Politics, Religion, and Society: Is the United States Experiencing a Period of Religious-Political Polarization?

This study investig ates th e effect of religious id entity o n U.S. Presi dential voter choice in o rder to determin e whether th is relationship chan ged over time. Th e research literature i s divided on t his question with sev eral investigators finding a positive trend in religious-political p olarization sin ce 19 80, and others fi nding no polarization. Th e stud y further addresses a pu tative li nk between so cial in equality and relig ious politics b y identifying the race, cl ass, and ge nder location of religiously influenced voters, using m ultiple cross sections from t he General S ocial Survey t o em pirically model P residential voting over t he period 1980 t o 2 008. T he findings demonstrate th at religious id entity in fluenced voter cho ice, and t hat th is influ ence in creased significantly and substantially across the study period. Second, that upper class whites are the source of religious partisan polari ty, and uppe r class whites beca me more po larized ove r t he period 1980 to 2008. The effect of gender on partisanship is less p ronounced, and overshadowed by social class and religious identity. The study findings demonstrate that religiously influenced Presidential voting reflects the political behavior of a relatively privileged component of the electorate.


Introduction
Social science research indicates that religious identity is a driver of political behavior within the United States.This finding is broadly confirmed, for example in empirical investigations of Presidential voter choice conducted since World War II (Converse, 1964;Greeley & Hout, 2006;Knoke, 1974;Manza & Brooks, 1997;Regenerus, Sikkink, & Smith, 1999;Sherkat & Ellison, 1999).There are multiple interpretations of this empirical relationship, perhaps because it is not obvious why, or how, voter choice is subject to religious identity, the former defined by the here and now, the latter by the hereafter (Billings & Scott, 1994;Woodberry & Smith, 1998).It is also the case that prominent European founders of social science predicted the demise of religion in favor of secular worldviews.Because comparative analysis indicates that religious beliefs and practices are more pervasive within the United States relative to most European nations (Inglehart & Baker, 2000), it is tempting to view Weber's (2004) and Durkheim's (2001) predictions of secularization as a product of their European experience.At least for the case of U.S. Presidential voter choice, the influence of religious identity appears not to have diminished.
In this study we investigate the trend line in the affect of religious identity on U.S. Presidential voter choice in order to determine whether this relationship has, in fact, changed over time.The research literature is divided on this question with several investigators finding a positive trend in religious-political polarization since 1980 (Gelman, Park, Shor, & Cortina, 2010;Fiorina, Abrams, & Pope, 2011:134;Layman, 2001;1997), and others finding no polarization (Brooks & Manza, 2004;Manza & Brooks, 1997).This task is both empirical and interpretive because an empirical trend is unimportant in the absence of substantive interpretation.However in fashioning an interpretation we are confronted by difficulties related to the broad import of a research question that touches upon several areas of social science inquiry including political science, political sociology, religious studies, the sociology of religion, and social stratification.The latter is pertinent because the period in question is defined by widening economic inequality, hence a computed trend line should allow for economic differences, given that these differences were changing across the study period.Research that spans multiple areas of social science inquiry is demanding because each area has its own preference set regarding research procedures, including differing approaches to measuring religious identity.
This study builds upon a model proposed by Hirschl, Booth, & Glenna (2009) (HBG) that defines religious identity effects on voter choice as contingent upon the individual's location within the stratification order.In this model the empirical link between religious identity and voter choice is allowed to vary independently by race, class, and gender.We view this as an appropriate framework, given the economic inequality trend mentioned in the prior paragraph, and the likelihood that individuals in different positions in the stratification order are exposed to differential economic change that could, at least potentially, influence voter choice.In this study the HBG model is tested in two new dimensions.First, the model's proposed measure of religious identity is tested for validity across three independent surveys.This was done because we are skeptical of the HBG measure, namely that it may lack a valid axis of interpretation among voters.Second, we extend the time horizon of the HBG model to facilitate a test for change over time.The present study comprises the 1980 to 2008 Presidential elections, and includes a test for time trends; the HBG analysis covered the 1980 to 2000 Presidential elections, and did not test for change over time.

Economic Inequality, Religion and Politics since 1970
If one reviews U.S. inequality indicators over the past four decades, it is apparent that inequality increases are substantial and sustained.The Gini coefficient of income inequality increased approximately 25 percent between 1970and 2010, from .35 to .44 (U.S. Census Bureau, 2011).The national income share of the top ten percent of families increased from 33 percent in 1972 to 50 percent in 2007 (Atkinson, Piketty, & Saez, 2011).And the national share of financial wealth owned by the richest 20 percent of families remained high, increasing slightly from 91 percent in 1983 to 93 percent in 2007 (Wolff, 2010).These trend lines were unanticipated by many social scientists, who since Kuznets (1955), expected that inequality would decrease, or remain stable, within highly developed countries such as the United States.
Various attempts have been made to link religiously influenced political activity to the failure of social policy and politics to reduce economic inequality.For example in an analysis of economic inequality since 1960, Hacker & Pierson (2010:179) argue that "the rise of Christian conservatism" partially explains the absence of pro-equality political response to economic inequality; however, these authors present no empirical evidence in support of this claim.Additionally there is a popular notion that middle and working class support for social equality has been attenuated by a political focus on non-economic "culture-war" issues, such as prayer in public schools and legalized abortion (Frank, 2004).However, in an analysis of the 1984 to 2004 National Election Study of Presidential Elections, Bartels (2005) found evidence that the middle and working classes continue to support social equality, defined as government spending, government jobs, and aid for minorities.Likewise the notion that Americans are engaged in a "culture war" (Hunter, 1991) pitting secularists and religious liberals against religious traditionalists has been judged to be lacking in empirical specificity and validity (Williams, 1997).
Because religious identity has been shown to have robust affects on voter choice, a focus on this relationship offers a measurable way to calibrate the relation of religion to politics.One way to proceed is not to ask if religious voting is obscuring economic inequality, blocking pro-equality politics, or contributing to a culture war, but rather to examine its distribution across the stratification order.Is religiously influenced voting more prominent within the upper classes, or within the lower classes, and how is this distribution changing over time?The stratification location of religiously influenced voters begins to define the context for questions about religion and economic inequality.If the poorest, least affluent voters are also the strongest religious voters, then this lends credence to the interpretation that economic inequality losers are interpreting their political options in religious terms.If, on the other hand, the most privileged voters are disproportionately the strongest religious voters, then this suggests something quite different, that the relatively privileged interpret their political options in religious terms.To identify these empirical relationships we measure religious identity effects on voter choice, and compare the density of this distribution across differing locations within the stratification order.We additionally measure trend lines in these differences: are these differences increasing, decreasing, or not changing?These elements define the empirical linkage between voter choice, religious identity, and the stratification order.
Measures of social class alone are insufficient indicators of the stratification order because of the connection between race and class established by U.S. history.In addition, the intersection of race, religion, and voter choice presents a puzzle that is of interest to researchers (Greeley & Hout, 2006: 72).We also include gender because it has been shown to be a critical dimension of the stratification order.Thus we endeavor to link religious identify effects on voter choice to position within the stratification order, defined by race, class, and gender.

Current Knowledge about Religious Identity Effects on Voter Choice
There are two social science theories linking voter choice to religious identity: ethnoreligious theory, and religious restructuring theory.Ethnoreligious theory postulates a set of linkages between religious traditions, ethnicity, and partisan politics that develop across time (Jensen, 1971;Kleppner, 1979).Although researchers acknowledge that net partisan choice varies marginally from election to election, the underlying linkages are theorized to remain intact.In this approach causality is empirically located at the level of church membership defined by ethnic and religious traditions that causally link to voter choice.Manza & Brooks (1997, 2004) are exemplars of ethnoreligious theory, and deploy empirical variants using the National Election Study within two analyses of Presidential voter choice, one for elections spanning 1960 to 1992, and another for the 1972 to 2000 elections.In both analyses they found high, but relatively stable levels of religious cleavage, confirming the ethnoreligious proposition regarding civil society linkages between religious traditions and partisan politics.In particular they found little partisan change among Catholics, Jews or conservative ("evangelical") Protestants.Their analysis identifies moderate partisan change among "mainline Protestants," who moved from the Republican to the Democratic Party.Thus to the extent religious-political polarization was found during these two time periods, it was in the opposite direction perceived by many political analysts and pundits who characterize religious political change in terms of conservative Protestants leaving the Democratic Party for the Republican Party (Philips, 2006).
The alternative theoretical model is religious restructuring theory that operates at the level of individual belief, and presumes that major religious traditions/denominations are split between individuals adhering to traditional religious authority, versus individuals adhering to relatively secular, or "progressive" authority (Wuthnow, 1988;Hunter, 1997).In a study of Presidential voter choice over the period 1980 to 1994, also using the National Election Study, Layman (1997) found increasing religious-political polarization using two variables derived from religious restructuring theory: "doctrinal orthodoxy" and "religious commitment."Layman measures doctrinal orthodoxy with a scale combining responses to questions about being "born again" with responses to a biblical authority question; religious commitment is constructed from questions about frequency of church attendance, and from responses to questions about religious salience.Over the study period the effect of these two constructs on voter choice increased, and the implication is that, over time, religiously influenced voters became more likely to vote Republican, whereas more secular individuals became more likely to vote Democratic.Consistent with Manza andBrooks (1997, 2004), Layman found no change over time with regard to the affect of religious tradition on voter choice.Comparing the Layman results to the Manza and Brooks results suggests that the two theories of religion and politics identify different trends and relationships in religious voting.
A study by Fiorina, Abrams, & Pope (2011) confirms a polar connection between religion, measured as church attendance, and voter choice between 1992 and 2004, using the National Election Study.Over this period individuals frequently attending church became stronger Republican voters, and individuals who attended infrequently, or not at all, became stronger Democratic voters.In interpreting their study results, Fiorina, Abrams, and Pope cite a related analysis by Bolce & DeMaio (1999) of a National Election Pilot Study that reports polarization between Christian fundamentalists and secular voters, specifically that "negative feelings toward Christian fundamentalists are a significant predictor of relative party assessment" (ibid:508).To the extent this study has validity, it suggests that religious polarization is gaining traction within the political sphere.Some researchers reason that the two theories describe different aspects of the same process, and combine the two theories to achieve a more comprehensive analysis.For example, Guth, Kellstedt, Smidt, & Green (2006) blend ethnoreligious theory and restructuring theory in an analysis of the 2004 Presidential Election, creating hybrid empirical measures of religious identity.Guth et al.'s (2006) analysis is based on a national survey of 4,000 respondents, interviewed before and after the 2004 election, and the analysis found that voter choice is predicted by hybrid categories derived from both theories.Thus for example, there is a voter choice gradient across "traditional," "centrist," and "modernist" individuals within the "mainline Protestant" tradition, and this gradient has a similar slope, but different mean levels, within the more Republican-leaning "evangelical Protestant" tradition (Guth et al., 2006: 228).These results are consistent with the proposition that ethnoreligious theory and restructuring theory predict voter choice using different domains of religiosity.Hirschl, Booth & Glenna (2009) (HBG) propose a model that combines elements from both theories.Their model is akin to religious restructuring theory insofar as it emphasizes biblical authority as a key component of religious identity.The emphasis is justified with reference to the Christian Bible's visibility within religious and secular spheres of American society, e.g. its function in legal and civic rituals (Kramnick & Moore, 2005).Second, their model of religious identity implies that the meaning of biblical authority varies by religious tradition, in particular between Catholics and Protestants that have differing institutional and historical legacies with regard to religious authority (Riesebrodt, 1993;Weber, 1958).Thus HBG deploy the variable "religious tradition" in combination with biblical authority.
The HBG empirical approach is inductive statistical modeling where voter choice is an unrestrained function of biblical authority, religious tradition, and stratification categories; inclusion of the stratification categories is further justified with reference to Durkheim (Hirschl, Booth, & Glenna, 2009: 929).Because Durkheim proposes that religious sentiments and rituals originate within the realm of the individual's material and mental dependency upon society, religious categories necessarily bear some relationship to economic and social categories.In practice this theory can be operationalized to the extent that a common symbolic system operates across secular and sacred spheres, and hence the HBG model depends upon a common axis of perception about biblical authority.For this reason we first assess the character of social interpretation of biblical authority before proceeding to analyze trends in the affect of religious identity on voter choice.

Data and Methods
This study reports analysis of three surveys.First, the General Social Survey (GSS) is a nationally representative, repeat cross-section sample of the English speaking, non-institutional population age 18 and over.The GSS component of the study analyzes voter choice for eight Presidential elections 1980 to 2008, and the data are derived from 20 annual or biannual surveys comprising 11,411 non-Hispanic white respondents, and 1,870 black respondents with complete sets of independent and dependent variables.
Second, the 2009 Cornell National Social Survey (CNSS) is a random sample of 1,000 households within the United States.One household member age 18 years of age and older is interviewed.The survey contains questions about Presidential voter choice in 2008, and repeats GSS survey questions used to construct the independent and dependent variables in the GSS analysis.In addition, it includes a new biblical authority scale designed by the study authors.Third, students taking "Introduction to Sociology" at Cornell University were surveyed during spring, 2010, in order to validate the biblical authority scale from the 2009 CNSS.
The dependent variable for the GSS/CNSS analyses is the binary voting preference for the Republican versus Democratic Presidential candidate.We exclude nonvoters and third party candidates from the analysis in order to focus upon candidate choice for the two major political parties.Other variables on the right hand side include relative family income used to proxy the individual's social class.Although income is not the favored approach for measuring social class, it is widely acknowledged to be implicated in class position (Domhoff, 2002;Perrucci & Wysong, 2008;Wright, 1996).We recode total family income into quartiles where quartile 4 is the top 25 percent, and roughly corresponds to Perrucci and Wysong's "comfort class."In 2005 dollars, the lower boundary for the top family income quartile is $75,000 per annum, and the upper boundary is top-coded at $150,000.The first family income quartile boundaries are zero to $22,499 per annum, second quartile boundaries $22,500 to $39,999, and third quartile boundaries $40,000 to $74,499.
Other variables in the analysis are gender, Southern region defined as categories 5-7 (South Atlantic, East South Central and West South Central) for the GSS variable, REGION, and race (white versus black).In the GSS, blacks are the sole minority group sufficiently large for multivariate statistical analysis.
The GSS/CNSS independent variable "religious identity" is operationalized by cross-classifying two measures: biblical authority and religious tradition.Biblical authority is operationalized by the answer to this multiple-choice question in the GSS (Davis, Smith, & Marsden, 2005: 197): 1. Which of these statements comes closest to describing your feelings about the Bible?a.The Bible is the actual word of God and is to be taken literally, word for word.
b.The Bible is the inspired word of God but not everything in it should be taken literally, word for word.
c.The Bible is an ancient book of fables, legends, history, and moral precepts recorded by men.
It is not necessary that the respondent actually read the Bible, or has direct textual knowledge of it.Rather, the variable measures whether the individual possesses "feelings" that the Bible is the literal word of God, versus the inspired word of God, or a book of fables.From the perspective Durkheim's (2001) theory of religion and society, responses "a" and "b" can be understood as communal projections of authority onto an object, in this case the Christian Bible.Because the Bible functions within secular as well as sacred spheres of American society, its authority is not limited to the sacred, and biblical literalism in particular has been found to be a symbolic resource for political mobilization (Riesebrodt, 2005).
The GSS/CNSS measure of religious tradition is derived from the GSS question, "What is your religious preference?" (Davis, Smith, & Marsden, 2005: 169), and is used to create a variable that corresponds to the major two religious traditions in American society: Protestants, divided between the largest denomination Baptist and other Protestant, Catholic, versus all other religious traditions and non-religious individuals.We tested our empirical models with various schemes for classifying religious tradition, including "mainline Protestant," and "evangelical Protestant," (Steensland et al., 2000), and found no substantive differences in the final results.
In addition to the GSS biblical authority question, participants in the CNSS were asked whether they agreed (yes or no) with the following four statements about the Bible: 2. The Bible should help guide political decisions.
3. The Bible is to be read literally.
4. The Bible is without contradiction.
5. The Bible is an authoritative document which has moral rules I must follow.
Thus, the CNSS data contains five categorical responses relating to respondents' views of biblical authority.These questions represent an attempt to recover information on the link between biblical authority, cognition, and behavior that is more encompassing than GSS Question 1.
As a validation experiment we asked students in "Introduction to Sociology" at Cornell University during the spring of 2010 to rate each of the responses to Question 1 and Statements 2-5, concerning biblical authority (11 responses in total).The students were asked to use a 5-point Likert scale with 1 representing a response indicating the respondent is a "non-Christian/atheist/religious cynic," and 5 indicating an "evangelical Christian/biblical literalist" respondent.

Analysis of Biblical Belief
In order to assess the dimensionality and structure of the CNSS biblical belief responses, we entered the CNSS data into a Multiple Correspondence Analysis (MCA) (Greenacre & Blasius, 2006).Multiple correspondence analysis (MCA) can be thought of as the analog of principle components analysis for categorical variables.The multiple correspondence map is a plot of the responses on the first two principle dimensions, computed from the data structure.This map is given in Figure 1 for the responses to the question, and statements, concerning views about the Bible.The map suggests an ordering of the responses on a 1-dimensional continuous scale, namely the 1 st principle coordinate.Thus, for example, responses "a" and "c" to Question 1 are, respectively, at the positive and negative extremes, while response "b" is in the middle of the scale.A "no" response to Statement 5 is close to the response "c" for Question 1 at the negative end of the scale.On the other hand a "yes" response to Statement 5 is in the positive part of the scale, but some distance from the positive extreme.The results in Figure 1 can be compared to the average ratings given by Cornell students to the eleven responses.
Table 1 shows how the students rated the eleven biblical belief responses, sorted according to average Likert score. 1 This ordering is the same as that based on the MCA 1 st principle coordinate with the exception that the order of responses 4.y and 3.y are switched.However, the scores for these two responses are almost identical both in the CNSS data and in the student survey.In fact, the spacing of the scores from the two datasets is remarkably similar.The main difference is that the students rate response "b" to question 1 closer to the higher (biblical literalist) end of the scale.The statistical results presented in Figure 1 and Table 1 support the conclusion that the belief set and ordering of biblical authority is widely shared and well comprehended within the United States.It is surprising that the CNSS sample responses comprise an item ordering that is closely replicated by a survey of college students.It is further noteworthy that GSS Question 1 encompasses the range of the 11-item scale, with Question 1 responses in the middle, and at both extremes.In order to further compare the two scales we conducted extensive multivariate estimations of a voter choice model using the full 11-item scale, versus multivariate tests using GSS Question 1 responses only, and these estimations are available from the authors upon request.The estimations suggest that both constructs perform in qualitatively similar ways, although the full 11-item construct explains relatively more model deviance.These multivariate results, together with the scale results, suggest that GSS Question 1 is a valid measure of biblical authority, and that biblical authority is widely shared and well comprehended within the United States across a range of belief sets from conservative Christian to secular.

Analysis of Voter Choice
Hirschl , Booth and Glenna (2009) analyzed data from the General Social Survey on voting patterns of white voters in the six presidential elections from 1980 to 2000.Here we extend their analysis to include the 2004 and 2008 elections, and modify their model to include both regional and time effects.
Table 2 gives the results of four different model fits to the 1980-2008 data. 2 The model denoted by HBG is Model 1 from Hirschl, Booth and Glenna (2009).The model HBG+R contains an additional predictor, a dummy indicator that the voter was from the southern US.Specifically, let denote the probability of voting for the Republican candidate (as opposed to the Democratic candidate) for a person of sex ( : 1=male, 2=male), religious tradition ( : 1=Baptist,2=other Protestant,3=Catholic,4=other), in biblical authority category (1=a, 2=b, 3=c), income quartile , from region ( : 1=south, 2=not south), then the model is given by , In model ( 1) is a biblical authority score ( ), is an income quartile score ( ), and is a dummy indicator for Catholics.
The inclusion of a southern regional indicator results in a very significant improvement in the fit, reducing the deviance by 26 at the cost of a single parameter.The associated coefficient estimate, 0.231 (se=0.045)implies that the odds of voting for the republican presidential candidate are about 26% higher among southern whites than those in other parts of the county, even after accounting for differences in income, sex, religious denomination and beliefs about biblical authority (a 95% confidence interval is 15% to 38%).The model HBG+RE includes both the southern regional indicator and election year as a categorical predictorseven dummy variables for elections 1984-2008, with 1980 being the baseline category.The inclusion of election as a predictor results in a dramatic improvement in the overall fit with the deviance being reduced by a further 285 points.However, it is interesting to note how the estimated coefficients for the variables in the HBG and HBG+R models are relatively unaffected by the inclusion of the election specific effects.
In order to examine the evolution of the HBG+R model over time one could fit the model to the data from each election separately and then plot the estimated log odds as function of time, controlling for the various model predictors.This approach is equivalent to including election year as a categorical factor crossed with every predictor in the HBG+R model, therefore multiplying the number of parameters by the number of elections.A more parsimonious modeling approach is to treat election year as a random factor with a single variance component associated with each crossed factor, and hence a fixed number of parameters regardless of the number of elections. 3This approach has the added benefit of smoothing the predicted log odds by borrowing strength across elections without assuming a particular parametric form for the trends over time.
Figure 2 displays the evolution of predicted log odds for non-southern, protestant, white, male voters, obtained from the mixed model, for each income quartile, and biblical authority.The figure highlights two points.First, there are large, election year, main effects, which explains the dramatic improvement in fit when election year was added as a predictor to the HBG+R model.For example, the overall tendency to vote Republican was high in 1984, Reagan's reelection year, but low in the 1996, Clinton's reelection year, relative to other election years.
Second, the biblical authority groups appear to be diverging over time, with a greater degree of divergence with increasing income.Figure 3 shows the same predicted log odds with the main effect of election subtracted out, and further suggests that the divergence is approximately linear in time.In modeling terms, the figures suggest there is a three-factor, income x biblical authority x time interaction, and that adding a single additional linear x linear x linear interaction parameter to the HBG+RE model is sufficient to describe the observed evolution in the log odds.The resulting model (denoted by HBG+RE3 in Table 2) reduces the deviance a further 50 points, and is preferred over all other models considered in terms of the AIC and BIC model selection criteria.Figures 4 and 5 give the analogous plots for the white, non-southern, Catholics.These plots show similar election year main effect patterns, and an overall tendency for Catholics to be less Republican than their Protestant counterparts.A striking difference between Catholics and Protestants is that the association between voting Republican and biblical authority in the 1980 election is in the opposite direction, with Catholic biblical literalists being the least likely Republican voters.However, the three-factor interaction noted in Figures 2 and 3 is also present among Catholics so that, over time, the direction of the association reverses, with the reversal occurring earlier as income increases.
Similar analysis of black voting patterns does not reveal any noticeable time trends with respect to biblical authority or family income.Black voters overwhelmingly favored the Democratic candidate in all Presidential elections from 1980 to 2008, with the only statistically significant factors being sex (black women are even less Republican than black men) and region (southern blacks are slightly more likely to vote Republican).

Discussion and Conclusion
This paper set out to measure the trend line in religious political polarization within the United States, and to determine the extent that polarization, if any is found, is more pronounced within the upper, versus the lower, reaches of the stratification order.Prior research is divided on the trend in religious political polarization since 1980, with some studies indicating no polarization, and other studies finding positive evidence.In addition there have been attempts to link religious political behavior to the politics of inequality, but these attempts generally lack firm empirical grounding, and have been judged to be unconvincing.The empirical analysis models the effect of religious identity on Presidential voter choice across the social class hierarchy (measured as relative family income), the race hierarchy, and the gender hierarchy.We reason that the stratification pattern of religious political polarization constitutes an empirical footing for interpreting the link between social inequality, religion, and political behavior.
The first analytical task is validating the measurement of religious identity, a necessary step given differences in the literature between the two most common measurement approaches.Our chosen measure is related to both approaches, yet also distinct insofar as it focuses upon biblical authority.The focus on biblical authority is justified by the Bible's presence within secular as well as religious realms of American society, potentially entering into symbolic preferences regarding how society should best be governed.
To assess the degree that biblical belief constitutes a valid concept for measuring religious identity, two surveys were undertaken.First, a nationally representative sample asked a set of five questions concerning biblical belief in relation to politics, personal decision-making, and social morality.Included within the five questions was the General Social Survey question about biblical authority, and it was found that responses to the five questions form a single-dimension scale with secular responses at one pole, religious responses at the opposite pole, and these two poles connected by a set of progressively religious/secular states.The second survey tested the biblical belief scale by asking a set of college students to rank the scale items, and the results confirmed the rankings.All of this suggests that people are well versed about biblical authority and social conduct, and assign a common ordering across a progressively religious/secular set of states.
The multivariate analysis of religious identity effects on voter choice invites the following interpretations.First, that religious identity was an influence on voter choice during the period 1980 to 2008 where individuals stating that the Bible is the literal word of God were strongly inclined to vote Republican, versus secular individuals believing that the Bible is a book of fables voting strongly Democratic.Second, that effect increased significantly across the study period, consistent with research finding religious political polarization during the same time period.Third, that religious identity linked to partisan polarity is concentrated within the upper reaches of the stratification order defined by upper income whites.There is no discernable effect of religious identity on voter choice among blacks who were strong Democratic voters across all categories of religious identity.This finding is consistent with the long-term trend in African-American partisanship that began shifting toward the Democrats following the 1932 election of Franklin Roosevelt; before then blacks were strong Republicans, the party symbolizing anti-slavery.Prior to 1932, Democrats were perceived by many African Americans to be implicated in racism and in supporting the Confederacy.Conversely whites, the dominant racial group, are sharply divided by religious identity, in particular within the upper levels of the income distribution.
The effect of gender on partisanship is less pronounced, and overshadowed by social class and religious identity.In some respects the study findings confirm the two dominant theories of religion and politics.The pattern of Protestant and Catholics partisanship is consistent with the ethnoreligious proposition that these two traditions evolved different partisan political linkages over time.Protestant partisanship is relatively less divided by social class, and more divided by biblical belief in comparison to Catholic partisanship that is more divided by social class, and less divided by biblical belief.We speculate that this difference is related to how political parties appeal to these two religious traditions, as well as to religious differences in theology and church organization.The data are consistent with an historical link between Catholicism and the Democrat Party emphasizing working class populism, and a moral imperative to redistribute to the lower classes.The link between Protestantism and the Republican Party presents little or no evidence for working class populism.
The study results are somewhat consistent with religious restructuring theory.Catholics and Protestants polarized on the basis of biblical belief where liberal interpreters increasingly favored the Democrats, and Republican partisanship strengthened among literal interpreters.Indeed, the finding of a positive time trend in partisan polarization related to trans-tradition religious identity is the epitome of religious restructuring theory.A limitation of religious restructuring theory, however, becomes apparent with regard to race.Black voters remained strong Democratic partisans across the study period, showing no inclination to polarize around religious identity.
The theoretical framework of this study provides an alternative interpretation to the ethnoreligious view of race.
Ethnoreligious researchers classify churches with predominately black congregations separately from white congregations, and theorize that the link between black churches and the Democratic Party is a function of race and historical experience.From our perspective, however, the experience of race is to some degree independent of religion, and the link between race and political partisanship is theorized as an outcome of domination/subordination within the stratification order.Religious identity empirically functions as a partisan influence for the dominant race, but not for the sub-dominant race, that votes as a block.This perspective resonates with Durkheim's social view of religion set forth in The Elementary Forms where religious consciousness reflects and reinforces social and economic relationships.
The study findings suggest that the influence of religious identity on Presidential voter choice is strengthening during a period of rising economic inequality.This influence is concentrated within upper class whites, and therefore not consistent with the term "culture war" because war implies that all of society is engaged.Secular/religious political polarization over the period 1980 -2008 in the United States reflects division within upper class whites, and is not a society wide phenomenon.Furthermore, this division affords an opportunity for the major political parties, and their candidates, to market themselves religiously without entertaining concrete economic policies that might attenuate economic inequality.In this sense we concur that religious politics are extraneous to the politics of economic inequality.Political activity aimed at reducing economic inequality would be strategically served by dialogue with the lower classes that are, in fact, the economic polarization losers, and therefore have the most to gain.
The U.S. evidence suggests that secularization predicted by earlier social scientists is premature at best, and perhaps altogether misplaced.Religious worldviews continue to play an active role in U.S. Presidential politics, and there is no evidence that this role will diminish in the short-term, or in the medium term.We believe that a fruitful theory for interpreting these phenomena is Durkheim's theory set forth in The Elementary Forms.The empirical link between religion and politics suggests that religion in the United States is not confined to formal institutions, but also is an element in constituting the national community.
Gelman et al. (2010) identify a nuance to theFiorina, Abrams, & Pope (2011) study utilizing a variety of data sources including the National Election Study, two National Annenberg Election surveys, and a Cooperative Congressional Election Study.Gelman et al. find evidence that partisan religious polarity characterizes upper income Presidential voters, where high-income church attenders are strong Republican partisans, versus high-income, non-attending voters that are strong Democratic partisans.Low-income voters, conversely, are not split by religious observance.Gelman et al. correlate this finding to the "red state/blue state" dichotomy where the former are characterized by church-going economic elites, and the latter by secular elites.

Figure 1 .
Figure 1.Multiple correspondence map of responses to question and statements concerning biblical authority: data labels correspond to responses to biblical authority question 1, and statements 2 -5, e.g., S5.n denotes a "no" response to statement 5, etc.The first principle coordinate explains 96.8% of inertia

Figure 2 .
Figure 2. Predicted log odds of voting for the Republican presidential candidate for non-southern, Protestant, white males, as function of time, income quartile and biblical authority response

Figure 3 .
Figure 3. Relative log odds of voting for the Republican presidential candidate for non-southern, Protestant, white males as function of time, income quartile and biblical authority response, after removing the main effect of election.(The main effect of election year has been subtracted out from Figure 2.)

Figure 4 .
Figure 4. Predicted log odds of voting for the Republican presidential candidate for non-southern, Catholic, white males as function of time, income quartile and biblical authority response

Figure 5 .
Figure 5. Relative log odds of voting for the Republican presidential candidate for non-southern, Catholic, white males as a function of time, income quartile and biblical authority response, after removing the main effect of election (The main effect of election year has been subtracted out from Figure 4.)

Table 1
Thus for example, although women in the aggregate are net Democratic partisans, upper class, white, biblical literalist, Protestant women are strong Republican partisans.The trend line in Protestant partisanship between 1980 and 2008 suggests growing religious-political polarization, where liberal interpreters of biblical authority became stronger Democratic partisans, and literal interpreters of biblical authority became stronger Republican voters.This polarization trend was intra-class salient within the upper income quartile, while diffusing downward over time to the lower reaches of the social class hierarchy.The time trend suggests that religious identity divides upper class Protestants, and this upper class divide has more or less faint reflections within the lower classes.Thus religious political polarization among white Protestants is characterized by intra-class polarity, and not inter-class polarity.
The trend line in Catholic partisanship is the same as the Protestant trend, although less Republican overall.Between 1980 and 2008, biblical literalist Catholics became stronger Republican voters, and liberal interpreters of biblical authority became stronger Democratic voters, with polarization between these two groups greatest within the top income quartile.Thus the trend in Catholic partisanship relative to income and biblical belief is the same as Protestant trend.When considering the Protestant and Catholic trend together, it suggests a common underlying trend in religiously linked, political partisanship.