Abstract
Using the linguistic software Linguistic Inquiry Word Count, we analyzed transcripts of group discussions of whether the words “under God” should be in the Pledge of Allegiance. We hypothesized that members with an extreme opinion would use less complex language and more you pronouns than other members. Furthermore, extreme members would have less influence when they used you pronouns or more complex language consistent with the illusion of understanding. Extreme members were more confident and perceived themselves as more knowledgeable, but they did not use less complex language than other members. When extreme members did use complex language, they were less influential. Extreme members used more you pronouns and use of you pronouns reduced their influence in the group. Groups containing at least one extreme member had a much lower level of complexity in their discourse than groups without extreme members. Results are situated within research in integrative complexity, illusion of understanding, and attitude extremity.
Considerable research has been devoted to understanding how extremity of attitudes, especially attitudes based on political ideologies, affect behavior and reflect beliefs. Research has consistently supported the premise that individuals with politically extreme attitudes tend to be more dogmatic, have more simplistic causal models, and have less integrative complexity when evaluating alternative viewpoints (Fernbach, Rogers, Fox, & Sloman, 2013; Tetlock, 1983; Toner, Leary, Asher, & Jongman-Sereno, 2013). While there are volumes of research on how groups can come to extreme decisions (e.g., Hinsz & Davis, 1984; Myers & Lamm, 1976; Sunstein, 2002), fewer researchers have examined how members with extreme attitudes behave in groups (Myers & Murdoch, 1972; Van Swol, 2009) or the effects of an extreme individual on the group as a whole. This study aims to address that question by looking at language use of extreme members during a group discussion and the influence of extreme members on the language used in the group. Language is an important marker of social cognition in groups (Hinsz, Tindale, & Vollrath, 1997; Semin, 2000) and can contribute to understanding social thought and behavior in groups. It can reflect at a granular level emotional states, intrapersonal processes, and interpersonal processes (Chung & Pennebaker, 2007; Holtgraves & Kashima, 2008). Language can both reflect these processes and influence them. Holtgraves and Kashima (2008, p. 73) state, “speakers’ language use can inadvertently affect the thoughts and actions of the speakers themselves, the recipients, and the collective that include both.” To examine extreme members and language use, we first summarize research on how extreme attitudes affect behavior and beliefs. Second, we examine the research on ideological extremity and integrative complexity. Finally, we relate this research to hypotheses about language use and extremity.
Extreme Attitudes
Extreme attitudes tend to be more accessible in memory (Eagly & Chaiken, 1993; Fazio, Chen, McDonel, & Sherman, 1982; Fazio & Williams, 1986; Lavine, Borgida, & Sullivan, 2000) and, consequently, more likely to affect and predict behavior than less extreme attitudes (Bright, 1997; Eagly & Chaiken, 1993; Fazio & Zanna, 1978; Petersen & Dutton, 1975; Petty, Haugtvedt, & Smith, 1995). Furthermore, there are a constellation of related variables where there are differences between people with extreme and nonextreme attitudes. People with extreme attitudes often view the topic as more important, are more committed to the issue, and are more resistant to change than people with less extreme attitudes (Eagly & Chaiken, 1993; Ewing, 1942; Judd & Brauer, 1995; Krosnick, Boninger, Chuang, Berent, & Carnot, 1993; Miller & Tesser, 1986; Osgood & Tannenbaum, 1955; Sarat & Vidmar, 1976; Van Swol, 2009). Van Swol (2009) found that members with extreme attitudes talked more and took more speaking turns during the group discussion. Van Hiel and Mervielde (2003) found that individuals with more extreme attitudes tend to have greater political interest. Yet despite this greater amount of participation in groups, higher perception of issue importance, and resistance to change, research on political issues has found that people who have extreme attitudes about an issue have not necessarily built these attitudes on a sophisticated and nuanced understanding of the attitudinal issue.
Extremity and Ideology
Researchers have explored how extremity of political ideology affects differences among individuals (e.g., Knutson, 1973; Levinson, 1968) and have primarily focused on less interactive communication from one individual such as speeches from politicians (e.g., Tetlock, 1983), online discussion boards or blogs (e.g., Anwar & Abulaish, 2015), or survey responses (e.g., Van Hiel, 2012). There is a need to understand how ideologically extreme members behave in a group discussion. Current research has supported the view that dependent on the issue at hand, extreme individuals at either end of the political spectrum can exhibit rigidity, belief superiority, and dogmatism (Toner et al., 2013; see also, Tetlock, 1984, 1986). For example, Toner et al. (2013) found that liberals had greater belief superiority on government aid and basing laws on religion while conservatives exhibited greater belief superiority on taxes and affirmative action. This supports the idea that those with extreme viewpoints on an issue are more inclined to hold their beliefs as superior for that issue.
There has also been considerable attention paid to the integrative complexity of arguments by those with extreme and moderate attitudes (e.g., Tetlock, 1983, 1984). Integrative complexity refers to the degree to which an individual considers and recognizes multiple perspectives. There is evidence that extreme members on either end of the political spectrum have lower integrative complexity than people who hold more moderate views on an issue (Tetlock, 1984, 2005). For example, Tetlock (1984) interviewed members of British parliament and found that extreme conservatives and socialists did not different greatly from one another in integrative complexity, and extreme members had significantly less integratively complex thinking about the issue than more moderate members of parliament. Tetlock’s (2005) research with professional forecasters found that forecasters he categorized as hedgehogs tended to be both more extreme and less integratively complex, regardless of ideological persuasion. These findings suggest that people who hold politically extreme views think about the issue in more black-and-white terms than their moderate counterparts. Research has found that people reach more extreme evaluations of people or social groups when they evaluate them on fewer dimensions (Linville, 1982), suggesting that lack of integrative complexity may drive extremity of attitudes. However, adopting an extreme view could also facilitate less integrative complexity as one selectively seeks out less nuanced information to support that extreme view (Hart et al., 2009).
Thus, individuals with politically extreme opinions are likely to believe their attitudes are superior, but have less integrative complexity in their thinking and beliefs. Fernbach et al. (2013) suggest that this lack of nuanced understanding of complex political issues is what underlies extremity of attitudes in those issues and overconfidence in the superiority of one’s understanding of the issue. They found that asking people with extreme attitudes to explain the causal mechanisms underlying the issue caused them to moderate their attitude on discovering that they understood the issue much less than they had supposed. Thus, when people evaluate an issue with less complex thinking, they are likely to reach more extreme attitudes toward the issue because their thinking lacks nuance and differentiation about the issue’s complexity. The belief superiority and lower complexity of thinking of individuals with extreme attitudes is likely to manifest itself in the language that they use in a group discussion and could influence the language used by the group as a whole.
Extremity and Language Use
We expect that individuals with extreme attitudes will use linguistic categories differently than individuals without extreme attitudes, and this could affect how much extreme members can exert influence on the group decision. Language communicates social cognition and also helps express meaning during group interaction. Previously much of the focus on groups and language has been relating language use and group performance, status, and group identity (e.g., Gonzales, Hancock, & Pennebaker, 2010; Kane & Rink, 2015; Van Swol & Carlson, 2015), and there has been little regarding how extreme members use language.
Researchers have identified several linguistic categories that can indicate depth and complexity of thought. These categories can indicate that the speaker is connecting different thoughts or actively evaluating an idea or event. Cognitive mechanism words reflect inference, cause, effect, and order (e.g., “cause,” “know,” “because,” “without”) and indicate more complex use of language, rational argumentation, and causal thinking (Gelfand et al., 2015; Pennebaker, Booth, & Francis, 2007; Pennebaker & King, 1999; Pennebaker, Mehl, & Niederhoffer, 2003; Tausczik & Pennebaker, 2010). For example, Gelfand et al. (2015) hypothesized that negotiators in the United States would use rational persuasion, reliance on facts, and logic in their negotiation strategies more than Egyptian negotiators, who would use more honor-based appeals. In support of this, Gelfand et al. (2015) found that U.S. negotiators used more cognitive mechanism language than Egyptian negotiators. Within cognitive mechanism words, exclusion words (e.g., “without,” “exclude”) are especially useful for determining when the speaker is making distinctions and insight words (e.g., “know,” “consider”) can signal that the speaker is engaging in a process of reappraisal (Tausczik & Pennebaker, 2010). In addition, conjunctions (e.g., “and,” “also”) link multiple ideas and thoughts together (Pennebaker, 2011). Tausczik and Pennebaker (2010) suggest that conjunctions and exclusion words are especially linked to integrative complexity because they may indicate differentiation between competing ideas and integration among ideas. Two other indicators of complexity are prepositions (e.g., “to,” “above”) and words per sentence (Tausczik & Pennebaker, 2010). For example, prepositions are often used more frequently in more abstract and complex articles (Hartley, Pennebaker, & Fox, 2003). Based on research on integrative complexity and extreme attitudes (Fernbach et al., 2013; Tetlock, 1984), during group discussion, we expect more extreme participants to be less cognitively complex in their language. As Tetlock (2005, p. 85) states, “It is hard to build up a lot of momentum for extreme predictions if one is slowed down by lots of buts and howevers.” Formally stated:
Fernbach et al. (2013) found that extreme members often had trouble explaining the causal mechanisms and reasoning behind the political issues for which they held extreme views, and as a result of this decreased sense of their own understanding, they often moderated their opinion and confidence after the attempt. This suggests that if extreme members do try to use more logical reasoning based on facts and complex language, they may be unsuccessful because it reveals that they have a more limited understanding of the issue than their confidence warrants. Therefore, we hypothesize that extreme members who use more indicators of complex language will have less influence in the group than extreme members who use fewer indicators of complex language. Furthermore, following from Fernbach et al. (2013), use of complex language should reduce extreme members’ confidence in their opinion in comparison with their prediscussion confidence. However, we do not hypothesize either of these relationships for other group members.
Furthermore, Holtgraves and Kashima (2008) note that people can influence the language used by others with whom they communicate. If extreme members are likely to exhibit dogmatism and attitude entrenchment (Toner et al., 2013), they may lower the level of discourse in the group as a whole, especially since we expect extreme members to dominate the discussion (Van Swol, 2009). Given that extreme members are expected to talk the most in the group, this could increase their level of language style matching (LSM). LSM (Gonzales et al., 2010) measures verbal mimicry on commonly used functions words like pronouns, conjunctions, and prepositions. Group members should mimic the language of members who talk more, and this could reduce the complexity of language used in groups with extreme members.
One linguistic category that may hold potential insight into cognitive processes and group dynamics are function words like pronouns. Function words—such as pronouns and articles—have been shown to reflect emotional states and features of social relationships (Chung & Pennebaker, 2007). Furthermore, since they are often used without much awareness, they are difficult for the speaker to control (Pennebaker, 2011). Pronouns are the most commonly used function word and account for about 14% of the words that people use while speaking (Pennebaker, Chung, Ireland, Gonzales, & Booth, 2007). Pronouns are also of importance because they mark a relationship or connection between the speaker and listener and can be indicative of attentional focus and status (Brown & Levinson, 1987; Kacewicz, Pennebaker, Davis, Jeon, & Graesser, 2014; Van Swol & Carlson, 2015). Excessive you statements can signal blaming others and distancing and differentiation (Kane & Rink, 2015; Simmons, Gordon, & Chambless, 2008). Use of you pronouns has been linked to more use of criticism and increased confrontation (Simmons et al., 2008). In marital interactions, having a we focus was associated with a more positive solution to problems, while there was a negative role of you (Simmons, Gordon, & Chambless, 2005). Pennebaker (2011, p. 175) calls you pronouns “the equivalent of pointing your finger.”
Extreme members are expected to enter the group discussion with greater confidence, higher perception of the issue’s importance, and more perceived knowledge than the less extreme members. Therefore, they are likely to want to showcase the perceived superiority of their view to the group. This could increase their use of you pronouns as they focus their attention on persuading other members and pointing out the superiority of their position; in other words, extreme members may have more finger-pointing when discussing their views.
However, extreme member’s perceived superiority of their view is not likely to be shared by other less extreme group members. By using you statements, extreme members deemphasize themselves and show a focus on other discussants by verbally “pointing a finger” at them. This may cause extreme members to be viewed more negatively by other group members, therefore hurting the extreme member’s ability to exert influence in the group. This effect is supported by past research as the use of you pronouns is associated with less influence in martial interactions (Simmons et al., 2005) and less influence for majority members in group discussions (Van Swol & Carlson, 2015). We expect to see a similar negative impact of use of you pronouns on extreme members’ influence on the group decision.
Finally, we pose a research question. We investigate whether there are more negative emotion words used in groups with extreme members. Although we do not have a strong theoretical basis, the presence of an extreme member may increase the level of negative emotion in the group. Therefore, we ask:
Method 1
Participants
Undergraduates at a large, public Midwestern university in the United States (males = 63, females = 167) participated in the study for extra credit in a communication course. 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). Same gender groups were used to avoid status expectations linked with gender that may affect participation rates (Balkwell & Berger, 1996).
Procedure
In pilot testing, eight controversial issues were pretested with 100 students enrolled in communication classes. Whether the words “under God” should be in the Pledge of Allegiance was selected as the topic with the strongest range of opinions and strongest feelings about the issue.
Participants were given informed consent on arrival and then asked their country of citizenship. Non-U.S. citizens were thanked and excused from this experiment. Participants were informed that the study examined how people discuss controversial issues in groups and were given a one-page sheet with information about whether the words “under God” should be in the Pledge of Allegiance. The information was from the nonpartisan website procon.org (http://undergod.procon.org), and participants were informed of this (see the appendix). The information in the one-page sheet reflected arguments on both sides of the issue. In addition, participants received a copy of the pledge and a 3-foot by 5-foot U.S. flag was hanging on the laboratory wall to highlight the salience of the issue. 2
After reading the one-page information sheet, participants were asked to rate their opinion on the issue on a scale 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 US Pledge of Allegiance). Participants also rated their confidence in the correctness of their views, issue importance, and knowledge of the issue on a 5-point scale from 1 (not at all confident, important, knowledgeable) to 3 (somewhat . . . ) to 5 (very . . . ).
Rating sheets were then collected, and participants brought into a room with their group. Every group member was assigned a spot at a rectangle table with a large visible letter to help identify that member for transcribing the group discussion. Group members were instructed: 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 answering any questions, the experimenter put a decision sheet in the table center and instructed the group to reach a consensual decision about the issue. The experimenter then left the room. The group decision and level of confidence was measured on the same scale used to measure individual opinion and confidence. Group discussion was recorded with a consumer grade digital camera in full view of the participants.
After the group came to a decision, the participants in the group were asked to recite the pledge according to the decision of the group (either with or without the words “under God”). Participants had been told before the discussion that this would be part of the postdiscussion procedure. Participants did not have to recite the Pledge of Allegiance. Thirteen participants in nine groups chose to either say the Pledge differently than their group or to not recite it at all.
Participants then filled out a questionnaire in which they stated their postdiscussion opinion and confidence in this opinion. After completion, they were debriefed, thanked, and dismissed. Group discussions were transcribed verbatim. Two groups were unable to be transcribed due to recording difficulties. Each transcript was edited (e.g., correcting misspellings or inserting codes for nonfluencies) by the second author to create a transcript for analysis within Linguistic Inquiry Word Count (LIWC) software. Furthermore, every group transcript was separated into a separate transcript for each group member. LIWC software (Pennebaker et al., 2007) has over 70 linguistic categories for classification of words for analysis. LIWC scores the percentage of words in each category that are in the transcript.
Results
First, we present differences between extreme group members and other group members in language use. Second, we present how the language used by extreme members affects their influence in the group. Finally, we examine the presence of extremity at the group level and examine how groups that contain members with extreme opinions differ from groups without extremity. Because individuals are nested within groups, most of the analysis uses hierarchical linear modeling (HLM7 software) to account for nonindependence in the observations. Level 1 variables are individual-level variables of group members that are nested within groups, and Level 2 variables are variables at the group level (for review, see Raudenbush & Bryk, 2002). Variables from LIWC output are nonstandardized and provide percentage of words from the category, except for word count which gives raw word count.
Extreme Members
We examined differences in how members rated the importance of the issue, confidence in opinion, and knowledge about the issue. We classified members into four groups based on their position: most extreme (rating of 1 or 9 on opinion scale), moderately extreme (8 or 2), somewhat moderate (7 or 3), and moderate (4, 5, or 6). Group members filled out these scales before the group discussion, so we used analysis of variance to examine individual differences. In support of Hypothesis 6, members with more extreme views rated the issue as significantly more important, F(1, 225) = 7.59, η2 = 0.09, p< .001. Pairwise comparisons found that the most extreme members rated the topic as significantly more important than other members (p< .05). Members with more extreme opinions rated their confidence in their opinion significantly higher, F(1, 224) = 33.80, η2 = 0.31, p< .001. Pairwise comparisons found that the most extreme members had higher confidence than other members (p< .001). Confidence and extremity significantly correlated (r = .44, p< .001). More extreme members rated themselves as more knowledgeable, F(1, 225) = 10.11, η2 = 0.12, p< .001. Pairwise comparisons found that the most extreme members and moderately extreme members rated themselves as more knowledgeable than less extreme members (p< .05), but the most extreme and moderately extreme members did not differ on their ratings of knowledge (p = .69).
The average opinion discrepancy from other members’ prediscussion opinions was calculated for every member. For example, if members in a four-person group had prediscussion opinions 3, 4, 5, and 8, the member with the opinion rating of 8 would have an average discrepancy score ((5 + 4 + 3)/3) of 4.00. Members with more extreme opinions had significantly higher discrepancy scores, F(1, 225) = 24.95, η2 = 0.25, p< .001. Pairwise comparisons found that the most extreme members had significantly higher discrepancy scores (p< .001) than other members. More extreme members also leaned more on average toward keeping the words “under God” in the pledge, F(1, 225) = 3.99, η2 = 0.05, p< .01. Pairwise comparisons found that the most extreme members differed significantly from moderate and somewhat moderate members (p< .05), but the most extreme members did not significantly differ from moderately extreme members (p = .48). Moderately extreme members only significantly differed from moderate members (rating = 4, 5, or 6; p< .05). See Tables 1 and 2, for means. See Figure 1, for a histogram of member positions.
Means and Standard Deviations of Most Extreme Group Members and Other Members.
Means and Standard Deviations of Four Groups of Group Members.
Distribution of prediscussion opinions.
Individuals are nested within groups, violating the assumption of independence between observations; thus, hierarchical linear modeling was used to analyze group members at the individual level of use of several language categories. A dummy variable of extremity was created for the most extreme (1) and other group members (0). The most extreme members were operationalized as those members who used the extreme ends of the scale when rating their opinion on the issue, that is, used a rating of 1 or 9. A binary interpretation of extremity may not tell the whole story, as participants at an 8 or 2 on initial opinion may not be extreme, but still very strongly opinionated. Thus, we created a variable to test the quadratic effect of extremity to examine extremity on a continuum (referred to as quadratic extremity; Box, 1954). All HLM analysis included group size as a variable to control for differences in the size of the group. All nondummy individual-level predictor variables are group mean centered and all group-level predictor variables are grand mean centered in accordance with recommendations for hierarchical linear modeling (Enders & Tofighi, 2007).
In support of previous research, the most extreme members had a higher word count than other members, γ10 = 139.37, standard error (SE) = 57.03, t(166) = 2.44, p = .02. The statistic 139.37 represents the average regression slope across groups; it is the average effect of a person being extreme on word count compared with the average value for that group. Using the quadratic extremity variable also found an effect of more extreme members having a higher word count, γ10 = 8.44, SE = 4.00, t(166) = 2.11, p = .04. In support of Hypothesis 7, the most extreme members used a higher percentage you pronouns than other members, γ10 = 0.70, SE = 0.23, t(166) = 3.04, p = .003. Using the quadratic extremity variable, members with more extreme opinions used more you pronouns, γ10 = 0.03, SE = 0.02, t(166) = 2.02, p = .042. We hypothesized (Hypothesis 1) that extreme members would have less complex language. We examined use of cognitive mechanisms words, γ10 = −0.06, SE = 0.65, t(166) = −0.10, p = .92; words per sentence, γ10 = −1.92, SE = 0.91, t(166) = −2.12, p = .035; prepositions, γ10 = 0.01, SE = 0.01, t(166) = 1.31, p = .19; and conjunctions, γ10 = 0.13, SE = 0.35, t(166) = 0.36, p = .72. Somewhat similar results were found using the quadratic extremity variable for cognitive mechanisms, γ10 = 0.00, SE = 0.05, t(166) = 0.08, p = .93; words per sentence, γ10 = −0.08, SE = 0.07, t(166) = −1.05, p = .30; prepositions, γ10 = −0.04, SE = 0.02, t(166) = −1.80, p = .07; and conjunctions, γ10 = −0.01, SE = 0.02, t(166) = −0.39, p = .70. Therefore, the most extreme members used fewer words per sentence. However, no other differences were found for use of linguistic indicators of complexity for member extremity. Therefore, Hypothesis 1 was largely unsupported.
Hypothesis 5 tested LSM. LSM scores (Gonzales et al., 2010) were created for nine linguistic categories of function words: auxiliary verbs (e.g., am, will, have), articles (e.g., the, a), adverbs (e.g., very, really, quickly), personal pronouns (e.g., I, them, her), impersonal pronouns (e.g., it, those), prepositions (e.g., with, above), negations (e.g., no, not, never), conjunctions (e.g., but, whereas), and quantifiers (e.g., few, many, much). Percentage of the words mentioned from each category is provided by LIWC. Using Gonzales et al.’s (2010) equation, for each group member, an LSM score was created that compares the member with the rest of the group members. For example, LSM for auxiliary verbs would be calculated for the three group members:
aux1 = 1 − (|aux1 − auxG|/(aux1 + auxG)),
aux2 = 1 − (|aux2 − auxG|/(aux2 + auxG)),
aux3 = 1 − (|aux3 − auxG|/(aux3 + auxG)),
auxG is the mean percentage of auxiliary verbs for the other members in the group. Using this equation, an LSM score was created for each member for each of the nine categories. The nine LSM scores were averaged for each member to create an overall LSM for every member. LSM scores run from 1 (perfect LSM) to 0 (no LSM). In support of Hypothesis 5, the most extreme members matched the language of other group members more than nonextreme members, γ10 = 0.02, SE = 0.01, t(166) = 2.48, p = .01. Furthermore, the quadratic extremity variable found that more extreme members had more LSM, γ10 = 0.00, SE = 0.00, t(166) = 2.29, p = .02. LSM and word count significantly correlated (r = .48, p < .0001), suggesting that extreme members had a higher LSM because they talked the most. See Tables 1 and 2, for means.
Influence
Hypothesis 8 predicted that members who used more you pronouns would have less influence, and that this effect would be especially strong for extreme members. Within HLM, we used percentage of you pronouns used, member (extreme, not extreme), and the interaction terms between you and member to predict influence in the group. Group size was included in all analyses as a control. We used comparative influence to measure influence (Van Swol, Braun, Acosta Lewis, Carlson, & Dimperio, 2015). Comparative influence is an absolute measure of a participant’s opinion before the group discussion subtracting the group’s consensual opinion (|pregroup individual opinion − group opinion|). The group consensual opinion is the opinion that members reached as a group through discussion. Then, the average absolute difference of the other participants from the group opinion is subtracted from this. For example, if a participant had a pregroup opinion of 7, the group’s consensual opinion was 6, and the mean absolute difference from the group opinion for the other participants is 3; then the comparative influence score of this participant is (|7 − 6| − 3) = −2. This measure helps determine how close a member’s prediscussion opinion is to the opinion reached by discussion by the group members in comparison with how close other member’s pregroup opinions are to the group consensual opinion. Larger negative numbers designate that a participant’s pregroup opinion is closer to the group consensual opinion than other participants; this would indicate more influence. Larger positive numbers indicate that the participant’s opinion was further from the group consensual opinion than the other participants; this would indicate less influence.
First, we ran a hierarchical null model with no predictors to decompose the variance between levels (Raudenbush & Bryk, 2002). The null model suggested that there was no significant between-group variance in influence without predictors in the model. For the hypothesized model, there was a significant interaction between use of you pronouns and type of member. In support of Hypothesis 8, extreme members had less influence when they used a higher percentage of you pronouns; whereas this relationship did not hold for other members. See Table 3.
Hierarchical Linear Modeling Predicting Influence From You Pronouns and Type of Member.
Note. SE = standard error.
p< .05. **p< .01. ***p< .001.
Next, using a similar analysis with cognitive mechanism words and member type as predictors of influence, we examined if there was a relationship between use of cognitive mechanisms words and less influence for extreme members. There was a significant interaction between use of cognitive mechanisms words and type of member, γ30 = 0.08, SE = 0.02, t(49) = 3.70, p< .001. Extreme members had less influence when they used a higher percentage of cognitive mechanism words, but this relationship did not hold for the other members. The same relationship held for words per sentence, use of conjunctions, and use of prepositions, γ30 = 0.11, SE = 0.03, t(49) = 3.37, p< .001; γ30 = 0.20, SE = 0.06, t(49) = 3.58, p< .001; γ30 = 0.17, SE = 0.05, t(49) = 3.63, p< .001, respectively. In support of Hypothesis 2, when extreme members used more indicators of language complexity (cognitive mechanism words, conjunctions, prepositions, and words per sentence), they had less influence in the group, whereas the relationship was not significant for nonextreme members.
We predicted (Hypothesis 3) confidence would decrease with more use of complex language for extreme members, but not for other members. First, using a measure of postdiscussion confidence in opinion minus prediscussion confidence in opinion (confidence post − confidence pre), we found that extreme members (M = −0.05, SD = 0.57) had reductions in confidence after discussion, while other members (M = 0.37, SD = 0.95) had increases in confidence, γ10 = 0.46, SE = 0.17, t(172) = 2.78, p = .006. Next, we examined whether there was an interaction in the relationship between change in confidence and use of cognitive mechanisms words for type of member, γ30 = −2.10, SE = 0.96, t(49) = 2.18, p = .034. For extreme members, use of cognitive mechanism words was related to decreases in confidence after group discussion, but the opposite was found for other members. There was a marginal interaction in the relationship between change in confidence and type of member for use of conjunctions, γ30 = −1.03, SE = 0.52, t(49) = 1.97, p = .055. For extreme members, use of conjunctions was related to decreases in confidence after the group, but for other members, use of conjunctions was related to increases in confidence. However, there was not a significant interaction in the relationship between change in confidence and member for words per sentence, γ30 = −1.19, SE = 3.02, t(49) = 0.40, p = .69; or a significant interaction for use of prepositions, γ30 = 0.64, SE = 0.56, t(49) = −1.15, p = .26. Therefore, extreme members had significantly reduced confidence after the discussion in comparison with other members, and this reduction in confidence is related to more use of cognitive mechanism words and conjunctions, but not words per sentence or prepositions. Therefore, there was some support for Hypothesis 3. Additional analysis on just the extreme members found that the negative relationship between use of cognitive mechanism words and reductions in confidence for extreme members held only in groups where the extreme members encountered dissent (at least one group member with a viewpoint on the other side of the midpoint), r = −.39, p = .034 (n = 30), and not in groups without dissent (n = 6), r = .55, p = .26.
Finally, we examined the interaction between LSM and type of group member on influence in the group. There was a significant interaction, γ30 = 2.01, SE = 0.52, t(49) = 3.88, p< .001. The most extreme members had less influence when they matched the language of the other group members.
Presence of Extreme Members in Groups
We examined the presence of a member with the most extreme opinion (1 or 9) in the group. First, we examined if nonextreme group members in groups that contained at least one extreme member used less complex language than members in groups without an extreme member. We compared percentage of cognitive mechanism words, γ01 = −2.07, SE = 0.61, t(48) = −3.39, p< .001; words per sentence, γ01 = −2.87, SE = 3.19, t(48) = −0.90, p = .37; prepositions, γ01 = 0.45, SE = 0.32, t(48) = 1.41, p = .14; and conjunctions, γ01 = −0.91, SE = 0.31, t(48) = −2.97, p = .005. Hypothesis 4 was largely supported; members in groups with extreme members used fewer linguistic indicators of complexity. 3
To answer the research question, there were no differences in negative and positive emotions between members in groups with no extreme members and nonextreme members in groups with an extreme member, γ01 = −0.14, SE = 0.11, t(48) = −1.32, p = .19, and γ01 = 0.22, SE = 0.25, t(48) = 0.85, p = .40, respectively. Although not hypothesized, groups that contained at least one extreme member had a higher percentage of nonfluencies, γ01 = 0.22, SE = 0.10, t(48) = 2.29, p = .026. Groups with extreme members did not have significantly longer discussions as measured by word count, γ01 = 47.86, SE = 55.72, t(48) = 0.86, p = .40. See Table 4, for means.
Means and Standard Deviations of Groups With and Without Extreme Members.
Discussion
In summary, group members with extreme opinions were more confident, viewed the issue as more important, perceived themselves as more knowledgeable, and talked more during the group discussion. They also used a higher percentage of you pronouns than less extreme members, and they were less influential when they used more you pronouns. However, with the exception of using fewer words per sentence than other members, extreme members did not have less complex language than other members. However, when extreme members used more complex language, they were less influential. Use of some indicators of complex language (cognitive mechanism and conjunctions) was related to reduced confidence after the group discussion for extreme members, but the opposite was found for other members. Extreme members had higher LSM scores, probably as a result of dominating the discussion. However, LSM did not increase their influence in the group, as it did for other members. Finally, group members in a group that contained a member with an extreme opinion had less complex language than groups without extreme members. The results are situated within theory below.
Extreme Members and Language
Extreme members were more confident, perceived themselves as more knowledgeable, perceived the issue as more important, and talked more than other members. This confirms past research about extremity and belief superiority (Fernbach et al., 2013; Tetlock, 2005) and level of participation (Van Swol, 2009). It also suggests that group members with extreme opinions hold their attitudes and behave in fundamentally different ways than other group members.
However, extreme group members, with the exception of words per sentence, did not use significantly less complex language than less extreme members. Some previous research on integrative complexity and extremity has used politicians and professional forecasters (Tetlock, 1984, 2005), and our use of undergraduates may not have produced the extremity or knowledge about the issue to find differences in complexity of language. Although, Tetlock (2005, p. 76) notes that, “the research literature offers no precise guidance on how far to the left or right one must go to observe hypothesized shifts in cognitive style,” so we do not have guidelines to determine if our participants were extreme enough. Yet the questionnaire data suggest they were different in confidence and other dimensions. We did provide our undergraduate participants with an informational sheet before the discussion, and this could have helped equalize the sophistication of language among participants.
Integrative complexity as conceived in this study is specific to the issue held by the participant. Lack of integrative complexity about a specific issue can be an underlying factor in forming extreme attitudes or extreme attitudes may lead to people seeking less nuanced and complex information to support their extreme attitude. However, this study did not have an independent assessment of integrative complexity of attitudes before the group discussion. This would have provided a stronger test of our hypothesis. Future research should try to replicate our results with an independent assessment of integrative complexity outside of the group discussion. One option is to use Fernbach et al.’s (2013) method of having participants explain the mechanism underlying their attitude.
Despite extreme members not using less complex language, we found that other group members used less complex language when there was an extreme member in the group than when there was not an extreme member. We had hypothesized that this would result from extreme members talking more and having higher LSM scores—which extreme members did. However, given that extreme members did not speak less complexly, LSM is not a satisfactory explanation. Something else was going on in the group that lowered the complexity of discourse for other members in the presence of extreme opinions.
One explanation may stem from research on group dynamics. Extreme members’ use of you pronouns may be indicative of a more confrontational, negative group communication dynamic (Yilmaz, 2014). This may be related to why more nonfluencies were used in groups with an extreme member. Research has found that group members are more cooperative and engage in more information sharing within a positive communication environment (Huffaker, Swaab, & Diermeier, 2011). With a more negative environment, members may share less information and offer more simplistic arguments rather than try to engage in a nuanced debate focusing on more differentiation and integration of complex information. Group members may need more psychological safety in order to open their positions up to nuance and differentiation (Burkhalter, Gastil, & Kelshaw, 2002; Edmondson & Roloff, 2008) and without that psychological safety, they may offer a less complex one-dimensional position to the group to avoid delving into a back and forth. More moderate participants may have felt the need to counter an extreme opinion with a clear and unqualified opinion of their own. More research is needed to understand the mechanism that lowers the level of discourse in the group in the presence of extremity.
In support of research by Fernbach et al. (2013), when extreme members used more complex language, they were less influential and had reductions in confidence. Fernbach et al. (2013) found that asking people with extreme attitudes to explain the causal reasoning behind their position caused them to moderate their opinion and reduce their confidence on discovering their lowered sense of understanding. This illusion of understanding may have been the underlying factor in our results. As extreme members used more complex language, they may have been more aware of their lack of understanding of the issue, especially in light of their initially high confidence and perception of knowledge before the group discussion. To paraphrase Tetlock (2005, p. 85), it is hard to defend extreme positions if one is weighted down “by lots of buts and howevers.” However, we cannot be sure how much of the change is due to an internal realization that they do not understand the issue or external push back from other members challenging their reasoning about the issue. Although the number of homogenous groups was small, we found a relationship between decreases in confidence and increased use of cognitive mechanism words only in groups that had dissent. Therefore, dissent should be examined in future research as a moderating variable in a data set with a larger number of homogeneous groups.
Generally, confident group members are more influential in groups (Sniezek, 1992; Sniezek & Van Swol, 2001; Zarnoth & Sniezek, 1997). Furthermore, most group members have increased confidence as a result of group discussion (Heath & Gonzalez, 1995). However, to our knowledge, research in groups has not examined how influential an initially confident member who decreases their confidence is in the group. Extreme members seem to be losing confidence as the discussion progressed, which could have reduced their influence in the group. Alternatively, possibly the extreme member’s lack of influence may have reduced their confidence. Often, there is a relationship, although weak to moderate and prone to overconfidence, between confidence and expertise (Sniezek & Van Swol, 2001; Van Swol & Sniezek, 2005). However, this relationship is weaker for extreme members who have higher confidence, but not necessarily greater understanding (Fernbach et al., 2013; Tetlock, 2005). This may have made high confidence a less effective influence mechanism.
Although this article examined language with LIWC analysis, the results have implications for work on sacred rhetoric and sacred values (Dehghani, Sagae, Sachdeva, & Gratch, 2014; Marietta, 2008, 2009). Members with extreme opinions may have more nonnegotiable convictions that are less grounded on reasoned consequences and have more arguments premised on morality. In other words, they may have more sacred values and as a result premise their arguments more on sacred rhetoric. Sacred rhetoric often is not effective for persuasion, but it can change the tone and intensity of discourse. Hence, if extreme members are using more sacred rhetoric, it could be a causal factor in groups with extreme members using less complex language because researchers have found that sacred rhetoric can decrease meaningful deliberation (Marietta, 2008).
You Pronouns
Given their greater confidence and perception of issue importance, we hypothesized that extreme members would use more you pronouns, as they focused on the other group members and engaged in “finger-pointing” to explain their “superior” position. In addition, in support of previous research (Simmons et al., 2005; Van Swol & Carlson, 2015), we found greater use of you pronouns was related to less influence for extreme members. We did not find this relationship between use of you pronouns and less influence for other members. Given their initially high confidence levels, perception of their own knowledge, and perception of the issue as important, use of you pronouns by extreme members may have been especially disconcerting to other group members. Furthermore, extreme members were prone to using more you pronouns already, so those who were on the high end for the extreme members may have engendered psychological reactance from other members. This article adds to research suggesting that you pronouns, especially mixed with perception of belief superiority, are indicative of a poor-quality relationship or influence attempt between the speaker and audience.
Language Style Matching
Research suggests that LSM should increase influence because of the role of mimicry. Research has found a relationship between speech mimicry and liking (Giles, 2008) and nonverbal mimicry and increased rapport in discussions (Chartrand & Bargh, 1999; Van Swol, 2003). Indeed, we found that for nonextreme members, LSM was positively related to influence in the group. However, we found the opposite pattern for extreme members. We had not hypothesized this interaction, so our reasoning is post hoc, but it poses an interesting question for future research. Because LSM was related to word count, this result may reflect that dominating the discussion was not increasing the influence of extreme members. This should be examined in future research.
Limitations and Future Research
This study did have several limitations. First, we used undergraduates who may not have had the most extreme opinions or knowledge on this issue. Indeed, Tetlock (2005) found that undergraduates were often worse forecasters of political issues, even when given information sheets about the issue. Whether our null results about no differences in complexity between extreme and nonextreme members would generalize to a more sophisticated sample is a question that should be addressed in future research. However, some of our other findings, like high confidence and decreases in confidence on using complex language for extreme members, does reflect some of the findings from Tetlock’s (2005) longitudinal study on professional forecasters. In some respects, our extreme members are similar to his hedgehogs. Tetlock (2005, p. 73) defines the hedgehog cognitive style as expressing “considerable confidence” in their knowledge and a more extreme opinion that is predicated on less integrative complexity.
Another limitation of an undergraduate sample is differences in motivation to discuss the issue. Some groups discussed the issue for 40 minutes before being asked to finish by the experiment, but most groups discussed the issue closer to 15 minutes. Some participants were passionate about the issue; one even wrote an editorial in the student newspaper the next day about the issue to continue the conversation (or win the argument). However, some students were more apathetic or less vocal. Future research should examine extremity and language in other contexts with other populations.
Conclusion
Like Tetlock’s (2005) hedgehogs, extreme members in this study entered the discussion with significantly higher confidence in their opinion and knowledge, but unlike Tetlock’s hedgehogs, we did not find that extreme members had less complexity. Yet during the course of discussion they had decreases in confidence, and when these extreme members used complex language, their influence and confidence waned. These results support Fernbach et al.’s(2013) research on the illusion of understanding and highlight the importance of deliberative debate and dissent among citizens when discussing political issues to moderate extremism (Sunstein, 2005).
Footnotes
Appendix
Should the Words “Under God” Be in the US 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 US 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 multi-faceted 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 |
Source. Should the words “under God” be in the US Pledge of Allegiance? Procon,org.
Acknowledgements
We would like to thank the anonymous reviewers for their comments and Brennan Harris and Riley Roehl for help editing the transcripts for LIWC.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was funded in part by a grant from the Wisconsin Alumni Research Foundation.