Journal of Personally and Social Psycholog> Copyright 1986 by the American Psychological Association. Inc.

1986. Vol. 50. No. 6 . ' l l 4 l - I I S I 0022-3 S14/86,'JOO.1'!

Group Polarization: A Critical Review and Meta-Analysis

Daniel J. Isenberg
Harvard Graduate School of Business Administration

This article critically reviews recent (1974-1982) group polarization studies that address themselves
to either one of the two primary explanatory mechanisms thought to underly group polarization,
namely social comparison and persuasive argumentation processes. A summary of the effect sizes of
21 published articles (33 independent effects) suggests that social comparison and persuasive argu-
mentation occur in combination to produce polarization, although the persuasive argumentation
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

effects tend to be larger. Attempts are made to reconcile the two positions, and some suggestions for
This document is copyrighted by the American Psychological Association or one of its allied publishers.

further research are offered.

In 1961 James Stoner observed that group decisions are riskier comparison processes (Sanders & Baron. 1977), and persuasive
than the previous private decisions of the group's members argumentation (Burnstein & Vinokur, 1977).
(Stoner, 1961). Since that time several hundred studies have 4. Finally, the polarization literature has made a substantial
shown that (a) the ''risky shift" is a particularly pervasive phe- contribution to social psychological theory. As one example,
nomenon: (b) on certain decisions groups are more cautious than Brown (1974) and Myers (1982) both have noted that the group
their members: and that (c) both risky and cautious shifts are polarization literature is significant for its emphasis on a counter-
special cases of a more general phenomenon—group-induced conformity effect because groups shift away from the average
attitude polarization (Moscovici & Zavalloni. 1969; Myers & attitude in the group rather than toward it.
Lamm, 1976). Group polarization is said to occur when an initial Another review of the polarization literature would be redun-
tendency of individual group members toward a given direction dant as several excellent reviews of the literature up to 1978
is enhanced following group discussion. For example, a group already exist (see Lamm & Myers, 1978; Myers, 1982; Myers &
of moderately profeminist women will be more strongly profem- Lamm. 1975. 1976; see also Clark, 1971: Pruitt. 1971a, 1971b).
inist following group discussion (Myers. 1975). Thus, on decisions The primary purpose of the present article is to critically review-
in which group members have, on the average, a moderate pro- in particular the recent literature (especially 1974-1982) that
clivity in a given direction, group discussion results in a more addresses itself to the dialogue between proponents of social
extreme average proclivity in the same direction. comparison and persuasive arguments as explanations of attitude
The group polarization literature is an encouraging example polarization (Burnstein & Vinokur, 1977; Sanders & Baron,
of how theoretically and practically meaningful phenomena in 1977). The mid-1970s serve as an appropriate point of departure
social psychology can be defined and explored through empirical because it was at this time that the debate concerning these two
research: explanatory processes became highly salient. Furthermore, it was
1. In recent years polarization research has been cumulative then that many researchers who had helped produce the hundreds
such that subsequent research studies have addressed the prob- of risky-shift studies began to veer away from further research
lems and issues identified by previous researchers. This has been in the field. At the time it appeared that the risky-shift was an
amply demonstrated by several excellent literature reviews interesting but limited and severely qualified phenomenon that
(Lamm & Myers, 1978; Myers, 1982; Myers & Lamm. 1976). had already outlived its theoretical usefulness (e.g., Cartwright,
2. Researchers have pursued lines of programmatic research 1973; Kleinhans & Taylor, 1976). In very recent years enthusiasm
rather than one-shot experimental studies (e.g., Blascovich & for group polarization research has again begun to wane: it is
associates: Myers & associates: Vinokur & Burnstein; Baron & hoped that an in-depth review will help reorient research in the
associates). field, in particular toward a third wave of research that will in-
3. Theoretical explanations of polarization phenomena have tegrate group polarization with other social psychological and
been disconfirmed (see Pruitt, 1971a. 1971b), thus focusing re- cognitive phenomena. This article will conclude by suggesting
search on an increasingly small number of explanatory mecha- four areas of integration. In addition to reviewing this recent
nisms. For example, in 1971 Pruitt identified 11 (overlapping) literature, attention will be paid to reporting the relative effect
explanatory mechanisms for choice shifts, whereas by 1978 this magnitudes of demonstrations of persuasive argument and social
list had been pruned and consolidated into 2 major ones, social comparison mediating mechanisms.

The author would like to thank Robert F. Bales. Roger Brown. David Social Comparison
Myers, Reid Hastie, and anonymous reviewers for their help and sugges-
tions at various stages of this research. Introduction
Correspondence concerning this article should be addressed to Daniel
J. Isenberg, Harvard Graduate School of Business Administration, Soldiers As far back as Brown's (1965) seminal discussion of the risky
Field Road. Boston, Massachusetts 02163. shift, one of the major explanations of choice shifts has been a
1141
 1142 DANIEL J. ISENBERG

social comparison explanation. According to this perspective, tween self-enhancement and humility. In reality, however, it may
people are constantly motivated both to perceive and to present be very difficult to distinguish these two mechanisms empirically.
themselves in a socially desirable light. In order to do this, an There are several overlapping variations on the pluralistic ig-
individual must be continually processing information about how norance and bandwagon effects, such as cultural values (Hong.
other people present themselves, and adjusting his or her own 1978), release mechanisms (Pruitt, 197la, 197Ib), self-presen-
self-presentation accordingly. Some versions of social comparison tation processes (Jellison & Arkin. 1977), specific values (Stoner,
theory state that many of us desire to be perceived as more fa- 1968), and self-anchoring (implicit in Festinger, 1954; Brown,
vorable than what we perceive to be the average tendency. Once 1974).'
we determine how most other people present themselves, we
present ourselves in a somewhat more favorable light. When all The Evidence
members of an interacting group engage in the same comparing
process, the result is an average shift in a direction of greater The major source of support for social comparison theory has
perceived social value. come from demonstrations that simple knowledge about other
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

There are two variations of the above sequence, one empha- group members' positions by itself can produce polarization ef-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

sizing the removal of pluralistic ignorance, and the other em- fects. These effects are called "mere-exposure" effects, and we
phasizing one-upmanship (bandwagon effects). According to the will use that terminology here. A number of studies have at-
pluralistic ignorance explanation (see Levinger & Schneider, tempted to establish that mere exposure to central tendency in-
1969; Pruitt. 197la; Schroeder. 1973: see also Isenberg, 1980), formation can be a sufficient cause of choice shifts (Baron &
individuals present their own positions as compromises between Roper. 1976: Blascovich & Ginsburg, 1974:Blascovich, Ginsburg.
two tendencies, the desire to be close to one's own ideal, and the & Howe, 1975, 1976; Blascovich. Ginsburg, & Veach, 1975:
desire not to be too deviant from the impression of the central Goethals&Zanna, 1979; Myers, 1978; Myers etal., 1980; Myers
tendency of the group. Prior to group discussion, group members etal.. 1977: see also Pruitt's reviews, 1971a, 1971b). I shall review
initially underestimate the group norm and. judging from the each study in some depth.
initial ratings of their own positions, are somewhat distant from Blascovich and associates. Blascovich and his associates
their ideal. During group discussion, individuals are exposed should be noted for their persistence in pursuing mere-exposure
more nearly to the true group norm, and thus a discrepancy effects, as well as for the fact that they use true risk-taking sit-
between how much better an individual is and would like to be uations in their research. In one study (Blascovich, Ginsburg, &
becomes apparent. Upon making a second set of choices, the Veach, 1975), three experimental conditions were formed for
individual group members shift closer to their ideal positions. blackjack playing. One was an individual condition, one was a
When this process takes place with most group members, an group-without-discussion condition, and the third was a group-
overall polarization is observed. with-discussion condition. Subjects were randomly assigned to
The assumption underlying the pluralistic ignorance expla- experimental conditions and played 20 hands of blackjack alone
nation of choice shifts is that pluralistic ignorance exists because in order to establish a baseline. Then each subject played 20
of a lack of accurate communication about the "true" beliefs of hands of blackjack in the experimental condition. In the group-
the majority of group members, although it is also likely that without-discussion condition, subjects heard each others' bets
pluralistic ignorance may be due to cognitive biases, such as but did not discuss them. In the group-with-discussion condition,
self-other differentials in person perception (see Jones & Nisbett. subjects placed one collective bet after trying to reach a consensus.
1972). There were two relevant findings from the Blascovich, Gins-
Other investigators have hypothesized a second explanation burg. and Veach (1975) study. First, the individual (no-group)
for the effects of social comparison processes on polarization
(e.g., Brown. 1974: Myers. 1978; Myers. Bruggink, Kersting, & 1
Schlosser, 1980; Myers, Wojcicki, & Aardema. 1977). These and A word about the latter is in order because it has been implied by
the literature but never explicitly stated. Festinger (1954) argued that one
other investigators hypothesize that people are motivated by a
reason people seek out comparison information is in order to define
desire to be different and distinct from other people in a valued social reality. For example, what it takes to be considered "intelligent"
direction (Fromkin, 1970). In addition, people are also motivated or "conservative" depends on an individual's comparison with how much
to present themselves somewhat more favorably than other peo- of these qualities the average individual possesses. If the average IQ is 75.
ple. In other words, w-e want to be different from as well as better then an IQ of 90 is intelligent. If the majority of the people in a given
than other people. Brown states, "To be virtuous . . . is to be population are against racial integration, then an individual favoring bus-
different from the mean—in the right direction and to the right ing is considered very liberal, even if he or she is against interracial mar-
degree" (1974, p. 469). When making initial ratings along a di- riage. In other words, these qualities are socially defined. This aspect of
mension, individuals give themselves a rating that is somewhat social comparison theory is usually overlooked in explaining polarization.
It is possible that the motivation underlying differentiation of the self
more favorable than the rating they presume the average group
from the presumed group norm is to maintain one's self-definition, not
member will give. When individuals directly or indirectly infer necessarily to enhance it. As Brown (1974) comments, "Giving advice
what the true norm is, they then "improve" their own ratings, in private, then, each participant means to be somewhat audacious. But
thus producing the overall choice shift. Whereas the mechanism how can he know how to be so since the situations are novel and invented?"
producing choice shifts in the pluralistic ignorance explanation Thus, information about the central tendency defines what response is
is a compromise between self-enhancement and conformity, the virtuous, nonvirtuous, risky, or cautious. Social reality and the social self
mechanism underlying bandwagon effects is a compromise be- are defined only through information about other people.
 GROUP POLARIZATION 1143

control did not increase the size of their bets between the first ment. The prediction from the mere-exposure hypothesis was
and second blocks of 20 hands, whereas both group conditions that for the attractive face, the second member's judgment should
did show marked increases. r(14) = .44 and .37 for the with- be more attractive than the first member's, and the reverse should
and without-discussion conditions, respectively.2 Second, neither be true for the unattractive face. In fact, there was a 1.75 to 1
group condition showed a greater polarization than the others tendency for the second member to polarize his or her judgments.
from block one to block two. This finding was extended by Blas- As Myers argued, it is rather difficult to imagine that novel or
covich et al. (1976). r(\2) = .44. valid arguments could be generated by the second member in
Blascovich. Ginsburg. and Veach (1975) and Blascovich et al. order to rationalize the shift to a more extreme judgment.
(1976) interpreted their studies as showing that norms of risk or A recent series of studies by Myers and his associates have
caution emerge, driving either risk or caution upward only when provided additional evidence for the mere-exposure hypothesis
the norm is observable by the individual. In agreement with (Myers, 1978; Myers et al., 1980: Myers et al., 1977). In the first
Jellison and Arkin (1977), these authors suggested that group of these studies (Myers et al.. 1977) the authors contrasted plu-
members associate ability and skill with riskiness (or caution) ralistic ignorance (Levinger & Schneider. 1969) and release theory
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

and thus become more risky (or more cautious) in order to appear (Pruitt. 1971a. 197 Ib) explanations of polarization by exposing
This document is copyrighted by the American Psychological Association or one of its allied publishers.

more able and skillful (see also Blascovich & Ginsburg. 1974; subjects to the group average response or to extreme responses
Blascovich, Ginsburg, & Howe, 1975). Nevertheless, one potential within the group. The awareness of group norms is assumed to
weak spot in their argument is that in order to claim ability- be the mediating mechanism in pluralistic ignorance (Levinger
attribution mediators they would have to explain why both shifts & Schneider, 1969). Awareness of extremes, which presumably
to risk and shifts to caution are found, and why either can be releases group members from the constraints of the perceived
interpreted as ability under different circumstances. group norms, is the mediating mechanism in release theory
Baron and Roper. Experiments by Baron and Roper (1976) (Pruitt. 197 la, 197 Ib). An attitude survey of 269 members of a
add support to the mere-exposure hypothesis and specifically church was conducted in two stages. In stage 1. 100 members
attempt to show how the direction of social value predicts the rated their own agreement/disagreement with 16 church-related
direction of polarization. In Experiment 1, subjects participated statements (e.g.. "Ministers should feel free to take a stand from
in an autokinetic study in which they were assigned to one of the pulpit on some political issue."). Three weeks later the re-
three conditions, depending on whether they were told that per- maining members (169) either (a) responded to the statements
ceiving smaller, larger, or better estimates of light movement was in a control condition: (b) were given the average of the previous
positively related to intelligence. Subjects in the three experi- 100 members' responses and then responded to the 16 statements;
mental conditions made their light estimates alone, for a 15-trial or (c) were given a frequency distribution of the previous 100
baseline, and then for an additional block of 15 trials as a group members' responses and then responded to the 16 statements.
in which members simply called their estimates out loud so that Those subjects in the average-exposure and frequency-exposure
other group members could hear. In a control condition, subjects condition showed significantly more extreme attitudes compared
were told that larger estimates were related to intelligence, but to the controls, ij = .23. r(265) = .17." The frequency-exposure
they made the second block of 15 estimates as individuals, without condition showed more extreme attitudes than the pretest baseline
being exposed to others' estimates. The results of this first study as well. There were no differences between subjects in the average
were mixed. In the larger intelligence condition, subjects' esti-
mates did polarize in the predicted direction (toward larger es-
timates), r(44) = .41, but they did not polarize in the other con- - Throughout this article I recomputed the / and F statistics to the
ditions. It appeared that there was an overall tendency for esti- more universal measure of effect size, r, which is the Pearson product-
mates to increase over blocks, but the authors did not report a moment correlation and therefore can be identically summed, averaged,
block main effect. These findings were replicated and extended and tested for significance. The formula for computing r from Ms r = t/
(l2 •*• df)1'2. For F(\. df), r = F/(F + <#')' 2 since t2 = F for degrees of
in subsequent experiments (Baron & Roper, 1976): /"(39) = .43.
freedom = 1. of/'(see Rosenthal. 1978). In several studies investigators
These studies are particularly interesting because the task that inappropriately tested specific hspotheses with the omnibus F using greater
subjects performed was an argument-poor one, in other words, than I dim the numerator. In each of these studies it was possible to test
one in which it was difficult for subjects to generate novel and a specific hypothesis with I d f i n the numerator, thus allowing a more
valid arguments either in favor of or against perceiving light mo- powerful test and allowing the computation of r. In all cases, an r greater
tion (see Vinokur & Burnstein. 1978b). It appears that this is at than 0 indicates an effect in the predicted direction. Further details of
least one example of a situation that one would find difficult to these calculations are available from the author.
3
explain without assuming some purely value-determined com- Nevertheless, it behooves the authors to explain within their framework
parison process (Sanders & Baron, I977). 3 why the smaller intelligence condition in Experiment 1 did not produce
Myers and associates. Another study (unpublished, reported the predicted shift.
4
This is only an estimate of r because the exact cell means are not
in Myers, 1982) demonstrated that exposure to others' positions
given in M\ers et al. (1977). but were estimated from the histogram
on argument-poor tasks can produce polarization. Dyads were presented in the article. Furthermore, the r reflects the effect size of the
shown slides of faces that had been judged previously as either difference between the two exposure conditions and the two control con-
attractive or unattractive. On each trial, one member of the dyad ditions. Basically, it is the effect size of the F ratio for the comparison
made a judgment of the attractiveness of the slides and announced between exposure and nonexposure. The SS (between) was calculated
the judgment out loud so that the other dyad member could from the estimated means, and given the F ratio from the article the SS
hear. Then the second member of the dyad made his or her judg- (error) was calculated.
 1144 DANIEL J. ISENBERG

and frequency-exposure conditions, thus giving an explanatory held up a card showing his or her position (I will refer to this
advantage to neither release nor pluralistic ignorance theories. condition as the partial-exchange condition). An information-
The first of two studies by Myers (1978) replicated and ex- exchange-of-position-flw/-ability condition (which I will refer to
tended the above findings by using eight Choice Dilemmas as the full-exchange condition) was like the partial-exchange
Questionnaire (CDQ)5 items, four risky and four cautious, and condition except that group members also shared their self-ratings
the eight traffic cases from Kaplan's (1977) study, four innocent of ability. (As expected, group members were similar in rating
and four guilty shifting cases. Exposure caused polarization themselves as above average in ability). In a control condition,
for both the CDQ items. r(\02) = .27, and the traffic cases. subjects simply rerated the CDQ items privately after reconsi-
r(!02)= .44. dering them for 10 min. Because the variances across conditions
It is commendable that these studies pit theoretical predictions were heterogeneous, nonparametric statistics were used to com-
against one another. Nevertheless, it should be pointed out that pare the number of groups polarizing to risk in each condition.
the above comparisons between release and pluralistic ignorance Overall, the amount of polarization varied quite substantially
explanations have two potential flaws. First, the manipulation is among conditions, x2 (3. -V = 32) = 10.42. p < .02, being greatest
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

confounded with the amount of information: Subjects in the in the full-exchange condition. More specifically, both the group-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

release conditions were exposed to a full frequency distribution, discussion and the full-exchange conditions showed greater po-
whereas subjects in the pluralistic ignorance condition were ex- larization to risk than did the partial-exchange condition (p <
posed to one number, an average. A second criticism is that dif- .05: my own reanalysis of the differences between the partial-
ferences between scores from release and pluralistic ignorance and full-exchange conditions showed x2 ( U -V = 16) = 6.35. p <
conditions may be attenuated by the low salience of the infor- .02. r( 14) = .63. After the experiment, group members rated the
mation (see Borgida & Nisbett, 1977). Whereas Myers found average group comember in terms of his or her overall talent,
polarization effects for both conditions when subjects in both of creativity, and ability, and it was found that indeed subjects in
those conditions were exposed to relatively abstract and unvivid the full-exchange condition saw other group members as more
information, exposure to more salient information about the similar than did subjects in the partial-exchange condition.
norm or extremes (for example, via video tape) may actually The authors interpreted these findings in support of the me-
lead to greater polarization for one or the other of the two con- diating role of perceived similarity of abilities in producing po-
ditions. The findings of Isenberg and Ennis (1981) can be inter- larization, although this interpretation raises two issues. First,
preted as showing that deviant group members (i.e.. extremes) what is the relation between risk taking and risk advocacy? The
are in fact particularly salient in the minds of other group mem- risk-ability link assumes that Person A will be a good comparison
bers.6 for risk taking for Person B if Person B perceives their risk-related
In a second experiment. Myers (1978) controlled for the effects abilities as similar. This was not the comparison performed by
of repeated measurements and had successive groups of subjects group members in the full-exchange condition, because risk tak-
rate the same eight CDQ items after having been exposed to the ing was not an issue, only the advocacy of risk (cf. Blascovich &
actual ratings of the previous group. As predicted, polarization Ginsburg. 1978). Second, a different interpretation might suggest
was greatest for the total of 60 subjects in the exposure conditions. another mediating mechanism, namely that exposure to others'
r(\ 16) = .33. Subsequent studies by Myers et al. (1980) have self-ratings enhanced source credibility: "I had better listen to
replicated and extended their demonstration of mere-exposure all of these (self-proclaimed) talented people who are advocating
effects. r(18) = .86" and r(21) = .43. more risk than I had expected them to."
Goet/ials and Zanna. Goethals and Zanna (1979) reported
one relatively recent attempt to demonstrate that normative pro- Informational Influences
cesses alone can produce polarization. Following reconceptual- Introduction
izations of Festinger's (1954) statement on social comparison
theory (e.g.. Goethals & Darley. 1977; Jellison & Arkin. 1977: Much research has been devoted to studying how the proc-
Jellison & Riskind. 1970). Goethals and Zanna argued that the essing of relevant information can affect group polarization (e.g..
mere-exposure hypothesis will onl\ be true when group members
believe that they are similar to one another on attributes that 5
The original risky-shift research bv Stoner (1961) used scenarios in
are related to the judgments being made. Since risk taking and which subjects read each scenario and then recommended how much
ability are perceived as related, group members who perceive risk thev thought the character in the scenario should take. Decision
themselves as similar in ability should polarize to greater risk scenarios involved chess moves, career shifts, professional choices, and
following group discussion. "Social comparison theory implies so forth.
6
that people will feel it is appropriate to take as much risk as A third qualification of the Myers studies is the possibilitv that release
others of equal ability but less risk than those who possess greater theory and pluralistic ignorance theory are in fact different, but the effect
ability" (p. 1470). A total of 137 subjects responded individually is so small as to be trivial. Both of Myers's studies showed frequencv-
exposure conditions to be slightlv more effective than average-exposure
to four CDQ items, rated themselves on overall "talent, creativity,
conditions in producing polarization, but the effects, simply and com-
and ability," and then were assigned to one of four conditions. bined, were nonsignificant.
In a group-discussion condition, groups of four subjects discussed " This is based on the comparison of the polarization in the three
the four CDQ items and then completed the items again in pri- experimental conditions versus the controls. The omnibus f\'3. 18) =
vate. In an information-exchange condition, groups of subjects 3.73. and given the 4 cell means 55 (between) was calculated and the
were exposed to each other's positions when each group member contrast performed.
 GROUP POLARIZATION 1145

Anderson & Graesser, 1976; Bishop & Myers. 1974: Ebbesen & priate information. PAT can predict the direction and extent of
Bowers, 1974: Kaplan, 1977: Kaplan & Miller. 1977: Madsen. choice shifts, be they polarizing or depolarizing (Vinokur &
1978: Vinokur & Burnstein, 1978a). The most sophisticated and Burnstein, 1978a). (b) PAT facilitates the conceptual integration
well-researched version of the information processing explanation of individual and group decision making, since the underlying
for choice shifts is persuasive arguments theory (PAT; e.g.. Burn- mechanism is the same for arguments processed privately or in
stein & Vinokur. 1975. 1977; Burnstein, Vinokur, & Trope, 1973: interaction with other people.
Madsen. 1978; Vinokur & Burnstein. 1974. 1978a). PAT holds
that an individual's choice or position on an issue is a function
of the number and persuasiveness of pro and con arguments that The Evidence
that person recalls from memory when formulating his or her
own position. Thus, in judging the guilt or innocence of a trial The evidence for the proposition that persuasive arguments
defendant, jurors come to predeliberation decisions on the basis alone can produce choice shifts and attitude polarization is quite
of the relative number and persuasiveness of proguilt and proin- strong and from the start has been one of the best-supported
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

nocence arguments. Group discussion will cause an individual explanations of polarization phenomena (see Pruitt. 197la.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

to shift in a given direction to the extent that the discussion 1971 b). However, the statement that only a persuasive argument
exposes that individual to persuasive arguments favoring that mechanism mediates choice shifts (e.g.. Burnstein & Vinokur.
direction. 1977) is premature on empirical and theoretical grounds (see
Since the notion of persuasiveness is so central to PAT, some Sanders & Baron, 1977, and Burnstein & Vinokur, 1977. for a
research has been devoted to ascertaining the characteristics of debate on this particular issue). I will summarize the major find-
arguments that make them persuasive. Burnstein (1982; Vinokur ings in support of PAT. emphasizing the most recent additions
& Burnstein. 1978b) persuasively argued that two factors deter- to the literature not covered in the reviews by Lamm and Myers
mine how persuasive a given argument will be. One factor is the (1978) and Myers and Lamm (1976). The evidence will be or-
perceived validity of the argument. How true is the argument? ganized around three hypotheses:
Does the argument fit into the person's previous views? Does 1. There is a correlation between the extent of polarization
the argument logically follow from accepted facts or assumptions? and the prior preponderance of pro and con arguments that are
The second factor determining persuasiveness is the perceived available to group members (the correlational hypothesis).
novelty of the argument. Does the argument represent a new way 2. Group polarization can be caused by manipulating the
of organizing information? Does the argument suggest new ideas? preponderance of pro and con arguments that are processed (the
Does the argument increase the perceiver's access to additional causal hypothesis).
information that is stored in memory? For example, the argu- 3. PAT is a necessary and sufficient cause of group polarization
ment. "Cigarette smoking is bad because it causes cancer in the whereas social comparison is neither necessary nor sufficient (the
smoker." is valid, but it is not novel anymore. The argument. exclusivity hypothesis).
"Cigarette smoking is bad because it causes cancer in nonsmokers Tlie correlational hypothesis. Several studies have shown that
who inhale the smoke when smokers are present." is relatively there is a good correlation between the preponderance of pro
novel. To the extent that the two arguments are perceived as and con arguments possessed by group members and the size
equally valid, the latter, more novel argument should be more and direction of the postdiscussion polarization. For example,
persuasive. Together, the perceived validity and perceived novelty Madsen (1978) in Experiment 1, had subjects in one condition
of an argument determine how influential that particular argu- generate arguments pro and con on public sex education either
ment will be in causing a choice shift. The novelty-persuasiveness in their own home state or in a geographically distant state. They
hypothesis has received experimental support (Vinokur & Burn- then rated the persuasiveness of the arguments. In the second
stein. 1978b). condition, eight groups of subjects completed pre- and postdis-
PAT seriously qualifies the risky shift phenomenon by making cussion ratings of their own support of sex education in their
shifts contingent upon the argument pool within the group. A own home state, while eight groups of subjects performed the
given group may or may not shift in a given direction, depending same task for sex education in a distant state. An index of average
upon the possession and expression of persuasive arguments persuasiveness (Vinokur & Burnstein. 1974) derived from ar-
during the group discussion. The role of novelty is particularly guments for and against public sex education was highly predic-
central. If arguments are presented that the individual group tive of the actual shifts towards greater or lesser support of public
member is already aware of, a shift in his or her position will sex education. /"(13) = .51. across 16 groups of subjects with one
not occur as a result of the discussion (Kaplan. 1977). If novel covariate.
persuasive arguments are presented that are opposite to the di- In Experiment 2, Madsen used the same paradigm but changed
rection initially favored by the group member, their position will the issue to be three CDQ-like scenarios involving drug usage.
shift in the opposite direction and depolarize (Kaplan. 1977: These three scenarios were crossed with a between-subjects ma-
Vinokur & Burnstein. 1978a). Thus, a juror who initially favors nipulation of issue importance. Again, one large group totaling
a guiltv verdict will come to favor a more guilty verdict if and 50 subjects generated and ranked the persuasiveness of pro and
only if he or she is exposed to novel persuasive arguments in con arguments, and 12 small groups totaling about 50 subjects
favor of guilt. completed pre- and postdiscussion ratings of their support of
The specificity of the process by which PAT produces choice the courses of action proposed in the scenarios. Again, there was
shifts lends it two major strengths as a theory: (a) given appro- a high correlation between the average persuasiveness index for
 1146 DANIEL J. ISENBERG

each scenario and the direction and magnitude of the observed Burnstein, 1974, I978a, 1978b). The greater the number of per-
choice shifts. r(4) = .82 and .64. by two alternative methods.8 suasive arguments that are novel or nonredundant in a group,
These studies and one by Ebbesen and Bowers (1974, Exper- the greater the impact of those arguments on group members.
iment 1). r(9) = .65. suggest a high correlation between the pre- Thus, the partially shared (novel) arguments will have the most
ponderance of pro versus con arguments and choice shifts (see impact.
also Bishop & Myers. 1974: Vinokur& Burnstein, 1974. Exper- In their second study, Vinokur and Burnstein (1978b) explored
iment I). Nevertheless, as Madsen points out. there are examples whether novel and valid arguments were in fact any more effective
of imperfect predictions from PAT. such as in his second exper- in causing shifts in predicted directions than non-novel argu-
iment where two shift directions were incorrectly anticipated. ments. In one condition, subjects received novel arguments that
He suggests that this may be due to the actual dynamics of how- were prorisk mixed with non-novel arguments that were pro-
persuasive arguments possessed by individual members may or caution. In the second condition, the same subjects received novel
may not work their way into the actual group discussion. Rather precaution and non-novel prorisk arguments. The prediction
than surveying all of the relevant arguments, groups tend to be from the novelty-persuasiveness hypothesis would be that shifts
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

rather selective in their pursuance of limited lines of argumen- would occur in the direction of the novel arguments. In fact, this
This document is copyrighted by the American Psychological Association or one of its allied publishers.

tation. Similarly, it has been observed in a number of studies was clearly the case, independent of whether the item was a typ-
that group members censor the arguments they put forth during ically risk- or caution-shifting item. In addition, risky items did
discussion in order to support the emerging group consensus (cf. shift to risk more than did cautious items, but the effect was
Ebbesen & Bowers, 1974. Experiment 2: Myers & Lamm. 1976. weaker than the predicted effect. r ( 5 l ) = .93 versus .74. The
pp. 619-620). effect of novelty was particularly strong for the neutral item,
The causal hypothesis. Clearly, the establishment of a strong which shifted significantly in the direction of the novel arguments
correlation between persuasive argument processing and group depending on whether they were prorisk or precaution.
polarization is impressive, but it does not demonstrate a causal The exclusivity hypothesis. A number of studies have at-
link. Accordingly, a number of studies have gone one step further tempted to show that PAT is necessary and sufficient to produce
and directly manipulated the preponderance of pro and con ar- polarization effects and that only PAT can account for these ef-
guments in order to bring about corresponding shifts (e.g.. Burn- fects. The most recent of these will be reviewed here (Burnstein
stein & Vinokur, 1973, 1975; Burnstein et al., 1973: Ebbesen & & Vinokur. 1975; Laughlin & Earley. 1982; Vinokur & Burnstein.
Bowers, 1974, Experiments 2 and 3: Kaplan. 1977: Kaplan & 1978a: see also Burnstein & Vinokur, 1973: Burnstein et al.,
Miller. 1977: Vinokur & Burnstein. 1978b). 1973). One study (Burnstein & Vinokur, 1975) was designed to
Ebbesen and Bowers (1974) in Experiment 3 had subjects listen show that exposure to others' positions (mere-exposure) causes
to 10 risky and cautious arguments, while systematically varying people to privately generate persuasive arguments, which in turn
the proportion of risky to cautious arguments from . I to .9. They produces polarization. These authors attempted to demonstrate
found that the correlation between this proportion and polar- that exposure to others' positions causes polarization only when
ization to risk was .98 across five different proportions (.1. .3. it stimulates the generation of persuasive arguments. In a within-
.5, .7. and .9). In other words, the higher the proportion, the subjects design. 12 groups of 5 subjects responded to three risky
greater the polarization to risk. When the proportion fell below CDQ items in three conditions (after having completed one CDQ
.5, the group polarized to caution. item as practice). In the major experimental condition, subjects
More recently, Kaplan and Miller (1977) showed that subjects responded to one of the three items, were exposed to each other's
tended to recall persuasive arguments that they had been exposed responses, privately generated arguments for and against risk for
to most recently rather than the ones they had been exposed to that item, and then responded again to the same CDQ item. An
first. They then composed 24 six-person groups, half of which exposure control condition was identical to the experimental
were in a redundant condition, and half of which were in a novel condition except that subjects privately generated arguments for
condition. Each subject in the redundant condition received six and against risk for a different item (the practice item). Thus,
arguments, and the arguments were in the exact same order for subjects in this condition were prevented both from thinking
each subject. Each subject in the novel condition received the about others' responses and from generating relevant arguments.
same six arguments, but in a given group every subject received A no-exposure control condition had subjects respond to a CDQ
the six arguments in a different order. If subjects showed a recency item, privately generate arguments, and respond to the same
effect and recalled the most recent argument, subjects in the CDQ item again. Thus, the first condition purportedly shows
redundant condition should recall the same argument, whereas the effects of exposure on argument generation, which is hy-
subjects in the novel condition should recall different arguments. pothesized to mediate the effects of exposure on polarization.
To the extent that recalled arguments were discussed more in The authors found a risky shift in the experimental condition,
the groups, subjects in the novel condition should be exposed to r( 11) = .89, a nonsignificant cautious shift in the exposure con-
more novel arguments and thus should shift more. As predicted, trol, r = .43,' and no shift in the no-exposure control, r = .03.
the novel arguments groups showed a greater polarization effect. The experimental condition polarized to risk significantly more
r(!40) = .67, although groups in both conditions polarized sig-
nificantly. r( 140) = .76. 8
The second correlation is a more conservative post hoc analysis that
Further studies have shown that group polarization is a func- I conducted on Madsen's data.
tion of an information pool within a group, where the pool con- 9
We are not told which item was used as a practice item, but it is
sists of partially shared persuasive arguments (Kaplan. 1977, Ex- conceivable that subjects generated precaution arguments that then gen-
periment 3: Kaplan & Miller, 1977; Madsen. 1978: Vinokur & eralized to the focal item, thus attenuating the usual polarization to risk.
 GROUP POLARIZATION 1147

than the other two conditions, r(22) = .68. In comparing the posite the other subgroup with the labels ("Risky Subgroup" and
balance of actual prorisk and procaution arguments generated "Cautious Subgroup") in front of them. For the other groups,
by subjects in the experimental and no-exposure conditions, it members simply sat together. In both conditions group members
was found that the weight and number of prorisk arguments was discussed each CDQ item with the instructions to attempt to
greater in the experimental conditions, whereas this was not true reach consensus. After responding to all of the items in seven
for the procaution arguments. Although the difference between different subgroupings. subjects made their postdiscussion ratings
prorisk and procaution arguments was significant for the exper- on all seven items. The measure of attitude polarization was
imental condition but not for the no-exposure condition, the postdiscussion minus prediscussion ratings. The measure of de-
difference between these conditions was apparently not significant polarization was the difference between the means of each of the
as should have been predicted by the model. two subgroups of 3 subjects each.1'
This is an admirable study in its use of content analysis, in The findings of Experiment 1 were clear: both polarization of
the strength of the findings, and in its specific predictions that the total groups of 6 and depolarization between the two
were tested by planned comparisons.'0 Nevertheless, there are subgroups of 3 occurred, the latter being two to three times as
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

some aspects of the study that are vulnerable to criticism. The large as the former. These effects apparently were equivalent for
This document is copyrighted by the American Psychological Association or one of its allied publishers.

major criticism is that Burnstein and Vinokur (1975) first dis- groups in the salient and nonsalient subgroup conditions (effect
torted the SCT position somewhat and then attacked the distor- sizes are not given, nor do we know if the effects were in the
tion. For example, in the exposure control condition the exper- predicted direction). Groups polarized to risk on risky items and
imenter prevented subjects from thinking after they had been to caution on cautious items, but overall the two subgroups
exposed to others' positions by immediately giving them a task showed a strong tendency to converge toward one another. r( 17) =
to generate arguments for a different item. In a very real sense .87. l3 Analyses of the subgroups showed that on the risky items
this was a distraction task. The implicit assumption is that social the cautious subgroups shifted more toward risk than did the
comparison processes require no thought. However, this is not risky subgroups shift toward caution. Likewise, on the cautious
the case; information processing in SCT must occur at two junc- items the risky subgroups shifted more toward caution than did
tures, the first being in processing the fact that others are different the cautious subgroups shift toward risk. On the neutral item
from what one had expected, and the second being in the cognitive both subgroups tended to shift (depolarize) toward each other.
calculus of how to be different from the average "in the right Experiment 2 replicated and extended these findings but in-
direction and to the right degree." Burnstein and Vinokur (1975) stead of the CDQ items, subjects responded to two value items
also assumed the role of strong emotion in social comparison (e.g., "Do you think capital punishment is not justified under
processes, stating that group members are supposed to be "dis- any circumstances or is justified for special cases of murder?"),
tressed" (p. 414). "disturbed," or "surprised" (p. 417) by dis- two personal taste items (e.g., "Is football or basketball more
covering that they are not as different from others as they had interesting to watch?"), and three factual items (e.g., "When will
previously thought. To my knowledge, nowhere is such emotion the LInited States become independent of foreign sources of en-
suggested by advocates of social comparison theory." ergy?"). Otherwise, the procedures were identical to those used
In one of the most thorough and innovative studies of group in Experiment 1. Again, there were apparently no effects of
polarization, Vinokur and Burnstein (1978a) argued that in most subgroup salience (again, no effect directions or sizes were re-
cases, PAT and SCT make similar predictions about the direction ported), both polarization (on six of the seven items) and de-
of polarization, except for the case when two subgroups with polarization occurred, and depolarization was stronger (on six
divergent positions (one pro-J and the other pro-K) try to reach of the seven items). In both cases, the exception was the capital
a decision. In this situation, SCT holds that each member of the punishment item. With two exceptions both subgroups shifted
pro-K. subgroup will compare himself to the members of his own toward each other, accounting for the large depolarization effect,
subgroup and then become more pro-K after discussion, whereas r(2Q) = .84 (see Footnote 13: again, the capital punishment item
members of the pro-J subgroup will become more pro-J. The was one of the two exceptions). Finally, combining the findings
result will be polarization between subgroups. PAT argues that from both experiments, depolarization was greatest for factual
within each subgroup most of the arguments favoring a given
alternative will have been shared already and thus there will be 10
One wonders wh> a 2 x 2 analysis of variance was not used to test
relatively few novel (and thus persuasive) valid arguments within the predicted interaction of condition with balance of prorisk/procaution
subgroups. Across subgroups, however, new arguments will be arguments.
heard, thus facilitating a shift toward the other subgroup, and " There is a confound in the comparison between the experimental
depolarization will be observed. and the no-exposure control conditions, namely that subjects always re-
In the first of two experiments, subjects first completed seven sponded to the no-exposure control condition first when participating in
CDQ items: four risky, two cautious, and one neutral. On the the experiment. Then the remaining two conditions were counterbalanced.
basis of their responses, experimenters composed several groups It is conceivable that subjects needed to warm up to the experimental
of six members for each item (in other words, groups were formed situation, and this caused them to generate arguments of different kinds
in the first trial (i.e.. the no-exposure control condition).
for one item, then re-formed for the next item, and so on). The i:
An additional group of subjects completed the two CDQ items and
criterion for forming each group was that there were two seven items used in Experiment 2 in order to control for regression to
subgroups of 3 subjects, and for the particular item to be discussed the mean.
the average responses for the two subgroups differed by approx- 13
This effect size is based upon the average of seven is for the depo-
imately 5 scale points (out of 10). For half of the groups, a salience larization score on each item, as well as on the harmonic mean. A slightly
manipulation had the three subgroup members sit together op- different number of groups was used for each of the seven items.
 1148 DANIEL J. ISENBERG

items (3.92), second for the CDQ items (2.44), third for taste perceived as invalid, and thus novel arguments (and rational dis-
(1.90), and fourth for value (1.01). cussion) will be unpersuasive. Without siding with either SCT
These two experiments are quite interesting primarily because or PAT. it seems that the perception of validity is one important
they are surprisingly the first to study polarization between conceptual link between the two.
subgroups. This remains a serious shortcoming in the group po- A final exclusivity study by Laughlin and Earley (1982) re-
larization literature. The authors interpret their results as showing portedly found stronger support for PAT than for SCT based on
that PAT makes an accurate prediction (depolarization) whereas the observation that, across CDQ items, repeated trials, and con-
SCT makes an inaccurate one (polarization between subgroups). ditions (individual vs. group), decisions taking the perspective of
A more cautious interpretation would be that here is yet another a hypothetical stranger were riskier than those taking the per-
demonstration of postdiscussion convergence (see Myers. 1982) spective of a friend or oneself. Why is this counter to SCT?
occurring along with an average shift, where the members most Laughlin and Earley argued that the hypothesized desire to be
extreme in the direction of polarization shift less compared to better than the average in SCT should manifest itself primarily
those most extreme in the opposite direction (see Ferguson & when making decisions from one's own perspective and not when
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Vidmar. 1971). Whether or not these two experiments actually making decisions from the stranger perspective. Thus when risk
This document is copyrighted by the American Psychological Association or one of its allied publishers.

test SCT depends upon how much we believe that the salience is valued, one should rate oneself as riskier than the stranger,
manipulation in fact caused group members to compare them- and when caution is valued one should rate oneself as more
selves to the other 2 members of their own subgroup. I think cautious.
this can be reasonably questioned given (a) the instructions to Curiously, Laughlin and Earley used the perspective main effect
"reach consensus." (b) the fact that all subjects were students at to test this hypothesis rather than the item by perspective inter-
the same university, and (c) that the students may have adhered action. Whereas they concluded that SCT has a problem because
to norms of conflict avoidance. Given the choice between being across items the self perspective is more cautious than the stranger
less extreme and avoiding conflict, versus comparing themselves perspective, there is nothing in SCT that would predict a per-
to their own subgroup and becoming more extreme, I think that spective main effect at all. The more appropriate item perspective
most students would choose the former, independent of any per- by interaction yields an F less than 1, F(\, 564) = .07. This
suasive arguments. might still be considered a problem for SCT. but certainly a
It is to the credit of PAT that its specificity allows anomalies relatively minor one.14
to be informative, and the anomaly of the capital punishment
item is instructive because it is the only item showing both sig- Summary and Conclusions
nificant polarization and no depolarization. Data presented in
the article indicate that the polarization was against capital pun- Table 1 summarizes the effects from the more recent studies
ishment, due primarily to nonliberal subjects becoming more showing effects either of mere-exposure or of persuasive argu-
liberal. According to PAT, this could occur only if (a) the liberal ments. The average effect sizes from each type of study are sub-
subgroup did not hear any new arguments for capital punishment, stantial, but the effect of persuasive argumentation is particularly
and (b) the nonliberal subgroup did hear novel arguments against strong (.746 average r vs. .436)." In many of the studies reported
capital punishment. There is no evidence that this was the case: here, both social comparison and persuasive argumentation are
in fact, the authors themselves argue that for value items, novel occurring simultaneously, but there are studies that show effects
arguments are essentially exhausted. A simpler alternative ex- of social comparison in argument-poor tasks (e.g.. Baron & Roper,
planation to the data on the capital punishment item is that 1976: see also Myers, 1982: Vidmar, 1974). There are also studies
discussion engaged a liberal norm in a liberal setting, and that that show effects of argumentation in comparison-poor tasks (e.g..
attitudes polarized accordingly, a typical finding in polarization
research (Myers, 1975: Myers & Bach, 1974). 14
The Vinokur and Burnstein (1978a) studies do pose a critical One source of potential support for the exclusivity hvpothesis has
been the relative lack of effects of mere-exposure to other group members"
question for group research in general, namely, under what cir-
positions compared to exposure to persuasive arguments. For example.
cumstances should polarization between groups occur? Would
Kaplan (1977. Experiment 2) found that when confronted with two con-
it be possible for polarization between groups to occur if both tradictory sources of influence, proincrimination persuasive arguments
groups (for example, Palestinians and Israelis) was exposed to and proexoneration ratings of group members, the persuasive arguments
the same arguments and the argument pool is exhausted? Ev- influenced subjects' subsequent ratings, whereas mere-exposure did not.
eryday experiences in families, labor relations, and international Care should be taken, however, in interpreting this finding as evidence
politics suggest that such argument-poor polarization can occur for the exclusivitv hypothesis. An alternative interpretation would be that
(Sherif, 1966). PAT itself suggests a way to understand how this a written list of proincrimination arguments would be more vivid and
might happen, namely that the arguments generated and proc- salient (Borgida and Nisbett. 1977) than a set of numbers representing
essed are novel but they are rejected as invalid (Vinokur & Burn- members' proexoneration positions.
15
Some care must be exercised in interpreting this difference in average
stein, 1978b). However, this puts us back in the arena of nor-
effect size. Data reported in several of the PAT studies (e.g.. Vinokur &
mative mechanisms because the perception of validity is by def- Burnstein. 1974) allow the calculation of the correlation between predicted
inition value based. This point is reinforced by the finding that and observed scores across CDQ or similar items. Given the lack of an
novelty leads to persuasiveness only when perceived validity is appropriate within subjects error term, it is not possible to calculate an
high (Vinokur & Burnstein, 1978b). Given the ubiquity of in- r that is directly comparable with the other«. Thus, several of the reported
tergroup phenomena (e.g., Sherif, 1966) we would expect that rs probably are higher than they would be based on an appropriate error
when an outgroup is perceived as "bad," their arguments will be term with <# based on number of subjects, not number of items.
 GROUP POLARIZATION 1149

Burnstein et al., 1973; Ebbesen & Bowers, 1974), that is, tasks of future experiments as well as the interpretation of past ones.
that do not permit the inference of group comembers' positions. For example, Burnstein & Vinokur (1977) cited the lack of mere-
In the cases of the argument-poor studies the effect sizes are very exposure effects in several studies, but reanalyses suggest effect
similar to the average effect size for mere-exposure studies. Like- sizes that are respectable (.3-.4) but insignificant perhaps as a
wise, for comparison-poor studies, the effect sizes are similar to result of the small number of subjects in the experiment. For
the average effect size for persuasive arguments studies. At this example. Clark and Willems (1969) found that information ex-
point in time there is very good evidence that there are two change of positions led to no shift in one of their conditions.
conceptually independent processes even though outside of the t(24) = 1.87, whereas the associated r(.36) is very similar to
laboratory they almost always co-occur. those in Table I. Thus, the conclusion that this is a failure to
The analysis of effect sizes is important for both the planning replicate is not necessarily founded in this study (a Type 2 error
may have been committed).
On a related point, many investigators have made it difficult
Table 1 to perform meta-analyses and at the same time impeded tests of
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Effect Sizes for Recent Group Polarization Studies on specific hypotheses by using omnibus F-tests (with df(num)
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Mere-Exposure and Persuasive Argumentation greater than 1). These make the translation to r difficult and test
Stud> r df only the general (and conceptually meaningless) hypothesis that
there is some significant amount of variance associated with the
Mere-exposure studies" independent variables. Given the maturity of the field and ac-
Baron & Roper (1976) cumulated knowledge of the effects of various stimuli, very spe-
Experiment 1 .41 44 cific a priori hypotheses can and should be stated using planned
Experiment 2 .43 39 contrasts. A discussion of this technique is beyond the scope of
Bell &Jamieson( 1970) .12 23 this article (see Winer, 1971), but any number of degrees of free-
Bell & Jamiesonf 1970) .21 23
Blascovich. Ginsburg. & Veach ( 1975) .37 14 dom in the numerator other than 1 should be a warning flag for
Blascovich, Ginsburg. & Howe (1975) .74 7 researchers (Rosenthal & Rosnow, 1984).
Blascovich. Ginsburg. & Howe (1976) .39 17
Blascovich & Ginsburg (1974) .29 27
Blascovich & Ginsburg (1974) .57 27 Integrative Questions
Clark &Willems( 1969) .52 24
Clark &Willems( 1969) .36 24 Given the support for both PAT and SCT as mediating pro-
Goethals& Zanna(l979) .63 14 cesses, it behooves investigators to develop theories that account
Myers (1978) for the interaction between SCT and PAT and that address the
Experiment 1 .27 102a factors that moderate the emergence of one or the other form of
Experiment 1 .44 102
Experiment 2 .33 116 influence. The following are four questions that suggest how to
Myers, Bach, & Schreiber (1974) .52 18 integrate PAT and SCT into a more conceptually coherent po-
Myers. Bruggink, Kersting, & Schlosser (1980) sition, a position that also serves to integrate group polarization
Experiment 1 .86 18 with other social psychological phenomena.
Experiment 2 .43 27
Myers, \\ojcicki. & Aardema ( 1977) .17 Under what conditions will group processes be more affected
265
Teger& Pruitt(l967) .51 18 by either rational argumentation or social comparison? One
Wallach& Kogan(1965) -.03 II potential moderating variable is decision characteristics. To the
Wallach& Kogan(1965) .37 11 extent that a decision has many factual or logical components
'a.t .436 we would expect rationality to be more prominent than social
Persuasive arguments studies desirability. The items that Vinokur and Burnstein (1978a)
showed to depolarize the most in group discussion (i.e., be less
Burnstein & Vinokur (1975) .68 22 susceptible to social comparison processes) were those involving
Burnstein. Vinokur, & Trope (1973) .39 252
Ebbesen & Bowers (1974) matters of fact.
Experiment 1 .65 9 A different potential moderating variable is ego involvement.
Experiment 3 .98 3 In a decision where group members are highly ego-involved sev-
Kaplan (1977) eral parameters change: (a) values are engaged; (b) attention is
Experiment I .44 88
Experiment 2 .53 88
constricted to a narrow range of information input and issues;
Kaplan & Miller (1977) .67 140 and (c) argument pools tend to be exhausted because ego-in-
Madsen(l978) volving issues have already been heavily processed by individuals
Experiment 1 .51 13 prior to discussion. We would expect capital punishment, fem-
Experiment 2 .57 4 inism, pacifism, and drug usage to be ego-involving compared
Vinokur & Burnstein (1974)
Experiment I .86 3 to the CDQ scenarios and questions of whether basketball or
Experiment 2 .93 3 football is the more interesting spectator sport. Thus social com-
Experiment 3 .84 3 parison should operate more strongly in the former situations,
.746 and persuasive argumentation should operate more strongly in
" These two correlations are not independent, thus the average of these the latter.
two correlations (.408) was used in computing the average r for all of the H 'hat are the causal paths among social desirability, persuasive
mere-exposure studies. argumentation, and attitude polarization? Myers and Lamm
 1150 DANIEL J. ISENBERG

(1976) suggested that social desirability can influence argumen- theoretical streams in social psychology as well as with the prac-
tation through action commitment, but that persuasive argu- tical requirements of functioning in real-life groups.
ments do not affect social desirability (Myers & Lamm. 1976,
Figure 1). The distinction between rationality and rationalization References
is germane here and needs to be paid attention to in particular
Anderson. N. H., & Graesser. C. C. (1976). An informational integration
by the persuasive arguments theorists. The former implies ar-
analysis of attitude change in group discussion. Journal of Personality
gumentation in the absence of social desirability, whereas the and Social Psychology. 34, 210-222.
latter suggests that social desirability can motivate persuasive Baron, R. S.. & Roper. G. (J976). Reaffirmation of social comparison
argumentation. Action commitment helps, but it is not necessary views of choice shifts: Averaging and extremity effects in an autokinetic
for the internal generation of arguments to reduce dissonance situation. Journal of Personality and Social Psychology, 33, 521 -530.
with socially desirable positions. Bell. P. D.. & Jamieson, R. B. (1970). Publicity of initial decision and
In overt group interaction, normative processes can do more the risky-shift phenomenon. Journal of Experimental Social Psychology,
than motivate group members to think of supporting arguments; 6. 329-345.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

they can also cause members to skew their spoken arguments to Bishop. G. D.. & Myers, D. G. (1974). Informational influences in group
This document is copyrighted by the American Psychological Association or one of its allied publishers.

discussion. Organizational Behavior and Human Performance. 12, 92-

favor one alternative over another. As Myers and Lamm (1976)
104.
argued, self-censoring probably occurs in group decision making.
Blascovich, J.. & Ginsburg, G. P. (1974). Emergent norms and choice
If so, we might hypothesize that the greater the pluralistic ig- shifts involving risk. Sociometry. 37, 205-218.
norance in a group prior to discussion, the greater the amount Blascovich, J.. & Ginsburg. G. P. (1978). Conceptual analysis of risk-
of self-censoring that will be observed as members attempt to taking in "risky-shift" research. Journal for the Theory of Social Be-
conform to the misperceived norm. As in release theory, when haviour. 8. 217-230.
the true norms are discovered, self-censoring will still occur, but Blascovich. J., Ginsburg, G. P.: & Howe, R. C. (1975). Blackjack and the
in the opposite direction as members strive to be better than the risky shift. II: Monetary stakes. Journal of Experimental Social Psy-
average. chology, I I . 224-232.
Hmv is information about group members' positions combined Blascovich, J.. Ginsburg, G. P.. & Howe. R. C. (1976). Blackjack, choice
shifts in the field. Sociometry, 39. 274-276.
into a mental concept of the "group norm"? What are the pro-
Blascovich, J., Ginsburg. G. P.. & Veach. T. L. (1975). A pluralistic ex-
cesses by which group members attend to, encode, and store planation of choice shifts on the risk dimension. Journal of Personality
information about these descriptive and prescriptive norms and Social Psychology, 31, 422-429.
within their group? This issue reverses the causal sequence in Borgida, E., & Nisbett. R. E. (1977). The differential impact of abstract
the second question above (normative —• cognitive) and suggests vs. concrete information on decisions. Journal of Applied Social Psv-
that informational processes can influence normative ones. Thus, chology, 7. 258-271.
certain characteristics of information such as vividness (Taylor Brown. R. (1965). Social psychology. New York: Free Press.
& Fiske, 1978) can influence how group members perceive their Brown. R. (1974). Further comment on the risky shift. American Psy-
group's norms. For example, it is possible that information about chologist, 29. 468-470.
extreme or deviant group members is cognitive!}1 overavailable Burnstein. E. (1982). Persuasion as argument processing. In H. Brand-
statter. J. H. Davis. & G. Stocher-Kreichgauer (Eds.), Contemporary
and will have an inordinate influence in determining perceptions
problems in group decision-making (pp. 103-124). New York: Academic
of norms. An obvious area of research is the relation between
Press.
social cognition and group polarization. Burnstein (1982) has Burnstein, E.. & Vinokur. A. (1973). Testing two classes of theories about
already begun to establish links between polarization and infor- group-induced shifts in individual choice. Journal of Experimental
mation processing. Social Psychology. 9. 123-137.
Hmv can group polarization phenomena be related to attitude Burnstein. E., & Vinokur. A. (1975). What a person thinks upon learning
change processes in general (e.g.. Sherif& Holland, 1961) and he has chosen differently from others: Nice evidence for the persuasive
attitude polarization in particular? One distinction typically arguments explanation of choice shifts. Journal of Experimental Social
made in the attitude change literature is between source and Psychology, 11,412-426.
Burnstein, E.. & Vinokur, A. (1977). Persuasive argumentation and social
message characteristics (McGuire, 1969). The debate between
comparison as determinants of attitude polarization. Journal of Ex-
PAT and SCT can be recast in these terms. For example, social
perimental Social Psychology. 13. 315-332.
comparison theorists indirectly focus on who or what is the source Burnstein. E., Vinokur. A.. & Trope. Y. (1973). Interpersonal comparison
of the norms that form the basis for social comparison. If the versus persuasive argumentation: A more direct test of alternative ex-
source is a reference group of similar others, then a perceived planations for group-induced shifts in individual choice. Journal of
discrepancy between own and others' positions should generate Experimental Social Psychology. 9, 236-245.
group polarization. Persuasive argument theorists seem to dis- Cartwright, D. (1973). Determinants of scientific progress: The case of
regard the source and differentiate among message characteristics, research on the risky shift. American Psychologist, 28, 222-231.
such as novelty and validity. When seen in this light, one reason Clark. R. D. (1971). Group-induced shift towards risk: A critical appraisal.
the two positions do not see eye to eye is that they are dealing Psychological Bulletin, 76, 251-270.
Clark. R. D., & VVillems, E. P. (1969). Where is the risky shift? Dependence
with different aspects of a more generalized process of attitude
on instructions. Journal of Personality and Social Psychology, 13. 215-
change.
221.
Quite unwittingly. James Stoner opened up an important sub- Ebbesen, E. B.. & Bowers, R. J. (1974). Proportion of risky to conservative
field of research in small group behavior that has resulted in arguments in a group discussion and choice shifts. Journal of Personality
several robust conclusions. It is hoped this research will continue and Social Psychology, 29, 316-327.
in such a way that it becomes integrated with other important Ferguson. D. A.. & Vidmar. N. (1971). Effects of group discussion on
 GROUP POLARIZATION 1151

estimates of culturally appropriate risk levels. Journal of Personality porary problems in group decision-making (pp. 125-161). New York:
and Social Psychology. 20, 436-445. Academic Press.
Festinger. L. (1954). A theory of social comparison processes. Human Myers, D. G., & Bach. P. J. (1974). Discussion effects on militarism-
Relations, 7. 117-140. pacifism: A test of the group polarization hypothesis. Journal of Per-
Fromkin. H. (1970). Effects of experimentally aroused feelings of undis- sonality and Social Psychology, 30. 741-747.
tinctiveness upon valuation of scarce and novel experiences. Journal Myers. D. G.. Bach. P. J.. & Schreiber, F. G. (1974). Normative and
of Personality and Social Psychology. 16. 521-529. informational effects of group interaction. Sociometry. 37, 275-286.
Goethals. G. R.. & Darley. J. (W). Social comparison theory. An at- Mvers. D. G.. Bruggink. J. B., Kersting, R. C.. & Schlosser, B. A. (1980).
tributional approach. In J. Suls & R. Miller (Eds.), Social comparison Does learning others' opinions change one's opinion? Personality and
processes: Theoretical and empirical perspectives (pp. 259-278). Social Psychology Bulletin. 6. 253-260.
Washington, DC: Hemisphere Press. Myers. D. G., & Lamm. H. (1975). The polarizing effect of group dis-
Goethals. G. R.. & Zanna. M. P. (1979). The role of social comparison cussion. American Scientist, 63. 297-303.
in choice shirts. Journal of Personality and Social Psychology. 3~. 1469- Myers, D. G.. & Lamm, H. (1976). The group polarization phenomenon.
1476. Psychological Bulletin. 83. 602-627.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Hong. L. K. (1978). Risky shift and cautious shift: Some direct evidence Myers. D. G., Wojcicki. S. B.. & Aardema, G. C. (1977). Attitude com-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

on the cultural-value theory. Social Psychology. 41. 342-346. parison: Is there ever a bandwagon effect? Journal of Applied Social
Isenberg, D. J. (1980). Levels of analysis of pluralistic ignorance phe- Psychology. 1. 341-347.
nomena: The case of receptiveness to interpersonal feedback. Journal Pruitl, D. G. (197 la). Choice shifts in group discussion: An introductory
of Applied Social Psychology, 10. 457-467. review. Journal of Personality and Social Psychology. 20, 339-360.
Isenberg. D. J.. & Ennis. J. G. (1981). Perceiving group members: A Pruitt, D. G. (197lb). Conclusions: Toward an understanding of choice
comparison of derived and imposed dimensions. Journal of Personality shifts in group discussion. Journal of Personality and Social Psychology.
and Social Psychology; 41. 293-305. 20.495-510.
Jellison. J.. & Arkin, R. (1977). Social comparison of abilities: A self- Rosemhal. R. (1978). Combining results of independent studies. Psycho-
presentation approach to decision-making in groups. In J. Suls & R. logical Bulletin. 85, 185-193.
Miller (Eds.). Social comparison processes.- Theoretical and empirical Rosenthal. R., & Rosnow, R. (1984). Essentials of behavioral research.
perspectives (pp. 235-258). Washington. DC: Hemisphere Press. New York: McGraw-Hill.
Jellison. J.. & Riskind. J. (1970). A social comparison of abilities inter- Sanders. G. S., & Baron, R. S. (1977). Is social comparison irrelevant for
pretation of risk-taking behavior. Journal of Personality and Social producing choice shifts? Journal of Experimental Social Psychology.
Psychology. 15. 375-390. 13. 303-314.
Jones, E. E.. & Nisbett. R. E. (1972). The actor and the observer: Divergent Schroeder, H. E. (1973). The risky shift as a general choice shift. Journal
perceptions of the causes of behavior. In E. E. Jones, E. E. Kanouse. of Personality and Social Psychology. 27. 297-300.
H. H. Kelley. R. E. Nisbett, S. Valins. & B. \Veiner(Eds.). Attribution. Sherif. M. (1966). In common predicament. Boston: Houghton Mifflin.
Perceiving the causes of behavior. Morristown. NJ: General Learning Sherif. M.. & Hovland. C. (1961). Social judgment. New Haven. CT: Yale
Press. University Press.
Kaplan. M. F. (1977). Discussion polarization effects in a modified jury Stoner, J. A. F. (1961). A comparison of individual and group decisions
decision paradigm: Informational influences. Sociomeir): 40, 262-271. involving risk. Unpublished master's thesis. Massachusetts Institute of
Kaplan. M. F. & Miller. C. E. (1977). Judgments and group discussion: Technology. Cambridge, MA.
Effect of presentation and memory factors on polarization. Sociomeln: Stoner, J. A. F. (1968). Risky and cautious shifts in group decisions: The
40. 337-343. influence of widely held values. Journal of Experimental Social Psy-
Kleinhans. B.. & Taylor. D. A. (1976). Group processes, productivity, chology. 4. 442-459.
and leadership. In B. Seidenberg & A. M. Snadowsky (Eds.). Social Taylor, S. E.. & Fiske, S. (1978). Salience, attention, and attribution: Top
psychology (pp. 407-434). New York: Free Press. of the head phenomena. In L. Berkowitz (Ed.), Advances in experimental
Lamm. H.. & Myers, D. G. (1978). Group-induced polarization of at- social psychology (\'o\. 11, pp. 250-288). New York: Academic Press.
titudes and behavior. In L. Berkowitz (Ed.). Advances in experimental Teger. A. I., & Pruitt, D. G. (1967). Components of group risk taking.
social psychology (Vo\. 2, pp. 147-195). New York: Academic Press. Journal of Experimental Social Psychology. 3. 189-205.
Laughlin, P. R.. & Earley. P. C. (1982). Social combination models, per- Vidmar, N. (1974). Effects of group discussion on category width judg-
suasive arguments theory, social comparison theory, and choice shifts. ment. Journal of Personality and Social Psychology. 29, 187-195.
Journal of Personality and Social Psychology. 42. 273-280. Vinokur. A.. & Bumstein. E. (1974). Effects of partially-shared persuasive
Levinger. G., & Schneider. D. J. (1969). Test of the "risk is a value" arguments on group-induced shifts: A group-problem-solving approach.
hypothesis. Journal of Personality and Social Psychology. I I . 165-169. Journal of Personality and Social Psychology. 29. 305-315.
Madsen. D. B. (1978). Issue importance and choice shifts: A persuasive Vinokur, A.. & Burnstein. E. (I978a). Depolarization of attitudes in
arguments approach. Journal of Personality and Social Psychology. 36. groups. Journal of Personality and Social Psychology, 36. 872-885.
1118-1127. Vinokur. A.. & Burnstein. E. (I978b). Novel argumentation and attitude
McGuire, W. J. (1969). The nature of attitudes and attitude change. In change: The case of polarization following group discussion. European
G. Lindzey & E. Aronson (Eds.). The handbook of social psychology Journal of Social Psychology, 8, 335-348.
(2nd ed.. pp. 136-314). Reading. MA: Addison-Wesley. Wallach. M. A.. & Kogan. N. (1965). The roles of information, discussion,
Moscovici. S., & Zavalloni. M. (1969). The group as a polarizer of attitudes. and consensus in group risk-taking. Journal of Experimental Social
Journal of Personality and Social Psychology. 12. I25-135. Psychology, I. 1-19.
Myers, D. G. (1975). Discussion-induced attitude polarization. Human Winer, B. J. 11971). Statistical principles in experimental design. New
Relations, 28, 699-714. York: McGraw-Hill.
Myers. D. G. (1978). Polarizing effects of social comparison. Journal of
Experimental Social Psychology, 14. 554-563.
Myers. D. G. (1982). Polarizating effects of social interaction. In H. Received July 3, 1984
Brandstatter, J. H. Davis. & G. Stocher-Kreichgauer (Eds.). Contem- Revision received March 22, 1985