Partisan Joiners: Associational Membership and Political Polarization in the United States (1974–2004)*

Abstract

Objectives

Associational life may foster political integration or amplify division, depending on how individuals partition themselves into groups and whether their multiple affiliations embed them into concentric or cross-cutting social circles. Starting from this premise, I relate trends in associational membership to political partisanship, and ask if there is any evidence of increased political polarization in the associative patterns of Americans.

Methods

Using GSS data (1974–2004) on affiliations to 16 types of groups, I plot trends and run multilevel models to examine changes over time in the partisan allegiances of group members and patterns of overlapping memberships.

Results

The often-lamented decline in group membership affects primarily the category of single-group members and is limited to a few types of groups. The density of the network of overlapping memberships has remained stable over time and there are no real changes in the patterns of shared memberships between group types, nor do Republicans and Democrats differ in their patterns of preferential affiliation. Although political partisanship does not drive patterns of group affiliation, group members, especially those affiliated with multiple groups, are more radical in their partisan identification than nonmembers, and most types of groups have become politically more heterogeneous over time.

Conclusion

The puzzling finding that group types are not becoming more partisan, while group members are, leads to the hypothesis (to be tested in future research) that civil society polarization is occurring at the level of actual groups, and not group types.

The academic debate on the impoverishment of political life, increased social isolation, and the threat of authoritarianism reached its height in the 1950s and 1960s, fueled by the dramatic experiences of World War II. A quite heterogeneous group of scholars, drawing from the European experiences of the previous decades and the changing nature of American community, reached discouraging conclusions about the rise of mass society and its consequences for democracy and social stability (Arendt, 1958; Mann, 1970; Kornhauser, 1959; Linz, 1978; see also Mills, [1956] 1970; Converse, 1964; Milgram, 1967). In contrast, theorists of political pluralism suggested that, in practice, representative democracies do not require large levels of mass participation, and that a multitude of interest groups, not an exclusive inner circle, have access to power. In a more optimistic vein, they argued that cross-cutting interests, intergroup competition, and institutional differentiation secure the openness of the democratic process (Truman, 1951; Dahl, 1961; Lowi, [1969] 1979; Galston, 2002).

Although animated by less dramatic concerns, over the last two decades sociologists and political scientists have produced ample evidence of the fact that many symptoms of the same (alleged) democratic malaise still affect U.S. society: from the polarization of Congress and the end of the era of bipartisan politics (Brady and Han, 2006; Poole and Rosenthal, 2007), to increasing social and political inequality (McCarty, Poole, and Rosenthal, 2006; Bartels, 2008), to the inconsistent and scarce participation of the mass public (Verba, Schlozman, and Brady, 1995; Delli Carpini and Keeter, 1996; Patterson, 2002), to the decline in associational membership (Putnam, 2000; Skocpol, 2003; but see Minkoff, 1997 and Wuthnow, 2002 for an alternative view, and Fischer, 2009 on social isolation). Nonetheless, in recent years, scholars have failed to systematically relate dynamics of political division and polarization to changes in associational life and civic engagement. Moreover, they have usually assumed that a vivid associational life fosters democratic outcomes—taking for granted that bowling together is better than bowling alone—thus failing to consider the actual consequences of the decline in civic participation on the level of political integration and social cohesion.

In this article I examine the interplay between associational membership and political partisanship, and ask if there is any evidence of increased political polarization in the associative patterns of Americans. The goal is to assess to what extent the polarization of party members and political activists is complemented by a similar dynamic in civic life. To address this question, I look at changes over time in the partisanship of associational groups, comparing the partisan allegiances of group members to those of nonmembers, and analyzing the network of interorganizational relations generated by overlapping memberships.

Looking at the intersection of different social spheres from a macro perspective, I conclude that U.S. civil society is not becoming more polarized. The analysis shows that the observed decline in civic engagement has not reduced the overall level of integration of civic networks, and that there are no signs of increased polarization in patterns of overlapping memberships: nor do divergent associational patterns emerge when comparing the subpopulations of Democrats and Republicans. Nonetheless, while political partisanship does not drive patterns of group affiliation, group joiners tend to be more partisan than the overall population: group members, especially those belonging to multiple groups, are more radical in their partisan identification than nonmembers, and most types of groups have become politically more heterogeneous over time. To make sense of these findings, in the discussion, I advance the hypothesis that polarization of civil society is occurring at the level of actual groups, and not group types.

Cross-Cutting Circles, Political Pluralism, and Political Subcultures

In his rejection of society as a holistic concept, George Simmel provided a vivid picture of the divisive tensions that individuals experience in contemporary societies (Simmel, [1908] 1950). For him, society is a constellation of forms of “sociation,” a place where different, often competing, tendencies combine into unresolved, unstable equilibria. Sociations are crystallized forms of interaction (such as the family, guild, church, social class, etc.), as well as less conspicuous, stable forms of relations, “that piece together the real life of society as we encounter it in our experience” (Simmel, [1908] 1950:9). Socialized individuals are in a dual relation with society. They are both part of it and against it; they are both acted upon and self-determined (Frisby, [1984] 2004).

According to Simmel ([1908] 1955), in traditional society, individuals were at the center of concentric circles, in which affiliation to one group was highly predictive of membership into other circles. In contrast, in modern society, individuals are at the intersection of a web of group affiliations in which membership in one group is less predictive of other affiliations. Individuals thus experience cross-cutting circles that exert diverse and often counteracting pressures on them, simultaneously allowing their self-actualization and limiting alternative realizations. The more the lines of social differentiation intersect, the more individuals experience freedom of choice, since the power of the group over the individual is weakened. Simultaneously, this very same process of individualization¹ brings about social isolation, alienation, and a sense of displacement.²

Peter Blau systematized Simmel’s suggestive notion of cross-cutting social circles and subjected it to empirical analysis. Blau was interested in the relationship between differentiation, inequality, and social integration and defined social integration not in terms of culture or shared values, but actual patterns of intergroup relations based on personal associations. His idea was that increasing social heterogeneity and differentiation might actually induce greater social integration in cases in which the multiple lines of social differentiation and sources of inequality are cross-cutting, rather than overlapping. Given that heterogeneity promotes intergroup relations and inequality enhances the probability of status-distant social relations, a seemingly paradoxical conclusion follows: that differentiation nourishes social integration (Blau, 1974; Blau and Schwartz, [1984] 1997).

Blau never explicitly referred to the theme of pluralism (cf. Collins, 1979), but it is easy to envision the close connection between his approach to social inequality and the classic liberal-pluralistic argument that inequalities and social divisions, as long as they unfold along nonoverlapping dimensions, are good for social stability.

The link between concentric versus cross-cutting micro environments and the theory of political pluralism becomes even more intelligible when we consider the macro-level structures that emerge from these two alternative micro experiences. In fact, at the macro level, concentric social circles translate into hierarchical vertical networks, in which individual affiliations are nested in a cumulative fashion. In a context in which people experience cross-cutting pressures, on the other hand, the emergent macro structure is likely to be an integrated one in which individuals span across different social circles. For instance, Lipset and Rokkan (1967) used a similar conceptualization in discussing the consolidation of social cleavages into exclusive political subcultures from which the ideological parties in Western democracies have emerged: exclusive political subcultures are created through the development of vertical networks of associations that ensure maximum loyalty to the party (or church) and protect the supporters from cross-cutting communications and pressures (Diani, 2000).

In sum, the origin and reinforcement of political identities is intertwined with the configuration of the web of group affiliations in which individuals are embedded. Civil society associations are a constitutive part of the political culture and serve as a “transmission belt” (Riley, 2005:290) between parties and citizens, contributing to the integration of local interests into the broader national political context (Lipset and Rokkan, 1967). From this perspective, associational life contributes to the social organization of political consent, a consent that can foster support for democratic and authoritarian regimes alike, depending on how people partition themselves into groups (Lipset and Rokkan, 1967; Gramsci, 1971; Allen, 1984; Koshar, 1986; Riley, 2005; Tilly, 2005). This approach thus differs from understandings of civil society as a sphere separate from, and alternative to, state, family, and market (Hegel, [1820] 1942; Parsons, 1951), and, moreover, challenges the Tocquevillian idea(l) that civic engagement univocally triggers democratic outcomes.

Looking at contemporary debates, substantial scholarship has been produced regarding the increased partisanship and polarization of U.S. politics. Similarly, different views concerning the decline in civic engagement and political participation have been articulated. Differently from the past, however, in recent years these two dynamics have rarely been addressed together.

Political Partisanship

There is virtually full agreement among scholars on the fact that political parties and politicians, in recent decades, have become more ideological on a broad set of political issues (McCarty, Poole, and Rosenthal, 2006; Layman, Carsey, and Menasce Horowitz, 2006). A similar dynamic has been observed among party activists (Stone and Rapoport, 1994; Saunders and Abramowitz, 2004). Both mechanisms of persuasion and mechanisms of selective recruitment/de-recruitment are at work (Saunders and Abramowitz, 2004:287). In addition, the growth, starting in the 1970s, of single-issue-based interest groups and activists has had a radicalizing effect on party primaries and legislative behavior in Congress (Brady and Han, 2006).

Whether similar polarization dynamics characterize public opinion as a whole is still a subject of discussion. Although there is scarce evidence of public opinion polarization in the population as a whole, or across sociodemographic subgroups (DiMaggio, Evans, and Bryson, 1996; Evans, 2003; Fiorina, Abrams, and Pope, 2005; Fiorina and Abrams, 2008), there is a consistent split across party lines: Republicans and Democrats have become increasingly more divided in their political preferences (Abramowitz and Saunders, 1998). The alignment of ideology and partisanship is often interpreted as a consequence of elite polarization—since parties are more polarized, they are now better at sorting individuals along ideological lines (Heterington, 2001; Levendusky, 2009; Baldassarri and Gelman, 2008)— but there is also evidence for issue-based change in party identification (Carsey and Layman, 2006). Importantly, after a period of decline in the importance of party identification and ideology, partisan loyalties have started to count more: in the mid-1990s, their impact on voting behavior reached its highest level in at least 50 years (Abramowitz and Saunders, 1998; Bartels, 2000; Hetherington, 2001).

This new era of partisan voting (Bafumi and Shapiro, 2009) might have the effect of handing over greater voice to political extremists and single-issue advocates, and amplifying dynamics of unequal representation (McCarty, Poole, and Rosenthal, 2006; Bartels, 2008; Baldassarri and Gelman, 2008; Levendusky, 2009). Such effects would be magnified if dynamics of increased partisanship and polarization were to occur along the lines of civic associations and organized groups. Thus it is crucial to turn our attention to patterns of associational affiliation and assess to what extent these patterns have changed over time.

Associational Life, Democracy, and Social Capital

The recent debate on changes in U.S. associational life took off with Putnam’s alarmed assessment of the decline in civic engagement and political participation (2000) and was followed by Skocpol’s investigation of the changing role of voluntary groups and associations (2003). Other scholars reached less dramatic conclusions, showing the overall stability of institutional forms of trust and social capital (Paxton, 1999; Wuthnow, 2002), the rise of new participatory forms (Minkoff, 1997; Wuthnow, 2002), and the continuity of collective action at the community level (Sampson et al., 2005).

Both those who are and are not concerned about current trends in civic engagement tend to take for granted that a vibrant associational life brings positive outcomes for the collectivity, despite admitting that associational membership is not in itself a guarantee of democratic outcomes. Historical research (Allen, 1984; Koshar, 1986; Berman, 1997; Riley, 2005) has similarly dismissed simplistic models that equate associational life to democratic outcomes. While focusing on single associations may be sufficient to study their integrative and educative effect on individuals (Verba, Schlozman, and Brady, 1995), in order to understand dynamics of interest representation and democratic deliberation it is necessary to take into account the interplay between associations and the overall web of interorganizational relations they generate. For example, Paxton has moved in this direction in her comparative research on the impact of associational memberships on social capital and generalized trust by distinguishing between associations that are connected to other associations through their members’ multiple membership and isolated associations (Paxton, 2002, 2007). Paxton shows that individual membership in connected associations is more likely to foster generalized trust than membership in isolated associations (2007) and that, while connected associations are likely to foster democratic outcomes, isolated associations can have a detrimental effect on democracy (2002). In studying the relationship between increased political partisanship and the decline in traditional forms of associational participation, this article will focus not only on the changing profile of the associations, but also on changes in the web of group affiliations they generate. The nature and quality of democratic governance depends on whether U.S. citizens are embedded in cross-cutting or concentric patterns of group membership.

Methodology and Data

There are two main strategies for studying associational networks. The first is to collect data on actual groups and map the web of shared memberships and interorganizational activities between them (Diani, 2003; Baldassarri and Diani, 2007). The second is to consider types of groups and look at patterns of overlapping memberships between them (McPherson, 1983; McPherson and Ranger-Moore, 1991; Cornwell and Harrison, 2004; Diani, 2009). The former captures the concrete web of organizational affiliations and the richness of the social and political context in which individuals and groups are embedded but, due to constraints of data collection, it is necessarily confined to the in-depth study of local communities, thus making it difficult to transcend the specificity of the context and claim generalizability for the findings. In contrast, mapping affiliations to group types makes it possible to use standard sampling techniques, thus ensuring the generalizability of the findings, although at the cost of substituting types of voluntary activities for actual groups (i.e., fraternal, religious, sports, etc.). In this article I adopt the latter approach, which seems more appropriate to answer basic questions concerning the political divides at the national level. Group types reflect cultural affinities and practices that are shared across the entire country, and are not subjected to the idiosyncrasies that one would find when looking at local communities.

Data on group types have been used by McPherson to develop an ecological model of group competition and adaptation that represents, to date, the most powerful attempt to model the growth, decline, and demographic dynamics of a population of voluntary groups (McPherson, 1983; Mc-Pherson and Ranger-Moore, 1991). Complementary to McPherson’s ecological model is a structural analysis of the web of overlapping memberships between groups that casts light on the level of interpenetration and potential cooperation between associative types. Cornwell and Harrison (2004) have moved in this direction by looking at the structural embeddedness of labor unions in the broader context of civil society organizations. I continue in this tradition and investigate the changes in the emergent interorganizational structure, with a specific focus on patterns of political polarization. I use the General Social Survey’s consistent information about associational participation in 16 different types of organizations over a time period of 30 years, from 1974–1994 and in 2004. The survey question is asked as follows.

I would like to know something about the groups and organizations to which individuals belong. Here is a list of various kinds of organizations. Could you tell me whether or not you are a member of each type? List of groups: Fraternal groups; Service groups; Veterans’ groups; Political clubs; Labor Unions; Sports groups; Youth groups; School service groups; Hobby or garden groups; School fraternities or sororities; Nationality groups; Farm organizations; Literary, art, discussion or study groups; Professional or academic societies; Church-affiliated groups; any other group.

This data set allows for the study of a complete typology of group affiliations over a long period of time, providing detailed individual-level information as well as an exhaustive categorization of group memberships. This aspect is central to research agendas that conceive of the web of group affiliation as emerging from the duality of persons and groups (Breiger, 1974). The most problematic shortcoming of the GSS data set, and of the research on associational participation in general, is that the list of associational types has not changed over time: it thus does not include newer types of associations, which are usually more appealing to younger generations, nor does it consider participatory forms that rely on new technologies like the Internet. Future data collections should seriously consider this issue.

I present the analysis in three parts. First, I summarize trends in associational membership, distinguishing between single membership and multiple memberships. Second, I ask whether there is increased polarization within group types by analyzing trends in the level of group partisanship and political homogeneity. The relationship between group membership and strength of political partisanship is further investigated through a multilevel model. Third, I study whether there is evidence of increasing polarization between group types by focusing on the web of interorganizational relations: I first describe trends of overlapping memberships between types of groups in the entire population and the subpopulations of Republicans and Democrats, and then model patterns of preferential relation between pairs of groups to assess the importance of political similarity in group affiliation.

The Network of Overlapping Memberships

Respondents’ multiple associational memberships will be used to investigate patterns of preferential attachment between groups—that is, whether certain types of associations are likely to share members but certain others tend to be mutually exclusive. The emergent network of overlapping memberships can have different configurations: on one end of the spectrum, one can imagine a society characterized by strong sectarian fragmentation— concentric membership—with strong cliques of interconnected groups and without bridges to other types of groups; on the other end, one can think of a society in which associational patterns are individualized—overlapping membership—and the resulting interorganizational network does not have any tendency toward preferential attachment (Baldassarri and Diani, 2007).

Starting from a bipartite (member-by-group) matrix that connects members to the groups they belong to, I built a group-by-group matrix based on the number of members each pair of groups has in common. The number of shared members is, first, a function of the overall size of the groups. Second, and more interesting to us, it is also the product of patterns of preferential affiliation. Unfortunately, it is very difficult to tell preference and group-size effects apart, and generally to derive a network structure from affiliation data. One strategy would be to draw a tie between two group types if the number of shared members meets a certain threshold: of course, the problem then becomes how to determine the threshold. Alternatively, one can compute a measure of preferential relation between group types.³

Since one of the primary interests of this analysis is to study variation in affiliation patterns, I opted for a directional measure of preferential relation between groups: the proportion of shared members that each group shares with any other group. Formally, the proportion of shared members that group A shares with group B (Prop_AB) is defined as the number of members that group A and group B have in common, divided by the number of members that group A shares with any other groups:

$${\mathit{Prop}}_{AB}=\frac{\mathit{num}.\mathit{shared}.{\mathit{memb}}_{AB}}{\mathit{num}.\mathit{tot}.{\mathit{memb}}_A-\mathit{num}.\mathit{single}.{\mathit{memb}}_A}.$$ (1)

This measure of overlapping membership is directional, so that Prop_BA is different from Prop_AB:

$${\mathit{Prop}}_{BA}=\frac{\mathit{num}.\mathit{shared}.{\mathit{memb}}_{AB}}{\mathit{num}.\mathit{tot}.{\mathit{memb}}_B-\mathit{num}.\mathit{single}.{\mathit{memb}}_B}.$$ (2)

A few specifications here: I opted for a directional measure and used proportions instead of raw numbers in order to control for group size: consider two groups, one with 100 members, the second with 10. They share five members. This means that the first shares 5 percent of its members with the second, while the second shares 50 percent of its members with the first. Nondirectional measures do not capture this difference. Second, I used the proportion of shared members instead of the proportion of total members because groups differ in their ratio of single and multiple members: there are some types of groups (like religious and unions) that have a large number of single members, while others have only a few.

Substantively, the proportion of shared members that a group receives from other groups depends on the total size of the group itself and is indicative of its capacity to attract members from other groups; it can thus be viewed as a measure of group appeal.

Analysis I: Trends in Membership

Scholars have often based their concerns about the decline in civic engagement on a downward trend in participation in voluntary groups and associations. At a first glance, survey data confirm this picture. Panel A in Figure 1 plots the proportion of GSS respondents who report being a member of at least one association: in 1974, they were 75 percent of the population, but by 2004, this number went down to 62 percent—a consistent, if not dramatic, drop.

FIGURE 1.

Time Trends in Associational Membership

Notes: x-axis: time (1974–2004). (A) Trend in the proportion of respondents that belong at least to one group. (B) Trend in the average membership per capita. (C) Trend in the number of memberships. (D) Trend in group membership by group. LOWESS curve fitted.

A closer analysis, however, suggests a more nuanced story. Although the proportion of people who do not belong to any association has increased, the average number of group memberships per capita does not show any clear trend, fluctuating between 1.6 to 2 over the period 1974–2004 (Figure 1, Panel B). The explanation for this finding is simple, as shown in Panel C: the decline in membership has occurred mostly among those who belong to a single group, while for those who are affiliated with multiple types of groups, there is no real decline, and there is even some instance of growth in the mid-1990s. Therefore, while there are fewer joiners, those who actually join are more likely to be affiliated with more than one group. In terms of the total number of group affiliations, this phenomenon has an interesting property: a 1 percent increase in the group of those who belong to five groups makes up for a 5 percent decrease in single-group members. Moreover, since the decline in membership has occurred among single-group members, this decline has virtually no impact on the web of interorganizational relations, since people who belong to only one group do not span across different types of associations.

Finally, the decline in membership did not occur evenly across group types. In fact, most types do not show drastic variations in the proportion of members. The exceptions are a steep decline for religious groups and unions, a lesser decline for fraternal groups, and a growth in professional membership: Panel D of Figure 1 reports the trends for these groups and time trends (LOWESS curves). Gray lines report the trend for all the other group types. Although the trend for the majority of groups is flat, religious decline is quite substantial, dropping from 42 percent to 31 percent in the last three decades. Similarly, unions have gone from 16 percent of the population in 1974, to 10 percent in 2004. In contrast, professional groups have grown from 13 percent in 1974 to 19 percent in 1994.

In conclusion, the decline in group membership is not a general pattern; rather, it is concentrated in the category of single-group members and occurs almost exclusively among specific types of association, namely, religious associations, unions, and fraternal groups. It should be noted that unions and religious associations show considerably higher levels of single-group membership compared to any other group. On average, 27 percent of union members and 26 percent of religious group members belong only to one group, but no other group has more than 15 percent single-group members, and the average across all groups is 11 percent.

Analysis II: Group Profile

I study trends in group partisanship by looking at the mean party identification of group members, and trends in political homogeneity by looking at the within-group variation in party identification. The analysis of partisanship will reveal whether group types are (and have become) disproportionally more Republican or, alternatively, Democratic than the rest of the population. The analysis of political homogeneity will reveal whether group types are (and have become) internally more united/divided with respect to the partisan allegiances of their members.

Both aspects are relevant to the study of the level of civil society polarization. If a group type has a clear partisan allegiance, its members and leaders will be more likely to openly support a party and certain political issues. Similarly, political partisanship is a more prominent aspect of group identity in politically homogeneous groups. In contrast, politically heterogeneous groups—groups in which the average difference in party identity between two randomly chosen members of the group is actually higher than the average difference between two randomly chosen members of the population— will have to downplay partisanship in order to maintain internal cohesion.

If group types show strong partisanship—for example, religious members are disproportionally Republican, or union members disproportionally Democratic—the intersection of associative patterns and political partisanship will resemble the image of concentric social circles, in which individual affiliations consistently cumulate into a strong identity. In contrast, if group types are politically heterogeneous, this would support the alternative image of cross-cutting social circles, in which individuals are situated at the intersection of different, even conflicting, affiliations, and accommodate their (weak) identities accordingly. Nonetheless, no conclusion can be drawn looking exclusively at group profiles, since this concept also concerns patterns of multiple affiliations.

The partisanship and political homogeneity of each group type is measured by looking at the mean and variance of its members’ party identifications, respectively. Party identification is measured on a seven-point scale, from strong Democrat (0) to strong Republican (6). The graphs in the left column of Figure 2 report trends in partisanship (average party identification), those in the right column report political homogeneity (variance in party identification). Each graph shows the time trend for group members (circles), nonmembers (asterisks), and all respondents (gray line). The first graph shows the trend for any possible group affiliation, while the graphs in the second through last rows show membership in specific groups: party, union, professional, religious, and sports associations. (These groups were selected because they cover a large variety of substantive interests and are large enough to reduce the impact of sample variability.)

FIGURE 2.

Trends in Group Partisanship (Left) and Political Homogeneity (Right) for Members (Black Circles), Nonmembers (Dark Gray Asterisks), and the Entire Population (Gray Line)

Note: LOWESS curve fitted.

These graphs tell a simple, but interesting, story: looking at trends in average party identification, we find little difference in the partisanship of members and nonmembers, and no evidence of growing partisanship in the last 30 years (with the only exception of union members). For instance, at any point in time, the average difference in party identification between members and nonmembers of religious groups is always less than 0.3 (on a seven-point scale), the average distance between members and nonmembers of professional associations is 0.4, and so on. In contrast, looking at trends in variance, we find that some group types are politically more heterogeneous than the population as a whole. The variance in party identification among group members is larger than among nonmembers. Moreover, the trend over time shows that group joiners are increasingly becoming more divided in their partisan affiliations, especially in the case of religious groups, professionals (starting in the 1990s), and, as previous research has already documented, among party members (note that the scale for this latter group is different). Union members are again an exception: they are politically more homogeneous than the population as a whole.

Group members tend to have stronger partisan identification than nonmembers. Yet, the observed relationship between associational membership and strength of political partisanship might be spurious and due to the greater social centrality of group members. A citizenry that is disproportionally white, affluent, male, well educated, and politically more committed might have both high levels of associational membership and strong partisanship. Moreover, in a diachronic perspective, socially marginal individuals might have left groups at a faster pace. If this is the case, the sociodemographic characteristics of group members should account for the strength of political partisanship and, once we control for sociodemographic characteristics, the relationship between associational membership and strong partisanship should disappear. To test this possibility, I regressed the strength of political partisanship (computed as a four-point scale, where a value of 0 is assigned to those respondents who took a central position on the Democratic-Republican seven-point scale, and 3 to the self-identified strong Democrats/Republicans) on a set of individual-level predictors: occupational prestige, years of education, age (four categories), gender, and race (white vs. nonwhite). The variable of interest is associational membership, measured in three categories (no membership, single membership, and multiple membership). A time variable and its interaction with associational membership are also included. To control for the fact that the relationship between associational membership and partisanship might be due to membership in specific group types (i.e., unions, parties, etc.), I added binary predictors (member/nonmember) for each of the 16 groups.

Results in Table 1 show that even taking into account background characteristics and group types, associational membership is significantly related to strong partisanship, and this association has increased in the last two decades. Moreover, the multiple-membership coefficient is substantially higher than the single-membership one: not only do group members have stronger partisanship than nonmembers, but individuals who belong to more than one group are also more partisan than people who belong to only one group. Single-group members are 0.11 points more extreme than people with no affiliations, while individuals who are involved in multiple groups are significantly more extreme than both nonmembers and single members (0.18 points). The time trend shows that the level of partisanship among nonmembers is decreasing (−0.04 per decade), while it is increasing among single-membership (0.06) and multiple-membership individuals (0.10). With respect to sociodemographic characteristics, people in their mid-40s and older, nonwhites, and, to a lesser extent, women tend to be more partisan. As expected, affiliation to certain groups is greatly conducive to political partisanship: members of political groups are half a point more extreme than nonmembers, followed by members of Greek organizations (0.12) and religious groups (0.08). We conclude that the relationship between associational membership and partisanship is not simply the byproduct of people’s sociodemographic profile. Yet, without proper data, we cannot argue concerning the causal directionality of the relationship. Likely, both processes of self-selection and interpersonal influence are at work, as in the case of Congress and political activists.

TABLE 1.

Coefficients from OLS Model

Variable Type	Variable Name	Coefficient	SE	t Value
	Intercept	1.281	0.046	27.784
Variables of Interest	Single membership	0.110	0.023	4.873
	Multiple membership	0.175	0.030	5.897
	Time	−0.040	0.016	−2.550
	Single membership * Time	0.055	0.024	2.347
	Multiple membership * Time	0.104	0.021	5.064
Sociodemographic Variables	Prestige	−0.013	0.006	−2.081
	Education	0.007	0.003	2.379
	Age 30–40	0.114	0.019	5.832
	Age 45–64	0.345	0.021	16.753
	Age 65 up	0.502	0.025	20.481
	Female	0.050	0.015	3.380
	Nonwhite	0.245	0.020	12.284
Group Types	Political	0.500	0.036	13.881
	Religious	0.079	0.020	4.279
	Professional	−0.044	0.024	−1.859
	Union	0.036	0.022	1.647
	Service	0.001	0.025	0.044
	Veterans	−0.020	0.029	−0.689
	Youth	−0.008	0.026	−0.305
	School	0.020	0.023	0.879
	Sport	−0.002	0.020	−0.106
	Farm	−0.016	0.037	−0.428
	Fraternal	−0.010	0.025	−0.394
	Greek	0.122	0.035	3.500
	Hobby	−0.054	0.025	−2.149
	Literary	−0.063	0.027	−2.381
	Nationality	−0.057	0.040	−1.411
	Other	0.007	0.024	0.318

Notes: Dependent variable: political extremism (0–3); 18,181 observations. R²=0.066, adjusted R²=0.065, F statistic: 45.77 on 28 and 18,152 df. The time variable is expressed in decades and centered in 1988 so that the intercepts and slopes can be more directly interpreted. See text for further description.

In sum, this section shows that there is no evidence of increased group partisanship. In contrast, there is strong evidence of increased political polarization within group types: on average, two members of the same group type are further apart in their political allegiances than two nonmembers. Further analyses have shown that this is actually more true for certain group types, like parties, but that even controlling for group type, there is a substantial relationship between political partisanship and associational membership. Both single-group membership and multiple-group membership are important predictors of political partisanship; nonetheless, members of more than one group are significantly more partisan than both nonmembers and single-group members. This relationship has also grown stronger over time. I conclude that group types are not becoming more partisan, while group joiners are becoming more so. I now consider whether this increased partisanship translates into greater polarization at the level of intergroup relationships through the analysis of the web of group affiliation.

Analysis III: Overlapping Memberships

In this section I model the tendency of each group to share members with other groups in order to describe trends in the patterns of overlapping membership of U.S. citizens. I consider the proportion of shared members that each group receives from other groups (cf. Equations (1) and (2)) as an indicator of group centrality in the web of civil society interorganizational relations. Time trends will reveal whether the interorganizational share of members has increased, remained stable, or decreased over time. I first consider the aggregate trend for each group, then I model the trend between group pairs.

The top graphs in Figure 3 report the trend over time of the average proportion of shared members that each group receives from other groups. In general, the plots show that time variations are rare, and are confined to a decline for religious, union, and fraternal groups, and a growth for professional groups. This trend closely resembles the trend observed in the overall proportion of members for each group (Figure 1, Panel D): as one might expect, group types that are losing members also receive a smaller share of members from other groups.

FIGURE 3.

Trend of the Average Proportion of Shared Members by Group

Notes: The top graph shows the trend of the average proportion of shared members that a group receives from other groups. The bottom graph shows the same measure as computed in the subpopulation of Democrats (gray) and Republicans (black). Because of the lower number of observations, I aggregate results for two consecutive years. As a consequence, for each subgroup, there are only eight data points. The plot for the Democrats is shifted to the right to better visualize the trend. x-axis: proportion of shared members received (25 to 70 for church, 5 to 50 for all other groups). y-axis: years (1974–2004).

One might wonder if this aggregate trend hides systematic differences between the associational patterns of Republicans and Democrats. For instance, Republicans might be connected to parties through religious, professional, or fraternal groups, while Democrats might be more likely to combine activism in union, school service, or literary groups with party affiliation. Results from further analysis reject this possibility, demonstrating that the subpopulations of Republicans and Democrats generate very similar webs of overlapping membership. The bottom graphs in Figure 4 show the proportion of received members for each group type, distinguishing between the civic networks generated by the associational patterns of Republicans (black) and Democrats (gray). In general, there are no real differences between the two subnetworks with respect to the capacity of groups to attract shared members; the exceptions are unions, which seem to play an important role of interconnection among Democrats and, to a lesser extent, school associations and, only recently, literary groups. In contrast, religious, professional, youth, political, and all other groups attract virtually the same average proportion of other groups’ members in both Republican and Democratic civic networks.

FIGURE 4.

Time Trend Estimates of the Proportion of Shared Members for Each Pair of Groups

Notes: The group-by-group matrix shows the estimates from the multilevel varying- intercept and slope model for the time change in the proportion of shared members for each pair of groups. Reading by the row, the trend in the proportion of shared members received by each group; by the column, the trend of the proportion of shared members given to other groups. Different shades of gray represent the intensity of the change: light gray for values that are > 0.5 or < −0.5, gray if >1 or < −1, dark gray if >2 or < −2, and black if >3 or < −3. The arrow shows the directionality of the change; thin arrows for estimates larger than one standard error; thick arrows for estimates larger than two standard errors.

Finally, I consider patterns of preferential relation between pairs of groups in order to assess, first, whether political similarity increases the proportion of shared members between groups and, second, whether changes in overlapping membership have produced more polarized civic networks. To this end, I model the proportion of shared members for each possible pair of groups over time.⁴ I run a multilevel model with varying intercept and slope, in which the unit of analysis is the proportion of shared members between each pair of groups for each year, and the second-level units are group pairs. This model produces estimates of the average proportion of members shared (intercept) and the time trend of this proportion (slope) for each of the 240 group pairs. The model includes a set of predictors that capture the extent to which groups differ in their sociodemographic and political composition. Namely, for each pair of groups I computed the difference between the two groups (in absolute values) in the average prestige level, years of education, age, and party identification, and the difference in the proportion of female and nonwhite members. The expectation is that the greater the difference between two groups on any of these dimensions, the smaller the ratio of shared members between the groups. Formally:

$${\mathit{Prop}}_{t,\mathit{pair}}={\alpha }_{\mathit{pair}}+{\beta }^0{X}_t^0+{\beta }_{\mathit{pair}\ast }t+{\epsilon }_{t,\mathit{pair}},$$ (3)

where Prop_t,pair is the proportion of members shared between any pair of groups at time t, β⁰ are estimated coefficients for the individual-level variables $X_t^0$ (differences in prestige, education, age, etc.), and α_pair and β_pair are, respectively, estimates of the average proportion of shared members (intercept) and time variation (expressed in decades) in the proportion of shared members (slope) for each pair of groups. Table 2 reports results from the model.

TABLE 2.

Fixed and Random Effects Estimates from the Varying-Intercept Varying-Slope Multilevel Model

Variable Type	Variable Name	Coefficient	SE	t Value
	Intercept	27.320	0.928	29.43
	Time	−0.645	0.143	−4.52
Sociodemographic Variables	Δ Prestige	−0.210	0.050	−4.19
	Δ Education	−0.907	0.230	−3.94
	Δ Age	−0.245	0.037	−6.65
	Δ Female	−9.199	1.291	−7.13
	Δ Nonwhite	−0.730	1.864	−0.39
Var of Interest	Δ Party ID	0.067	0.323	0.21
Random Effects	Name	SD	Correlation
	Group-pairs (intercept)	13.049
	Time (trend)	1.525	−0.018
	Residual	4.569

Notes: 16 years, 240 group pairs, 3,840 observations. AIC 23,933, loglik −11,954, deviance 23,903. See text for a description.

In line with previous scholarship (e.g., McPherson, 1983; McPherson and Ranger-Moore, 1991), I find that sociodemographic similarities between groups are related to a greater ratio of shared members. Coefficients for prestige, education, age, and gender are all negative and significant, suggesting that differences between groups along these dimensions consistently reduce their propensity to share members. Interestingly, this does not apply to differences in partisanship: whether a group is disproportionally Republican (or Democratic) is irrelevant in determining with which groups it shares members. In other words, the analysis shows that the political creed has no importance in defining the patterns of multiple affiliation of U.S. citizens.

Finally, results from this model allow us to assess whether changes in overlapping membership have produced more polarized civic networks—a situation in which group types tend to exchange members with a certain subset of group types, and are less likely to exchange members with another subset of group types. Alternatively, we might find that group types are becoming more (or less) embedded into a broader, cohesive civic network over time. If they are, they would receive a greater (or lower) proportion of other groups’ members from most group types. From Table 2 we observe that there is ample variation in the average proportion of shared members, with about two-thirds of the pairs falling in the interval between 15 percent and 41 percent (intercept ± 1 SD). More interestingly, the time trend is negative: the proportion of shared members decreased by 0.7 percentage points per decade. There is nonetheless some variation between group pairs, since about two-thirds of them are in the interval from −2.2 to +0.8.

To assess whether changes have induced greater division, or have occurred evenly across the board, I examine the results at a greater level of detail. Figure 4 plots the time trend estimates for the 240 group pairs. Different shades of gray represent the intensity of the change. The arrow shows the direction of the change: thin arrows for estimates larger than one standard error, thick arrows for estimates larger than two standard errors. Reading by row, cells report the trend in the proportion of shared members received by each group; by column, the trend of the proportion of shared members that each group shares with other groups.

With the single exception of professional groups, which have increasingly attracted members from other groups, most of the groups are either stable or show a decline in the proportion of shared members. Moreover, whatever the trend is, it is generalized. Religious, union, and fraternal groups, and youth associations to some extent, are experiencing a decline in popularity that is consistent across the board. Similarly, professional groups are gaining members from most other groups. This observation, along with the fact that the trends observed here essentially reproduce general trends in the popularity (size) of the groups, leads to the conclusion that there have been no real changes in the patterns of preferential association of U.S. citizens, and thus in the civic networks they generate.

Conclusions

An integrated society is not a society in which conflict is absent; rather, it is one in which conflict expresses itself through nonencompassing interests and identities. Associational life may either foster political integration or amplify division, depending on how individuals partition themselves into groups and the web of overlapping memberships that they generate. The cross-cutting, bridging function of civil society associations can be either boosted or limited by the structure of interorganizational relations in which they are embedded. Starting from this premise, I looked at the relationship between trends in associational membership and political partisanship of U.S. citizens over the last four decades. The main goal was to assess whether civil society associational patterns show evidence of increased political polarization—namely, whether groups have become more partisan and whether the web of interorganizational relations has become more polarized.

The news is that not very much has changed in the associational patterns of U.S. citizens over the last 40 years. The study confirms the decline in group membership denounced by other scholars, but with two important caveats: this decline involves almost exclusively a few specific groups— religious, unions, and fraternal groups—and affects only single-group members. This latter aspect is very important: the fact that the volume of multiple membership has not declined guarantees that the overall relational density of civic networks has remained stable. In other words, there are as many group members connecting different types of groups today as there were 40 years ago. Second, an extensive analysis of the evolution of overlapping memberships over time has revealed that there are no real changes in the patterns of association of U.S. citizens, nor do Republicans and Democrats differ in their patterns of preferential affiliation.

Has there been “much ado about nothing,” then? Maybe not. The analysis of group profiles based on the partisan allegiances of their members has shown that while group types are not becoming more partisan, group joiners are becoming more so. Specifically, group members, especially those with multiple memberships, have grown consistently more extreme in their political identities, even as group types are becoming more ideologically heterogeneous.

The most plausible explanation of this phenomenon—although, sadly, one that available data do not allow us to consider—is that partisan alignment is occurring at the level of groups, and not group types. To use the categories of Lipset and Rokkan (1967), it might be the case that vertical networks are becoming more insulated, and therefore better at protecting “the supporters from cross cutting communications and pressures” (Lipset and Rokkan, 1967:15–16). Greater levels of vertical network insulation can in fact explain the increased partisanship of group members.

In a [loosely insulated] system there is low membership crystallization; most of the participants tend to be tied to organizations and environments exposing them to divergent political pressures. By contrast in a highly [insulated] system there is high membership crystallization; most of the participants tend to be exposed to messages and persuasive efforts in the same general direction in all their “24-hour-7-day” environments. (Lipset and Rokkan, 1967:17)

This picture is consistent with our findings. While, in the aggregate, group types show little variation with respect to their partisan orientation, parallel dynamics of insulation along party lines are likely to induce increasing partisanship among group members at the individual level, due to the mechanisms of self-selection and reinforcement typical of insulated systems.

This can simultaneously explain why group members are more likely to hold extreme political views in a context in which group membership is declining. According to Simmel, individuals in contemporary societies are exposed to the risk of being “pulled apart” by their different, and unique, memberships, allegiances, or identities. The very same trend toward greater individual freedom increases the possibility of social isolation: those who are at the intersection of conflicting social circles might be induced to withdraw from social life. Indeed, empirical evidence suggests that those individuals whose interpersonal networks involve greater heterogeneity or political disagreement have lower levels of political participation (Lipset, 1963; Knoke, 1990; Mutz, 2002).

The working hypothesis for future research is the following: individuals who are politically more moderate because of their intersecting, and sometimes even conflicting, interests and allegiances will tend to withdraw from civic engagement at a faster pace than subjects embedded in more coherent associational experiences. Two mechanisms are expected to drive the observed growth in the level of political extremism of group members: a mechanism of self-selection, according to which politically moderate individuals are more likely to opt out than their more extreme counterparts, and a mechanism of reinforcement, in which the group experience itself is likely to nurture political extremism, since members’ ideological commitment is strengthened by the position of the (remaining) members.