Polit Behav (2008) 30:367389 DOI 10.

1007/s11109-008-9053-1 ORIGINAL PAPER

American Republican Religion? Disentangling the Causal Link Between Religion and Politics in the US
Stratos Patrikios

Published online: 5 February 2008 Springer Science+Business Media, LLC 2008

Abstract Recent research in American political behavior has examined at length the link between evangelical Protestants and the Republican Party. These works however do not consider the idiosyncratic nature of religiosity in the US, and insist on treating religion as an unmoved mover with respect to political contexts. The question posed herein is: during the participation of religious communities in partisan politics, should we expect politics to eventually constrain religious behavior? Motivated by a political social identity approach, I use American National Election Study panel data and structural equation modeling techniques to explore the untested possibility that religious and political factors are linked through reciprocal causation. Conditional upon religious and temporal context, ndings highlight the causal impact of ideology and partisanship in shaping religious behavior. Keywords Religious politicization Church attendance Party identication Ideology Social identity theory

Introduction The nexus between religion and politics in violent and non-violent conicts tends to generate global scholarly and popular attention. American society in particular serves as a proverbial case, where the peaceful but vocal participation of religious
With apologies to Alexis de Tocqueville. Electronic supplementary material The online version of this article (doi:10.1007/s11109-008-9053-1) contains supplementary material, which is available to authorized users. S. Patrikios (&) Department of Government, University of Strathclyde, McCance Building, 16 Richmond St., Glasgow G1 1XQ, UK e-mail: e.patrikios@strath.ac.uk

123

368

Polit Behav (2008) 30:367389

populations in the political process is now widely considered an endemic phenomenon (Leege and Kellstedt 1993; Green et al. 2003). The relationship has been identied as a cleavage, a concept describing the translation of objective social divisions into enduring political conicts, with original reference to the formation of West European party systems (Lipset and Rokkan 1967). Given the interest and concerned tone of most secular social science on the matter, it is surprising that research has tended to probe the role of religion in politics relative only to the impact of stable religious variables on political behavior. This sociological interpretation of politics expects that exogenous religious processes (e.g. exposure to church contexts) shape political behavior (e.g. partisanship), and explains relevant trends accordingly. Recent studies in this direction have debated the drift of mainline Protestants away from the Republican Party, the stability of Catholic traditional support for the Democrats, but above all the entrance of evangelical Protestants into the Republican electoral base in the 1980s and their transformation into an efcient political machine in the 1990s, as described in Fig. 1 (e.g. Moen 1994; Wilcox 1996; Guth et al. 1997; Layman 1997, 2001; Manza and Brooks 1999; Bolzendahl and Brooks 2005). The present study examines the relationship between evangelical Protestantism and the Republican Party in recent decades. However, it avoids the aforementioned sociological reductionism and its limited focus on the political impact of religious factors, such as churches, para-church organizations, lobbies and demographics. Instead the following discussion shifts to the effects of politicization on religious behavior itself, dened here as church attendance. In this way, I pursue a more nuanced direction in the study of cleavages, and propose that the products of the infusion of religion into politics are not restricted to the electoral realm, but can potentially transform religion into a secular/political phenomenon. According to this path, the inuence of religious practice on political behaviora conventional

Fig. 1 Partisanship among frequently attending evangelicals, 19602004. Source: 19602004 ANES Cumulative File (excluding African Americans). Note: Independent category includes leaners. Frequent attendance is dened as regular until 1968, and as almost every week or more thereafter

123

Polit Behav (2008) 30:367389

369

assumption adopted by most political scientistscan be supplemented by a reverse effect, whereby religious practice becomes constrained by political/partisan concerns (cf. Sartori 1969; Kriesi 1998). Michael Hout and Claude Fischers sociological study (2002) is to my knowledge the only published quantitative effort that attempts to explore this expectation, as an investigation of political pressures on individual religiosity in the context of the recent religious politicization in the US. The investigation centers on apostasy, i.e. the phenomenon of Christians dropping out of church. The authors argue that the conservative religious politicization of the 1990s caused the following backlash: ideologically liberal and moderate Christians abandoning conservative denominations. Hout and Fischer interpret part of this movement as a reaction against the Christian Rights political agenda, and the prominent place occupied by conservative Protestantism in GOP ranks (2002, pp. 181, 185). The present article builds on Hout and Fischers effort, updating it in three ways. First, the authors provide only an indirect test of their expectation using crosssectional data; the models presented herein avoid assumptions of temporal precedence by turning to panel data, and provide therefore a more rigorous test of the hypothesis. Second, the 2002 work focuses on how personal ideological orientation determines apostasy. In what follows, I elaborate on this idea and develop an additional explanation rooted in partisan inuences, on the basis of a well-documented phenomenon in realignment research. This refers to the sortingout, i.e. the overlap between ideology and partisanship, and the polarization experienced between the two major parties since the late 1970s. In this development, culturally liberal Republicans abandoned the GOP and conservative Democrats followed a similar movement away from their party (e.g. Poole and Rosenthal 1984; Abramowitz and Saunders 1998; Levendusky 2005). Finally, my work focuses on changes in religious practice and not on apostasy, since the low number of apostates in the datasets is inadequate for multivariate analysis, and relevant repeated measures are missing. My expectation is that the politically charged American religious landscape, in particular the close association in the public mind of evangelical Protestants with social conservatives and the Republican Party, can lead individuals to react by altering their religious behavior. Religious politicization, loosely dened as the perceived afnity of certain religious populations with certain parties or ideological camps, creates the potential to make some believers minimize their attendance at certain churches and others to increase their church-going. This process is summarized as an extrinsically political religion fuelled by ideology and partisanship. The two competing expectations are specied as: (i) an attendance effect on politics (the sociological approach) and (ii) an ideological/partisan effect on attendance (the political religion phenomenon). Based on social identity theory with a strong emphasis on partisanship (Greene 1999, 2002, 2004; cf. Green et al. 2002), I use panel data from the Michigan/ American National Election Study (ANES) pool, which both precede and overlap with the religiously charged eras of the mid-1990s and early 2000s. Drawing on the political behavior literature, particularly on methodological advances regarding the ` -vis other political concerns, a series of exogeneity of party identication vis-a

123

370

Polit Behav (2008) 30:367389

cross-lagged effects models tests competing expectations. Results support the political religion hypothesis, either ideological or partisan, and open the way for a novel understanding of the continuing impact of the political process even on nonpolitical variables, and of the worldly nature of religious behavior. Due to the absence of relevant data on religious afliation and identication, the analysis is based on the interplay of political factors (ideology and partisanship) with church attendance. The argument begins with the American religious exceptionalism thesis, and continues by reviewing a recent attempt to revitalize the party identication concept as social identication. The former discussion supports the use of religious variables as endogenous to the political process, while the latter allows for the possibility of partisan and ideological effects on religious variables. The analysis then continues with a consideration of methodological problems, followed by an interpretation of ndings and their implications for research on electoral politics and religion.

American Religion in the Left-Hand Side Based on an assumed equivalence, the use of religious explanations in American political research often reects an idealized European experience of religion as a stable, politically exogenous phenomenon. These features are treated as adequate reasons for the use of religious variables as stable demographics. Sociologists of religion however have long agreed on the idiosyncratic nature of American piety, especially among native Protestants. First, the civil religion phenomenon (Bellah 1967) challenges the idea that Christian faith in the US merely reects theological concerns. Civil or general religion represents faith in the American way of life and the status of the US as Gods chosen nation. This infusion of religion with secular values provides a rst step for recognizing the partly secular component in the religiosity of American Christians. Second, the religious market metaphor claims that the existence of a plethora of churches in America (religious pluralism) creates the conditions that reinforce religiosity among the American population (Finke and Stark 1992). This deregulated setting implies that churches have to tailor/improve their products and even offer less spiritual benets, in order to increase demand. Again, it appears that the religious package put forward by American churches, especially the less centralized Protestant ones, has a demystied quality, one that is infused with secular benets. Finally, the character of the products offered by American religious institutions goes hand in glove with the motives behind the preferences of American believers. In a pluralistic religious market located within a heavily consumerist society, it is reasonable to suggest that the public experience of religiosity is closer to that of voluntary association, and not of passive socialization (Newport 1979). The idea of religion as an active, conscious choice sounds less eccentric when we consider one phenomenon with prominence in the relevant literature: denominational switching (Roof and McKinney 1987). Switching from one church to another, also known as religious mobility, concerns the unstable character of religious choice and consumption in the United States, especially within American Protestantism.

123

Polit Behav (2008) 30:367389

371

This type of mobility, practiced by up to a third of Americans at some point in their life, supports an image of religion as a self-selective, dynamic choice that cannot be treated as permanent (Warner 1993; Loveland 2003). Overall, the American religious exceptionalism thesis challenges the use of religious factors as stable demographics, and instead maintains that religiosity is a volatile choice, open to secular inuences. This study uses this understanding and posits that in specic circumstances, part of these inuences can take the form of ideological and partisan concerns.

A Political Explanation of American Religiosity The theoretical justication of my argument is based on a social identity conceptualization of party identication, which can also be extended to identication with ideological camps. I argue that this provides an explanation of how politicizationeither on ideological or partisan linesinitiates a transformation of religion. The Michigan school and its disciples have treated identication with a party as a multifaceted concept: evaluation of partisan objects, perceptual screen for interpreting incoming stimuli, and psychological expression of group membership (e.g. Campbell et al. 1960; Miller and Shanks 1996). The concept has been subsequently challenged as endogenous to policy preferences, ideological predispositions, evaluations of candidates and government performance (e.g. Nie et al. 1979; Page and Jones 1979; Fiorina 1981). Recent theoretical and methodological advances however, have reinstated the traditional view of party identication as a long-term predisposition, which affects voters evaluations and perceptions of political reality and even their core political predispositions, without being affected by them in the short term (Green and Palmquist 1990; Green et al. 2002; Goren 2005). Miller and Shanks summarize this view: one of the roles of the church, or the party, is to provide structure to the ordinary persons understanding of the external world[and] cues for normative assessments of the outside world (1996, p. 121). Instead of being treated as a simple attitude towards a political object, party identication can be conceptualized through social identity theory (SIT) as psychological belonging to a social group (for an overview of theoretical and measurement issues see Greene 2002, 2004). In general, the group psychological effects identied by SIT are termed social identication and self-categorization. These correspond to a denition of the self according to group characteristics, and to an exaggeration of differences between own group and other groups in order to achieve a positive self-concept, respectively (Tajfel 1981; Long and Spears 1997). For SIT, an individual that feels closer to a group tends to internalize group membership by seeing herself through group stereotypes and not through personal characteristicsI am Democrat/Catholic/Hispanic/Liberal. The cognitive process of self-categorization explains how SIT works, that is, through intergroup social classication (Turner 1985). People assign social objects into us/them categories: the us category represents the in-group, where people feel they belong, while the them category stands for the out-group, where non-members are located. In this process, individuals attempt to make the in-group more

123

372

Polit Behav (2008) 30:367389

distinctive than the out-group by conducting biased comparisons (stereotypes) with members of the out-group. According to social identity and self-categorization theory then, individuals will (a) tend to perceive themselves less as unique units and more as group members; (b) cement their impulse for a positive self-image on exaggerated comparisons with out-group members; and (c) follow in-group standards in attitude and produce groupy behavior (Hogg and Terry 2000, p. 121). The three points are interrelated in the sense that self-perceived membership to the same group expects conformity with shared in-group standards (and against out-group standards) in order to achieve greatest possible perceived inter-group distinctiveness. The original exposition of the party identication concept in The American Voter already contained an emphasis on this belonging dimension of partisanship, in the sense of an us versus them distinction, besides its function as a perceptual screen or political attitude (Campbell et al. 1960, pp. 1335). The SIT approach towards partisanship then expects that citizens who experience greater group identication with fellow partisans (gauged with scales that measure feelings of belonging to a social group) will tend to internalize in-party and out-party stereotypes, exhibit increased engagement in partisan behavior, like rally attendance, and exaggerate differences between us and them, even after controlling for the effect of the traditional party identication variable (Greene 2004). The above opens the way for examining the overlooked effect of party identication on individual religious characteristics. Typically, parties raise the salience of social-political links by sending out clear references via election manifestos, candidate speeches and policy proposals. Famous among many examples is the 1992 Republican National Convention, especially the culture war speech by Pat Buchanan, which represents a symbolic milestone in the partys effort to associate itself with conservative family values, popular among theologically conservative populations. This effort has not been limited to the 1990s, as evident by an inspection of the news since 2000, i.e. since George W. Bushs election in the White House. The President, a born-again Christian, has consistently stressed the link between Republicanism and theologically conservative religion. For instance, in what appears to be a conservative reading of Christian faith he hindered federal funding of prochoice groups abroad in 2001, while taking steps to promote funding for religious service organizations. Bush has also openly declared his opposition to same-sex marriage by favoring a constitutional amendment that made it illegal in 2004. It is not surprising then that evangelical leader Jerry Falwell described Bushs reelection in 2004 and the role played by evangelicals in it as: a slam dunk as the Church of Jesus Christ made the difference in initiating the return of this nation to moral sanity and the Judeo-Christian ethic (quoted in Layman and Hussey 2005, p. 1). s that directly or indirectly identify two The outcomes of such recurring cliche labels as naturally connected (Republicanevangelical Protestant or conservativeevangelical Protestant) can be twofold. If we focus on the religious group as our building block, as political science normally does opting for a sociological interpretation of politics, the expectation is that members of specic religious communities will conform to group norms and move closer to the prescribed party (cf. Green et al. 2002). Other things being equal, evangelicals will tend to move

123

Polit Behav (2008) 30:367389

373

closer to the GOP, which constitutes only one interpretation of the trend described in Fig. 1. Yet, members of religious groups are also attached to partisan groups, identities that become particularly energized during election periods. If we concentrate on the partisan in-group, self-categorization theory expects that labels associated with the in-party will guide its members towards adopting similar meanings of reality and desirable preferences, or actions. If in-group (party) members are Republicans, and assuming that the link between social group and party is very prominent, members will be exposed to the stereotype (the social imagery of the party), which connects, for instance, Republicanism with evangelical Protestantism, and Democrats with Catholicism, or in more recent years, with secularism. In this case, religiosity becomes an in-group (party) norm and according to SIT, group identiers will be more likely to follow this religious norm. In a self-selective process, Republicans will tend to stress their Republicanism by intensifying their commitment to Evangelical churches; alternatively, Democrats will be under pressure to distance themselves from this religious environment, which contradicts their partisan in-group norms.1 A similar rationale could apply to ideological camps (Lau 1989; Deaux et al. 1995). According to Hout and Fischer (2002), the conservative avor dominating evangelical churches since the 1970s has driven non-conservative evangelicals away from those churches. Using SIT, the distancing of moderate/liberal believers from evangelical churches can be interpreted as the result of groupy behavior. Here, liberal evangelicals would be expected to follow in-group (liberal) standards and react to the conservative turn of their churches by avoiding religious services. Conversely, conservative evangelicals would tend to adopt in-group (conservative) standards and strengthen their links with evangelical churches by attending more frequently. In politicized periods, the connection of evangelicalism with political conservatism would have attendance become the norm of the ideological (conservative) in-group, with the reverse holding for the opposite (liberal) ingroup. This process should accompany the conventionally assumed inuence of church on ideological orientation. Measurement and Analytic Strategy The starting point in my effort to test the hypothesized bidirectional causal link between religiosity and political variables is a specic eld in the party identication scholarship, which explores the possibility of reciprocal causation between partisanship and other political variables (Jackson 1975; Markus and Converse 1979; Page and Jones 1979; Layman and Carsey 2002; Goren 2005; Carsey and Layman 2006). This motivation, coupled with the theoretical emphasis of SIT on partisan groups, justies the preference given by the following methodological discussion to partisanship over ideology. The same analytic framework however is used for assessing the effect of ideology on church attendance.
1

This stereotyping might seem a simplication of realityfor example, not all Republicans or conservatives are evangelicals. SIT stresses the point that members perceive in- and out-groups as homogenous, and not that they actually embody these stereotypes (Huddy 2003).

123

374

Polit Behav (2008) 30:367389

To evaluate the religious inuence on politics and the parallel political inuence on religious behavior, I use ANES panel data. Three electoral cycles are covered: 19721976, 19921996, and 20002004.2 Following common practice (e.g. Lenski 1963) African Americans have been excluded from the analysis. This segment appears only in low numbers in ANES samples, and also is distinctive in terms of historical and demographic characteristics, and organizational autonomy within the Protestant family. The variable that operationalizes social identication with a party is the ANES partisanship item: a seven-point scale, with high scores showing Republicanism. Although not a perfect measure of psychological group identication, the Michigan item could be considered an adequate indicator of such processes when no alternative is available (Greene 2002). The ideological self-placement scale is selected to tap identication with the liberal and conservative camps; high scores on the seven-point scale represent conservatism. I use the original root question, and not the summary three-point measure, which also allocates leaners. This ensures that when examining reciprocity between ideology and attendance, both scales have a high number of points. Due to limited item availability in the panels, the dimension of religion examined is church attendance, a ve-point scale with high scores indicating frequent attendance.3 Table 1 contains summary statistics for intra-group trends across waves. With the exception of Hout and Fischer (2002), almost all studies of the relationship between religion and politics, in the American case and beyond, propose unidirectional causal effects from religious variables to political ones (for instance, see contributions in Leege and Kellstedt 1993). In assuming the sole existence of this ow of causality, these studies are content to model cross-sectional data, drawn at a single point in time. Considering the typical conjecture in such research, namely that religious variables represent xed personal characteristics, temporally prior and exogenous to the political process, analyses without a temporal dimension seem to serve the purpose. Yet, the untested assumption of unidirectional causation can easily be evaluated when temporal precedence is embedded in the data, i.e. when repeated measurement of the same individuals is available across time (Finkel 1995, pp. 223).4

Data used in the present study were made available by the Inter-University Consortium for Political and Social Research. The author holds sole responsibility for their analysis and interpretation. Results from supplementary analyses mentioned in the text are available from the author. Throughout the study, I avoid using the 19561960 ANES panel, since the detailed measure that differentiates among Protestant denominations and is required for sub-group analysis had not yet been introduced in 1956. Also, in the examination of the reciprocal link between ideology and attendance, the 20002004 panel is dropped, since there is no repeated measure of ideological self-placement in 2004. Note that the item may be subject to misreporting due to social desirability effects, a problem however that cannot be corrected with ANES data (e.g. Hadaway et al. 1993). Establishment of causal precedence is possible with cross-sectional data, when simultaneous effects are assumed between two variables (Finkel 1995). This however requires the use of unrealistic modeling options in the form of instrumental variables. According to theory, instruments should be very strong predictors of one endogenous variable, but not related to the second endogenous variable. Instrumental variables are difcult to locate in social surveys, unless designers had planned ahead and included such indicators in the questionnaire.

123

Polit Behav (2008) 30:367389

375

Table 1 Descriptive statistics for 19721976, 19921996, 20002004 panels: means (standard deviations) Variable [scale] Catholic Mainline Evangelical

Descriptive statistics for 19721976 panela Party ID 1972 [06] Party ID 1976 [06] Ideology 1972 [17] Ideology 1976 [17] Attendance 1972 [15] Attendance 1976 [15] Party ID 1992 [06] Party ID 1996 [06] Ideology 1992 [17] Ideology 1996 [17] Attendance 1992 [15] Attendance 1996 [15] Party ID 2000 [06] Party ID 2004 [06] Attendance 2000 [15] Attendance 2004 [15]
a

2.18 (1.90) 2.11 (1.88) 4.18 (1.16) 4.25 (1.19) 3.73 (1.49) 3.63 (1.48)
b

3.37 (2.02) 3.29 (1.98) 4.30 (1.19) 4.51 (1.23) 3.00 (1.41) 2.94 (1.40) 3.97 (1.73) 4.11 (1.95) 4.58 (1.30) 4.79 (1.21) 2.71 (1.51) 2.72 (1.44) 3.22 (2.11) 3.35 (2.25) 2.91 (1.46) 2.94 (1.51)

2.84 (1.98) 2.66 (1.96) 4.62 (1.03) 4.81 (1.14) 3.41 (1.49) 3.50 (1.48) 3.28 (1.98) 3.49 (2.02) 4.59 (1.53) 4.76 (1.33) 3.33 (1.59) 3.16 (1.59) 3.35 (1.97) 3.63 (2.16) 3.16 (1.63) 3.28 (1.58)

Descriptive statistics for 19921996 panel

2.54 (1.80) 2.59 (2.05) 4.16 (1.35) 4.18 (1.35) 3.37 (1.56) 3.31 (1.55) 3.13 (2.06) 3.30 (2.24) 3.18 (1.56) 2.89 (1.58)

Descriptive statistics for 20002004 panelc

Source: 19721976 ANES panel. Note: Higher scores indicate Republicanism (6 is strong Republican), conservatism (7 is extremely conservative) and frequent attendance (5 is every week)
b Source: 19921996 ANES panel. Note: Higher scores indicate Republicanism, conservatism and frequent attendance c

Source: 20002004 ANES panel. Note: Higher scores indicate Republicanism and frequent attendance. No repeated measurement available for ideological self-placement

Still, most investigations remain innocent as to what happens to religious characteristics once they are exposed to partisan politics. The causal relationship proposed in this analysis takes this step and identies two mutually reinforcing effects: a religious effect on politics (the orthodox assumption in political science), and a political effect on religion. Their combination constitutes a feedback phenomenon. The causal link between religiosity and partisanship is specied as a cross-lagged effects model with the following structure (see Fig. 2): PartyIDt b1PartyIDt1 b2Attendancet1 e1t Attendancet b3Attendancet1 b4PartyIDt1 e2t 1 2

Coefcient b1 in Eq. 1 represents the impact of partisanship at time t - 1 on partisanship at time t. This is the stability coefcient, showing how volatile partisanship is in the course of two consecutive presidential elections, net of the

123

376

Polit Behav (2008) 30:367389

A. Unidirectional effect
Political variable (t-1) Political variable (t)

Attendance (t-1)

Attendance (t)

B. Feedback effect
Political variable (t-1) Political variable (t)

Attendance (t-1)

Attendance (t)

Fig. 2 Model comparison: unidirectional versus feedback effects. Note: The sole causal assumption of studies working with cross-sections is represented by coefcient a in panel A. The feedback model adds the political effect on attendance represented by coefcient b. Errors, correlations and controls have been excluded for simplicity

inuence of attendance at time t - 1. An equivalent logic applies to b3 regarding church attendance as the dependent variable in Eq. 2. Coefcient b2 is the crosslagged effect of attendance at time t - 1 on partisanship at time t, net of the effect of partisan stability. The equation then predicts changes in partisanship. The same applies to b4 regarding attendance as the dependent variable. For the effect of ideology on attendance, the partisanship variable is replaced by the ideology measure. The two effects underlying the politicization of religion are: rst, that attendance shapes partisanship (or ideology); second, that partisanship (or ideology) reinforces attendance, i.e. the political religion effect. If the latter is correct, we should not only observe b2 [ 0 in Eq. 1, but also b4 [ 0 in Eq. 2. Alternatively, if only the sociological view that dominates electoral studies is true, we should only observe b2 [ 0 in Eq. 1, and b4 = 0 in Eq. 2. As will be argued below in more detail, I expect that the political religion phenomenon emerges under specic circumstances, namely when the social image of the party or ideological camp is exceptionally salient.

123

Polit Behav (2008) 30:367389

377

Models were estimated with AMOS 6.0 and full-information-maximum likelihood (Arbuckle 2005).5 The analysis does not report coefcients from the above causal specication (Eqs. 1 and 2) for all subgroups across all panels. An explicit test is proposed, which directly compares alternative models. Specically, I contrast the t of the feedback model (the political-religion phenomenon) to that of a unidirectional model, which postulates that the only effect taking place during religious politicization moves from attendance to ideology/partisanship (see Fig. 2). This constitutes a more explicit evaluation of the unidirectional hypothesis versus the reciprocal one (Bollen 1989, pp. 2912). The comparison is conducted through a chi-square difference test, since the constrained model with unidirectional effects is nested within the unconstrained model that proposes the feedback. The difference between the chi-square values of the two nested models is itself distributed as a chi-square value with degrees of freedom equal to the constraints imposed on the second model. For reasons of presentational parsimony, I only present results when the reciprocal hypothesis (unrestricted model, where the ideological/partisan effect on attendance is freely estimated) shows a signicantly better t against the traditional, unidirectional expectation (restricted model, with the political effect constrained to equal 0). For comparison, results that do not support the feedback hypothesis appear in the Appendix (Supplementary electronic material). The hypothesis of lagged instead of synchronous effects between the main variables denes the recursive character of the model, and makes identication simple. A specication with synchronous effects would suggest that the two main factors inuence each other at a single point in time. Such models break the condition of independent variables being uncorrelated with the residual, since the cause is at the same time inuenced by its effect (Finkel 1995, p. 32). In this case regression estimates would be biased, therefore analysis must turn to the use of instrumental variables (see note 4). Due to heroic assumptions and the gradual nature of most social psychological effects, I consider such non-recursive models a less plausible scenario. In any case, researchers that employ feedback models of ANES data suggest that the selection of lagged over synchronous effects makes no difference to the estimation of causal relationships (Goren 2005; Carsey and Layman 2006). Finally, I have specied additional elements in the cross-effects models, in accordance with the literature. First, the error terms in Eqs. 1 and 2 are correlated. This allows estimating whether the two dependent variables at time t share at least one omitted independent variable (Kline 1998, p. 101). If the disturbances were left unrelated, that would represent the less plausible assumption that the two dependent variables do not share any common, unobserved, causes. This is especially unwarranted, when one considers that partisanship, ideology and religiosity are
5

FIML does not exclude missing cases from analysis for respondents included in the models. Under the assumption of data missing at random, the procedure produces superior estimates to either listwise or pairwise deletion, or mean imputation (Arbuckle 2005). Regarding panel effects, Bartels study (1999) shows that the ANES design does not suffer from serious panel attrition and conditioning, with the exception of campaign interest and turnout variables.

123

378

Polit Behav (2008) 30:367389

outcomes of the same pre-adult socialization process. Standard demographic controls were added as predictors of ideology/partisanship and religiosity at time t, together with their inter-correlations (see notes in Tables 3 and 5).

The Impact of Context A central element in SIT is that social identities are dynamic, i.e. their strength and inuence depends on temporal and social context (Hogg and Terry 2000). Huddys critique of SIT (2003) emphasizes such situational inuences, which result in the variable salience of identities relative to settings. In essence, context can result in individuals switching from one social identity to another. Consider the example of a politician who openly stresses differences between two ethnic groups during her campaign. Here, identication with ethnic groups is likely to become the dominant identity among the public. In another instance, the participation of an in-group member as a candidate in elections could also raise the visibility of the in-group and out-group demarcation (Huddy 2003, pp. 533, 543). The point of context-dependent identity salience guides the selection of multiple-group estimation, discrete timelags, and emphasizes the importance of presidential election campaigns in highlighting the religion-politics nexus. In the following, I justify these three features of the models. First, cross-lagged effects were estimated separately for each major religious subculture in the US. The classication of churches into three major religious groups (Catholic, mainline and evangelical Protestant) follows a standard categorization scheme, which allocates denominations into broader religious families (see Steensland et al. 2000).6 Contextual differences among the three groups make this decision essential (cf. Layman and Green 2005). For instance, the lack of vocal electoral participation within Roman Catholicism is one element distinguishing this religious tradition in the US, and making the expectation of substantial effects produced by the linking of religion with politics less plausible. Regarding analytic discrimination between mainline and evangelical Protestant denominations, a main element of my hypothesis is that politicization takes place to different extents and in different periods for the two groups. Research generally agrees that the initially higher socioeconomic standing of mainline Protestants has led to a different timing of the political mobilization phenomenon compared with evangelicals, who entered the political arena at a later phase (Roof and McKinney 1987; Finke and Stark 1992; Manza and Brooks 1999). This justies a separate estimation of the hypothesized relationships for mainline and evangelical Protestants.

Steensland et al. classify nondenominational Protestants as evangelicals according to their church attendance (frequent attendance suggests evangelicalism) (2000, p. 316). Since the present analysis is built on the church attendance variable, adopting the above practice would have introduced a biased logic: religiosity would feature both as an independent/dependent variable within religious groups and as a stratication criterion across groups. Facing the risk of introducing an amount of unwanted variability in the groupings, nondenominational Protestants were assigned to the evangelical group in recent decades, irrespective of their observance.

123

Polit Behav (2008) 30:367389

379

The phenomenon of religious behavior driven by partisan (or ideological) concerns is hypothesized to emerge within a religious community where ideological or partisan identities are increasingly salient. I locate this condition in the post1980s political era among evangelicals, who were at the forefront of electoral politics during two periods covered by the data.7 Specically, the mid-1990s were a time of extreme cultural polarization and salient religious and political identities. Similarly, the early 2000s have been a period of equally prominent connections between the GOP and conservative Protestantism, and witnessed a reborn Christian heading the Republican ticket in an atmosphere of cultural polarization. It remains an empirical question however whether the feedback hypothesis holds among evangelicals for the period covered by the 19721976 data. This was a time when evangelical Protestants were already ideologically energized by the Social Issue, but not yet explicitly attracted by the GOP. Hence, the link between Republicanism and evangelical Protestantism had not yet been consolidated in the public mind. After all, 1976, the Year of the Evangelical, was dened by the Democratic candidacy of Jimmy Carter. The stratication by religious tradition unfortunately encounters practical obstacles, which render it susceptible to a degree of criticism. ANES panels normally measure denominational afliation only in the rst wave, under the obvious assumption that switching does not take placeor that the phenomenon is irrelevant to politics. For instance in the 20002004 panel (repeated measures of denominational afliation not available), missing repeated measurement of afliation requires denition of religious groups as follows: individuals are assigned to the Catholic group, for example, if they are members of this group in the rst wave of the panel. The obvious assumption is that these individuals remain members of the same religious community in subsequent waves. In other words, division of the sample into religious traditions assumes stability in religious preference across waves, an expectation challenged by the switching phenomenon. It would have been more reassuring to dene each religious group by selecting respondents that consistently belonged to the same church across panel waves, but this was not feasible especially for the 20002004 study. In this sense, it is a possibility that some participants dened as Catholics or evangelicals in the rst wave may have already dropped out of religion by the following wave or converted. Second, the time distance between measurements is crucial to the detection of causal effects. The problem of specifying the most plausible, if any, delay between cause and effect does not have a straightforward solution (Asher 1983; Bollen 1989). It depends partly on the theoretical conception of the causal effects and partly on data restrictions. The use of ANES three-wave panel data constrains my selection of time-intervals between cause and effect, since the rst and nal waves are
7

The effects reported in this paper hold even with more homogenous stratications. For instance, if models are estimated within born-again evangelicals (measure not available in 19721976), the partisan religion effect is intensied. I do not emphasize this result however, for two main reasons. First, subsample size further decreases. Second, the born-again item is a subjective measure, less reliable than denominational membership. Estimates are insensitive to an additional test: the successive dropping of each control variable from the analysis.

123

380

Polit Behav (2008) 30:367389

administered close to presidential elections, while the middle one is collected in a mid-term election. One obvious strategy is to specify the effects as continuous across time (Finkel 1995, p. 16). It seems plausible to hypothesize that the effect of religiosity on political variables takes place evenly distributed across time, and not during some arbitrarily dened time intervals, for instance, every two or four years. Yet, this apparently reasonable assumption expects too much on behalf of citizens. The idea that voters actively follow political affairs, and continuously update their views in meaningful ways loses most of its plausibility against empirical reality (Zaller 1992). An alternative choice is to model discrete time-lags, with the expectation that citizens tend to reassess the link between religion and politics only in specic temporal contexts. Election campaigns appear to be ideal as such. In Edelmans words elections are rituals and draw attention to common social ties (1964, p. 3). During these periods, candidate speeches, advertisements and every day discussions bring the component parts of the political process to the forefront; voters become actively engaged, while ideological and partisan identities obtain increased salience (Campbell et al. 1966; Clarke and Stewart 1998). One practical implication is that it is reasonable to suggest that the political religion phenomenon is an evaluation activated by specic situations during specic periods, and does not take place constantly across time. Third, the focus on presidential elections is not only necessary for pragmatic reasons, in particular the absence of mid-term measurement for church attendance in the 19921996 data, but it is also supported for substantive reasons. In justifying the exclusion of the mid-term panel wave from the analysis, I argue that citizens re-evaluate the relationship between politics and religion mainly in the abstract context of presidential elections and not in the more specic, local context of mid-terms. Differences between presidential and congressional electoral settings support this decision (e.g. Campbell 1960; Davidson and Oleszek 2004). For the above reasons, the same lag (two consecutive presidential elections) was selected for all panels. The two-wave specication with observed indicators assumes that indicators are perfect measures of the underlying concepts and prohibits correction for random measurement error. The addition of a measurement model would minimize noise in the variables, but requires either three-wave panel data (for single indicators) or multi-item constructs (Finkel 1995). Still, even without a threewave specication and correction for measurement error, the subjectivehence less reliableideological and party identication variables exert a non-vanishing inuence on religious behavior.8

In an earlier stage of the analysis, I used three-wave models with the 19721976 and 20002004 datasets, which employed a Wiley-Wiley specication with single-indicator latent variables. These produced a similar picture of causal relationships as the one presented in this study. However, in some cases AMOS returned inadmissible gures, such as negative variances. I attribute this to small and homogenous subsamples, which possibly hindered the efcient estimation of complex models with latent variables.

123

Polit Behav (2008) 30:367389

381

Results The analysis begins with a test of Hout and Fischers thesis (2002), adjusted to ANES panel data. This predicts changes in religious practice due to ideological concerns, focusing on the link between political ideology and church attendance. The psychological groups of interest identied by SIT are Liberals versus Conservatives. Which competing assumptionreciprocity or unidirectionality, i.e. an ideological political religion or the sociological modelbest approximates the causal patterns observed in the data? Instead of visually comparing the t of the feedback hypothesis against the t of the unidirectional hypothesis, Table 2 summarizes results from the v2 difference test.9 This test directly compares the feedback hypothesis of ideologically driven attendance against the more traditional, unidirectional expectation of religious effects on political variables. Highlighted entries show which part of the religious population experiences the political religion phenomenon in each period. In both periods, the religious community that practices religiosity partly as political behavior is the evangelical group. I consider this a product of the entrance of religious populations into ideological camps, especially the conservative ideological mobilization of evangelicals by the Social Issue since the early 1970s. In this case, members identifying with opposing ideological groups tend to adjust their religious exposure accordingly, with liberals avoiding ideologically conservative churches and vice versa. Table 3 presents in detail the reciprocal models identied above and the coefcients connecting ideology and attendance (a seven- and ve-point scale, respectively). In the 1970s, it seems that apart from the impact of lagged evangelical attendance on changing ideological orientation, we also observe a signicant political religion effect over and above the inuence of control variables and lagged attendance. In this case, lagged ideology shapes evangelical attendance, with the more conservative tending to increasingly attend evangelical churches and vice versa for the more liberal (b = .162). On average, extreme conservatives (seven on the ideology scale) increase their attendance by .97 points compared to extreme liberals (one on the ideology scale), i.e. by almost one point on the attendance scale. We witness a stronger occurrence of the same effect in the 1990s (b = .274), whereby extreme conservatives increase their attendance by 1.64 scale points compared to extreme liberals. The discussion now turns to the posited occurrence of a partisan political religion, i.e. the shaping of church attendance by party identication. Table 4 summarizes results from the v2 difference test, this time using the partisanship variable in place of ideological self-placement. The psychological groups of interest identied by SIT in
9 Goodness-of-t is assessed with the following criteria (Arbuckle 2005): the v2 test/degrees of freedom ratio, in which values less than 5 are desirable (or 3 for stricter evaluations); Bollens incremental t index (IFI), which makes adjustments for sample size and the complexity of the model, taking into account degrees of freedom, should score close to 90 and above (or .95 and above for more conservative evaluations); Bentlers comparative t index (CFI), which again accounts for small sample sizes and should be greater than .90 (or .95 for stricter evaluations); nally, the root mean square error of approximation (RMSEA), where values should be lower than .08 (or .05 for more conservative evaluations). All models testedfeedback and unidirectional, for all groups, during all periodshad an acceptable t to the variances/covariances encountered in the data.

123

382

Polit Behav (2008) 30:367389

Table 2 (Ideological) Political religion: feedback versus unidirectional effects Model comparison Dv2 19721976 ANES Catholic Mainline Evangelical 19921996 ANES Catholic Mainline Evangelical Sources: 19721976, 19921996 ANES Note: Feedback models with a better t than unidirectional models are highlighted. The v2 difference statistic compares nested models: a feedback model against a constrained unidirectional model, which only suggests religious effects on politics. A signicant v2 statistic indicates that the constrained model (unidirectional effects) is highly unlikely to be valid. In regression analysis terminology, this suggests that the effect of lagged ideology on attendance is signicant. When the statistic is insignicant, the same effect is probably a misspecication. Finally, the 20002004 panel study does not contain a repeated measure for ideological self-placement in 2004 * p \ .10; ** p \ .05 .138 2.577 4.506** .009 .305 9.639** Ddf 1 1 1 1 1 1

Table 3 (Ideological) Political religion in recent election cycles Evangelical (19721976) Stabilities (Time 1 ? Time 2) Ideology ? Ideology Attendance ? Attendance Cross-lags (Time 1 ? Time 2) Ideology ? Attendance Attendance ? Ideology Summary statistics N v2/df IFI, CFI, RMSEA 244 91.201/47 (1.941, p = .000) .961, .957, .062 106 58.047/47 (1.235, p = .130) .985, .982, .047 .162 (.113/.076)** .092 (.119/.046)** .274 (.265/.086)*** .205 (.239/.062)*** .245 (.220/.068)***a .608 (.613/.051)*** .571 (.642/.078)***a .707 (.705/.069)*** Evangelical (19921996)

Sources: 19721976, 19921996 ANES panels Note: Maximum likelihood estimates. Parentheses contain standardized coefcients/standard errors. Controls included for age, male dummy, education dummies (reference category is high school or less), union member in household dummy, marital status dummies (never married is baseline), region dummies (northeast is baseline), children in household dummy, and income. Estimates for disturbance correlations, unanalyzed relationships, and control effects are omitted for clarity
a

Ideological stability seems to be articially low due to the use of the root question; this decision ignores leaners, hence introduces an amount of variability. What remains is that relative to attendance, ideology still exerts a signicant pressure ceteris paribus.

* p \ .10; ** p \ .05; *** p \ .01

123

Polit Behav (2008) 30:367389 Table 4 (Partisan) Political religion: feedback versus unidirectional effects Model comparison Dv2 19721976 ANES Catholic Mainline Evangelical 19921996 ANES Catholic Mainline Evangelical 20002004 ANES Catholic Mainline Evangelical Sources: 19721976, 19921996, 20002004 ANES panels .171 7.608** .118 .354 .575 4.395** .412 .053 3.719 **

383

Ddf 1 1 1 1 1 1 1 1 1

Note: Feedback models with a better t than unidirectional models are highlighted. The v2 difference statistic compares nested models: a feedback model against a constrained unidirectional model, which only suggests religious effects on politics. A signicant v2 statistic indicates that the constrained model (unidirectional effects) is highly unlikely to be valid. In regression analysis terminology, this suggests that the effect of lagged partisanship on attendance is signicant. When the statistic is insignicant, the same effect is probably a misspecication * p \ .10; ** p \ .05

this case are dened as Democrats versus Republicans. Highlighted entries indicate which population experienced the political religion phenomenon in each period. Once again, attendance seems to be inuenced by political (partisan) concerns among evangelical Protestants in the 1990s and 2000s, yet the feedback effect is not present for this constituency in the 1970s. I attribute this to the absence of partisan polarization during that period. That is, the 1970s were still a time when evangelical support for the GOP was not yet salient, and the Christian Right had not emerged as a major force in American elections. Consistently therefore with the context-dependent nature of SIT expectations, evangelicals do not appear to experience partisan pressures on their attendance prior to the particization of their churches.10
10 According to the test in Table 4, a partisan religion also emerges among mainline Protestants in the 1970s, whereby partisanship appears to constrain church attendance. One explanation of this could lie with the political mobilization of mainline churches during the turbulent 1960s and 1970s, as dened by the Civil Rights movement and the Vietnam War (e.g. Wald and Calhoun-Brown 2007). Yet, due to methodological problems in model estimation for this subsample in the 1970s, I choose instead to emphasize the absence of the political religion effect from the evangelical subsample, which is the focus of the present study. Specically, the cross-effects have different signs for mainline Protestants in the 19721976 model: lagged partisanship has a positive effect on changing attendance, while lagged attendance has a counter-intuitive, negative effect on changing partisanship (see Supplementary electronic material, Table C). This is a suppressor effect, whereby two variables are positively correlated, but direct effects are negative or vice versa. Suppression occurs for two main reasons (Smith et al. 1992). First, suppression happens when an additional predictor is entered in the model, i.e. the true relationship between the variables is in the opposite direction than the one indicated by their correlation. If the difference in sign can be explained theoretically, the effect can be retained. Second, suppression can be the result of multicollinearity, an inherent problem in cross-lagged models, which use repeated measures of often very stables variables. The problem is aggravated here by the stratication of the sample into homogenous subsamples. I nd the second explanation more plausible. This justies my concern in reading too much into this result.

123

384 Table 5 (Partisan) Political religion in recent election cycles Evangelical (19921996) Stabilities (Time 1 ? Time 2) Party ID ? Party ID Attendance ? Attendance Cross-lags (Time 1 ? Time 2) Party ID ? Attendance Attendance ? Party ID Summary statistics N v2/df IFI, CFI, RMSEA 119 65.233/47 (1.388, p = .040) .976, .973, .057 .110** (.141/.052) .156** (.121/.072) .784*** (.786/.064) .791*** (.789/.059)

Polit Behav (2008) 30:367389

Evangelical (20002004)

.903*** (.817/.045) .738*** (.752/.049) .079** (.096/.041) .247*** (.187/.054) 200 98.046/47 (2.086, p = .000) .960, .957, .074

Source: 19921996, 20002004 ANES panels Note: Maximum likelihood estimates. Parentheses contain standardized coefcients/standard errors. Controls included for age, male dummy, education dummies (reference category is high school or less), union member in household dummy, marital status dummies (never married is baseline), region dummies (northeast is baseline), children in household dummy, and income. Estimates for disturbance correlations, unanalyzed relationships, and control effects are omitted for clarity * p \ .10; ** p \ .05; *** p \ .01

Table 5 presents the actual estimates connecting the main variables within the reciprocal models. Evangelical Protestants go through a round of intense religious politicization in the periods covered by the 19921996 and the 20002004 panels.11 The conventionally assumed impact of lagged evangelical attendance on changing partisanship is both substantially and statistically signicant in both periods (b = .156 and b = .247, respectively). The substantial interpretation of this result is that in a period of increasing connections between the GOP and conservative Protestantism, frequent exposure to evangelical churches seems to move congregants closer to the Republican Party. Conversely, less frequent exposure to evangelical churches seems to move congregants closer to the Democrats. However, lagged partisanship also signicantly affects attendance in both eras (b = .110 and b = .079, respectively). On average in the 1990s, strong Republicans (six on the partisanship scale) increase their attendance to evangelical churches by .66 points compared to strong Democrats (zero on the scale). In the 20002004 case, strong Republicans increase their attendance to evangelical churches by .47 points compared to strong Democrats. As expected therefore, ceteris paribus, the more Republican among evangelicals were driven towards participating more frequently in their churches, while the more Democratic were motivated to participate less. Results support a Republican pull behind increasing attendance in evangelical churches (or a Democratic push away from the same churches). Importantly, this

11 Note that the detection of signicant effects especially post-1990s obtains greater leverage due to small subsample sizes (higher probability for a Type II Error).

123

Polit Behav (2008) 30:367389

385

phenomenon emerges within the constituency that has been at the forefront in a polarized partisan environment.12 By and large, the causal argument and ndings presented above suggest that when the link connecting a religious group with an ideological camp or political party becomes institutionalized in the public mind, religious behavior in that group can be partly shaped by a political calculus. This effect is particularly evident for the most salient social group in both the conservative backlash to the social upheaval of the 1960s and 1970s, and later in the Republican electoral base: evangelical Protestants. I consider the emergence of the ideological effect before the emergence of the partisan effect as a robust basis for this conclusion. As a new issue conict emerged in the political scene of the 1970s based on traditional social values, ideological political religion was the case for evangelicals in that period (see the effect of ideological demarcation in Table 2). Later on, as party elites responded and adapted to this new division, partisan political religion appears to have followed (see the effect of partisan demarcation in Table 4) (cf. Carmines and Stimson 1989; Layman 2001; Wald and Calhoun-Brown 2007). The intense linking of a church with a political ideology or party produces an autonomous outcome, which could eventually alter the composition of the religious constituency. Even though available datasets can only indirectly support this conclusion, it seems that if, for instance, Democrats reduce their attendance to evangelical churches, and Republicans increase it, political homogeneity within these churches may not be a mere effect of theology or demography.

Discussion The rising voice of evangelical Protestants, mobilized by abortion and other social issues in the 1970s, and later expressed by the GOP, is considered today an important component of the American political system. The present analysis contributes an alternative theoretical and empirical explanation of this phenomenon, based on the political reinforcement of religious patterns in American politics. The alternative proposed and tested moves beyond reductionist sociostructural accounts of the cleavage phenomenon, which merely focus on political consequences of religious factors. Expanding on Hout and Fischers (2002) empirical demonstration of how religious communities become affected by their vocal participation in political conicts, I provide more robust evidence of this transformation, in terms of causal inference. Results shed light on the role of ideological and partisan group identication in generating changes in the religious behavior of evangelical Protestants. In a process more or less ignored by prior research, which insists on
12 To test whether interactions are signicant, i.e. whether effects differ across religious communities, I t two models for each panel: rst I allow all parameters to differ across groups. In the second case, I constrain cross-lagged effects to be equal across groups. The relative validity of the two assumptions can be evaluated with a v2 difference test, i.e. the difference in v2 values between the two hypotheses. This will indicate which model ts the data better (Bollen 1989, p. 292). Results show that the intergroup differences observed in Tables 2 and 4 are signicant.

123

386

Polit Behav (2008) 30:367389

investigating the conventional effect of social structure on politics, this analysis suggests that ideology and partisanship may be able to construct religious communities, by boosting movement within and perhaps eventually across these communities. Certainly, the results presented here are only a preliminary attempt to address an ignored social psychological process. Findings are not without methodological difculties, since the designing teams of ANES surveys have worked under the assumption that religion affects politics and not vice versa. More ne-tuned survey questions coupled with contextual and other in-depth analyses should delve deeper into the reinforcing connection between religiosity and political variables. One suggestion in that direction would be the consistent repeated measurement of denominational membership for the same individuals across time, which can assess if this indicator is also subject to ideological and partisan inuences. Such an analysis could establish if changing attendance within politicized churches eventually leads to apostasy or switching. Further research could also help understand the effect of this phenomenon on voting choice. This step can evaluate whether political participation and volatility in voting intentions varies between believers who are pulled closer to church or pushed further away from it due to political concerns. Bearing these weaknesses in mind, what does the analysis of panel data tell us about religious politicization? The empirical picture suggests that since the 1970s, the ideological and partisan mobilization of evangelicals has pulled some members closer to the church (especially conservatives and Republicans), while pushing others away (liberals and Democrats). Worshiping in a theologically conservative church seems to eventually functionat least in partas a symbolic expression of conservatism and Republican partisanship, whereby, other things being equal, conservatives and Republicans tend to attend church because they see this practice as conrmation of ideology and partisanship and as demarcation from the out-group (liberals and Democrats). One can only speculate at this point, but the long-term consequence of this political religion could lead to an ideological and partisan sorting-out within politicized churches, a process that should reinforce the trend described by Fig. 1. Finally, apart from offering a less simplied portrayal of religious politicization, the political religion phenomenon has substantial implications for our understanding of religion in modern society and the continuing importance of partisan attachments in contemporary American politics. In relation to religion, it appears that American piety is partly shaped by worldly components. By focusing on surface indicators, research usually ignores this phenomenon and instead concludes that secularization is evidently not taking place in the United States, with its high religious observance and growing religious mobilization in recent political periods (Leege and Kellstedt 1993). Even so, this article reinforces the possibility that the basis of Christian faith in America is exposed to secular concerns (cf. Bellah 1967). Regarding parties, the set of empirical ndings presented opens up the potential of treating some religious phenomena as politically endogenous, exposed to the pressures of a partisan environment. This is to say that partisan social identities may not be consequential purely for political processes (Campbell et al. 1960; Green

123

Polit Behav (2008) 30:367389

387

et al. 2002; Greene 2004), but that they may affect non-political aspects of social life as well. In the end, contradicting the proposition that parties are in decline (Nie et al. 1979; Aldrich 1995), and that partisanship is merely shorthand for policy preferences and short-term evaluations (Fiorina 1981), the existence of a political religion supports a view of party identication as a major component of mass belief systems.
Acknowledgments Earlier versions of the paper were presented at the Joint Sessions of the European Consortium for Political Research (Workshop: Politicizing Socio-Cultural Structures), Helsinki, May 2007, and the annual meeting of the Society for the Scientic Study of Religion, Tampa, November 2007. I am indebted to Mark Shephard, John Curtice, Robert Johns, the editors and two anonymous reviewers for advice and comments. I also want to thank Christopher J. Carman and Wouter van der Brug for helpful suggestions.

References
Abramowitz, A., & Saunders, K. (1998). Ideological realignment in the U.S. electorate. Journal of Politics, 60, 634652. Aldrich, J. H. (1995). Why parties? The origin and transformation of party politics in America. Chicago: University of Chicago Press. Arbuckle, J. (2005). Amos 6.0 users guide. Chicago, IL: SmallWaters Corporation. Asher, H. (1983). Causal modeling. Newbury Park, CA: Sage. Bartels, L. M. (1999). Panel effects in the American national election studies. Political Analysis, 8, 120. Bellah, R. (1967). Civil religion in America. Daedalus, 96, 121. Bolzendahl, C., & Brooks, C. (2005). Polarization, secularization or differences as usual? The denominational cleavage in U.S. social attitudes since the 1970s. Sociological Quarterly, 46, 4778. Bollen, K. (1989). Structural equations with latent variables. New York: Wiley. Campbell, A. (1960). Surge and decline: A study of electoral change. Public Opinion Quarterly, 24, 397418. Campbell, A., Converse, P., Miller, W., & Stokes, D. (1960). The American voter. New York: Wiley. Campbell, A., Converse, P., Miller, W., & Stokes, D. (1966). Elections & the political order. NY: Wiley. Carmines, E., & Stimson, J. (1989). Issue evolution. Princeton: Princeton UP. Carsey, T., & Layman, G. (2006). Changing sides or changing minds? Party identication and policy preferences in the American electorate. American Journal of Political Science, 50, 464477. Clarke, H., & Stewart, M. (1998). The decline of parties in the minds of citizens. Annual Review of Political Science, 1, 357378. Davidson, R., & Oleszek, W. (2004). Congress and its members (9th ed.). Washington: CQ Press. Deaux, K, Reid, A., Mizrahi, K., & Ethier K. (1995). Parameters of social identity. Journal of Personality and Social Psychology, 68, 280291. Edelman, M. (1964). The symbolic uses of politics. Urbana: University of Illinois Press. Finke, R., & Stark, R. (1992). The churching of America, 17761990: Winners and losers in our religious economy. New Brunswick: Rutgers UP. Finkel, S. (1995). Causal analysis with panel data. Thousand Oaks, CA: Sage. Fiorina, M. (1981). Retrospective voting in American national elections. Yale UP. Goren, P. (2005). Party identication and core political values. American Journal of Political Science, 49, 881896. Green, D., & Palmquist, B. (1990). Of artifacts and partisan instability. American Journal of Political Science, 34, 872902. Green, D., Palmquist B., & Schickler, E. (2002). Partisan hearts and minds: Political parties and the social identities of voters. New Haven: Yale UP. Green, J. C., Rozell, M., & Wilcox, C. (Eds.). (2003). The Christian right in American politics: marching to the millennium. Washington: Georgetown UP. Greene, S. (1999). Understanding party identication: A social identity approach. Political Psychology, 20, 393403.

123

388

Polit Behav (2008) 30:367389

Greene, S. (2002). The social-psychological measurement of partisanship. Political Behavior, 24, 171197. Greene, S. (2004). Social identity theory and party identication. Social Science Quarterly, 85, 136153. Guth, J. L., Green, J. C., Smidt, C., Kellstedt, L., & Poloma, M. (1997). The bully pulpit: The politics of Protestant clergy. Lawrence: University of Kansas Press. Hadaway, K., Marler, P. L., & Chaves, M. (1993). What the polls dont show: A closer look at church attendance. American Sociological Review, 58, 741752. Hogg, M. A., & Terry, D. J. (2000). Social identity and self-categorization processes in organizational contexts. Academy of Management Review, 25, 121140. Hout, M., & Fischer, C. (2002). Why more Americans have no religious preference. American Sociological Review, 67, 165190. Huddy, L. (2003). Group identity and political cohesion. In D. O. Sears, L. Huddy, & R. Jervis (Eds.), Oxford handbook of political psychology. Oxford: Oxford UP. Jackson, J. E. (1975). Issues, party choices and presidential votes. American Journal of Political Science, 19, 161186. Kline, R. B. (1998). Principles and practice of structural equation modeling. NY: Guilford. Kriesi, H. (1998). The transformation of cleavage politics. The 1997 Stein Rokkan Lecture. European Journal of Political Research, 33, 165185. Lau, R. (1989). Individual and contextual inuences on group identication. Social Psychology Quarterly, 52, 220231. Layman, G. (1997). Religion and political behavior in the United States: The impact of beliefs, afliations, and commitment from 1980 to 1994. The Public Opinion Quarterly, 61, 288316. Layman, G. (2001). The great divide: Religious and cultural conict in American party politics. New York: Columbia UP. Layman, G., & Carsey, T. (2002). Party polarization and party structuring of policy attitudes: A comparison of three NES panel studies. Political Behavior, 24, 199236. Layman, G., & Green, J. C. (2005). Wars and rumours of wars: The contexts of cultural conict in American political behaviour. British Journal of Political Science, 36, 6189. Layman, G., & Hussey, L. (2005). George W. Bush and the Evangelicals. Presented at the Conference on A Matter of Faith? Religion and the 2004 Presidential Election, University of Notre Dame. Leege, D. C., & Kellstedt, L. A. (Eds.). (1993). Rediscovering the religious factor in American politics. Armonk, N.Y.: M.E. Sharpe. Lenski, G. (1963). The Religious Factor. NY: Doubleday. Levendusky, M. (2005). Sorting in the U.S. mass electorate. Presented at the MPSA Annual Meeting. Lipset, S. M., & Rokkan, S. (1967). Cleavage structures, party systems, and voter alignments: An introduction. In S. M. Lipset, & S. Rokkan (Eds.), Party systems and voter alignments. NY: Free Press. Long, K. M., & Spears, R. (1997). The self-esteem hypothesis revisited: Differentiation and the disaffected. In R. Spears, P. Oakes, N. Ellemers, & S. Haslam (Eds.), The social psychology of stereotyping and group life. Oxford: Blackwell. Loveland, M. T. (2003). Religious switching: Preference development, maintenance, and change. Journal for the Scientic Study of Religion, 42, 147157. Manza, J., & Brooks, C. (1999). Social cleavages and political change. Oxford: Oxford UP. Markus, G, & Converse, P. (1979). A dynamic simultaneous equation model of electoral choice. American Political Science Review, 73, 10551070. Miller, W. E., & Shanks, J. M. (1996). The new American voter. Cambridge: Harvard UP. Moen, M. C. (1994). From revolution to evolution: The changing nature of the Christian right. Sociology of Religion, 55, 345357. Newport, F. (1979). The religious switcher in the US. American Sociological Review, 44, 528552. Nie, N., Verba, S., & Petrocik, J. (1979). The changing American voter. Cambridge: Harvard UP. Page, B., & Jones, C. (1979). Reciprocal effects of policy preferences, party loyalties, and the vote. American Political Science Review, 73, 10711090. Poole, K. T., & Rosenthal, H. (1984). The polarization of American politics. Journal of Politics, 46, 10611079. Roof, W. C., & McKinney, W. (1987). American mainline religion: Its changing shape and future. New Brunswick, N.J.: Rutgers UP. Sartori, G. (1969). From the sociology of politics to political sociology. In S. M. Lipset (Ed.), Politics and the social sciences. NY: Oxford UP.

123

Polit Behav (2008) 30:367389

389

Smith, R., Ager, J., & Williams, D. (1992). Suppressor variables in multiple regression/correlation. Educational and Psychological Measurement, 52, 1729. Steensland, B., Park, J., Regnerus, M., Robinson, L., Wilcox, W. B., & Woodberry, R. (2000). The measure of American religion: Toward improving the state of the art. Social Forces, 79, 291318. Tajfel, H. (1981). Human groups and social categories. Cambridge: Cambridge UP. Turner, J. C. (1985). Social categorization and the self-concept: A social cognitive theory of group behavior. In E. Lawler (Ed.), Advances in group processes: Theory and research Vol. 2. Greenwich, CT: JAI Press. Wald, K., & Calhoun-Brown, A. (2007). Religion and politics in the United States (5th ed.). Lanham, MD: Rowman & Littleeld. Warner, R. S. (1993). Work in progress toward a new paradigm for the sociological study of religion in the United States. American Journal of Sociology, 98, 10441093. Wilcox, C. (1996). Onward Christian soldiers: The religious right in American politics. Boulder, CO: Westview. Zaller, J. (1992). The nature and origins of mass opinion. New York: Cambridge UP.

123