Do Political Beliefs Drive Environment Selection?

The Selection Problem

Researchers’ ability to establish unbiased relationships between social networks and political behavior is frustrated by the reality that individuals construct their own social worlds. Individuals choose with whom to discuss politics ¹ and this choice is characterized by homophily, or likes associating with likes (see McPherson et al., 2001 for a summary). ² Homophily can occur based on race and ethnicity, gender, age, religion, education levels, occupation, social class, social/structural position, behavior, attitudes, beliefs, abilities, and aspirations, among others. Most important for our purposes, homophily on political orientations exists when individuals choose to associate with individuals with similar political orientations or select social environments because of their political preferences (Knoke, 1990; Huckfeldt & Sprague, 1995, 2004; Bishop, 2008). This becomes a problem (deemed the “selection problem”) because it is difficult to differentiate between an individual influencing their friend’s candidate choice and the extent to which shared political preferences or other preferences correlated with politics, such as race and social class, lead to similar candidate choice among individuals.

Most research on social influence on politics remains agnostic about the reasons – political or otherwise – for which individuals form relationships in the first place. However, it might be reasonable to expect that political beliefs drive relationship formation in at least three ways. First, even though a majority of the time individuals select into environments for reasons other than politics (Walsh, 2004; Sinclair, 2012; Lazer et al., 2010; Minozzi, Song et al., 2020), those reasons are often correlated with political ones. In fact, research has established that there are political dimensions to personality traits (Gerber et al., 2011), religious beliefs (Putnam & Campbell, 2012; Margolis, 2018), education levels (Miller & Shanks, 1996), race, ethnicity, gender, and sexual orientation (Box-Steffensmeier et al., 2004; Egan, 2019), region of country (Gelman, 2008), and even facial features (Olivola & Todorov, 2010).

And, many of these factors are consequential for the types of contexts and friendships that individuals select into. Friendships, for example, are formed based on shared personalities, experiences, and identities (for a review, see Fehr, 2008). Education levels and social class can constrain the types of people in an individual’s social circle.

Second, in some cases, individuals’ beliefs about politics directly factor into context selection (what Minozzi, Song et al., 2020 call “purposive” selection). Shared political preferences are important predictors of discussion partners. Strong partisans, especially, are likely to select politically like-minded discussion partners (Bello & Rolfe, 2014). Individuals with a larger gap in feelings between in- and out-group members (strong, positive feelings toward their in-group and strong, negative feelings for the out-group) are more likely to discuss politics with those who are politically like-minded (Hutchens et al., 2019; see also Huckfeldt et al., 2004).

Shared political preferences may be especially important in the selection of romantic partners. A high degree of political congruence has been demonstrated in spousal relationships, and this congruence can increase over time (Jennings & Niemi, 1968; Stoker & Jennings, 2008). In their experimental analysis of an on-line dating community, Huber and Malhotra (2017) demonstrate that potential dating partners are rated more favorably and are more likely to be contacted when they share political preferences. The potential mate’s political affiliation rivals other consequential predictors of selection, such as education levels (see also, Klofstad et al., 2012). Political preferences also affect whom individuals rate as attractive (Nicholson et al., 2016).

Beyond political discussants and romantic partners, political preferences are important for the selection of a slew of other social environments. Individuals report not wanting to become friends with and being upset if their children married individuals from the opposing political party (Iyengar et al., 2012). Individuals also express preferences to live in neighborhoods with co-partisans and sort into neighborhoods with others of similar values, religious beliefs, and political preferences (Hui, 2013; Bishop, 2008; but see Makse et al., 2014 for some nuance).

Lastly, there is a growing amount of evidence to suggest that political beliefs, especially party identification, may be even more consequential drivers of decisions about which social environments to select into in today’s polarized political climate than in the past, due to increased partisan social sorting (Mason, 2018). Specifically, if more and more of individuals’ social identities overlap with their partisan identities, then the “mega” identity of partisanship should become an even more important determinant of the social environments individuals select into (Mason, 2015, 2018; Egan, 2019).

Taken together, this body of evidence suggests that, in certain cases, political beliefs may be important predictors of selection into social environments. Still, the causal direction in these studies is (mostly) unclear. Do individuals select discussants of the same party because of their shared political beliefs (homophily), or do they develop similar beliefs because they discuss politics together (influence)? And, how would we know?

Identification Strategies

In the section that follows, three strategies utilized to isolate causal effects in network studies are discussed. Though they by no means represent the universe of solutions, they are among the most common.

Theory: The Role of Political Beliefs in Environment Selection

If political beliefs are significant predictors of environment selection, then past evidence demonstrating social influence effects needs to be (re)interpreted with this in mind. If political beliefs are inconsequential predictors of environment selection, then – even though confounding between influence and homophily still exists in observational settings – we can rest more solidly on the implicit assumptions in our work. Either way, understanding the extent of political homophily in network formation is a fruitful exercise.

I argue that individuals are changed by their social environments precisely because they choose them, and that politics is consequential for these choices (Santoro, 2017). Individuals actively (though, perhaps unconsciously) create influence environments and are consequentially influenced within those settings. In some cases, political orientations indirectly or incidentally – vis a vis gender, social class, religion, and etc. – predict selection into social environments (Walsh, 2004; Minozzi, Song et al., 2020). In other cases, political beliefs directly inform selection. As reviewed previously, individuals’ political beliefs are consequential in the selection of romantic partners, political discussants, and neighborhoods, among others (Bishop, 2008; Iyengar et al., 2012; Huber & Malhotra, 2017). And, in today’s polarized political climate, political beliefs may play an even bigger role in environment selection than in the past (Mason, 2018).

However, selection is contextual (Bond & Sweitzer, 2018); I do not expect that political beliefs are consequential predictors of the selection of all socio-political environments, but especially for politically confirming ones. Which environments are considered confirming is conditional on an individual’s own party and ideological identification. For individuals who identify as liberals or with the Democratic Party, these environments might include atheist, environmental, racial, and LGBTQ organizations along with Democratic or liberal organizations. For conservatives and Republicans, these environments might include religious or morality groups along with Republican and conservative organizations (see Research Design, Data & Measures for further discussion).

Furthermore, political beliefs may play a stronger, more direct role when an individual chooses close friends or a romantic partner, for example, than when they choose which work colleague to have lunch with (Mutz & Martin, 2001; Mutz, 2006; Mutz & Mondak, 2006; Lazer et al., 2010). Similarly, political factors may be more important in deciding where to live than in choosing an occupation (Bishop, 2008).

It remains to be seen, however, why political beliefs – specifically party and ideological identification – are consequential for the selection of socio-political environments. Conceptualizing party identification squarely within social identity theory (Tajfel, 1981; Tajfel & Turner, 1979), as recent work as done (Green et al., 2004; Mason, 2015, 2018; Egan, 2019), provides a path forward. In this perspective, party identification is a strong, affective attachment to a political party, capable of driving action, emotion, and bias (Campbell et al., 1960; Huddy, 2013; Albertson & Gadarian, 2015). Similarly, ideological identification may capture more than beliefs about the role of government in society; it may also be identity based – capturing individuals affiliation with liberal and conservative groups (Mason, 2018). ⁴ As these identities increasingly overlap with other social identities, such as race, religion, and sexual orientation, individuals’ biological and evolutionary proclivities to associate with their in-group and differentiate themselves from the out-group are heightened (Roccas & Brewer, 2002; Mason, 2018).

Furthermore, phenomena of in-group love and out-group hate help explain why individuals associate with members of their own groups and reject those from different groups (Brewer, 1999; Greenwald & Pettigrew, 2014; Weisel & Bohm, 2015). As social groups become more distinctive from one another – as is the case with the Democratic and Republican Parties in the United States – group members face increasing consequences for violating norms of group behavior. From this account, avoiding contact with individuals with opposing political beliefs or exposure to differing social contexts not only protects identities but also demonstrates dedication to one’s own team.

All of this should be especially true for strong partisans and ideologues, which I define as those individuals who have the strongest attachments to their political party and ideological group (Campbell et al., 1960; Mason, 2018). These individuals are more likely to adhere to group behavior standards, take action on behalf of the group, become angry when the group is threatened, and exhibit bias in information processing (Bartels, 2002; Cohen, 2003; Huddy, 2013; Huddy et al., 2015; Ahler & Sood, 2018). They may also be more likely to be motivated by directional goals, to seek out congruent information sources including partisan news, and to discuss politics with individuals who have similar political views (Stroud, 2010; Bello & Rolfe, 2014; Hutchens et al., 2019). And, feelings of hostility toward out-partisans are more pronounced when out-party members are depicted as more ideologically extreme (Druckman et al., 2022; Homola et al., 2022). Strong partisans and ideologues may also be more knowledgeable about which social environments will be supportive of their political views (what I call “confirming” social environments) and which environments will expose them to alternate perspectives as well as be more likely to have politically aligned social identities (Mason, 2018).

Accordingly, I expect that the strongest political identifiers exhibit the greatest preferences for political homophily and create social environments that reinforce their political beliefs.

Hypothesis 1: Individuals with strong political beliefs are more likely to select into social environments that reinforce those beliefs than are those with weak political beliefs.

In other words, I expect strong Democrats and liberals (and Republicans and conservatives) to be more likely to select into environments that reinforce their political beliefs than weak or moderate identifiers. Beyond political beliefs, an individual’s interest in politics should also be an important determinant of environment selection. Specifically, I expect:

Hypothesis 2: Individuals with greater interest in politics are more likely to select into political social environments than are those with less interest.

Of course, once individuals select into social networks and contexts, they are influenced in those settings. Over time, movement into social environments that reinforce individuals’ political beliefs may lead them to adopt stronger political beliefs (Hutchens et al., 2019). However, this may more likely be a reinforcing effect of homophily rather than evidence of social influence, especially for those with strong political beliefs. On the other hand, movement into social environments that contradict individuals’ political beliefs may lead to weaker viewpoints over time, or even for them to leave that particular social environment. Thus, a feedback loop exists in which individuals actively select into social environments, are influenced within those environments, and then their updated political orientations drive selection into future socio-political contexts.

Hypothesis 3: The beliefs of strong partisans and ideologues are reinforced in confirming social environments.

Evidence for Hypothesis 3 would provide support for the argument that individuals choose to be changed (i.e., individuals purposely select into environments – based to some extent on political beliefs – in which they are influenced). However, it could also be the case that individuals with weak or less crystalized political beliefs select into social environments and become politically liberal or conservative as a result. After all, foundational work on the political effects of social networks demonstrate that influence most often occurs in the presence of political disagreement (when individuals are consistently exposed to competing considerations, or cross-pressures) (Lazarsfeld et al., 1948; Berelson et al., 1954; see also, Huckfeldt et al., 2004; Mutz, 2006; Klofstad et al., 2013). Thus, I expect individuals with less developed political views to be politically influenced in certain social environments, which leads to the fourth and final hypothesis.

Hypothesis 4: The beliefs of weak partisans and ideologues are strengthened in confirming social environments. ⁵

Research Design, Data & Measures

In order to understand the extent to which political beliefs predict environment selection, I need to know (a) the political beliefs of individuals before selection occurs, (b) the environments that individuals select into, and (c) the political beliefs after the selection takes place. Accordingly, I follow adult college students as they select into campus organizations throughout their first year in college at a large midwestern university – first contacting them before the start of their first-year.

College is an especially good setting for many reasons. First, it provides “a well-defined social system with clear boundaries” (Ognyanova, 2020, p. 547). This is especially important when trying to determine network effects, which are challenging to attribute to a network when no clear boundary of that network exists. Second, college represents one of the few times in life when individuals’ social networks experience a significant change (Newcomb, 1943; Hobbs, 2019; Samaraki et al., 2014). Furthermore, decisions made in the college environment represent some of the first selection decisions independent of parental influence and can persist throughout individuals’ adult lives (Newcomb et al., 1967). This is not to say that college students are necessarily representative of the wider population; the study does reflect a specific time and place. But, I believe that the depth of understanding about the role of political beliefs in environment selection is worth sacrificing the breadth of its’ application in this case. ⁶

All incoming first-year students at a large, midwestern research university were eligible to participate in this study. ⁷ From the universe of incoming first-year students in the Fall of 2016, 1000 were randomly selected and invited to participate in the research study via e-mail. Individuals were eligible to receive a small monetary incentive for their participation. Appendix A provides more detail about the undergraduate population as well as demographic information about the student sample.

From this sample, 407 individuals participated in the first survey wave, which ran from August to October 2016. Wave I, then, captures their first weeks on campus as well as the lead up to the November 2016 presidential election. The second survey wave (n = 302) began at the very beginning of participants second semester in college. It took place between January and March 2017 and followed up with the students who completed the Wave I survey as well as those who did not respond in order to check for potential bias in the Wave I sample. Wave II covers the first few months of Donald Trump’s tenure as president. The third survey wave (n = 238) followed up with individuals after their first full year in college and at the start of their second year; specifically, it launched in August and concluded in September of 2017. Thus, I have collected information throughout individuals’ entire first year of college, looking forward to their second year.

While 74% of Wave I respondents responded to Wave II (and 79% of Wave II respondents responded to Wave III), the attrition between survey waves is potentially problematic for inferences regarding the political determinants of environment selection. Specifically, it could be possible that political independents and moderates are more likely to drop out of a political survey than strong partisans, artificially increasing support for the hypotheses. In Appendix F, the causes and outcome of the attrition between survey waves is explored. Individuals who dropped out between Waves I and II (and between Waves II and III) were ideologically similar to those that remained in the panel, but individuals who remained were more likely to be U. S. citizens. This appendix also presents results from Heckman selection models that account for the non-response process.

Each survey wave contained multiple measures of participants’ social contexts and social networks. Social contexts refer to the physical locations or places where social influence occurs. Measures of the social context include student dormitories and student organizations. Social networks refer to the interpersonal relationships of peers, friends, and family in which social influence occurs. Measures of social networks include a general measure of the partisanship of their close friends as well as two political discussants. The surveys also included multiple measures of political beliefs, including the standard measures for party and ideological identification, as well as issue opinions and candidate feeling thermometers, among others.

Because, at the outset of the study, panelists are not yet embedded in the college setting, the design credibly accounts for the confound between network formation and network effects. Using this nascent network approach (Lazer et al., 2010; Ognyanova, 2020; Minozzi, Neblo et al., 2020), I assume that the baseline political attitudes observed in the first wave are not the consequence of influence from the socio-political environment of interest. In other words, I assume that Wave I attitudes are devoid of social influence effects from the college environment. With this assumption, I can go on to examine whether subsequent interactions between individuals and the social environment contribute to the change in their political beliefs. Certainly, this is a reasonable assumption given that all panelists are incoming first-year students and that the first survey wave took place from the week preceding their arrival on campus to their first few weeks on campus. Furthermore, individuals do not join campus groups right away but experience several weeks of student involvement fairs and recruitment events.

To be sure, Wave I political beliefs are not devoid of social influence effects entirely. We know, for example, that the political attitudes of young adults are influenced by the political beliefs of their parents, among others (Jennings & Niemi, 1968; Jennings et al., 2009). My claim is that Wave I political attitudes are devoid of social influence effects related to the college environment. Of course, it could still be possible that the environments that individuals join in college are a reflection of their previous social contexts and relationships, and that network ties precipitate their joining of these environments. ⁸

Results: The Role of Political Beliefs and Interest in the Selection of Partisan Environments

Do individuals’ political beliefs affect which types of environments they select into? An exploration into the reasons individuals select into social contexts is important primarily because if previously established political beliefs predict context selection, then shared similarities - and not social influence - may account for shared political beliefs among individuals in social settings.

I expect that individuals with strong beliefs about and politics are more likely to base decisions on which contexts to select into directly based on political reasons than individuals with not as strong beliefs. To understand the impact of political beliefs on environment selection, I first look at the role of political beliefs in predicting the intention to join explicitly partisan groups and then at predicting actually joining these groups. Results: The Role of Political Beliefs and Interest in the Selection of Non-Partisan Social Environments looks at the extent to which beliefs about politics drive the selection of non-partisan social environments.

At the very least, political beliefs should be consequential predictors of the selection of explicitly political environments. Respondents in Wave I were asked if they had any interest in joining political groups at the university and could select response options which included the College Democrats, College Republicans, Multi-Partisan Coalition, Young Americans for Liberty, Other, or no interest in joining a political group. ⁹ The College Democrats and College Republicans are explicitly partisan political groups, while the Multi-Partisan Coalition and the Young Americans for Liberty (YAL) are explicitly non-partisan political groups according to their mission statements.

From this question, three dichotomous variables were constructed. The join partisan groups measure takes the value of ‘1’ if the respondent indicated that they wanted to join one of the two explicitly partisan groups (the College Democrats and Republicans), and a ‘0’ otherwise. Seventy respondents indicated that they planned to join one of these groups in Wave I (40 College Democrats; 30 College Republicans). Because reasons for selection into politically partisan groups could vary between Democrats and Republicans, the decision to join the College Democrats and the College Republicans are also modeled separately (results are provided in Appendix B).

Hypothesis 1 predicts that political beliefs will be consequential predictors of these selection decisions. From the seven-point party identification measure, a partisan strength measure was created that takes a value of ‘3’ if an individual identities strongly with the Democratic or Republican Parties, a value of ‘2’ if an individual identifies not very strongly with the Democratic or Republican Parties, a value of ‘1’ if an individual leans toward the Democratic or Republican Parties, and a value of ‘0’ if they are true independents. Thus, higher numbers indicate stronger identification with a political party and lower numbers indicate weaker identification. A similar measure was created for individuals’ ideological identification. Ideological strength scored a value of ‘3’ if the participant identifies as Extremely conservative or liberal, ‘2’ if liberal or conservative, ‘1’ if slightly liberal or conservative, and ‘0’ if the participant chose moderate or “haven’t thought about this much”. ¹⁰

Results of the intention to join partisan groups, both Democratic and Republican are presented in Figure 1. The left-hand panel of this figure presents results from a logistic model with robust standard errors of the intention to join a partisan group in Wave I as a function of an individual’s strength of party identification, strength of ideological identification, interest in politics, ¹¹ and binary indicators for gender (female = 1) and race (white = 1). This model incorporates both political and non-political factors which may explain engagement in social environments. We know, for example, that gender and racial identities are correlated with political ones (Egan, 2019; Mason, 2018). In the supplementary materials, these models are expanded to test for alternate explanations.OPEN IN VIEWER

Taking the exponent of the coefficient for the strength of party identification (e^0.74) - calculating the odds ratio – results demonstrate that the odds of intending to join a partisan group are 2.10 times greater for individuals who identify strongly with a political party than for those individuals who identify not very strongly with a political party. The odds for ideological strength and interest are even bigger (2.23 and 2.66, respectively). The model (the full results of which are provided in Appendix B) explains 31% of the variation in the intention to join a partisan group. ¹²

Of course, indicating that you are interested in joining a partisan group and actually joining the group are two different things. The longitudinal nature of the study allows us to verify if and which groups were actually joined. Specifically, while 39 individuals who participated in Waves I, II, and III of the survey indicated that they planned to join an explicitly partisan group in Wave I, only seven reported actually joining these groups in Wave III (3 joined the College Democrats and 4 joined the College Republicans).

I run analyses on these actual joiners, although the low number of actual joiners (n = 7) makes inference difficult. Results are displayed in the right-hand panel of Figure 1. I model whether or not an individual actually joined a partisan group (the College Democrats and College Republicans) in Wave III as a function of Wave I measures of the partisan and ideological strength, their interest in politics, and the binary indicators for gender and race. While the coefficient for partisan strength is no longer significant (and in the opposite hypothesized direction), the coefficients for ideological strength and interest in politics retain their significance despite low statistical power to find effects. For individuals who identify as extremely liberal or conservative the odds of joining a partisan group are 2.66 times greater than for individuals who identify simply as liberal or conservative.

The variable for interest in politics is an even more important determinant of actually joining a political group. The coefficient for political interest is significant and in the expected direction in both models. Specifically, for a one unit increase in interest – going from “not at all interested” in the 2016 presidential election to “very little” interest in the election – the odds of joining a partisan group increase by a factor of 4.14 (right panel of Figure 1). And, this is intuitive given the population. For college-age students, interest in politics – over and above their partisan and ideological identifications – drives intention and, to an even larger extent, actual joining of political groups because their political beliefs are not yet crystallized. Much of this crystallization occurs in college; whereas, political beliefs might be more central to these decisions among older adults. Again, while the low number of actual joiners (n = 7) clouds our understanding of the role of political beliefs in selecting into partisan environments, there is initial – though limited – evidence to suggest that interest in politics drives selection into political groups. ¹³

These results are intuitive. Of course we expect individuals with more crystallized beliefs and greater interest in politics to be more likely to join partisan political groups. And yet, social network research stays quiet about whether or not political beliefs drive relationship formation. These analyses demonstrate that political interest and beliefs are a part of the decision to join partisan and political environments for individuals who are more strongly attached to a political party and ideological group and, especially, among those who have more interest in politics. In the “real world” these groups might represent local and state-wide party organizations.

Results: The Role of Political Beliefs and Interest in the Selection of Non-Partisan Social Environments

However, I am not solely interested in joining political groups, but in joining politically confirming social environments. For our purposes, politically confirming social environments are operationalized as the contexts and networks that, while not explicitly partisan (like, for example, the College Democrats and College Republicans) should be confirming of a particular political viewpoint, in light of social sorting. Jost et al. (2008), for example, look at a wide range of differences in the explicit and implicit preferences underlying political ideology. They find that liberals, on average, are more favorable to atheists, gay unions, environmentalists, and vegetarians. Whereas, conservatives, on average, are more favorable toward fraternities and sororities, Christians and religious people, and traditional institutions, such as marriage and family. Incorporating this work with Mason’s (2018) work on polarization and identities, I have chosen to categorize historically white sororities and fraternities, religious organizations, such as Campus Crusade, Pro-Israel organizations, as well as anti-abortion and gun rights organizations as right-leaning confirming social groups. LGBTQ, ethnic and racial organizations, such as the Black Student Association, and environment and atheist organizations were categorized as left-leaning confirming social environments. ¹⁴ I believe that these groups represent the social sorting described in previous work (Mason, 2015, 2018; Egan, 2019). ¹⁵

Wave II participants were asked to list the groups they joined the semester before (in Fall 2016). Importantly, this joining occurred between the first and second survey. From this list in Wave II and a similar list in Wave III, three binary variables were created. The first represents whether or not an individual joined an ideologically confirming social group (both liberal and conservative). The variable join confirming groups takes on a value of ‘1’ if the respondent indicates that they joined a confirming group in Fall 2016 and a ‘0’ otherwise. From this variable, two additional variables were created. Liberal confirming group takes a value of ‘1’ if the respondent joins a liberal confirming student organization and a ‘0’ otherwise. Conservative confirming group takes a value of ‘1’ if the respondent joins a conservative confirming student organization and a ‘0’ otherwise. In Wave II, 32 individuals reported joining confirming groups, which represents only a small minority of the 302 respondents in the Wave II survey. Sixteen individuals joined at least one liberal confirming group, and sixteen individuals joined at least one conservative group (though, a few individuals joined more than one) in Wave II. ¹⁶

Figure 2 models the likelihood of joining liberal (left panel) and conservative (right panel) groups in Wave II as a function of an individual’s party identification, ideological identification, interest in politics, and binary indicators for gender and race. Importantly, the standard, seven-point measures of party and ideological identification are used here because I model decisions to join liberal and conservative social groups separately. Party and ideological identification should work oppositely across panels in Figure 2, where I expect negative coefficients in the model for joining liberal groups (as ‘1’ indicates both Strong Democrat and Extremely Liberal) and expect positive coefficients in the model for joining conservative groups (as ‘7’ indicates Strong Republican and Extremely Conservative). ¹⁷ This is the case for the coefficient for ideological identification, but not for party identification (which is the opposite direction in the model of joining conservative social groups). The odds of joining a conservative social environment are over two times higher for individuals who identify as extremely conservative than for individuals who identify simply as conservative (a one unit increase in the independent variable); whereas, the odds of joining a liberal group decrease by a factor of 0.596 (e^−0.52) the more conservative an individual identifies as. ¹⁸ While interest in politics increases the odds of joining a liberal social group, it is not a significant predictor of joining a conservative group.OPEN IN VIEWER

I look more specifically at the effect of ideological identification on the probability of joining a liberal or conservative confirming social environment in Figure 3. Using the same models shown in Figure 2, I calculate the probability of joining liberal and conservative groups in Wave II given specific values of individuals’ ideological identification in Wave I. The left-hand panel of Figure 3 demonstrates that as individuals more strongly identify as liberal, their probability of joining a liberal confirming group increases. Individuals who identify as extremely liberal have an 12% probability of joining a liberal group while moderates have a 3% chance. This is especially true for the probability of joining a conservative group (right panel of Figure 3), where stronger identification as a conservative increases individuals’ probability of joining conservative groups. Specifically, individuals who identify as extremely conservative have a 39% chance of joining a conservative confirming social group, compared to a 6% chance for moderates.OPEN IN VIEWER

Results depicted in Figures 2 and 3 demonstrate that political beliefs, especially ideological identification, can be important predictors of joining environments that confirm individuals’ previously established beliefs about politics, confirming Hypothesis 1. In Appendix C, I demonstrate that the results for ideological identification persist when modeling the likelihood of joining liberal and conservative social groups in Spring 2017 (in participants’ second college semester; measured in Wave III) as well as with a combined measure of liberal and conservative groups joined in the Fall of 2016 (Wave II measure) and Spring 2017 (Wave III measure). I also look at the robustness of these results to including other measures that may affect the relationship between political beliefs and joining these social environments, such as religious affiliation and knowledge of politics in Appendix C. ¹⁹

For additional robustness checks, I modeled individuals decisions to join non-political social groups; specifically, the decision to join academic (n = 75) and sports (n = 61) organizations. If the hypotheses are correct, I should find that interest in politics and political beliefs are not important determinants of the selection of academic or sports organizations. And, as depicted in Appendix D, Table 17, partisan and ideological beliefs are not important determinants of these decisions. Interestingly, interest in politics increases the likelihood of joining an academic group.

Finally, I look at whether political beliefs are important predictors of the close friends and important matters discussants. Results, provided in Appendix D, demonstrate that political beliefs, in most cases – and not interest in politics – are consequential predictors of being close friends with individuals who share your political beliefs. Individuals who identify more strongly as liberal or Democrats (as conservatives or Republicans) are more likely to have close friends who are almost all Democrats (Republicans) and are more likely to discuss important matters with Democratic (Republican) discussants, confirming results of prior research (Bello & Rolfe, 2014; Huckfeldt et al., 2004; McClurg, 2006). That being said, I cannot be sure if these friendships were formed before arriving on campus or if they were formed afterward. If the friendships were formed before respondents arrived on campus, then the network nascent inferential approach does not hold. This is in contrast to the measures of joining a politically liberal or conservative leaning social group, where the joining occurred after arrival on campus (and after the first survey).

Political beliefs predict selection not only into explicitly political environments but also into non-political groups associated with individuals’ religious, racial, and sexual identities, among others. I find initial – though, again, limited – evidence that, for the most ideological individuals, political beliefs drive selection into certain contexts. While interest in politics is a consistently meaningful predictor of the intention to join and, especially, actual joining of political groups, it does not consistently influence individuals’ decisions to join politically confirming social groups, providing mixed support for Hypothesis 2.

Results: The Impact of Joining Politically Confirming Social Contexts on Individual Political Beliefs

I now turn to an analysis of whether or not the selection of politically confirming contexts affects individuals’ beliefs about politics. Specifically, does an individual’s presence in a politically confirming social environment predict their future political beliefs? To answer this question, I estimate whether participants’ self-selected social context is significantly associated with their political beliefs in Wave III controlling for their beliefs in Wave I (the “pre-exposure” measure) along with controls for race and gender. Specifically, I model participants partisan and ideological identification in Wave III using both linear regression and ordered logit models. As results are similar across models, the linear estimates are depicted for ease of interpretation in Figure 4. Results for both sets of models are available in Appendix E. ²⁰ OPEN IN VIEWER

In Figure 4, I model Wave III measures of party identification (left panel) and ideology (right panel) as a function of their Wave I beliefs, the Wave II measures of whether they joined a liberal and conservative social context and controls for gender and race (coefficients of the control variables are not shown in Figure 4). I focus on results from the model of ideology. Joining a liberal group is significantly associated with identifying as liberal and joining a conservative group is associated with identifying as conservative in Wave III even when controlling for their ideology and party identification in Wave I.

Joining a liberal or conservative group remains a significant predictor of the change in ideology between Waves I and III, where joining a liberal (conservative) group in Wave II is associated with more liberal (conservative) identification between the first and third survey waves. Results persists with controls for interest in and knowledge of politics and religious affiliation (see Appendix E). ²¹ Thus, there is initial support for a reinforcing effect whereby joining politically confirming social contexts leads to stronger affiliation as liberal and conservative. While individuals with stronger ideological identification are more likely to join these groups, it could also be the case that individuals with weaker, or less developed, political beliefs join these environments and become politicized, pointing to an influence effect.

To tease this apart, I analyze individuals with different political beliefs separately. First, I model the change in party and ideological identification between Waves I and III for independents and moderates only (Appendix E, Table 26). Individuals who identified as political independents in Wave I and joined conservative groups in Wave II, identified as more strongly Republican over time (no independents joined liberal groups). Individuals who identified as moderates in Wave I and joined liberal groups in Wave II, identified as more strongly liberal over time. The other coefficients in the models are in the expected direction but do not reach conventional levels of significance. Importantly, results of this effort offer only suggestive evidence of an influence effect among political moderates and independents due to low statistical power.

Looking at party leaners and slight ideological identifiers (Appendix E, Tables 27-28), Republican leaners and slight conservatives who join conservative groups identify more strongly with the Republican Party and as conservatives over time; however, this pattern does not occur among individuals who identify as slightly liberal or lean Democratic.

What about individuals with the strongest political beliefs? I have hypothesized that their political beliefs are reinforced in confirming social environments. Results from Table 29 (Appendix E) offer initial, though heavily qualified, evidence that the beliefs of the strongest Democratic and liberal individuals are reinforced in liberal social environments (though these are not significant effects in models of the change in party identification). Interestingly, strong and not very strong Democrats who join a conservative group (a non-confirming social group) become more strongly Democratic over time. The strongest Democratic identifiers may resist conflicting considerations, further entrenching their political beliefs.

Results of individuals who identify most strongly with the Republican Party and as conservatives (Table 30, Appendix E) depict an opposite effect, however. In Table 30 (Appendix E), not very strong and strong Republicans who join confirming (conservative) social environments (no strong identifiers joined liberal environments) identify as more strongly Democratic over time – opposite of expectations. This helps explain the negative coefficients for party identification in the models of joining conservative confirming social environments (see Results: The Role of Political Beliefs and Interest in the Selection of Partisan Environments and Results: The Role of Political Beliefs and Interest in the Selection of Non-Partisan Social Environments) and could be indicative of the liberal influence of the college environment (Newcomb et al., 1967). Again, I caution from making inferences from results that are dependent only on a handful of individuals in each category that join these groups.

Taken together, parsing out the behaviors of political independents and moderates, weak partisan and ideological identifiers, and the strongest identifiers, there is suggestive evidence that those in the middle are influenced by their social environments while the beliefs of the strongest identifiers are reinforced (though not in all cases and even in opposite directions).

While I expected both party and ideological identification to drive context selection and that selection to effect both sets of political beliefs, results demonstrate consistent effects of ideological identification over and above party identification. This could be because the measure of politically confirming social environments directly taps into ideology in its operationalization more so than it does party identification. Possibly, with a larger sample size, party identification would play more of a role in context selection. Finally, it is feasible that ideology – and not party identification – may drive context selection because it taps into distinct (though obviously correlated) aspects of individuals’ political experience. Specifically, while a college students’ affective attachment to a political party (party identification) may still be forming (Huddy, 2013), their views about how government should operate in society (ideology) may be more fixed. The importance of ideology, over and in conjunction with party identification, is echoed in recent working looking at effects of affective polarization (Druckman et al., 2022; Homola et al., 2022).

Discussion

If individuals are changed by their social environments, it is because they allow themselves to be. The evidence presented here demonstrates that individuals make decisions on which contexts and networks to select into at least in part based on their personal political beliefs and interest in politics. Utilizing a novel three-wave panel study of first-year students at a large midwestern university, I demonstrate that individuals’ previously established beliefs about politics – specifically their ideological beliefs – inform selection into non-political social environments. Individuals with strong beliefs about and interest in politics choose social environments that align with their political beliefs, and it is in these environments that their attitudes about politics are affected. The temporal nature of the data and the study’s design (observing individuals before they are embedded in the social environments of interest via a network nascent approach) help us begin to tease apart the causal direction of these effects.

These results suggest that political homophily drives network selection for those with the strongest ideological beliefs about politics. Strong liberals and Democrats join politically confirming social environments and their beliefs are reinforced in those settings; whereas, the political beliefs of weaker identifiers and political moderates and independents change significantly. There is suggestive evidence, then, that influence is more likely to occur among members of the elusive “middle”, which echoes foundational work in this area (Lazarsfeld et al., 1948; Berelson et al., 1954). However, much work drops or discounts those individuals who identify as moderates or independents and those who report not knowing their political beliefs. Results from this analysis suggest that focusing solely on partisans and ideologues may result in claiming influence effects when homophily actually drives results. Accordingly, then, I argue that there should be more explicit acknowledgement that political beliefs drive context selection for the strongest political identifiers in some settings: specifically, those settings which align with individuals’ political beliefs.

The design employed here first contacts individuals before they are embedded in social contexts, tracks environment selection over time, and can measure the change in political attitudes as a result of those selections via a network nascent approach (Lazer et al., 2010). In fact, it is difficult to imagine a real-world setting, outside of the university, where you can uniformly establish the timing at which social environments are selected. However, while this design improves upon inferences from cross-sectional and, even, longitudinal data, it still cannot isolate causal effects.

Of course, a primary limitation of this design is the external validity. This study took place at specific location at a specific time – at a midwestern university in the lead-up and aftermath of the 2016 presidential election – and the results reflect this reality. The advantage of starting the panel with first year college students is that they are in a mostly new environment – making new associational network choices for the first time, and participants are not embedded in the university environment when contact is first made.

Furthermore, an individual’s time in college is constrained in time, representing only a very small portion of their adult lives. Individuals often experiment with different aspects of their identity, both political and otherwise. Certainly, these realities place further limitations on the external validity of the findings. However short and fluid, I believe that college is a very consequential time in an individual’s life as foundational works in social networks and political behavior attest (e.g., Newcomb et al., 1967; Jennings & Niemi, 1968).

While entrance into college is a unique time in individuals’ lives, it is not without parallels. Individuals who take a new job in a new place and company also face similar circumstances – moving into a new social environment in which they were not previously embedded. Of course, these individuals make this decision based on a host of factors, some of which may include some ties to the new social environment (e. g., knowing a friend who enjoys living in that city) – similar to decisions on which college to attend. Despite obvious limitations to the generalizability of the findings, I believe this design provides a first step toward understanding the extent to which political beliefs drive selection into social contexts and a blueprint for replication and extension.

A second limitation of this design, which impedes the inferences made, is the small sample size, providing low power to find effects. While caution needs to be taken in inferring too much from these results – due to both external validity and low power concerns – I believe that, if anything, the effect of political beliefs on context selection, as found here, are understated. That persistent and robust effects were found at all given the small sample size and sometimes crude measures of social contexts gives us more confidence in the results. Results aside, the task of understanding the role of political beliefs in the selection of social networks and contexts is worth exploring.

As mentioned in the introduction, a related, but slightly different issue arises from the reality that individuals share common environments, which can impact individuals in similar ways. College roommates, for example, may become similar because they share a common dorm and campus environment and not because they have influenced one another. While the research design was designed to address the selection problem, it does not take into account influences from the shared environment. Because panel participants were all incoming first-year students at the same university, it is possible that this group of individuals were exposed to common, unobserved factors that act to influence both their likelihood of joining politically confirming social environments as well as their political beliefs.

Future work should extend the empirical results presented in this article. Do results apply only toward formal organizations or do they apply toward other contexts as well? What about the selection of friends? While the data used in the current study include other measures of social networks (important matters discussants and the partisanship of their friends) and contexts (college dormitories), I cannot be certain, especially with friendships, when these relationships were formed – whether they originated before individuals’ arrival on campus or afterward. Finally, I believe that the time, effort, and money necessary to employ this design on a larger scale is absolutely worth the effort. While this analysis provides a necessary first step, replication of these results in other samples and in other periods of time is imperative.

Supplemental Material