Out of Sync, Out of Society: Political Beliefs and Social Networks

Abstract

Who is more likely to be isolated from society in terms of political beliefs? To answer this question, we measure whether individuals’ beliefs are “out of sync” – the extent to which their views differ with their contemporaries – and examine how the level of synchronization is associated with the size of important-matter and political-matter discussion networks. The results show that people with weaker belief synchronization are more likely to have smaller important-matter discussion networks. However, additional analyses of political-matter discussion networks show that weaker belief synchronization is associated with smaller networks only among those without a high school diploma and even provides some advantage in maintaining larger networks for the college-educated. Overall, the results imply that political beliefs that are “out of sync” correspond to the individual being “out of society,” whereas the aspects of “out of society” are quite different among educational groups.

1. Introduction

Scholars have described social isolation in various ways: the decrease in community-based trust and civic engagement (Putnam, 2001); few social identities (Thoits, 1983); the dissolution of the traditional family structure (McLanahan & Sandefur, 1994); the shrinkage in the number of people with whom you can discuss important matters (McPherson, Smith-Lovin, & Brashears, 2006). Social isolation is not only related to the decrease of social activities but also the disconnection to specific social groups through clustering of social actors. For example, network researchers have consistently emphasized that there is noticeable clustering based on socioeconomic traits such as income or education, which induces a stronger differentiation among social strata and a more severe inequality at a population level (DiMaggio & Garip, 2012). From this perspective, it is essential to explore the micro-mechanisms that create population-level difference in social isolation and disconnection among groups.

Political beliefs are closely related to social networks as one of the common topics raised in every discussion with potential network members (Bearman & Parigi, 2004; Bennett, Flickinger, & Rhine, 2000; Gerber, Huber, Doherty, & Dowling, 2012a; Lee & Bearman, 2017). While many political scientists and sociologists have examined the role of social networks in promoting political participation, voting behavior, and political attitudes (Berelson, Lazarsfeld, & McPhee, 1954; Huckfeldt, Mendez, & Osborn, 2004; Huckfeldt & Sprague, 1987; Knoke, 1990; Lazarsfeld, Berelson, & Gaudet, 1948; Lazer, Rubineau, Chetkovich, Katz, & Neblo, 2010; Levitan & Visser, 2009; Mutz, 2002; Visser & Mirabile, 2004), there has been only a few empirical studies investigating how political beliefs foster or hinder the cultivation of social networks.

In this study, we propose a novel theoretical perspective to explain the linkage between political beliefs and social networks and test our models with the empirical data. Previous studies on political beliefs have considered a set of beliefs as a “system” (Converse, 1964; Lakoff, 1996) or “network” (Boutyline & Vaisey, 2017) because of their systemic organization as a whole. Political beliefs are structured by making a metaphor to other value systems in the world (e.g. “nation as family” metaphor, see Lakoff (1996)) or aligned by social constraint (Converse, 1964) such as political party affiliation (Baldassarri & Gelman, 2008; Baldassarri & Goldberg, 2014) or political ideology (Boutyline & Vaisey, 2017). In the multiple processes of structuring, what hides behind the networks among beliefs are human actors. If many people report similar attitudes about political beliefs, those beliefs are expected to covary and closely react to one another’s change. In this study, we approach political beliefs in the opposite direction: we assess the relationship among people by focusing on the organization of multiple political beliefs. If a pair of people share similar opinions over various political issues, they are expected to interact with and influence each other in terms of political beliefs and cultural values. We call this space where individuals are related through political beliefs as belief space.

By measuring the average distance to other people in belief space, we assess whether individuals’ beliefs are “out of sync” with their contemporaries and examine how the level of synchronization is related to individuals’ network size. For this aim, we adopt the data from the General Social Survey and the American National Election Study which have a rich set of political belief items and include survey modules for revealing the number of network members with whom the respondent discuss important (or political) matters. According to the results, a lower level of synchronization is associated with a smaller important-matter discussion network, which suggests that political beliefs that are “out of sync” correspond to the individual being “out of society.” When examining discussion networks focusing on political matters, however, we find no significant relationship with belief synchronization, which is mainly due to its strong interaction with education: a lower level of belief synchronization results in a decrease in network size only for those with no high school diploma, whereas the college-educated enjoy more political discussion even with weaker synchronization. Our findings suggest that there are two stories of “out of sync” and “out of society”: people with low education are likely to be isolated in both general and political senses, whereas those with high education stay connected at least from the perspective of political discussion.

1–1. Belief Space and Belief Synchronization

Blau (1977) first proposed the idea of social space for describing the changes in social structure in modern society. In the pre-modern society where politics, economics, and civil society were strongly unified, the life trajectories of individuals could easily be explained with a few sociodemographic characteristics. However, people in modern society started to enjoy diversified social lives due to a loosening of connections among social institutions, which makes it important to examine multiple demographic and socioeconomic categories together for understanding the heterogeneity in life trajectories.¹ By adopting the concept of Blau space, scholars could explain how the crosscutting of multiple sociodemographic categories fosters or constrains the social interaction among individuals (McPherson, 1983; McPherson & Ranger-Moore, 1991). For example, we can infer that people with the same level of educational attainment and from the same occupational group are more likely to encounter and share everyday lives than those from different educational and occupational backgrounds. These major organizational bases of socioeconomic dimensions generate foci where people find their target of social interactions and cultivate social relationships, which becomes an essential mechanism of network generating processes in modern society (Blau & Schwartz, 1984; Feld, 1981).

The idea of social space can also be found in Bourdieu’s works (1984, 1989). Unlike Blau space where the pre-existing sociodemographic categories are directly adopted as structural constraints on human agencies, Bourdieu argues that the social structure can only be perceived and presented by relative positions in the multi-dimensional social space consisting of various forms of capital. Individuals can get a sense of their social positions over the distributions of multiple resources such as economic, cultural, and symbolic capital, and try to possess a set of resources that can differentiate themselves from others in the society. From this perspective, the “social distance” in the space has special importance: by drawing an analogy to geographic space, Bourdieu explains that “the closer the agents, groups or institutions which are situated within this space, the more common properties they have; and the more distant, the fewer.” (Bourdieu 1989: 16).

Borrowing the concept of social space from Blau and Bourdieu, we propose belief space comprising multiple dimensions of political beliefs. By situating individuals in the space based on their beliefs about various political issues, we can assess where they are, and how far they are, from other people in the society – in other words, how strongly they are synchronized with other people in terms of political beliefs. Political beliefs may not directly constrain the social interactions among individuals as the sociodemographic categories do in Blau space. However, political beliefs, as an organized system of world views (Boutyline & Vaisey, 2017; Converse, 1964), may have a substantial implication not only on political behaviors but also on a wide range of cultural habits and lifestyles (DellaPosta, Shi, & Macy, 2015). From this perspective, we expect that individuals would consider political beliefs as a part of cultural resources that can be utilized for assessing their relative position and search for potential social network members in society.

1–2. Mechanism 1: Homophily

How does the level of belief synchronization matter for social networks? We suggest the foremost mechanism is network homophily, a similarity-based network generating principle (McPherson, Smith-Lovin, & Cook, 2001). The homophily mechanism can be explained in two steps using the concept of social space: baseline homophily and inbreeding homophily (McPherson et al., 2001). We can infer baseline homophily from a direct constraint of demographic dimensions: for example, the intersection of age and residential area intrinsically limits the potential of meeting up with people from different age groups and other districts. Inbreeding homophily, on the other hand, occurs at a more micro level: under the constraints of age and place, people are more likely to interact with someone similar with themselves due to the cross-cutting of additional socioeconomic dimensions and individual inclination for making social ties with similar people. The combination of these two processes leads to the clustering of similar people and the separation of different ones. For this reason, a pair of actors who are closer to each other in Blau space have a higher probability of being friends (McPherson & Ranger-Moore, 1991).

While homophily based on socioeconomic background generally results in cultural similarity among social network members (McPherson et al., 2001), homophily based on values and attitudes can be an independent mechanism of organizing social networks. If social structure determines patterns of baseline homophily, people may actively use their cultural traits for inbreeding homophily within a given structure. Some level of similarity in knowledge is a necessary condition for social interaction since one can exchange new information only based on common concepts and values they share (Carley, 1991; Mark, 2003). Similar values and cultural tastes are helpful not only for easing the difficulty in communication but also promoting conversation by providing more topics that can be discussed, which results in an increase in the level of intimacy and attraction between similar dyads (Collins, 2004; DiMaggio, 1987; Lazarsfeld & Merton, 1954; Rivera, 2012). We can explain homophily based on political attitudes in a similar way: people with similar political beliefs may feel comfortable with their thoughts and behaviors, which would result in a high probability of being friends (Downs, 1957; Huckfeldt & Sprague, 1987). Given a certain level of homophily, people with high levels of belief synchronization have an advantage in cultivating social networks due to their proximity to potential network members, whereas those with only a few individuals in their local area of belief space may be more constrained in establishing large networks.

Even though the notion of value homophily is well acknowledged among social network scholars, there has been skepticism about the practical importance of beliefs themselves. The main concern is that the association between values and social networks may be not consequential and largely driven by baseline homophily based on structural dimensions. For example, McPherson et al. (2001: 429) argue that “it is unclear whether this homophily is due to actual political similarity or similarity on other social characteristics that are correlated with political beliefs.” DellaPosta et al. (2015) also point out that a large amount of association among political ideology, beliefs, and lifestyles may be spurious due to confounding by socioeconomic status and the tendency to have similar attitudes among the same socioeconomic groups (i.e., network autocorrelation). According to this argument, it is essential to consider the baseline sociodemographic characteristics that fundamentally foster the similarity in political beliefs and cultural habits. If one’s structural position strongly constrains his or her chance of interacting with those with similar beliefs, the observed association between belief synchronization and social network size would be mostly spurious. To address this issue, we adopt a broad set of structural factors that can confound the relationship between belief synchronization and test if belief synchronization is an independent predictor of social network size.

1–3. Mechanism 2: Moderate or Polarized Opinions

As discussed above, a person can obtain a high level of belief synchronization when he or she has a logic of organizing multiple beliefs that are common to contemporaries. However, we can think about a more straightforward way of achieving belief synchronization: if a person has moderate opinions over multiple political topics, he or she could automatically reduce the average distance to other people in the society. If the distribution of belief items roughly follows the normal distribution (i.e., dense around the moderate or neutral stance and sparse in the extreme ends), respondents who provide moderate opinions to more belief items would have more people in their proximity in belief space and be advantageous in maintaining higher levels of belief synchronization. Even when the distribution is polarized (i.e., sparse around the mean and dense in the extreme ends), moderate opinions can be effective in staying close to both ends, whereas those with polarized opinions may suffer from weak synchronization with the opposite end.

The number of moderate opinions can also be problematic due to its association with social networks through the confounding of unobserved individual characteristics. For example, studies show that the attitudes towards political issues are strongly influenced by personality traits such as authoritarianism (Hetherington & Weiler, 2009) and the Big Five personality traits (Gerber, Huber, Doherty, & Dowling, 2011; Gerber, Huber, Doherty, Dowling, & Ha, 2010). Specifically, a study by Gerber, Huber, Doherty, and Dowling (2012b) shows that people who are more extroverted, more agreeable, and less open to experience are more likely to have strong partisanship, which implies that moderate opinions may originate from less outgoing but more creative personality traits. Considering that personality is an important basis of establishing social networks (Burt, Kilduff, & Tasselli, 2013; Fang et al., 2015), it is important to remove confounding bias originated from a tendency for holding moderate opinions.

In this study, we are mainly interested in people with similar opinions to their contemporaries, regardless of their opinions being moderate or polarized. For this aim, we employ two analytic strategies. First, when assessing the relationship between belief synchronization and social network size, we condition on the number of moderate responses for clarifying the role of having similar opinions with others and reducing bias due to unmeasured information about individual characteristics. Second, we examine two different measurements for belief synchronization in the study of social networks: 1) the average distance to all other people in belief space and 2) the percentage of people in the proximate distance. Since the second method concentrates on the local space of each individual, it would balance out the difference between moderate and polarized opinions.

1–4. Mechanism 3: Heterogeneity in Social Networks

According to studies on political attitudes and social networks, however, homophily is not the only mechanism which governs the social network formation. Scholars have consistently observed that people usually maintain some level of heterogeneity in political beliefs in their social networks (Huckfeldt et al., 2004; Huckfeldt, Sprague, & Levine, 2000; Mutz, 2002; Visser & Mirabile, 2004). For example, according to a representative study on the US population after the 2000 presidential election, less than half of voters have entirely homogenous networks where all network members support the same candidate as the respondent, whereas more than one-third have at least one supporter for the candidate from the opposite party (Huckfeldt et al., 2004).

While the heterogeneity in social networks can be a result of modernization by which the functions of social institutions are diversified and people inevitably cooperate with those from different socioeconomic backgrounds (Durkheim, 1983), we can find one potential mechanism from a behavioral principle called selective disclosure. Selective discourse includes two behavioral tendencies regarding the selection of discussion topics. First, people are likely to select political issues on which they have similar opinions with their discussion partners. Second, when people find the discrepancy in political beliefs during the discussion, they consciously filter discussion topics and keep silent on controversial issues for avoiding unnecessary conflicts (Baldassarri & Bearman, 2007; Cowan & Baldassarri, 2018; Huckfeldt & Sprague, 1987). By screening discussion topics in order for promoting social interactions among individuals with diverse thoughts, people can maintain social relationship with those with different values.

From this point of view, the positive association between belief synchronization and social network size may be weaker than expected from the mechanism of homophily. Even though we do not directly test the level of heterogeneity in social networks, we expect that the inclination of staying connected with people with different beliefs can be a significant mechanism that can reduce the influence of belief synchronization on social networks.

1–5. Mechanism 4: Education as a Moderator

One compelling structural dimension associated with belief synchronization and social networks is education. As a major social institution for transferring knowledge and culture to the young generation, education has a substantial influence on shaping political attitudes in modern society. Given that higher levels of education are highly correlated with a bigger size and a wider range of social networks (Campbell, Marsden, & Hurlbert, 1986; Huang & Tausig, 1990; Marsden, 1987; McPherson et al., 2006), we expect that education would play an essential role in explaining how belief synchronization leads to the cultivation of social networks.

Previous studies on political behaviors emphasize two distinctive features of educated people. First, a longer duration of educational training increases the amount of civic and political knowledge (Carpini & Keeter, 1996; Popkin & Dimock, 1999), which leads to better understanding of politics and public policies and promotes their participation in political activities (Dee, 2004; Galston, 2001; Mayer, 2011). According to studies of belief systems, the advantage of education is from the cultivation of abilities to interpret abstract and complicated concepts through the lens of a dominant ideological system in the society, which results in better organized and aligned political beliefs among those with higher education (Boutyline & Vaisey, 2017; Converse, 1964; Gordon & Segura, 1997). These findings imply that people with higher education would be more advantageous in maintaining higher belief synchronization by situating themselves along the line of standard ideological system in belief space using their deeper knowledge about political matters. In this case, education is a confounder that fosters a spurious association between belief synchronization and social networks.

Second, education transfers not only knowledge but also cultural values such as civic liberty and tolerance to diversity (Golebiowska, 1995; Phelan, Link, Stueve, & Moore, 1995; Weakliem, 2002). The value of tolerance can be derived from the organized political beliefs of highly-educated people: using a more sophisticated belief system and better knowledge about the political geography in the society, people with higher education can better understand heterogeneity in opinions and become tolerant toward extremes and minorities (Bobo & Licari, 1989; Carpini & Keeter, 1996; Gabennesch, 1972; Lipsitz, 1965). Tolerance also arises from the direct experience of heterogeneity: educational institutions provide an environment where students from diverse socioeconomic backgrounds should interact and cooperate, which would automatically train to be open to diverse thoughts and behaviors (Mutz, 2002; Stouffer, 1955). According to this argument, highly educated people would be better at communicating with those with heterogeneous people, which echoes previous findings that higher education leads to social networks of more diverse backgrounds. In such a case, education may be a moderator that strengthens or weakens the association between belief synchronization and social networks.

The moderating role of education is noteworthy in light of recent studies reporting the limited role of political attitudes in cultivating social networks. Lazer et al. (2010), using longitudinal social network data from 164 individuals, show that the similarity in political attitudes is not a significant predictor of being socially connected, which correspond to the findings from Levitan and Visser (2009) that the effect of attitudinal similarity is inconsistent across issue domains in constructing new social networks. Both studies argue that the similarity in opinions seems to emerge through social interaction rather than causes the selection of social network members. Given that the participants of two studies were recruited from undergraduate and graduate programs, we infer that the strength of value homophily can be largely weaker along with the increase in the level of education. Instead, the social environment open to diversity may foster more discussion among those with heterogeneous backgrounds and opinions. By employing the interaction between belief synchronization and education, we check if education matters as a moderator in the process of social network generation.

2. Methods

2–1. Data

The data are from two nationally representative and repeated cross-sectional studies in the USA: the General Social Survey (GSS) and the American National Election Study (ANES). There are two reasons for using the data from these two studies. First, the GSS and ANES have incorporated an extensive set of political belief items over time. Previous studies on political polarization and the changes in political belief systems in the USA heavily depended on the data from both studies (Baldassarri & Gelman, 2008; Baldassarri & Goldberg, 2014; Boutyline & Vaisey, 2017; DiMaggio, Evans, & Bryson, 1996; Evans, 2003; Park, 2018). Using those political belief items, we construct belief space where individuals are positioned based on the combination of their political beliefs. Second, both studies incorporated similar egocentric network modules, which enables us to investigate the relationship between political beliefs and social network characteristics. In this study, we assess social networks using the network modules from the 1985, 1987, and 2010 GSS,² and 2000 ANES³ to test if the position in belief space corresponds to the size of social networks.⁴

2–2. Belief Space

Belief space is Euclidean space consisting of multiple dimensions of political beliefs.⁵ Individuals are situated in the space based on their answers to the political belief items, which make it possible to assess how close or distant they are to each other in terms of political beliefs. In this study, we construct year-specific belief spaces for measuring the distance between each pair of people in the same period.⁶ The construction of the belief space follows three steps:

1. We select belief items that are comparable to the previous studies on political belief systems by Baldassarri and Goldberg (2014) and Boutyline and Vaisey (2017). While these two studies are based on the data from ANES, we adopted as many corresponding items regarding economic, civic, and moral issues as possible from the GSS, which results in 47, 47, 61, and 46 belief items for the 1985, 1987, 2010 GSS and 2000 ANES respectively (see Table A1 in the appendix).⁷

2. We identify the dimensions of the belief space using principal component analysis (PCA) (Pearson, 1901; Wold, Esbensen, & Geladi, 1987).⁸ PCA is a commonly-used method for transforming a set of variables into a fewer number of linearly-independent vectors which explain as much of total variance in the original data as possible. The main reason for using principal component scores rather than the original belief items is that the unbalanced number of questions among belief domains can induce a bias when calculating the position in belief space. For example, if the model incorporates more belief items about moral issues than about civic issues, the distance between person A and B in belief space would be closer when they have more similar attitudes toward moral issues than to civic issues. For addressing this issue, we retain principal components that are expected to represent significant belief domains in a given period and standardize component scores so that every issue domain has zero mean and unit variance. We identified 12, 12, 18, and 10 principal components from the 1985, 1987, 2010 GSS and 2000 ANES respectively, and adopted the standardized component scores as dimensions for each year-specific belief space (see Table A2–5 in the appendix).

3. Before calculating the dimension scores, we estimate missing values in political belief items through multiple imputation (MI).⁹ We can identify the coordinate in the political belief space only when the respondents answered all political belief items, which leads to the exclusion of many individuals with only a few missing values. This is especially the case for the 2010 GSS where the majority of political belief items were assigned to randomly-selected subsamples, which results in no survey participant with the full information about all 61 belief items.¹⁰ To address this issue, we generate 100 datasets with imputed missing values under the assumption that the data missing occurred completely at random – especially for the most missing values in the GSS data – or only depending on the observed answers to political belief items, and the joint distribution among belief items follows multivariate normal distribution (Allison, 2001; Schafer, 1997). As a result, we imputed 8%, 5%, 28%, and 3% of values among the overall political belief items in the 1985, 1987, 2010 GSS and 2000 ANES respectively.¹¹ Using the components from PCA and the datasets from MI, we calculate the dimension scores that comprise year-specific belief spaces.

2–3. Sync Score

In this study, we propose a measurement called sync score for assessing the relative position in the belief space. We calculate the Euclidean distance between each pair of individuals and take the opposite-signed values of the average distance to all other individuals in belief space. Considering that the scales of year-specific belief spaces are not comparable due to different numbers of dimensions, we use the standardized version of the opposite-signed average distance.¹² A one-unit increase in sync score is comparable to a one-standard-deviation decrease in the average distance, which is the same as a one-standard-deviation increase in the level of belief synchronization with other people in terms of political beliefs. We expect that individuals with higher sync scores would have higher probabilities of meeting people with similar political attitudes, which would lead to larger social networks through the homophily mechanism.

Our measurement of synchronization considers the distance to every other person in belief space, which is unrealistic from the perspective that people may search for their partners for social interaction within a limited social space where their everyday lives take place. This is especially problematic when differentiating between the roles of belief synchronization and belief moderation since those with polarized beliefs would automatically have lower scores due to their weak sync with the opposite end. For this reason, we propose several different versions of sync scores that place less weight on people who are so remote in belief space that may not influence the probability of being socially connected. First, we measure the percentage of people who are situated in the proximity of each individual. We assume that a pair of people are proximate if the distance between the two is shorter than the 20^th percentile among all pairs in belief space.¹³ When using this measurement, two people with the same number of potential network members in proximity would have the same level of synchronization regardless of their distance to other individuals in belief space. We mainly focus on findings from the original method based on Euclidean distance, while we report selective results from the method based on the percentage and provide full tables in the appendix.

Second, considering that life trajectories are strongly constrained by structural factors (McPherson, 1983; McPherson & Ranger-Moore, 1991), we calculate sync scores within the same socioeconomic groups. For example, sync scores for those with a college degree are from the average distance to those with a college degree rather than to every other people in belief space. Similarly, we calculated different sync scores based on education, age (five groups: ≤20s; 30s; 40s; 50s; ≥60s), gender, race/ethnicity, region, religion, and political ideology. Interestingly, all those measurements are highly correlated (i.e., >0.8) with the original sync score, which implies that our sync score is not constrained by other socio-demographic dimensions. This is mainly due to our strategy for assessing sync scores: since we incorporate more than 40 belief items each year and equalize the influence of issue domains by standardizing principal component scores, the role of contentious belief items that might vary widely across different socio-demographic dimensions is weakened. For this reason, our sync scores are not good at picking up individuals with a few extreme opinions but effective in identifying those with a combination of out-of-sync political beliefs across many issue domains. Due to a strong correlation among different versions of sync scores, the estimates from our main analyses do not meaningfully differ by type of measurement, and we focus on the results from our original sync score in the rest of the paper.

2–4. Social Network Size

Our measurement of social network size is based on the count of important-matter discussion network members using the egocentric network module adopted in the 1985, 1987, and 2010 GSS. The module starts with the following question: “From time to time, most people discuss important matters with other people. Looking back over the last six months - who are the people with whom you discussed matters important to you?” Scholars argue that this module is effective in revealing close and frequently-contacted social networks that are expected to provide intimacy, normative influence, and information (Burt, 1984; Marsden, 1987; McPherson et al., 2006). The number of network members ranges from zero to six, which is our measurement for the important-matter network size.

For checking the mechanism in detail, we measure the size of political-matter discussion networks using the data from the 1987 GSS and 2000 ANES. The egocentric network module included in the 2000 ANES originally focuses on political matters as follows: “From time to time, people discuss government, elections, and politics with other people. I’d like to ask you about the people with whom you discuss these matters. These people might or might not be relatives. Can you think of anyone?” The module elicits up to four people, and we adopt the number of those members as the political-matter discussion network size. Even though the GSS surveys do not incorporate an independent survey module for political-matter discussion networks, the 1987 GSS network module probes the frequency of discussion about political matters with a six-level scale (1=almost daily; 2=at least weekly; 3=at least monthly; 4=at least yearly; 5=less than yearly; 6=never) up to the first three important-matter discussion network members. We count the number of network members with a monthly or more frequent discussion about political issues, which is used as the political-matter discussion network size in the 1987 GSS.¹⁴

2–5. Control Variables

For examining the relationship between sync score and social network size, we need an extensive set of controls regarding structural factors that may confound the process of value homophily. For this aim, we control for basic demographic factors which have been incorporated in previous studies of structural homophily (McPherson et al., 2001; Smith, McPherson, & Smith-Lovin, 2014): age (linear and quadratic effects), gender, race/ethnicity, geographic region, marital status, education, income, and religion. We use different racial/ethnic categories for the GSS and ANES: as for the GSS, we include two dummy variables for black and other racial groups with reference to white, whereas we add three dummies for non-Hispanic black, Hispanic, and other racial groups in models for the ANES. Geographic region is coded with nine categories – New England, Middle Atlantic, East North Central, West North Central, South Atlantic, East South Central, West South Central, Mountain, and Pacific. Marital status is considered as three categories: married/partnered, widowed/divorced/separated, and never married. We differentiate four levels of educational attainments (<High school; high school; some college; ≥college) and four income groups based on total family income in the last year before taxes (<$10,000; ≥$10,000; <$25,000; ≥$25,000; refusal or missing). Religion is coded into four categories of Protestant, Catholic, other religions, and no religion.

In this study, we are interested in the variation in social connection according to the level of similarity in political opinions to contemporaries. For focusing on the role of similarity per se, we condition on political party affiliation and political ideology, which may be deeply involved in the level of social connection through the organization of political attitudes (Boutyline & Vaisey, 2017; Converse, 1964). Political party affiliation is re-categorized into four groups of the Democratic, independent, Republican and other parties/missing, while the political ideology is assessed by four categories of liberal, moderate, conservative, and missing.

Lastly, for checking if the relationship between sync score and social network size results from the tendency for taking a moderate or neutral stance over controversial topics, we count the number of moderate opinions over belief items. We consider the items with a scale of at least three choices and assume to have a moderate opinion if a person reports the middle value among an odd number of choices or the middle two values among an even number of choices.¹⁵ In order to eliminate the difference in belief items across the surveys, we standardize the number of moderate opinions within each survey year.

2–6. Analytic Strategy

We provide the results from three steps of analyses. First, we perform Ordinary Least Squares (OLS) regression analyses of the sync score on demographics to get a sense of our proposed measurement. Second, we perform Poisson regression analyses for assessing the relationship between important-matter discussion network size and sync score. We consider three sequential models: i) regression of social network size on only sync score, ii) regression with other control variables, iii) regression with controls and the interaction of education and sync score. Third, we add the level of moderate opinions over political issues in the models and check if the association between sync score and social network size still holds. All the results from the regression analyses are adjusted for the survey design and the variability among 100 imputed datasets (Rubin, 1987). While we perform separate analyses for each year of survey, we additionally do the analyses using the pooled datasets in each step of analyses above to check the overall trends.¹⁶ When doing the analyses using the pooled datasets, we additionally control for year fixed effects for purging out the differences in year-specific belief spaces and the effect of year-specific unobserved factors.

3. Results

3–1. Description of Sync Score and Social Network Size

Table 1 contains descriptive statistics of sync score and social network size. Since the sync scores vary by imputed dataset due to the changes in the dimension scores, we report the statistics from five randomly selected datasets among 100 imputed data. The results show that the distribution of standardized sync scores is skewed left with the 1^st percentile between −3.08 and −2.57 and the 99^th percentile between 1.67 and 1.96, which is consistent across imputed datasets from each survey year (also see Figure 2). As for sync scores based on the percentage of proximate individuals, the average sync score is about 20% across the surveys – this is mechanically determined by our threshold of the 20^th percentile – while the 1^st and 99^th percentiles fall within a range from 0.2–1.6% to 49.9–57.6%. The correlation between two versions of sync scores is higher than 0.9 across all survey years, while the correlation with the level of moderate opinions is slightly stronger for sync scores based on the average distance than those from the percentage measure (0.23 vs. 0.20 in GSS 1985; 0.19 vs. 0.16 in GSS 1987; 0.35 vs. 0.33 in GSS 2010; 0.33 vs. 0.32 in ANES 2000; the results of bivariate analyses are not shown in the table). The results imply that the distance measure is more influenced by polarized beliefs than the percentage measure, whereas the difference is nearly negligible given the strong correlation between two measures.

Table 1.

Descriptive Statistics of Sync Score and Social Network Size

Note. Sync scores are from five randomly selected datasets among 100 imputed datasets in each survey year. SD Standard deviation; 1st 1^st percentile; 99th 99^th percentile.

Figure 2. Distribution of Sync Score.

Note. Kernel density is estimated using the pooled 100 imputed datasets and the Epanechnikov kernel with a bandwidth of 0.1. Outliers with sync scores lower than the 1^st percentile or higher than the 99^th percentile in each imputed dataset are excluded.

As for important-matter discussion networks, the size of networks has slightly decreased over time from 3.0 in 1985 to 2.5 in 1987 and 2010. The mean of political-matter discussion network size is 1.2 for the 1987 GSS with a maximum of 3, and 1.9 for the 2000 ANES with a maximum of 4.

3–2. Sync Score and Socio-demographic Factors

Table 2 contains the results from OLS regression analyses of sync score on demographic factors. Models from (1) to (4) show the estimates from each survey, while Model (5) is based on the pooled data over four surveys. As for age and gender, we do not find any consistent relationship with sync score across the surveys, whereas the analysis of the pooled data shows a significant increase in sync score for female (β=0.089, p=0.013). However, the association is substantially small considering that the standard deviation of sync scores is one in all survey years. People with household income of $25k or higher has a slightly higher sync score than those with income lower than $10k (β=0.124, p=0.065). However, the association was not consistent across the surveys. We find consistent patterns of relationship from race/ethnicity and education across the surveys: blacks and lower-educated people are more likely to have lower sync scores than white and higher-educated counterparts. The pooled-data analysis shows that there is a 0.22 decrease for blacks (p<0.001) and a 0.36 (p<0.001) decrease for those without high school diploma compared with whites and college-educated people. We observe no significant and consistent relationship between sync score and marital status and religion (not shown in the table).

Table 2.

OLS Regression of Sync Score (average distance) on Individual Characteristics

Sample	(1)	(2)	(3)	(4)	(5)
	GSS 1985	GSS 1987	GSS 2010	ANES 2000	Pooleda
Age (10 years)	0.065 (0.097)	0.048 (0.135)	0.017 (0.154)	−0.106 (0.113)	−0.028 (0.066)
Age2 (10 years)	−0.005 (0.010)	−0.012 (0.013)	−0.004 (0.014)	0.007 (0.011)	0.000 (0.006)
Female	0.025 (0.058)	0.146* (0.056)	−0.006 (0.091)	0.171* (0.065)	0.089* (0.035)
Household income (ref.: <$10k)
≥$10k and <$25k	−0.031 (0.088)	0.066 (0.092)	0.171 (0.211)	−0.018 (0.113)	0.040 (0.061)
≥$25k	−0.029 (0.098)	0.160 (0.112)	0.357+ (0.190)	0.018 (0.134)	0.124+ (0.067)
Missing	0.086 (0.125)	−0.018 (0.134)	0.090 (0.226)	−0.069 (0.139)	0.018 (0.076)
Race/ethnicity (ref.: White)
Blacks	−0.403** (0.118)	−0.090 (0.065)	−0.313* (0.139)	−0.295** (0.105)	−0.219*** (0.052)
Hispanic				−0.265* (0.129)
Others	−0.015 (0.191)	−0.413* (0.201)	−0.248 (0.158)	−0.350* (0.156)	−0.235** (0.082)
Education (ref.: ≥College)
<High school	−0.284** (0.094)	−0.400*** (0.094)	−0.596*** (0.150)	−0.137 (0.134)	−0.359*** (0.061)
High school	−0.141+ (0.077)	−0.326*** (0.079)	−0.316** (0.113)	−0.174* (0.071)	−0.261*** (0.045)
Some college	−0.052 (0.094)	−0.175* (0.081)	−0.183 (0.125)	0.062 (0.069)	−0.106* (0.048)
Political party affiliation (ref.: Democrat)
Independent	0.018 (0.072)	−0.141+ (0.075)	0.013 (0.107)	−0.099 (0.096)	−0.051 (0.044)
Republican	0.047 (0.069)	−0.068 (0.073)	−0.132 (0.121)	−0.184* (0.078)	−0.087* (0.044)
Other party/missing	−0.056 (0.249)	−0.556* (0.232)	−0.325 (0.241)	−0.074 (0.114)	−0.168+ (0.096)
Political ideology (ref.: Liberal)
Moderate	−0.021 (0.067)	0.044 (0.070)	0.021 (0.116)	0.227** (0.074)	0.069 (0.042)
Conservative	−0.208** (0.071)	−0.157* (0.076)	−0.216 (0.131)	0.022 (0.093)	−0.133** (0.048)
Missing	−0.356* (0.170)	−0.138 (0.140)	−0.341 (0.436)	−0.114 (0.143)	−0.193* (0.093)
N	1519	1787	1243	1518	6067

Note. Standard errors in parentheses. All models are adjusted for the survey design and the variability among 100 imputed datasets. Coefficients for intercepts and control variables including geographic region, marital status and religion are not shown in the table.

^aHispanics in the 2000 ANES are included in the category for other racial/ethnic groups. The model controls for year fixed effects.

⁺p<.01;

^*p<.05;

^**p<.01;

^***p<.001.

We additionally consider the association with political party affiliation and political ideology. We observe lower sync scores for those affiliated with the Republican Party (β=−0.087, p=0.049) and conservative political ideology (β=−0.133, p=0.006) than Democrats and liberals. However, these associations are not large enough to make substantially meaningful changes in sync scores.

The results from sync scores based on the percentage of proximate individuals follow similar patterns: we find a 2.6% decrease in people in proximity for blacks (p<0.001), a 4.9% decrease for those without high school diploma (p<0.001), a 1.1% decrease for Republicans (p=0.036), and a 2.1% decrease for conservatives (p=0.001) (see Table A6 in the appendix).

In sum, we find evidence that those with lower socioeconomic status – especially low levels of education – are more likely to have lower sync scores. We also observe that partisanship and political ideology show statistically meaningful relationships with sync scores. However, the size of the associations is not practically large in most cases. This is partly due to the characteristic of our sync score, as discussed in the method section, that the inclusion of many belief items and standardization of dimensions scores result in a similar distribution of sync score across socio-demographic dimensions. However, it is still notable that the level of education has a strong and positive association with sync score compared with other individual factors. Considering that our sync score becomes lower when people have a more unusual combination of political beliefs, the results repeat previous findings that higher education leads to more organized and aligned political beliefs (Boutyline & Vaisey, 2017; Converse, 1964).

3–3. Important-matter Discussion Network Size and Sync Score

Table 3 shows the incident rate ratios (IRR) and standard errors from Poisson regression of important-matter discussion network size on sync score using the GSS datasets. We additionally report the marginal effects of sync score when fixing other controls at the mean for the ease of interpretation (marginal effects are not shown in the table). We start from models for seeking the association between network size and sync score without control variables. As seen in Model (1), (4), and (7), the association is positive and significant at the level of 0.05 across all survey years. The analysis of pooled data in Model (10) shows that a one-unit increase in sync score leads to 1.06 times larger social networks (p<0.001), while the marginal effect at the mean is 0.16.

Table 3.

Poisson Regression of Important-Matter Discussion Network Size on Sync Score (average distance)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
Sample	GSS 1985			GSS 1987			GSS 2010			Pooleda
Control	No	Yes	Yes	No	Yes	Yes	No	Yes	Yes	No	Yes	Yes
Sync × Education	No	No	Yes	No	No	Yes	No	No	Yes	No	No	Yes
Sync	1.043* (0.017)	1.009 (0.014)	1.046 (0.029)	1.055*** (0.016)	1.014 (0.014)	0.970 (0.025)	1.103** (0.035)	1.064+ (0.033)	1.040 (0.053)	1.062*** (0.013)	1.023* (0.011)	1.018 (0.023)
Education (ref.: ≥College)
<High school		0.696*** (0.033)	0.701*** (0.033)		0.741*** (0.035)	0.741*** (0.035)		0.767** (0.075)	0.786* (0.076)		0.728*** (0.027)	0.732*** (0.027)
High school		0.795*** (0.020)	0.801*** (0.021)		0.864*** (0.031)	0.849*** (0.032)		0.898+ (0.054)	0.891+ (0.056)		0.849*** (0.021)	0.848*** (0.021)
Some college		0.918* (0.031)	0.927* (0.034)		0.896** (0.031)	0.881** (0.033)		1.042 (0.066)	1.033 (0.066)		0.940* (0.025)	0.939* (0.026)
Sync × <High school			0.969 (0.049)			1.118* (0.047)			1.150 (0.131)			1.047 (0.039)
Sync × High school			0.953 (0.034)			1.019 (0.038)			1.009 (0.083)			0.986 (0.030)
Sync × Some college			0.949 (0.042)			1.064 (0.044)			1.024 (0.079)			1.006 (0.034)
N	1519			1787			1243			4549

Note. Standard errors in parentheses. Incident rate ratios are reported. All models are adjusted for the survey design and the variability among 100 imputed datasets. Coefficients for intercepts and control variables including age, age squared, gender, race/ethnicity, geographic region, marital status, household income, religion, political party identification, and political ideology are not shown in the table.

^aAll models control for year fixed effects.

⁺p<.01;

^*p<.05;

^**p<.01;

^***p<.001.

When including control variables, the association becomes weaker and insignificant at the level of 0.05 in all year-specific analyses but remains statistically significant in the pooled-data analysis (IRR=1.023, p=0.040) with a marginal effect of 0.06. As for education, we observe a clear advantage for highly educated people in social network size: looking through the marginal effects, those with no high school degree, high school degree, and some college have 0.83, 0.46, and 0.18 smaller social networks than college-educated people. Compared with education, the size of the association for sync score is not large.

According to Model (3), (6), (9), and (12), the interaction between sync score and education shows inconsistent patterns across survey years. While the results from the 1985 GSS in Model (3) shows that lower education leads to slightly lower returns from sync score, Model (6) for the 1987 GSS shows that those without high school diploma enjoy 1.12 times larger returns from a one-unit increase in sync score than the college-educated (p=0.010). Model (9) for the GSS 2010 shows similar patterns with the 1987 GSS, whereas the interaction terms are not statistically significant at the level of 0.05. Due to inconsistency among surveys, the pooled-data analysis provides little evidence for the interaction between sync score and education.

The results from sync scores based on the percentage of proximate individuals are reported in Table A7. The patterns are similar with the findings using sync score based on the average distance: according to Model (10) based on the pooled data, a 10% increase in proximate people in belief space leads to 1.05 times larger networks (p<0.001), which decreases to 1.02 (p=0.044) when including control variables in Model (11). We observe no meaningful interaction between sync score and education.

In sum, the results show that a higher sync score is associated with a larger size of important-matter social networks, while we find weak evidence for the interaction between sync score and education.

3–4. Political-Matter Discussion Network Size and Sync Score

In Table 4, we report the results from the Poisson regression of political-matter discussion networks using the 1987 GSS and 2000 ANES. As seen in the table, the association between network size and sync score without control variables is positive in both Model (1) from the GSS and (4) from the ANES, which results in a 0.07 increase in network size from a one-unit increase in sync score in the pooled data (IRR=1.051, p=0.009). When including control variables in Model (2) and (5), however, the size of association largely decreases, which leads to only a 0.01 increase in network size in the pooled data (IRR=1.005, p=0.780). Interestingly, we find a meaning interaction between sync score and education in both surveys: Model (3) and (6) show that a one-unit increase in sync score leads to 1.17 times larger returns (p=0.033) in the GSS and 1.13 larger returns in the ANES (p=0.253) for those without high school diploma than the college-educated, which results in larger returns for those with no high school diploma (IRR=1.164, p=0.008) and high school diploma (IRR=1.080, p=0.038) than those with college degree in the pooled-data analysis. Similar results are found when considering sync score based on the percentage of proximate people in belief space: as seen in Table A8, the results from pooled datasets show no meaningful relationship between political-matter discussion network size and sync score in Model (8) (IRR=0.996, p=0.765), while we observe larger returns for those with no high school diploma in Model (9) (IRR=1.113, p=0.020).

Table 4.

Poisson Regression of Political-Matter Discussion Network Size on Sync Score (average distance)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Sample	GSS 1987			ANES 2000			Pooleda
Control	No	Yes	Yes	No	Yes	Yes	No	Yes	Yes
Sync × Education	No	No	Yes	No	No	Yes	No	No	Yes
Sync	1.084** (0.030)	1.017 (0.025)	0.964 (0.035)	1.017 (0.026)	0.982 (0.021)	0.927** (0.023)	1.051** (0.020)	1.005 (0.017)	0.946* (0.020)
Education (ref.: ≥College)
<High school		0.578*** (0.051)	0.579*** (0.050)		0.546*** (0.065)	0.547*** (0.064)		0.569*** (0.037)	0.570*** (0.036)
High school		0.768*** (0.056)	0.755*** (0.055)		0.785*** (0.047)	0.787*** (0.047)		0.765*** (0.034)	0.758*** (0.033)
Some college		0.910 (0.061)	0.895 (0.061)		0.976 (0.036)	0.975 (0.036)		0.935* (0.031)	0.927* (0.030)
Sync × <High school			1.171* (0.085)			1.130 (0.119)			1.164** (0.065)
Sync × High school			1.041 (0.054)			1.106+ (0.063)			1.080* (0.039)
Sync × Some college			1.051 (0.062)			1.059 (0.048)			1.056 (0.036)
N	1787			1518			3305

Note. Standard errors in parentheses. Incident rate ratios are reported. All models are adjusted for the survey design and the variability among 100 imputed datasets. Control variables including age, age squared, gender, race/ethnicity, geographic region, marital status, household income, political party identification, and political ideology are not shown in the table.

^aHispanic category in the 2000 ANES is combined with the category for other racial/ethnic groups. Network size in the 2000 ANES is top-coded at 3. All models control for year fixed effects.

⁺p<.01;

^*p<.05;

^**p<.01;

^***p<.001.

For ease of interpretation, we plot the interaction between sync score and education based on the results from the pooled-data analyses in Figure 1. As shown in the graph of important-matter discussion network size on the left, the positive slope of belief synchronization gets flatter when the level of education is higher, although the interaction is not statistically significant at the level of 0.05. Regarding political-matter discussion networks on the right, the patterns of networking are quite different among educational groups: we observe that an increase in sync score leads to more political discussants only for those with no high school diploma, while people with college degree maintain larger political discussion networks along with a decrease in sync score. These interaction effects result in a wider educational gap in discussion network size especially among those with lower levels of belief synchronization: when the sync score decreases from 2 to −2, the difference in network size between a college degree and no high school diploma gets larger from 0.35 (=1.55–1.20) to 1.17 (=1.97–0.80). Those results imply that the meaning of “out of sync” is different among educational groups, especially when considering discussion networks focusing on political matters. It is probable that higher levels of tolerance and openness to value heterogeneity among highly educated people may provide more chances to talk about unconventional political thinking, whereas the mechanism of homophily strongly constrains political discussion among those with low education.

Figure 1.

Predicted Size of Important and Political Matter Discussion Networks: Interaction between Sync Score and Educational Attainment

3–5. Re-examination with Moderate Opinions

Does the association between belief synchronization and social networks originate from the selection of moderate opinions over issue domains? We employ two strategies for answering this question. First, we adopt the percentage of proximate individuals in belief space which may reduce the disadvantage of having polarized opinions in scoring high on belief synchronization. As shown above, the results from the percentage measure are not substantively different from those based on the average distance in belief space, which imply that our findings are not fully explained by the tendency of being moderate in political issues.

Second, we directly include the level of moderate opinions in our analytic models. Table 5 shows the results from the pooled-data analyses of sync score and network size when adding moderate opinions. According to Model (1), a one-standard-deviation increase in moderate opinions leads to a 0.28 increase in sync score (p<0.001), which shows that the level of moderate opinions is positively associated with sync score even after controlling for structural factors. Additionally, we observe that the association between sync score and education gets weaker: the gap between the college-educated and those without a high school diploma is 0.29, which is smaller than 0.36 when not conditioning on moderate opinions in Model (5) in Table 2. This is mainly due to the positive association between moderate opinions and education: according to OLS regression of moderate opinions on the level of education, the college-educated have a 0.46 higher level of moderate opinions than those without a high school degree (p<0.001) (the pooled GSS and ANES data are used, no control variables are included except year fixed effects, the results are not shown in the table). However, as seen in Model (1), people with a high school diploma or lower levels of education have significantly lower sync scores than the college-educated even after conditioning on moderate opinions, while the results for important-matter and political matter network size in Model (2)-(5) show that the level of moderate opinions is not a significant predictor of network size and makes no meaningful difference in our main findings. Table A9 also shows that a one-standard-deviation increase in moderate opinions leads to a 3.19% increase in the percentage of people in the proximate distance, whereas the inclusion of moderate opinions does not make any meaningful change in the relationship between social networks and sync scores in Model (2)-(5).

Table 5.

Regression of Sync Score (average distance) and Network size with the Level of Moderate Opinions

	(1)	(2)	(3)	(4)	(5)
Outcome	Sync score	Important-matter NS		Political-matter NS
Model	OLS	Poissona		Poissona
Sample	All datasetsb	All GSS datasets		GSS 1987 + ANES 2000bc
Sync × Education	No	No	Yes	No	Yes
Sync		1.026* (0.012)	1.021 (0.023)	1.007 (0.018)	0.948* (0.021)
Education (ref.: ≥College)
<High school	−0.288*** (0.061)	0.726*** (0.027)	0.730*** (0.027)	0.568*** (0.037)	0.569*** (0.036)
High school	−0.201*** (0.045)	0.848*** (0.021)	0.846*** (0.021)	0.764*** (0.033)	0.757*** (0.032)
Some college	−0.054 (0.046)	0.939* (0.025)	0.937* (0.026)	0.934* (0.031)	0.926* (0.030)
Sync × <High school			1.046 (0.039)		1.164** (0.065)
Sync × High school			0.985 (0.030)		1.080* (0.039)
Sync × Some college			1.005 (0.034)		1.057 (0.036)
Moderate opinions	0.276*** (0.019)	0.988 (0.011)	0.989 (0.011)	0.990 (0.018)	0.990 (0.018)
N	6067	4549		3305

Note. Standard errors in parentheses. All models are adjusted for the survey design and the variability among 100 imputed datasets. Control variables including age, age squared, gender, race/ethnicity, geographic region, marital status, household income, political party identification, and political ideology are not shown in the table. All models control for year fixed effects.

^aIncident rate ratios are reported;

^bHispanic category in the 2000 ANES is combined with the category for other racial/ethnic groups;

^cNetwork size in the 2000 ANES is top-coded at 3.

⁺p<.01;

^*p<.05;

^**p<.01;

^***p<.001;

NS Network size.

According to the results, our measures of belief synchronization show a positive association with the tendency for having moderate opinions over political issues, whereas the tendency does not explain the relationship between belief synchronization and network size. These results imply that the difference in network size is mainly due to the level of similar opinions with potential network members rather than the level of moderate or polarized opinions.

4. Discussion

This study aims to test if the degree of synchronization with contemporaries in terms of political beliefs is associated with the size of social networks. Using political belief items from the GSS and ANES, we construct belief spaces where all individuals are placed based on their political opinions and assess the level of belief synchronization by the average distance to other people in belief space. We summarize the results in three points. First, the level of synchronization is higher for individuals from higher socioeconomic status, especially those with higher education. Second, higher levels of synchronization is associated with more network members with whom individuals can discuss important matters. Third, when considering the discussion networks focusing on political matters, we observe heterogeneous trends among educational groups: a positive association between belief synchronization and network size is clear among people with low education, whereas lower sync scores are associated with larger political discussion networks among the college-educated.

This study is based on the idea of social space proposed by Blau (1977) and Bourdieu (1984, 1989) to explore new dimensions in the determinants of social network size. Our measurement for synchronization indirectly measures the size of potential social network members within proximity in belief space. If individuals purposely or unwittingly are concerned with cultural values and habits including political beliefs when searching for discussion partners, higher levels of synchronization would lead to higher probabilities of meeting up with those of similar political beliefs in everyday lives. The results support our hypothesis on the link between the position in belief space and social network size, especially when considering discussion networks that are not confined to political matters. The correspondence to social networks with a broad definition suggests that our measurement for synchronization may not be confined to political matters but extends to a broad set of issues, given that political attitudes are closely associated with other cultural dimensions such as leisure activities or aesthetic tastes (DellaPosta et al., 2015).

The analyses also show that the association between the level of synchronization and network size is not strong, compared with other socio-demographic factors such as education. This is partly due to our strategy for measuring the level of synchronization: we incorporate a large number of belief items and balance the influence of issue domains by standardizing dimension scores, which may decrease variation from hot-button topics. However, considering that the political polarization in the American public is limited to a few contentious issues or a small number of political activists (DiMaggio et al., 1996; Evans, 2003; Fiorina & Abrams, 2008; Park, 2018), our measurement may better reflect the nature of political attitudes than a few selective topics of interest. The small influence of belief synchronization also echo the concerns by McPherson et al. (2001) that value homophily might be largely confounded by other sources of structural homophily, considering that the inclusion of control variables largely weakens the association between social network size and sync score. Despite the small role of belief synchronization, it is worth noting that the position in belief space still matters even after conditioning on a broad set of structural factors that may strongly constrain individual life trajectories.

The results from political-matter discussion networks show a weak association with belief synchronization on average, which is mainly due to a strong interaction with the level of education. The correspondence between belief synchronization and network size still holds among those with no high school diploma, whereas a combination of unusual political beliefs leads to a tendency toward cultivating larger social networks for those with high levels of education. One possible explanation for this finding is that education may be related to different behavioral mechanisms in establishing political-discussion social networks. Previous studies show that longer experience in education would expose individuals to people from diverse backgrounds, which leads to higher levels of tolerance to diverse political opinions (Mutz, 2002; Stouffer, 1955). If the openness to diversity could lead to the ability to establish and manage social connections with those with different opinions (Baldassarri & Bearman, 2007; Cowan & Baldassarri, 2018), weak belief synchronization may not be problematic for highly-educated people.

Nonetheless, it is still surprising that the association between sync scores and political-matter discussion network size is negative rather than null among the college-educated. Does this mean that a set of uncommon beliefs are more advantageous in maintaining more extensive networks of political discussion? It is premature to answer to this question since our outcome reflects only the respondent’s tendency toward identifying network members but not the chance of being identified by other potential discussants in belief space. If an uncommon combination of political opinions requires detailed explanation and justification, there is a reason to expect that those with low sync scores would talk more about politics and identify more network members discussing political matters. This process of convincing others of the value of one’s beliefs may be possible only for highly educated people with a good knowledge about politics and in the social environment that permits diverse ways of thinking, whereas the same set of opinions can signal maladaptation or deviance among those with low education whose level of openness to diversity is expected to be low. However, this does not guarantee that the college-educated people talk about political matters with those with unusual beliefs. Since our analyses are based on egocentric network modules for revealing respondent-perceived network members, our findings can only be understood from the perspective of egos.

The educational gradient is not notable in the analysis of social networks discussing important matters. We observe that the slope of sync scores goes closer to zero but not to negative values along with the increase in education level, which shows that more investment in political discussion does not necessarily lead to the growth of general discussion networks. These results imply that those two types of networks have a trade-off rather than an additive relationship among highly educated people: more communication about political matters may lead to a less frequent discussion about other issues, whereas those with less focus on political topics may cultivate larger social networks regarding a broader range of everyday issues. Despite the limited advantage of political discussion in cultivating general social networks, it is noteworthy that those with higher education have larger important-matter and political-matter discussion networks at every level of belief synchronization, and the educational gap becomes larger when the level of synchronization is lower. Scholars have consistently reported that higher levels of education lead to larger social networks, whereas they do not provide a satisfactory explanation about the mechanism in detail (Marsden, 1987; McPherson et al., 2006; Moore, 1990). Our analyses provide two possible mechanisms. First, higher education leads to higher levels of belief synchronization with other people in society, which would increase the probabilities of encountering potential network members with similar cultural traits. Second, highly educated people can enjoy a decent level of discussion even with low levels of belief synchronization, which results in larger social networks compared with those with low education who can maintain a moderate level of discussion only with strong sync. Both cases imply that social networks of highly educated people are less constrained by the position in belief space, whereas “out of sync” has a clear correspondence to “out of society” for those with low education.

The re-examination with belief moderation (or polarization) shows that, despite its significant correlation with belief synchronization, the level of belief moderation does not fully explain our original findings of the association between social networks and sync scores. These results imply that our measurement of belief synchronization captures unique aspects of political attitudes that are not explained by political polarization. We summarize our findings in two regards. First, from the perspective of individuals, the polarization of political attitudes itself does not necessarily lead to a decrease in belief synchronization. According to our findings, polarized opinions are problematic in making friends only when their combination is quite unusual, compared with that of contemporaries. Second, from a macro perspective, the polarization of political attitudes can decrease societal-level belief synchronization only when the polarization is prevalent but not aligned among various belief domains. Since this study adopts the standardized version of sync scores for purging out the variation arising from different survey designs, we could not test the time trends of belief synchronization in detail. Instead, we briefly check if the distribution of sync scores has changed over time in Figure 2. As seen in the figure, we do not find a notable change in the left tails of the distribution. On the other hand, the right tails appear shorter in the 2000s than in the 1980s, whereas the percentage of people with higher sync scores than average (=0) is larger in the 2000s (58% in 2000; 57% in 2010) than in the 1980s (54% in 1985; 52% in 1987). Overall, we find little evidence of belief de-synchronization, which may reflect that the political polarization in the US is weak and limited to a few issue domains (DiMaggio et al., 1996; Evans, 2003; Fiorina & Abrams, 2008; Park, 2018).

This study is not without limitations.

1. The foremost one is that we could not directly test the mechanisms of value homophily in detail since i) we have only a few survey items for constructing social spaces based on other cultural dimensions,¹⁷ and ii) the social network data are from only the respondents (egos) and do not include detailed information about network members (alters). If we had information about social networks and a rich set of political and cultural items from both egos and alters, we could directly investigate i) the activation of value homophily and heterophily, ii) the distinctive role of each political or cultural dimension, and iii) the difference in levels of homophily between network members identified by egos and people who designate egos as their network members.

2. The results may partly depend on the selection of belief items used to construct the belief space. Even though we integrate items from over-represented domains into a reduced number of dimensions and standardize the component scores for balancing the relative importance among domains when calculating dimension scores, this method still has some limitation since i) the standardized component scores can underestimate (or overestimate) the explanatory power of some dimensions which explain a large (or small) amount of variation in political beliefs in the empirical world, and ii) it is possible to miss some under-represented domains in the first stage of thresholding based on the size of the eigenvalue due to a small number of items. Researchers should test if our measurement for synchronization in the belief space works well in other datasets.

3. Even though we reduce potential confounding bias due to unobserved personality traits by controlling for the level of moderate opinions, we could still expect confounding from the correlation between belief synchronization and unobserved individual traits (e.g. personality traits). Since this is the first study proposing the measurement of belief synchronization, it is unclear how the measurement would be associated with personality traits and other unobserved individual characteristics. Future studies with a rich set of personality measurements may be able to address this issue.

4. We do not pursue causal inference in this study: as previous studies emphasize, people readjust their political attitudes during social interaction with their discussion partners, which in turn influences the selection of social network members in the future (Huckfeldt et al., 2004; Lazer et al., 2010; Levitan & Visser, 2009; Visser & Mirabile, 2004). While this study only considers the cross-sectional relationship between social networks and political beliefs, it will be an important task for future studies to examine the dynamic interplay of social networks and cultural contexts which results in the coevolution of both.

Despite those limitations, we expect that this study will provide new insights into the study of social networks and culture. Belief synchronization posits that the same sets of attitudes can have different meanings and consequences along with the change in the distribution of attitudes over time. Even though this idea of relative social position is not new, there has been little effort to elaborate on the mechanism by which the position is involved in the social lives of people. In this study, we propose the measurement of relative position in belief space, called sync score, which is easily calculable using the existent survey datasets with belief items and show that the level of belief synchronization corresponds to the level of social connection. One of significant findings in this study is that we identify a group of people who are not detected solely by structural factors or polarized beliefs but truly disadvantaged in the perspective of social connection: low-educated people with an uncommon combination of beliefs. We believe that further exploration of this group would clarify the cultural mechanism of the social gradient in “out of sync” and “out of society”.

Appendix

Table A1.

Political Belief Items

Belief item	GSS question	ANES question	GSS 1985	GSS 1987	GSS 2010	ANES 2000
Environment spending	National spending on environment	Federal spending on environmental protection	1	1	1	1
Health spending (g)	National spending on health		1	1	1	0
Health spending 1 (a)		Support for government insurance plan	0	0	0	1
Health spending 2 (a)		Federal spending on AIDS research	0	0	0	1
Education spending	National spending on education	Federal spending on public schools	1	1	1	1
Health & education spending (a)		The government should provide fewer services in health and education	0	0	0	1
Racial-problem spending	National spending on conditions of blacks	Federal spending on aid to blacks	1	1	1	1
Military spending	National spending on military	Federal spending on defense	1	1	1	1
Welfare spending	National spending on welfare	Federal spending on welfare	1	1	1	1
Crime spending	National spending on crime control	Federal spending on dealing with crime	1	1	1	1
Foreign-aid spending	National spending on foreign aid	Federal spending on foreign aid	1	1	1	1
Poor-aid spending	The government should improve living standards	Federal spending on aid to poor people	0	1	1	1
Social-security spending	National spending on social security	Federal spending on social security	1	1	1	1
Child-care spending	National spending on childcare	Federal spending on childcare	0	0	1	1
Tolerance 1–1 (g)	An anti-religionist should be allowed to speak in public		1	1	1	0
Tolerance 1–2 (g)	An anti-religionist should be allowed to teach in a college		1	1	1	0
Tolerance 1–3 (g)	A book written by an anti-religionist should not be removed from a public library		1	1	1	0
Tolerance 2–1 (g)	A racist should be allowed to speak in public		1	1	1	0
Tolerance 2–2 (g)	A racist should be allowed to teach in a college		1	1	1	0
Tolerance 2–3 (g)	A book written by a racist should not be removed from a public library		1	1	1	0
Tolerance 3–1 (g)	A communist should be allowed to speak in public		1	1	1	0
Tolerance 3–2 (g)	A communist should be allowed to teach in a college		1	1	1	0
Tolerance 3–3 (g)	A book written by a communist should not be removed from a public library		1	1	1	0
Tolerance 4–1 (g)	A militarist should be allowed to speak in public		1	1	1	0
Tolerance 4–2 (g)	A militarist should be allowed to teach in a college		1	1	1	0
Tolerance 4–3 (g)	A book written by a militarist should not be removed from a public library		1	1	1	0
Tolerance 5–1 (g)	A homosexual should be allowed to speak in public		1	1	1	0
Tolerance 5–2 (g)	A homosexual should be allowed to teach in a college		1	1	1	0
Tolerance 5–3 (g)	A book written by a homosexual should not be removed from a public library		1	1	1	0
Tolerance (a)		Tolerance to those with different moral standards	0	0	0	1
Death penalty	Favor death penalty for murder	Favor death penalty for murder	1	1	1	1
Gun	Favor gun permits	The government should make it more difficult for people to buy a gun	1	1	1	1
Gender 1 (g)	Women are not suited for politics		1	0	1	0
Gender 2 (g)	Mother working doesn’t hurt children		1	0	1	0
Gender 3 (g)	Preschool kids suffer if mother works		1	0	1	0
Gender 4 (g)	Better for man to work, woman tend home		1	0	1	0
Gender (a)		Men and women should have equal roles	0	0	0	1
Abortion 1 (g)	Abortion should be allowed when there is a strong chance of birth defect		1	1	1	0
Abortion 2 (g)	Abortion should be allowed when woman wants no more children		1	1	1	0
Abortion 3 (g)	Abortion should be allowed when mother’s health is at risk		1	1	1	0
Abortion 4 (g)	Abortion should be allowed when cannot afford more children		1	1	1	0
Abortion 5 (g)	Abortion should be allowed when pregnant due to rape		1	1	1	0
Abortion 6 (g)	Abortion should be allowed when woman does not wish to marry father		1	1	1	0
Abortion 7 (g)	Abortion should be allowed for any reason		1	1	1	0
Abortion (a)		Abortion should be legal	0	0	0	1
Homosexuality 1 (g)	Homosexuality is wrong		1	1	1	0
Homosexuality 2 (g)	Homosexuals should have right to marry		0	0	1	0
Homosexuality 1 (a)		Homosexuals in the armed forces	0	0	0	1
Homosexuality 2 (a)		Job discrimination against homosexuals	0	0	0	1
Affirmative action	Preferential hiring for blacks	Preferential hiring for blacks	0	0	1	1
Racial problems 1	The government should help blacks	The government should help blacks	0	1	1	1
Racial problems 2	Blacks overcome prejudice without favor	Blacks overcome prejudice without favor	0	0	1	1
Racial problems 3	Differences due to discrimination	Slavery & discrimination makes blacks stay in lower classes	1	0	1	1
Racial problems 4 (g)	Differences due to lack of education		1	0	1	0
Racial problems 5 (g)	Differences due to lack of will		1	0	1	0
Racial problems 4 (a)		Blacks have gotten less than they deserve	0	0	0	1
Racial problems 5 (a)		If blacks would try harder, they could be just as well off as whites	0	0	0	1
Racism 1	Blacks are lazy	Blacks are lazy	0	0	1	1
Racism 2	Blacks are unintelligent	Blacks are unintelligent	0	0	1	1
Racism 3 (g)	Blacks are poor		0	0	1	0
Racism 4 (g)	Neighborhood half black		0	0	1	0
Racism 3 (a)		Blacks are untrustworthy	0	0	0	1
Inequality 1 (g)	Differences in income too large		0	1	0	0
Inequality 2 (g)	Inequality exists to benefit rich		0	1	0	0
Inequality 3 (g)	Large income differences necessary for prosperity		0	1	0	0
Inequality 1 (a)		Our society should make everyone equal	0	0	0	1
Inequality 2 (a)		We have gone too far in pushing equal rights in this country	0	0	0	1
Inequality 3 (a)		It is one of big problems not to give everyone an equal chance	0	0	0	1
Inequality 4 (a)		This country would be better off if we worried less about equality	0	0	0	1
Inequality 5 (a)		Unequal chance is not a big problem	0	0	0	1
Inequality 6 (a)		Equality would lead to fewer problems	0	0	0	1
Limited government 1 (g)	The government should reduce income differences		0	1	1	0
Limited government 2 (g)	The government should solve national problems		0	1	1	0
Limited government 3 (g)	Government’s role in the economy: cutting government spending		1	0	0	0
Limited government 4 (g)	Government’s role in the economy: less regulation of business		1	0	0	0
Limited government 5 (g)	Government’s role in the economy: price controls		1	0	0	0
Limited government 6 (g)	Government’s role in the economy: wage controls		1	0	0	0
Limited government 1 (a)		The less government, the better	0	0	0	1
Limited government 2 (a)		Free market can handle economic problems	0	0	0	1
Limited government 3 (a)		Government has become bigger due to private problems	0	0	0	1
Tax (g)	The amount of federal income tax is too high		1	1	1	0
Tax (a)		The expected federal budget surplus should be used to cut taxes	0	0	0	1
Religiosity	R considers self a religious person	Do you consider religion to be an important part of your life?	0	0	1	1
Biblical literalism	The bible is the word of god	The bible is the word of god	1	1	1	1
Immigration	Should # of immigrants to US increase	Should # of immigrants to US increase	0	0	1	1
Parenting 1	Desirable child qualities: obey	Desirable child qualities: obedience vs. Self-reliance	0	1	1	1
Parenting 2 (g)	Desirable child qualities: think for self		0	1	1	0
Parenting 3 (g)	Desirable child qualities: work hard		0	1	1	0
Parenting 4 (g)	Desirable child qualities: help others		0	1	1	0
Parenting 2 (a)		Desirable child qualities: independence vs. Respect for elders	0	0	0	1
Parenting 3 (a)		Desirable child qualities: curiosity vs. Good manners	0	0	0	1
Parenting 4 (a)		Desirable child qualities: being considerate vs. Well behaved	0	0	0	1

Note. (g) is for items that are included only in the GSS. (a) is for items that are included only in the ANES.

Table A2.

Principal Components from the 1985 GSS

Belief item	[1]	[2]	[3]	[4]	[5]	[6]	[7]	[8]	[9]	[10]	[11]	[12]	Unexplained
Environment spending			0.203				0.223			−0.282			0.577
Health spending (g)			0.252	0.204			0.284						0.515
Education spending			0.240	0.208			0.282						0.524
Racial-problem spending			0.345										0.441
Military spending										0.474			0.495
Welfare spending			0.278										0.639
Crime spending				0.265								−0.310	0.597
Foreign-aid spending											0.517	−0.211	0.422
Social-security spending			0.218	0.240			0.230					0.304	0.506
Tolerance 1–1 (g)	0.213												0.432
Tolerance 1–2 (g)										0.249			0.375
Tolerance 1–3 (g)	0.227												0.433
Tolerance 2–1 (g)	0.216												0.374
Tolerance 2–2 (g)	0.223												0.417
Tolerance 2–3 (g)	0.215					0.220			0.220				0.310
Tolerance 3–1 (g)						0.272			0.377				0.265
Tolerance 3–2 (g)	−0.203									0.224			0.448
Tolerance 3–3 (g)	0.201								0.238				0.320
Tolerance 4–1 (g)	0.219										0.258		0.406
Tolerance 4–2 (g)	0.231								−0.271				0.331
Tolerance 4–3 (g)	0.210												0.392
Tolerance 5–1 (g)	0.243								−0.262				0.255
Tolerance 5–2 (g)	0.230								−0.232				0.272
Tolerance 5–3 (g)	0.240												0.306
Death penalty											−0.214		0.717
Gun										−0.266		−0.487	0.489
Gender 1 (g)					0.202					−0.243			0.651
Gender 2 (g)					0.342			0.207					0.373
Gender 3 (g)					0.383			0.261					0.313
Gender 4 (g)					0.383								0.334
Abortion 1 (g)		0.316					−0.254	0.316					0.259
Abortion 2 (g)		0.352											0.169
Abortion 3 (g)		0.245					−0.320	0.376					0.289
Abortion 4 (g)		0.361											0.212
Abortion 5 (g)		0.312					−0.247	0.283					0.312
Abortion 6 (g)		0.365											0.156
Abortion 7 (g)		0.336						0.217					0.177
Homosexuality 1 (g)											0.306		0.519
Racial problems 3			0.255	−0.308									0.437
Racial problems 4 (g)				−0.351					−0.221				0.445
Racial problems 5 (g)			0.224	−0.245									0.485
Limited government 3 (g)						0.545			−0.250				0.380
Limited government 4 (g)						0.405							0.496
Limited government 5 (g)			−0.216	−0.234	0.334				0.317				0.236
Limited government 6 (g)					0.358		0.276	0.212	0.327				0.256
Tax (g)						0.265					0.435	0.386	0.448
Biblical literalism													0.580
Eigenvalue	9.422	3.459	3.092	1.891	1.674	1.463	1.395	1.325	1.260	1.148	1.065	1.024

Note. Loadings whose absolute valaues are higher than 0.2 are presented.

Table A3.

Principal Components from the 1987 GSS

Belief item	[1]	[2]	[3]	[4]	[5]	[6]	[7]	[8]	[9]	[10]	[11]	[12]	Unexplained
Environment spending													0.677
Health spending (g)					0.212								0.580
Education spending				0.226	0.201		−0.234						0.551
Racial-problem spending		0.292				−0.230				−0.203			0.431
Military spending					0.373						0.246	−0.250	0.509
Welfare spending		0.223											0.669
Crime spending				0.239	0.281			0.265				0.280	0.536
Foreign-aid spending						−0.252					0.490	−0.248	0.439
Poor-aid spending		0.306											0.478
Social-security spending					0.240			0.239					0.584
Tolerance 1–1 (g)	0.219												0.424
Tolerance 1–2 (g)						0.252							0.353
Tolerance 1–3 (g)	0.217												0.460
Tolerance 2–1 (g)	0.220												0.457
Tolerance 2–2 (g)	0.226												0.449
Tolerance 2–3 (g)	0.220						−0.217						0.372
Tolerance 3–1 (g)				−0.262		0.297	−0.284						0.268
Tolerance 3–2 (g)													0.488
Tolerance 3–3 (g)	0.204						−0.258						0.311
Tolerance 4–1 (g)	0.221												0.449
Tolerance 4–2 (g)	0.228						0.262						0.340
Tolerance 4–3 (g)	0.202						0.275	0.227					0.337
Tolerance 5–1 (g)	0.240						0.279						0.265
Tolerance 5–2 (g)	0.230						0.219						0.372
Tolerance 5–3 (g)	0.234						0.214						0.343
Death penalty		0.216											0.686
Gun						−0.251						0.640	0.392
Abortion 1 (g)			0.276					0.221	0.202				0.271
Abortion 2 (g)			0.295										0.182
Abortion 3 (g)			0.237					0.341	0.292				0.270
Abortion 4 (g)			0.312										0.215
Abortion 5 (g)			0.266					0.239	0.222				0.284
Abortion 6 (g)			0.296										0.164
Abortion 7 (g)			0.296										0.182
Homosexuality 1 (g)													0.582
Racial problems 1		0.307											0.438
Inequality 1 (g)					−0.225	0.352							0.428
Inequality 2 (g)						0.282			−0.254		0.236		0.505
Inequality 3 (g)					−0.318								0.693
Limited government 1 (g)		0.256				0.243							0.515
Limited government 2 (g)		0.274											0.523
Tax (g)				0.244					−0.204		−0.417	−0.333	0.468
Biblical literalism													0.638
Parenting 1				−0.261	0.300		0.224			−0.222	−0.340		0.248
Parenting 2 (g)				0.323						−0.414	0.288		0.272
Parenting 3 (g)								0.216	−0.340	0.635			0.152
Parenting 4 (g)					−0.307		−0.264	0.334	0.514	0.203			0.183
Eigenvalue	8.767	3.809	3.265	1.646	1.537	1.455	1.338	1.280	1.237	1.150	1.058	1.025

Note. Loadings whose absolute valaues are higher than 0.2 are presented.

Table A4.

Principal Components from the 2010 GSS

Belief item	[1]	[2]	[3]	[4]	[5]	[6]	[7]	[8]	[9]	[10]	[11]	[12]	[13]	[14]	[15]	[16]	[17]	[18]	Unexplained
Environment spending				0.236											−0.294				0.451
Health spending (g)		0.215		0.220															0.444
Education spending				0.318															0.471
Racial-problem spending		0.228																	0.401
Military spending															0.237	0.231			0.476
Welfare spending														0.207				−0.359	0.493
Crime spending				0.292				0.204						0.331					0.420
Foreign-aid spending																	−0.345		0.517
Poor-aid spending		0.267												0.250					0.327
Social-security spending				0.279														−0.390	0.449
Child-care spending																		−0.210	0.522
Tolerance 1–1 (g)																			0.347
Tolerance 1–2 (g)														0.229	0.233				0.275
Tolerance 1–3 (g)																			0.278
Tolerance 2–1 (g)									0.258	0.232									0.251
Tolerance 2–2 (g)								0.263	0.206			−0.256							0.241
Tolerance 2–3 (g)							0.202												0.295
Tolerance 3–1 (g)							0.309												0.248
Tolerance 3–2 (g)	0.207																		0.236
Tolerance 3–3 (g)											0.255						0.203		0.245
Tolerance 4–1 (g)								0.266		−0.228									0.269
Tolerance 4–2 (g)					0.299			0.211											0.265
Tolerance 4–3 (g)					0.357			0.224							0.281				0.223
Tolerance 5–1 (g)														0.257					0.328
Tolerance 5–2 (g)					0.225			0.302											0.307
Tolerance 5–3 (g)							−0.256										0.223		0.284
Death penalty											0.200					0.218			0.556
Gun				0.208									−0.256						0.542
Gender 1 (g)										0.223						0.363			0.476
Gender 2 (g)				0.255		−0.214		0.292											0.338
Gender 3 (g)				0.305		−0.308		0.206											0.250
Gender 4 (g)				0.207		−0.246													0.397
Abortion 1 (g)			−0.286								0.276								0.296
Abortion 2 (g)			−0.261																0.161
Abortion 3 (g)			−0.250							−0.214	0.270						−0.254		0.263
Abortion 4 (g)			−0.281																0.150
Abortion 5 (g)			−0.289								0.251								0.282
Abortion 6 (g)	0.200		−0.285																0.118
Abortion 7 (g)	0.202		−0.266																0.140
Homosexuality 1 (g)	0.208																	−0.224	0.355
Homosexuality 2 (g)	0.203																	−0.252	0.387
Affirmative action		0.205							0.221							−0.244			0.476
Racial problems 1		0.256																	0.334
Racial problems 2				−0.203															0.390
Racial problems 3															0.298				0.501
Racial problems 4 (g)											0.240			0.239					0.392
Racial problems 5 (g)							−0.237												0.489
Racism 1						0.348			0.209										0.327
Racism 2						0.405											0.324		0.363
Racism 3 (g)					−0.282	0.256			0.210	0.200				0.247					0.394
Racism 4 (g)						0.239		0.266											0.513
Limited government 1 (g)		0.249										−0.225							0.396
Limited government 2 (g)		0.272																	0.356
Tax(g)											0.231			0.227	0.312	−0.501	0.304		0.256
Religiosity																		0.206	0.429
Biblical literalism					0.206														0.429
Immigration											−0.246								0.486
Parenting 1					0.322					−0.253									0.248
Parenting 2 (g)							−0.443			0.276		0.310						0.209	0.107
Parenting 3 (g)									0.303			−0.218	0.558						0.113
Parenting 4 (g)							0.381		−0.279			−0.432	−0.204						0.145
Eigenvalue	9.482	6.186	3.674	2.188	1.810	1.738	1.552	1.500	1.435	1.375	1.294	1.264	1.251	1.165	1.115	1.033	1.016	1.004

Note. Loadings whose absolute valaues are higher than 0.2 are presented.

Table A5.

Principal Components from the 2000 ANES

Belief item	[1]	[2]	[3]	[4]	[5]	[6]	[7]	[8]	[9]	[10]	Unexplained
Environment spending										0.235	0.643
Health spending 1 (a)							−0.339				0.536
Health spending 2 (a)						0.219					0.609
Education spending							0.245				0.525
Health & education spending (a)											0.516
Racial-problem spending	0.221										0.455
Military spending						0.203				0.432	0.489
Welfare spending					−0.311						0.526
Crime spending		0.226					0.271				0.594
Foreign-aid spending					−0.235					0.306	0.545
Poor-aid spending					−0.219						0.493
Social-security spending		0.224						−0.219			0.516
Child-care spending											0.503
Tolerance (a)						0.249	−0.238	0.285			0.544
Death penalty									0.208		0.649
Gun											0.589
Gender (a)									0.214		0.605
Abortion (a)			0.263								0.612
Homosexuality 1 (a)			0.203					0.281		−0.258	0.455
Homosexuality 2 (a)								0.321			0.520
Affirmative action											0.554
Racial problems 1	0.207										0.537
Racial problems 2											0.485
Racial problems 3											0.515
Racial problems 4 (a)			−0.221								0.508
Racial problems 5 (a)											0.454
Racism 1			0.299	0.395							0.312
Racism 2			0.318	0.442							0.241
Racism 3 (a)			0.298	0.446							0.287
Inequality 1 (a)					0.288					0.282	0.559
Inequality 2 (a)	0.201				0.261						0.531
Inequality 3 (a)					0.208	0.203					0.520
Inequality 4 (a)					0.282			−0.288			0.398
Inequality 5 (a)					0.311		0.241	−0.328			0.471
Inequality 6 (a)					0.286	0.312			−0.234		0.425
Limited government 1 (a)						0.315					0.388
Limited government 2 (a)						0.347					0.443
Limited government 3 (a)						0.354					0.416
Tax (a)						0.205				−0.377	0.619
Religiosity							0.331	0.297	−0.291		0.392
Biblical literalism		0.300	−0.219						−0.228		0.397
Immigration											0.562
Parenting 1		0.275							0.328		0.482
Parenting 2 (a)		0.202							0.447		0.504
Parenting 3 (a)		0.301							0.333		0.471
Parenting 4 (a)		0.235									0.650
Eigenvalue	7.553	3.503	2.692	1.818	1.414	1.345	1.302	1.176	1.089	1.067

Note. Loadings whose absolute valaues are higher than 0.2 are presented.

Table A6.

OLS Regression of Sync Score (% in proximity) on Individual Characteristics

	(1)	(2)	(3)	(4)	(5)
Sample	GSS 1985	GSS 1987	GSS 2010	ANES 2000	Pooleda
Age (10 years)	0.247 (1.200)	0.038 (1.316)	1.073 (2.234)	−0.989 (1.496)	−0.365 (0.786)
Age2 (10 years)	−0.016 (0.119)	−0.079 (0.121)	−0.150 (0.203)	0.037 (0.144)	−0.007 (0.075)
Female	0.362 (0.740)	1.712** (0.612)	−0.286 (1.399)	2.222* (0.869)	1.092* (0.461)
Household income (ref.: <$10k)
≥$10k and <$25k	−0.305 (1.063)	0.963 (0.958)	0.919 (2.824)	−0.987 (1.511)	0.271 (0.714)
≥$25k	−0.023 (1.195)	2.059 (1.269)	3.424 (2.590)	−0.032 (1.893)	1.436+ (0.825)
Missing	1.132 (1.543)	−0.121 (1.488)	−0.324 (3.100)	−0.610 (1.803)	0.198 (0.921)
Race/ethnicity (ref.: White)
Blacks	−4.373*** (1.239)	−0.737 (0.798)	−4.884* (2.009)	−3.799** (1.311)	−2.647*** (0.645)
Hispanic				−3.653* (1.541)
Others	0.164 (2.162)	−2.926+ (1.632)	−4.270* (2.102)	−3.769* (1.808)	−2.877** (0.927)
Education (ref.: ≥College)
<High school	−3.462** (1.133)	−5.425*** (1.108)	.8.944*** (2.118)	−1.464 (1.831)	−4.865*** (0.787)
High school	−1.830+ (0.938)	−3.977*** (0.938)	−5.314** (1.697)	−2.104* (0.872)	−3.565*** (0.591)
Some college	−0.774 (1.133)	−2.531* (1.092)	−2.501 (1.954)	0.636 (0.914)	−1.652* (0.670)
Political party affiliation (ref: Democrat)
Independent	−0.042 (0.924)	−1.345+ (0.758)	0.398 (1.625)	−0.114 (1.185)	−0.274 (0.542)
Republican	0.639 (0.877)	−1.055 (0.793)	−2.021 (1.741)	−2.106* (0.962)	−1.126* (0.532)
Other party/missing	−0.711 (2.593)	−6.962** (2.256)	−5.198+ (3.075)	−0.257 (1.476)	−2.178+ (1.202)
Political ideology (ref: Liberal)
Moderate	−0.382 (0.865)	0.454 (0.869)	0.138 (1.807)	2.497* (0.973)	0.720 (0.562)
Conservative	−2.568** (0.910)	−1.965* (0.847)	−3.610+ (1.967)	−0.565 (1.104)	−2.055*** (0.603)
Missing	−4.703* (1.923)	−2.981* (1.331)	−3.848 (6.047)	−0.857 (1.952)	−2.688* (1.109)
N	1519	1787	1243	1518	6067

Note. Standard errors in parentheses. All models are adjusted for the survey design and the variability among 100 imputed datasets. Coefficients for intercepts and control variables including geographic region, marital status and religion are not shown in the table.

^aHispanics in the 2000 ANES are included in the category for other racial/ethnic groups. The model controls for year fixed effects.

⁺p<.01;

^*p<.05;

^**p<.01;

^***p<.001.

Table A7.

Poisson Regression of Important-matter Discussion Network Size on Sync Score (% in proximity)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
Sample	GSS 1985			GSS 1987			GSS 2010			Pooleda
Control	No	Yes	Yes	No	Yes	Yes	No	Yes	Yes	No	Yes	Yes
Sync × Education	No	No	Yes	No	No	Yes	No	No	Yes	No	No	Yes
Sync (10%)	1.034* (0.014)	1.004 (0.011)	1.026 (0.023)	1.048*** (0.010)	1.010 (0.009)	0.985 (0.020)	1.064** (0.023)	1.039+ (0.021)	1.025 (0.033)	1.049*** (0.009)	1.017* (0.009)	1.013 (0.017)
Education (ref.: ≥College)
<High school		0.695*** (0.033)	0.734** (0.070)		0.741*** (0.035)	0.642*** (0.055)		0.766** (0.075)	0.632** (0.103)		0.728*** (0.027)	0.681*** (0.046)
High school		0.795*** (0.020)	0.847* (0.053)		0.863*** (0.032)	0.847* (0.066)		0.899+ (0.054)	0.889 (0.113)		0.849*** (0.021)	0.866* (0.048)
Some college		0.918* (0.032)	0.983 (0.091)		0.896** (0.031)	0.806* (0.067)		1.040 (0.066)	0.998 (0.131)		0.941* (0.025)	0.920 (0.059)
Sync × <High school			0.976 (0.042)			1.076* (0.032)			1.121 (0.087)			1.037 (0.030)
Sync × High school			0.971 (0.028)			1.003 (0.029)			1.002 (0.052)			0.989 (0.023)
Sync × Some college			0.970 (0.036)			1.046 (0.032)			1.016 (0.051)			1.010 (0.026)
N		1519			1787			1243			4549

Note. Standard errors in parentheses. Incident rate ratios are reported. All models are adjusted for the survey design and the variability among 100 imputed datasets. Coefficients for intercepts and control variables including age, age squared, gender, race/ethnicity, geographic region, marital status, household income, religion, political party identification, and political ideology are not shown in the table.

^aAll models control for year fixed effects.

⁺p<.01;

^*p<.05;

^**p<.01;

^***p<.001.

Table A8.

Poisson Regression of Political-matter Discussion Network Size on Sync Score (% in proximity)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Sample	GSS 1987			ANES 2000			Pooleda
Control	No	Yes	Yes	No	Yes	Yes	No	Yes	Yes
Sync × Education	No	No	Yes	No	No	Yes	No	No	Yes
Sync (10%)	1.065** (0.022)	1.004 (0.019)	0.970 (0.031)	1.002 (0.020)	0.980 (0.017)	0.940* (0.022)	1.031* (0.014)	0.996 (0.013)	0.958* (0.017)
Education (ref.: ≥College)
<High school		0.576*** (0.051)	0.460*** (0.073)		0.545*** (0.065)	0.457*** (0.095)		0.567*** (0.037)	0.459*** (0.051)
High school		0.766*** (0.055)	0.713* (0.099)		0.784*** (0.047)	0.713*** (0.069)		0.763*** (0.033)	0.697*** (0.050)
Some college		0.909 (0.061)	0.849 (0.112)		0.976 (0.036)	0.858+ (0.073)		0.934* (0.031)	0.842** (0.055)
Sync × <High school			1.122+ (0.069)			1.094 (0.088)			1.113* (0.051)
Sync × High school			1.029 (0.049)			1.048 (0.048)			1.041 (0.032)
Sync × Some college			1.027 (0.049)			1.065+ (0.039)			1.048+ (0.029)
N	1787			1518			3305

Note. Standard errors in parentheses. Incident rate ratios are reported. All models are adjusted for the survey design and the variability among 100 imputed datasets. Control variables including age, age squared, gender, race/ethnicity, geographic region, marital status, household income, political party identification, and political ideology are not shown in the table.

^aHispanic category in the 2000 ANES is combined with the category for other racial/ethnic groups. Network size in the 2000 ANES is top-coded at 3. All models control for year fixed effects.

⁺p<.01;

^*p<.05;

^**p<.01;

^***p<.001.

Table A9.

Regression of Sync Score (% in proximity) and Network size with the Level of Moderate Opinions

	(1)	(2)	(3)	(4)	(5)
Outcome	Sync score	Important-matter NS		Political-matter NS
Model	OLS	Poissona		Poissona
Sample	All datasetsb	All GSS datasets		GSS 1987 + ANES 2000bc
Sync × Education	No	No	Yes	No	Yes
Sync (10%)		1.019* (0.009)	1.015 (0.017)	0.997 (0.014)	0.959* (0.018)
Education (ref.: ≥College)
<High school	.4.049*** (0.799)	0.726*** (0.027)	0.681*** (0.046)	0.566*** (0.037)	0.459*** (0.051)
High school	−2.873*** (0.590)	0.848*** (0.021)	0.866* (0.048)	0.762*** (0.033)	0.696*** (0.050)
Some college	−1.053 (0.655)	0.939* (0.025)	0.919 (0.059)	0.934* (0.031)	0.842** (0.055)
Sync × <High school			1.036 (0.030)		1.113* (0.051)
Sync × High school			0.989 (0.023)		1.041 (0.032)
Sync × Some college			1.009 (0.026)		1.049+ (0.029)
Moderate opinions	3.189*** (0.228)	0.990 (0.011)	0.990 (0.011)	0.993 (0.018)	0.993 (0.018)
N	6067	4549		3305

Note. Standard errors in parentheses. All models are adjusted for the survey design and the variability among 100 imputed datasets. Control variables including age, age squared, gender, race/ethnicity, geographic region, marital status, household income, political party identification, and political ideology are not shown in the table. All models control for year fixed effects.

^aIncident rate ratios are reported;

^bHispanic category in the 2000 ANES is combined with the category for other racial/ethnic groups;

^cNetwork size in the 2000 ANES is top-coded at 3.

⁺p<.01;

^*p<.05;

^**p<.01;

^***p<.001;

NS Network size.