (error avoidance and error management) are preconditions for optimal
performance of multistep rule-induction tasks. The more a correspondence be-
tween interim hypotheses and the available evidence is ensured, the more
efficiently can rule induction proceed. Furthermore, post hoc error management
is a strategy with high demonstrability in the group collaborative context because
it becomes self-evident during social episodes of mutual error checking. In
contrast, error avoidance is a strategy with low demonstrability in the group
collaborative context because it is performed privately (i.e., before hypothesis
explication), and thus it is not evident to other group members during collective
rule induction.
For the subtask of inducing correct hypotheses in multistep rule induction,
various strategies have been investigated in the literature (e.g., Laughlin et al.,
1997). Here we concentrate on the so-called “focusing” strategy, which describes
an effective and complex attribute-testing procedure for concept attainment
(Bruner et al., 1956; Laughlin, 1968; McGlynn & Schick, 1973). Focusing is
defined as a two-step attribute-testing strategy. A test of an initially held hypoth-
RULE INDUCTION 625
esis leads to the generation of a new hypothesis in which at least one attribute of
the original hypothesis has changed. A positive test result (i.e., confirmation of
the new hypothesis) indicates that the changed attribute is irrelevant, whereas a
negative instance (i.e., disconfirmation of the new hypothesis) indicates that the
changed attribute is relevant for hypothesis development (Laughlin, 1968,1973).
Complex hypothesis testing strategies, such as focusing, are less self-evident and
demonstrable in a collaborative group context than error-checking strategies such
as error management. As a more or less direct manifestation of Popper’s
Falsification Principle (Johnson, 1971; Klayman & Ha, 1987; Laughlin, 1973),
focusing is comparatively complex and difficult to understand. For a previously
unknown task context, as the rule-induction paradigm from Laughlin et al. (1991)
can be for university student participants, verbal explanation, extra effort, and
time are required to demonstrate the profitable application of focusing to others
in a group collaborative context. On the other hand, error management is
self-evident in the group collaborative context. Hypotheses that are proposed to
the group by one group member can be “checked” by others. When an incon-
sistency is diagnosed and pointed out in public, the underlying error-checking
strategy (and its effectiveness) is apparent and does not need to be further
elaborated or explicitly discussed with others.
In summary, generating plausible solutions in multistep rule-induction tasks is
assumed to be subject to general G-I transfer because in the group collaborative
context, effective error-checking strategies apparent in social episodes are self-
evident (high demonstrability) and thus can be adopted by group members. In
contrast, generating correct solutions is not assumed to be subject to general G-I
transfer because effective strategies of hypothesis testing, such as focusing, are
not self-evident during collective task performance (low demonstrability).
Hypothesis 1: General G-I transfer in rule induction occurs for error checking rather than
for hypothesis testing strategies.
Because collective error checking should increase the rejection rate of non-
plausible interim hypotheses during multistep collective induction, we further
expect that group experience improves the quality of final solutions (i.e., lower
proportions of nonplausible final group solutions).
In summary, the present study addresses assumptions derived from prior
research and theorizing about small-group problem solving and group-member
learning. The basic assumption is that the demonstrability of effective task
performance strategies predicts general positive G-I transfer. More specifically,
we test the hypothesis that collaborative rule induction facilitates individual
learning of error-checking strategies rather than of hypothesis testing strategies.
Furthermore, it is expected that whole groups benefit from experience in collec-
tive induction. More specifically, groups experienced in collective induction
perform collective error checking better than inexperienced groups.
RULE INDUCTION 627
TABLE 1
Experimental Design
Pretest Posttest
Training condition Nominal Training phase Nominal Group
Individual training I I I I I I I I I G
Mixed training I G I G I G I G I G
Task 1 2 3 4 5 6 7 8 9 10
METHOD
Participants and Design
Participants were 132 students (48 women and 84 men) from German univer-
sities and technical colleges in Munich who received DM 30 (about $16) for their
participation. The experimental design was a mixed-factorial design with indi-
vidual training versus mixed individual and group training as a between-group
factor and repeated measurement (pretest and posttest) as a within-group factor.
Three stages of the experiment are distinguished (see Table 1): (a) the pretest at
Task 1 for measuring individual performance in a nominal group setting, (b) the
training phase from Task 2 to Task 8 in which the experimental conditions were
introduced, and (c) the posttest measuring individual performance in a nominal
group setting (Task 9) and whole-group performance in a collaborative setting
(Task 10).
Experimental task. The rule-induction paradigm as described in Laughlin et al.
(1991) was used (see also Hollingshead, 1998; Laughlin & Hollingshead, 1995).
A rule had to be induced that partitioned a deck of 52 playing cards with four
suits (clubs ⫽ C, diamonds ⫽ D, hearts ⫽ H, spades ⫽ S) of 13 cards (ace ⫽ 1,
two ⫽ 2, . . . , jacks ⫽ 11, queens ⫽ 12, kings ⫽ 13) into examples and
nonexamples of the rule. The instructions indicated that the rule could be based
on suit, number, color (red ⫽ r, black ⫽ b), or any combination of numerical and
logical operations on these attributes (e.g., odd ⫽ o, even ⫽ e). The rule sequence
length consisted of either three or four cards. Each trial began with a card that
was known to be an example of the rule. There were four types of rules: (a)
combination of suits (e.g., S-S-H-C), (b) combination of colors (e.g., r-r-b), (c)
combination of odd and even numbers (e.g., e-o-e), and (d) combination of color
and odd versus even numbers (e.g., ro-bo-re). The sequence length of the latter
type always amounted to three cards. A total of 10 different rules existed: r-r-b-b,
r-r-b, be-ro-be, bo-re-be, ro-so-re, S-D-S-D, e-o-e, C-C-S, e-e-o-e, and S-S-H-C.
The most difficult rule was S-S-H-C (32% solution rate) and the easiest rule was
r-r-b-b (71% solution rate). All 10 rules were permuted randomly across the 10
tasks in a way that two three-person groups, one for each experimental condition,
shared the same random sequence of tasks (sibling groups). Thus, task difficulty
628 BRODBECK AND GREITEMEYER
was randomly distributed across task positions and balanced between experi-
mental conditions.
that is, examples of the rule stayed at the right of the last example and nonex-
amples were placed below the last card played. No further feedback was given
during task performance. A maximum of 10 steps per task was allowed (each step
comprised individual hypothesis formation, card selection, and feedback). Fi-
nally, each player informed the experimenter about the final hypothesis and was
given feedback whether it corresponded with the target rule. This final outcome
feedback was made public to all group members as part of a cooperative reward
scheme (see below).
Procedure for collaborative performance of collective induction. Sets of three
participants performed the task collaboratively as a group. Confronted with one
card on the table that was an example of the rule, each group member noted an
interim hypothesis on a private protocol sheet. Then, group members decided
collectively on a group hypothesis (noting it in the group protocol) and on which
of the 52 cards to play (also noting it in the group protocol) in order to test their
group hypothesis. Next, the experimenter informed the group whether the se-
lected card was an example or nonexample of the rule by placing it on the
respective position. A maximum of 10 steps (individual and group hypothesis
formation, group card selection, and group feedback) per task was allowed.
Finally, the experimenter was informed about the final group hypothesis and gave
feedback about its correctness. A similar cooperative reward scheme as for
individual training was used (see below).
Time restriction. Individuals and groups worked under the same time restric-
tion of 10 min for each task. In the instruction, it was stated that time is irrelevant
to performance evaluation but that time limits would be strictly enforced and that
making use of all 10 steps would be most beneficial to problem solving. Under
time restriction, an increase in effectiveness of task performance strategies means
making more efficient use of the information given during either individual or
collective task performance.
Cooperative goal structure. In order to induce a sense of solidarity and
cooperation and to ensure comparable levels of group goal commitment and
motivation among members of all groups, the following circumstances were
introduced in both experimental conditions: (a) physical copresence of two other
group members, (b) the expectation of future collaboration between members, (c)
feedback on the correctness of the individuals’ and the groups’ final solutions
after each task, and (d) cooperative structure of monetary reward for correct
solutions per task. 1 Several measures were taken after the training phase to
control for motivational differences that might result from the different group
1
In addition to the financial reward given for participation, each individual was paid a bonus of up
to 15 German Marks (about $8) based on group performance in all 10 tasks. Per task, a maximum of
15 points could be achieved, either 5 points for every individual’s correct hypothesis in the individual
task format or 15 points for a correct group hypothesis in the collaborative task format. The number
of points achieved in each condition could reach a total maximum of 150, and 0.10 German Marks
per point were given to the group to be equally shared among group members. Points credited to
individual group members and the whole group for each task were displayed on a flip chart in such
630 BRODBECK AND GREITEMEYER
Dependent Variables
Performance measures. As noted in the individual and group protocol, correct,
plausible, and nonplausible hypotheses were assessed for each task step (interim
hypotheses), for the final solution (final hypothesis), for each person during
individual and collective rule induction, and for each group during collective rule
induction. A correct hypothesis exactly matches the arbitrarily selected rule. A
plausible hypothesis means that the assumed rule matches the available evidence
(the cards played) but not the correct rule. A nonplausible hypothesis means that
the available evidence does not match the proposed rule.
Individual error checking. Error checking was defined as either anticipative
error avoidance or post hoc error management. Error avoidance was operation-
alized as generating no nonplausible interim hypothesis throughout the task
before stating the final hypothesis. In this case, a raw score of 1.00 for optimal
error checking was given. When at least one nonplausible interim hypothesis
occurs during task performance, error management comes into play. It can be
performed either immediately at the next step of the task, at any other later stage,
or not at all throughout the task. Error management is operationalized as the
immediacy of correcting nonplausible interim hypotheses that occur during task
performance. If a nonplausible interim hypothesis is corrected promptly in a
subsequent task step (n ⫹ 1), error checking is again optimal (raw score ⫽
1.00). However, if a nonplausible hypothesis is not corrected at step n ⫹ 1 but
at step n ⫹ 2, a raw score of .50 (1 error/2 steps) is given. If a nonplausible
hypothesis is not corrected until step n ⫹ 3, the raw score is .33 (1 error/3 steps)
and so on. If nonplausible hypotheses are not corrected at all, a raw score of .00
results. The error-management formula also covers the case in which several
nonplausible hypotheses occur during one task. For example, if the first non-
a way that it was clear to the participants who had solved which problems and how many points were
earned by the whole group and each group member.
RULE INDUCTION 631
reliability was established with two independent raters who coded the same N ⫽
40 tasks randomly chosen from the nominal pretest and posttest across the two
experimental conditions. The estimated interrater reliability was r ⫽ .96. In
order to obtain a proportional focusing score (distributed from .00 to 1.00), the
number of individual hypotheses that satisfied the two rules was divided by the
number of hypotheses proposed during one complete task.
Collective focusing. For collective focusing the same calculation as for indi-
vidual focusing was conducted for the group interim hypotheses. There is only
one measure for collective focusing because in the group context only the group
as a whole selects a card for hypothesis testing.
RESULTS
Logic of Analysis
Because we considered at least four collective tasks to be necessary for
significant effects of social interaction on learning (cf. McGlynn & Schick, 1973;
McGlynn et al., 1995) planned comparisons for pre- and posttest performance
between the experimental training conditions were conducted. For descriptive
and explorative purposes, performance data during the training phase were also
assessed. In Table 2, performance data for use of strategies (error checking and
focusing) and quality of final solutions (correct, plausible, and nonplausible) are
described across tasks (pretest, training phase, and posttests). Overall we were
interested in benefits to group members and to whole groups from mixed training
as compared to individual training. The analyses concentrating on group-member
benefits include an evaluation of how assigned training affects individual stra-
tegic behavior and the quality of final solutions. Furthermore, error-checking
strategies are explored as potential mediator variables for the hypothesized
association between mixed training and low proportions of nonplausible final
solutions. Analyses concentrating on group benefits include an examination of
how assigned training affects collective strategic behavior and the quality of final
group solutions. Furthermore, it is explored whether the hypothesized general
G-I transfer for individual error checking extends to group performance.
TABLE 2
Proportional Data for Individual and Group Performance by Individual
versus Mixed Training across Tasks
Pretest Posttest
Training condition
dependent variable Nominal Training phase Nominal Group
Individual training I I I I I I I I I G
Ind. error checking .84 n .87 n .86 n .91 n .82 .91
Group error checking — — — — — — — — — .85
Mutual error
checking — — — — — — — — — .69
Individual focusing .17 n .17 n .26 n .31 n .29 —
Group focusing — — — — — — — — — .29
Correct solution .36 .42 .51 .48 .38 .59 .68 .60 .62 .59
Plausible solution .55 .38 .28 .38 .47 .35 .26 .32 .27 .36
Nonplausible solution .09 .20 .21 .14 .15 .06 .06 .08 .11 .05
Mixed training I G I G I G I G I G
Ind. error checking .73 .85 .82 n .76 n .86 .82 .93 .88
Group error checking — .87 — n — n — .93 — .98
Mutual error
checking — .75 — .71 — .79 — .62 — .90
Individual focusing .12 — .21 — .20 — .29 — .24 —
Group focusing — .15 — n — n — .21 — .32
Correct solution .33 .32 .51 .41 .55 .53 .47 .68 .55 .82
Plausible solution .50 .45 .35 .45 .32 .33 .44 .23 .42 .18
Nonplausible solution .17 .23 .14 .14 .14 .14 .09 .09 .03 .00
Task 1 2 3 4 5 6 7 8 9 10
FIG. 1. Immediacy of individual error checking per task (upper part of figure) and mean
proportions of individual focusing per task step (lower part) as a function of training condition
(individual vs mixed) and task (nominal pretest and nominal posttest).
partial 2 ⫽ .00, p ⫽ .943] resulted (see lower part of Fig. 1). Focusing
increased significantly from pretest to posttest in both training conditions: in the
mixed-training condition [pretest: mean proportion ⫽ .12, SD ⫽ .10; posttest:
mean proportion ⫽ .24, SD ⫽ .15; F(1, 21) ⫽ 14.30, partial 2 ⫽ .40, p ⫽
.001] and in the individual-training condition [pretest: mean proportion ⫽ .17,
SD ⫽ .42; posttest; mean proportion ⫽ .29, SD ⫽ .14; F(1, 21) ⫽ 7.95,
partial 2 ⫽ .28, p ⫽ .01]. At the posttest, no significant difference in focusing
between training conditions was shown [F(1, 42) ⫽ 1.12, 2 ⫽ .03, p ⫽
.296]. Trend analysis with a repeated-measures ANOVA across pretest, training
phase, and posttest (Tasks 1, 3, 5, 7, and 9; see Table 2) yielded a significant
linear change in focusing for the mixed-training condition [mean proportions .12,
.21, .20, .29, .24; F(1, 21) ⫽ 13.48, p ⬍ .001] and the individual-training
condition [mean proportions .17, .17, .26, .31, .29; F(1, 21) ⫽ 12.77, p ⫽
.002]. No significant higher order trend (quadratic or cubic) was found.
In summary, the findings correspond with Hypothesis 1. On the one hand,
individuals improve their error-checking performance more with mixed training
than with individual training (i.e., general G-I transfer) and on the other hand, in
the use of focusing strategies, individuals do not benefit differentially from the
training conditions.
Individual benefits in quality of final solutions. We examined differences in
quality of final individual solutions before and after either individual training or
mixed training (see Table 2 for nominal pretest and posttest results). Again,
group means of individual performance data were used and the proportional data
per group were arcsin-transformed. Planned comparisons were conducted by two
repeated-measures analyses of variance with group versus mixed training as the
between-groups factor and pretest (Task 1) versus posttest (Task 9) as the
within-groups factor; once for proportions of nonplausible final solutions and
once for proportions of correct final solutions.
For nonplausible final solutions, no main effect for the treatment factor [F(1,
42) ⫽ 0.00, partial 2 ⫽ .00, p ⫽ 1.000], a marginally significant main effect
for the repeated-measures factor [F(1, 42) ⫽ 4.06, partial 2 ⫽ .09, p ⫽ .050],
and a significant two-way interaction [F(1, 42) ⫽ 6.72, partial 2 ⫽ .14, p ⫽
.013] resulted. The form of the two-way interaction is shown in the lower part
of Fig. 2. Follow-up comparisons showed that in the mixed-training condition,
nonplausible final solutions decreased significantly from pretest (mean propor-
tion ⫽ .17, SD ⫽ .20) to posttest (mean proportion ⫽ .03, SD ⫽ .10) [F(1,
21) ⫽ 10.23, partial 2 ⫽ .33, p ⫽ .004]. In the individual training condition,
no significant change from pretest (mean proportion ⫽ .09, SD ⫽ .15) to
posttest (mean proportion ⫽ .11, SD ⫽ .19) was evident [F(1, 21) ⫽ 0.17,
partial 2 ⫽ .01, p ⫽ .682]. At the posttest, the difference in proportions of
nonplausible final solutions between training conditions was close to marginal
significance [F(1, 42) ⫽ 2.78, 2 ⫽ .06, p ⫽ .103]. Trend analysis with a
repeated-measures ANOVA across pretest, training phase, and posttest (Tasks 1,
3, 5, 7, and 9; see Table 2) yielded a significant linear change in nonplausible
636 BRODBECK AND GREITEMEYER
FIG. 2. Proportions of nonplausible (lower part of figure) and correct final solutions (upper part)
as a function of training condition (individual vs mixed) and task (nominal pretest and nominal
posttest).
final solutions for the mixed-training condition [mean proportions .17, .14, .14,
.09, .03; F(1, 21) ⫽ 15.09, p ⬍ .001] but not for the individual-training
condition [mean proportions .09, .21, .15, .06, .11; F(1, 21) ⫽ 1.28, p ⫽ .270].
No significant higher order trend (quadratic or cubic) was found.
For correct final solutions, no significant main effect for the treatment factor
[F(1, 42) ⫽ 1.08, partial 2 ⫽ .02, p ⫽ .305], a significant main effect for the
repeated-measures factor [F(1, 42) ⫽ 9.83, partial 2 ⫽ .19, p ⫽ .003], and
no significant two-way interaction [F(1, 42) ⫽ 0.20, partial 2 ⫽ .01, p ⫽
.660] resulted (see upper part of Fig. 2). Correct final solution increased from
pretest to posttest in both training conditions: in the mixed-training condition
[pretest mean proportion ⫽ .33, SD ⫽ .23; posttest mean proportion ⫽ .55,
SD ⫽ .32; F(1, 21) ⫽ 4.14, partial 2 ⫽ .17, p ⫽ .055] and in the
individual-training condition [pretest mean proportion ⫽ .36, SD ⫽ .29; posttest
mean proportion ⫽ .62, SD ⫽ .35; F(1, 21) ⫽ 5.69, partial 2 ⫽ .21, p ⫽
.027]. At the posttest, no significant difference in correct final solutions between
RULE INDUCTION 637
training conditions was shown [F(1, 42) ⫽ 0.77, 2 ⫽ .02, p ⫽ .384]. Trend
analysis with a repeated-measures ANOVA across pretest, training phase, and
posttest (Tasks 1, 3, 5, 7, and 9; see Table 2) yielded no significant linear change
in correct final solutions for the mixed-training condition [mean proportions .33,
.51, .55, .47, .55; F(1, 21) ⫽ 2.35, p ⫽ .140]. However, for the individual-
training condition a significant linear change was evident [mean proportions .36,
.51, .38, .68, .62; F(1, 21) ⫽ 7.26, p ⫽ .014]. No significant higher order trend
(quadratic or cubic) was found.
In summary, when compared to individual-training participants, mixed-train-
ing participants improve particularly in terms of rejecting nonplausible final
solutions rather than in terms of detecting correct final solutions. The pattern of
findings for quality of final individual solutions is in accord with Hypothesis 1
and with the findings obtained by examining individual task performance strat-
egies.
Mediational properties of error-checking strategies. We examined which of
the two error-checking strategies determined the mixed-training profit for non-
plausible final solutions in the individual posttest. Remember that error checking
was conceptually differentiated into error avoidance and error management.
Optimal error avoidance was operationalized as the lack of nonplausible interim
hypothesis proposition throughout task performance before stating the final
hypothesis. Optimal error management was operationalized as prompt correction
of nonplausible interim hypotheses occurring during task performance. Our aim
was to test whether anticipative error avoidance or post hoc error management
serves as a mediator for the association between mixed training and reduced
proportions of nonplausible final hypotheses demonstrated above. This research
question was examined using the individual participant as the unit of analysis
because causal relationships between particular behavioral strategies of partici-
pants and their individual outcome performance were tested. Moreover, because
some of the analyses involve a subset of people (who commit errors) scattered
across groups, controlling for nonindependence of participants’ responses due to
common group membership by using group means (Kenny et al., 1998) was not
feasible.
Post hoc error correction is contingent upon the occurrence of interim non-
plausible hypotheses during task performance. In other words, error correction
can occur only after error avoidance has failed. Thus, for logical reasons, the
mediating properties of error avoidance had to be tested first. According to Baron
and Kenny (1986) significant bivariate relationships between the three variables
that are usually part of a mediator model—the independent, the dependent, and
the mediator variable—need to be demonstrated before testing for mediation is
justified. The quality of final solution, when examined on the individual level of
analysis, is a dichotomous variable (in our case, 1 ⫽ nonplausible vs 0 ⫽
plausible or correct), hence between-group comparisons were conducted with
2 ⫻ 2 chi-squared cross-classification analysis using r as the measure of asso-
ciation [for 2 ⫻ 2 contingency tables r is identical to the (phi) measure]. Error
638 BRODBECK AND GREITEMEYER
2
Logistic regression is more appropriate than linear regression because the criterion variable is
dichotomous. Because logistic regression does not deliver estimates of association similar to the more
commonly known coefficients (  , R 2 , and ⌬R 2 ), linear regression results are reported in the text. By
using logistic regression the same results were obtained. Without controlling for post hoc error
correction, the training condition is associated with nonplausible final hypotheses (R ⫽ ⫺.16,
Wald ⫽ 2.98, p ⫽.084). After controlling for post hoc error correction, the training condition is
unrelated to proportion of nonplausible final hypotheses (⌬R ⫽ .00, Wald ⫽ 0.55, p ⫽ .457).
RULE INDUCTION 639
FIG. 3. Error management as a mediator of the relationship between mixed training and
nonplausible final solutions at the nominal posttest. Standardized -weights are reported (**p ⫽
.001; ***p ⬍ .001; §p ⫽ .053).
3
Conducting a MANOVA for all three error-checking variables was not feasible. The group-level
criterion for group error checking is contaminated with the average individual error-checking
performance per group. In a multivariate analysis, one cannot control for multilevel cross-contami-
nation between dependent variables. Furthermore, the SDS data for error correction are dichotomous
and thus do not meet the assumptions for inclusion in analysis of variance.
640 BRODBECK AND GREITEMEYER
TABLE 3
Frequencies of Plausible and Nonplausible Interim Group Hypotheses for Each Distribution
of Plausible and Nonplausible Interim Group Member Hypotheses
DISCUSSION
Individual Benefits from Prior Group Performance
This study’s findings of group-member benefits from mixed (group and
individual) training extend the first demonstration of a general positive G-I
transfer reported by Stasson et al. (1991) to the rule-induction subtask of
generating plausible and rejecting nonplausible hypotheses. Stasson et al. (1991)
assumed general positive G-I transfer to occur for tasks that approach the
“truth-wins” criterion in group performance due to high demonstrability of
task-relevant concepts in the group context (p. 27). In our study, this assumption
was supported for self-evident strategies in rule induction by showing general
positive G-I transfer for error management (high demonstrability) and not for
error avoidance (low demonstrability) and focusing behavior (low demonstrabil-
ity). We infer that individual learning benefits from prior group performance (in
the form of mixed training) are more likely to occur for strategies that are of high
demonstrability in a group collaborative context than for strategies of low
demonstrability.
The additionally obtained mixed training benefit for rejecting nonplausible
final solutions was mediated by improvements in error management rather than
in error avoidance. This finding is very intuitive. Before idiosyncratic errors can
be checked by groups and the error-management strategy implicit in this social
642 BRODBECK AND GREITEMEYER
episode can be adopted by other group members, errors first need to be made and
they need to be pointed out in public. It is very difficult to learn from errors that
are avoided or not publicly expressed. Thus, error avoidance is good for group
performance but does not impact on group-member learning.
Some implications of the above results for theory and research are tenable.
Laughlin and Jaccard (1975) demonstrated the profitability of vicarious learning
when observing dyads that performed rule-induction tasks collectively. However,
they did not show exactly what was being observed and learned that could
mediate the performance-enhancing effect. It is tenable that individuals profit
from vicarious experiences made during collective induction by observing social
episodes in which the effective use of particular task performance strategies is
made evident. For the results presented here, it is reasonable to assume that the
error-checking strategy (i.e., monitoring the correspondence between a proposed
hypothesis and the available evidence) was adopted via vicarious experience of
collective error-checking behavior. Future research could investigate whether the
factors known to enhance vicarious learning and the processes involved (cf.
Bandura, 1986) apply when social episodes of collective induction are observed.
It would be also of interest to discern whether active involvement in such social
episodes (e.g., public suggestion of a hypothesis that is pointed out to be
nonplausible) enhances the learning effect as opposed to passive observation, as
studied by Laughlin and Jaccard (1975).
By drawing on the concept of demonstrability (Laughlin & Ellis, 1986), some
further research questions can be derived. The concept of demonstrability plays
a central role in the theory of collective decision making presented by Laughlin
and Hollingshead (1995). The theory specifies several factors that enhance the
demonstrability of facilitative individual contributions during collective task
performance: (a) group consensus about a valid conceptual system (e.g., a theory,
an algorithm, a technical language); (b) the amount of information available on
the task; (c) the ability, motivation, and time invested by group members to
identify others’ contributions to task performance that may turn out to be more
beneficial than their own contributions; and (d) the ability, motivation, and time
invested by group members to convince others of the benefits of their own
contributions as compared to contributions of others. In the light of the factors (c)
and (d), the present study’s finding of a general positive G-I transfer for error
management and not for focusing might be a consequence of the time restriction
of 10 min per task that was imposed on all groups. The groups did not have
enough time to demonstrate the benefits of the complex focusing strategy
because this would have required verbalization, extra effort, and time. Because
the error-management strategy is self-evident, no additional time and effort was
needed for demonstrating the benefits that result from using it.
Future research might investigate whether it is possible to raise the demon-
strability of complex effective hypothesis testing strategies in a way that results
in positive general G-I transfer. The theoretical and empirical work by Doise and
Mugny (1979) and Nastasi and Clements (1992) about group-member learning
RULE INDUCTION 643
group potential, process gain should result. However, this type of synergy would
not be obtained “free of charge.” Group-member learning requires, for instance,
repeated group task performance, extra effort and cognitive investments by group
members, and time. The possible reason why the assumption of process gain
forwarded by Shaw (1932) and others (e.g., Collins & Guetzkow, 1964) received
only fragmentary empirical support might lie in the preferred use of ad hoc
groups in experimental studies that performed only one or two tasks collabora-
tively, thereby preventing facilitative effects of group-member learning to be-
come manifest as a performance surplus on later occasions (for further experi-
mental evidence of this effect and a theoretical account, see Brodbeck &
Greitemeyer, 2000).
Differences between individual and group performance have often been at-
tributed to motivational variables (for reviews, see Kerr, 1983; Paulus & Dzin-
dolet, 1993; Shepperd, 1993). In our study, the different group settings (nominal
vs collaborative) might have caused motivational differences that were con-
founded with learning benefits from mixed training. Therefore, the same moti-
vational circumstances were induced in both experimental conditions: (a) co-
presence of two other group members, (b) cooperative reward structure, (c)
feedback on individual and group performance to all group members, and (d) the
expectation to work collaboratively in future tasks. Additionally, several vari-
ables were measured to statistically control for motivational differences between
the training conditions. Apart from one exception, which was contrary to the
assumed confounding effect, no significant differences were found. This raises
our confidence in the assumption that during mixed training participants actually
learn to use more effective strategies for rejecting nonplausible hypothesis in rule
induction than do individual-training participants.
CONCLUSION
This study extends previous research on group-member learning and problem
solving in two ways. First, results suggest that individual learning benefits from
prior group performance are more likely to occur for subtasks that involve
strategies of high demonstrability in the group context (error checking) than for
subtasks that involve strategies of low demonstrability (error avoidance and
focusing). Second, the evidence suggested that the social combination of indi-
vidual resources can improve as a result of prior experience in collaborative
problem solving, particularly in terms of improving collective error checking.
Our findings suggest that face-to-face interaction in the course of repeatedly
performing problem-solving tasks facilitates group-member learning and group
learning.
REFERENCES
Anderson, N., & Thomas, H. D. C. (1996). Work group socialization. In M. A. West (Ed.), Handbook
of work group psychology (pp. 423– 449). Chichester, UK: Wiley.
Argote, L. (1993). Group and organizational learning curves: Individual, system and environmental
components. Special Issue: Social processes in small groups: I. Theoretical perspectives. British
Journal of Social Psychology, 32, 31–51.
Argote, L., & McGrath, J. E. (1993). Group processes in organizations: Continuity and change. In C.
646 BRODBECK AND GREITEMEYER
Cooper & I. T. Robertson (Eds.), International review of industrial and organizational psychol-
ogy (Vol. 8, pp. 333–389). Chichester, UK: Wiley.
Azmitia, M., & Montgomery, R. (1993). Friendship, transactive dialogues, and the development of
scientific reasoning. Social Development, 2, 202–221.
Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Englewood
Cliffs, NJ: Prentice-Hall.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social
psychological research: Conceptual, strategic, and statistical considerations. Journal of Person-
ality and Social Psychology, 51, 1173–1182.
Brodbeck, F. C. (1996). Criteria for the study of work group functioning. In M. A. West (Ed.),
Handbook of work group psychology (pp. 285–315). Chichester, UK: Wiley.
Brodbeck, F. C., & Greitemeyer, T. (2000). A dynamic model of group performance: Considering the
group members’ capacity to learn. Group Processes & Intergroup Relations, 3, 159 –182.
Brown, R. (1988). Group processes: Dynamics within and between groups. Oxford, UK: Blackwell.
Bruner, J. S., Goodnow, J. J., & Austin, G. A. (1956). A study of thinking. New York: Wiley.
Collins, B. E., & Guetzkow, H. (1964). A social psychology of group processes for decision-making.
New York: Wiley.
Crott, H. W., Giesel, M., & Hoffmann, C. (1998). The process of inductive inference in groups: The
use of positive and negative hypothesis and target testing in sequential rule-discovery tasks.
Journal of Personality and Social Psychology, 75, 938 –952.
Davis, D. D., & Harless, D. W. (1996). Group vs individual performance in a price-searching
experiment. Organizational Behavior and Human Decision Processes, 66, 215–227.
Davis, J. H. (1969). Individual-group problem solving, subject preference, and problem type. Journal
of Personality and Social Psychology, 13, 362–374.
Davis, J. H. (1973). Group decision and social interaction: A theory of social decision schemes.
Psychological Review, 80, 97–125.
Doise, W., & Mugny, G. (1979). Individual and collective conflicts of centrations in cognitive
development. European Journal of Social Psychology, 9, 105–108.
Frese, M., & Zapf, D. (1994). Action as the core of work psychology: A German approach. In H. C.
Triandis (Ed.), Handbook of industrial and organizational psychology (2nd ed., pp. 271–340).
Palo Alto, CA: Consulting Psychologists Press.
Gigone, D., & Hastie, R. (1997). Proper analysis of the accuracy of group judgments. Psychological
Bulletin, 121, 149 –167.
Gruenfeld, D. H., & Hollingshead, A. B. (1993). Sociocognition in work groups: The evolution of
group integrative complexity and its relation to task performance. Small Group Research, 24,
383– 405.
Hare, A. P., Blumberg, H. H., Davies, M. F., & Kent, M. V. (1994). Small group research: A
handbook. Norwood, NJ: Ablex.
Hastie, R. (1986). Review essay: Experimental evidence on accuracy. In G. Owen & B. Grofman
(Eds.), Information pooling group decision making (pp. 129 –157). Westport, CT: JAI.
Hertz-Lazarowitz, R., & Miller, N. (Eds.). (1992). Interaction in cooperative groups: The theoretical
anatomy of group learning. Cambridge, UK: Cambridge Univ. Press.
Hill, G. W. (1982). Group versus individual performance: Are N ⫹ 1 heads better than one?
Psychological Bulletin, 91, 517–539.
Hinsz, V. B., Tindale, R. S., & Vollrath, D. A. (1997). The emerging conceptualization of groups as
information processors. Psychological Bulletin, 121, 43– 64.
Hollingshead, A. B. (1998). Group and individual training: The impact of practice on performance.
Small Group Research, 29, 254 –280.
Jerusalem, M., & Schwarzer, R. (1981). Fragebogen zur Erfassung von Selbstwirksamkeit [Self-
Efficacy Scale]. Forschungsbericht. Berlin: Freie Universität.
Johnson, E. S. (1971). Objective identification of strategy on a selection concept learning task.
Journal of Experimental Psychology Monograph, 90, 167–196.
RULE INDUCTION 647
Kenny, D. A., & Judd, C. M. (1986). Consequences of violating the independence assumption in
analysis of variance. Psychological Bulletin, 99, 422– 431.
Kenny, D. A., Kashy, D. A., & Bolger, N. (1998). Data analysis in social psychology. In D. T.
Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (4th ed., Vol. 1,
pp. 233–265). New York: McGraw-Hill.
Kerr, N. L. (1983). Motivation losses in small groups: A social dilemma analysis. Journal of
Personality and Social Psychology, 45, 819 – 828.
Klayman, J., & Ha, Y. W. (1987). Confirmation, disconfirmation, and information in hypothesis
testing. Psychological Review, 94, 211–228.
Lamm, H., & Trommsdorff, G. (1973). Group versus individual performance on tasks requiring
ideational proficiency (brainstorming): A review. European Journal of Social Psychology, 3,
361–388.
Laughlin, P. R. (1968). Focusing strategy for eight concept rules. Journal of Experimental Psychol-
ogy, 77, 661– 669.
Laughlin, P. R. (1973). Selection strategies in concept attainment. In R. L. Solso (Ed.), Contemporary
issues in cognitive psychology: The Loyola Symposium (pp. 277–311). Washington, DC:
Winston and Sons.
Laughlin, P. R. (1980). Social combination processes of cooperative problem-solving groups on
verbal intellective tasks. In M. Fishbein (Ed.), Progress in social psychology (pp. 127–155).
Hillsdale, NJ: Erlbaum.
Laughlin, P. R. (1988). Collective induction: Group performance, social combination processes, and
mutual majority and minority influence. Journal of Personality and Social Psychology, 54,
254 –267.
Laughlin, P. R., & Barth, J. M. (1981). Group-to-individual and individual-to-group in problem
solving transfer. Journal of Personality and Social Psychology, 41, 1087–1093.
Laughlin, P. R., Bonner, B. L., & Altermatt, T. W. (1999). Effectiveness of positive hypothesis
testing in inductive and deductive rule learning. Organizational Behavior and Human Decision
Processes, 77, 130 –146.
Laughlin, P. R., Chandler, J. S., Shupe, E. I., Magley, V. J., & Hulbert, L. G. (1995). Generality of
a theory of collective induction: Face-to-face and computer-mediated interaction, amount of
potential information, and group versus member choice of evidence. Organizational Behavior
and Human Decision Processes, 63, 98 –111.
Laughlin, P. R., & Ellis, A. L. (1986). Demonstrability and social combination processes on
mathematical intellective tasks. Journal of Experimental Social Psychology, 22, 177–189.
Laughlin, P. R., & Hollingshead, A. B. (1995). A theory of collective induction. Organizational
Behavior and Human Decision Processes, 61, 94 –107.
Laughlin, P. R., & Jaccard, J. J. (1975). Social facilitation and observational learning of individuals
and cooperative pairs. Journal of Personality and Social Psychology, 32, 873– 879.
Laughlin, P. R., Magley, V. J., & Shupe, E. I. (1997). Positive and negative hypothesis testing by
cooperative groups. Organizational Behavior and Human Decision Processes, 69, 265–275.
Laughlin, P. R., & Sweeney, J. D. (1977). Individual-to-group and group-to-individual in problem
solving transfer. Journal of Experimental Psychology: Human Learning and Memory, 3,
246 –254.
Laughlin, P. R., VanderStoep, S. W., & Hollingshead, A. B. (1991). Collective versus individual
induction: Recognition of truth, rejection of error, and collective information processing.
Journal of Personality and Social Psychology, 61, 50 – 67.
Levine, J. M., Resnick, L. B., & Higgins, E. T. (1993). Social foundations of cognition. Annual
Review of Psychology, 44, 585– 612.
Liang, D. W., Moreland, R., & Argote, L. (1995). Group versus individual training and group
performance: The mediating factor of transactive memory. Personality and Social Psychology
Bulletin, 21, 384 –393.
Lorge, I., Fox, D., Davitz, J., & Brenner, M. (1958). A survey of studies contrasting the quality of
648 BRODBECK AND GREITEMEYER
group performance and individual performance, 1920 –1957. Psychological Bulletin, 55, 337–
372.
McGlynn, R. P., & Schick, C. (1973). Dyadic concept attainment as a function of interaction format,
memory requirements, and sex. Journal of Educational Psychology, 65, 335–340.
McGlynn, R. P., Tubbs, D. D., & Holzhausen, K. G. (1995). Hypothesis generation in groups
constrained by evidence. Journal of Experimental Social Psychology, 31, 64 – 81.
Nastasi, B. K., & Clements, D. H. (1992). Social-cognitive behaviors and higher-order thinking in
educational computer environments. Learning and Instruction, 2, 215–238.
Paulus, P. B., & Dzindolet, M. T. (1993). Social influence processes in group brainstorming. Journal
of Personality and Social Psychology, 64, 575–586.
Reagan-Cirincione, P. (1994). Improving the accuracy of group judgment: A process intervention
combining group facilitation, social judgment analysis, and information technology. Organiza-
tional Behavior and Human Decision Processes, 58, 246 –270.
Schwartz, D. L. (1995). The emergence of abstract representations in dyad problem solving. The
Journal of the Learning Sciences, 4, 321–354.
Shaw, M. E. (1932). A comparison of individuals and small groups in the rational solution of complex
problems. American Journal of Psychology, 44, 491–504.
Shepperd, J. A. (1993). Productivity loss in performance groups: A motivation analysis. Psycholog-
ical Bulletin, 113, 67– 81.
Stasser, G., Kerr, N. L., & Davis, J. H. (1989). Influence processes and consensus models in
decision-making groups. In P. B. Paulus (Ed.), Psychology of group influence (pp. 279 –326).
Hillsdale, NJ: Erlbaum.
Stasson, M. F., Kameda, T., Parks, C. D., Zimmerman, S. K., & Davis, J. H. (1991). Effects of
assigned group consensus requirement on group problem solving and group members’ learning.
Social Psychology Quarterly, 54, 25–35.
Staudinger, U. M., & Baltes, P. B. (1996). Interactive minds: A facilitative setting for wisdom-related
performance? Journal of Personality and Social Psychology, 71, 746 –762.
Steiner, I. D. (1972). Group process and productivity. New York: Academic Press.
Stroebe W., Diehl, M., & Abakoumkin, G. (1996). Social compensation and the Köhler effect:
Toward a theoretical explanation of motivation gains in group productivity. In E. Witte & J. H.
Davis (Eds.), Understanding group behavior: Small group processes and interpersonal relations
(pp. 37– 65). Mahwah, NJ: Erlbaum.
Tindale, R. S. (1993). Decision errors made by individuals and groups. In J. Castellan (Ed.), Current
issues in individual and group decision making research (pp. 109 –124). Hillsdale, NJ: Erlbaum.
Wagner, J. A. (1995). Studies of individualism– collectivism: Effects on cooperation in groups.
Academy of Management Journal, 38, 152–172.
Wegner, D. M. (1986). Transactive memory: A contemporary analysis of the group mind. In B.
Mullen & G. R. Goethals (Eds.), Theories of group behavior (pp. 185–208). New York:
Springer-Verlag.
Whitney, K. (1994). Improving group task performance: The role of group goals and group efficacy.
Human Performance, 7, 55–78.
Winer, B. J. (1962). Statistical principles in experimental design. New York: McGraw-Hill.
Zaccaro, S. J., & McCoy, M. C. (1988). The effects of task and interpersonal cohesiveness on
performance of a disjunctive group task. Journal of Applied Psychology, 18, 837– 851.
Zapf, D., Brodbeck, F. C., Frese, M., Peters, H., & Prümper, J. (1992). Errors in working with office
computers: A first validation of a taxonomy for observed errors in a field setting. International
Journal of Human Computer Interaction, 4, 311–339.