Escolar Documentos
Profissional Documentos
Cultura Documentos
This research was supported by National Science Foundation Grants SES-9121690 and SBR-
9708547 and by Office of Naval Research Contract N00014-86-K-0322. I thank my collaborators
in this research program on collective induction, T. William Altermatt, Bryan L. Bonner, John S.
Chandler, Alan L. Ellis, Gail C. Futoran, Andrea B. Hollingshead, Lorne G. Hulbert, Vicki J.
Magley, Richard P. McGlynn, Thomas A. Shippy, Ellen I. Shupe, and Scott W. VanderStoep.
Address correspondence and reprint requests to Patrick R. Laughlin, Department of Psychology,
University of Illinois, 603 E. Daniel Street, Champaign, IL 61820. E-mail: plaughli@s.psych.
uiuc.edu.
50
0749-5978/99 $30.00
Copyright q 1999 by Academic Press
All rights of reproduction in any form reserved.
COLLECTIVE INDUCTION 51
the predictions, the hypotheses are rejected or revised. They may exchange
hypotheses and evidence with other groups or individuals who are working on
the same problem. These and many other cooperative groups, such as auditors,
securities analysts, intelligence analysts, and art experts, engage in collective
induction, the cooperative search for descriptive, predictive, and explanatory
generalizations, rules, and principles.
This article presents 12 postulates on collective induction. Postulates 1
through 8 were previously proposed by Laughlin and Hollingshead (1995) and
Laughlin (1996), who give further references and relationships to other re-
search. Postulates 1–6 set collective induction within the general social combi-
nation approach to group decision making. Postulates 7 and 8 formalize the
social combination processes of group hypothesis formation in collective induc-
tion. Postulates 9–12 summarize research on collective versus individual induc-
tion, influence in simultaneous collective and individual induction, the relative
importance of multiple hypotheses and multiple evidence, and the relative
effectiveness of positive and negative hypothesis tests. After we present and
briefly describe Postulates 1–8 we describe the rule induction paradigm of
this research program. We then summarize the evidence for Postulates 7–12.
Finally, we consider the history and fundamental emphasis of the social combi-
nation approach to small group performance.
group problems and decisions. One correct member (truth wins) suffices on
eureka or insight problems, remote verbal associations, and mathematical
problems. Two correct members (truth-supported wins) suffice on world knowl-
edge (Hastie, 1986) items such as vocabulary, analogies, and general informa-
tion. Simple majorities are necessary on choice dilemma problems, evaluative
attitudinal judgments, and preferences for bets and gambles. Two-thirds major-
ities prevail on mock jury decisions.
The objective is to induce a rule based on standard playing cards with four
suits (clubs, diamonds, hearts, spades) of 13 cards (ace 5 1, two 5 2, three 5
3, . . . , king 5 13). The rule may be based on suit (e.g., “diamonds”), number
(e.g., “eights”), or any combination of numerical and logical operations on suit
and number (e.g., “even diamonds alternate with odd spades”). The problem
begins with a card that is known to be an example of the rule.
Each trial consists of three stages. First, each group member records his or
her own hypothesis. Second, the group proposes a group hypothesis. Third, the
group chooses one of the 52 cards (repetitions allowed). The experimenter then
informs the group whether the selected card is an example or nonexample of
the rule. Examples are displayed to the right of the initial card, and nonexam-
ples are displayed below the last card played. The experimenter does not indi-
cate whether the member and group hypotheses are correct or incorrect until
after the final trial (typically 10). Discussion is completely free within the
groups. No group decision rule (e.g., unanimity, majority) for either hypotheses
or card plays is imposed or implied by the instructions. Several decks of cards,
sorted by suits in ascending order from ace to king, are available, so the same
card may be played as many times as desired. Table 1 gives an illustration for
the correct rule “two diamonds and two clubs alternate.”
Assume a four-person group. Denote the correct hypothesis as C, plausible
hypotheses as P, and nonplausible hypotheses as N, following each C, P, or N
by the number of group members (1 to 4) who propose it. For example, C3P1
indicates that three members propose the correct hypothesis and one member
proposes a plausible hypothesis. C1P2P1 indicates that one member proposes
TABLE 1
Illustration of Card Plays and Hypotheses
8D 2D 2C 6C 4D 9D
AH 3C
QH 2H
6D
Hypotheses
1. Even diamonds
2. Even diamonds and odd hearts
3. Even diamonds and even hearts
4. Diamonds below jack
5. Eights and twos
6. Diamonds and clubs
7. Diamonds and clubs
8. Even diamonds and clubs
9. Two diamonds and two even clubs alternate
10. Two diamonds and two clubs alternate
Final: Two diamonds and two clubs alternate
Note. C 5 clubs, D 5 diamonds, H 5 hearts, S 5 spades. Correct rule is two diamonds and two
clubs alternate. Examples are placed to right of previous example; nonexamples are placed below
last card played.
COLLECTIVE INDUCTION 55
the correct hypothesis, two members propose the same plausible hypothesis,
and one member proposes another plausible hypothesis. There are 38 possible
distributions of member hypotheses: C4, C3P1, C3N1, . . . , N1N1N1N1. These
38 distributions are the rows of a social combination model.
There are 11 possible types of group hypotheses on each trial: the single
correct hypothesis; a plausible hypothesis proposed by four, three, two, or one
group member; an emergent plausible hypothesis not proposed by any group
member (P0); a nonplausible hypothesis proposed by four, three, two, or one
group member; or an emergent nonplausible hypothesis not proposed by any
group member (N0). These 11 possible types of group hypotheses, C, P4, P3,
P2, P1, P0, N4, N3, N2, N1, and N0, are the columns of a social combination
model. Table 2 gives the predicted probabilities of group responses for distribu-
tions of member hypotheses from Postulates 7 and 8.
Table 3 gives the results of model testing for Postulates 7 and 8 (POST) and
five other plausible models for the aggregate data of 616 four-person groups
in seven experiments (Laughlin, 1988, 1992; Laughlin, Magley, & Shupe,
1997, control condition; Laughlin & McGlynn, 1986; Laughlin & Shupe, 1996;
Laughlin, Shupe, & Magley, 1998, emergent group and emergent member
condition; Laughlin, VanderStoep, & Hollingshead, 1991, one-array condition).
These five models were: (1) equiprobability; (2) majority, plurality otherwise,
proportionality otherwise; (3) proportionality; (4) correct-supported wins, ma-
jority otherwise, plurality otherwise, proportionality otherwise; (5) correct
wins, majority otherwise, plurality otherwise, proportionality otherwise.
The predicted proportions of each group hypothesis for each distribution of
member hypotheses were determined for each model by multiplying the pre-
dicted probabilities of each group hypothesis by the respective row sum and
dividing by the total number of group hypotheses. Table 3 gives the results of
Kolmogorov–Smirnov one-sample tests of the predicted and obtained propor-
tions for each model. Inspection of Table 3 indicates lower Dmax values for
POST than for each of the other models. Using the number of groups (616),
the critical value of Dmax at p . .20 is .04. Table 3 indicates that POST provides
an acceptable fit whereas each of the other five models may be rejected.
These tests were for the overall goodness of fit of Postulates 7 and 8 and
the other five social combination models for the 38 distributions of member
preferences (C4, C3P1, . . . , N1N1N1N1). Further chi-square tests of the pre-
dicted and obtained frequencies of group hypotheses from Postulates 7 and 8
were then conducted for each of the 38 distributions of member preferences.
As indicated in Tables 2 and 3, there are 18 member distributions with at least
one correct member (C4, C3P1, . . . , C1N1N1N1), and 20 distributions with
no correct member (P4, P3P1, . . . , N1N1N1N1). The difference between pre-
dicted and obtained frequencies was nonsignificant for 15 of the 18 distributions
with at least one correct member, and significant ( p , .10) for C2P1P1, C1P3,
and C1P1P1P1. Comparison of Tables 2 and 3 indicates more correct group
hypotheses than predicted in each of these three distributions, indicating a
prepotency of the correct hypothesis over plausible hypotheses beyond the
number of advocates of the correct hypothesis.
56 PATRICK R. LAUGHLIN
TABLE 2
Predicted Probabilities of Group Hypotheses for Distributions of Member
Hypotheses from Postulates 7 and 8
Group hypothesis
Distribution C P4 P3 P2 P1 P0 N4 N3 N2 N1 N0
TABLE 3
Obtained Probabilities of Group Hypotheses for Distributions of Member
Hypotheses for 616 Four-Person Groups
Group hypothesis
Distribution C P4 P3 P2 P1 P0 N4 N3 N2 N1 N0 Sum
and pluralities determined the group hypothesis somewhat more than predicted
when no group member proposed the correct hypothesis.
In summary, Postulates 7 and 8 summarize a complex but orderly process
of group hypothesis formation in collective induction. This process involves
voting (explicit or implicit), turntaking, demonstration, and emergent re-
sponses for different distributions of member hypotheses, rather than any
single one of these processes. Postulates 7 and 8 provided a good overall fit for
the 38 distributions of member preferences while the predictions of five other
models could be rejected. Further tests within each of the 38 distributions of
member preferences indicated that Postulates 7 and 8 provided a good fit for
15 of the 18 distributions with at least one correct member, with the other
three distributions indicating a prepotency of the correct hypothesis relative
to plausible hypotheses. When no member proposed the correct hypothesis,
majorities had somewhat less influence and pluralities somewhat more influ-
ence than predicted.
Postulates 7 and 8 predict that the final group hypothesis will be correct if
at least one member proposes the correct hypothesis on some trial, and incorrect
if no member proposes the correct hypothesis on some trial (see Laughlin &
Hollingshead, 1995, for the derivation of this prediction). Over the 616 groups
the conditional probabilities of a correct final group hypothesis were .91 if some
member proposed the correct hypothesis on some trial and .01 if no member
proposed the correct hypothesis on any trial, supporting both predictions.
information and time the groups would perform at the level of the first (best)
individuals for both correct hypotheses and nonplausible hypotheses. Figure
1 (bottom) gives the proportions of correct and nonplausible hypotheses for
the groups and first, second, third, and fourth individuals. The groups and
first individuals did not differ significantly for either correct or nonplausible
hypotheses.
The design of these two experiments does not indicate whether the improve-
ment with increasing information is due to the increasing number of hypothe-
ses, the increasing number of card plays (evidence), or both. Accordingly,
Laughlin, Bonner, and Altermatt (1998) compared four-person groups and four
independent individuals with four arrays of card choices, but with instructions
to propose one group hypothesis, two different group hypotheses, or four differ-
ent group hypotheses on each trial. This assesses the value of multiple hypothe-
ses while providing a constant large amount of potential information. As indi-
cated in Fig. 2 (top), the proportion of correct hypotheses did not differ for the
groups and first individuals, and both the groups and first individuals had a
higher proportion of correct hypotheses than each of the second, third, and
60 PATRICK R. LAUGHLIN
FIG. 2. Proportions of correct hypotheses in Laughlin, Bonner, and Altermatt, 1998 (top), and
in Laughlin, Chandler, Shupe, Magley, and Hulbert, 1995 (bottom).
fourth individuals. There was no significant difference between one, two, and
four hypotheses, indicating that the previous results of Laughlin et al. (1991)
were due to increasing evidence rather than increasing hypotheses.
In summary, given sufficient information and time, collective induction is
comparable to the induction of the best of an equivalent number of individuals.
This extends the large literature comparing groups and the average individual
(see Hastie, 1986; Hill, 1982; Kelley & Thibaut, 1969; Steiner, 1972) to the
more demanding comparison of groups and an equivalent number of individu-
als. Indeed, to our knowledge no other research apart from this program has
compared group performance with an equivalent number of individuals, so the
results are unprecedented. In the classic typology of Steiner (1972), collective
induction with sufficient information is a disjunctive group task.
FIG. 3. Proportions of correct hypotheses in Laughlin and McGlynn, 1986 (top); in Laughlin,
1992 (middle); and in Laughlin, 1988 (bottom).
COLLECTIVE INDUCTION 63
hypotheses from other individuals or groups had less effect, just as multiple
hypotheses did not increase the number of correct hypotheses in the experiment
of Laughlin et al. (1998). The group members were able to generate sufficient
hypotheses, so that additional hypotheses from other individuals or groups did
not appreciably improve performance. Groups such as scientific research teams
or auditing teams may be able to generate sufficient hypotheses but lack the
evidence to evaluate them, so that further evidence from other individuals or
groups is relatively more valuable than further hypotheses.
In summary, multiple evidence, whether generated within the group or re-
ceived from outside the group, is more effective than multiple hypotheses,
whether generated within the group or received from outside the group.
FIG. 4. Proportions of patterns indicating group influence on individual and individual influ-
ence on group by correct hypotheses in Laughlin and McGlynn, 1986 (top), and in Laughlin,
1992 (bottom).
Postulate 12: Positive hypothesis tests are generally more effective than
negative hypothesis tests in collective induction.
In rule induction problems a positive hypothesis test selects or generates
evidence that the group or individual expects to be an example if the proposed
hypothesis is correct, whereas a negative hypothesis test selects or generates
evidence that the group or individual expects to be a nonexample if the proposed
hypothesis is correct. Are positive or negative hypothesis tests more effective
in inducing the correct rule?
In an excellent theoretical analysis Klayman and Ha (1987) distinguished
five types of hypotheses by the relationship between a proposed hypothesis
and the correct hypothesis. To illustrate for the current rule induction problems,
assume the correct answer “two diamonds alternate with two clubs” and the
sequence of four examples 8D 2D 8C 2C. Embedded hypotheses are plausible
COLLECTIVE INDUCTION 65
and a subset of the correct hypothesis (e.g., “two even diamonds alternate with
two even clubs”). Overlapping hypotheses are plausible and based on other
relationships (e.g., “even diamonds and even clubs”). Surrounding hypotheses
are plausible and a superset of the correct hypothesis (e.g., “two red alternate
with two black”). Disjoint (nonplausible) hypotheses are inconsistent with the
evidence (e.g., “hearts”). Target hypotheses are correct. Klayman and Ha (1987)
analyze the possible inferences of ambiguous verification or conclusive falsifica-
tion from positive and negative tests of the five types of hypotheses followed
by examples or nonexamples.
In addition to these inferences of ambiguous verification and conclusive
falsification of hypotheses, Laughlin, Magley, and Shupe (1997) emphasized
the importance of examples. Since examples provide further evidence for what
the correct rule is, whereas nonexamples indicate what the correct rule is not,
examples should make the correct relationships more likely to be perceived,
tested, and adopted. In particular, Laughlin et al. (1997) proposed that positive
tests of the overlapping hypotheses which predominate on early trials would
be more likely to result in examples than would negative tests.
In five conditions four-person groups were instructed to use positive and/or
negative hypothesis tests on four separate arrays of cards on each of 10 trials.
In the PPPP condition the groups were instructed to use a positive test (P) of
their current hypothesis on each of the four arrays. In the PPPN condition the
groups were instructed to use positive tests (P) on Arrays 1, 2, and 3, and a
negative test (N) on Array 4. Similarly, in the PPNN, PNNN, and NNNN
conditions the groups were instructed to use positive or negative tests on the
four arrays. Reasoning that positive tests are more likely to lead to further
examples than are negative tests, and that examples are more useful than
nonexamples in inducing the correct rule because they provide further evidence,
Laughlin, Magley, and Shupe (1997) predicted that the number of correct
hypotheses would correspond to the proportion of positive tests on the four
arrays. Figure 5 (top) gives the proportion of correct hypotheses. There was a
higher proportion of correct hypotheses if the groups were instructed or allowed
to use positive hypothesis tests on at least two arrays (PPPP, PPPN, PPNN,
Control), with no significant differences between these four conditions, support-
ing the predicted effectiveness of positive hypothesis testing.
Experiment 2 used more difficult rules, alternations of triples of two different
suits, such as “three diamonds alternate with three clubs.” As in Experiment
1, there were four arrays and 10 trials of group member hypotheses, group
hypothesis, and card selections. There were four conditions of instructions to
use positive tests (P) or negative tests (N) on the first five trials and the second
five trials: (1) positive tests on the first 5 and positive tests on the second 5
(PP), (2) positive tests on the first 5 and negative tests on the second 5 (PN),
(3) negative tests on the first 5 and positive tests on the second 5 (NP), (4)
negative tests on the first 5 and negative tests on the second 5 (NN). In a
fifth condition (control) the groups could use any combination of positive and
negative tests on the four arrays on all trials. Figure 5 (bottom) gives the
66 PATRICK R. LAUGHLIN
FIG. 5. Proportions of five types of hypotheses in Laughlin, Magley, and Shupe, 1997, Experi-
ment 1 (top) and Experiment 2 (bottom).
Social combination approaches assume that small group processes are orderly
rather than capricious mappings from a distribution of member preferences to
a collective response. Lorge and Solomon (1955) used the binomial theorem to
predict the probability of group solution from known individual probabilities
of solution and the truth-wins assumption that groups recognize and adopt a
correct answer if proposed by at least one group member. Thomas and Fink
(1961) extended the social combination approach by using the multinomial
theorem for decisions with three or more alternatives and testing two further
process assumptions in addition to the truth-wins assumption. Smoke and
Zajonc (1962) introduced the concept of group decision schemes as formaliza-
tions of assumptions of the group process and considered further decision
schemes, such as majorities of different orders and oligarchies of different sizes.
In the elegant matrix algebra formulation that inspires this special issue, Davis
(1973) integrated these two fundamental ideas of applying the multinomial
theorem to determine the expected composition of group member preferences
from a known probability distribution of individual preferences and formalizing
different assumptions of the group process as social decision schemes. Any
theory of small group performance that specifies the probabilities of each group
response for each distribution of member preferences may be formalized as a
social decision scheme (social combination model) and the predictions of the
theory competitively tested against other theories.
The current Postulates 1 and 2 propose that social combination models for-
malize the processes by which groups map a distribution of member preferences
to a group response. Postulates 3 and 4 propose a basic intellective–judgmental
continuum of group tasks and four conditions of demonstrability that underlie
intellective tasks. Postulate 5 summarizes the orderly relationship between
the number of group members who advocate a group response and the demon-
strability of the response that emerges from the work of many researchers
with a social combination approach. Postulates 6, 7, and 8 specify the social
combination processes in collective induction, a complex but orderly process
that involves voting (explicit or implicit), turntaking, demonstration, and emer-
gent responses for different distributions of member hypotheses. Postulates 9
through 12 summarize the results of research on other issues in collective
induction. Future research may extend the postulate set to further issues.
Just as Postulates 7 and 8 specify the particular social combination process
in collective induction and Postulates 9 through 12 summarize research on
68 PATRICK R. LAUGHLIN
further issues, other researchers in the social combination approach have both
specified the particular social combination process on classes of group tasks
and extended the research to other issues. For example, the other articles of this
issue both clearly establish that the fundamental social combination process on
mock jury decisions is two-thirds majority, with a defendant protection bias
when there is not a two-thirds majority, and extend the research to issues such
as dynamic models, procedural variables, confidence, group member character-
istics, and continuous (rather than discrete) decisions. This competitive testing
of theories and extension to other issues indicates the fundamental emphasis
of the social combination approach on social behavior in an era when social
psychology is predominantly the study of the social mind.
REFERENCES
Davis, J. H. (1973). Group decision and social interaction: A theory of social decision schemes.
Psychological Review, 80, 97–125.
Hastie, R. (1986). Review essay: Experimental evidence on group accuracy. In G. Owen & B.
Grofman (Eds.), Information pooling and group accuracy (pp. 129–157). Westport, CT: JAI.
Hill, G. W. (1982). Group versus individual performance: Are N 1 1 heads better than one?
Psychological Bulletin, 91, 517–539.
Kelley, H. H., & Thibaut, J. W. (1969). Group problem solving. In G. Lindzey & E. Aronson (Eds.),
The handbook of social psychology (Vol 4, pp. 1–101). Reading, MA: Addison–Wesley.
Klayman, J., & Ha, Y-M. (1987). Confirmation, disconfirmation, and information in hypothesis
testing. Psychological Review, 94, 211–228.
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. (1992). Performance and influence in simultaneous collective and individual induc-
tion. Organizational Behavior and Human Decision Processes, 51, 447–470.
Laughlin, P. R. (1996). Group decision making and collective induction. In J. Davis & E. Witte
(Eds.), Understanding group behavior: Vol. 1. Consensual action by small groups (pp. 61–80).
Mahwah, NJ: Erlbaum.
Laughlin, P. R., Bonner, B. L, & Altermatt, T. W. (1998). Collective versus individual induction with
single versus multiple hypotheses. Journal of Personality and Social Psychology, 75, 1481–1489.
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 mathe-
matical 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., 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., & McGlynn, R. P. (1986). Collective induction: Mutual group and individual
influence by exchange of hypotheses and evidence. Journal of Experimental Social Psychology,
22, 567–589.
Laughlin, P. R., & Shupe, E. I. (1996). Intergroup collective induction. Organizational Behavior
and Human Decision Processes, 68, 44–57.
COLLECTIVE INDUCTION 69
Laughlin, P. R., Shupe, E. I., & Magley, V. J. (1998). Effectiveness of positive hypothesis testing
for cooperative groups. Organizational Behavior and Human Decision Processes, 73, 27–38.
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.
Lorge, I., & Solomon, H. (1955). Two models of group behavior in the solution of Eureka-type
problems. Psychometrika, 29, 139–148.
Popper, K. R. (1959). The logic of scientific discovery. New York: Basic Books.
Smoke, W. H., & Zajonc, R. B. (1962). On the reliability of group judgments and decisions. In
J. H. Criswell, H. Solomon, & P. Suppes (Eds.), Mathematical methods in small group process
(pp. 279–290). Stanford, CA: Stanford Univ. Press.
Steiner, I. D. (1972). Group process and productivity. New York: Academic Press.
Thomas, E. J., & Fink, C. F. (1961). Models of group problem solving. Journal of Abnormal and
Social Psychology, 63, 53–63.