Direct and indirect effects of group discussion on consensus

Abstract

Participants within groups of three were pre‐tested on a set of issues, discussed a subset of the issues, and were then post‐tested on all the issues. Discussion led to significant group consensus on issues that were discussed by participants. Consensus was also observed for issues that were not discussed by the group, but were related to the discussed issues. Groups did not move toward consensus on non‐discussed issues that were unrelated to the discussed issues. We discuss implications of our results for theories of cognitive consistency, and for the formation of culture within large and small groups.

Imagine having a discussion with colleagues about the newest medical developments for Parkinson's disease which involve the use of stem cells. Such an interaction may change or strengthen your pre‐existing attitude toward stem cell research, but might you also then think further about related issues such as the ethics or legality surrounding such research, or the related issues of cloning or abortion? Might there be conditions under which we would expect such change? As group‐level social influence takes place, and individuals are persuaded by others' arguments regarding the topics at hand, might this same influence affect individuals' attitudes on related topics? The goal of this paper is to explore the idea that group discussions may not only have direct effects on one's attitudes, opinions, and beliefs, but may also affect non‐discussed attitudes, opinions, and beliefs that are related to discussed issues.

A ubiquitous phenomenon in social psychology is that people within groups influence each other, and groups therefore often move toward consensus. Due to such motivations as achieving social reality and group locomotion (Festinger, 1950), and through the processes of normative and informational influence (Deutsch & Gerard, 1955), diversity in groups tends to become reduced over time (Latané & Bourgeois, 2000). Since Sherif's (1936) classic work showing that consensus emerges among groups of people rating the apparent movement of light in studies of the autokinetic effect, research has consistently shown that groups move toward consensus after communicating on a variety of issues, including perceptual judgments (McCord, 1948), aesthetic preferences (Mausner, 1953), and social and political opinions (Bourgeois, 2002; Huguet, Latané, & Bourgeois, 1998).

This reduction in diversity, termed consolidation, is a key prediction of Dynamic Social Impact Theory (DSIT, Latané, 1996), and the evidence above suggests that when people within a group discuss issues over time, diversity within the group is often reduced. DSIT suggests that culture may become self‐organized from the bottom up, such that individuals influence each other and each other's preferences, leading to clusters of people who share similar preferences and show correlated opinions for initially unrelated issues. In this way, culture gradually develops from repeated social influence at the local level. As we interact with proximal and similar others on a daily basis, we become more similar to them in our views on various topics. In other words, individual‐level social influence determines group‐level phenomena; the more people who respond to this individual‐level social influence, the more consolidation and clustering that will occur. This can occur at various levels of resolution (e.g., in neighborhoods, towns, cities, states, and countries), and global outcomes of clustering and consolidation can result after years of this inevitable individual‐level influence. Data on consumer attitudes by Weiss (2000) even show how consumer preferences (e.g., spam and salsa consumption) can be reliably predicted based on one's region of residence within the U.S.

Most of the evidence for the predictions of DSIT thus far, however, has been provided by having groups discuss a specific set of issues. Similarly, traditional methodologies in the group consensus literature typically first entail a pre‐test to assess individuals' attitudes, then a group interaction, followed by a post‐test of the same issues. Consensus (or consolidation) is then operationalized as a reduction in diversity within the group from pre‐test to post‐test (e.g., Latané & Bourgeois, 2000). In such paradigms, it is not typically expected that non‐discussed items will change following discussion; in fact, they are often included as a control condition to which consensus on the discussed issues is compared (e.g., Huguet et al., 1998).

However, McGuire (1960, 1990) has posited that our attitudes and beliefs are not held in a vacuum, but rather exist within the larger context of interrelated thought systems. Thus, when we change attitudes or beliefs on some issues as a result of discussing them with others, our preference for consistency (e.g., Festinger, 1957; Heider, 1958) may lead us to also change some of our other existing beliefs in order to maintain consistency in our thought systems. Surprisingly, there has been very little research testing the effects of such indirect social influence on cognitively proximal beliefs, and even less within the context of group discussions.

McGuire (1960) showed that when persuasive communications changed the beliefs of high‐school seniors and college freshmen, the communications also had indirect effects on logically related but unmentioned beliefs. This effect was also demonstrated with college students by Dillehay, Insko, and Smith (1966), and Hastie and Rawson (2004). For example, Hastie and Rawson showed that direct attempts to change the belief “An increasing proportion of students will attend universities in their own states, rather than out‐of‐state universities” also had indirect effects on the logically related antecedent belief “Out‐of‐state tuition fees will go much higher at state universities” as well as the logically related consequent belief “Opportunities to broaden one's range of friends in college will be reduced.” Hastie and Rawson suggested that these related beliefs changed in order to maintain consistency within participants' thought systems.

Crano and colleagues (Alvaro & Crano, 1997; Crano & Alvaro, 1998; Crano & Chen, 1998) have shown that the process of minority influence within groups may lead to indirect attitude change on non‐discussed issues even when people do not change their opinions on focal issues. Tests of their leniency contract model of minority influence have shown that when a minority argues in favor of one issue (e.g., gays in the military), participants often changed their attitudes on related issues (e.g., gun control), but not on unrelated issues. Crano and Chen (1998) showed that when strong persuasive messages are attributed to minorities, attitude change on focal issues was delayed, but group members did change their opinions on related issues immediately. Such indirect attitude change is not necessarily due to logical relationships among issues, but rather ideological ones (Crano & Alvaro, 1998).

Although not directly related to changes in belief structures, several other studies have shown that as groups interact on specific decision‐making tasks, the group's influence may extend to other related tasks as well. For example, Stasson and Hawkes (1995) showed that groups working collectively on a judgment not only reached consensus on the task at hand, but also exhibited convergence among judgment strategies (i.e., which relevant evidence was considered to be most important) on other related tasks that were made individually following the group discussion. Participants within three‐person groups were presented with four cues about each of several students in an introductory psychology class, and asked to predict each student's score on a final exam. The four cues were scores on English and Math achievement tests, high‐school rank, and gender. There was a large group influence on participants' judgment strategies; in other words, groups significantly tended toward consensus in judgment strategies. For example, if a given group decided that achievement tests were the most important evidence to use on the group task, individuals from that group also weighted this evidence more heavily on subsequent individual judgments. Therefore, consensus in judgment also carried over to individual judgments made after the group discussion, suggesting that the group interaction had indirect effects on a related but separate judgment.

Conway (2004) found that as participants within 4–6‐person groups interacted to complete word puzzles, groups came to consensus on mood measures and estimates of time perception. This consensus emerged despite the fact that group members discussed neither their moods nor their perceptions of time. Conway suggested that the consensus on time perception was mediated by mood; that is, working in groups affected the mood of group members, which in turn led to consensus on the perception of how quickly time passed. This study provides another example showing how group consensus can emerge not only on a focal task, but also on related judgments.

The current study investigated the extent to which group‐level social influence on discussed issues may also affect attitudes toward other topics of varying relatedness. Multi‐dimensional scaling allowed us to determine the relatedness among a set of issues, some of which were discussed within groups and some of which were not discussed. The research question guiding the current study was: will groups come to consensus, not only on issues being discussed by the group members, but also on related issues that were not discussed?

METHOD

Participants

A total of 61 groups of three college undergraduates (N = 183) participated in the study in exchange for extra credit or to partially fulfill a course requirement. The sample was relatively young (M age = 20.89, SD = 4.08), mostly female (127 F, 54 M, and 2 did not specify gender), and mostly Caucasian (92%).

Materials and procedure

Participants arrived at the experiment in groups of up to 10 and were randomly assigned to smaller groups of 3. When the number of people who showed up was not a multiple of three, extra participants were run through an alternative study. After reading and signing an informed consent form, all participants individually and privately completed a packet including a demographic questionnaire, the NEO‐FFI personality inventory (as a filler task, not included in analyses), and a 30‐item attitude pretest. The items in the attitude pretest (see table 1) were generated for this study, and assessed a variety of attitudes and beliefs ranging from campus topics to more general social and political issues. Participants rated their attitude or belief regarding each statement on a 6‐point scale from −3 (strongly disagree) to +3 (strongly agree), with no 0 midpoint.

Table 1. Results for Issue Set 1

Issues	F	p	ICC
Discussed issues
a. People on welfare are lazy and should just get a job like the rest of us.	1.57	.069	.36
b. Women never invite sexual harassment.	2.63	.001	.62
c. Dogs are much better pets than cats.	1.55	.073	.36
d. The national health problem of HIV should be discussed in schools starting in kindergarten.	1.96	.013	.49
e. There should be mandatory marriage counseling before divorce is granted.	2.04	.009	.51
f. Women in combat should have to take birth control pills.	2.59	.001	.61
Related items
g. It is a good idea to automatically charge students a fee each year for an activities pass so they can get into sporting events.	1.31	.190	.24
h. The bars in Laramie do not enforce underage drinking laws.	1.45	.111	.31
i. Research on the cloning of humans is a waste of time.	0.86	.665	−.16
j. Television/film companies should prohibit cartoon violence.	1.22	.257	.18
k. The death penalty should be abolished in Wyoming.	1.06	.419	.05
l. The government is withholding information from the public on extra‐terrestrial life forms.	1.86	.022	.46
m. Affirmative action should be abolished.	1.70	.043	.41
n. UW should close a full week before Thanksgiving break.	1.69	.044	.41
o. Professors should not be allowed to give students any homework or tests the week before finals.	0.75	.807	−.34
p. UW students should get in free to athletic events.	1.28	.210	.22
q. Pornography on the Internet needs to be completely banned.	1.44	.119	.30
r. Skateboarding should be banned from university grounds.	1.09	.382	.08
Unrelated items
s. Men and women should share the expense of birth control	1.29	.202	.22
t. People do not choose to be homosexual.	1.25	.227	.20
u. All campus jobs should be open to all students regardless of financial need	1.33	.171	.25
v. Dining centers should offer more choices in hot meals for vegetarians	1.15	.319	.13
w. UW needs more racial diversity on its campus.	0.88	.645	−.14
x. There should be more computer labs on campus.	1.23	.245	.19
y. Scientists really have a cure for AIDS and are not telling us.	0.88	.642	−.14
z. Cloning another human being is completely unethical	0.76	.793	−.32
aa. Abortion ends the life of a human being	1.12	.349	.10
bb. The government should crack down on illegal immigrants in Wyoming.	1.50	.092	.33
cc. Money is the answer to all problems.	0.73	.831	−.38
dd. Anti‐Terrorism efforts should be the top priority in the U.S.	0.83	.704	−.20

Prior to this study, the items were pre‐tested on a sample of undergraduates drawn from the same population (N = 105) in order to determine which issues would be likely to generate substantive discussions. In order to choose items that showed adequate opinion diversity at the group level, we selected issues that had substantial variability of opinion at the group level (variance>2). Of the 30 items, we chose 12 for discussion.

We performed two replications within the study, such that 31 of the three‐person groups discussed six of these items and 30 groups discussed a different set of six. table 1 shows the six items discussed by the groups in Replication 1, and table 2 shows the six items discussed by those in Replication 2.

Table 2. Results for Issue Set 2

Issues	F	p	ICC
Discussed issues
g. It is a good idea to automatically charge students a fee each year for an activities pass so they can get into sporting events.	1.35	.165	.26
l. The government is withholding information from the public on extra‐terrestrial life forms.	2.54	.001	.61
x. There should be more computer labs on campus.	6.04	.000	.83
o. Professors should not be allowed to give students any homework or tests the week before finals.	1.69	.044	.41
z. Cloning another human being is completely unethical	1.71	.041	.41
cc. Money is the answer to all problems.	1.32	.182	.24
Related items
h. The bars in Laramie do not enforce underage drinking laws	1.07	.402	.07
a. People on welfare are lazy and should just get a job like the rest of us.	0.86	.660	−.16
i. Research on the cloning of humans is a waste of time	1.26	.225	.21
j. Television/film companies should prohibit cartoon violence.	0.69	.860	−.44
k. The death penalty should be abolished in Wyoming.	1.26	.223	.21
b. Women never invite sexual harassment.	0.99	.495	0
d. The national health problem of HIV should be discussed in schools starting in kindergarten.	0.97	.512	−.02
m. Affirmative action should be abolished.	1.32	.186	.24
bb. The government should crack down on illegal immigrants in Wyoming.	1.17	.297	.15
v. Dining centers should offer more choices in hot meals for vegetarians	2.24	.005	.55
w. UW needs more racial diversity on its campus.	1.06	.420	.05
n. UW should close a full week before Thanksgiving break.	1.11	.360	.10
c. Dogs are much better pets than cats.	1.16	.312	.14
p. UW students should get in free to athletic events.	1.84	.026	.46
y. Scientists really have a cure for AIDS and are not telling us.	1.53	.087	.35
aa. Abortion ends the life of a human being	1.73	.040	.42
Unrelated items
s. Men and women should share the expense of birth control	1.19	.282	.16
t. People do not choose to be homosexual.	0.69	.863	−.45
u. All campus jobs should be open to all students regardless of financial need	1.01	.469	.01
r. Skateboarding should be banned from university grounds.	1.27	.213	.21
q. Pornography on the Internet needs to be completely banned.	1.20	.275	.16
e. There should be mandatory marriage counseling before divorce is granted.	1.26	.222	.21
f. Women in combat should have to take birth control pills.	1.28	.208	.22
dd. Anti‐Terrorism efforts should be the top priority in the U.S.	0.93	.577	−.08

Following completion of the questionnaire packet, note‐cards specifying a room number were randomly distributed to the participants so that each group of three participants was assigned to one of the experiment rooms. During each experimental session, there were at least as many experimenters as there were three‐person discussion groups. To begin the discussion phase of the study, each three‐person group was given the following instructions:

In just a minute, you'll be discussing various topics with the other members of your group. In total you will be discussing six topics for three minutes each. Prior to each topic change I will ask you to stop discussing the one you're on and to change to a different topic. At that time I'll walk back into the room and give you the next topic. Please be sure to stay on topic for the entire three minutes. Feel free to speak openly about the issues with your other group members, we will be recording your conversations on tape but will not associate your names with your comments. I am going to read you the first topic, and then you may begin.

The experimenter kept time with a stopwatch and entered the room every 3 minutes to shift the group to a new discussion topic by reading it aloud to the group. The six topics were presented in a random order to each group. After discussing all six topics, participants were then given another copy of the 30‐item attitude and belief questionnaire, and asked to indicate their attitudes on the same scales that were used on the pre‐test. All participants were then thanked and debriefed.

RESULTS

Assessing relatedness among items

Before testing for direct and indirect effects of group discussion, we first determined which non‐discussed items on the pre‐test were and were not related to the discussed items using multidimensional scaling. Following a technique similar to that utilized by Alvaro and Crano (1997) and Breckler, Pratkanis, and McCann (1991), we considered proximity in multidimensional space to be an indicator of which non‐discussed items were related to the discussed items. figure 1 shows the spatial distribution (in two‐dimensional space) of the 30 items (labeled by the letters from table 1) from the pre‐test. We adopted the default two‐dimensional space because it seemed more readily interpretable than higher‐order spaces. The horizontal dimension seems to reflect a construct similar to the degree of acceptable control over the decisions of individuals, whereas the vertical dimension appears to reflect a sexually liberal versus conservative dimension. Euclidian distances were calculated from the correlation matrix of the 30 attitude items. Tables 1 and 2 provide a thorough summary of the results from the analyses for both replications.

Figure 1 Discussed, related, and unrelated issues in multidimensional space.

Whether or not any given non‐discussed item was proximal to a discussed issue was determined independently by the two authors by visually inspecting figure 1, and we agreed on 45 of the 48 issues across the two replications (94%). Disagreements were resolved by discussion. Following Alvaro and Crano (1997) and Breckler et al. (1991), items that were close together in space were considered to be cognitively proximal. For example, half the participants discussed item z (Cloning another human being is completely unethical); for these participants, item aa (Abortion ends the life of a human being) was considered to be a related item. This method was used separately within each of the two replications to determine which non‐discussed items were related or unrelated to discussed items, and the specific related and unrelated items will be described separately within each of the replications. table 1 shows which items were related and unrelated to the six discussed items from participants in Replication 1, and table 2 shows the same information from participants in Replication 2.

Replication 1

Using multidimensional scaling as described above, we determined, by observing proximities, which of the 24 non‐discussed items were related to each other (see table 3). Overall, 12 items were determined to be related to the 6 discussed items, and 12 of the items were determined to be unrelated to the discussed items.

Table 3. Discussed, related, and unrelated items in Replication 1

Discussed item	Related items
A	G, H, I, J
B	L, M, D, K
C	N, O, P
D	B, L, M, K
E	Q
F	R
Unrelated itemsS, T, U, V, W, X, Y, Z, AA, BB, CC, DD

Test of consensus

We tested whether or not groups were coming to consensus on discussed and non‐discussed items by conducting MANOVAs with group as a predictor of participants' attitudes. Significant Fs were interpreted as evidence for group consensus; specifically, that there was more within‐ than between‐group similarity in attitudes than would be expected by chance (see Conway, 2004).

An alternative, more conservative approach would have been to average across each of the discussed, related, and unrelated items, and compare the magnitude of the group as a predictor of attitudes among each in a repeated‐measures analysis. We decided to conduct the MANOVAs in order to increase power to detect effects; for example, because people discussed six issues, using a MANOVA and follow‐up univariate tests allowed us more power to detect significant effects than if we had averaged across items. Also, averaging across items would not have allowed us to test directly the predicted consensus effects. Because we were predicting consensus on specific issues, it does not make logical sense to average across issues before testing for group consensus.

Intraclass correlations (ICCs) were computed by the following formula suggested by Conway (2004): (MSB–MSW)/MSB (MSB = mean squared deviation between groups, MSW = mean squared deviation within groups). Therefore, as groups moved toward consensus on any given issue, the mean square between groups became larger relative to the mean square within groups, which led to larger ICCs. These ICCs can be interpreted as effect sizes showing the strength of group consensus, with higher positive scores reflecting more consensus. Negative ICCs suggest more within‐ than between‐group variability than would be expected by chance, and ICCs around 0 reflect a lack of consensus.

Before testing for group consensus on discussed, related, and unrelated items, we first examined whether discussion groups predicted participants' attitudes on the pretest. Because groups had not yet interacted, we did not expect any significant group effects; in other words, this could be considered a test of random assignment. We entered all 30 items on the pretest into a MANOVA with group as a predictor of attitudes, and found no significant effects, MANOVA F(870, 1537) = 1, p = .46. Regarding univariate effects, the group was not a significant predictor of any of the 30 items on the pretest, and the average ICC was .00.

Discussed items

Overall, groups tended to come to consensus on the six discussed items, MANOVA F(180, 366) = 1.77, p<.001. Regarding the univariate tests of group as a predictor of each of the six discussed attitudes, four of the effects were significant (p<.05), all six ICCs were positive, and the average effect size (ICC) was .49.

Related non‐discussed items

Groups also tended to come to consensus on the 12 non‐discussed items that were related to the discussed items, MANOVA F(360, 684) = 1.20, p = .02. The group was a significant predictor of 4 of the 12 related items, 10 of the 12 ICCs were positive, and the average effect size (ICC) was .18.

Unrelated non‐discussed items

In contrast, groups did not reach significant consensus on the 12 non‐discussed items that were not related to the six discussed items, MANOVA F(360, 708) = 1.02, p = .43. The group was not a significant predictor of any of the unrelated items, 7 of the 12 ICCs were positive, and the average effect size (ICC) was .02.

Replication 2

Groups in Replication 2 discussed the six items listed in table 2. Again, by using multidimensional scaling, we used proximities as determined by multidimensional scaling to categorize the 24 non‐discussed items into related and unrelated items. Overall, 16 items were determined to be related to the 6 discussed items, and 8 of the items were determined to be unrelated to the discussed items. See table 4 for a full depiction of the relatedness of specific issues to one another for replication 2.

Table 4. Discussed, related, and unrelated items in Replication 2

Discussed item	Related items
G	H, A, I, J
L	B, M, D, K
X	V, W, BB
O	P, C, N
Z	AA
CC	Y
Unrelated itemsS, T, U, R, Q, E, F, DD

Test of consensus

Again, we tested whether or not groups were reaching consensus (i.e., clustering) on discussed items by conducting MANOVAs, with the group as a predictor of post‐discussion attitudes. Significant Fs were interpreted as evidence for group consensus. Effect sizes (ICCs) were calculated to indicate the magnitude of group consensus. Among these groups, we again first determined whether or not the group was a significant predictor of participants' attitudes on the pretest (i.e., before discussion) by conducting a MANOVA, with group as a predictor of attitudes. Because the groups had not yet interacted, we did not expect any significant effects. There was no evidence that the group significantly predicted attitudes before discussion across the 30 items, MANOVA F(900, 1590) = 1.03, p = .31. In examining the univariate statistics, the group significantly predicted attitudes on one of the 30 items, and the average effect size (ICC) was .05.

Discussed items

Overall, groups tended to come to consensus on the six discussed items, MANOVA F(174, 354) = 2.00, p<.001. Regarding the univariate tests of group as a predictor of each of the six discussed attitudes, four of the effects were significant (p<.05), all six ICCs were positive, and the average effect size (ICC) was .46.

Related non‐discussed items

Groups also tended to come to consensus on the 16 non‐discussed items that were related to the discussed items, MANOVA F(464, 896) = 1.17, p = .03. The group was a significant predictor of 4 of the 16 related items, 12 of the 16 ICCs were positive, and the average effect size (ICC) was .15.

Unrelated non‐discussed items

In contrast, groups did not reach significant consensus on the eight non‐discussed items that were not related to the six discussed items, MANOVA F(232, 480) = 1.10, p = .19. The group was not a significant predictor of any of the unrelated items, six of the eight ICCs were positive, and the average effect size (ICC) was .05.

DISCUSSION

In addition to showing that discussion can lead to group consensus, this is the first study to demonstrate the effects of group discussion on non‐discussed items. Not only did group discussion direct groups toward consensus on discussed items, but this discussion also indirectly affected participants' attitudes on related, yet non‐discussed issues. We found no evidence of an effect on unrelated non‐discussed issues. Perhaps social influence in general is more far‐reaching than once thought. This study provides preliminary evidence that influence attempts may not only have their intended direct effects, but also can have indirect effects, which may or may not be intended.

Another interesting result was that we found indirect effects on items that were shown to be cognitively proximal, yet without any clear logical connection. For example, it seems quite logical that after groups discuss the issue “Professors should not be allowed to give students any homework or tests the week before finals” they reach consensus not only on this focal topic but also on the issue “UW students should get in free to athletic events.” These issues seem to be logically related; both involve the university providing a service to students. However, through multidimensional scaling, we also determined that the issue “Scientists really have a cure for AIDS and are not telling us” was cognitively proximal to the focal topic “Money is the answer to all problems.” Although there is no apparent logical connection, groups that discussed the latter also tended towards consensus on the former.

Our indirect consensus effects seem especially strong considering the fact that there was no overt pressure for groups to come to agreement on the issues being discussed. Previous research showing indirect attitude change involved explicit attempts to change participants' beliefs (e.g., Dillehay et al., 1966; Hastie & Rawson, 2004). In the present study, groups were simply encouraged to discuss their opinions with other group members. An explicit goal to reach consensus on focal issues may increase consensus effects on indirect issues as well, and future research should test this possibility.

Some of our results may provide support for McGuire's dynamic model, which suggests that attitude change on one issue may lead to indirect attitude change on related issues in order to maintain consistency in our thought systems. This result would also be predicted by more recent models of constraint satisfaction (e.g., Simon & Holyoak, 2002), which argue that the attitudes held by individuals may influence the attitudes these individuals may hold on other related issues. Relationships may be based on a number of underlying structural dynamics, for example semantic relatedness or logical coherence of two issues. However, we did not directly test these models, but rather determined relationships among issues post hoc. One could easily make a priori predictions about indirect change as a function of constraint satisfaction by having groups discuss items with a pre‐established relationship (e.g., several items drawn from the same personality scale).

One fruitful avenue for future research would be to examine the process involved in indirect social influence. There are a number of potential mechanisms suggested by previous research that may have led to influence on issues related to the discussed issues but not on unrelated issues. For example, some of the related issues may have been logically inferred from discussed issues, as McGuire (1960) has shown. This explanation seems to describe some, but not all, of the change found on related issues within our study. The example discussed above, in which discussing an issue regarding whether professors should be permitted to give exams just before finals week may logically lead to change in other attitudes related to university policies, seems to provide evidence for McGuire's model. However, many of the related issues on which we found indirect attitude change do not have any clear logical connections. Therefore, indirect social influence may have occurred due to other mechanisms. For example, McGuire (1960) also suggests that people may change their opinions on related issues in order to reduce dissonance that may have arisen after changing their attitudes on discussed issues. That is, one's positions on attitude objects that are initially consistent with each other may become inconsistent due to attitude change on discussed issues. Although our method did not allow a direct test of this idea, we believe that it is unlikely that the indirect attitude change found in the present study was due to this mechanism. Tedeschi, Schlenker, and Bonoma (1971) have proposed that such dissonance reduction would only be likely to occur when people are concerned with the attributions others may make regarding their potentially inconsistent attitudes. Because our participants expressed their attitudes privately both before and after discussion, dissonance reduction due to impression management concerns does not seem to have played a role.

Another potential explanation for our results may follow from a spreading activation model of attitude change. That is, if two issues are semantically associated in one's mind (even with no logical relationship between the issues), attitude change on a discussed issue may lead to change on an associated issue that was not discussed. Although we are unable to find any research showing that spreading activation may lead to indirect social influence, this remains an intriguing possibility, and is consistent with the predictions of Simon and Holyoak's (2002) constraint satisfaction model.

Within our three‐person groups, it seems likely that participants changed their opinions due to both majority and minority influence processes. As discussed earlier, Crano and colleagues (Alvaro & Crano, 1997; Crano & Alvaro, 1998) have shown that minority influence processes can lead to indirect attitude change, and this mechanism may provide another explanation for the indirect social influence found in the current study. Because we did not record the group discussions, we were unable to tease apart the relative impact of majority and minority influence processes within our groups. Each of these mechanisms may have potentially been operating within our discussion groups, and we plan to include tests of the processes that may be involved in indirect attitude change in future studies.

We feel that our results may have implications for the culture of both small and large groups. Dynamic social impact theory (Latané, 1996) argues that cultural elements emerge from everyday communication among members of a group or society. We have argued (Latané & Bourgeois, 2001) that cross‐cultural differences in attitudes (Weiss, 2000) and personality (McCrae, 2001) may be explained from such a bottom‐up perspective, and our indirect consensus effects may provide a potential mechanism. For example, we (Latané & Bourgeois, 2001) found that discussing items drawn from a personality scale (the Big 5) led to local consensus effects within large spatially distributed groups, yet it is unlikely that people typically discuss such scale items in their everyday lives. Thus, conversations among people with differing opinions may lead one to change not only one's opinion on discussed items, but also on related, non‐discussed items. In this way, opinions, consensus, and culture may all emerge through repeated interpersonal interactions, in which social influence acts on both individuals' focal and peripheral attitudes and beliefs.