Discussion of Shared Information Can Increase the Influence of Divergent Members

Abstract

Based on the theoretical frameworks of information-sharing in groups and the linear discrepancy model, this study highlights the importance of communicating shared information for a divergent member to influence a group. Participants received information concerning whether “under God” should be in the Pledge of Allegiance. After stating individual opinions, they discussed the issue in small groups and came to a group decision on a continuous, ordered scale. Low divergent members, who had opinions closer to the average of other group members, had more influence than high divergent members. Group members with high divergence were more confident and talked more than others. However, there was no relationship between the amount divergent members talked or their confidence level and their amount of influence. Highly divergent group members who mentioned more shared information were more influential and came across as more knowledgeable.

Keywords

minority influence divergent group communication shared information linear discrepancy model social judgment theory

As groups discuss information and attempt to reach consensus on a task with a continuum of alternatives (Ohtsubo, Masuchi, & Nakanishi, 2002), low divergent members (group members who hold similar opinions on the topic and are greater in number than members who hold highly divergent opinions) exert substantial influence, especially in cases of judgmental tasks, tasks that do not have a correct answer (Davis, Kerr, Sussmann, & Rissman, 1974; Laughlin, 2011; Laughlin & Ellis, 1986). Highly divergent opinion holders (group members whose opinions are most different from other group members on a continuous ordered scale) often have their views excluded (Davis et al., 1997; Ohtsubo et al., 2002). The linear discrepancy model (Boster, Mayer, Hunter, & Hale, 1980) posits a reason for this. Because members are influenced by each argument they hear during a group discussion and because group members will usually state arguments that agree with their position, groups usually come to consensus and converge on a position near the average of the group, assuming that speaking terms are fairly evenly distributed (Boster et al., 1980). Thus, it is natural that the group’s ultimate decision converges toward the opinion closer to that of most group members.

This does not mean that highly divergent members have no influence, however. Group members holding a highly divergent opinion can influence other group members. The linear discrepancy model (Boster et al., 1980) states that “the greater discrepancy between the opinion advocated (by the argument) and the opinion held by the listener, the greater the listener’s opinion change” (p. 166). However, Boster and Mayer’s (1984) model has found that the perception of highly divergent members’ arguments is affected by normative influence and perceived as lower quality than majority arguments, thus reducing their effect on opinion change. Therefore, one of the challenges facing highly divergent members when trying to influence the rest of the group is framing their arguments in a way that increases the credibility of the argument (Boster, Hunter, & Hale, 1991).

In this article, we use the theoretical framework of the linear discrepancy model (Boster et al., 1980) to examine what types of communication increase credibility and influence for members with highly divergent opinions. In doing so, we attempt to answer the call by researchers for more inquires examining communication and minority influence (Levine & Russo, 1995; Smith, Tindale, & Dugoni, 1996; Tindale, Davis, Vollrath, Nagao, & Hinsz, 1990). Specifically, Davis et al. (1974) stated that research with social decision schemes and social judgment schemes is “indifferent to the important question of whether a member changes his preference as a function of discussion [ . . . ]. This is an empirical question that should be met by experimental estimates, not assumptions” (p. 269). Despite posing this challenge more than 40 years ago, previous research usually has not examined what types of communication can assist a member with a divergent opinion in influencing the rest of a group.

In addition, we attempt to add to the literature by investigating influence in a decision context with continuous group judgments; as Bonner, Gonzalez, and Sommer (2004) noted, “Although there is a rich literature dealing with categorical group decision modeling, far less is known about continuous group judgments”(p. 157). Many previous studies of minority influence, majority influence, and divergent opinions have used tasks with two or more discrete choices (e.g., guilty vs. not guilty; green slide vs. blue slide). Though tasks with dichotomous choice options have real-world relevance (e.g., juries), such choices make compromise difficult, as a group can only select one choice or the other. With continuous alternatives, compromise is possible but introduces more complications in the interactions between minority and majority members and often blurs the distinction between a minority and majority (Miller, 1989). With multiple, continuous alternatives, a divergent member, though not necessarily prevailing, may nevertheless exert subtle influence (Wood, Lundgren, Ouellette, Busceme, & Blackstone, 1994) because the group decision does not have to choose between two, discrete alternatives (Moscovici, 1980). A majority group member does not have to flip sides to be influenced by a highly divergent opinion. Using a judgmental task with continuous alternatives, this study investigates the role communication plays for a highly divergent member attempting to influence a group decision.

In the following sections, we summarize research with the linear discrepancy model and research on shared information in groups and then offer hypotheses for the present study.

Communication of Information

The linear discrepancy model postulates that message discrepancy should increase the influence of a message on the target’s attitudes (Boster et al., 1980), and some research has supported this (for review, see Boster et al., 1991). Other research has found less influence when the message is more discrepant with the target’s opinion (e.g., Sherif & Hovland, 1961; for review, see Boster et al., 1991). In explanation of this pattern, Boster et al. (1991) stipulated that in order for a source of a discrepant message to be influential, the source must be perceived to have at least a moderate level of credibility. Furthermore, subsequent research (Boster & Mayer, 1984) has found that arguments from more discrepant sources may be perceived as lower quality due to normative influence.

In groups, normative and information influence are often confounded, so that information presented by a group member is judged based on its conformity to group norms and majority viewpoints. Boster and Mayer (1984) manipulated the quality of arguments presented by minority and majority members to try to determine if informational influence is more important to minority members. Surprisingly, Boster and Mayer’s (1984) manipulation of argument quality had little effect on perceptions of argument quality; normative influence had a greater effect in increasing perception of argument quality, thus strengthening majority positions. Minority members are still able to influence: Garlick and Mongeau (1993) found that minority members could be more persuasive if presenting arguments perceived to be of high quality. The difference, however, demonstrates the disadvantage that minority members face. While majority members can influence through argument quality and normative influence (and the interaction between the two), minority members must depend more on the strength of their arguments alone.

Research in social judgment theory (Hovland & Sherif, 1980; Sherif, Sherif, & Nebergall, 1965) makes a similar argument, predicting increasing influence of a persuasive message with increasing discrepancy from the target’s attitude (Boster et al., 1980), up to a point. That point is when the message crosses the threshold from latitude of acceptance to latitude of rejection. A message in the latitude of rejection often produces perceptual distortions (i.e., contrast effect) that cause the message or argument to be viewed as more discrepant and less acceptable, reducing persuasion. Normative influence of the majority position may affect whether an argument is perceived in the latitude of acceptance or rejection. A majority member may be willing to move toward an argument when they perceive it is from another majority member, but then put that argument in the latitude of rejection when it is perceived as coming from a minority member.

Another factor that likely affects whether an argument will be considered in the latitude of acceptance or rejection is whether or not the argument is shared. Research concerning exchange of information in groups has distinguished between shared information (information that all members possess and have in common before group discussion) and unshared information (information known to only one member prior to discussion; Stasser & Titus, 1985). In general, group members discuss and focus on shared information more than unshared information (for review, see Brodbeck, Kerschreiter, Mojzisch, & Schulz-Hardt, 2007; Prahl, Dexter, Braun, & Van Swol, 2013; Wittenbaum, Hollingshead, & Botero, 2004).

Focusing on shared information is valuable for group members. Members who mention shared information are viewed as more competent and knowledgeable (Wittenbaum, Hubbell, & Zuckerman, 1999), are more influential (Kameda, Ohtsubo, & Takezawa, 1997; Van Swol & Seinfeld, 2006), and participate more (Sargis & Larson, 2002) than group members who mention more unshared information. Wittenbaum et al. (1999) noted that members who mention shared information are viewed as more competent and knowledgeable because mentioning shared information mutually enhances members’ knowledge by reinforcing their evaluation of their own knowledge and the speaker’s knowledge. Shared information can be socially validated and is easier to trust. Therefore, group members who mention shared information often elicit positive responses and social validation from others in the group (Wittenbaum & Bowman, 2004). Because discussion of shared information can positively affect perceptions of competence and influence, group members who mention more shared information should have increased influence on the group.

Because shared information helps shape all members’ opinions before the group discussion (Gigone & Hastie, 1993) and because it is perceived as more credible, it should be especially effective for highly divergent members whose information and arguments may otherwise be dismissed and rated as lower quality due to normative influence processes (Boster & Mayer, 1984) and perceptual distortions of perceiving the information in the latitude of rejection (Hovland & Sherif, 1980; Sherif et al., 1965). Thus, shared information may be especially important if highly divergent members are to succeed in influencing others. For example, Kameda et al. (1997) found that minority members who discussed more shared information were more influential in groups. Smith, Dykema-Engblade, Walker, Niven, and McGough (2000) further determined that appealing to common religious beliefs shared by the entire group enhanced the influence of a group member harboring a minority opinion. Similarly, Van Swol and Seinfeld (2006) reported that minority members, but not majority members, who discussed more shared information were more influential than those discussing less shared information.

The goals of the member may also matter, as posited in Wittenbaum et al.’s (2004) motivated information-sharing model. This model suggests that member goals moderate what type of information is discussed and how it is received by other group members. One important goal to consider is the need for social validation. Divergent members of groups elicit less social validation via agreement with other members, but mentioning shared information can provide them with some degree of social validation by virtue of mutual enhancement. Wittenbaum and Bowman (2004, Study 1) observed that, when need for accuracy was low, the need for social validation diminished and mutual enhancement was not important, but when need for accuracy was high, increasing social validation through mutual enhancement was important. Divergent members presumably should have a greater need for social validation, and other group members should be more highly motivated to try to validate a divergent member’s information because the divergent opinion cannot be validated through majority support and conformity to group norms. One mechanism by which shared information provides social validation is perceptions of the speaker’s knowledge of the topic.

The Present Study

The distinction between shared and unshared information is controlled in laboratory experiments in which participants engage in a fictitious task. In these types of experiments, availability of information is manipulated: All members are provided with some information, while other information is given to just one member (Wittenbaum et al., 2004). For example, participants might be asked to make a hiring recommendation by selecting between several applicants. All participants have access to applicant resumes (shared information), and each participant has access to a unique interview transcript (unshared information).

Though this type of experiment has been used as an effective tool to investigate shared and unshared information, it does not reflect real-world issue discussions. In the real world, information cannot be divided neatly into shared and unshared categories, nor can it be determined ante tempore what a discussion participant knows or does not know about the issue at hand. As such, the current experiment uses a different procedure. In our study, participants read an article concerning whether the words “under God” should be in the U.S. Pledge of Allegiance.¹ The pro and con information was evenly distributed. Thus, no information was given to one group member only, and shared information offered equal amounts of support and opposition to “under God.” In addition to this shared information, some group members brought up information they acquired from other sources, of which other group members were not previously aware. We call this “own information” and avoid the term “unshared information” because we cannot control for own information the way unshared information is controlled in a typical information-sharing experiment.

Past research has indicated that group members who focus on unshared information are less influential than members who focus on more shared information (Sargis & Larson, 2002). However, as “own information” differs from how unshared information has been operationalized in previous research, we are unsure how members will react to it when mentioned. In light of previous research involving unshared information (Sargis & Larson, 2002), we could predict that group members who mention more own information will have less influence in the group. However, as own information is not given to participants by the researcher and is learned from previous experiences with the issue, mentioning own information could be an indicator of expertise and thereby increase influence.

We expect that communication of shared information will be more important to highly divergent members because shared information can be validated and will increase the perception of the divergent member as knowledgeable. This will help avoid normative processes of dismissing information from highly divergent members as lower in quality. We examine influence as a measure of how much of the group decision is pulled toward the divergent member’s opinion. We expect that divergent members who mention more shared information will be perceived as more knowledgeable by the group and more influential. In addition, we anticipate that the effect of mentioning shared information on influence will be mediated by other group members’ perceptions of the divergent member as more knowledgeable. We do not expect any differences in the discussion of shared or own information between divergent and non-divergent members. Rather, we expect the discussion of shared information to have a stronger effect on the influence of divergent members. The following hypotheses and research question formally state these expectations.

Hypothesis 1: There will be a positive relationship between percentage of shared information mentioned and influence for divergent members, but not for non-divergent members.

Hypothesis 2: Other group members will perceive divergent members who mention a greater percentage of shared information as more knowledgeable than divergent members who mention less.

Hypothesis 3: Other group member perception of divergent member knowledge will mediate the effect of shared information on influence.

Research Question 1: Will there be a relationship between percentage of own information mentioned and influence in the group, and will there be differences by type of member?

In addition to examining how discussion of information affects influence in the group, we investigate differences in the ways divergent members communicate. As discussed above, group members come to the discussion with previous opinions and varying degrees of knowledge about the issue. Just as their knowledge differs, group members may differ in their behavior during the group discussion. For example, in an experiment in which groups discussed the legalization of marijuana, Van Swol (2009) found that extreme group members talked more in the group. Furthermore, evidence indicates that people with extreme attitudes are more involved with and committed to the issue, view the issue as more important, and have more confidence in their attitude than do less extreme members (Eagly & Chaiken, 1993; Judd & Brauer, 1995; Krosnick, Boninger, Chuang, Berent, & Carnot, 1993; Millar & Tesser, 1986; Sunstein, 2003). Divergent group members with more extreme opinions should talk more and have more confidence than other members (Sniezek & Henry, 1989). This may help them to avoid being marginalized from the group. As such, we offer this final hypothesis.

Hypothesis 4: The most divergent members in a group will be more confident, talk more, and view the issue as more important than other group members.

Method

Participants and Design

Participants (males = 63, females = 167) were undergraduate students from a large, public Midwestern university in the United States receiving extra credit in communication courses for participation. Participants were randomly assigned groups based on their availability. There were 29 four-person groups (22 female, 7 male), 21 five-person groups (14 female, 7 male), and 3 three-person groups (3 female, 0 male). The choice of same-sex groups was to avoid any status expectations linked with sex that could affect level of participation (Balkwell & Berger, 1996).

Procedure and Materials

To select a controversial topic, eight issues were pre-tested with a brief survey of around 100 students enrolled in communication classes. Survey-takers were asked for their opinion (on a scale of 1-9) for each issue. “Under God” in the Pledge of Allegiance was used because there was a strong range of opinions and strong feelings on the issue in our student sample.

Upon arrival at the lab, participants provided informed consent. Then, they were asked about their citizenship. Non-U.S. citizens were thanked and excused from the experiment. Participants were told that the study concerned how people discuss controversial issues in groups and were asked to read a one-page sheet with background information relating to whether the words “under God” should remain in the Pledge of Allegiance. The information came from a non-partisan website (Should the words “under God” be in the US Pledge of Allegiance?, n. d., http://undergod.procon.org) and was credited as such (see appendix). It reflected both sides of the debate. Participants also received a copy of the Pledge. A large (3 ft × 5 ft) U.S. flag was hanging in the laboratory to make the issue more salient.

After reading the information, participants indicated their attitude toward the issue on a 9-point scale with response options from 1 (I completely oppose the words “under God” appearing in the U.S. Pledge of Allegiance) to 5 (unsure) to 9 (I completely support the words “under God” appearing in the U.S. Pledge of Allegiance). Following this activity, they indicated their confidence in the correctness of their views, importance of the issue, and knowledge of the issue on a 5-point scale with response options that ranged from 1 (not at all confident, important, knowledgeable) to 3 (somewhat . . .) to 5 (very . . . ).

After participants finished, ratings sheets were collected, and participants were brought together as groups. Each member was assigned to a spot at a table labeled with a large visible letter to help distinguish their contributions during transcription. Participants were told

Now please discuss the issue as a group and make a group decision about the issue. We ask that you discuss the issue for at least 10 minutes, but you may continue up to 30 minutes. After the group discussion, you will be asked to voluntarily say the Pledge of Allegiance either with or without the words “under God” based on your group’s decision. You are free to decline to say the Pledge for any reason, which you do not have to tell us. All interaction will be videotaped. Any questions?

After reading the instructions and answering questions, the experimenter put a group decision sheet in the middle of the table, instructed the members to reach a decision on the issue, and then left the room. The group’s decision and confidence in the correctness of that decision was recorded on the same 9-point scale used for individual participant ratings.

Once discussion ended and the group had made a decision, the members (voluntarily) faced the flag and recited the Pledge the way that their group had decided (with or without “under God”). Participants did not have to recite the Pledge of Allegiance, and some group members (n = 13) chose to say it differently from their group decision or not to say it at all. This happened in 9 groups, as in some groups more than one member chose not to conform to the group decision.

Finally, the participants individually indicated their attitude toward the topic and their confidence in the correctness of their attitude on the same scale previously used. They also completed a questionnaire on which they rated each group member (excluding self) for knowledge about the topic and influence during discussion on a 5-point scale with response options ranging from 1 (not at all knowledgeable, influential) to 3 (somewhat . . . ) to 5 (very . . . ). After completing the questionnaire, participants were debriefed and thanked. During debriefing, they were told that the study was investigating the influence and communication of the most extreme member in the group compared with other members.

Measures

Because we are using a task with continuous, ordered alternatives, we define difference of opinion in terms of divergence. The divergent member is the member in the group whose opinion is furthest on average from the other group members. This member is often a minority member, as defined by which side of the scale group members’ judgments fall on. For example, in a group in which members’ opinions are 9, 7, 7, 5, and 3 on the ordered scale of 1 to 9, the member whose opinion is 3 is the most divergent member and a minority member. In cases where groups were more homogeneous and all opinions were grouped on one side of the scale (including the midpoint of 5), the most divergent member was not a minority member in the sense of being on the other side of the scale midpoint than the group. For example, in a group with members’ opinions as 3, 2, 2, 2, and 5, the member whose opinion is 5 is the most divergent member. Because members did not give dichotomous decisions, we used most divergent member as the member in most disagreement with the majority of group members.

In order to measure influence, we developed two measures that reflect variance within the group. The first measure was comparative influence. This took the absolute value of a member’s pre-group opinion minus the group opinion and subtracted from it the average absolute difference of the other group members from the group opinion. For example, if a member chose 6, the group opinion was 5, and the average difference from the group opinion for the other members in this group was 2, then this member had a comparative influence score of −1.00. Larger negative numbers indicated that a member’s pre-group opinion was closer to the group opinion than other members in the group; larger positive numbers indicated that their opinion was further from the group opinion than others.

Our second measure was residual influence (Braun, 2012). It uses the mean of members’ pre-group opinion (pre-group mean) as the standard and measures how much the group opinion moves from this standard in the direction of an individual member’s opinion. In essence, it measures the influence an individual group member has to pull the group away from the pre-group mean. This is represented by

\frac{[\begin{array}{l} (Group opinion - pre-group mean) \times \\ (Individual opinion - pre-group mean) \end{array}]}{pre-group mean}

For this measure, higher positive numbers indicated more influence. Both measures are derived from work by Rashotte and Smith-Lovin (1997) who measure influence as the difference between a group member’s expected and actual change in opinion. Expected change is a group member’s pre-discussion opinion minus the group’s average pre-discussion opinion; actual change is the member’s pre-discussion decision minus the group’s final decision. Rashotte and Smith-Lovin’s formula produces a unique value for each group, not for each group member. Therefore, we used measures that accounted for how close a group member was to the group opinion in comparison with other members, so that we could assess the influence each individual group member had relative to others.

Coding

Two graduate students transcribed all interactions verbatim; the first author checked over their work. Two interactions were not recorded due to technical difficulties, resulting in transcripts for 51 groups.

A two-layered coding process was used in preparing the transcripts. First, two undergraduate coders coded them into idea units. These coders, blind to experimental hypotheses, put brackets around each idea unit to indicate where a new idea began and ended. Idea units were based on grammatical definitions of simple sentences and single clauses. Guetzkow’s (1950) U is a measure of unitizing disagreement. It is computed by subtracting the amount of units coded by the second coder from the number of units coded by the first coder and dividing this difference by the total amount of units coded by both coders across all discussions. Lower numbers (close to zero) indicated acceptable levels of reliability. The coding was reliable (U = .0316; agreement higher than 96%).

One of the authors reviewed the transcripts and resolved any discrepancies between the two coders involving the idea units. In the second stage of coding, two different graduate student coders received the transcripts with the brackets around idea units in the transcripts. They were to code each idea unit as (1) statement of opinion: statement of their view (e.g., “I’m a 7.” “I don’t think it [under God] should be in there.”), (2) information from experimental sheet: information from the sheet or alluding to the sheet (see appendix), (3) new information: information related to the discussion of whether or not the phrase should be in the Pledge and not from the experimental sheet, and (4) affirmation of another group member: talk meant to show agreement (e.g., “Yeah,” “Me too”). After the first round of coding, Cohen’s kappas were too low (.65). Therefore, the two coders met with the first author to go over several transcripts and then recoded all the transcripts. For this second round, Cohen’s kappa (.91) was good. For each idea unit coded as shared information, the two coders labeled which of the 16 statements given to participants that the piece of shared information corresponded to (see appendix). Initial agreement was 86.7% and discrepancies were resolved by the first author. For each idea unit coded as new information, the two coders labeled it as pro-“under God” (e.g., “Taking out under God might get people really upset, so upset that they might send their kids to a religious school”), anti-“under God” (e.g., “The whole under God stuff was added because people thought communists wouldn’t say it, it was McCarthyism”), and neutral (e.g., “We said the Pledge during grade school, but I don’t think we said it during high school.”). Coders were instructed to code other information as neutral if they were unsure about whether it was for or against keeping “under God” in the Pledge. Coders agreed on 71.5% of coding for pieces of new information, and as before, the first author resolved all discrepancies.

Results

The results section is organized as follows: First we provide a manipulation check on divergent members. Next, using the influence measure we created, we determine if there is a relationship between discussion of shared information and influence for divergent members and whether this relationship is mediated by divergent members being perceived as more knowledgeable. We also check if there is a relationship between discussion of own information and influence. Because members are nested in groups and individual member data violate the assumptions of independence of observations, we use hierarchical linear modeling (HLM7 software) and multilevel modeling for many of the analyses. When the predictor variables are continuous, we use hierarchical linear modeling. For hierarchical linear modeling, we grand-mean-centered Level-2 variables and group-mean-centered Level-1 variables (Enders & Tofighi, 2007). When the predictor variables are categorical, we use multilevel modeling and include group ID as a random factor with correlated error variance to account for non-independence.

Manipulation Check

We calculated the mean opinion discrepancy from other group members’ pre-discussion opinions for each member. For example, if a four-person group consisted of members whose pre-discussion opinions were 5, 3, 4, and 8, 8 would be the member with the highest divergence, with an average of 4.00 from others. The most divergent member² was significantly more discrepant in opinion from others in the group (M = 3.82, SD = 1.33) than the less discrepant members (M = 2.28, SD = 0.87), F(1, 223) = 99.76, p < .001, η² = .31. This confirms that divergent member’s opinions were diverging significantly from the rest of the group. Of the 58 members who were the most discrepant members, 31 were the sole minority in the group, 4 were a minority with one more fellow minority member, 6 were in the majority, 9 were in homogeneous groups (all members on one side of the scale, including the midpoint), and 7 were in plurality groups (equal number of members on both sides of the scale).

Hypothesis 1

Divergent members, whose opinions reflected the highest discrepancies, were compared with other group members for influence in the group. For the first two analyses, we used hierarchical linear modeling to determine whether or not discussion of shared information and own information contributed to increased influence. Percentage of shared information mentioned and percentage of own information mentioned correlated (r = .27, p < .001) and were tested separately. To test Hypothesis 1, the prediction variables were shared information (percentage of shared information discussed), member status (0 = non-divergent member, 1 = divergent member), and interaction terms between divergent member status and shared information as Level 1 predictors. Percentage of shared information discussed was the number of pieces of shared information mentioned divided by the number of statements coded for the group in which the participant was a member. For example, a participant who mentioned 14 pieces of shared information (including repetitions) and was in a group with 201 coded idea units during the group discussion was coded as stating 7% shared information. The criterion variable was comparative and residual influence measures (in two separate analyses) with group size and gender included as Level 2 control variables.

Following the recommendation of Raudenbush and Bryk (2002), a hierarchical null model with no predictors was conducted to decompose the variance. See Table 2. The results provided evidence that there was not significant between-group variance in influence. The intraclass correlation for both influence measures was 0, indicating that none of the variance in influence was explained by between-group differences. Comparing deviances from the null model with the hypothesized model for residual influence for shared information (255.41 – 210.85) found that the experimental model controlled for more relevant variance between groups (χ² = 44.56, p < .05), and similarly, the hypothesized model for comparative influence (857.27 – 720.40) accounted for more variance between groups than the null model (χ² = 160.39, p < .001). Therefore, we conducted analyses including our Level 1 predictors. The results revealed that for both residual influence and comparative influence, the effect of member status was significant, indicating that non-divergent members had more influence. There was a significant effect of divergent member and shared information. Divergent members who mentioned a higher percentage of shared information were more influential. There was not a significant effect of non-divergent member and shared information, indicating that there was no relationship between influence and mentioning shared information for non-divergent members. See Table 1 for means and Table 2 for measures. In support of Hypothesis 1, the relationship between increased influence and discussion of shared information only held for members with the highest divergence scores.

Table 1.

Means and Standard Deviations of Group Members.

	Most divergent member (n = 58)		Other group members (n = 172)
	M	SD	M	SD
Comparative influence	1.71	1.68	−0.56	1.13
Residual influence	−0.10	0.67	0.03	0.29
Confidence in opinion	4.11	0.82	3.72	0.82
Knowledgeable	2.95	0.72	2.82	0.87
Importance of issue	3.03	0.97	2.66	0.98
% of words spoken	26%	0.11	22%	0.11
% of shared information	2.90%	0.002	2.20%	0.003
% of shared information supporting opinion	1.80%	0.02	1.38%	0.01
% of shared information against opinion	1.04%	0.01	0.73%	0.01
% of own information	6.90%	0.006	5.60%	0.003
% of own information supporting opinion	1.97%	0.02	1.56%	0.02
% of own information against opinion	0.95%	0.01	0.64%	0.01
% of opinion statements	5.37%	0.04	4.54%	0.03
% of affirmations	4.00%	0.03	3.45%	0.03
Other rated confidence	3.93	0.71	3.87	0.61
Other rated knowledge	3.53	0.57	3.51	0.57
Rated influence in discussion	3.53	0.72	3.38	0.68
% of time speak in group	29.96%	12.09	25.94%	10.28
Move toward group opinion	1.09	1.62	0.53	1.03

Table 2.

Hierarchical Linear Modeling Predicting Influence From Shared Information and Member Status.

	Residual influence				Comparative influence
	Null model		Multilevel model		Null model		Multilevel model
Model	γ	SE γ	γ	SE γ	γ	SE γ	γ	SE γ
Intercept	−0.00	0.03	−0.00	0.03	0.01	0.11	0.00	0.08
Member status			−0.36**	0.11			3.01***	0.34
Information shared %			−1.30	2.00			−2.84	6.10
Shared × Status interaction			8.75*	3.48			−20.55*	10.49
Group gender			−0.00	0.07			0.00	0.21
Group size			−0.00	0.05			0.01	0.15
U₀ SD	0.00		0.00		0.01		0.01
Log likelihood	−127.7		−105.4		−428.6		−348.4
Deviance	255.41		210.85		857.27		696.88

p < .05. **p < .01. ***p < .001.

Hypotheses 2 and 3

To test Hypothesis 2, we examined whether mentioning shared information related to other group members’ perceiving those with the highest divergence score as more knowledgeable. We used the mean of other group members’ ratings in the post-questionnaire (not at all [1] to very [5]) of the divergent member as the measure of knowledge about the issue and perceived influence. There was a significant positive relationship between percentage of shared information mentioned and perceived knowledge about the issue (r = .46, p < .001) and a significant positive relationship between percentage of shared information mentioned and perceived influence (r = .50, p < .001). Therefore, Hypothesis 2, that divergent members who discuss more shared information will be perceived as more knowledgeable, received support.³

Using the Preacher and Hayes (2004) bootstrap approach to obtain confidence intervals (CIs), we attempted to determine if other members’ perceptions of divergent members’ knowledge concerning the issue mediated the relationship between percentage of shared information mentioned and influence (both residual and comparative influence). For residual influence, a 95% bias corrected bootstrap CI contained zero (95% CI = [−3.21, 5.41]; Z = 0.50, p = .61), which indicated that the indirect effect was not significant and mediation did not occur. Similarly, for comparative influence, a 95% bias corrected bootstrap CI contained zero (95% CI = [−14.33, 7.06]; Z = −0.67, p = .51), which indicated that the indirect effect was not significant and mediation did not occur. Hence, there was no support for Hypothesis 3.

Research Question 1

We ran the same HLM7 tests for percentage of own information mentioned. For residual influence, no significant findings emerged. For comparative influence, there was a significant effect for member status, but not for own information. There was no statistically detectable effect of own information on influence for either deviant or non-deviant members. See Table 3.

Table 3.

Hierarchical Linear Modeling Predicting Influence From Own Information and Member Status.

	Residual influence				Comparative influence
	Null model		Multilevel model		Null model		Multilevel model
Model	γ	SE γ	γ	SE γ	γ	SE γ	γ	SE γ
Intercept	−0.00	0.03	−0.00	0.03	0.01	0.11	0.00	0.09
Member status			−0.29	0.16			2.43***	0.49
Information own %			−1.38	0.97			4.26	2.91
Own × Status interaction			2.55	2.15			−1.65	6.42
Group gender			−0.00	0.07			0.00	0.21
Group size			−0.00	0.05			0.01	0.15
U₀ SD	0.00		0.00		0.01		0.01
Log likelihood	−127.7		−118.4		−419.7		−346.2
Deviance	255.41		236.85		857.27		692.38

p < .05. **p < .01. ***p < .001.

Hypothesis 4

Next, we examined how members with the highest divergence scores differed from the other members in self-ratings prior to group discussion using one-way analysis of variance to test differences. The members with the highest divergence scores were more confident in their opinions, F(1, 223) = 9.59, η² = .04, p = .002, and rated the issue as more important, F(1, 223) = 5.99, η² = .03, p = .015. They did not rate themselves as more knowledgeable, F(1, 223) = 1.01, η² = .01, p = .32. See Table 1 for means. The most discrepant members (32%) were more likely than the other members (11%) to have used the extreme ends of the scale (1 or 9) to indicate their opinions, F(1, 223) = 15.31, η² = .06, p < .001. Finally, the most discrepant members (M = 5.02, SD = 2.96) were more likely than other members (M = 5.71, SD = 2.09) to oppose the words “under God” appearing in the Pledge, F(1, 223) = 3.69, η² = .02, p = .056, but effects were marginal.

Using linear mixed models, we addressed whether or not the most divergent members were communicating differently from the other members. First, we examined what percentage of the words they accounted for in the group. In support of Hypothesis 4, the most divergent members spoke more than other participants, γ₁₀ = −0.04, SE = 0.02, t(216) = −2.09, p = .04. The most divergent members mentioned a significantly higher percentage of shared information, γ₁₀ = −0.007, SE = 0.003, t(218) = −2.26, p = .025, but not shared information in support of their opinions,⁴ γ₁₀ = −0.004, SE = 0.002, t(194) = −1.89, p = .06, or shared information in opposition to their opinions, γ₁₀ = −0.003, SE = 0.002, t(189) = −1.81, p = .07. The most divergent members mentioned significantly more own information, γ₁₀ = −0.013, SE = 0.006, t(218) = −2.01, p = .046, but not significantly more own information in support of their opinions, γ₁₀ = −0.004, SE = 0.003, t(198) = −1.20, p = .23, or own information in opposition to their opinions, γ₁₀ = −0.003, SE = 0.002, t(198) = −1.74, p = .08. There were no differences between members on statements of opinion, γ₁₀ = −0.01, SE = 0.005, t(218) = −1.55, p = .12, or statements of affirmation, γ₁₀ = −0.01, SE = 0.004, t (218) = −1.21, p = .23. The only significant findings for communication were that the most divergent members talked more and discussed both more shared and own information than the average of the non-divergent members. See Table 1 for means.

We compared the group mean of percentage of shared information discussed for each member with the percentage of own information discussed in a repeated measures (information) analysis of variance. A higher percentage of own information (M = 6.18%) was discussed than shared information (M = 2.38%), F(1, 47) = 80.16, η² = .63, p < .001.

Confidence and Participation

Next, we checked for effects of confidence and participation on influence. Using hierarchical linear modeling, we predicted comparative and residual influence measures (in two separate analyses) with participation (percentage of words of discussion), confidence in pre-discussion opinion, member status (0 = non-divergent member; 1 = divergent member), and interaction terms between divergent member status and participation and confidence. Group size and gender of group were included as a control in all analyses as Level 2 variables. For neither comparative nor residual influence was the relationship significant. Level of confidence and participation also showed no significant relationship to influence in the group. See Table 4.

Table 4.

Hierarchical Linear Modeling Predicting Influence From Confidence, Member Status, and Participation.

	Residual influence				Comparative influence
	Null model		Multilevel model		Null model		Multilevel model
Model	γ	SE γ	γ	SE γ	γ	SE γ	γ	SE γ
Intercept	−0.00	0.03	−0.00	0.03	0.01	0.11	0.00	0.09
Member status			0.00	0.40			1.82	1.18
Confidence			−0.04	0.05			0.15	0.14
Participation			−0.17	0.33			0.21	0.97
Confidence × Status interaction			−0.09	0.09			0.33	0.27
Participation × Status interaction			0.96	0.69			−3.26	2.06
Group gender			−0.00	0.07			0.00	0.21
Group size			−0.00	0.05			0.01	0.15
U₀ SD	0.00		0.00		0.01		0.01
Log likelihood	−127.7		−132.0		−419.7		−363.1
Deviance	255.41		264.00		857.27		726.19

p < .05. **p < .01. ***p < .001.

Additional Analyses of Divergent Members

Using linear mixed models, we attempted to determine how much the most divergent member was influenced by the group relative to other members. We assessed how much members’ post-discussion opinion moved toward the group opinion in comparison with their pre-discussion opinion. For example, if a member’s pre-discussion opinion was 3, the group decision was 8, and the member’s post-discussion opinion was 5, then the member was noted as moving 2 intervals toward the group opinion. Members with the highest divergence scores moved toward the group opinion more than other members, γ₁₀ = −0.56, SE = 0.17, t(177.08) = −3.19, p = .002. Highly divergent members recited the Pledge differently from the group more often (15% of the time) than others (4%), χ²(df = 1) = 6.20, p = .013.

Discussion

This study addressed how highly divergent members may influence a group’s decision concerning a controversial issue. Divergent members were more confident and participated more. Mentioning more shared information made divergent members more influential (though the effect was not found for other members), and discussing more shared information raised perceptions that the divergent members were more influential and knowledgeable. Contrary to expectations, however, greater perception of knowledge did not mediate the effect of shared information on actual influence for divergent members, and confidence and participation were not significantly related to influence. Furthermore, the findings reveal that divergent members were less influential than other members; this finding supports previous research (Davis et al., 1997; Ohtsubo et al., 2002).

This article adds to research involving divergence of opinions in several ways. First, we replicated an important finding that group members with divergent opinions can use shared information to enhance their influence in a group. We were able to replicate this using a different type of task from those normally used in information-sharing studies, in which group members were defined by divergence rather than their minority/majority status. Second, we examined influence in interacting groups, rather than how people react when they are merely told an opinion is divergent or non-divergent. Researchers (Levine & Russo, 1995; Smith et al., 1996; Tindale et al., 1990) have stressed the importance of examining communication and influence in interacting groups, and this article has contributed to that end. Third, we created a behavioral measure of influence rather than relying on participant perception of influence in the post-discussion questionnaire. Fourth, this article integrates research in the area of the linear discrepancy model and information-sharing in groups and addresses motivated information-sharing theory’s (Wittenbaum et al., 2004) assertion that member goals affect how group members attend to and process information. Finally, we found that discussing more shared information enhances influence among divergent members and tested the role of other members’ perception of knowledge as a mediator.

Influence and Divergent Members

We hypothesized that highly divergent members in the group would behave differently from other members, and consistent with our predictions and with previous research (Krosnick et al., 1993; Van Swol, 2009), the most divergent members were more confident, perceived the task as more important, and participated more than other members. We would not expect these findings to generalize to all types of group tasks, however. We selected a controversial issue related to current events and for which participants, especially ones with extreme opinions, came to the group discussion with pre-existing attitudes about separation of church and state and the role of religion in society. Divergent members may have had prior experience defending their more extreme viewpoints concerning the Pledge or about separation of church and state and could have become more confident and talkative as a result. We would not expect a group member with a divergent opinion about a fictitious task that is first introduced during the experiment to be more confident and view the issue as more important than other members.

Although we determined that divergent members were behaving differently from other members, there was no significant relationship between their behavior and influence. Divergent members participated more during discussion and had more confidence in their pre-discussion opinion than other participants, but there was no significant effect of participation or confidence on influence in the group. In both cases, however, the results were in the predicted direction. Previous research has indicated that participants who participate more in a group usually have more influence on group outcomes (Bonito & Hollingshead, 1997; Bottger, 1984; Henningsen & Henningsen, 2015; Sorrentino & Boutillier, 1975; for an exception, see Yuan, Bazarova, Fulk, & Zhang, 2013), but we did not replicate this. The old adage advises one to never discuss religion or politics. Our topic included both. Anecdotal evidence from watching the videos suggested that the undergraduate participants generally did not want to impose their religious views on others. Because our task included a subjective preference about religious values, participation in and of itself may not have been a strong source of influence. Rather, similarity of religious values may have been important. It is possible that group members who talk more have more influence, if their views do not diverge too much from the rest of the group (Henningsen & Henningsen, 2015). This could be tested in future research.

The results for residual influence and comparative influence indicated that, in terms of the overall group decision and in comparison with other members, highly divergent members were less influential. For example, comparative influence reflects how close a member’s pre-group opinion is to the group opinion, in comparison with the other group members. The results for comparative influence showed that the most divergent members failed to prevail in the group discussion. That is, the group did not endorse the opinion of the most divergent member. Therefore, comparatively, their pre-group opinion was still farther from the group than other members. This supports previous research on influence in groups reaching consensus on tasks with a continuum of alternatives (Davis et al., 1997; Ohtsubo et al., 2002).

Communication of Information and Influence

Highly divergent members who mentioned a higher percentage of shared information were more influential, but the introduction of shared information was not related to influence for the other members. The importance of discussing shared information supports research involving centrality (Kameda et al., 1997) and mutual enhancement (Wittenbaum et al., 1999), and our results were consistent with those of Van Swol and Seinfeld (2006).

When group members hear shared information mentioned, the information is mutually enhancing and viewed as more important, relevant, and accurate because another member thought this information important enough to mention (Wittenbaum & Bowman, 2004). We hypothesized that communication of information should matter for divergent members because their information may be perceived as lower quality due to normative influence processes (Boster & Mayer, 1984). Therefore, they would have a higher need for social validation of their information to be perceived as credible and knowledgeable. The need for social validation may underlie the differences between divergent and other members. Social validation underlies mutual enhancement, and researchers have found that when social validation concerns are low, mutual enhancement is reduced (Wittenbaum & Bowman, 2004). The need for social validation is lower for group members whose opinions are shared by others. Their opinion is socially validated by agreement from others, so members already have a sense that they are correct, and their information is likely to be perceived as of higher quality due to normative influence. Additional social validation emerging from discussing shared information is unnecessary for group members with non-diverging opinions. For highly divergent members, however, social validation may be important because their need for social validation is not satisfied by others’ agreement with them. If divergent members can receive social validation by mentioning shared information, then they may increase their credibility. In support of this view, other group members viewed divergent members who mentioned more shared information as more knowledgeable. However, perception of knowledge about the topic did not mediate the effects of shared information on influence.

One question that needs to be resolved in future research is the factor mediating between discussion of shared information and greater influence for divergent members. We had hypothesized that perception of knowledge would mediate this, but it was not significant. Possibly, a more comprehensive measure of knowledge and competence could tap into a broader perception of the divergent member and could mediate. However, other measures like perception of similarity or likeableness could act as mediators and should be tested in future research. The linear discrepancy model (Boster et al., 1980) emphasizes how each argument can influence other group members. It may be the perception of the information and arguments that are presented is what mediates influence and not the perception of the communicator. Therefore, research should have participants rate information after discussion to determine its effect (Van Swol, 2007).

We did not assess how the communication goals of highly divergent members affect their discussion of shared information. Wittenbaum et al. (2004) noted that members’ goals can have an impact on the type of information they choose to share with the group, and members with more validation goals may repeat more shared information after witnessing other members react positively to it. This suggests that information-sharing can be strategic and measuring participants’ goals offers a direction for future research.

There were no effects of the discussion of own information on influence. Possibly, it is shared information that is driving the results for mutual enhancement and social validation (Wittenbaum et al., 1999) and increased influence (Kameda et al., 1997; Van Swol & Seinfeld, 2006). However, we do not want to infer too much from the results for own information because we did not have control over participants’ own information, and much of the own information was neutral in position. Other participants may not have responded strongly toward such information. Furthermore, in the typical information-sharing experiment, in which unshared information is distributed by the experimenter, participants have no reason to feel pride in having it. However, people who possess own information may feel pride in possessing the information because acquiring it may have required effort (e.g., reading about the topic), show special access (e.g., interaction with high-status others such as professors, from whom one learned about the topic), or reflect greater intellect. In some cases, therefore, discussing one’s own information may offer a strategy for winning approval. It may be that discussing own information had competing influences on perception of the speaker. In some cases, it reduced perception of competence, ostensibly because of a lack of social validation, but in other cases, the information could have conveyed expertise. In conclusion, people with divergent opinions may suspect that highlighting their unique expertise may help them achieve influence in the group, but our results do not support this.

A comparison of the percentage of own information and shared information individuals mentioned in groups found that participants mentioned a much higher amount of own information. Because we could not control the amount of own information available for participants to mention, we do not conclude that our results are contrary to previous research that has found that groups discuss a higher percentage of shared information than unshared information (Stasser & Titus, 1985). Rather, we contend that in many naturally occurring discussions, the role of own information may be quite large as participants have lots of their own information available at their disposal. Furthermore, participants may have been more comfortable mentioning their own information than newly learned shared information.

Future Research and Limitations

Influence and divergence of opinion were central to this study, but the distribution of opinions in the group and how they interact with the influence of divergent members should be examined in future research. In the present inquiry, the opinions were not strongly skewed toward either pro or con content; this is why the present topic was selected. For other pre-tested issues, like medical marijuana, we found a strong favorable bias. For an issue like this, a divergent member who opposes medical marijuana may have a very different sort of influence in a group than a divergent who is closer to the trend (Maass, Clark, & Haberkorn, 1982). Therefore, it is likely that a combination of task and social norms determines how influential a divergent member is in the group.

One of the limitations of this study was that although we assessed participants’ attitudes on a scale, they could have perceived the choice as binary. Participants were either pro or con “under God,” or in some cases neutral. However, participants were still able to compromise and shift position during the discussion on the continuous, ordered scale. This allowed members to move in a certain direction without flipping their opinion from pro to con; therefore, a divergent member could shift toward the other group members without perceiving themselves as completely changing.

Conclusion

This article integrates research in the area of linear discrepancy model and information-sharing in groups. This study used a controversial task involving the often taboo subjects of religion and politics and found that divergent members were able to have influence with shared information and appear more knowledgeable. The implications for other topics are wide reaching. In arenas from organizations, politics, and even family discussions, one may find oneself with an opinion that diverges significantly from others, and much past research has found that divergent members are often not too influential (Levine & Prislin, 2013). Therefore, having a strategy to appeal to shared information provides an important communication tool to divergent members, and given the now firm grounding for this finding, books and articles aimed at applied audiences should suggest this as a viable strategy.

Footnotes

Appendix

Should the Words “Under God” Be in the U.S. Pledge of Allegiance?

About this topic
The Pledge of Allegiance was first written in 1892 by Francis Bellamy, a Baptist Minister and Socialist, for the patriotic family magazine The Youth’s Companion. At that time, it read “I Pledge Allegiance to my Flag and the Republic for which it stands; one nation indivisible, with liberty and justice for all.”
The Pledge officially became part of the U.S. Flag Code in 1942. In 1954, the words “under God” were added to the Pledge by Congress with approval from President Dwight Eisenhower.
A 2001 lawsuit, filed by Michael A. Newdow, contested the inclusion of the phrase “under God” in the Pledge of Allegiance, citing a violation of church-state separation principles.
To clarify this multifaceted issue, here are pro and con responses to the central questions in the words of key proponents and opponents in the debate.
Pro under God	Con under God
Pro: Some proponents argue that the United States was created from the Christian principles of the Founding Fathers and as such the Pledge should respect the country’s heritage. Others say the U.S. Constitution protects freedom of religion and not freedom from religion. Many advocates of including “under God” in the Pledge point out that polls show at least 80% of Americans support it, that federal law already contains 22 references to “God,” and that Presidents swear an oath of office ending with “so help me God.” Many others claim the incorporation of religious language is a reflection of the U.S. civic culture and not a promotion of religion.	Con: Some opponents argue that church and state should be kept separate as the Founding Fathers intended. Others say the phrase “under God” in the Pledge places “undue coercion” on young children, thus violating the Establishment Clause of the First Amendment. They also declare that the U.S. Constitution protects minority rights against majority will. Many advocates of removing “under God” point out that the phrase was not written into the original pledge and that the opposition to returning to the original pledge is proof that “under God” is a religious symbol and not merely a secular practice.

Acknowledgements

We would like to thank Heena Shin, Leo Stroe, Rebecca Rogers, Ellen Meinholz, and Brennan Harris for help with formatting and coding the transcripts.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by a grant from the Wisconsin Alumni Research Foundation and the Hamel family.

Notes

Author Biographies

Lyn M. Van Swol is a Professor in the Communication Arts department at the University of Wisconsin-Madison. Her research interests include advice, deception, and information sharing in groups.

Michael T. Braun is a research specialist at the Children and Family Research Center at the University of Illinois at Urbana-Champaign. He evaluates statewide child protective services initiatives. His research focuses on family communication and communication technology use among at-risk families and families with children involved in the child welfare system.

Emily E. Acosta Lewis is an assistant professor of Communication and Media Studies at Sonoma State University. She received her PhD in Communication Arts and MA in Journalism and Mass Communication from University of Wisconsin-Madison and her BA in Communications (Public Relations) from San Diego State University. She is a faculty-in-residence for a freshman living learning community and oversees the student-run public relations firm on campus.

Cassandra L. Carlson is an Assistant Professor of Communication Studies at California Polytechnic State University in San Luis Obispo, CA. Her broader research explores communication approaches to decision-making, perception, and empathy, particularly in the contexts of interpersonal persuasion and social influence.

Giovanna Dimperio is a digital marketer working in the world of social media and strategy for Fortune 500 brands. She applies her research on decision-making and interpersonal persuasion in creating social media strategies and campaign execution for agencies, brands and retailers, and is currently employed by Lands’ End.

References

Balkwell

J. W.

Berger

(1996). Gender, status, and behavior in task situations. Social Psychology Quarterly, 59, 273-283. doi:10.1037/0033-295X.94.3.369

Bonito

J. A.

Hollingshead

A. B.

(1997). Participation in small groups. In Burleson

B. R.

(Ed.), Communication yearbook (Vol. 20, pp. 227-261). Newbury Park, CA: Sage.

Bonner

B. L.

Gonzalez

C. M.

Sommer

(2004). Centrality and accuracy in group quantity estimations. Group Dynamics, 8, 155-165. doi:10.1037/1089-2699.8.3.155

Boster

F. J.

Hunter

J. E.

Hale

J. L.

(1991). An information processing model of jury decision-making. Communication Research, 18, 524-547. doi: 10.1177/009365091018004004

Boster

F. J.

Mayer

M. E.

(1984). Choice shifts: Argument qualities and social comparisons. Communication yearbook, 8, 393-410. doi: 10.1080/23808985.1984.11678583

Boster

F. J.

Mayer

M. E.

Hunter

J. E.

Hale

J. L.

(1980). Expanding the persuasive arguments explanation of the polarity shifts: A linear discrepancy model. Communication yearbook, 4, 165-176.

Bottger

P. C.

(1984). Expertise and air time as bases of actual and perceived influence in problem-solving groups. Journal of Applied Psychology, 69, 214-221. doi:10.1037/0021-9010.69.2.214

Braun

M. T.

(2012). Social influence in groups. Unpublished manuscript, University of Wisconsin-Madison.

Brodbeck

F. C.

Kerschreiter

Mojzisch

Schulz-Hardt

(2007). Group decision-making under conditions of distributed knowledge: The information asymmetries model. Academy of Management Review, 32, 459-479. doi:10.5465/AMR.2007.24351441

10.

Davis

J. H.

Kerr

Sussmann

Rissman

A. K.

(1974). Social decision schemes under risk. Journal of Personality and Social Psychology, 30, 248-271. doi:10.1037/h0036617

11.

Davis

J. H.

Zarnoth

Hulbert

Chen

X. P.

Parks

Nam

(1997). The committee charge, framing interpersonal agreement, and consensus models of group quantitative judgment. Organizational Behavior and Human Decision Processes, 72, 137-157. doi: 10.1006/obhd.1997.2733

12.

Eagly

A. H.

Chaiken

(1993). The psychology of attitudes. Forth Worth, TX: Harcourt Brace.

13.

Enders

C. K.

Tofighi

(2007). Centering predictor variables in cross-sectional multilevel models: A new look at an old issue. Psychological Methods, 12, 121-138. doi:10.1037/1082-989X.12.2.121

14.

Garlick

Mongeau

P. A.

(1993). Argument quality and group member status as determinants of attitudinal minority influence. Western Journal of Communication, 57, 289-308. doi: 10.1080/10570319309374455

15.

Gigone

Hastie

(1993). The common knowledge effect: Information sharing and group judgment. Journal of Personality and Social Psychology, 65, 959-974. doi: 10.1037/0022-3514.65.5.959

16.

Guetzkow

(1950). Unitizing and categorizing problems in coding qualitative data. Journal of Clinical Psychology, 6, 47-58. doi:10.1002/1097-4679

17.

Henningsen

D. D.

Henningsen

M. L. M.

(2015). The role of expertise and expressiveness in decision-making groups. Communication Quarterly, 63, 463-479. doi: 10.1080/01463373.2015.1058284

18.

Hovland

C. I.

Sherif

(1980). Social judgment: Assimilation and contrast effects in communication and attitude change. Westport, CT: Greenwood.

19.

Judd

C. M.

Brauer

(1995). Repetition and attitude extremity. In Petty

R. E.

Krosnick

J. A.

(Eds.), Attitude strength: Antecedents and consequences (pp. 43-72). Mahwah, NJ: Lawrence Erlbaum.

20.

Kameda

Ohtsubo

Takezawa

(1997). Centrality in sociocognitive networks and social influence: An illustration in a group decision-making context. Journal of Personality and Social Psychology, 73, 296-309. doi:10.1037/0022-3514.73.2.296

21.

Krosnick

J. A.

Boninger

D. S.

Chuang

Y. C.

Berent

M. K.

Carnot

C. G.

(1993). Attitude strength: One construct of many related constructs? Journal of Personality and Social Psychology, 65, 1132-1151. doi:10.1037/0022-3514.65.6.1132

22.

Laughlin

P. R.

(2011). Group problem solving. Princeton, NJ: Princeton University Press.

23.

Laughlin

P. R.

Ellis

A. L.

(1986). Demonstrability and social combination processes on mathematical intellective tasks. Journal of Experimental Social Psychology, 22, 177-189. doi:10.1016/0022-1031(86)90022-3

24.

Levine

J. M.

Prislin

(2013). Majority and minority influence. In Levine

J. M.

(Ed.), Group processes (pp. 135-163). New York, NY: Psychology Press.

25.

Levine

J. M.

Russo

(1995). Impact of anticipated interaction on information acquisition. Social Cognition, 13, 293-317. doi:10.1521/soco.1995.13.3.293

26.

Maass

Clark

R. D.

Haberkorn

(1982). The effects of differential ascribed category membership and norms on minority influence. European Journal of Social Psychology, 12, 89-104. doi:10.1002/ejsp.2420120107

27.

Millar

M. G.

Tesser

(1986). Thought-induced attitude change: The effects of schema structure and commitment. Journal of Personality and Social Psychology, 51, 259-269. doi:10.1037/0022-3514.51.2.259

28.

Miller

C. E.

(1989). The social psychological effects of group decision rules. In Paulus

P. B.

(Ed.), Psychology of group influence (2nd ed., pp. 327-355). Hillsdale, NJ: Lawrence Erlbaum.

29.

Moscovici

(1980). Toward a theory of conversion behavior. In Berkowitz

(Ed.), Advances in experimental social psychology (Vol. 13, pp. 209-239). New York, NY: Academic Press.

30.

Ohtsubo

Masuchi

Nakanishi

(2002). Majority influence process in group judgement: Test of the social judgment scheme model in a group polarization context. Group Processes & Intergroup Relations, 5, 249-261. doi: 10.1177/1368430202005003005

31.

Prahl

Dexter

Braun

M. T.

Van Swol

L. M.

(2013). Review of experimental studies in social psychology of small groups related to operating room and anesthesia department decisions with best answers. Anesthesia & Analgesia, 117, 1221-1229. doi: 10.1213/ane.0b013e3182a0eed1

32.

Preacher

K. J.

Hayes

A. F.

(2004). SPSS and SAS procedures for estimating indirect effects in simple mediation models. Behavior Research Methods, Instruments, & Computers, 36, 717-731. doi:10.3758/BF03206553

33.

Rashotte

L. S.

Smith-Lovin

(1997). Who benefits from being bold? The interactive effects of task cues and status characteristics on influence in mock jury groups. Advances in Group Processes, 14, 235-255.

34.

Raudenbush

S. W.

Bryk

A. S.

(2002). Hierarchical linear models (2nd ed.). Thousand Oaks, CA: Sage.

35.

Sargis

E. G.

Larson

J. R.

Jr. (2002). Informational centrality and member participation during group decision making. Group Processes & Intergroup Relations, 5, 333-347. doi:10.1177/1368430202005004005

36.

Sherif

Hovland

C. I.

(1961). Social judgment: Assimilation and contrast effects in communication and attitude change. New Haven, CT: Yale University Press.

37.

Sherif

Nebergall

(1965). Attitude and attitude change: The social judgment-involvement approach. Philadelphia, PA: W. B. Saunders.

38.

Should the words “under God” be in the US Pledge of Allegiance? (n.d.). Retrieved from http://undergod.procon.org/viewresource.asp?resourceID=000062

39.

Smith

C. M.

Dykema-Engblade

Walker

Niven

T. S.

McGough

(2000). Asymmetrical social influence in freely interacting groups discussing the death penalty: A shared representations interpretation. Group Processes & Intergroup Relations, 3, 387-401. doi:10.1177/1368430200003004004

40.

Smith

C. M.

Tindale

R. S.

Dugoni

B. L.

(1996). Minority and majority influence in freely interacting groups: Qualitative versus quantitative differences. British Journal of Social Psychology, 35, 137-149. doi:10.1111/j.2044-8309.1996.tb01088.x

41.

Sniezek

J. A.

Henry

R. A.

(1989). Accuracy and confidence in group judgment. Organizational Behavior and Human Decision Processes, 43, 1-28. doi:10.1016/0749-5978(89)90055-1

42.

Sorrentino

R. M.

Boutillier

R. G.

(1975). The effect of quantity and quality of verbal interaction on ratings of leadership ability. Journal of Experimental Social Psychology, 11, 403-411. doi:10.1016/0022-1031(75)90044-X

43.

Stasser

Titus

(1985). Pooling of unshared information in group decision making: Biased information sampling during discussion. Journal of Personality and Social Psychology, 48, 1467-1478. doi:10.1037/0022-3514.48.6.1467

44.

Sunstein

(2003). Why societies need dissent. Cambridge, MA: Harvard University Press.

45.

Tindale

R. S.

Davis

J. H.

Vollrath

D. A.

Nagao

D. H.

Hinsz

V. B.

(1990). Asymmetrical social influence in freely interacting groups: A test of three models. Journal of Personality and Social Psychology, 58, 438-449. doi:10.1037/0022-3514.58.3.438

46.

Van Swol

L. M.

(2007). Perceived importance of information: The effects of mentioning of information, shared information bias, ownership bias, reiteration, and confirmation bias. Group Processes & Intergroup Relations, 10, 241-258. doi: 10.1177/1368430207074730

47.

Van Swol

L. M.

(2009). Extreme members and group polarization. Social Influence, 4, 185-199. doi:10.1080/15534510802584368

48.

Van Swol

L. M.

Seinfeld

(2006). Differences between minority, majority, and unanimous group members in the communication of information. Human Communication Research, 32, 178-197. doi:10.1111/j.1468-2958.2006.00008.x

49.

Wittenbaum

G. M.

Bowman

J. M.

(2004). A social validation explanation for mutual enhancement. Journal of Experimental Social Psychology, 40, 169-184. doi:10.1016/S0022-1031(03)00091-X

50.

Wittenbaum

G. M.

Hollingshead

A. B.

Botero

I. C.

(2004). From cooperative to motivated information sharing in groups: Moving beyond the hidden profile paradigm. Communication Monographs, 7, 59-83. doi: 10.1080/0363452042000299894

51.

Wittenbaum

G. M.

Hubbell

A. P.

Zuckerman

(1999). Mutual enhancement: Toward an understanding of the collective preference for shared information. Journal of Personality and Social Psychology, 77, 967-978. doi:10.1080/0363452042000299894

52.

Wood

Lundgren

Ouellette

J. A.

Busceme

Blackstone

(1994). Minority influence: A meta-analytic review of social influence processes. Psychological Bulletin, 115, 323-345. doi:10.1037/0033-2909.115.3.323

53.

Yuan

Y. C.

Bazarova

N. N.

Fulk

Zhang

Z. X.

(2013). Recognition of expertise and perceived influence in intercultural collaboration: A study of mixed American and Chinese group. Journal of Communication, 63, 476-497. doi:10.1111/jcom.12026