Professional Documents
Culture Documents
Pooling of Unshared Information in Group Decision Making: Biased Information Sampling During Discussion
Pooling of Unshared Information in Group Decision Making: Biased Information Sampling During Discussion
Pooling of Unshared Information in Group Decision Making: Biased Information Sampling During Discussion
Decision-making groups must often choose is rarely a systematic and balanced exploration
from a set of specified alternatives (e.g., guilty of the relevant issues. On the contrary, it is
and not guilty for a jury, a set of applicants often thematic and consensus confirming;
for a selection committee), and this choice is that is, discussion tends to focus on particular
typically preceded by discussion of the merits issues and to support an existing or emergent
of each alternative. One goal of discussion is consensus (cf. Fisher, 1980). Such patterns
to achieve a consensus among the group's may counter effective pooling of information
members. Another goal is to pool members' and may perpetuate biases that members
expertise and knowledge. In principle, pooling bring to the group.
information permits a group decision that is In this article, we explore the dynamics of
more informed than the decisions of members discussion within the framework of an infor-
acting individually. In particular, discussion mation sampling model. This model high-
can perform a corrective function when lights the role of the pregroup information
members individually have incomplete and distribution, a summary of which group
biased information but collectively can piece members are exposed to what information
together an unbiased picture of the relative before discussion. The distribution of infor-
merits of the decision alternatives. Notwith- mation among group members may give rise
standing the potential of discussion to serve to several types of bias. Informational bias
such a corrective function, group discussion occurs when individual group members are
may often fall short of its potential. Discussion given partial sets of information that do not
reflect the balance of available supporting
arguments for the various decision alterna-
Portions of this study were reported at the 91st annual tives. Such biased sets of information may,
meeting of the American Psychological Association, An-
aheim, California, 1983. We thank Alison Karas, Jerry
in turn, result in preferential bias: individual
Kasai, Scott Snell, and Amelia Tynan for assisting in members' preferring alternatives at the onset
data collection and analysis. We also thank James Davis of discussion that they would not prefer if
and Norbert Kerr for reading and commenting on an they had complete information. Finally, the
earlier draft of this article.
sampling model suggests that the pregroup
Requests for reprints should be sent to Garold Stasser,
Department of Psychology, Miami University, Oxford, biases in information and preference may act
Ohio 45056. to bias the content of subsequent discussion.
1467
1468 GAROLD STASSER AND WILLIAM TITUS
Thus the content of discussion tends to reflect natives, A and B. Suppose that there exist
but not to correct biases introduced by the seven items of information favoring A (de-
distribution of information over group mem- noted as a,, 82, . . • , 3 6 , and a7) and four
bers before discussion. items of information favoring B (denoted as
b i , b 2 , bj, and b4). For ease of discussion, we
assume that these 11 items of information
Pregroup Information Distributions are equally convincing. In Table 1 we present
four ways in which this information could be
For many topics of group discussion and distributed across the group members before
decision, we can think of a set of relevant discussion. In Case 1, all information is
information and arguments (cf. Burnstein & shared; that is, every member is aware of the
Vinokur, 1977). Members are usually aware seven pro-A and four pro-B pieces of infor-
of some subset of these arguments before the mation. In this case, we would expect that
group convenes and, on the basis of this each member would prefer A before discus-
subset, have a tentative preference for one of sion and thus an initial unanimous consensus
the alternatives. During group discussion, would occur. Group discussion would serve
members exchange arguments and reevaluate primarily to reassure the members that they
their initial preferences. The information and were aware of all the critical information.
arguments that members collectively bring to This reassurance might increase members'
the group guide the emergence of a consensus confidence in their preferences, but discussion
in two ways. On the one hand, such infor- would provide little impetus for modifying
mation shapes the initial preferences of the initial preferences.
group members; these initial preferences de- In Case 2, some information about both
fine the degree of consensus that exists at the A and B is unshared. We refer to this case as
onset of discussion. On the other hand, the an unbiased distribution of unshared infor-
content of discussion is drawn from the ex- mation because each individual possesses a
isting pool of information and arguments, ratio of pro-A to pro-B that is nearly identical
and the balance of arguments favoring one to the ratio in the total information pool.
position relative to another partly determines Members bring three pro-A and two pro-B
which preferences are likely to be strengthened items to the group and, if information ex-
or changed during the course of discussion. change is exhaustive, would leave the group
It is important to consider not only how with seven pro-A and four pro-B items. Thus
much supporting information exists for each members should prefer A both before and
alternative but also how this information is after group discussion.
distributed across a group's members before In Case 3, three of the pro-A items of
their discussion. In the extreme, there are information (viz., a5, a^ and a7) are unshared,
two kinds of pregroup information distribu- but all of the pro-B information is shared.
tions: shared information is familiar to all Each group member still has more pro-A
group members, whereas unshared informa- than pro-B information; however, from each
tion is held by only one of the group's member's perspective, the advantage to A
members. (Of course, there are degrees be- appears proportionately less than if they were
tween these two extremes in that an argument aware of all of the available information.
or fact may be shared by several but not all Thus members would be likely to bring a
members; for our purposes, we shall only relatively weak preference for A to the group's
consider these two extreme cases.) Burnstein discussion, but discussion could strengthen
and Vinokur (1977) made a similar distinc- these initial preferences because each member
tion in their persuasive arguments theory of could inform the others of a pro-A argument
group polarization. We will use an example of which they were previously unaware. As
suggested by their analysis to illustrate the suggested by Burnstein and Vinokur (1977),
distinction between shared and unshared in- pregroup distributions of information similar
formation distributions. to those in Cases 2 and 3 would probably
Consider a simple case in which a three- result in group polarization (cf. Myers &
person group must decide between two alter- Lamm, 1976); that is, group discussion should
UNSHARED INFORMATION AND GROUP DECISION MAKING 1469
Table 1
Some Possible Pregroup Distributions of Seven Pro-A (af) and Four Pro-B (b-,) hems
of Information Over Three Group Members
Group member
Item
position X Y Z
Pro-A ai, a 2 , a3, 34, as, 36, 37 ai, a 2 , 33, 34, as, 3&, 37 3], 32, 33, 34, 3$, 3&, 3T
Pro-B b,,b 2 , b,, b. b , , b 2 , b 3 , b4 b i , b 2 , bj.b*
Pro-A
Shared a, a, a,
Unshared a2, a3 34, 3s ae, a7
Pro-B
Shared b, b, b,
Unshared b, b3 b.
Pro-A
Shared 3], 32, 33, 34 3|, 32, 33, 34 3|, 32, ^3, 34
Unshared a, 36 37
Pro-B" b.^.bj, b4 b,,b 2 , b,, b. b,,b2,b3,b4
Pro-A
Shared
Unshared 36, 37
Pro-B" b , , b 2 , bj, I b,,b 2 , b,,
enhance the initial predispositions of the biased against A could conclude discussion
members to select A. favoring A.
Case 4 in Table 1 is a more extreme Thus in all cases presented in Table 1,
version of Case 3. In this version, only one exhaustive information exchange could result
pro-A item of information (viz., a,) is shared in members' preferring A after discussion
and the other six are unshared; thus group and in the group's choosing A as its decision.
members should tend to prefer B rather than Nonetheless, Case 4 presents the most chal-
A at the onset of discussion because they lenging situation. In this case, for the group
each possess only three pro-A pieces of in- to eventually choose the best alternative (best
formation. When pregroup distributions of in terms of the total available information1),
information are sufficiently biased to shift
pregroup preferences in this way, we refer to
them as severely biased (as opposed to mildly 1
In this article, we use term best in a very restricted
biased distributions that tend to weaken but way. The best alternative, in our sense, is the alternative
not change initial preferences). An extension that is supported by the preponderance of the information
of the logic of Burnstein and Vinokur's (1977) that is collectively av3il3ble to the group. We do not
me3n that this alternative is correct. Indeed, criteria of
persuasive arguments theory suggests that correctness are usually undefined or unavailable for the
group discussion could counter this initial kinds of decision tssks that we sre considering (viz.,
bias in favor of B. By exchanging information, judgmental tasks; cf. Laughlin, 1980). Furthermore, we
each group member can gain several items recognize that in practice the information that is collec-
tively available to a particular group may be incomplete,
of new pro-A information but no new pro-B and thus the best alternative, relative to the informational
information. Ideally, even a group having a resources of that group, may not be the best alternative
pregroup distribution of information severely in view of a more complete set of information.
1470 GAROLD STASSER AND WILLIAM TITUS
the initial preferences of the members must information during discussion but sample a
be changed, whereas in the other cases a subset of the information to contribute to
group would choose the best alternative by discussion. At the level of the individual
simply acting on the initial predispositions of member, sampling is biased by the member's
the members. As a result, group discussion current preference. Preference consistent in-
is critical to effective decision making when formation is more salient and thus more
pregroup distributions of information are se- likely to be recalled during discussion (cf.
verely biased against the best alternative. Fishbein & Ajzen, 1975.) Furthermore, the
Case 4 also illustrates a situation in which contribution of recalled information to dis-
a group's decision should be better than the cussion is probably governed by a norm of
decisions of individuals acting alone. A po- advocacy: a social expectation that group
tential advantage of group versus individual members will actively advocate their preferred
decision making is that groups can pool alternative during debate. At the level of the
information and thus make a more informed group, sampling is biased by the number of
decision. In practice, groups are often com- members who are cognizant of a given piece
posed of members who are believed to have of information. The more members there are
unshared information (e.g., experts or repre- who have been exposed to an item of infor-
sentatives of special-interest populations). This mation, the more likely it is that at least one
strategy would have the greatest potential of them will recall and mention it. If sampling
benefit when the total information pool favors of the content of discussion is biased in these
one alternative but individuals have infor- ways, then it is also the case that the amount
mation that is severely biased in favor of of discussion favoring an alternative depends
another alternative. However, the following on the number of members who prefer that
biased sampling model of group discussion alternative.
suggests that this potential advantage of group Regarding the severely biased pregroup
decision making may often be unrealized. information distribution that is illustrated by
Case 4 in Table 1, the biased sampling model
Biased Sampling Model of predicts that group discussion will be biased
Group Discussion in favor of alternative B even though there
exists more information, in total, favoring A
The biased sampling model identifies sev- over B. This bias in discussion is expected
eral sources of bias in unstructured, face-to- for two reasons. First, group members will
face discussion when a group is confronted tend to prefer B initially and, as a result, will
with a consensus requirement. Moreover, we be predisposed to argue in favor of B. Second,
are concerned primarily with decision tasks shared information favors B even though the
for which there is no commonly accepted total set of information favors A. In summary,
system of logic that would lead to an unam- Case 4 illustrates a situation in which the
biguously correct decision. In Laughlin's collective decision of a group could be better
(1980) terms, we are restricting our attention (i.e., more consistent with the implications
to judgmental rather than intellective decision of all the available information) than the
tasks. For such tasks, according to the model, decision of any member acting alone. How-
the content of discussion is biased in two ever, the biased sampling model suggests that
ways. First, discussion is biased in favor of groups will often fail to realize their potential,
shared information: An item of information especially when pregroup distributions are
is more likely to enter discussion if it is biased sufficiently to shift initial preferences
shared rather than unshared. Second, discus- away from the best alternative, as in Case 4.
sion is biased in favor of the current prefer- The role of initial consensus in determining
ences of group members: An item of infor- the outcome of group discussion and decision
mation is more likely to enter discussion if it is emphasized by the biased sampling model.
favors rather than opposes the existent pref- This emphasis is not new in the group deci-
erences of group members. sion-making literature (cf. Davis, 1973; Stas-
According to the biased sampling model, ser, Kerr, & Davis, 1980) but deserves some
group members rarely exhaust their store of elaboration in the context of our study. In
UNSHARED INFORMATION AND GROUP DECISION MAKING 1471
terms of the initial consensus, Case 4 illus- via pre-experimental ratings. Thus Candidate A was the
trates the worst of all possible cases: Group best candidate in the sense that his profile contained
more of the consensually valued attributes of a student
members possess sets of information that are body president. We denned three experimental conditions
not only biased against the best alternative according to how the information about the candidates
but are also consistently biased in favor of was distributed over the 4 group members before discus-
another alternative. Thus we expect an initial sion. In the shared condition, participants read descrip-
consensus that favors an alternative other tions that contained all of the profile information about
each candidate. Two unshared conditions were used; in
than the best one. Limiting our example to both, a participant was given only partial information
two alternatives precluded illustrating another about each candidate. However, the distribution of infor-
possibility. If more than two decision alter- mation across a group's members was designed so that a
natives exist, it is possible for pregroup dis- group, collectively, had all of the information and poten-
tially could recreate the complete candidate profiles
tributions of information to be biased against during discussion. In the unshared/consensus condition,
the best alternative but not consistently, across positive information about A and negative information
members, biased in favor of another alterna- about B were unshared (i.e., given to only one member)
tive. In other words, none of the group mem- before discussion in order to bias initial preferences
bers may support the best alternative but, at against A and for B. In the unshared/conflict condition,
the same strategy for distributing positive information
the same time, they may not agree on any about A was used; however, negative information about
one of the remaining alternatives. Such a both B and C was unshared with the intent of shifting
lack of an initial consensus may avoid dis- pregroup preferences away from A but, at the same time,
cussion that uniformly supports one alterna- avoiding a strong initial consensus for either B or C.
tive and promote more extensive and exhaus-
tive sampling of the total available informa- Subjects
tion. In this way, initial disagreement or Miami University students participated in order to
conflict may actually encourage more effective partly fulfill a research experience requirement of intro-
information exchange. Others have noted the ductory psychology courses. Assignment to experimental
apparent value of conflict in group decision conditions and to decision-making groups within condi-
tions was random. There were 72 subjects in the shared
making (e.g., Fisher, 1980; Janis, 1972). and unshared/conflict conditions and 84 in the unshared/
We designed our study to examine two consensus condition.
implications of the biased sampling model of
group discussion. First, when pregroup dis- Materials
tributions of information are severely biased
against one alternative, group discussion tends The candidate profiles contained 16 items of infor-
mation about each of the three candidates. This infor-
to enhance rather than erode this initial bias. mation consisted of biographical data (e.g., extracurricular
Second, discussion is more likely to counter activities and hobbies, academic classification and major,
an initial bias when there is disagreement grade point average) and positions on local and university
that is due to conflicting patterns of infor- issues such as dorm policies (e.g., visitation hours, dorm
assignments), academic policies (e.g., class evaluations,
mation across group members than when
course scheduling), and student social life (e.g., program
pregroup information is consistently biased board activities, local drinking ordinances). On the basis
in favor of one alternative. We examined of an independent sample's pre-experimental ratings of
these predictions by comparing the distribu- candidate characteristics and policy statements, we con-
tions of pregroup preferences with the distri- structed the profiles to include a specific number of
positive, negative, and neutral items of information.
butions of group decisions and postgroup Positive items are those that were rated, on the average,
individual preferences. as both desirable and important attributes for a candidate
by pretest participants, whereas negative items were rated
as both moderately undesirable and important. (In order
Method
to avoid unrealistic profiles, extremely undesirable attri-
Overview butes were not used.) Neutral items either were judged
unimportant or received neutral desirability ratings. The
University students read descriptions of three hypo- profile for Candidate A contained eight positive, four
thetical candidates for student body president and then neutral, and four negative items, whereas the profiles for
met in 4-person groups to decide which candidate was Candidates B and C contained four positive, eight neutral,
best suited for the position. The profile of Candidate A and four negative items.
contained more positive and fewer neutral attributes than The descriptions that were read by participants before
the profiles of Candidates B and C. The valences (positive, group discussion were based on the overall profiles and
neutral, or negative) of profile attributes were determined included a summary of biographical information followed
1472 OAROLD STASSER AND WILLIAM TITUS
Results Table 3
Relative Frequencies of Pregroup and Postgroup
Pre- and Postgroup Individual Preferences Preferences and Group Decisions
The distributions of pregroup preferences, Candidate
given in Table 3, are significantly different
among the experimental conditions, x2(4, N = Condition B C
228) = 51.44, p < .001. In the shared con-
Pregroup preferences
dition, Candidate A was the popular choice
Shared info. .67 .17 .17 72
as expected. By comparison, Candidate A Unshared info./
was significantly less popular in the unshared consensus .25 .61 .14 84
conditions, x 2 U, N = 228) = 51.63, p < Unshared info./
.001, and Candidate B was more popular, conflict .21 .46 .33 72
Postgroup preferences
X 2 (l, N = 228) = 27.95, p < .001. Thus the Shared info. .85 .04 72
method of distributing unshared information Unshared info./
had the intended effect of reducing pregroup consensus .20 .75 .05 83"
support for A and increasing support for B. Unshared info./
conflict .17 .47 .36 72
Furthermore, Candidate C was more popular
Group decisions
in the unshared/conflict condition than in Shared info. .83 .11 .06 18
the unshared/consensus condition, x20, Unshared info./
JV = 156) = 7.91, p < .005. In summary, consensus .24 .71 .05 21
pregroup support shifted from A to B in the Unshared info./
conflict .12 .53 .35 17"
unshared/consensus condition, whereas in the
unshared/conflict condition, both B and C • One participant failed to report a postgroup preference.
b
gained support at the expense of A. One group failed to reach a decision.
One can assess the effects of group discus-
sion on individual preferences by comparing
pregroup with postgroup preference distri- tion than in the unshared conditions, x 2 (!>
butions. The biased sampling model suggests TV = 227) = 89.51, p < .001, and B was chosen
that candidates having the predominance of more frequently in the unshared conditions
support before discussion will maintain or than in the shared conditions, x2( 1, N =
even gain support during discussion. This 227) = 51.17, p < .001. Finally, C was more
expectation follows because the content of popular in the unshared/conflict condition
discussion is presumed to reflect the initial than in the unshared/consensus condition,
allegiances of group members. If, however, X 2 (l, N = 155) = 24.17, p < .001.
groups do effectively pool members' infor- In sum, the patterns of differences obtained
mation, group discussion should substantially for pregroup preferences tend to be exagger-
increase support for Candidate A in the ated in the postgroup data. Our expectation
unshared conditions. was that conflicting patterns of pregroup in-
The distributions of postgroup preferences, formation would increase the likelihood of
given in Table 3, are consistent with the discussion shifting preferences toward A when
biased sampling model predictions. The dif- information was unshared; this expectation
ferences among the experimental conditions was not supported. The postgroup support
observed in the pregroup preferences are for A is not significantly different between
enhanced rather than mitigated in the post- the unshared/consensus and unshared/conflict
group preferences. For example, in the shared conditions.
condition, the proportion of participants pre-
ferring A increased from .67 to .85, and in
the unshared/consensus condition the pro- Group Decisions
portion preferring B increased from .61 to
.75. Overall, the distributions of postgroup The distributions of group decisions in
preferences are significantly different, \2(4, Table 3 parallel closely the distributions of
N= 227) = 122.00, p < .001. Again, A was postgroup preferences. Overall, the distribu-
chosen more frequently in the shared condi- tions of decisions are significantly different
1474 GAROLD STASSER AND WILLIAM TITUS
ducing shifts of preference. The point is that the few groups who chose A) the one for
shifts of preference occur in the direction of whom members were given all of the positive
the initial consensus because members tend information before discussion. Unshared in-
to recall and contribute arguments and facts formation, in these conditions, was typically
that support their existent preferences. The (a) negative information about the winning
content of discussion serves not only to give candidate and (b) positive and negative infor-
members new information but also to change mation about one or both of the losing can-
the salience of old information. If shifts of didates. Thus if group discussion had cor-
preference and the emergence of a group rected the bias introduced by the patterns of
decision are dependent on the content of shared and unshared information in the un-
discussion, there should be a concomitant shared conditions, this disproportionate recall
increase in the salience of information that of positive-winning information should have
supports the winning candidate from pre- to been reduced in postdiscussion recall. How-
postdiscussion. Thus information that sup- ever, visual inspection of Figures Ic and le
ports the winning candidate should be more suggests that this bias in recall increased from
likely to be recalled after than before discus- pre- to postdiscussion.
sion. A 3 X 2 X 2 X 2 (Experimental Condi-
The numbers of positive and negative items tion X Winning vs. Losing Candidate X Pos-
recalled before and after discussion were tab- itive vs. Negative Valence X Prediscussion vs.
ulated separately for the winning and losing Postdiscussion) ANOVA of the recall scores
candidates. Because Candidate A had more
positive attributes than B and C, and because
0 POSITIVE
there was always more total information about • NEGATIVE
t- 40
recall of positive attributes of the winning Z
O 30
candidate is noticeably higher in the unshared cc
conditions than in the shared condition. This £ 20 ' _
effectively exchanged information during dis- more individuals possess the same resource
cussion, the collective choice would have does not increase the total set of available
been based on less biased information. resources but does increase the probability
Extending Steiner's (1972) process-loss of that resource being used" (p. 72). Further-
model of group productivity, Shiflett (1979) more, according to the biased sampling
proposed a general model that analyzes group model, members' initial preferences are a
performance in terms of resources, trans- transformer variable in Shiflett's formulation.
formers, and outputs. Resources include Initial preferences may mediate the contri-
knowledge and abilities that are relevant to bution of information to discussion in either
performing a task. Transformers are variables of two ways. First, members' recall may be
that mediate the ways in which member biased because preference-consistent infor-
resources are combined or weighted to yield mation is more salient than preference-incon-
the group output (e.g., a solution to a problem, sistent information. Second, members may
a judgment, or a decision). For our decision- tend to advocate or defend their initial pref-
making groups, the information in the can- erence and thus bias their contributions to
didate profiles can be viewed as resources, discussion even if their recall is not biased.
and the collective choice of the best-qualified In sum, initial preferences may act as trans-
candidate as the group output. Transformers former variables by introducing selective recall
would include variables that affect the likeli- or selective contribution of information dur-
hood that information is introduced into ing discussion.
discussion and thus is reflected in a group's In many contexts, a desirable goal may be
decision. Shiflett (1979) also distinguished to increase the weight of uniquely held infor-
unique resources (resources held by one mation in the determination of a group's
member) from redundant resources (resources judgment. One suspects that unshared infor-
held by all members). In our case, unique mation may often be just as important, or
resources are items of information unshared even more important, than commonly held
before discussion, and shared information information in arriving at a collective choice.
constitutes redundant resources. We suggest For example, if a group is composed of
that one criterion of group productivity in a members who have differing areas of exper-
judgmental task could be the extent to which tise, consideration of unshared information
unique resources are considered in a group's may be very critical to the quality of the
final judgment. Viewed in this way, the de- group's final decision. Similarly, groups whose
cisions of groups in the shared conditions are members represent disparate points of view
a standard against which to judge the decisions or special-interest populations may err by
of groups in the unshared conditions. If focusing on their shared perspectives and
unique and redundant informational re- thereby negating any advantage that accrues
sources were weighted equally, there should from multiple sources of diverse input. Our
be no difference in the distributions of group results suggest that unstructured discussion
decisions between the shared and unshared in the face of a consensus requirement may
conditions. However, our results suggest that often fail as a means of combining unique
unique or unshared information had little informational resources. Ironically, our anal-
impact on groups' decisions. ysis also suggests that this failure to consider
The biased sampling model of group dis- unique information is most likely when the
cussion identifies several possible mechanisms unique information counters the prevailing
that may result in commonly held information sentiment in the group and could change its
receiving more weight than uniquely held final decision.
information in a group's decision. One pos-
sibility follows directly from Shiflett's (1979) References
distinction between unique and redundant
resources: The more members there are who Anderson, N. H., & Graesser, C. (1976). An information
integration analysis of attitude change in group discus-
are exposed to an item of information, the
sion. Journal of Personality and Social Psychology, 34,
more likely it is that at least one of them will 210-222.
recall and mention it during discussion. In Burnstcin, E., & Vinokur, A. (1977). Persuasive argu-
Shiflett's (1979) words, "The fact that two or mentation and social comparison as determinants of
1478 GAROLD STASSER AND WILLIAM TITUS
attitude polarization. Journal of Experimental Social laughlin, P. R., & Earley, P. C. (1982). Social combination
Psychology, 13, 315-332. models, persuasive arguments theory, social comparison
Davis, J. H. (1973). Group decisions and social interaction: theory, and choice shift. Journal of Personality and
A theory of social decision schemes. Psychological Social Psychology, 42, 273-280.
Review, 80, 97-125. Myers, D. G., & Lamm, H. (1976). The group polarization
Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention phenomenon. Psychological Bulletin, 83, 602-627.
and behavior: An introduction to theory and research. Myers, J. L., DiCecco, J. V., & Lorch, R. F., Jr. (1981).
Reading, MA: Addison-Wesley. Group dynamics and individual performances: Pseu-
Fisher, B. A. (1980). Small group decision making: dogroup and quasi-/7 analyses. Journal of Personality
Communication and the group process (2nd ed.). New and Social Psychology, 40, 86-98.
York: McGraw-Hill. Shiflett, S. C. (1979). Toward a general model of small
Hoffman, L. R., & Maier, N. R. F. (1964). Valence in group productivity. Psychological Bulletin, 86, 67-79.
the adoption of solutions by problem-solving groups: Stasser, G., & Davis, J. H. (1981). Group decision making
Concept, method, and results. Journal of Abnormal and social influence: A social interaction sequence
and Social Psychology, 69, 264-271. model. Psychological Review. 88, 523-551.
Janis, I. L. (i 972). Victims ofgroupthink. Boston: Hough- Stasser, G., Kerr, N. L., & Davis, J. H. (1980). Influence
ton Mifllin. processes in decision-making groups: A modeling ap-
Kaplan, M. F. (1977). Discussion polarization effects in proach. In P. Paulus(Ed-), Psychology oj group influence
a modified jury paradigm: Informational influences. (pp. 431-477). Hillsdale, NJ: Erlbaum.
Sociometry, 40. 261-271. Steiner, I. D. (1972). Group process and productivity.
Laughlin, P. R. (1980). Social combination processes of New York: Academic Press.
cooperative problem-solving groups on verbal intellec-
tive tasks. In M. Fishbein (Ed.), Progress in social
psychology (Vol. 1, pp. 127-155). Hillsdale, NJ: Erl- Received February 1, 1984
baum. Revision received July 6, 1984