Pulling Closer and Moving Apart

Abstract

This article reconciles two seemingly incompatible expectations about interpersonal interaction and social influence. One theoretical perspective predicts that an increase in interaction between two actors will promote subsequent convergence in their attitudes and behaviors, whereas another view anticipates divergence. We examine the role of political identity in moderating the effects of interaction on influence. Our investigation takes place in the U.S. Senate—a setting in which actors forge political identities for public consumption based on the external constraints, normative obligations, and reputational concerns they face. We argue that interaction between senators who share the same political identity will promote convergence in their voting behavior, whereas interaction between actors with opposing political identities will lead to divergence. Moreover, we theorize that the consequences of political identity for interpersonal influence depend on the local interaction context. Political identity’s effects on influence will be greater in more divided Senate committees than in less divided ones. We find support for these hypotheses in analyses of data, spanning over three decades, on voting behavior, interaction, and political identity in the Senate. These findings contribute to research on social influence; elite integration and political polarization; and identity theory.

Keywords

social influence social identity social networks ideology political networks polarization

The consequences of interpersonal interaction for social influence—for example, how contact between individuals and groups can lead to shifts in their respective attitudes, beliefs, or behavior—have been long-standing concerns for sociologists and social psychologists (De Groot 1974; French 1956; Marsden 1981). Indeed, the interplay of interaction and influence has animated research on topics as wide-ranging as attitude change in small groups (e.g., Friedkin 1999), risk-taking in individual versus group settings (e.g., Cartwright 1971), the polarization of public opinion and political ideology (e.g., Baldassarri and Gelman 2008), and the role of elites in promoting similarity in corporate behavior (e.g., Mizruchi 1989).

Across these diverse contexts, a central line of inquiry has united research on social influence: How do proximity and the frequency of interaction affect the tendency for actors’ attitudes, beliefs, or behavior to either converge or diverge? One set of conceptual arguments emphasizes convergence. Friedkin (1993:862–63) succinctly summarizes this view: “Frequent communication tends to embed opinions in a supporting fabric of arguments and information. . . . Hence the pressure toward uniformity of opinions that arises from a comparison of opinions . . . is likely to be more pronounced and sustained when issue-related communication is frequent.” An alternative perspective highlights divergence and group polarization—stemming, for example, from social comparisons (e.g., Baron et al. 1996), persuasive arguments (e.g., Burnstein 1982), and repeated attitude expressions (e.g., Brauer, Judd, and Gliner 1995).

In line with attempts to integrate these disparate research traditions (e.g., Baldassarri and Bearman 2007; Friedkin 1999), we seek to deepen our understanding of the conditions under which heightened interaction leads to convergence or divergence in attitudes and behaviors. Building on the burgeoning literature on political networks (McClurg and Lazer 2014; Parigi and Bearman 2008; Parigi and Sartori 2014), we situate our investigation in the context of the U.S. Senate from 1973 to 2009. Although social attitudes in the United States have, with few exceptions, not become more polarized in recent years (DiMaggio, Evans, and Bryson 1996; Fiorina and Abrams 2008), the U.S. Senate has witnessed marked increases in political polarization during this period (Hetherington 2001; McCarty, Poole, and Rosenthal 2006). As Poole and Rosenthal (1997:232) presciently concluded from their seminal study of voting behavior in Congress: “The degree of polarization in Congress is approaching levels not seen since the 1890s. . . . Intense conflict between . . . parties will continue.” Given that senators have coherent, visible, and resonant political identities and seek to exert influence over one another through interaction, the U.S. Senate is an apt setting for the study of interaction, identity, and influence.

In this article, we strive to make three main contributions. First, we provide an account of increased polarization in the U.S. Senate that complements prevailing explanations, such as the geographic sorting of voters along partisan lines and the growing importance of political activists who are themselves more polarized (Theriault 2008). We do not seek to adjudicate among these macro-level explanations; rather, our aim is to uncover the microfoundations of these behavioral shifts. To do so, we draw on a rich dataset of period-to-period changes in the distance between pairs of senators on a well-established measure of political voting behavior (Poole and Rosenthal 1997) and their proximity to one another in physical and social space, as defined by their seat locations on the chamber floor (Chown and Liu 2014) and committee co-membership, respectively. Second, we help reconcile the competing theoretical perspectives on interaction and behavior change by highlighting the moderating role of political identity. To preview our findings, we show that greater interaction among senators sharing the same political identity resulted in convergence in their subsequent attitudes and behaviors, whereas an increase in contact among senators with opposing political identities led to divergence. Finally, our work shows how the local context in which interaction occurs can moderate the effects of identity on social influence. We report evidence that political identity played a role in voting behavior change in more divided Senate committees but did not matter in less divided ones.

Theory

Network-Analytic Approaches to Social Influence

As Marsden and Friedkin (1993:127) note, “The study of social influence is a strategic arena for social network research; it links the structure of social relations to attitudes and behaviors of the actors who compose a network.” Early conceptual work in this vein began with simple mathematical formulations that described the structural origins of group consensus (De Groot 1974; French 1956; Harary 1959). Later work accounted for group outcomes that fell short of full consensus (Friedkin and Johnson 1990; Marsden 1981).

A prominent formulation, referred to as social influence network theory (Friedkin 1998), suggests that a person’s attitudes and behaviors can change endogenously through the influence of others and exogenously by the conditions that formed their original views. We draw on a recent conceptualization (Baldassarri and Bearman 2007), which assumes that actors hold multiple opinions on diverse issues, actors’ overall perspectives can be characterized by aggregating their views on these issues, actors’ views are susceptible to influence from all other actors with whom they are connected, and attitude and behavior change occurs through interaction with others who have similar or opposing views.

Political Identity in the U.S. Senate

Identity is a core feature of social life and a key theoretical construct in psychology (Ellemers and Haslam 2012; Hogg and Turner 1985; Tajfel and Turner 1979), sociology (Burke and Stets 2009; Hogg and Ridgeway 2003; McCall and Simmons 1978; Stryker and Burke 2000), and political science (Fowler and Kam 2007; Ringe, Victor, and Gross 2013; Schildkraut 2011; Shayo 2009). Identity involves “the recognition of and participation in a web of social relations or communities that envelop the self and through which individuals feel themselves as identical with others” (Berezin 2001:84). It is often multifaceted, as people relate to and derive meaning from multiple communities of similar selves (Agnew and Brusa 1999; Calhoun 1993). People possess private identities, which are informed by emotion and tradition, and public identities, which are governed by interest and rationality. We focus on a particular kind of identity—political identity—that has private and public elements (Berezin 2001; Kanazawa 2000; Somers 1993).

In the U.S. Congress, senators forge political identities for public consumption based on external constraints, such as promises they made on the campaign trail and to key donors; normative obligations, such as commitments they made to party leaders and the history of prior support they received from colleagues; and reputational concerns, such as the extent to which they are known for hewing to and reinforcing the party line. In the Senate, perhaps the most salient manifestation of political identity is party affiliation. Each of the two main parties—Republican and Democratic—has a collective political reputation, shared leadership, a common party line, a history of supporting and sanctioning same-party colleagues, and a common opponent in the form of the other party. At the same time, political identities in the Senate can transcend party boundaries. For example, two senators from states with shared economic interests or matching levels of religiosity may construct similar political identities; senators whose constituents have competing economic, social, or moral interests, however, may also fashion opposing identities.

Political Identity and Social Influence

Once an identity—including a political identity—is deemed salient, the normative pressures defined by that identity lead people to think and act in ways that conform to the norms of the identity group. That is, when activated, identities create the conditions for social influence (for a review, see Abrams and Hogg 1990). Specifically, the process unfolds in the following steps: (1) people define themselves as members of a social category—for example, a group defined by party affiliation; (2) they learn the expected and desirable behaviors correlated with membership in that category; (3) they assign the norms and attributes of the category to themselves through a process of internalization; and (4) their behavior changes to conform to the norm as their category membership becomes more salient (Turner and Reynolds 2012).

Experimental evidence from social psychology supports this conceptualization. For example, participants in an experimental study were better able to recall information they thought relevant to the group with which they identified than information considered not relevant to their group (Maitner et al. 2010). Similarly, in another study, participants were more likely to change their attitudes in response to arguments they perceived were made by members of their identity group rather than arguments they thought were made by people not in their identity group (Mackie and Cooper 1984). Outside the laboratory setting, a longitudinal field study found that students’ political views were more likely to converge when they shared a positive tie than when they did not share a positive tie (Lazer et al. 2010). Finally, agent-based simulations of group dynamics show that interaction among actors with a shared identity, in the form of a positive relationship, reduces the ideological distance between them (Kitts 2006; Macy et al. 2003).

In the U.S. Senate, identity-based sources of social influence are amplified because of the fundamental oppositional nature of the two main political parties. In a sense, we can think of senators from the same party as having a positive tie to one another in terms of their political identity—even if they have an acrimonious personal relationship. Conversely, senators across the aisle in terms of party affiliation have a negative tie to one another regardless of whether they have an amicable or antagonistic personal relationship (cf. Almquist and Butts 2013). These positive and negative ties of political identity create strong normative pressure to conform to the expected attitudes and behaviors of a senator’s party or other salient political identity group. Because senators (like most people) have a desire to maintain a positive self-concept, thinking or acting in ways that violate these expectations creates cognitive dissonance, which in turn causes attitudes and behaviors to fall in line with expectations (Aronson 1968; Festinger 1957; Stone and Cooper 2001; Thibodeau and Aronson 1992). Thus, when two people sharing the same political identity come into greater contact with one another, their shared political identity becomes more salient and the normative pressure to reach alignment in thoughts and action increases. Moreover, strategic and public self-presentations, of the kind that senators routinely undertake when they interact with each other, tend to produce lasting changes in views of the self and behavior changes that persist across social settings (Schienker, Dlugolecki, and Doherty 1994). We therefore argue that an increase in interpersonal interaction between senators sharing the same political identity will lead to convergence in voting behavior:

Hypothesis 1a: The more senators with the same political identity interact with one another, the more their subsequent voting behavior will tend to converge.

How will an increase in contact between people with opposing political identities change their attitudes and behaviors? One set of insights comes from the agent-based simulations mentioned earlier. These studies show that interaction among actors who are negatively tied to one another increases distance and leads to group polarization (Kitts 2006; Macy et al. 2003). Another perspective suggests that interaction among people with opposing views can result in compromise or conflict, depending on their relative positions on a broader set of topics. As Baldassarri and Bearman (2007:792) note: “If [two actors] have contrasting views on a focal issue, but share similar opinions on the remaining issues, they compromise by reducing their commitment on the salient issue, thus moving closer to each other. In contrast, if they disagree on other issues, their commitment to the focal issue is reinforced and their opinions diverge further.” In already polarized settings, such as the U.S. Senate, heightened interaction among senators with opposing political identities will lead to divergence in voting behavior because senators’ conflicting identities will become more salient and the normative pressure to move further apart in thoughts and action will intensify. Subject to the scope condition of preexisting polarization, we expect the following:

Hypothesis 1b: The more senators with conflicting political identities interact with one another, the more their subsequent voting behavior will tend to diverge.

Local Context: Degree of Past Division in Committees

Having contended that political identity will moderate the effects of interaction on social influence, we turn next to exploring how the local context in which interaction occurs can amplify or dampen this effect. We posit that group contexts in which members have experienced past division across identity group lines will tend to make political identities more salient. By contrast, political identities will matter less in group contexts where members have collaborated effectively in the past across identity group lines. Thus, political identity should play a more important role in moderating the effects of interaction on identity in the former context, relative to the latter.

In the U.S. Senate, the committees on which senators serve and deliberate on key issues represent one of the most prominent contexts for local interaction. Within these committees, senators discuss, debate, and amend preliminary pieces of legislation, or bills, which are assigned to them by the broader legislative body. Committee deliberations represent the most significant hurdle for the passage of bills: once a bill passes through committee, it has a high likelihood of becoming law (Fenno 1973). Committees are venues in which senators wrangle with one another over bills but also seek to collaborate by co-sponsoring bills that matter to them, their constituents, and other key stakeholders. Because tradition dictates that party representation on committees should reflect the overall composition of the Senate, committees are key sites for within- and cross-party political interaction.

Committees vary considerably in membership, with some consisting of senators who have a track record of co-sponsoring many bills across party lines and others whose members co-sponsor bills primarily within party lines. We can think of these different kinds of committees as varying in their level of past division. The former is less divided because those senators have a history of working effectively across party lines; the latter is more divided because these committee members collaborate primarily within party lines.

We suggest that the more divided a Senate committee is, the more likely it is to create a local interaction context in which senators’ political identities become salient. In a divided committee, where senators do not collaborate extensively across party lines, the political fissures between the two main parties will be more evident to committee members.¹ Thus, the tendency for political identity to moderate the effects of interaction on voting behavior convergence will be amplified when senators’ interactions take place in the context of more divided committees. By contrast, interaction in the context of less divided committees is less likely to trigger identity-based sources of social influence because senators in such settings are surrounded by colleagues who have collaborated effectively across party lines. Together, these arguments suggest the following:

Hypothesis 2: The tendency for political identities to influence the convergence or divergence of senators’ voting behavior will be greater in more divided Senate committees than in less divided committees.

Data and Methods

Empirical Setting: The U.S. Senate

To test these hypotheses, we examined interactions, group contexts, and voting behavior changes in the Senate, the upper house of the U.S. national legislature. This empirical setting is especially well-suited to our theoretical aims for four reasons. First, senators’ voting behavior is part of the public record, and the (changing) distance in voting behavior between each pair of senators can be readily observed. Second, senators interact with one another in multiple, observable settings, including the Senate chamber and a range of committees (Deering and Smith 1997). We can therefore derive two distinct indicators of interpersonal interaction. Third, we can characterize the local interaction context of different committees as more or less divided by drawing on the record of past bill co-sponsorships between each pair of same- and different-party senators. Finally, because of the two-party system in the United States, senators construct political identities that are often in close alignment with or direct opposition to one another. In summary, the U.S. Senate represents a strategic site for the study of political identity, interaction, and influence.

Our hypotheses focus on the tendency for interaction to lead to the convergence or divergence of voting behavior—that is, we theorize about changes in the voting distance between two senators. We thus analyze interpersonal influence at the level of dyads rather than individuals. Dyads are the appropriate unit of analysis for another key reason: rather than assuming that actors have the same level of influence over all other actors, we allow for heterogeneity in actors’ influence over other actors (Friedkin 1993).²

Our analyses cover the years 1973 to 2009. This time frame allows us to collate a rich dataset that encompasses a period of stable, two-party rule, with both parties occupying the White House at different times, and neither party outright dominating the other.

Dependent Variable

Because senators’ votes on contentious issues—for example, regulation of business practices, tax policy, and abortion rights—are often highly correlated, one can distill these votes into a composite measure of voting behavior (cf. Baldassarri and Bearman 2007). We adopted Nokken and Poole’s (2004) variant of a widely used measure of voting distance in the U.S. Congress: Dynamic Weighted Nominate, or DW-Nominate. This approach extrapolates from each senator’s observed voting record to map each senator’s behavior in a given Congress onto a point in Euclidean space. In colloquial terms, one can think of this space as spanning the spectrum from liberal to conservative (Poole and Rosenthal 1997). Senators’ composite voting behavior can range from −1 (especially liberal) to 1 (especially conservative). We used this behavioral score to construct our dependent variable: the absolute voting distance between two senators within a given Congress, which can range from 0 to 2. Figure 1 depicts the distribution of this variable.

Figure 1.

Distribution of Senators’ Aggregate Voting Behavior on Spectrum from Liberal to Conservative

Independent Variables

We constructed two measures that proxy for the degree of interaction between senators. The first is based on geographic proximity. A long literature—dating back to Bossard’s (1932) seminal work on marriage selection and Festinger, Schachter, and Back’s (1950) treatise on spatial configurations—suggests that geographic proximity is one of the most salient determinants of interaction. To examine the effects of geographic proximity, we focus on the Senate chamber—a large, 16-by-26 meter room in which each senator is assigned a desk. In this chamber, senators debate legislation, work to build consensus, and vote on bills. Every two years, after an election, the chamber map is redrawn, and chamber desks are rebolted to new locations on the floor. By tradition, each party is apportioned a side of the chamber floor, and senators sequentially choose their desks in order of seniority. Figure 2 presents a graphic representation of the chamber. Using archives of seating locations, we mapped each senator’s desk location onto Cartesian space, and we used these coordinates to construct the precise chamber distance between each pair of senators. Senators whose desks were closer together were more likely to interact with one another (Chown and Liu 2014). Hence, our first measure of interaction between senators is chamber distance. Figure 3 depicts the distribution of this variable.

Figure 2.

U.S. Senate, Map of the Chamber Floor

Figure 3.

Distribution of Chamber Distance between Dyadic Pairs

Our second measure considers the extent to which two individuals had overlapping sites of interaction within the Senate. Specifically, we counted the number of committees that two senators served on together, committee co-memberships, within a Congress (Nelson and Stewart 2011). Senate committees are formally constituted subgroups of the larger legislative body. After initial legislation in the form of a bill is drafted, it is then referred to the appropriate committee, which gathers information, holds hearings, and revises the bill. Most bills fail to garner sufficient support at the committee stage and are never voted on by the full Senate. Because a significant amount of senatorial work is conducted in committees, these groups represent a major locus of interpersonal interaction and influence.

Our theory suggests that the effects of interaction may lead to convergence or divergence of voting behavior, depending on individuals’ political identities. We use individuals’ party affiliation (i.e., Democrat, Republican, or Independent) as an indicator of their political identity.³ Party affiliations define senators’ core electorate and also trace their most likely alliance partners and sources of opposition within the chamber. Members of the same party typically seek to craft a common political agenda and are apt to lend one another political support, whereas members of different parties are more likely to oppose one another. As a reflection of the importance of party affiliation, members of the two parties are seated in separate halves of the Senate chamber and are designated as majority or minority members of a committee. Our measure of shared or contrasting political identity is same party, an indicator set to one for dyads in which both senators had the same party affiliation and zero otherwise. To test Hypothesis 1, we interacted our measures of the intensity of interaction, chamber distance and committee co-membership, with our indicator variable of shared political identity, same party. In supplemental analyses described in the Results section, we generated an alternative measure of political identity based on the degree of religiosity among constituents of the state that each senator represents.

In developing Hypothesis 2, we theorized that contextual features of the groups in which interaction occurs will shape the tendency toward ideological convergence or divergence. To measure the level of past division that occurred in a committee, we used data on bill co-sponsorships among U.S. senators (Fowler 2006a, 2006b). Bill co-sponsorships are public endorsements of support that one senator makes for another’s policy initiatives. In our data, within any given Congress, about 45 percent of bills received one or more co-sponsors.

To capture the history of division within a committee, we generated the average number of bill co-sponsorships per dyadic pair on the committee, separated out by bill co-sponsorships within the same political party versus those across party lines. For each committee, we then measured the ratio of within-party to cross-party co-sponsorships (see Figure 4). When this ratio is close to 1, bill co-sponsorships are just as likely to occur between senators from the same party as between senators from opposing parties. In such committees, cross-party division is low. By contrast, a ratio greater than 1 indicates that bill co-sponsorships are more likely to occur between senators from the same party than between senators from opposing parties. For these committees, division is expected to be high. As Figure 4 illustrates, committees vary considerably in past division. To capture this variation, we implemented a median split of committees based on their ratio of within- to cross-party co-sponsorships and then created two separate counts of shared committees of each type between pairs of senators.⁴ Finally, we created interaction terms of these committee co-membership counts and our measure of shared political identity, same party, to test Hypothesis 2.

Figure 4.

Distribution of Committee Division

Estimation

To examine the impact of changes in interactions on voting distance between two senators, we regressed voting distance in the next period on the degree of interpersonal interaction that occurred within the Senate in the present period. Formally, our baseline regression model is represented as follows:

\begin{array}{l} E [y_{i j t + 1} | X_{i j t}] = β_{0} + β_{1} C h a m b e r D i s t a n c e_{i j t} \\ + β_{2} C o m m m i t t e e C o - m e m b e r s h i p_{i j t} \\ + β_{3} X_{i j t} + δ_{t} + γ_{i j} + ε_{i j t}, \end{array}

where y is the voting distance between i and j in the subsequent Congress, t + 1; chamber distance is the geographic distance between senators i and j in meters in Congress t; committee co-membership is a count of the committees shared by i and j in Congress t; X is a vector of control variables; δ_t represents fixed effects for each Congress; and γ_ij corresponds to dyad fixed effects.

The inclusion of dyad fixed effects allows us to focus on within-dyad variation. In this specification, all time-invariant characteristics of individuals—for example, gender, party affiliation, starting ideology, or cohort—as well as characteristics of dyads—for example, whether two senators were of the same gender, were elected at the same time, or shared the same ideology at the time they were elected—net out. Put differently, we linked changes in voting distance, our dependent variable, to changes within the dyad (e.g., chamber distance or committee co-membership) over time. We controlled for any residual effects of increasing seniority in dyads by including the sum of their senate tenures and their changing personal relationship over time by including their history of bill co-sponsorships (logged). The inclusion of Congress fixed effects accounts for unobserved time heterogeneity—for example, years in which members elected to the Senate held especially extreme ideological views.

Non-independence of Observations

The error terms in regressions of dyadic network data will be correlated across observations—a problem referred to as network autocorrelation (i.e., clustering). Failure to account for clustering can lead to under-estimated standard errors and over-rejection of hypothesis tests. To address this issue, we employed a variance estimator that enables cluster-robust inference (Cameron, Gelbach, and Miller 2011). This approach to adjusting standard errors is appropriate for analysis of social network data (Dahlander and McFarland 2013; Kleinbaum, Stuart, and Tushman 2013). We considered, but decided against, stochastic actor-based models of the kind estimated using the software program SIENA (Steglich, Snijders, and West 2006). These models assume a dichotomous dependent variable and become more difficult to estimate as the number of time periods increases (Snijders, van de Bunt, and Steglich 2010); thus, they are not appropriate for analyzing a dataset in which the dependent variable is a distance measure, particularly one that spans such a long time period.

Accounting for the Role of Selection onto Committees

Although regressions with dyad fixed effects account for unobserved heterogeneity among individual senators and pairs of senators, questions about how senators sort themselves onto committees could still undermine our claims about the role of political identity in moderating the effects of interaction on influence. For example, senators from the same party might choose to jointly serve on a committee because they expect their behaviors to later converge. Alternatively, a senator from one party might choose to serve on a committee to block the influence of a senator from a different party whose voting patterns diverge from her own. We address these concerns in two ways: (1) by reviewing institutional features and norms of the U.S. Senate that make it unlikely senators chose or were assigned to committees in ways that could provide an alternative explanation for our findings; and (2) by explicitly accounting for selection onto committees in supplemental regression analyses.

The committee assignment process in the Senate has three goals: (1) ensuring that each committee is staffed with the requisite number of senators; (2) responding to the preferences of individual senators; and (3) limiting direct personal conflict among senators who have overlapping preferences. After each election, the Senate Committee on Committees determines the size of each committee. Each party’s share of committee seats reflects the overall party composition of the Senate in that Congress. Although each party allocates its members to committees independently, both parties adhere to long-standing norms. First, members can retain their committee assignments for as long as they desire. Second, both parties defer to seniority when it comes to resolving competing demands. Finally, senators are limited to two major committee assignments and one minor committee assignment (exceptions are sometimes granted) (Deering and Smith 1997).

As a consequence of these norms, senators often do not receive their preferred committee assignments. For example, Senator Dan Quayle related the following experience:

You literally sit around in a room with a sheet of paper in front of you and pick committees in order of seniority. Foreign Relations had no vacancies. It would have been my first choice. My next choice was Finance. [Three other senators] took it before me, so I missed it. I said “Armed Service.” It covers some of the same problems as Foreign Relations. . . . On the second round, I wanted Governmental Affairs. But I sat there watching, and I saw that I could be third ranking on Labor and Human Resources. I noticed everyone was shying away from it. I didn’t have any interest in it, to tell you the truth. I hadn’t even thought of it. But if I were third ranking—and Bob Stafford retired next year, I could be second ranking—assuming Orrin Hatch is reelected. So I said to myself right there, “Why not take it? There are a lot of important policies there—education, employment, labor. I took it on the spot.” (Fenno 1989:23–24)

Concerns about anticipated alignment or misalignment in voting behavior did not factor into Senator Quayle’s account or other comparable accounts of how members chose or were assigned to their committees.

Although these institutional features significantly reduce concerns about the threat to causal inference from the process by which senators are sorted onto committees, we nevertheless conducted supplemental analyses that directly account for this selection mechanism. In particular, we used an empirical approach that has gained acceptance in biostatistics and has since diffused to the social sciences: Inverse Probability of Treatment Weighted (IPTW) estimation (Hernán, Brumback, and Robins 2001; Robins, Hernán, and Brumback 2000). This estimator is related to propensity-score matching (Rubin 2006) and similarly assumes that selection into treatment is based on observable characteristics (Azoulay, Ding, and Stuart 2009).

We implemented the procedure in three steps. First, for each dyad in our sample, we estimated the probability of serving together on at least one committee as a function of same party-Democrat, same party-Republican, same cohort, same gender, and same state. Next, we computed the inverse of this predicted probability. Finally, in the models pertaining to committee co-membership (i.e., models reported in Tables 5 and 6), we weighted observations by the inverse of these predicted probabilities. In effect, this methodology created a quasi-random sample, giving more weight to atypical observations and less weight to typical ones.

Results

We begin with a description of the data. Table 1, Panel A describes characteristics of the 276 senators in our dataset. The median senator was male and joined the Senate in 1979 (i.e., the 96th Congress). On average, 12 senators entered each Congress, although this number ranges widely in our sample, from 6 to 21. Across the entirety of our dataset, senators were evenly divided between political parties, and voting behavior was centered at 0. Panel B in Table 1 describes characteristics of the 64,856 senator dyads in our dataset. The median senator pair had a voting behavioral distance of .417, a chamber distance of approximately 10 meters, and sat on .57 committees together. Table 2 shows the distribution of dyads by number of shared committees. Over 54 percent of dyads had no shared committees, and 35 percent had just one. Less than 1 percent of dyads served on three or more committees together.

Table 1.

Descriptive Statistics

	Mean	SD	Min.	Max.
Panel A: Senator Characteristics (N = 276)^a
Female	.043	.204	0	1
Democrat	.486	.501	0	1
Republican	.504	.501	0	1
First Congress	96.82	7.576	77	109
Last Congress	104.3	6.839	94	112
Tenure (in Congresses)	2.547	3.210	1	18
Voting behavior index	.012	.383	−.700	.992
Panel B: Senator-Dyad Characteristics (N = 64,856)^b
Same gender	.912	.283	0	1
Same party	.498	.500	0	1
Same state	.010	.099	0	1
Voting distance	.417	.311	.00	1.81
Chamber distance (meters)	1.04	5.620	.68	24.0
Committee co-membership (count)	.570	.707	0	5
Committee division	1.234	.178	.38	1.94

Sample includes the 276 senators included in the dyad-level regressions.

Sample includes 19,038 unique dyads (i.e., ij) and 64,856 observations (i.e., ijt). For average committee division, we report averages only for the 29,532 dyads that had one or more shared committees.

Table 2.

Distribution of Dyads by Committee Co-membership

Number of ij Committee Co-memberships	Frequency	Percent
0	35,324	54.47
1	22,901	35.31
2	5,892	9.08
3	694	1.07
4	41	.06
5	4	.01

Table 3 describes patterns in how the distance in voting behavior among pairs of senators changed over time. Across all Congresses, the mean distance between senators was .42 (on a scale ranging from 0 to 2), although this distance increased significantly over the observation period (column 1). Much of this increase in polarization was driven by cross-party (column 3) rather than within-party (column 2) changes. This increasing polarization could be due to incumbents shifting their behaviors to adopt more extreme positions, newcomers becoming increasingly radicalized, or both. To further illuminate these trends, we examined these changes within two subsets of dyads: senators who entered Congress before the observation period (i.e., preexisting senators, column 5) and entering senators (column 7). Aggregate increases in cross-party polarization arose from increases in voting distance in both sets of dyads.

Table 3.

Mean Voting Distance by Type of Dyad, over the Observation Period

	(1) All Dyads	(2) Same-Party Dyads	(3) Different-Party Dyads	(4) Preexisting Same-Party Dyads	(5) Preexisting Different-Party Dyads	(6) First-Year Same-Party Dyads	(7) First-Year Different-Party Dyads
94th Congress	.401	.235	.569	.247	.559	.163	.794
95th Congress	.389	.230	.556	.252	.535	.197	.630
96th Congress	.359	.174	.545	.179	.524	.183	.556
97th Congress	.381	.198	.555	.211	.539	.217	.621
98th Congress	.377	.200	.555	.211	.535	.144	.683
99th Congress	.387	.213	.559	.236	.527	.122	.783
100th Congress	.389	.171	.606	.192	.536	.141	.638
101st Congress	.398	.171	.625	.216	.635	.161	.585
102nd Congress	.407	.174	.642	.197	.634	.300	1.080
103rd Congress	.421	.173	.671	.211	.632	.165	.870
104th Congress	.445	.180	.708	.176	.615	.222	NA
105th Congress	.468	.184	.753	.192	.709	.188	.800
106th Congress	.473	.179	.767	.188	.725	.162	.749
107th Congress	.466	.176	.751	.159	.920	.185	.708
108th Congress	.447	.163	.715	.091	.744	.112	.687
109th Congress	.481	.182	.770	.093	.672	.256	.937

Note: Mean voting distances are based on Nokken and Poole’s (2004) measure. We define preexisting dyads as dyads composed of senators who both entered before the 96th Congress. First-year dyads consist of senators who both entered in the given Congress. For the 104th Congress, all entering first-year senators were Republicans.

Our empirical analyses sought to explain shifts in voting distance stemming from changes in interaction, using two distinct indicators of interaction: geographic distance in the Senate chamber and committee co-membership. The mean chamber distance between senators was 10 meters. Given the allocation of chamber seats by party, the mean intra-party distance was six meters, and the mean inter-party distance was 14 meters (see Figure 3). Between Congresses, the median senator moved 1.7 meters on the floor; relocations tended to occur in the early years of a senator’s tenure. After one’s fifth Congress (i.e., 10 years of service), a senator typically settled into a fixed desk position on the chamber floor. The mean number of committee co-memberships was .5.

Table 4 reports the first set of analyses pertaining to Hypotheses 1a and 1b: increased interaction, as indicated by greater physical proximity, between senators sharing the same salient political identity will lead to subsequent convergence in voting behavior, whereas increased interaction between senators with oppositional political identities will lead to divergence. In Model 1, the baseline, we first examine the effects of interaction on voting distance, without taking into account senators’ political identities. Both chamber distance and committee co-membership have a statistically significant effect. The two coefficients both indicate that the more two senators interacted with each other, the less similar their voting behaviors became in the subsequent period. In interpreting the effects, note that chamber distance and committee co-membership are opposing indicators of interaction: an increase in chamber distance is associated with a decrease in interaction, whereas an increase in committee co-membership suggests more interaction.

Table 4.

OLS Regressions of Distance in Voting Behavior on Interaction and Political Identity (Hypotheses 1a and 1b, Interaction Based on Chamber Distance)

	(1)	(2)	(3)	(4)
Chamber Distance	−.002***	−.004***	−.004***
Chamber Distance	(.000)	(.001)	(.001)
Committee Co-membership	.003*	.003*	.003*	.003***
Committee Co-membership	(.001)	(.001)	(.001)	(.001)
Chamber Distance x Same Party		.005***
Chamber Distance x Same Party		(.001)
Chamber Distance x Same Party-Republican			.005***
Chamber Distance x Same Party-Republican			(.001)
Chamber Distance x Same Party-Democrat			.005***
Chamber Distance x Same Party-Democrat			(.001)
Chamber Distance (0 to 2 meters)				.055***
Chamber Distance (0 to 2 meters)				(.016)
Chamber Distance (2 to 10 meters)				.038***
Chamber Distance (2 to 10 meters)				(.007)
Chamber Distance (10 to 15 meters)				.019***
Chamber Distance (10 to 15 meters)				(.004)
Chamber Distance (0 to 2 meters) x Same Party				−.062**
Chamber Distance (0 to 2 meters) x Same Party				(.020)
Chamber Distance (2 to 10 meters) x Same Party				−.043**
Chamber Distance (2 to 10 meters) x Same Party				(.014)
Chamber Distance (10 to 15 meters) x Same Party				−.019
Chamber Distance (10 to 15 meters) x Same Party				(.012)
Prior Bill Co-sponsorships (Log)	−.006	−.006	−.006	−.005
Prior Bill Co-sponsorships (Log)	(.004)	(.004)	(.004)	(.004)
Constant	.443***	.450***	.450***	.415***
Constant	(.015)	(.015)	(.016)	(.014)
Congress Fixed Effects	Yes	Yes	Yes	Yes
Dyad Fixed Effects	Yes	Yes	Yes	Yes
Prob. > Chi²	5.9e-199	9.2e-239	1.7e-238	1.9e-226

Note: N = 64,856. Chamber distance (75 to 100 percent) is the excluded category in Model 4. The summed tenure of i and j is included but not shown. Standard errors are clustered in two dimensions: by i and by j.

p < .05; **p < .01; ***p < .001 (two-tailed tests).

Our next set of analyses sought to unpack the aggregate effect of increased interaction on voting behavior change by considering the role of political identity. In Model 2 of Table 4, we interacted chamber distance with the political identity indicator, same party. The main effect of same-party membership, which was time-invariant, was subsumed by the dyad fixed effects. Results indicate that when two senators from the same party moved closer together on the chamber floor, their voting distance in the next period decreased. By contrast, when two senators from different parties moved closer together, their voting distance increased. For senators from the same party, a one-standard-deviation decrease in chamber distance resulted in a 10.4 percent decrease in voting distance, whereas for different-party senators, a one-standard-deviation decrease resulted in a 2.7 percent increase in voting distance.⁵

Although these changes may seem modest, they are substantively meaningful when we consider the baseline stability in senators’ voting behavior. For the median cross-party dyad, there was only a 14.1 percent change over the observation period. Thus, a 2.7 percent increase in voting distance represents nearly one-fifth of the lifetime change in voting behavior in the median cross-party dyad. Another way to interpret the magnitude of this effect is to consider how it would affect the location of a focal dyad on the distribution of dyads by voting distance. If the median (i.e., 50th percentile) same-party dyad moved one standard deviation closer together in the senate chamber, that dyad would move to the 45th percentile of voting distance in the next period. In summary, changes in senators’ physical proximity on the chamber floor yield modest yet perceptible shifts in their subsequent voting distance.

Model 3 in Table 4 decomposes the same-party variable into same party-Republican and same party-Democrat and reveals no significant differences in political identity dynamics between the two major parties. Finally, Model 4 does not assume a linear relationship between chamber distance and the subsequent change in voting behavior. Instead, we introduced indicators that allow us to flexibly identify the effects of especially close versus less close proximity on voting distance. Results indicate that the effects of chamber distance are more pronounced when senators come into especially close physical proximity, with effects tapering beyond 10 meters.

Model 5 in Table 5 reports results of the second set of analyses relating to Hypotheses 1a and 1b: the effects of committee co-memberships and political identities on voting behavior convergence or divergence. Consistent with Table 4, we find that same-party senators converged in their subsequent voting behavior when they experienced an increase in committee co-memberships. For median same-party dyads, each additional committee they served on together led to a 5.8 percent decrease in their subsequent voting distance. By contrast, different-party senators who served on more committees together tended to diverge in voting behavior. For the median different-party dyad, each additional committee co-membership was associated with a 1.1 percent increase in voting distance. These results are consistent for both parties, although Democrats exhibit greater influence on one another in voting behavior convergence than do Republicans (Model 6); the difference between these two coefficients is significant at the .005 level.

Table 5.

OLS Regressions of Distance in Voting Behavior on Interaction and Political Identity (Hypotheses 1a and 1b, Interaction Based on Committee Co-membership and Integrated Model)

	(5)	(6)	(7)	(8)
Chamber Distance	−.002***	−.002***	−.002***	−.004***
Chamber Distance	(.000)	(.000)	(.000)	(.001)
Committee Co-membership	.007**	.007**		.006**
Committee Co-membership	(.002)	(.002)		(.002)
Same Party x Committee Co-membership	−.009***			−.006**
Same Party x Committee Co-membership	(.002)			(.002)
Same Party-Republican x Committee Co-membership		−.005*
Same Party-Republican x Committee Co-membership		(.002)
Same Party-Democrat x Committee Co-membership		−.013***
Same Party-Democrat x Committee Co-membership		(.003)
Single (1) Committee Co-membership			.006*
Single (1) Committee Co-membership			(.002)
Multiple (>1) Committee Co-membership			.017***
Multiple (>1) Committee Co-membership			(.004)
Single (1) Committee Co-membership x Same Party			−.009**
Single (1) Committee Co-membership x Same Party			(.003)
Multiple (>1) Committee Co-membership x Same Party			−.019***
Multiple (>1) Committee Co-membership x Same Party			(.005)
Chamber Distance x Same Party				.005***
Chamber Distance x Same Party				(.001)
Prior Bill Co-sponsorships (Log)	−.006	−.006	−.006	−.006
Prior Bill Co-sponsorships (Log)	(.004)	(.004)	(.004)	(.004)
Constant	.443***	.443***	.443***	.449***
Constant	(.015)	(.015)	(.015)	(.016)
Congress Fixed Effects	Yes	Yes	Yes	Yes
Dyad Fixed Effects	Yes	Yes	Yes	Yes
Prob. > Chi²	3.1e-203	5.7e-206	6.6e-205	4.1e-240

Note: N = 64,856. No committee co-membership (i.e., zero shared committees) is the excluded category in Model 7. The summed tenure of i and j is included but not shown. Standard errors are clustered in two dimensions: by i and by j.

p < .05; **p < .01; ***p < .001 (two-tailed tests).

Model 7 in Table 5 does not assume a linear relationship between shared committees and voting behavior change; rather, we introduced two indicator variables to disentangle the effects of having one shared committee from that of having two or more shared committees. Results indicate a 4.5 percent decrease in voting distance when two senators from the same party went from serving on zero committees together to serving on one committee together. The same change in committee co-memberships led to a .9 percent increase in voting distance for different-party dyads. When same-party dyads went from having one shared committee to two or more shared committees, they experienced a further 5.4 percent decrease in voting distance. This same change in committee co-memberships led to a further 1.7 percent increase in voting distance for different-party dyads. Finally, Model 8 is an integrated model that considers how changes in chamber distance and committee co-memberships jointly affect voting behavior change. The hypothesized effects are robust to the inclusion of both measures of interpersonal interaction. In summary, Models 1 through 8 provide strong support for Hypotheses 1a and 1b.⁶

Table 6 reports results that speak to Hypothesis 2: the effects of political identity and interaction on voting behavior change will be greater in more divided committees than in less divided committees. Model 10 disaggregates committees into those above or below the median level of division and then interacts these two committee counts with the political identity indicator, same party. In support of Hypothesis 2, the convergence or divergence in voting behavior holds only for changes in co-memberships on more divided committees (Model 10). For median same-party dyads, each increase in the number of divided committees served on together was associated with a 10.3 percent decrease in subsequent voting distance. For the median different-party dyad, each increase in the number of divided committees served on together was associated with a 1.4 percent increase in voting distance. The effect is not statistically significant for changes in co-memberships on committees with a more equitable ratio of same- to cross-party bill co-sponsorships. Moreover, these results are robust to the inclusion of chamber distance (Model 11).

Table 6.

OLS Regressions of Distance in Voting Behavior on Committee Co-memberships and Political Identity, by Committee Division (Hypothesis 2)

	(9)	(10)	(11)
Chamber Distance	−.002***	−.002***	−.004***
Chamber Distance	(.000)	(.000)	(.001)
Committee Co-membership Count (Less Divided)	.004	.004	.003
Committee Co-membership Count (Less Divided)	(.002)	(.003)	(.003)
Committee Co-membership Count (More Divided)	.002	.009**	.008**
Committee Co-membership Count (More Divided)	(.002)	(.003)	(.003)
Committee Co-membership Count (Less Divided) x Same Party		−.001	.000
Committee Co-membership Count (Less Divided) x Same Party		(.003)	(.003)
Committee Co-membership Count (More Divided) x Same Party		−.015***	−.012***
Committee Co-membership Count (More Divided) x Same Party		(.004)	(.003)
Chamber Distance x Same Party			.005***
Chamber Distance x Same Party			(.001)
Prior Bill Co-sponsorships (Log)	−.006	−.006	−.006
Prior Bill Co-sponsorships (Log)	(.004)	(.004)	(.004)
Constant	.443***	.442***	.449***
Constant	(.015)	(.015)	(.016)
Congress Fixed Effects	Yes	Yes	Yes
Dyad Fixed Effects	Yes	Yes	Yes
Prob. > Chi²	2.9e-202	1.8e-211	3.2e-247

Note: N = 64,856. Less divided committees are those in which the ratio of same- to cross-party bill co-sponsorship rates is below the median; more divided committees have ratios above the median. The summed tenure of i and j is included but not shown. Standard errors are clustered in two dimensions: by i and by j.

p < .05; **p < .01; ***p < .001 (two-tailed tests).

To account for selection of senators onto committees, we implemented an Inverse Probability of Treatment Weighted (IPTW) estimation approach. We first estimated the likelihood that two senators would serve on a committee together and then, in the second-stage, inverse weighted each observation by its predicted probability. This approach nets out observable correlates of joint selection to a committee. Results from this analysis (not reported) show only negligible changes to our results in Tables 5 and 6 and assuage concerns about endogeneity arising from senators self-selecting onto committees based on anticipated changes in voting behavior.

Finally, we conducted a number of supplemental analyses (see the online supplement) to ensure that our results are robust to alternative specifications and modeling choices. Our results are substantively unchanged when we estimate random-effects models with time-invariant covariates, include time-varying covariates such as whether members of a dyad were committee chairs in a given Congress, add controls for triadic closure, or model the dynamics of interaction and influence at the individual, rather than dyadic, level.

Empirical Extension: Political Identities Based on the Religiosity of Senators’ Constituents

So far, our analyses provide evidence that one type of political identity, party affiliation, moderates the effects of interaction on behavioral convergence or divergence. To demonstrate the generalizability of these findings, we examined whether comparable moderation effects could be detected using a different measure of political identity. In particular, we conducted a supplemental analysis based on the degree of religiosity among constituents of the state each senator represented. Our theory predicts that senators from states with comparable rates of religiosity will construct matching political identities with respect to divisive social issues, whereas senators from states with very different rates of religiosity will construct opposing political identities with respect to the same issues.

We derived our state-level religiosity measure from responses to a Gallup poll that asked respondents the extent to which “religion is an important part of their daily life.”⁷ This measure of religiosity is not highly correlated with party affiliation: although statistically significant, the correlation between being Republican and greater state religiosity is only .063. Thus, political identities defined using this religiosity measure are distinct from political identities based on party affiliation.

Because our theory is predicated on matching or oppositional identities, we identified the subset of senator dyads whose political identities with respect to constituent religiosity would be either closely aligned or opposing. To do so, we first arrayed states on a religiosity index that ranged from 0 to 100 and placed them into quartile bins. Next, we restricted our sample to dyads of senators representing states with either very high (top quartile) or very low (bottom quartile) rates of religiosity. That is, we dropped senators from states with moderate (middle two quartiles) rates of religiosity. Then we created an indicator variable, same state religiosity, set to one for dyads in which both members came from states with very high or very low rates of religiosity. Because we dropped the middle of the distribution, the reference category is dyads from states with sharply contrasting rates of religiosity. This analytic approach sought to parallel the identity dynamics that we theorized about and empirically assessed using party affiliation. In both cases, the indicator variable represents a positive tie in terms of political identity (regardless of whether the two senators had a positive or negative interpersonal relationship), and the reference category indicates a negative tie in terms of political identity.

Table 7 reports results of these analyses. Model 12 provides the baseline and again shows a negative relationship between chamber distance and voting distance in the next period. Model 13 includes the interaction of same state religiosity and chamber distance, which is negative and statistically significant.⁸ These results lend further support for Hypotheses 1a and 1b, suggesting that a one-standard-deviation decrease in chamber distance (that is, an increase in interaction) is associated with a 2.8 percent decrease in voting distance for senators from states with the same level of state religiosity. By contrast, a one-standard-deviation decrease in chamber distance (that is, an increase in interaction) induced a 4.4 percent increase in voting distance for senators from states with sharply contrasting rates of religiosity. Model 14 indicates that these effects are robust to the inclusion of same party and its interaction with chamber distance.

Table 7.

OLS Regressions of Distance in Voting Behavior on Interaction and Political Identity (Hypotheses 1a and 1b, Political Identity Based on Religiosity of Senators’ Constituents)

	(12)	(13)	(14)
Data	i and j represent states with 0 to 25 percent or 75 to 100 percent religiosity
Chamber Distance	−.002**	−.003**	−.005***
	(.001)	(.001)	(.001)
Committee Co-membership	.003	.003	.003
	(.002)	(.002)	(.002)
Chamber Distance x Same State Religiosity		.002*	.002*
		(.001)	(.001)
Chamber Distance x Same Party			.006***
			(.001)
Prior Bill Co-sponsorships (Log)	−.005	−.005	−.006
	(.005)	(.005)	(.005)
Constant	.471***	.470***	.483***
	(.024)	(.024)	(.025)

Congress Fixed Effects	Yes	Yes	Yes
Dyad Fixed Effects	Yes	Yes	Yes
Prob. > Chi²	2.77e-65	7.91e-66	3.93e-84

Note: N = 14,851. The regression only includes dyads where i and j represent states that are in either the least (0 to 25 percent) or the most religious (75 to 100 percent) quartile. The variable same state religiosity is an indicator variable set to 1 if both i and j represent states with the same religiosity quartile (i.e., both 0 to 25 percent or both 75 to 100 percent), and set to 0 otherwise (e.g., one state is 0 to 25 percent and one is 75 to 100 percent). The summed tenure of i and j is included but not shown. Standard errors are clustered in two dimensions: by i and by j.

p < .05; **p < .01; ***p < .001 (two-tailed tests).

Finally, in supplemental analyses (not reported), we found that when the sample was constructed based on dyads that came from states that were less starkly polarized in constituent religiosity (e.g., when dyads from states with the same religiosity were compared to dyads from states that were only one or two quartiles apart rather than three quartiles apart in religiosity), these effects were attenuated and not statistically significant. Taken together, these analyses support the contention that political identities—specifically political identities that match or are opposed to one another—moderate the effects of interaction on the tendency for attitudes and behaviors to converge or diverge.

Discussion

This article aimed to contribute to our understanding of the conditions under which interaction between actors leads to greater similarity in their attitudes and behavior and the conditions under which it promotes dissimilarity. Our empirical context was the U.S. Senate, where senators forge political identities for public consumption in response to the external constraints, normative pressures, and reputational concerns they face. We first argued that greater interaction between senators with the same political identity will promote subsequent convergence in their voting behavior. Next, we posited that more interaction between senators with opposing political identities will lead to divergence in their voting behavior. Finally, we considered how the local context in which interaction occurs can affect the extent to which political identity motivations are made salient and thereby condition the effect of political identity on influence. We argued that the effects of political identity on interaction and voting behavior change will be greater in more divided interaction contexts than in less divided contexts.

Empirical support for these propositions came from analyses of interaction, identity, and influence in the U.S. Senate from 1973 to 2009. Using two distinct indicators of political identity, we demonstrated that, as the level of interaction between senators changed, their voting behavior converged or diverged as a function of their respective political identities. For political identity based on party affiliation, we found remarkable consistency in these patterns across two disparate indicators of interpersonal interaction—committee co-membership and physical proximity. Finally, our results indicate that the committee co-membership effect was concentrated in interactions that took place within more divided Senate committees but was not operative in less divided committees.

Limitations

That our findings prove robust across two distinct measures of political identity and two different indicators of social interaction bolsters confidence in our conclusions. Moreover, the use of stringent statistical controls (e.g., dyad fixed effects) helps address concerns about unobserved heterogeneity among individual senators and senator pairs. Nevertheless, concerns about the role of selection—for example, the possibility that senators sought to change their level of interaction with others because they anticipated moving closer or further apart in voting distance—cannot be fully eliminated. To address this concern, we explicitly modeled the probability of two senators sorting into the same committee and then weighted our observations by the inverse of these probabilities. Reassuringly, this analysis replicates our main findings. Still, our models cannot account for unobserved, time-varying attributes of individuals or dyads that might have affected the dynamics of social influence. Thus, we cannot completely rule out the possibility of potentially confounding, unobserved attributes.

Moreover, because the U.S. Senate represents a specialized institutional setting, we urge caution in generalizing this article’s findings to other contexts. That the hypothesized pattern of convergence and divergence held not only for political identity defined by party affiliation, but also for identity based on the religiosity of a senator’s constituents, suggests that our findings may generalize to social contexts in which oppositional groups with public identities come into contact with one another and seek to influence each other’s views and behavior. Examples of such contexts include labor-management relations and enforcement of environmental regulations. We see great promise in future research that extends our theory and empirical approach into such settings.

Contributions

This study makes a number of noteworthy contributions. First, our findings help reconcile two seemingly inconsistent expectations about the tendency of increased interaction to promote convergence or divergence of attitudes, beliefs, and behavior. Indeed, Bonacich and Lu (2012:216) list the question of “how groups become polarized or how two groups can become more and more different” as among the most important unsolved problems in sociology. By highlighting the moderating role of identity, this work informs research across a range of sociological subfields on when and how interaction leads to polarization.

Scholars of opinion change in groups have, for example, noted the absence of consistent empirical evidence for negative influence—a mechanism that can help account for group polarization (for a review, see Mäs and Flache 2013). Yet much of the empirical evidence on negative influence comes from laboratory settings, in which subjects lack a shared history, do not strongly dislike outgroup members, and discuss issues that are not especially important to them (Krizan and Baron 2007). The lack of consistent empirical support for negative influence has led mathematical sociologists to propose alternative mechanisms—for example, based on homophily and the content of communication among similar or dissimilar people—that can account for polarization (Mäs and Flache 2013). The present study does not discount these alternative mechanisms. It does, however, provide compelling evidence that negative influence, which has heretofore been theorized and modeled in agent-based simulations and experiments (Mäs, Flache, and Kitts 2014), can be observed in interactions that take place in an important real-world setting where people have shared histories, feel animosity toward outgroup members, and grapple with issues about which they care deeply. Moreover, whereas prior work theorizes about the role of identity in negative influence (Kitts 2000), this study provides empirical evidence of the role of oppositional political identities in social influence.

Similarly, a common theme in institutional theory centers on how interactions among actors in the same field promote convergence in their behaviors. Indeed, fields are often defined as settings where actors “partake of a common meaning system and . . . interact more frequently and fatefully with one another than with actors outside the ﬁeld” (Scott 2001:84, emphasis added). Research on geographic communities, for example, considers how interaction among elites promotes the convergence of corporate behaviors within communities that host multiple corporate headquarters (for a review, see Tilcsik and Marquis 2013). Within such communities, executives from different firms are assumed to interact with one another and then to “look to the actions of other locally headquartered companies for standards of appropriateness” (Marquis, Glynn, and Davis 2007:927). Along the same lines, upper-class social clubs in communities are thought to “provide institutionalized informal settings in which elites are socialized and socially controlled to adhere to normative business attitudes and behaviors” (Kono et al. 1998:868). Our findings suggest the need to rethink and potentially revise these assumptions. The tendency for an executive to emulate the actions of another executive from a different locally headquartered firm or to adopt the same norms and attitudes about business may depend on the extent to which the two executives share a salient public identity and on the local context in which their interactions take place.

Our work also provides novel insight for research on the coevolution of networks and political attitudes and ideology (e.g., Huckfeldt, Johnson, and Sprague 2004; Mutz 2002). For example, in a longitudinal field study, Lazer and colleagues (2010) found that social, rather than task, ties among students were associated with subsequent convergence in their political ideology. They concluded that “persuasion may be more a function of affect than information transfer, and persuasion is unlikely to be a function merely of interaction frequency” (p. 267). Our findings suggest the need to broaden this proposition to account for the roles of public identity and the local context of interaction. We would expect, for example, that social ties would promote convergence of political attitudes when two students share the same political identity, but would either have no effect or a negative effect on political alignment when two students have opposing political identities.

Second, these findings contribute to research on integration and disunion among elites (e.g., Hetherington 2001; McCarty et al. 2006; Mizruchi 1989; Poole and Rosenthal 1997). We provide insight into these dynamics in a setting that has served as perhaps the most important focal point for the study of elite polarization: the U.S. Senate. Whereas the polarization literature tends to focus on macro-structural factors, such as geographic sorting of voters along partisan lines (Theriault 2008), that have contributed to increased polarization in the Senate, our study illuminates a complementary set of explanations: the microfoundations of polarization that arise through subtle, day-to-day interactions among senators with shared or opposing political identities. Although compositional shifts in the Senate—for example, the tendency for people holding more extreme views to enter the chamber and for moderates to exit—likely account for much of the increase in polarization, our work shows how interactions among senators during their periods of joint service in the chamber may have exacerbated this tendency.

Our theoretical arguments also provide an intriguing hint about the prospects for future political polarization. Were the macro-structural forces somehow reversed, such that the Senate comprised people with more moderate views, our arguments suggest that the tendency toward polarization resulting from interaction between senators from the two main parties might also diminish. Recall that we imposed an important scope condition on the tendency for people with opposing political identities to diverge in attitudes and behavior following interaction on a contentious issue: that they also tend to disagree on other issues beyond the focal issue. Thus, if senators from different parties had more common ground, interaction between them need not lead to further polarization.

Finally, the present study informs social psychological research on identity and influence (e.g., Baron et al. 1996; Hogg and Abrams 2003; Hogg and Ridgeway 2003; Maitner et al. 2010). Whereas this literature tends to focus on social influence between individuals in one-off interactions in laboratory settings (for a review and critique of this literature, see Mason, Conrey, and Smith 2007), our investigation shows how the desire for a positive self-concept and the pressure to conform to norms consistent with salient identities (Aronson 1968; Thibodeau and Aronson 1992) operate in real-world settings, where actors are embedded in networks, have multiple public identities, and are engaged in repeated interactions over time (cf. Srivastava and Banaji 2011). Moreover, we bring to research on identity and influence fresh insight about the role of the structural context. Our findings indicate that identity matters for social influence in certain group settings—for example, senate committees with a history of past division—but not in others.

In summary, this article illustrates the value of bringing together insights from sociology, political science, and social psychology in the study of the dynamics of social influence. This cross-disciplinary exchange promises to yield novel insight into how attitudes, beliefs, and behaviors form and evolve and about the interpenetration of identity and social structure.

Footnotes

Acknowledgements

Both authors contributed equally. We thank the ASR editors and reviewers, Jenny Chatman, Andreea Gorbatai, Heather Haveman, Geoff Leonardelli, Ming Leung, Sanaz Mobasseri, Paul Pierson, Jo-Ellen Pozner, Chris Rider, Eliot Sherman, Toby Stuart, András Tilcsik, Noam Yuchtman, Pavel Zhelyazkov, participants of the University of Chicago Organization Theory Workshop for Junior Faculty, the ION Conference, the Political Networks Conference, the Academy of Management Annual Meeting, and the American Sociological Association Annual Meeting, for helpful comments and feedback. The usual disclaimer applies.

Notes

Christopher C. Liu is an Assistant Professor of Strategic Management at the Rotman School of Management, University of Toronto. His research examines spatial networks, how geography shapes the ability of individuals and organizations to enter into preferential relationships, and network positions. In addition to studying political actors, Chris’s current projects include the communication patterns of life scientists, the structure of their laboratories, and the spatial determinants of firm and regional knowledge. Chris holds a DBA in Technology and Operations Management from Harvard Business School and a PhD in Biology from MIT.

Sameer B. Srivastava is an Assistant Professor at the Haas School of Business at University of California, Berkeley. His research examines the dynamics of social networks within organizations and their consequences for individual attainment. His current projects include a study of organizational culture and socialization using a computational linguistics approach; a field experiment to assess the network effects of a program designed to increase the social belonging of women in science and engineering careers; and ongoing investigations of the cognitive underpinnings of social capital activation. Sameer holds a PhD in Sociology and Organizational Behavior from Harvard University.

References

Abrams

Dominic

Hogg

Michael A.

1990. “Social Identification, Self-Categorization, and Social Influence.” European Review of Social Psychology 1(1):195–228.

Agnew

John

Brusa

Carlo

. 1999. “New Rules for National Identity? The Northern League and Political Identity in Contemporary Northern Italy.” National Identities 1(2):117–33.

Almquist

Zack W.

Butts

Carter T.

2013. “Dynamic Network Logistic Regression: A Logistic Choice Analysis of Inter- and Intra-Group Blog Citation Dynamics in the 2004 US Presidential Election.” Political Analysis 21(4):430–48.

Aronson

Elliot

. 1968. “Dissonance Theory: Progress and Problems.” Pp. 5–27 in Cognitive Consistency Theories: A Source Book, edited by Abelson

R. P.

Aronson

McGuire

W. J.

Newcomb

T. M.

Rosenberg

M. J.

Tannenbaum

P. H.

Skokie, IL: Rand McNally.

Azoulay

Pierre

Ding

Waverly

Stuart

Toby E.

2009. “The Impact of Academic Patenting on the Rate, Quality, and Direction of (Public) Research Output.” Journal of Industrial Economics 57(4):637–76.

Baldassarri

Delia

Bearman

Peter

. 2007. “Dynamics of Political Polarization.” American Sociological Review 72(5):784–811.

Baldassarri

Delia

Gelman

Andrew

. 2008. “Partisans without Constraint: Political Polarization and Trends in American Public Opinion.” American Journal of Sociology 114(2):408–446.

Baron

Robert S.

Hoppe

Sieg I.

Kao

Chaun Feng

Brunsman

Bethany

Linneweh

Barbara

Rogers

Diane

. 1996. “Social Corroboration and Opinion Extremity.” Journal of Experimental Social Psychology 32(6):537–60.

Berezin

Mabel

. 2001. “Emotions and Political Identity: Mobilizing Affection for the Polity.” Pp. 83–98 in Passionate Politics: Emotions and Social Movements, edited by Goodwin

Jasper

J. M.

Poletta

Chicago: University of Chicago Press.

10.

Bonacich

Phillip

Philip

. 2012. Introduction to Mathematical Sociology. Princeton, NJ: Princeton University Press.

11.

Bossard

James H. S.

1932. “Residential Propinquity as a Factor in Marriage Selection.” American Journal of Sociology 38(2):219–24.

12.

Brauer

Markus

Judd

Charles M.

Gliner

Melissa D.

1995. “The Effects of Repeated Expressions on Attitude Polarization During Group Discussions.” Journal of Personality and Social Psychology 68(6):1014–1029.

13.

Burke

Peter J.

Stets

Jan E.

2009. Identity Theory. New York: Oxford University Press.

14.

Burnstein

Eugene

. 1982. “Persuasion as Argument Processing.” Pp. 103–124 in Group Decision-Making, edited by Davis

J. H.

Stocher-Kreichgauer

New York: Academic Press.

15.

Burris

Val

. 2005. “Interlocking Directorates and Political Cohesion among Corporate Elites.” American Journal of Sociology 111(1):249–83.

16.

Calhoun

Craig

. 1993. “Nationalism and Ethnicity.” Annual Review of Sociology 19:211–39.

17.

Cameron

A. Colin

Gelbach

Jonah B.

Miller

Douglas L.

2011. “Robust Inference with Multi-way Clustering.” Journal of Business and Economic Statistics 29(2):238–49.

18.

Cartwright

Dorwin

. 1971. “Risk Taking by Individuals and Groups: An Assessment of Research Employing Choice Dilemmas.” Journal of Personality and Social Psychology 20(3):361–78.

19.

Chown

Jillian D.

Liu

Christopher C.

2014. “Geography and Power in an Organizational Forum: Evidence from the U.S. Senate Chamber.” Strategic Management Journal. doi: 10.1002/smj.2209.

20.

Dahlander

Linus

McFarland

Daniel A.

2013. “Ties That Last: Tie Formation and Persistence in Research Collaborations over Time.” Administrative Science Quarterly 58(1):69–110.

21.

De Groot

Morris H.

1974. “Reaching a Consensus.” Journal of the American Statistical Association 69(345):118–21.

22.

Deering

Christopher J.

Smith

Steven S.

1997. Committees in Congress. Washington, DC: Congressional Quarterly Press.

23.

DiMaggio

Paul

Evans

John

Bryson

Bethany

. 1996. “Have Americans’ Social Attitudes Become More Polarized?” American Journal of Sociology 103(3):690–755.

24.

Dreiling

Michael

Darves

Derek

. 2011. “Corporate Unity in American Trade Policy: A Network Analysis of Corporate-Dyad Political Action.” American Journal of Sociology 116(5):1514–63.

25.

Ellemers

Naomi

Haslam

S. Alexander

. 2012. “Social Identity Theory.” Pp. 379–98 in Handbook of Theories of Social Psychology, Vol. 2, edited by Van Lange

P. A. M.

Kruglanski

A. W.

Higgins

E. T.

Thousand Oaks, CA: Sage.

26.

Fenno

Richard F.

Jr.

1973. Congressmen in Committees. Boston: Little, Brown.

27.

Fenno

Richard F.

Jr.

1989. The Making of a Senator: Dan Quayle. Washington, DC: CQ Press.

28.

Festinger

Leon

. 1957. A Theory of Cognitive Dissonance. Stanford, CA: Stanford University Press.

29.

Festinger

Leon

Schachter

Stanley

Back

Kurt

. 1950. “The Spatial Ecology of Group Formation.” Pp. 141–61 in Social Pressure in Informal Groups, edited by Festinger

Schachter

Back

Stanford, CA: Stanford University Press.

30.

Fiorina

Morris P.

Abrams

Samuel J.

2008. “Political Polarization in the American Public.” Annual Review of Political Science 11:563–88.

31.

Fowler

James H.

2006a. “Connecting the Congress: A Study of Cosponsorship Networks.” Political Analysis 14(4):456–87.

32.

Fowler

James H.

2006b. “Legislative Cosponsorship Networks in the US House and Senate.” Social Networks 28(4):454–65.

33.

Fowler

James H.

Kam

Cindy D.

2007. “Beyond the Self: Social Identity, Altruism, and Political Participation.” Journal of Politics 69(3):813–27.

34.

French

John R. P.

Jr.

1956. “A Formal Theory of Social Power.” Psychological Review 65(3):181–94.

35.

Friedkin

Noah E.

1993. “Structural Bases of Interpersonal Influence in Groups: A Longitudinal Case Study.” American Sociological Review 58(6):861–72.

36.

Friedkin

Noah E.

1998. A Structural Theory of Social Influence. New York: Cambridge University Press.

37.

Friedkin

Noah E.

1999. “Choice Shift and Group Polarization.” American Sociological Review 64(6):856–75.

38.

Friedkin

Noah E.

Johnson

Eugene C.

1990. “Social Influence and Opinions.” Journal of Mathematical Sociology 15(3–4):193–206.

39.

Harary

Frank

. 1959. “A Criterion for Unanimity in French’s Theory of Social Power.” Pp. 168–82 in Studies in Social Power, edited by Cartwright

Ann Arbor, MI: Institute for Social Research.

40.

Hernán

Miguel A.

Brumback

Babette

Robins

James M.

2001. “Marginal Structural Models to Estimate the Joint Causal Effect of Nonrandomized Treatments.” Journal of the American Statistical Association 96(454):440–48.

41.

Hetherington

Mark J.

2001. “Resurgent Mass Partisanship: The Role of Elite Polarization.” American Political Science Review 95(3):619–31.

42.

Hogg

Michael A.

Abrams

Dominic

. 2003. “Intergroup Behavior and Social Identity.” Pp. 407–431 in The Sage Handbook of Social Psychology, edited by Hogg

M. A.

Cooper

Thousand Oaks, CA: Sage.

43.

Hogg

Michael A.

Ridgeway

Cecilia L.

2003. “Social Identity: Sociological and Social Psychological Perspectives.” Social Psychology Quarterly 66(2):97–100.

44.

Hogg

Michael A.

Turner

John C.

1985. “Interpersonal Attraction, Social Identification and Psychological Group Formation.” European Journal of Social Psychology 15(1):51–66.

45.

Huckfeldt

Robert

Johnson

Paul E.

Sprague

John

. 2004. Political Disagreement: The Survival of Diverse Opinions within Communication Networks. Cambridge, UK: Cambridge University Press.

46.

Kanazawa

Satoshi

. 2000. “A New Solution to the Collective Action Problem: The Paradox of Voter Turnout.” American Sociological Review 65(3):433–42.

47.

Kitts

James A.

2000. “Mobilizing in Black Boxes: Social Networks and Participation in Social Movement Organizations.” Mobilization: An International Journal 5(2):241–57.

48.

Kitts

James A.

2006. “Social Influence and the Emergence of Norms Amid Ties and Amity and Enmity.” Simulation Modelling Practice and Theory 14(4):407–422.

49.

Kleinbaum

Adam

Stuart

Toby E.

Tushman

Michael L.

2013. “Discretion within Constraint: Homophily and Structure in a Formal Organization.” Organization Science 24(5):1316–36.

50.

Kono

Clifford

Palmer

Donald

Friedland

Roger

Zafonte

Matthew

. 1998. “Lost in Space: The Geography of Corporate Interlocking Directorates.” American Journal of Sociology 103(4):863–911.

51.

Krizan

Zlatan

Baron

Robert S.

2007. “Group Polarization and Choice-Dilemmas: How Important Is Self-Categorization?” European Journal of Social Psychology 37(1):191–201.

52.

Lazer

David

Rubineau

Brian

Chetkovich

Carol

Katz

Nancy

Neblo

Michael

. 2010. “The Coevolution of Networks and Political Attitudes.” Political Communication 27(3):248–74.

53.

Mackie

Diane M.

Cooper

Joel

. 1984. “Attitude Polarization: Effects of Group Membership.” Journal of Personality and Social Psychology 46(3):575–85.

54.

Macy

Michael W.

Kitts

James A.

Flache

Andreas

Benard

Stephen

. 2003. “A Hopﬁeld Model of Emergent Structure.” Pp. 162–73 in Dynamic Social Network Modeling and Analysis: Workshop Summary and Papers, edited by Breiger

Carley

K. M.

Pattison

Washington, DC: National Academies Press.

55.

Maitner

Angela T.

Mackie

Diane M.

Claypool

Heather M.

Crisp

Richard J.

2010. “Identity Salience Moderates Processing of Group-Relevant Information.” Journal of Experimental Social Psychology 46(2):441–44.

56.

Marquis

Christopher

Glynn

Mary Ann

Davis

Gerald F.

2007. “Community Isomorphism and Corporate Social Action.” Academy of Management Review 32(3):925–45.

57.

Marsden

Peter V.

1981. “Introducing Influence Processes into a System of Collective Decisions.” American Journal of Sociology 86(6):1203–1235.

58.

Marsden

Peter V.

Friedkin

Noah E.

1993. “Network Studies of Social Influence.” Sociological Methods and Research 22(1):127–51.

59.

Mäs

Michael

Flache

Andreas

. 2013. “Differentiation without Distancing: Explaining Bi-Polarization of Opinions without Negative Influence.” PLoS ONE 8(11):1–17.

60.

Mäs

Michael

Flache

Andreas

Kitts

James A.

2014. “Cultural Integration and Differentiation in Groups and Organizations.” Pp. 71–90 in Perspectives on Culture and Agent-Based Simulations, Vol. 3, Studies in the Philosophy of Sociality, edited by Dignum

Dignum

Cham, Switzerland: Springer International Publishing.

61.

Mason

Winter A.

Conrey

Frederica R.

Smith

Eliot R.

2007. “Situating Social Influence Processes: Dynamic, Multidirectional Flows of Influence within Social Networks.” Personality and Social Psychology Review 11(3):279–300.

62.

McCall

George J.

Simmons

J. L.

1978. Identities and Interactions: An Examination of Human Associations in Everyday Life. New York: Free Press.

63.

McCarty

Nolan

Poole

Keith T.

Rosenthal

Howard

. 2006. Polarized America: The Dance of Ideology and Unequal Riches. Cambridge, MA: MIT Press.

64.

McClurg

Scott D.

Lazer

David

. 2014. “Political Networks.” Social Networks 36:1–4.

65.

Mizruchi

Mark S.

1989. “Similarity of Political Behavior among Large American Corporations.” American Journal of Sociology 95(2):401–424.

66.

Mizruchi

Mark S.

1992. The Structure of Corporate Political Action: Interfirm Relations and Their Consequences. Cambridge, MA: Harvard University Press.

67.

Mizruchi

Mark S.

Marquis

Christopher

. 2006. “Egocentric, Sociocentric, or Dyadic? Identifying the Appropriate Level of Analysis in the Study of Organizational Networks.” Social Networks 28(3):187–208.

68.

Mutz

Diana C.

2002. “Cross-Cutting Social Networks: Testing Democratic Theory in Practice.” American Political Science Review 96(1):111–26.

69.

Nelson

Garrison

Stewart

Charles

III . 2011. Committees in the US Congress, 1992–2010. Washington, DC: CQ Press.

70.

Nokken

Timothy P.

Poole

Keith T.

2004. “Congressional Party Defection in American History.” Legislative Studies Quarterly 29(4):545–68.

71.

Parigi

Paolo

Bearman

Peter S.

2008. “Spaghetti Politics: Local Electoral Systems and Alliance Structure in Italy, 1984–2001.” Social Forces 87(2):623–49.

72.

Parigi

Paolo

Sartori

Laura

. 2014. “The Political Party as a Network of Cleavages: Disclosing the Inner Structure of Italian Political Parties in the Seventies.” Social Networks 36:54–65.

73.

Poole

Keith T.

Rosenthal

Howard

. 1997. Congress: A Political-Economic History of Roll Call Voting. New York: Oxford University Press.

74.

Ringe

Nils

Victor

Jennifer Nicoll

Gross

Justin H.

2013. “Keeping Your Friends Close and Your Enemies Closer? Information Networks in Legislative Politics.” British Journal of Political Science 43(3):601–628.

75.

Robins

James M.

Hernán

Miguel A.

Brumback

Babette

. 2000. “Marginal Structural Models and Causal Inference in Epidemiology.” Epidemiology 11(5):550–60.

76.

Rubin

Donald B.

2006. Matched Sampling for Causal Effects. New York: Cambridge University Press.

77.

Schienker

Barry R.

Dlugolecki

David W.

Doherty

Kevin

. 1994. “The Impact of Self-Presentations on Self-Appraisals and Behavior: The Power of Public Commitment.” Personality and Social Psychology Bulletin 20(1):20–33.

78.

Schildkraut

Deborah J.

2011. “National Identity in the United States.” Pp. 845–65 in Handbook of Identity Theory and Research, edited by Schwartz

S. J.

Luyckx

Vignoles

V. L.

New York: Springer Science+Business Media.

79.

Scott

Richard W.

2001. Institutions and Organizations. Thousand Oaks, CA: Sage Publications.

80.

Shayo

Moses

. 2009. “A Model of Social Identity with an Application to Political Economy: Nation, Class, and Redistribution.” American Political Science Review 103(2):147–74.

81.

Snijders

Tom A. B.

van de Bunt

Gerhard G.

Steglich

Christian E. G.

2010. “Introduction to Stochastic Actor-Based Models for Network Dynamics.” Social Networks 32(1):44–60.

82.

Somers

Margaret

. 1993. “Citizenship and the Place of the Public Sphere: Law, Community, and Political Culture in the Transition to Democracy.” American Sociological Review 58(5):587–620.

83.

Srivastava

Sameer B.

Banaji

Mahzarin R.

2011. “Culture, Cognition, and Collaborative Networks in Organizations.” American Sociological Review 76(2):207–233.

84.

Steglich

Christian

Snijders

Tom A. B.

West

Patrick

. 2006. “Applying SIENA: An Illustrative Analysis of the Coevolution of Adolescents’ Friendship Networks, Taste in Music, and Alcohol Consumption.” Methodology: European Journal of Research Methods for the Behavioral and Social Sciences 2(1):48–56.

85.

Stone

Jeff

Cooper

Joel

. 2001. “A Self-Standards Model of Cognitive Dissonance.” Journal of Experimental Social Psychology 37(3):228–43.

86.

Stryker

Sheldon

Burke

Peter J.

2000. “The Past, Present, and Future of an Identity Theory.” Social Psychology Quarterly 63(4):284–97.

87.

Tajfel

Turner

J. C.

1979. “An Integrative Theory of Intergroup Conflict.” Pp. 33–53 in The Social Psychology of Intergroup Relations, edited by Austin

W. G.

Worchel

Monterey, CA: Brooks-Cole.

88.

Theriault

Sean M.

2008. Party Polarization in Congress. New York: Cambridge University Press.

89.

Thibodeau

Ruth

Aronson

Elliot

. 1992. “Taking a Closer Look: Reasserting the Role of the Self-Concept in Dissonance Theory.” Personality and Social Psychology Bulletin 18(5):591–602.

90.

Tilcsik

András

Marquis

Christopher

. 2013. “Punctuated Generosity: Events, Communities, and Corporate Philanthropy in the United States, 1980–2006.” Administrative Science Quarterly 58(1):111–48.

91.

Turner

John C.

Reynolds

Katherine J.

2012. “Self-Categorization Theory.” Pp. 399–417 in Handbook of Theories of Social Psychology, Vol. 2, edited by Van Lange

P. A. M.

Kruglanski

A. W.

Higgins

E. T.

Thousand Oaks, CA: Sage.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.01 MB

0.02 MB

0.27 MB