―Polarized Churches: The Political Sorting of American Religious Behavior‖

Jason Husser PhD Candidate Vanderbilt University Abstract: The heightened political information associated with American evangelicalism leads to politics becoming a key factor for organizational satisfaction with churches. In turn, people increase or decrease in frequency of attendance dependent on their agreement with the political position of their church; evangelical Republicans rise in church attendance while comparable Democrats decline. This leads to politically polarized churches. I test the causal relationship using the 20002004 NES Panel Survey to conclude that partisan identification has a strong influence on church attendance of evangelical Protestants but church attendance does not influence partisan identification. I show that religious behavior is more susceptible to political change when a person holds religion less centrally or politics more centrally. I then find religious belief does not drive the religious behavior gap between Democrats and Republicans. I extend the analysis to the early 1970s and find that partisan influences on attendance are a contemporary phenomenon.

0

In 1986, about eight percent of white Southern Baptists considered themselves to be strong Republicans. By 2004, strong Republicans had tripled to 25 percent, becoming largest grouping. Meanwhile, the proportion of strong Democrats attending Southern Baptist churches fell by almost half (GSS). What explains the rising political homogeneity in some churches? A common response is that churches have persuaded their members to become more extreme. This answer is consistent with a long history of social science research treating the relationship between religion and politics as unidirectional. Foundational works only analyzed religion for its effects on political attitudes and voting. However, the relationship between religion and politics has grown more complex for some people because of two trends. First, religion in the United States is increasingly fluid. Almost half of Americans change their religious affiliation in their lifetimes (Pew 2009), and nearly four out of five vary their church attendance from childhood to adulthood (GSS 2007). People make choices about not only the role of faith in their lives, but also which religious organization they accept and how intensely they are involved with that organization. Second, some churches increasingly take controversial positions on political issues. And, the rising polarization of politics means people are likely to have strong feelings about these positions. The result for some people is that politics has become a major factor of satisfaction and integration with churches as organizations. These trends weaken the commonly held assumption that religion affects politics but politics does not affect religion. I offer an alternative explanation about why religious affiliations have become more politically homogenous. People whose political preferences clash with the dominant political direction of a church become less involved with that church. Conversely, participation increases among those whose politics resonate with the dominant political positions

1

espoused in and about church. Specifically, I explore whether the intensifying divisiveness of moral and social issues has led to religion being more than just a cause of vote choice and policy opinion. The following findings suggest that partisan orientation now affects religious behavior. For some people, religious allegiances certainly dominate political preferences. However, the overall relationship between religion and politics is more complicated. I argue that a substantial number of Americans choose to attend or sort into churches reflecting their political values, attitudes and worldviews embodied by partisanship. This pattern likely results in an intensification of polarization within the United States. Using panel survey data, I find results consistent with this theory that show strong effects of party on religious behavior but not significant effect of religious behavior on party.

II. How Politics Can Influence Religious Behavior Individuals receive political information about a church on several levels. Most obviously, pastors discuss politically relevant topics during church services. Church leaders and church organizations can communicate politics directly, indirectly, during services, outside of church in the community at large, and even unintentionally. Informal social networks within churches also transmit political information (Djupe and Gilbert 2009). Everyday members of churches are capable of delivering political information in small, yet potent doses. Ostensibly apolitical moral judgments can carry political implications. Additionally, people glean political information about churches from the broad social environment. Stereotypes about the denomination or larger religious group could act as heuristics like partisanship. For instance, many people immediately recognize Southern Baptist churches as organizations with a decidedly conservative political bend.

2

Political information about churches can trigger two mechanisms that then change religious behavior. One mechanism is psychological; political messages in church can cause either affirmation or unpleasantness. The other mechanism is social; politics can enhance or compromise social integration in a church. Both mechanisms operate by changing organizational satisfaction. Political messages in church sometimes deviate from people‘s prior understanding of the world. The clash of messages from religion with a person‘s prior political understanding can compromise a person‘s ability to preserve a concept of self that is stable, competent, and morally good (Festinger 1957). Because such instances are psychologically unpleasant or ―dissonant,‖ it can lead to change, either in religious or political behavior. One way to reduce this dissonance is to neutralize a source of conflicting information. For people conflicted by a church, this means evaluating the church experience less favorably than they would if they rarely experienced discomfort. In the diminished favorability of the church, the individual becomes a member of an organization providing declining satisfaction. Conversely, not all political information in a church is contradictory or dissonant to all people. Some members of a church may criticize gay marriage legalization in front another member with a similar opinion. Likewise, a pastor might deliver a sermon admonishing a court decision before a similarly minded congregant. These examples of consonant political messages could enhance well-being by providing a narrative affirming prior beliefs. Therefore, consonant political messages can increase satisfaction with a church. In addition to these psychological or information processing mechanisms, politics can influence churches through a social mechanism. Prior research establishes that church-based social networks carry political information (Djupe and Gilbert 2009) and that politics influences

3

friendship patterns (McPherson 2001). Politics can strength or weaken social ties in a church in several ways. First, church members can directly clash over differences on issues. People who agree with fellow members are more likely to have reinforcing conversations (Mutz 2006) and less likely to disagree with coreligionists about politics. Second, seemingly apolitical statements about topics such as sexuality or gambling often are enmeshed with contentious political implications. Third, political minorities in a church may feel like they just do not ―fit in‖ with fellow church goers. Social networks are often more cohesive when members display ideological homogeneity (McPherson et al 2001). People in line with the dominant attitudes of their congregation are more likely to corroborate with coreligionists on a wide-array of non-political matters such as business and social outings, especially if the world continues to sort along political lines. And, those who agree about politics are likely to agree about other aspects of life as well (Hillygus and Shields 2008). Taken as a whole, these weakened social ties likely translate to weakened organizational satisfaction. In short, political messages associated with religion can influence social ties and psychological wellbeing. These factors in turn can reduce or increase satisfaction with a church as an organization. When organizational satisfaction is high or improving, people are more likely to be heavily invested in their church. When organizational satisfaction is low or declining, people respond with exit, neglect, voice, or loyalty (Hirschman 1970; Rusbult et al 1988). Overall, this process of exit, voice, and neglect manifests itself as changes in religious behavior. Exit is the abandonment of an organization- leaving or switching churches. Voice is church conflict that attempts to change the organizational aspects that are causing dissatisfaction. Neglect involves disengagement from an organization without exit. Forms of neglect in church

4

include declines in attendance, donating, volunteering, and other forms of organizational investment.

III. Why American Religion is Politically Responsive The contemporary United States is a ripe environment for politics to influence religion for six reasons. First, Americans have freedom of religion. The government does not mandate religious expression, though social pressures can still make some religions at least semiinvoluntary (Hunt and Hunt 2000). Second, the United States has a functioning religious economy. Individuals are free to select the churches that best match their preferences. However, market options and competitiveness vary based on geography and religious tradition (Finke and Stark 1992; Iannocconne 1998). Third, partisan politics and religious life forcefully entangled over the last quarter-century (Hunter 1991; Layman 2001; Hetherington 2010; Hetherington and Weiler 2009). Religious leaders such as Jerry Falwell became vigorous policy advocates and opinion leaders. Two leaders, Mike Huckabee and Pat Robertson, even mounted credible presidential campaigns. Fourth, political information associated with churches is ubiquitous. People connect politics with churches through religious elites, everyday churchgoers, and the media. Fifth, these political messages generally have a clear direction. When contemporary Christian organizations take political stances they tend to be consistently right-leaning. Sixth, political issues associated with churches are potent. The subject matter is emotional, salient, and easily-understood (Carmines and Stimson 1989). Abortion, gay marriage, and other issues commonly on the agenda of religious organizations clearly have this tendency to affect people, a tendency strengthened by partisan polarization.

5

IV. Who Changes Religious Behavior because of Politics Not all American churchgoers change their religious behavior because of politics. Two characteristics make people more prone to change. The first is religious tradition. Evangelical Protestants are susceptible to change while mainline Protestants and Catholics are not. Partisan preferences can influence the religious behavior of evangelical Protestants for several reasons. Evangelical churches have greater clarity in their political direction. Political messages associated with evangelicals are much more common. And, evangelicals typically care deeply about certain political issues. Furthermore, evangelicals have more market choices, closer substitute churches, and fewer social pressures against ―shopping‖ for churches (Smith and Sikkink 2003; Sherkat 2001). Consequently, the religion of evangelicals is uniquely responsive to politics. As evidence that environmental heuristics are clearer about evangelicals, more than twice as many people could not offer opinion on Lutherans or Presbyterians than those who had no opinion on Baptists (Barna Research Group 1992). Furthermore, ample evidence shows evangelical religious elites are more conservative and are more likely to express their conservatism (Hunter 2010; Hunter 1991). Another reason evangelical Protestants have to greater susceptibility to partisan influence is that Evangelicals tend to embrace literal or strict interpretations of the Bible (Hoffman and Bartowski 2008). This often translates to an emphasis on strict acceptance of biblicallyassociated social prescriptions, many of which have political implications (Barker and Carman 2000; Wolkomir et al 1997). The dogmatic nature of these affiliations means that they often offer themselves as holding the indelible truth on sociopolitical issues. Consequently, incongruent political messages in those churches are harder to ignore or not integrate into a person‘s thinking

6

than in less dogmatic churches. It is easier for a member to let go of some disagreeable statements if the denomination does not ask members to make infallible scriptural interpretations the primary arbiter of most of that member‘s cognitions. Furthmore, mainline Protestant clergy tend to discuss issues like homosexuality in less divisive language (Olson and Cadge 2002). The second is centrality of politics and the centrality of religion. According to Converse (1964), beliefs that are more likely to change are those that are less central. A person who holds political beliefs highly central is one who cares deeply about those beliefs and has integrated them in a less mutable understanding of the world. Similarly, a person who holds religion highly central to his or her life is one whose religious beliefs, belongings, and behaviors are wellestablished and stable. Those who hold religion more central to their lives are less likely to change while those who hold politics more central and more likely to change. Within any tradition, the religious behavior of some is more responsive to politics than that of others. Loyalty, the continued involvement with and support of an organization despite its problems, is a more likely response than neglect to declining satisfaction when politics is less important to a person or religion is more important. This is not to say that politics is the foremost force guiding religious behavior. On the contrary, scholars identify several groups of factors that influence religion. The first group is family socialization (Bader and Desmond 2006; Elison and Sherkat 1981; Ozorak 1989; Hunsberger 1984; Dudley and Laurent 1989; Nelsen 1981, Perkins 1987, and Ellison and Sherkat 1990; Bao et al 1999; Nelsen 1980). The second group is composed of demographic explanations such as gender and age (Krause 2003; Krause 2005; Nash and Berger 1962; Kalmijin 1998; Alston 1979; Batson et al 1993; Smith et al 1998; Shipley 1996; Hunt and Hunt 2000). Finally, the third hinges on contextual circumstances faced by the individual such as

7

church characteristics and religious market forces (Dyck and Starke 1999; Ammerman 1995; Kelley 1977; Iannacconne 1994; Stark and Finke 2000; Mark 2003; Stern 1999; Chaves 2006). Consequently, politics cannot explain all variance in religious dependent variables. Other nonpolitical factors will decide many people‘s religious behavior and belonging. Nonetheless, politics in churches influences a significant portion of decision making of a significant portion of the population.

V. Research Design Tests of the hypotheses generated by this theory rely on multivariate analysis of observational data. Observational data provides advantages of external validity, common availability, considerable temporal variation, and an extensive range of variables. The unit of analysis throughout is individual people. Individual level data better operationalizes the microindividual level theory. Data at the level of churches, affiliations, or geographic regions might offer some insights. However, reliable aggregate data on religious behavior are very limited (though see the Religious Congregations and Membership Studies for example), and conclusions drawn from them are prone to ecological fallacies (King 1997). Public opinion surveys provide the best available data on individuals‘ religious behavior. As with any instrument, surveys are not perfect. Surveys of religious behavior face two special problems. First, survey respondents often over-report their frequency church attendance, largely out of a social desirability bias (Hadaway et al 1993; Marler and Hadaway 1997; Hadaway et al 1998; also see Woodbury 1998; Caplow 1998; Hout and Greely 1998). Second, members of fundamentalist religious groups have a relatively high non-response rate leading to potential sampling bias (Sherkat 2007). Neither of these concerns is condemning. No evidence is

8

available suggesting that error associated with over-reporting of attendance correlates strongly with key independent variables like party identification. Over-reporting error is more of a concern in overtime comparison of aggregate religious behavior than in the analysis here. At worst, non-response rates among some fundamentalists constrain degrees of freedom. As with concern about over reporting attendance, non-response bias is relevant for making aggregate comparisons. Under and over reporting are important for studies asking if the country as a whole is secularizing or if a certain religious group is growing or declining. However, they have limited relevance for the generalizability of individual level findings. Each model operates through the rather austere assumption that political messages in and about evangelical Protestant churches in the United States have a strong general tendency to favor Republican politics. Accordingly, the theoretical expectation is for those respondents who identify more closely with the Republican Party to have stronger measures of commitment to their church. Conversely, those who identify as Democrats will have weakened commitment to their church as an organizations. This will result in changes in religious behavior, particularly in attendance. Undoubtedly, a monolithic assumption about the political content within evangelical churches is imprecise. The ―evangelical left‖ is a small but vocal counterpart to the right-wing Christian bloc. Prominent examples of left leaning evangelical leaders include Jim Wallis of Sojourners, sociologist turned pastor Tony Campolo, Duke Divinity professor and theologian Stanley Hauerwas, and, to many, President Jimmy Carter. While members of the evangelical left commonly agree with the evangelical right on major matters of belief, they often emphasize social justice and individual spiritual development over efforts at changing secular government and society (Hunter 2011).

9

The overall distribution of evangelical churches strongly leans to the right. Evangelicals who favor Democrats and information associating their churches with the left are still rare. People seek much more information about evangelicals on the right than they do about comparable evangelicals on the left. Over twelve times more people in the United States searched Google using the name of the Republican Pat Robertson than with the name of the Democrat supporting Jim Wallis.1 Due to the skewed distribution of partisan associations with evangelicals, the error in considering evangelical churches as transmitters of Republican messages is very small. The error does not compromise causal analysis. The imprecision should only dilute the relationship between party identification and church attendance. Democrats would not face dissonance and Republicans would not find consonance in liberal churches. Likely, this leads to an underestimation of the effects of party, making significant findings harder to obtain. However, the underestimation is not a major concern because the primary purpose of the analysis is to test the existence of a basic causal relationship rather than an estimation of the exact magnitude of partisanship‘s influence on religious behavior. All models include a measure of party identification and a dummy variable indicating if the respondent belongs to a church within an evangelical denomination. These variables are also interacted. Interactive variables allow testing the conditional influence of partisanship. The interacted variable reflects party‘s influence among evangelicals, though it requires interpretation in concert with its component variables (Kam and Franzese 2007). The interaction provides an 1

These counts were calculated using Google Trends (www.google.com/trends). The counts

reflect searches in 2008. Count data are only available as ratios between terms rather than raw numbers. Ripberger (2011) finds Google search traffic to have convergent validity with issue coverage in the New York Times.

10

advantage over splitting survey samples to only evangelicals in that the non-interacted variable for party reflects the influence of partisanship among non-evangelicals. I limit samples to whites who are Protestant or Catholic. This isolates variation to allow meaningful comparisons of evangelicals with the reference category, Catholics and nonevangelical Protestants. I do not analyze the behavior of non-Protestants and non-Catholics for four reasons. First, the most politically active religious organizations in the United States are Catholic or Protestant. Second, sample sizes of non-Catholics and non-Protestants are too small for meaningful conclusions. Cell sizes would be in the single digits for many groups. For example, only four Muslims appear in one of the datasets (NES 2000-2004 Panel). Third, generalizations about the political information in and about non-Catholic and non-Protestant churches are much less compelling; it is difficult to know the partisan leanings of small groups such as Buddhists; and many non-Catholics and non-Protestants are first generation immigrants without consistent party ties (Lee 2008). Fourth, attendance patterns of other groups can be less responsive to organizational satisfaction. As a result, the meaning of attendance is less comparable. I do not analyze the behavior of non-whites for the latter two reasons. Because many churches are racially homogenous (Dougherty 2003), African Americans who identify as evangelical Protestant are unlikely to be members of Republican favoring churches. Similarly, African Americans have a significantly higher mean frequency of attendance with arguably less discretion (Hunt and Hunt 2000). The investigation in this chapter has several parts. First, I show a powerful effect of partisanship on religious behavior through analysis of the 2000-2004 NES panel survey. Second, I find that church attendance does not have a similar influence on partisanship, ruling out concerns of simultaneity. Third, I continue analyzing the 2000-2004 NES panel survey to test

11

who is more likely to change religious behavior because of politics. Finally, I consider the alternative hypothesis that differences in belief between Democrats and Republicans drive differences in attendance.

VI. Establishing Causation – Panel Models of Party and Church Attendance Religious leaders, the broad environment, and church based social networks transmit political messages that members associate with their church as an organization. Those who attend church frequently have higher exposure to this same political information. Usually these messages have persuasive intent. Going to church regularly may change a person‘s mind about politics. Consequently, intense religious behavior could also influence partisan attachments. This presents a thorny problem for a research design. The same process that theoretically leads to attendance change could also lead to partisan change. Testing this theory requires serious attention to simultaneity.

Fortunately, causality

concerns are surmountable. Panel data are well suited to treat the basic causal problem. Panel surveys ask questions to the same people at multiple points in time. In doing so, they measure both religious behavior and political attitudes of the same people over at least two points in time. In this case, I use measures of the independent variables taken prior to the measure used as dependent variables. Panel data overcome estimation difficulties in several ways. Finding a significant relationship between a dependent variable and an independent variable measured at a time prior to the dependent variable –a lagged variable– is meaningful evidence that the independent variable is truly driving the relationship. This extends from the reasoning that future behavior does not influence past behavior. Similarly, not finding such a relationship is strong evidence

12

that the independent variable does not influence the dependent variable. A lagged measure of the dependent variable further strengthens finding (Keele 2006). It provides a baseline that obviates the likelihood that past correlation is driving future correlation. This negates the chance that contemporaneous measures of the dependent variable simply mirror past levels. For statistical reasons, it is akin to measuring change in the dependent variable. The American National Election Study conducted a panel survey in three waves, first in 2000, second in 2002, and third in 2004. Each face-to-face interview within the surveys is part of a high-quality and nationally representative sample. The period between 2000 and 2004 is of key interest as it coincides with events that brought religion into political contexts. In particular, the George W. Bush administration and surrounding salient issues galvanized many religious communities (Campbell and Putnam 2010; Jacobson 2010). I employ a two-part test of the influence of partisanship on religion using the 2000-2004 American National Election Panel study. First, I establish that lagged partisanship has a significant effect of non-lagged church attendance even when controlling for lagged church attendance and a host of other variables. Then, I find that lagged church attendance does not have a significant effect on contemporaneous partisanship. In brief, partisanship influences church attendance but church attendance does no influence partisanship. These findings provide considerable support for the argument that the causal arrow only points one way- from partisanship to religious behavior. This is a direction somewhat contrary to conventional scholarly understanding.

The Influence of Party on Church Attendance --- FIGURE 1 ABOUT HERE ---

13

A simple comparison of means shows major differences between partisans in their proclivity to change in frequency of church attendance between 2000 and 2004. Regardless of partisan affiliation, most respondents declined at least slightly in attendance frequency over the four years in the survey. However, Democrats declined much more than Republicans did. Figure 1 groups by tradition individuals‘ self-reported attendance in 2004 subtracted by their self-reported attendance in 2000. For presentation purposes, I multiplied these values by -1 so that higher values indicate greater decline. The maximum potential decline is five (a person who went from attending more than weekly to never). The average decline across the entire survey sample is 0.05 with a standard deviation of 1.17. True independents and party-leaning independents are not part of the comparisons. On average, evangelical Republicans declined a paltry 0.08. However, evangelical Democrats declined 0.67. This is statistically distinct (p