Personality Traits, Disagreement, and the Avoidance of Political Discussion: Putting Political Discussion Networks in Context Alan S. Gerber Yale University Professor Department of Political Science Institution for Social and Policy Studies 77 Prospect Street, PO Box 208209 New Haven, CT 06520-8209 [email protected]

Gregory A. Huber Yale University Professor Department of Political Science Institution for Social and Policy Studies 77 Prospect Street, PO Box 208209 New Haven, CT 06520-8209 [email protected]

David Doherty Loyola University Chicago Assistant Professor Department of Political Science 1032 W. Sheridan Road Coffey Hall, 3rd Floor Chicago, IL 60660 [email protected]

Conor M. Dowling Yale University Postdoctoral Associate Institution for Social and Policy Studies 77 Prospect Street, PO Box 208209 New Haven, CT 06520-8209 [email protected]

Personality Traits, Disagreement, and the Avoidance of Political Discussion: Putting Political Discussion Networks in Context

ABSTRACT: Social networks play a prominent role in the explanation of many political phenomena. Using data from a nationally representative survey of registered voters conducted around the 2008 U.S. presidential election, we document three findings. First, we show that during this period people discussed politics as frequently as (or more frequently than) other topics such as family, work, sports, and entertainment with frequent discussion partners. Second, the frequency with which a topic is discussed is strongly and positively associated with reported agreement on that topic among these same discussion partners. Supplementary experimental evidence suggests this correlation arises because people avoid discussing politics when they anticipate disagreement. Third, we show that Big Five personality traits affect how frequently people discuss a variety of topics, including politics. Some of these traits also alter the relationship between agreement and frequency of discussion in theoretically expected ways. This suggests that certain personality types are more likely to be exposed to divergent political information.

Keywords: Political Discussion; Social Networks; Personality Traits; The Big Five

People are affected by their family and friends, co-workers and neighbors, and others in countless ways.1 These social networks have been found to influence emotions, weight, smoking behavior, and a variety of other attitudes, attributes, and behaviors (see, e.g., Christakis and Fowler 2009). The effects of social (or, “discussion”) networks in the political realm have also been widely documented (e.g., Berelson, Lazarsfeld, and McPhee 1954; Huckfeldt and Sprague 1987). In fact, discussion “networks are now assumed to be a central part of the explanation of political dynamics in a wide range of phenomena” (Heaney and McClurg 2009, 732), including: political information acquisition2, vote choice3, partisan defection4, political participation5, and levels of political awareness and tolerance.6 Yet, some basic questions about the networks themselves remain unanswered. First, little is known about how frequently politics is discussed in these networks compared with other topics, such as work or sports. Second, it is not clear how the level of agreement about a topic between two potential discussion partners affects how frequently people discuss that topic. Do people only discuss politics when they are confident that their preferences are aligned with their discussion partners? If so, then the idea that discussion networks serve as a source of challenging new information is questionable. Third, we have little evidence regarding what types of people are most inclined to talk about particular topics. For instance, there may be basic, individual-level traits that lead some people to enjoy discussing politics, even if the discussion may lead to conflict (see, e.g., Mondak 2010; Mutz 2002a; 2002b). If this is the case, then exposure to challenging political information may depend on nonpolitical characteristics of the individual. This paper provides empirical evidence on each of these questions using a nationally representative survey sample of registered voters in the U.S. To our knowledge, this data provides the 1

This research was funded by Yale’s Center for the Study of American Politics and Institution for Social and Policy Studies. Data and supporting materials necessary to reproduce the numerical results will be made available at http://huber.research.yale.edu/ upon publication. 2 See, for example, Berelson, Lazarsfeld, and McPhee (1954); Huckfeldt (2007); Huckfeldt, Beck, Dalton, and Levine (1995); Huckfeldt, Mendez, and Osborn (2004); Huckfeldt and Sprague (1987). 3 See, for example, Berelson, Lazarsfeld, and McPhee (1954); Huckfeldt and Sprague (1991). 4 See, for example, Beck (2002); Sinclair (2009). 5 See, for example, Jang (2009); Lake and Huckfeldt (1998); Leighley (1990); McClurg (2003; 2006); Mutz (2002a) Nickerson (2008); Pattie and Johnston (2009); Scheufele, Nisbet, Brossard, and Nisbet (2004). 6 See, for example, Huckfeldt, Mendez, and Osborn (2004); Jang (2009); Mutz (2002b); Mutz and Mondak (2006). 1

first estimate of how much people talk and agree about politics relative to other (non-political) discussion areas, in two types of networks: 1) family and 2) non-family.7 Our survey employs network generators (described below) that ask about frequent conversation partners, enabling us to assess how we interact with those we are closest to. We present three main findings from these data, which add to previous work that examines the frequency with which people discuss various topics in different settings (e.g., at home, at work, or at a place of worship; see, Wyatt, Katz, and Kim 2000). First, we find that, compared with six other topics (family, work, god/religion, sports, food, and entertainment), during the course of the 2008 presidential election, politics was the most talked about topic among family members and a close second to work in the non-family discussion network. Second, reported agreement (between the respondent and her/his frequent discussion partner) on a topic is strongly associated with how frequently a topic is discussed in both family and non-family networks. This is true for all topics, including politics: The more people agree on political matters with their frequent discussion partners, the more they talk about it with them. (As with any cross-sectional analysis, determining the direction of causality is difficult. We provide suggestive experimental evidence about causality in our concluding section, where we also discuss the limitations of our approach.) Thus, even when frequent discussion partners have opposing viewpoints, the dyad may not experience extensive exposure to these views.8 For example, among those who report mostly agreeing with their family discussion partner about politics, 71% say they discuss politics often; among those who report mostly disagreeing, only 29% say they discuss politics often. A potential implication of the general preference for avoiding disagreement is that even people in diverse social networks may only occasionally be exposed to, and thus influenced by, their discussion partners’ differing positions (see, e.g., Levitan and Visser 2008; 2009; Visser and Mirabile 2004). To be clear, this is not to say that individuals whose discussion networks include individuals with divergent opinions are never exposed to these opinions or 7

We present evidence from both types of networks given recent debate about whether results are affected by the method of selecting (or, “naming”) discussion partners in the course of a survey (e.g., Huckfeldt, Levine, Morgan, and Sprague 1998; Klofstad, McClurg, and Rolfe 2009). 8 Huckfeldt and Mendez (2008) find evidence of this using the Indianapolis-St. Louis Study data. Our work corroborates this finding using a nationally representative sample. 2

are unaffected by them. Instead, the findings suggest that patterns of topic selection in discussions with these individuals differ substantially from those that emerge with discussants where less disagreement is present. Third, we show that Big Five personality traits are associated with the frequency with which a topic is discussed (particularly between family members), even after controlling for the level of agreement (and demographic characteristics). Psychological research has found that these personality traits are stable individual-level variations in how people respond to the stimuli they encounter in the world. Thus, examination of how personality is associated with communication can offer important insight into how fundamental psychological characteristics shape the social forces and information people encounter. For example, in the family network, an individual’s level of Extraversion—“the degree to which a person needs attention and social interaction” (Costa and McCrae 1992, 9)—is positively associated with discussing politics. In addition to direct associations between the Big Five and frequency of discussion of various topics, we also find that some of these traits affect the relationship between agreement and frequency of discussion in theoretically expected ways. For instance, in the family network, an individual scoring high on Emotional Stability (characterized as being “relaxed under stressful conditions”) is less likely to avoid discussing politics in the face of disagreement than a less Emotionally Stable individual. This suggests that certain types of individuals prefer political agreement more than others and, as a consequence, are less likely to be exposed to divergent political information. Overall, our results imply that not everyone is equally likely to experience cross-cutting discourse, even in heterogeneous networks. Exposure to such discourse is hypothesized to increase political awareness (Huckfeldt, Mendez, and Osborn 2004) and tolerance (Mutz 2002b) and also to encourage changes in attitudes in the face of new information (Levitan and Visser 2008; 2009). Although some people have a general proclivity to discuss politics, we find that others are particularly likely to only discuss politics when they already agree with their frequent discussion partners. In short, it appears that personality traits are associated with whether people expose themselves to political disagreement. Our results also place politics within the broader context of the myriad topics individuals can talk about and 3

add to the growing body of evidence that individual predispositions, such as the Big Five, are important determinants of political attitudes and behaviors (see, e.g., Gerber, Huber, Doherty, Dowling, and Ha 2010; Mondak, Hibbing, Canache, Seligson, and Anderson 2010). In the next section, we briefly discuss relevant literature on political discussion networks. We then describe the Five-factor Model (FFM) of personality—the Big Five: Extraversion, Agreeableness, Conscientiousness, Emotional Stability, and Openness—and discuss our expectations regarding how these traits affect an individual’s general inclination to discuss politics as well as whether that individual is likely to shy away from disagreement. Next, we discuss our survey data and present our empirical results. We conclude with a discussion of the implications and important limitations of our findings. 1. Political Discussion and Disagreement For some time, researchers have noted that people tend to have similar attributes and views, including political views, to those with whom they regularly interact (Berelson, Lazarsfeld, and McPhee 1954; Finifter 1974; Huckfeldt and Sprague 1987; 1995; McPherson, Smith-Lovin, and Cook 2001; Mutz 2002b). Evidence of such homophily—“birds of a feather flocking together”—in social networks suggests that even if politics (or any topic for that matter) is discussed, new information may not be acquired because people are only sharing like-minded views. There is also evidence, however, that political discussion networks are not strictly homogeneous (Huckfeldt, Beck, Dalton, and Levine 1995; Huckfeldt, Johnson, and Sprague 2002; Huckfeldt and Sprague 1995). In fact, there is some evidence to suggest that political homogeneity may not be the norm in many networks (Huckfeldt, Mendez, and Osborn 2004). Evidence of heterogeneity in political discussion networks most often rests on measures of candidate preference/vote choice (e.g., Huckfeldt and Sprague 1987; Huckfeldt, Mendez, and Osborn 2004) or party identification as reported by respondents and their discussion partners (e.g., Huckfeldt and Mendez 2008).9 When diversity is measured in this fashion, people appear to have relatively diverse networks, which suggests people will be exposed to a range of opinions. However, we know little about

9

Exceptions include Mutz (2002a; 2002b) who uses measures similar to ours (see below), asking people about the extent to which they agree with their discussion partners about political matters. 4

the extent to which people talk to their discussion partners about politics when they disagree with them. If people tend to avoid discussing subjects they disagree about, the potential for a diverse network to lead to sustained exposure to alternative views will not be realized. Huckfeldt and Mendez’s (2008) analysis of a sample of over 1,000 registered voters in the Indianapolis, IN and St. Louis, MO metropolitan areas provides some evidence that an individual discusses political matters less frequently the more disagreement there is between her and her discussion partners (also see Mutz 2002b, footnote 9). Based on this previous work, we expect the relationship between reported agreement on a topic, such as politics, and the frequency of discussion of that topic to be positive in our national sample of registered voters. Are some people more affected by disagreement (and therefore less willing to discuss politics in the face of it) than others? Our survey included questions used to measure an individual’s underlying personality traits, which we discuss in the next section. There is some evidence from previous work on political discussion networks that individual-level preferences other than personality traits may help determine the extent to which an individual discusses politics. In particular, Mutz’s (2002a; 2002b) work on the effects of “cross-cutting” (i.e., heterogeneous) social networks demonstrates that a person’s level of awareness of opposing political views is predicted by the interaction between their level of exposure to dissonant political views and their “civil orientation toward conflict” (this individual characteristic is measured as the combination of an individual’s [1] preference for social harmony and [2] conflict orientation; Mutz 2002b, 118). According to this account, exposure to cross-cutting political views via discussion results in greater awareness of opposing views, especially for those individuals who value social harmony but are also willing to have a spirited discussion. By contrast, we investigate how disagreement and an individual’s general personality dispositions (apart from politics) shape exposure to disagreement in the first place. In this way, our analysis complements Mutz’s work by offering insight into the dynamics of initial exposure to potentially informative opposing viewpoints. 2. The Five-factor Model of Personality Traits The particular personality characteristics we consider as potentially explaining both a willingness to discuss different topics and whether to do so in the face of disagreement are measures of the Big Five 5

traits. Although there is no fully comprehensive way to conceptualize and measure an individual’s personality, a strong consensus has emerged in psychology that a Five-factor Model (FFM) provides an appropriate and comprehensive way of measuring a person’s dispositional personality traits (see, e.g., John and Srivastava 1999; McAdams 1995). These dispositional traits—Extraversion, Agreeableness, Conscientiousness, Emotional Stability (sometimes referred to by its inverse, Neuroticism), and Openness (to Experience)—are thought of as variations in basic individual-level tendencies (McCrae and Costa 1996) and are defined in greater detail below. The Big Five appear to be rooted in genetic factors (Bouchard 1997; Plomin, DeFries, McClearn, and McGuffin 1990; Van Gestel and Van Broeckhoven 2003) and are highly stable through the life cycle (Costa and McCrae 1992; Gosling, Rentfrow, and Swann 2003; Caspi, Roberts, and Shiner 2005; but see Srivastava, John, Gosling, and Potter 2003). An additional benefit of examining Big Five personality traits is that, from a theoretical perspective, these dispositional traits are broad, general orientations that are thought to shape individuals’ responses to the vast array of stimuli—political and non-political alike—that people encounter in the world. Thus, the Big Five stand in contrast to many other measures of personality that have been used to predict political outcomes. For example, Hayes, Scheufele, and Huge (2006) find that individuals who score higher on a Willingness to Self-Censor scale (see Hayes, Glynn, and Shanahan 2005a, 2005b)—i.e., individuals who are “more influenced by the climate of opinion when choosing whether or not to voice an opinion” (Hayes, Scheufele, and Huge 2006, 259)—are less likely to take part in public political activities (e.g., to attend a political meeting). An individual’s willingness to self-censor is conceptually much closer to the amount and type of discussion an individual engages in than the much broader Big Five traits. Despite their breadth, research finds that Big Five traits predict an array of attitudes and behaviors, from job and school performance to alcohol and tobacco consumption, to musical tastes and the way we dress, to our overall health and a variety of other behaviors and attitudes (see, e.g., Goldberg 1993; Gosling 2008; Graziano and Eisenberg 1997; Hogan and Ones 1997; John and Srivasta 1999; McCrae 1996; Ozer and Benet-Martinez 2006; Paunonen and Ashton 2001; Watson and Clark 1997). The strength and robustness of findings from research applying the FFM has led to a great deal of enthusiasm 6

among researchers and, recently, the Big Five have also been shown to predict a number of political outcomes. Most notably, there is a great deal of evidence that Big Five traits (Openness and Conscientiousness in particular) are associated with party identification and political ideology (Carney, Jost, Gosling, and Potter 2008; Gerber, Huber, Doherty, Dowling, and Ha 2010; Gerber, Huber, Doherty, and Dowling 2011a; Mondak and Halperin 2008). There is also some evidence that Big Five traits are associated with voting for particular parties or candidates (Caprara, Barbaranelli, and Zimbardo 1999; Caprara, Schwartz, Capanna, Vecchione, and Barbaranelli 2006; Rentfrow, Jost, Gosling, and Potter 2009; Schoen and Schumann 2007), various participatory acts, such as voter turnout (Gerber, Huber, Doherty, Dowling, Raso, and Ha 2011; Mondak 2010; Mondak and Halperin 2008; Mondak, Hibbing, Canache, Seligson, and Anderson 2010), and political interest and knowledge (Gerber, Huber, Doherty, and Dowling 2011b; Mondak 2010; Mondak and Halperin 2008). 2.1 The Big Five traits, political discussion, and disagreement To our knowledge, only a few previous studies have examined the relationships between an individual’s Big Five traits and the frequency with which that individual engages in political discussion (see Hibbing, Ritchie, and Anderson 2010; Mondak 2010, chapter 4; Mondak and Halperin 2008), and no work has examined the relationships between the Big Five and the frequency an individual discusses other topics in comparison to politics. However, there is also other related work on the Big Five or other individual-level characteristics and social networks. For instance, using an innovative research design that leverages the random assignment of college students to a dorm roommate, Klofstad (2009) finds that the positive association between “civic talk” (discussion of politics and current events) and “civic participation” is more pronounced for individuals who are predisposed to engage in civic talk (measured by the individual’s amount of civic discussion and participation in high school). In addition, a few researchers have investigated how the Big Five are linked to various aspects of social networks. Mondak et al. (2010) measure the relationships between the Big Five traits and social network size. They find that Openness and Extraversion are positively associated with the size of an individual’s social network and that Conscientiousness and Emotional Stability are negatively associated 7

with social network size. They also find that low levels of Agreeableness and high levels of Extraversion strengthen the relationship between social network size and exposure to cross-cutting political discourse (also reported in Mondak 2010, 119). These results suggest that personality traits may condition the relationship between level of reported agreement with a frequent discussion partner and the frequency with which politics is discussed, a topic we investigate directly below. (In other words, personality traits may condition both who is in one’s social network [Mondak et al. 2010] and whether and how frequently politics is discussed with a particular discussion partner.) 2.2 Hypotheses about how Big Five traits are related to choice of discussion topic Building on this prior work, we present several hypotheses regarding the relationships between the Big Five and what people discuss. These hypotheses are based on the nature of political discussion (e.g., it can be contentious), psychologists’ descriptions of the Big Five traits, the results reported in Hibbing, Ritchie, and Anderson (2010), Mondak (2010, chapter 4), and Mondak and Halperin (2008), and (where appropriate) other related work.10 We discuss each trait separately and offer a specific prediction for how it should be associated with the frequency of political discussion.11 Extraversion is defined by “the degree to which a person needs attention and social interaction” (Costa and McCrae 1992, 9). Extraverts are active and assertive, while their counterparts (introverts) are more reserved. This need for social interaction may drive those scoring high on Extraversion to discuss a variety of topics, including politics, more frequently than those scoring low on this trait. The prior work on personality and political discussion is largely supportive of such a prediction. For example, Mondak and Halperin (2008) find Extraversion to be positively associated with the frequency of political discussion in each of their three survey samples.12 Therefore, we expect Extraversion to be positively associated with the frequency of political discussion.

10

One important difference between our work and prior work is that we focus on discussion of politics and other topics with two specific discussion partners, whereas prior work examines the frequency a person discusses politics generally (e.g., “number of days in the past week” in Mondak and Halperin) or local politics (in Hibbing, Ritchie, and Anderson). 11 For reasons of brevity and analytical clarity, we do not offer a specific hypothesis for every discussion topic. 12 This association is only statistically significant in two of their three samples. 8

Agreeableness is defined by “the degree to which a person needs pleasant and harmonious relations with others” (Costa and McCrae 1992, 9). Individuals scoring high on Agreeableness tend to be compassionate, good-natured, and eager to cooperate, while those scoring low on Agreeableness are hardheaded and skeptical. Given that individuals scoring high on Agreeableness are eager to cooperate, we expect them to be less likely to discuss contentious topics, such as politics. Mondak and Halperin (2008) find mixed evidence for this expectation (also see Hibbing, Ritchie, and Anderson 2010; Mondak 2010, chapter 4): In one of their samples Agreeableness is negatively and statistically significantly associated with the frequency of political discussion, in the other two it is statistically insignificant. Conscientiousness is defined by “the degree to which a person is willing to comply with conventional rules, norms, and standards” (Costa and McCrae 1992, 9). Those scoring high on Conscientiousness are well-organized and goal-oriented, while those scoring low on Conscientiousness are easygoing and careless. We do not have a strong expectation for how Conscientiousness will relate to the frequency with which a person discusses politics. We might expect those scoring high on Conscientiousness to discuss politics more frequently if they view it as a social norm that should be followed. Alternatively, it is difficult to imagine a person who is easygoing and careless being unwilling to discuss politics, so we might expect there to be no difference in the frequency of political discussion between those scoring low and high on Conscientiousness. Mondak and Halperin (2008) find small and statistically insignificant associations between Conscientiousness and the frequency with which a person discusses politics in two of their samples; in their other sample, they find a positive association between Conscientiousness and frequency of political discussion (also see Hibbing, Ritchie, and Anderson 2010). Conversely, Mondak (2010, chapter 4) reports negative associations between Conscientiousness and political discussion. Given both the lack of clear theoretical arguments and the uncertainty in previous research, we do not have offer a prediction for Conscientiousness. Emotional Stability is defined by “the degree to which a person experiences the world as threatening and beyond his/her control” (Costa and McCrae 1992, 9). The more Emotionally Stable are secure, hardy, and relaxed under stressful conditions, while their counterparts, the more neurotic, tend to 9

be anxious, sensitive, and easily upset. Those scoring high on Emotional Stability report being more interested in politics and are more politically knowledgeable (see Gerber, Huber, Doherty, and Dowling 2011b). They are also less likely to regard interpersonal or politically conflictual interactions as threatening (Antonioni 1993; Gerber, Huber, Doherty, Dowling, Raso, and Ha 2011). Therefore, we expect Emotional Stability to be associated with greater frequency of political discussion.13 Finally, Openness is defined by “the degree to which a person needs intellectual stimulation, change, and variety” (Costa and McCrae 1992, 9). Those scoring high on Openness tend to have broad interests and be imaginative, while those scoring low on this trait are more practical and traditional. The need for intellectual stimulation and variety may drive a person scoring high on Openness to discuss politics more frequently because it is a topic that can challenge people to think about alternative opinions. Moreover, there is evidence that more Open individuals are more politically interested and knowledgeable (see Gerber, Huber, Doherty, and Dowling 2011b; Mondak and Halperin 2008). Mondak and Halperin (2008) and Mondak (2010, chapter 4) also find a positive and statistically significant association between Openness and frequency of political discussion in all three of their samples. Therefore, we expect Openness to be positively associated with frequency of political discussion. 2.3 Hypotheses about how Big Five traits moderate the relationship between disagreement and discussion In addition to direct relationships between the Big Five and the frequency of political discussion between a respondent and her discussion partner, the Big Five may also moderate the relationship between reported agreement on politics and the frequency of political discussion (between the respondent and her discussion partner). Recall that we expect the relationship between reported agreement on a topic, such as politics, and the frequency of discussion of that topic to be uniformly positive. The clearest theoretical basis we have for expecting Big Five traits to moderate this relationship is for Agreeableness. Compared to those scoring low on Agreeableness (individuals who are more “hardheaded” and 13

Mondak and Halperin (2008) do not find support for this hypothesis. Emotional Stability is negatively associated with frequency of political discussion in all three of their samples, although the negative relationship is statistically significant in only one sample (also see Mondak 2010, chapter 4). Hibbing, Ritchie, and Anderson (2010) do not report any statistically significant associations between Emotional Stability and political discussion. 10

“competitive”), we expect those scoring high on Agreeableness (individuals who are more “eager to cooperate”) to be less comfortable with disagreement, and thus, less likely to discuss political matters with people they disagree with. We also expect individuals scoring high on Emotional Stability (who are more “relaxed under stressful conditions”) to be less repelled by disagreement compared to individuals scoring low on Emotional Stability (who are more “sensitive” and “prone to feelings that are upsetting”). Similarly, we expect the assertiveness that is characteristic of individuals scoring high on Extraversion to result in these individuals being more willing to engage in political disagreements. In other words, we expect less Emotionally Stable, less Extraverted, and more Agreeable individuals to be more averse to political disagreements and, therefore, particularly unlikely to discuss politics if they disagree with others. By contrast, we do not have clear hypotheses for how Conscientiousness or Openness will moderate the relationship between agreement on political matters and the discussion of politics. 3. Research Design Our primary survey data come from the 2007-2008 Cooperative Campaign Analysis Project (CCAP: Jackman and Vavreck 2009). The CCAP is an Internet-based survey conducted by YouGov/Polimetrix that uses a combination of sampling and matching techniques to approximate a random digit dialing sample. (See Section 1 of the Supporting Information document for further details on sample construction.) The final CCAP sample is nationally representative of registered voters in the U.S. All the descriptive statistics and analyses presented below use the sampling weights provided by YouGov/Polimetrix. The CCAP was fielded as a panel survey with five waves before the November 2008 election (the pre-election waves) and one post-election wave. Our questions about respondents’ discussion networks were part of the October (or, last pre-election) wave of the study, and were completed by 682 respondents.14 Demographic characteristics, including the Big Five, were collected in December 2007, except where otherwise noted. See Section 2 of the Supporting Information document 14

In total, 20,000 respondents participated in the Common Content of the CCAP. Our discussion networks battery appeared on our Private Content module, which 863 respondents were assigned to complete. Of these, 682 (79%) completed our entire discussion battery and form the sample we use in the analyses presented below. 11

for a complete list of variables, including question wording and coding rules for each variable. In addition to the CCAP survey, we conducted a follow-up survey in February of 2011 (N=1,235), which was also fielded by YouGov/Polimetrix (details about question wording in the follow-up survey are in Section 2 of the Supporting Information document). The data from this follow-up survey provide a way for us to more carefully examine several issues related to the measurement of discussion frequency and to investigate the robustness of several of our findings. We note some of the key findings from this survey in the text and discuss the results in more detail in Section 4 of the Supporting Information document. 3.1 Discussion networks battery The discussion networks battery that we placed on the CCAP permits us to document the frequency with which people talk about a variety of topics and their perceived level of agreement with their discussion partner on the same topics. We do so for the respondent’s (1) familial (“Which person in your immediate family do you talk with most frequently?”) and (2) non-familial (“Who, outside of your immediate family, do you talk with most frequently?”) network.15 In the case of the non-family network, respondents were asked to list the first name of their three “most frequent conversation partners.” One name was then selected at random for the respondent to answer the discussion frequency and topic agreement questions. We note that this generator therefore yields a range of discussion partners, from the most frequent partner one-third of the time to the third most frequent—a more “distant” person—onethird of the time.16 After giving the name of a person, respondents were then asked how often they talk with that individual (“Daily or more than daily,” “A few times a week,” “Once a week,” or “Once a month or less”). Following this, respondents were asked, “How often do you talk with [name] about each of the following topics?” (“Never,” “Rarely,” “Sometimes,” or “Often”) and then “Do you think that [name] agrees with your viewpoints on these topics?” (“Mostly agrees,” “Agrees and Disagrees,” “Mostly

15

For each respondent, the questions about discussion with the immediate family member appeared prior to the questions about discussion with the person outside of the immediate family. 16 We replicated this non-family discussion partner name generator in our February 2011 follow-up survey and asked respondents whether the discussion partner was a relative (from outside of their immediate family), a coworker, a friend, or a neighbor. There, we find that 19.9% of respondents were assigned to a non-family discussant partner who was a relative, 15.5% to a co-worker, 59.5% to a friend, and 5.1% to a neighbor. 12

disagrees,” and “Don’t know/Don’t talk about”). These questions were asked about seven topics, which appeared in a grid in the following order: (1) “Family/kids/spouse/dating,” (2) “Work,” (3) “God/religion,” (4) “Sports,” (5) “Food,” (6) “Current events/politics,” and (7) “TV/entertainment/celebrities.” The name generators used in this study have two features that might affect which of the respondent’s discussion partners are elicited. First, the name generators were designed to identify respondents’ frequent discussion partners. As expected (and as Figure 1, discussed in more detail below, shows), respondents did report frequently discussing a variety of topics with these discussion partners. Second, although we did not specifically ask people to name individuals with whom they discuss “important matters” or “political matters” as is typically done in prior work (see, e.g., Klofstad, McClurg, and Rolfe 2009), our name generator nevertheless did identify (frequent) political discussion partners. We were able to assess this because after the full battery described above was completed, we asked respondents who they “talk with most frequently about politics and current events” and less than 6% of respondents reported someone other than a person they had already named.17 Thus, even though we did not ask for respondents to identify people with whom they discuss political (or important) matters, our name generator appears to have successfully identified these individuals at a high rate.18 Before conducting a more detailed statistical analysis, we first show the relationship between disagreement and frequency of discussion. For each of the seven topics, Figure 1 displays the average Frequency of Discussion (Panel A), which ranges from zero (“never”) to three (“often”), and average Agreement (Panel B), which ranges from -1 (“mostly disagrees”) to 1 (“mostly agrees”). (See Table A in the Supporting Information document for a complete set of summary statistics.) Three aspects of these figures are noteworthy. First, politics was the most talked about topic among family members (gray bars;

17

The full question wording was: “Who do you talk with most frequently about politics and current events?” The response options were: “Family person named above ([family member name])”; “One of [3] people you named above ([Person A], [Person B] or [Person C])” or “Someone else”. 18 This is perhaps not surprising given recent work that finds that political discussion networks tend to be part of individuals’ general discussion networks, not a distinct, separate group of individuals (Lazer, Rubineau, Chetkovich, Katz, and Neblo 2010; Levitan and Visser 2009). 13

all differences between politics and other topics significant at p<.05 except family [p=.252]) and a close second to work among non-family members (white bars; all differences between politics and other topics significant at p<.05 except work [p=.243]).19 Second, although people talk to their family members more than their non-family members about each topic (i.e., the gray bars are higher than the white bars for each topic; all within topic differences are significant at p<.05), there is remarkable consistency between family members and non-family members in terms of which topics are most frequently discussed. For example, sports is by far the least talked about topic, with religion and entertainment also talked about infrequently compared to politics, work, and family. Third, average levels of agreement on topics are, for the most part, closely related to their average frequency of discussion. The lowest levels of agreement are for sports and entertainment, which are discussed relatively infrequently compared to the other topics. Work and family receive the highest levels of agreement, and are discussed frequently. For the purpose of the analysis that follows we create a variable for both relative discussion frequency and relative agreement for each frequent discussion partner (family and non-family). These measures account for the total amount of talking or agreement the respondent engages in with her discussion partner. Specifically, for each topic we subtract the average on all other topics from the respondent’s score on that topic to create, for example, measures of Relative Frequency of Political Discussion and Relative Agreement on Politics. We use these measures of relative discussion frequency and relative agreement because they account for individual-level measurement error in response to the questions. For example, some respondents could say “agree” (or, “disagree”) for every topic because their impression of agreement differs from other respondents. The relative measures serve to scale the responses for each individual relative to their baseline (average) evaluation.20

19

The frequency of political discussion is presumably strongly affected by the fact that the survey was fielded during a presidential campaign. However, in the follow-up survey we find that politics is still a common topic of discussion (in February 2011). On average, people reported discussing politics more frequently than entertainment, sports, and religion, but less frequently than work, family, and food (all differences significant at p<.05). 20 Relative Frequency of Political Discussion = Frequency of Political Discussion – Sum of Discussion Frequency for Six Topics Other than Politics/6; Relative Agreement on Politics = Agreement on Political Topics – Sum of Agreement for Six Topics Other than Politics/6. The results we present are largely robust to using non-relative measures of agreement and discussion frequency. 14

3.2 Ten-Item Personality Inventory (TIPI) To measure the Big Five traits, we used the Ten-Item Personality Inventory (TIPI: Gosling, Rentfrow, and Swann 2003). A number of batteries have been developed to measure the Big Five trait domains, with each consisting of a list of adjectives (e.g., “temperamental”; Goldberg 1992) or phrases (e.g., “Sometimes I do things on impulse that I later regret”; Costa & McCrae 1992) and asking the respondent to rate how well each adjective or phrase describes the individual whose personality is being rated (typically the respondent). Researchers then use these ratings to calculate scores for each of the Big Five traits. While there are a variety of instruments that can be used to measure the Big Five, ranging from brief batteries of 10 items (such as the TIPI) to batteries that use dozens (Big Five Inventory [BFI]: John et al. 1991; Mini-Markers: Saucier 1994; NEO-Five-Factor Inventory [NEO-FFI]: Costa & McCrae 1992) or even hundreds (NEO-Personality Inventory-Revised [NEO-PI-R]: Costa & McCrae 1992; International Personality Item Pool [IPIP]: Goldberg et al. 2006) of items, the TIPI is ideal in the survey context because its length and speed of administration make it feasible where longer batteries, such as the BFI or NEO-PI-R are not. Additionally, Gosling et al. compared the performance of the TIPI to much longer tests and found that TIPI scores are highly correlated with those obtained from longer instruments (2003, see Tables 6 & 9). Responses to the TIPI also appear to capture meaningful across-individual variation in personality traits even when measured in explicitly political surveys and appear to be unaffected by political events (Gerber, Huber, Doherty, and Dowling 2011c).21 The TIPI asks respondents to report whether “I see myself as” characterized by a series of 10 trait pairs (e.g., “Extraverted, enthusiastic”) using a 7-point scale ordered from Disagree Strongly to Agree Strongly. Each Big Five trait is measured by responses to two trait pairs, with one trait pair for each trait reverse scored to mitigate problems of acquiescence bias. Responses to these 10 questions are used to score a respondent’s personality for each of the Big Five traits. In the analysis presented below, each Big Five trait is standardized to have a mean of 0 and standard deviation of 1. 21

The correlations among Big Five traits are reported in section 7 of the Supporting Information document. Section 3 of that document also provides a brief discussion of the tradeoffs researchers face when deciding which personality battery to use. 15

3.3 Model specification In our simplest model specification we regress Relative Frequency of Discussion (for a given topic) on Relative Agreement (on that same topic) and a vector of controls. (1) Relative Frequency of Discussiont,i = B0 + B*Relative Agreementt,i + C*Controlsi + e, where Relative Frequency of Discussion and Relative Agreement vary by topic (t) and respondent (i) and Controls varies by respondent (i), and includes how often the respondent talks with their discussion partner (excluded category is “Once a month or less”), gender of respondent (indicator for female), race of respondent (White [excluded category], Black, Hispanic, or Other), age of respondent, age-squared (to allow for non-linearity in the effects of age), and educational attainment of respondent (excluded category is high school diploma or equivalent). We estimate this equation for each topic and for each respondent’s family and non-family discussion partner, employing weights to obtain representative population estimates. This model specification and those that follow do not account for any reverse causality in which frequency of discussion of a topic influences actual or perceived agreement on the topic (see Huckfeldt and Mendez 2008). To test for the direct effects of personality on the frequency of discussion for these seven topics, we add our measures of the Big Five to the equation (1) specification: (2) Relative Frequency of Discussiont,i = B0 + B*Relative Agreementt,i + C*Controlsi + D*Personalityi + e. Then, for our analysis of the differential effect of personality by relative agreement on the topic, we extend the equation (2) specification by interacting Personality with Relative Agreement. (3) Relative Frequency of Discussiont,i = B0 + B*Relative Agreementt,i + C*Controlsi + D*Personalityi + E*Personalityi*Relative Agreementt,i + e. One potential concern with the equation (3) specification is that it assumes that the associations between personality and frequency of discussion vary only by agreement on the topic—e.g., they do not also vary across gender or racial groups. In order to account for the possibility of such relationships, which recent work examining political ideology has shown is a distinct possibility (Gerber, Huber, Doherty, Dowling,

16

and Ha 2010), we also present results from specification (3’) in Section 7 of the Supporting Information document. (3’) Frequency of Discussiont,i = B0 + B*Relative Agreementt,i + C*Controlsi + D*Personalityi + E*Personalityi*Relative Agreementt,i + F*Controlsi*Relative Agreementt,i + G*Controlsi*Personalityi + e. This specification includes all possible interactions between Relative Agreement and Controls, as well as the interactions between Controls and Personality. It therefore minimizes concerns that any estimated relationship between Personality and Relative Agreement proxies conditional effects of either Relative Agreement or Personality on other measured controls. 4. Analysis 4.1 Agreement on issue and frequency of discussion Table 1 presents the results of model specification (1), which addresses the question of whether citizens are more likely to discuss an issue the more their views are in agreement, for the respondent’s familial (Table 1a) and non-familial (Table 1b) network. We find support for our expectation that, on average, people discuss a topic more frequently the more their views on that topic are in agreement with their discussion partner’s views. The relative agreement coefficient is positive and statistically significant for all seven topics in both the family and non-family discussion network. Thus, in 14 unique specifications accounting for seven different discussion topics and two different types of discussion partners, we uniformly find that (relative) disagreement is associated with (relative) frequency of discussion. The magnitudes of these effects are also quite large. For example, a two standard deviation increase in relative agreement on politics predicts an increase of about two-thirds of a standard deviation for how frequently politics is discussed relative to mean topic discussion levels (see column [7] of both tables) in both types of networks.22 This relationship is comparable to the relationship between relative agreement and relative discussion frequency in the other topic areas, and is robust to alternative model specifications—including 22

For example, in the family network, the coefficient for relative agreement on politics = 0.343, SD = 0.071. Thus, a two standard deviation shift in relative agreement = 0.343 + (0.071*2) = 0.485. The standard deviation of the outcome measure, relative political discussion = 0.675. 0.485 is approximately 70% of .675. 17

a) excluding indicators for how often the pair talks (column [8]), b) using absolute reported agreement instead of relative agreement (column [9]), c) using absolute discussion frequency and absolute agreement measures (column [10]), and d) estimating the column (10) model using ordered probit (column [11]). In short, politics looks quite similar to the other topics in terms of the extent to which the frequency with which it is discussed is associated with agreement. Previous work has shown that politics tends to be discussed more the more people are in agreement (see, e.g., Huckfeldt and Mendez 2008), and our evidence suggests that politics is not atypical in this regard. 4.2 Big Five personality traits and frequency of discussion Next, we examine the role Big Five personality traits play in shaping people’s patterns of discussion. Table 2 presents the results of model specification (2) using frequency of political discussion as the dependent variable. This analysis allows us to examine whether certain Big Five traits make people more inclined to talk about politics. (Table C in the Supporting Information document shows the results of model specification (2) for the remaining six topic areas for both the family and non-family networks.) Before discussing the results associated with our hypotheses regarding relationships between the Big Five and frequency of discussion, we highlight one other aspect of the results shown in Table 2 that is also noteworthy: The associations between relative agreement on an issue and the relative frequency with which that issue is discussed are largely unchanged by the inclusion of the Big Five in both networks. The coefficient on issue agreement in the family network moves from .34 in Table 1a to .31 in Table 2 (column [1]), and the coefficient on issue agreement in the non-family network changes from .43 in Table 1b to .44 in Table 2 (column [6]). This suggests that the relationship between (relative) agreement and (relative) frequency of discussion does not arise from the Big Five simultaneously causing people to both agree (or report agreement) about politics and discuss (or report discussing) it more. Turning to our predictions regarding the relationship between personality and political discussion, we find in column (1) of Table 2, as expected, that Extraversion and Emotional Stability are positively associated with the relative frequency of political discussion in family discussion networks, most likely because they are drawn to social interaction and are not easily upset, respectively. Also, as expected, 18

Agreeableness is negatively associated with discussing politics with family members across columns (1)(5), which mirror the alternative specifications presented in columns (7)-(11) of Table 1a and 1b. This suggests that, consistent with expectations, more Agreeable individuals shy away from political discussion because they recognize that it can be a contentious topic that can result in disagreements. We find a positive association between Openness and frequency of political discussion, however, this coefficient is not statistically significant (p-value = .239; all p-values mentioned in the text are twotailed.) We did not have clear expectations about Conscientiousness, and find that it is negatively associated with discussing politics, but indications of statistical significance change across model specifications in columns (1)-(5). These associations between the Big Five and (relative) frequency of discussion are much more pronounced among family members than in the non-family network, however. In columns (6)-(10) of Table 2, we find that none of the five traits are statistically significantly associated with frequency of political discussion in the non-familial network, whereas three (Extraversion, Agreeableness, and Emotional Stability) are consistently statistically significant in the familial network. We discuss the different pattern of results for the family and non-family discussion networks in Section 5 of the Supporting Information document. 4.3 Big Five Traits and the relationship between agreement and frequency of political discussion Table 3 presents the results of model specification (3), in which we interact each of the Big Five traits with reported agreement on politics, both for the respondent’s familial (columns [1]-[2]) and nonfamilial (columns [3]-[4]) network. This specification permits us to test whether the Big Five alter the relationship between agreement on politics and the frequency with which politics is discussed. We outlined three expectations for how the Big Five might moderate the relationship between agreement and frequency of political discussion. We expected less Extraverted, more Agreeable, and less Emotionally Stable individuals to be more averse to disagreement. In other words, we expected the positive association between agreement on a topic and frequency of discussion of that topic to be stronger for these individuals. If this is the case, then we should observe a positive coefficient on the interaction term 19

Agreeableness X (relative) agreement and negative coefficients for Extraversion and Emotional Stability X (relative) agreement. Beginning with the results for the familial network, we find that Extraversion and Emotional Stability moderate the relationship between relative agreement on politics and relative frequency of political discussion in the manner we expected. In columns (1)-(2) of Table 3 the maximum p-value for the interaction term between Emotional Stability and relative agreement is .139 and the maximum p-value for the interaction term between Extraversion and relative agreement is .032. The relationship between relative agreement and relative frequency of political discussion is stronger among individuals scoring low on Emotional Stability (who tend to be more anxious). This may be because they are more likely to become upset when they find themselves in political disagreements. Likewise, among individuals scoring low on Extraversion (who are more “reserved”), those who disagree with their family discussant are particularly unlikely to discuss politics frequently, presumably because they lack the assertiveness of their more Extraverted counterparts. These findings are largely robust to the same alternative model specifications that we presented in Tables 1 and 2: excluding indicators for how often the pair talks in column (2), using absolute reported agreement instead of relative agreement (Supporting Information Table D, column [1]), using absolute discussion frequency and absolute agreement measures (Supporting Information Table D, column [2]), and estimating the model using absolute discussion frequency and absolute agreement measures using ordered probit (Supporting Information Table D, column [3]).23 In terms of our other hypothesis, although the coefficient for Agreeableness interacted with relative agreement is in the expected (positive) direction—perhaps reflecting the eagerness to cooperate that characterizes individuals scoring high on this trait—it is not statistically significant at conventional levels. (The p-value in column [1] is .168.) We offered no hypothesis for how Conscientiousness might alter the relationship between agreement and frequency of discussion, and find that the coefficient for this trait’s interaction with (relative) agreement is small and statistically insignificant across columns (1)-(2). 23

These results are also robust to our alternative model specification (3’) in which we interact all of the controls with the Big Five personality traits and all of the controls with relative agreement (although the interaction between Emotional Stability and relative agreement is not statistically significant). See Supporting Information Table E. 20

Finally, we offered no hypothesis for how Openness might change the relationship between agreement and frequency of discussion, and find a positive and insignificant effect. Turning to the non-family network (columns [3]-[4] of Table 3), we find consistent, but unexpected evidence that Openness moderates the relationship between agreement on politics and frequency of political discussion. The coefficient for the interaction between Openness and (relative) agreement is positive and statistically significant in both model specifications, suggesting that more Open individuals discuss politics more frequently the more they agree about politics with their non-family discussion partner. (This finding is robust to the alternative model specifications presented in columns [4]-[6] of Supporting Information Table D, but falls short of statistical significance in columns [3]-[4] of Supporting Information Table E where the maximum p-value is .154.) Additionally, Extraversion and Emotional Stability moderate the relationship between agreement on politics and frequency of political discussion in the same (expected) manner as in the familial network. However, their moderating effects are not statistically significant in this network (although the moderating effect for Emotional Stability is statistically significant in some of the alternative model specifications presented in Supporting Information Tables D and E). Finally, contrary to expectations the coefficient on Agreeableness X relative agreement is negative in this network, but indistinguishable from 0. To provide a sense of the magnitude of the statistically significant effects, we calculated the effect of a two standard deviation (SD) increase in relative agreement on relative frequency of political discussion separately for individuals scoring low (one standard deviation below the mean) and high (one standard deviation above the mean) on each personality trait. These estimates are based on the models reported in column (1) and column (3) of Table 3. In the family network, a two SD increase in relative agreement predicts a .62 and .58 unit increase in relative discussion frequency for those scoring low on Extraversion and Emotional Stability, respectively, but only a .19 and .22 unit increase for those scoring high on these same traits. In other words, the effect of relative agreement on relative frequency of political discussion in the familial network is roughly three times larger for individuals scoring low on Extraversion and Emotional Stability than those scoring high on the same traits. (In the non-family 21

network the corresponding, but not statistically significant numbers are approximately .88 [for those scoring low on Extraversion] and .88 [for those scoring low on Emotional Stability] and about .47 [for those scoring high on Extraversion] and .47 [for those scoring high on Emotional Stability].) There is also a clear difference between individuals scoring low and high on Openness, particularly in the non-family network where a two SD increase in relative agreement predicts a 1 unit increase in relative discussion frequency for those high on Openness, but only a .27 unit increase for those low on Openness. In sum, the analysis presented in Table 3 confirms two of the three hypotheses we offered and reveals one relatively robust and unexpected effect (for Openness in the non-family network). More generally, it provides additional evidence that personality traits alter who is exposed to political disagreement and provides evidence that personality shapes discussion within existing social networks (also see Mondak 2010, 119; Mondak et al. 2010). 5. Discussion The preceding analysis illustrates, first, that although politics is only one of many topics people discuss, it is a topic that is discussed quite often (particularly during presidential campaigns). We also show that reported agreement on a given topic—including politics—predicts the extent to which that topic is discussed. Thus, although political discussions do occur frequently, those conversations are substantially more likely to occur among discussion partners who tend to agree about political matters (also see Huckfeldt and Mendez 2008). Our findings therefore suggest that even if a discussion partner does hold some opposing political viewpoints, political discussions with that partner may focus on topics where both parties already agree, limiting exposure to opposing viewpoints. Recent work finds that holding similar political views does not appear to be the reason behind the creation of new communication networks (Lazer, Rubineau, Chetkovich, Katz, and Neblo 2010; Levitan and Visser 2009). Our results complement this prior work because they suggest that when people have frequent discussion partners with whom they happen to disagree about political matters, one of the ways these relationships can be maintained is through regulation of the frequency of political discussions. Just as political matters are not the only (or perhaps even the primary) consideration that determines who ends 22

up in a person’s communication network, discussants can select from a menu of topics to discuss in order to maintain their relationship in the face of potential disagreement. Our analysis of the direct and moderating effects of Big Five personality traits on frequency of (political) discussion showed that these relationships differed between the two types of networks we examined. Differences in Extraversion, Agreeableness, and Emotional Stability were associated with the frequency that politics was the topic of conversation in the familial network. While this suggests that personality plays a role in the extent to which people discuss politics (i.e., certain individuals are predisposed to discuss politics), this finding is not replicated in the non-family network. In fact, in the non-family network none of the five traits were associated with frequency of political discussion. Last, we found that the Big Five traits moderate the relationship between agreement and frequency of discussion. Most notably, we find evidence that more Extraverted and more Emotionally Stable individuals were less responsive to disagreement (particularly with their family member). That is, the positive relationship between agreement and frequency of discussion was weaker for these types of individuals. Interestingly, we also find evidence that the relationship between agreement and frequency of discussion was particularly strong for individuals higher on Openness to Experience. This moderating effect is particularly pronounced in the non-familial discussion network. Thus, our analysis provides evidence that certain Big Five traits are associated with variation in who is exposed to political disagreement and, as a consequence, whose attitudes are most likely to be changed by discussion (e.g., Visser and Mirabile 2004). As we have noted, the results reported in Table 2 linking personality and frequency of discussion are different for the family and non-family networks. Because we did not have specific expectations that the relationships between Big Five traits and frequency of discussion would differ across familial and non-familial discussion partners we wish to be cautious about drawing too much from these findings. However, there are a number of plausible explanations for these differences, including fewer choices about who the most frequent familial discussion partner will be, differences in the process of topic selection across networks, and genetic factors, each of which we discuss in more detail in Section 5 of the 23

Supporting Information document. We emphasize that these explanations are speculative, and that future research could more rigorously examine these possibilities. Before concluding, we note that our analysis has several important limitations related to the use of observational survey data and the context of our survey. First, because our analysis relies on observational survey data, we have attempted to refrain from strong causal claims about the relationship between reported agreement on politics and frequency of political discussion. There is some evidence that talking about politics more frequently with a discussion partner results in greater agreement over time (see Huckfeldt and Mendez 2008). Although we interpret our findings as suggesting disagreement drives topic selection, our design does not allow us to rule out this possible reverse causality. Additionally, we note that people may maintain discussion partners because they hold similar views and part ways with discussion partners who hold disparate ones (see McPherson, Smith-Lovin, and Cook 2001). To address the issue of causal direction we also included a survey experiment on our follow-up survey that provides preliminary evidence that disagreement causes less frequent discussion. The experiment asked respondents to imagine that they had two friends who would be meeting for the first time and provided basic information about the friends’ approaches to parenting, taste in food, and political preferences. Whether the two hypothetical friends agreed on each of these topics was randomly assigned. Respondents were then asked whether they would recommend the two individuals discuss each of four topics—food and dining, parenting, politics, and the weather—when they met. Support for talking about a topic was measured on four-point scales ranging from “Definitely avoid” to “Definitely talk about.” We then examined whether (randomly assigned) agreement on each of the three topics mentioned in the descriptions of the hypothetical individuals affected the frequency of recommended discussion for each topic. The results of separate OLS regressions for each of the four topics are presented in Supporting Information Table G, where we find that people were significantly more likely to suggest that the individuals discuss a topic on which they agreed. Randomly assigned agreement on a topic (e.g., politics) is a statistically significant and strong predictor of whether the topic (e.g., politics) should be discussed. Furthermore, agreement on politics also appears to substitute for discussion of other topics, as it is 24

statistically significant and negative predictor of each of the other three topics (which is not the case for the other two randomly assigned topics, parenting and food and dining). This survey experiment is of course hardly definitive evidence that there is no important reverse causality, but rather suggests that the causal arrow at least flows in one direction: people recommend avoiding discussion of topics where there is disagreement. A limitation of the survey experiment is that it is possible that discussing topics where there is disagreement should be avoided in initial meetings, but these topics are stimulating areas to explore in established relationships. In a different experimental set-up Hayes, Glynn, and Shanahan (2005b) find that people are less willing to express their true opinion to a (hypothetical) group of people who hold views that differ from them compared to a group of people who hold views similar to their own across a range of issues (death penalty, affirmative action, and environmental activism). Although limited in scope, these experiments suggest that people tend to see avoiding topics that are likely to lead to disagreement as a good social strategy. Another question we asked on the CCAP survey also addresses this concern. The question asked respondents whether they prefer to avoid sensitive topics in general, not just politics, “which people may disagree about” (see the Supporting Information document for question wording and coding details). In a model where this measure is regressed on the Big Five and demographic controls (gender, race, age, and education), we find that more Extraverted and more Emotionally Stable individuals are less likely to state a preference for avoiding sensitive topics (see column [4] of Supporting Information Table F). In other words, more Extraverted and more Emotionally Stable individuals report a greater willingness to discuss sensitive topics. This suggests that although avoiding topics that are likely to lead to disagreement may be a generally good social strategy, there are personality characteristics of the individuals that predict such an avoidance strategy. These are the same traits for which we found a weaker relationship between agreement and frequency of discussion. But, in contrast to our (unexpected) finding that individuals high on Openness are more likely to avoid discussing political matters in the presence of disagreement (particularly in the non-family network), we find that Openness is associated with a greater reported

25

preference for disagreement on sensitive topics and lower reported avoidance of discussing sensitive topics. A second limitation of our analysis is that all our measures are based on self-reports and may be subject to measurement error. For example, our measures of the Big Five personality traits rely on responses to the TIPI. Random measurement error in independent variables like the Big Five tends to attenuate the estimated relationships between these measures and outcomes. Bias due to systematic measurement error may arise, however, if errors in these measures are correlated with respondent attitudes or behaviors that are correlated with the outcome variables and not controlled for in the statistical analysis. If, for instance, responses to the TIPI are influenced by perceptions of social desirability, and the degree to which the measures are affected is correlated with omitted factors that are also correlated with the measured outcome variables, the regression estimates will be biased. A third limitation is that the CCAP was fielded during a heated presidential election campaign and the sample consists of only self-reported registered voters. Our follow-up survey, however, was conducted in February 2011, outside of an active campaign period and included unregistered individuals. Nonetheless, the results presented here should be replicated in different contexts and with different sampling strategies. Future work could also consider other psychological characteristics that may be relevant to the extent to which an individual discusses politics. For example, willingness to self-censor (Hayes, Glynn, and Shanahan 2005a) and need for cognition (Cacioppo and Petty 1982; Cacioppo, Petty, and Kao 1984; Cohen, Stotland, and Wolfe 1955)—an individual’s tendency to “engage in and enjoy effortful thought” (Sadowski and Cogburn 1997, 307)—also likely shape the extent to which individuals discuss politics, an often engaging topic that sometimes requires effortful thought. These limitations notwithstanding, this paper contributes to our understanding of the role discussion networks may play in shaping political attitudes by situating politics in the broader context of discussion that occurs in people’s everyday lives. Our findings also add to the growing body of evidence that personality traits, and Big Five personality traits in particular, affect how people participate in the political process, including influencing whether they are willing to participate at all. 26

References Antonioni, David. 1993. “Relationship between the Big Five Personality Factors and Conflict Management Styles.” International Journal of Conflict Management 9: 336-355. Beck, Paul A. 2002. “Encouraging Political Defection: The Role of Personal Discussion Networks in Partisan Desertions to the Opposition Party and Perot Votes in 1992.” Political Behavior 24: 309337. Berelson, Bernard R., Paul F. Lazarsfeld, and William N. McPhee. 1954. Voting: A Study of Opinion Formation in a Presidential Campaign. Chicago: University of Chicago Press. Bouchard, Thomas J., Jr. 1997. “The Genetics of Personality.” In Handbook of Psychiatric Genetics, ed. Kenneth Blum and Ernest P. Noble. Boca Raton, FL: CRC Press, Inc., 273-296. Cacioppo, John T., and Richard E. Petty. 1982. “The Need for Cognition.” Journal of Personality and Social Psychology 42: 116-131. Cacioppo, John T., Richard E. Petty, and Chuan Feng Kao. 1984. “The Efficient Assessment of Need for Cognition.” Journal of Personality Assessment 48: 306-307. Caprara, Gian Vittorio, Claudio Barbaranelli, and Philip G. Zimbardo. 1999. “Personality Profiles and Political Parties.” Political Psychology 20: 175-197. Caprara, Gian Vittorio, Shalom Schwartz, Cristina Capanna, Michele Vecchione, and Claudio Barbaranelli. 2006. “Personality and Politics: Values, Traits, and Political Choice.” Political Psychology 27: 1-28. Carney, Dana R., John T. Jost, Samuel D. Gosling, and Jeff Potter. 2008. “The Secret Lives of Liberals and Conservatives: Personality Profiles, Interaction Styles, and the Things They Leave Behind.” Political Psychology 29: 807-840. Caspi, Avshalom, Brent W. Roberts, and Rebecca L. Shiner. 2005. “Personality Development: Stability and Change.” Annual Review of Psychology 56: 453-484. Christakis, Nicholas A., and James H. Fowler. 2009. Connected: The Surprising Power of Our Social Networks and How They Shape Our Lives. Boston: Little, Brown. Cohen, Arthur R., Ezra Stotland, and Donald M. Wolfe. 1955. “An Experimental Investigation of Need for Cognition.” Journal of Abnormal and Social Psychology 51: 291-294. Costa, Paul T., and Robert R. McCrae. 1992. NEO PI-R. Professional Manual. Odessa, FL: Psychological Assessment Resources, Inc. Finifter, Ada W. 1974. “The Friendship Group as a Protective Environment for Political Deviants.” American Political Science Review 68: 607-625. Gerber, Alan S., Gregory A. Huber, David Doherty, and Conor M. Dowling. 2011a. “Personality and the Strength and Direction of Partisan Identification.” Political Behavior DOI: 10.1007/s11109-0119178-5. Gerber, Alan S., Gregory A. Huber, David Doherty, and Conor M. Dowling. 2011b. “Personality Traits and the Consumption of Political Information.” American Politics Research 39: 32-84. Gerber, Alan S., Gregory A. Huber, David Doherty, and Conor M. Dowling. 2011c. “Assessing the Stability of Psychological and Political Survey Measures.” Yale University, Typescript. Gerber, Alan S., Gregory A. Huber, David Doherty, Conor M. Dowling, and Shang E. Ha. 2010. “Personality and Political Attitudes: Relationships Across Issue Domains and Political Contexts.” American Political Science Review 104: 111-133. Gerber, Alan S., Gregory A. Huber, David Doherty, Conor M. Dowling, Connor Raso, and Shang E. Ha. 2011. “Personality Traits and Participation in Political Processes.” Journal of Politics 73: 692706. Goldberg, Lewis R. 1992. “The Development of Markers for the Big-Five Factor Structure. Psychological Assessment 4: 26-42. Goldberg, Lewis R. 1993. “The Structure of Phenotypic Personality Traits.” American Psychologist 48: 26-34. Goldberg, Lewis R., John A. Johnson, Herbert W. Eber, Robert Hogan, Michael C. Ashton, C. Robert 27

Cloninger, and Harrison G. Gough. 2006. “The International Personality Item Pool and the Future of Public-Domain Personality Measures.” Journal of Research in Personality 40: 84-96. Gosling, Samuel D. 2008. Snoop: What Your Stuff Says about You. New York: Basic Books. Gosling, Samuel D., Peter J. Rentfrow, and William B. Swann, Jr. 2003. “A Very Brief Measure of the Big-Five Personality Domains.” Journal of Research in Personality 37: 504-528. Graziano, William G., and Nancy Eisenberg. 1997. “Agreeableness: A Dimension of Personality.” In Handbook of Personality Psychology, eds. R. Hogan, J. A. Johnson, and S. R. Briggs. San Diego, CA: Academic Press, 795-824. Hayes, Andrew F., Carroll J. Glynn, and James Shanahan. 2005a. “Validating the Willingness to SelfCensor Scale: Individual Differences in the Effect of the Climate of Opinion on Opinion Expression.” International Journal of Public Opinion Research 17: 443-455. Hayes, Andrew F., Carroll J. Glynn, and James Shanahan. 2005b. “Willingness to Self-Censor: A Construct and Measurement Tool for Public Opinion Research.” International Journal of Public Opinion Research 17: 298-323. Hayes, Andrew F., Dietram A. Scheufele, and Michael E. Huge. 2006. “Nonparticipation as SelfCensorship: Publicly Observable Political Activity in a Polarized Opinion Climate.” Political Behavior 28: 259-283. Heaney, Michael T., and Scott D. McClurg. 2009. “Social Networks and American Politics: Introduction to the Special Issue.” American Politics Research 37: 727-741. Hibbing, Matthew V., Melinda Ritchie, and Mary R. Anderson. 2010. “Personality and Political Discussion.” Political Behavior DOI: 10.1007/s11109-010-9147-4. Hogan, Joyce, & Deniz. S. Ones. 1997. “Conscientiousness and Integrity at Work.” In Handbook of Personality Psychology, eds. R. Hogan, J. A. Johnson, and S. R. Briggs. San Diego, CA: Academic Press, 849-870. Huckfeldt, Robert. 2007. “Unanimity, Discord, and the Communication of Public Opinion.” American Journal of Political Science 51: 978-995. Huckfeldt, Robert, Paul Allen Beck, Russell J. Dalton, and Jeffrey Levine. 1995. “Political Environments, Cohesive Social Groups, and the Communication of Public Opinion.” American Journal of Political Science 39: 1025-1054. Huckfeldt, Robert, Paul E. Johnson, and John Sprague. 2002. “Political Environments, Political Dynamics, and the Survival of Disagreement.” Journal of Politics 64: 1-21. Huckfeldt, Robert, Jeffrey Levine, William Morgan, and John Sprague. 1998. “Election Campaigns, Social Communication, and the Accessibility of Perceived Discussant Preference.” Political Behavior 20: 263-294. Huckfeldt, Robert, and Jeanette Morehouse Mendez. 2008. “Moths, Flames, and Political Engagement: Managing Disagreement within Communication Networks.” Journal of Politics 70: 83-96. Huckfeldt, Robert, Jeanette Morehouse Mendez, and Tracy Osborn. 2004. “Disagreement, Ambivalence, and Engagement: The Political Consequences of Heterogeneous Networks.” Political Psychology 25: 65-95. Huckfeldt, Robert, and John Sprague. 1987. “Networks in Context: The Social Flow of Political Information.” American Political Science Review 81: 1197-1215. Huckfeldt, Robert, and John Sprague. 1991. “Discussant Effects on Vote Choice: Intimacy, Structure, and Interdependence.” Journal of Politics 53: 122-158. Huckfeldt, Robert, and John Sprague. 1995. Citizens, Politics, and Social Communication. New York: Cambridge University Press. Jackman, Simon, and Lynn Vavreck. 2009. THE 2008 COOPERATIVE CAMPAIGN ANALYSIS PROJECT, Release 2.1. Palo Alto, CA USA: YouGov/Polimetrix, Inc., [producer and distributor]. Jang, Seung-Jin. 2009. “Are Diverse Political Networks Always Bad for Participatory Democracy? Indifference, Alienation, and Political Disagreements.” American Politics Research 37: 879-898. John, Oliver P., Eileen M. Donahue, and Robert L. Kentle 1991. The Big Five Inventory-Versions 4a and 28

54. Berkeley, CA: University of California, Berkeley, Institute of Personality and Social Research. John, Oliver P., and Sanjay Srivastava. 1999. “The Big Five Trait Taxonomy: History, Measurement, and Theoretical Perspectives.” In Handbook of Personality: Theory and Research, ed. Lawrence A. Pervin and Oliver P. John. New York: Guilford Press, 102-138. Klofstad, Casey A. 2009. “Civic Talk and Civic Participation: The Moderating Effect of Individual Predispositions.” American Politics Research 37: 856-878. Klofstad, Casey A., Scott D. McClurg, and Meredith Rolfe. 2009. “Measurement of Political Discussion Networks: A Comparison of Two ‘Name Generator’ Procedures.” Public Opinion Quarterly 73: 462-483. Lake, Ronald La Due, and Robert Huckfeldt. 1998. “Social Capital, Social Networks, and Political Participation.” Political Psychology 19: 567-584. Lazer, David, Brian Rubineau, Carol Chetkovich, Nancy Katz, and Michael Neblo. 2010. “The Coevolution of Networks and Political Attitudes.” Political Communication 27: 248-274. Leighley, Jan E. 1990. “Social Interaction and Contextual Influences on Political Participation.” American Politics Research 18: 459-475. Levitan, Lindsey Clark, and Penny S. Visser. 2008. “The Impact of the Social Context on Resistance to Persuasion: Effortful versus Effortless Responses to Counter-Attitudinal Information.” Journal of Experimental Social Psychology 44: 640-649. Levitan, Lindsey Clark, and Penny S. Visser. 2009. “Social Network Composition and Attitude Strength: Exploring the Dynamics within Newly Formed Social Networks.” Journal of Experimental Social Psychology 45: 1057-1067. McAdams, Dan P. 1995. “What Do We Know When We Know a Person?” Journal of Personality 63: 365-396. McClurg, Scott D. 2003. “Social Networks and Political Participation: The Role of Social Interaction in Explaining Political Participation.” Political Research Quarterly 56: 448-464. McClurg, Scott D. 2006. “The Electoral Relevance of Political Talk: Examining Disagreement and Expertise Effects in Social Networks on Political Participation.” American Journal of Political Science 50: 737-754. McCrae, Robert R. 1996. “Social Consequences of Experiential Openness.” Psychological Bulletin 120: 323-337. McCrae, Robert R., and Paul T. Costa, Jr. 1996. “Toward a New Generation of Personality Theories: Theoretical Contexts for the Five-factor Model.” In The Five-factor Model of Personality: Theoretical Perspectives, ed. J. S. Wiggins. New York: Guilford. McPherson, Miller, Lynn Smith-Lovin, and James M. Cook. 2001. “Birds of a Feather: Homophily in Social Networks.” Annual Review of Sociology 27: 415-444. Mondak, Jeffery J. 2010. Personality and the Foundations of Political Behavior. New York: Cambridge University Press. Mondak, Jeffery J., Damarys Canache, Mitchell A. Seligson, and Matthew V. Hibbing. 2011. “The Participatory Personality: Evidence from Latin America.” British Journal of Political Science 41: 211-221. Mondak, Jeffery J., and Karen D. Halperin. 2008. “A Framework for the Study of Personality and Political Behaviour.” British Journal of Political Science 38: 335-362. Mondak, Jeffery J., Matthew V. Hibbing, Damarys Canache, Mitchell A. Seligson, and Mary R. Anderson. 2010. “Personality and Civic Engagement: An Integrative Framework for the Study of Trait Effects on Political Behavior.” American Political Science Review 104: 85-110. Mutz, Diana C. 2002a. “The Consequences of Cross-Cutting Networks for Political Participation.” American Journal of Political Science 46: 838-855. Mutz, Diana C. 2002b. “Cross-Cutting Social Networks: Testing Democratic Theory in Practice.” American Political Science Review 96: 111-126. Mutz, Diana C., and Jeffery J. Mondak. 2006. “The Workplace as a Context for Cross-Cutting Political 29

Discourse.” Journal of Politics 68: 140-155. Nickerson, David W. 2008. “Is Voting Contagious? Evidence from Two Field Experiments.” American Political Science Review 102: 49-57. Ozer, Daniel J., and Verónica Benet-Martinez. 2006. “Personality and the Prediction of Consequential Outcomes.” Annual Review of Psychology 57: 401-421. Pattie, C. J., and R. J. Johnston. 2009. “Conversation, Disagreement and Political Participation.” Political Behavior 31: 261-285. Paunonen, Sampo V., and Michael C. Ashton. 2001. “Big Five Factors and Facets and the Prediction of Behavior.” Journal of Personality and Social Psychology 81: 524-539. Plomin, Robert, John C. DeFries, Gerald E. McClearn, and Peter McGuffin. 1990. Behavioral Genetics: A Primer. New York: W.H. Freeman & Company. Rentfrow, Peter J., John T. Jost, Samuel D. Gosling, and Jonathan Potter. 2009. “Statewide Differences in Personality Predict Voting Patterns in 1996-2004 U.S. Presidential Elections.” In Social and Psychological Bases of Ideology and System Justification, eds. John T. Jost, Aaron C. Kay, and Hulda Thorisdottir. New York: Oxford University Press, 314-347. Sadowski, Cyril J., and Helen E. Cogburn. 1997. “Need for Cognition in the Big-Five Factor Structure.” The Journal of Psychology: Interdisciplinary and Applied 131: 307-312. Saucier, Gerard. 1994. “Mini-Markers: A Brief Version of Goldberg's Unipolar Big-Five Markers.” Journal of Personality Assessment 63: 506-516. Scheufele, Dietram A., Matthew C. Nisbet, Dominique Brossard, and Erik C. Nisbet. 2004. “Social Structure and Citizenship: Examining the Impacts of Social Setting, Network Heterogeneity, and Informational Variables on Political Participation.” Political Communication 21: 315-338. Schoen, Harald, and Siegfried Schumann. 2007. “Personality Traits, Partisan Attitudes, and Voting Behavior. Evidence from Germany.” Political Psychology 28: 471-498. Sinclair, Betsy. 2009. “The Multi-Valued Treatment Effects of Political Networks and Context: When Does a Democrat Vote Like a Republican?” Typescript. University of Chicago. Srivastava, Sanjay, Oliver P. John, Samuel D. Gosling, and Jeff Potter. 2003. “Development of Personality in Early and Middle Adulthood: Set Like Plaster or Persistent Change? Journal of Personality and Social Psychology 84: 1041-53. Van Gestel, S., and C. Van Broeckhoven. 2003. “Genetics of Personality: Are we Making Progress?” Molecular Psychiatry 8: 840-52. Visser, Penny S., and Robert R. Mirabile. 2004. “Attitudes in the Social Context: The Impact of Social Network Composition on Individual-Level Attitude Strength.” Journal of Personality and Social Psychology 87: 779-795. Watson, David, and Lee Anna Clark. 1997. “Extraversion and Its Positive Emotional Core.” In Handbook of Personality Psychology, eds. R. Hogan, J. A. Johnson, and S. R. Briggs. San Diego, CA: Academic Press, 767-793. Wyatt, Robert O., Elihu Katz, and Joohan Kim. 2000. “Bridging the Spheres: Political and Personal Conversation in Public and Private Spaces.” Journal of Communication 50: 71-92.

30

Table 1a. Family Network: Agreement on Issue and Frequency of Discussion (specification 1) (1) (2) (3) (4) (5) (6)

Agreement on issue with family, relative to mean

Freq. discuss Freq. discuss Freq. discuss Freq. discuss Freq. discuss family with work with god with family, sports with food with family, relative family, relative relative to family, relative family, relative to mean to mean mean to mean to mean 0.318 0.510 0.461 0.560 0.294 [0.083]*** [0.087]*** [0.062]*** [0.089]*** [0.064]***

Freq. discuss entertainment with family, relative to mean 0.425 [0.083]***

Agreement on issue with family (-1=DA;0=A & DA,DK;1=A) Talk once a week Talk a few times a week Talk daily Female = 1 Black = 1 Hispanic = 1 Other (Native American,Asian,Mixed,Other) = 1 Age (in years) Age^2/100 Educ
-0.493 [0.282]* -0.338 [0.224] -0.340 [0.205]* 0.107 [0.089] 0.039 [0.168] 0.131 [0.164] -0.107 [0.120] 0.043 [0.025]* -0.041 [0.025] 0.193 [0.182] 0.109 [0.135] 0.043 [0.154] -0.086 [0.138] 0.103 [0.138] -0.432 [0.668] 682 0.071

-0.051 [0.257] -0.073 [0.204] -0.243 [0.180] 0.013 [0.081] -0.369 [0.141]*** -0.177 [0.158] 0.224 [0.148] 0.047 [0.018]*** -0.065 [0.019]*** 0.094 [0.193] 0.138 [0.127] 0.233 [0.151] 0.247 [0.110]** 0.303 [0.124]** -0.340 [0.464] 682 0.241

-0.291 [0.230] 0.274 [0.217] 0.114 [0.194] 0.016 [0.073] 0.303 [0.124]** 0.045 [0.162] 0.026 [0.167] -0.008 [0.015] 0.009 [0.016] 0.249 [0.222] -0.042 [0.103] 0.320 [0.127]** -0.054 [0.109] -0.072 [0.128] -0.276 [0.404] 682 0.163

-0.193 [0.282] -0.268 [0.292] 0.025 [0.245] 0.004 [0.097] 0.163 [0.142] 0.119 [0.154] -0.114 [0.194] -0.039 [0.022]* 0.048 [0.023]** -0.419 [0.272] -0.151 [0.149] -0.164 [0.188] -0.146 [0.163] -0.102 [0.165] -0.037 [0.598] 682 0.127

0.386 [0.319] 0.359 [0.263] 0.427 [0.253]* -0.047 [0.063] -0.105 [0.083] 0.037 [0.122] 0.026 [0.133] -0.008 [0.013] 0.009 [0.013] 0.095 [0.151] -0.096 [0.106] -0.259 [0.119]** -0.239 [0.094]** -0.248 [0.100]** 0.217 [0.392] 682 0.100

0.612 [0.248]** 0.262 [0.206] 0.446 [0.168]*** -0.051 [0.078] 0.254 [0.109]** -0.136 [0.151] -0.028 [0.176] -0.041 [0.017]** 0.040 [0.017]** -0.176 [0.200] -0.010 [0.118] -0.285 [0.165]* 0.082 [0.119] -0.172 [0.134] 0.398 [0.439] 682 0.117

Note: OLS/ordered probit (column 11) regression coefficients with robust standard errors in brackets. Weighted analysis. Cut points for ordered probit model in column (11) are: .31, .90, and 2.13. * significant at 10%; ** significant at 5%; *** significant at 1% (two-tailed) Source: 2008 CCAP.

(7)

(8)

(9)

(10)

(11)

Freq. discuss Freq. discuss Freq. discuss Freq. discuss Freq. discuss politics with politics with politics with politics with politics with family family family, relative family, relative family, relative (0=Never;3=Oft (0=Never;3=Oft to mean to mean to mean en) en) 0.343 0.324 [0.071]*** [0.075]*** 0.242 0.256 0.450 [0.061]*** [0.059]*** [0.094]*** 0.085 0.05 0.411 0.569 [0.326] [0.346] [0.246]* [0.335]* -0.179 -0.201 0.405 0.625 [0.317] [0.332] [0.244]* [0.331]* -0.393 -0.447 0.489 0.790 [0.307] [0.326] [0.231]** [0.304]*** -0.028 -0.065 -0.013 -0.018 0.001 [0.071] [0.075] [0.074] [0.073] [0.130] -0.258 -0.239 -0.232 -0.24 -0.342 [0.124]** [0.132]* [0.131]* [0.192] [0.269] -0.002 -0.016 0.024 -0.163 -0.249 [0.134] [0.130] [0.138] [0.134] [0.214] 0.023 0.093 0.038 0.124 0.194 [0.155] [0.208] [0.162] [0.152] [0.329] 0.006 0.004 0.009 0.035 0.056 [0.015] [0.015] [0.016] [0.019]* [0.029]** -0.001 0.002 -0.005 -0.031 -0.049 [0.016] [0.015] [0.016] [0.019] [0.029]* -0.033 -0.031 -0.006 -0.024 -0.081 [0.226] [0.242] [0.231] [0.197] [0.364] 0.050 0.087 0.09 -0.07 -0.064 [0.108] [0.117] [0.112] [0.121] [0.192] 0.117 0.129 0.142 0.301 0.613 [0.095] [0.100] [0.096] [0.113]*** [0.247]** 0.215 0.223 0.207 0.178 0.382 [0.086]** [0.085]*** [0.089]** [0.103]* [0.183]** 0.164 0.198 0.154 0.07 0.146 [0.089]* [0.092]** [0.089]* [0.106] [0.188] 0.433 0.199 0.292 1.059 [0.475] [0.362] [0.505] [0.524]** 682 682 682 682 682 0.151 0.097 0.127 0.127

Table 1b. Non-Family Network: Agreement on Issue and Frequency of Discussion (specification 1) (1) (2) (3) (4) (5) (6)

Agreement on issue with non-family, relative to mean

(7)

Freq. discuss Freq. discuss politics with Freq. discuss Freq. discuss Freq. discuss Freq. discuss Freq. discuss entertainment family with non- work with non- god with non- sports with non- food with non- with non-family, non-family, relative to relative to family, relative family, relative family, relative family, relative family, relative to mean to mean mean mean to mean to mean to mean 0.467 0.736 0.588 0.608 0.493 0.897 0.432 [0.082]*** [0.116]*** [0.087]*** [0.112]*** [0.112]*** [0.127]*** [0.092]***

(8) Freq. discuss politics with non-family, relative to mean 0.446 [0.098]***

Agreement on issue with non-family (-1=DA;0=A & DA,DK;1=A) Talk once a week Talk a few times a week Talk daily Female = 1 Black = 1 Hispanic = 1 Other (Native American,Asian,Mixed,Other) = 1 Age (in years) Age^2/100 Educ
-0.009 [0.170] -0.087 [0.157] -0.328 [0.163]** 0.464 [0.078]*** 0.109 [0.148] -0.029 [0.149] -0.139 [0.153] 0.002 [0.019] -0.003 [0.019] -0.549 [0.191]*** -0.057 [0.120] -0.202 [0.152] -0.118 [0.122] -0.109 [0.119] 0.197 [0.494] 682 0.188

0.303 [0.146]** 0.332 [0.127]*** 0.302 [0.131]** -0.199 [0.090]** -0.257 [0.137]* -0.174 [0.236] -0.147 [0.166] 0.052 [0.019]*** -0.062 [0.021]*** -0.021 [0.186] 0.241 [0.142]* 0.000 [0.171] 0.235 [0.127]* 0.061 [0.158] -0.801 [0.456]* 682 0.238

-0.106 [0.114] -0.111 [0.108] -0.184 [0.112] 0.251 [0.088]*** 0.127 [0.186] 0.092 [0.173] -0.020 [0.167] -0.008 [0.017] 0.008 [0.018] 0.696 [0.327]** 0.054 [0.104] 0.056 [0.141] -0.026 [0.108] -0.045 [0.141] -0.189 [0.391] 682 0.169

-0.199 [0.152] -0.252 [0.128]** -0.188 [0.138] -0.677 [0.098]*** 0.032 [0.152] -0.217 [0.149] -0.374 [0.214]* -0.012 [0.017] 0.016 [0.018] -0.157 [0.325] -0.096 [0.125] 0.151 [0.181] -0.027 [0.134] 0.085 [0.155] 0.208 [0.383] 682 0.200

0.059 [0.132] -0.009 [0.084] 0.203 [0.093]** 0.040 [0.074] -0.039 [0.089] 0.140 [0.120] 0.234 [0.209] 0.014 [0.015] -0.009 [0.016] -0.291 [0.226] -0.159 [0.110] -0.148 [0.150] -0.181 [0.114] -0.152 [0.130] -0.369 [0.319] 682 0.120

-0.108 [0.126] -0.120 [0.125] 0.093 [0.119] 0.177 [0.085]** 0.121 [0.125] 0.051 [0.149] 0.473 [0.155]*** -0.039 [0.018]** 0.039 [0.018]** 0.050 [0.289] -0.112 [0.111] -0.191 [0.161] -0.085 [0.113] -0.084 [0.124] 0.638 [0.407] 682 0.194

Note: OLS/ordered probit (column 11) regression coefficients with robust standard errors in brackets. Weighted analysis. Cut points for ordered probit model in column (11) are: -.77, .04, and 1.21. * significant at 10%; ** significant at 5%; *** significant at 1% (two-tailed) Source: 2008 CCAP.

0.042 [0.138] 0.258 [0.111]** 0.081 [0.106] -0.048 [0.080] -0.052 [0.110] 0.163 [0.155] 0.041 [0.130] -0.012 [0.019] 0.017 [0.019] 0.262 [0.216] 0.102 [0.116] 0.281 [0.163]* 0.174 [0.119] 0.235 [0.112]** 0.377 [0.457] 682 0.145

-0.038 [0.080] -0.042 [0.114] 0.155 [0.156] 0.071 [0.126] -0.009 [0.019] 0.014 [0.019] 0.274 [0.219] 0.127 [0.122] 0.297 [0.160]* 0.164 [0.123] 0.242 [0.113]** 0.427 [0.462] 682 0.129

(9)

(10)

(11)

Freq. discuss Freq. discuss Freq. discuss politics with politics with politics with non-family, non-family non-family relative to (0=Never;3=Oft (0=Never;3=Oft mean en) en)

0.356 [0.074]*** -0.02 [0.154] 0.24 [0.119]** 0.049 [0.114] -0.047 [0.082] -0.04 [0.114] 0.216 [0.159] 0.093 [0.136] -0.005 [0.020] 0.010 [0.019] 0.283 [0.213] 0.131 [0.119] 0.284 [0.162]* 0.183 [0.127] 0.213 [0.116]* 0.089 [0.494] 682 0.133

0.446 [0.077]*** 0.41 [0.174]** 0.718 [0.139]*** 0.744 [0.140]*** -0.016 [0.086] 0.001 [0.165] 0.091 [0.135] 0.271 [0.148]* -0.003 [0.019] 0.000 [0.018] 0.325 [0.220] -0.004 [0.138] 0.179 [0.186] 0.247 [0.131]* 0.281 [0.124]** 1.367 [0.470]*** 682 0.221

0.624 [0.103]*** 0.479 [0.207]** 0.872 [0.168]*** 0.951 [0.177]*** -0.039 [0.116] 0.019 [0.232] 0.142 [0.193] 0.553 [0.284]* -0.005 [0.025] 0.000 [0.025] 0.51 [0.344] 0.002 [0.172] 0.255 [0.247] 0.334 [0.171]* 0.374 [0.171]**

682

Table 2. Big Five Personality Traits and Frequency of Political Discussion (specification 2) (1)

Agreement on issue with family/non-family, relative to mean Agreement on issue with family/non-family (-1=DA;0=A & DA,DK;1=A) Talk once a week Talk a few times a week Talk daily Female = 1 Black = 1 Hispanic = 1 Other (Native American,Asian,Mixed,Other) = 1 Age (in years) Age^2/100 Educ F

(2)

Family Network (3)

(4)

(5)

(6)

Freq. discuss Freq. discuss Freq. discuss Freq. discuss Freq. discuss Freq. discuss politics with politics with politics with politics with politics with politics with family family non-family, family, relative family, relative family, relative (0=Never;3=Oft (0=Never;3=Oft relative to to mean to mean to mean en) en) mean 0.309 0.292 0.436 [0.065]*** [0.069]*** [0.088]*** 0.218 0.215 0.403 [0.054]*** [0.051]*** [0.090]*** 0.065 0.048 0.375 0.504 0.056 [0.300] [0.317] [0.203]* [0.295]* [0.137] -0.210 -0.218 0.363 0.557 0.257 [0.284] [0.297] [0.197]* [0.280]** [0.116]** -0.397 -0.432 0.459 0.765 0.094 [0.277] [0.293] [0.185]** [0.254]*** [0.109] -0.026 -0.061 -0.013 -0.015 -0.009 -0.042 [0.070] [0.075] [0.073] [0.071] [0.131] [0.086] -0.269 -0.247 -0.24 -0.261 -0.419 -0.041 [0.127]** [0.136]* [0.133]* [0.193] [0.284] [0.111] 0.006 -0.012 0.03 -0.14 -0.226 0.148 [0.132] [0.126] [0.136] [0.135] [0.220] [0.154] 0.053 0.119 0.064 0.163 0.274 0.032 [0.128] [0.167] [0.132] [0.129] [0.289] [0.137] 0.008 0.007 0.011 0.033 0.053 -0.009 [0.014] [0.014] [0.015] [0.017]* [0.028]* [0.017] -0.003 -0.001 -0.005 -0.029 -0.045 0.014 [0.015] [0.014] [0.015] [0.017]* [0.028] [0.017] -0.086 -0.09 -0.062 -0.086 -0.162 0.261 [0.211] [0.221] [0.214] [0.178] [0.343] [0.213] 0.013 0.036 0.046 -0.094 -0.108 0.097 [0.105] [0.114] [0.109] [0.113] [0.191] [0.113] 0.047 0.054 0.069 0.212 0.493 0.274 [0.102] [0.106] [0.102] [0.116]* [0.258]* [0.167] 0.146 0.141 0.137 0.123 0.297 0.161 [0.088]* [0.089] [0.090] [0.103] [0.188] [0.121] 0.116 0.14 0.106 0.028 0.107 0.229 [0.093] [0.096] [0.093] [0.104] [0.189] [0.112]** 0.111 0.103 0.113 0.171 0.312 -0.004 [0.039]*** [0.040]*** [0.040]*** [0.037]*** [0.067]*** [0.046] -0.062 -0.064 -0.072 -0.086 -0.15 0.006 [0.037]* [0.042] [0.037]* [0.038]** [0.073]** [0.050] -0.056 -0.074 -0.063 -0.02 -0.054 -0.039 [0.035] [0.038]* [0.034]* [0.033] [0.065] [0.041] 0.082 0.079 0.074 0.104 0.171 0.031 [0.038]** [0.039]** [0.038]* [0.038]*** [0.067]** [0.046] 0.049 0.067 0.053 0.012 0.059 -0.014 [0.042] [0.047] [0.043] [0.038] [0.071] [0.046] 0.449 0.174 0.307 1.192 0.315 [0.451] [0.347] [0.473] [0.488]** [0.431] 682 682 682 682 682 682 0.201 0.152 0.181 0.199 0.147 4.910 4.650 5.290 6.950 34.510 0.270 0.000 0.000 0.000 0.000 0.000 0.930

(7)

Non-Family Network (8)

Freq. discuss politics with non-family, relative to mean 0.448 [0.095]***

-0.034 [0.087] -0.03 [0.116] 0.139 [0.156] 0.062 [0.134] -0.006 [0.018] 0.01 [0.018] 0.27 [0.218] 0.115 [0.118] 0.285 [0.166]* 0.141 [0.124] 0.232 [0.113]** 0.003 [0.046] 0.013 [0.051] -0.055 [0.041] 0.04 [0.046] -0.006 [0.048] 0.367 [0.436] 682 0.134 0.490 0.780

(9)

0.360 [0.075]*** -0.003 [0.149] 0.244 [0.122]** 0.062 [0.114] -0.039 [0.088] -0.02 [0.115] 0.205 [0.159] 0.081 [0.143] -0.003 [0.018] 0.008 [0.018] 0.283 [0.212] 0.122 [0.115] 0.28 [0.164]* 0.172 [0.127] 0.206 [0.114]* -0.005 [0.047] -0.01 [0.051] -0.043 [0.042] 0.013 [0.045] -0.008 [0.046] 0.021 [0.463] 682 0.136 0.260 0.940

0.444 [0.079]*** 0.411 [0.166]** 0.706 [0.142]*** 0.74 [0.141]*** -0.029 [0.094] 0.019 [0.165] 0.094 [0.139] 0.264 [0.149]* -0.005 [0.018] 0.002 [0.018] 0.338 [0.217] 0.019 [0.134] 0.182 [0.186] 0.252 [0.133]* 0.309 [0.124]** 0.065 [0.045] 0.038 [0.054] -0.019 [0.043] 0.022 [0.046] -0.058 [0.052] 1.43 [0.456]*** 682 0.227 0.660 0.660

Note: OLS/ordered probit (columns 5 and 10) regression coefficients with robust standard errors in brackets. Weighted analysis. Cut points for ordered probit model in column (5) are: -.03, .60, and 1.93; column (10) are: -.89, -.07, and 1.10. * significant at 10%; ** significant at 5%; *** significant at 1% (two-tailed) Source: 2008 CCAP.

(10)

Freq. discuss Freq. discuss Freq. discuss politics with politics with politics with non-family, non-family non-family relative to (0=Never;3=Oft (0=Never;3=Oft mean en) en)

0.622 [0.104]*** 0.477 [0.199]** 0.86 [0.172]*** 0.95 [0.177]*** -0.069 [0.126] 0.043 [0.233] 0.142 [0.198] 0.546 [0.285]* -0.008 [0.024] 0.003 [0.024] 0.541 [0.341] 0.036 [0.170] 0.271 [0.249] 0.348 [0.174]** 0.424 [0.172]** 0.097 [0.062] 0.06 [0.072] -0.02 [0.060] 0.014 [0.062] -0.079 [0.070]

682 3.680 0.600

Table 3. Moderating Effect of Big Five on Relationship between Agreement and Frequency of Political Discussion (specification 3) Family Network Non-Family Network (1) (2) (3) (4)

Agreement on issue with family/non-family, relative to mean Talk once a week Talk a few times a week Talk daily Female = 1 Black = 1 Hispanic = 1 Other (Native American,Asian,Mixed,Other) = 1 Age (in years) Age^2/100 Educ F

Freq. discuss Freq. discuss Freq. discuss Freq. discuss politics with politics with politics with non- politics with nonfamily, relative to family, relative to family, relative to family, relative to mean mean mean mean 0.298 0.282 0.422 0.429 [0.062]*** [0.068]*** [0.081]*** [0.084]*** 0.072 0.080 [0.276] [0.124] -0.202 0.249 [0.261] [0.113]** -0.405 0.084 [0.253] [0.109] -0.043 -0.076 -0.075 -0.070 [0.068] [0.073] [0.079] [0.080] -0.249 -0.233 -0.039 -0.031 [0.129]* [0.139]* [0.111] [0.116] -0.022 -0.038 0.177 0.170 [0.137] [0.131] [0.150] [0.152] 0.036 0.105 0.062 0.091 [0.128] [0.167] [0.134] [0.134] 0.007 0.006 -0.012 -0.009 [0.014] [0.014] [0.017] [0.017] -0.002 0.000 0.016 0.013 [0.015] [0.015] [0.017] [0.017] -0.069 -0.078 0.248 0.259 [0.202] [0.214] [0.205] [0.206] 0.060 0.081 0.095 0.112 [0.100] [0.109] [0.108] [0.112] 0.073 0.076 0.253 0.268 [0.104] [0.109] [0.157] [0.157]* 0.160 0.150 0.162 0.147 [0.087]* [0.091]* [0.107] [0.110] 0.132 0.155 0.209 0.214 [0.094] [0.098] [0.108]* [0.110]* 0.107 0.100 0.003 0.010 [0.036]*** [0.037]*** [0.043] [0.043] -0.055 -0.057 0.004 0.011 [0.037] [0.041] [0.046] [0.047] -0.062 -0.079 -0.045 -0.060 [0.034]* [0.038]** [0.040] [0.040] 0.076 0.073 0.038 0.045 [0.036]** [0.037]* [0.046] [0.045] 0.046 0.065 0.000 0.010 [0.040] [0.045] [0.042] [0.043] -0.157 -0.162 -0.126 -0.127 [0.071]** [0.075]** [0.089] [0.091] 0.094 0.073 -0.022 -0.014 [0.068] [0.072] [0.085] [0.088] 0.004 0.039 0.113 0.103 [0.069] [0.081] [0.072] [0.072] -0.133 -0.112 -0.129 -0.137 [0.071]* [0.075] [0.087] [0.089] 0.116 0.106 0.255 0.266 [0.085] [0.101] [0.094]*** [0.095]*** 0.459 0.189 0.376 0.442 [0.443] [0.354] [0.412] [0.409] 682 682 682 682 0.226 0.174 0.178 0.166 1.390 1.140 2.220 2.200 0.230 0.340 0.050 0.050

Note: OLS regression coefficients with robust standard errors in brackets. Weighted analysis. * significant at 10%; ** significant at 5%; *** significant at 1% (two-tailed) Source: 2008 CCAP.

Figure 1a. Average Frequency of Discussion, by Topic

Average Frequency of Talk with Discussion Partner (0=never; 1=rarely; 2=sometimes; 3=often)

3

2

1

0 Family

Work

Religion

Frequency discuss topic with family member

Sports

Food

Entertainment

Frequency discuss topic with non-family member

Politics

Figure 1b. Average Agreement, by Topic

Average Agreement on Topic with Discussion Partner (-1=Mostly Disagrees; 0=Agrees and Disagress & DK; 1=Mostly Agrees)

0.75

0.50

0.25

0.00 Family

Work

Religion

Agreement on topic with family member

Sports

Food

Entertainment

Agreement on topic with non-family member

Politics