The Heterogeneity of Southern Distinctiveness

Transcription

1 The Heterogeneity of Southern Distinctiveness Steven White April 28, 2012 Thanks to Robert Shapiro, Justin Phillips, Jeffrey Lax, and Robert Erikson for helpful comments. Thanks also to the Columbia University Economics Department for providing access to their computing cluster. This paper uses the 2004 and 2008 National Annenberg Election Survey (NAES) data, available online from the Annenberg Public Policy Center of the University of Pennsylvania at

2 Abstract This paper documents and assesses subregional variation among white southerners in presidential voting behavior and a variety of issue attitudes. I demonstrate that whites in the South remain consistently distinct from those in the rest of the nation, but heterogeneously so: whites in the Deep South are generally far more conservative than their Peripheral South neighbors. I also assess how the region s disproportionate concentration of born-again Christians can confound assessments of regional and state coefficients when properly accounted for in regression models. By demonstrating the continuing distinctiveness of the South, the significant variation present within the region, and the interrelationship of region and religion, these results have theoretical and methodological implications for the study of American politics.

3 In 2000 and 2004, the Democratic candidate for president received none of the South s 153 Electoral College votes. Many liberals in turn advocated that the Democratic Party whistle past Dixie and shift its focus to friendlier territory in the West (Schaller 2006). Yet in 2008, Barack Obama won Virginia, North Carolina, and Florida, suggesting Democrats might have a chance in the region after all. However, this improved showing was not consistent across the South. Many of the few areas where Obama performed worse than John Kerry among white voters were in the Deep South. In some states, like Alabama, his share of the white vote was barely half that of Kerry s, with only about one in ten white voters casting their ballot for Obama (Nossiter 2008). The context behind all this was the polarizing presidency of George W. Bush, a former governor of the largest southern state and the most famous born-again Christian in American political history. This confluence of factors is directly related to debates about the role of regionalism in contemporary American politics. In this paper, I test certain conventional theories of southern politics, as well as introduce a few new ones. 1 Briefly summarized, the results are as follows: First, the conventional southern dummy variable is almost always consistently significant across issue areas, but, second, it obscures substantial variation: In line with Key (2006 [1949]), the Deep South is generally far more distinctive than the Peripheral South, and arguments against the utility of exploring divisions within the South (e.g., Shafer and Johnston 2006) are at the very least overstated. Finally, when controlled for, the disproportionate concentration of born-again Christians in the southern states significantly alters assessments of the southern effect on certain culture wars issues especially policy-specific ones like bans on gay marriage and abortion suggesting the importance of unraveling regional effects from religious ones. By demonstrating the continuing significance of the South and the substantial variation 1 Of course, I do not claim to be testing all hypotheses about southern politics. The goals stated above necessarily limit the ground I can successfully cover in a single paper. 1

4 within the region and integrating the study of born-again Christian religious identification firmly into the study of southern political behavior these results have clear implications for studies of regionalism in American politics, as well as the role of religion in individual attitudes. Background and Literature Review The South has long been a distinctive region in American politics. Key described in great detail a one-party solid South defined to a large degree by racial politics, especially the maintenance of Jim Crow (2006 [1949]). Since then a two-party South has emerged see Lamis (1984) for a book-length treatment with the white South increasingly going Republican. Black and Black distinguish between two great white switches, the first in presidential voting habits of southern whites accompanying Barry Goldwater s Deep South victories in 1964, with the second in partisan identification of southern whites coming in the middle of Ronald Reagan s presidency (2002: 4). The Democratic Party s southern constituency has changed dramatically. A party originally created by racist southern white men to enhance and maintain their perceived interests has now become the political home of African Americans, liberal and moderate whites, and Hispanics, as Merle Black puts it (2004: 1001). Fitting with Abramowitz and Saunders s (1998) description of the increasing fit between ideology and partisanship in the electorate, white conservatives in the South became Republicans. Naturally enough, by 2000 a majority of southern whites in the American National Election Studies (ANES) survey considered themselves Republicans, while only one in three considered themselves Democrats (Knuckey 2006: 57). Contemporary political scientists are split on whether the South remains meaningfully different from the rest of the United States, and if so how it does. First, given the significant amount of political change in the postwar South, the degree to which the South as a whole 2

5 remains distinct is debated. Bartels (2000) suggests there no longer exists regional variation in the relationship between partisanship and presidential vote choice. However, Hillygus and Shields find that, despite recent arguments that the South is losing its historical distinctiveness, in the 2000 election, the voting calculus of Southern voters differs from the rest of the electorate (2008: 507). They argue that not only are southerners more conservative, but this conservatism plays a larger role in individual-level decision-making (Ibid.: ). Hillygus and Shields like much of the literature in American politics treat the South as one homogeneous region, a second issue I address. This decision is generally made without explicit theoretical rationale. A notable exception is Shafer and Johnston (2006), who directly attack the idea of dividing the South into subregions. They argue the Peripheral South s growth has lead to it becoming the South more generally, while the Deep South s demographic decline has left it a residual (Ibid.: 133). In his review of the southern politics subfield, however, Kousser rejects this critique in strong terms, declaring it unwarranted (2010: 377). A second goal of this paper is to assess precisely the utility of such subregional distinction in contemporary attitudes and behavior. I also examine the role of religious identification in shaping these regional distinctions. While much of the academic literature on religion and American politics falls into the culture war debate (Hunter 1991; McConkey 1998; Fiorina et al. 2006) which in political science has tended to relate to debates about partisan polarization and red states versus blue states (Abramowitz and Saunders 2008; Fiorina et al. 2008) recent work in sociology provides a more in-depth look at the effects of specific religious beliefs. Lewis and Huyser de Bernardo (2010), for instance, study Evangelicals and argue tradition is a better predictor of political attitudes while identification is a better predictor of partisanship. Froese et al. (2008) use General Social Survey (GSS) data to argue that belief in a punitive God leads to a tension between religious faith and support for civil liberties. Political scientists have also started to more seriously consider the role of religious identification. Barker et al. (2008) 3

6 argue that as belief in biblical authority increases, so too does the salience of foreign policy issues relative to domestic ones. Other work has challenged conventional media narratives about the political ramifications of the Christian Right. Claassen and Povtak (2010), for example, use the American National Election Studies (ANES) data to demonstrate that increased voter turnout of Evangelicals in 2004 was not a function of campaign dynamics the conventional story suggests the Bush reelection team was especially adept at turning out these voters but rather was simply the latest in a long-term trend driven by demographic change among Evangelicals. Such religious identification is not randomly distributed across the country. Figure 1 plots the percentage of whites in each state that identify as born-again (in 2004). While 40 percent of white Americans overall call themselves born-again Christians, this mean value obscures the vast range. Connecticut and Mississippi represent the two extremes, with only 12 percent of Connecticut whites identifying as born-again Christians, compared to 78 percent in Mississippi. The graph also demonstrates clear regional clustering. Five of the ten least born-again states are in New England and four others are in northern states like New York and New Jersey (heavily Mormon Utah is fifth). By contrast, seven of the ten most born-again states are in the South, with the other three being the southern border states of Oklahoma, Kentucky, and West Virginia. The top three spots are clustered in the Deep South in particular Mississippi, South Carolina, and Alabama with the more conservative Peripheral South states Arkansas and Tennessee coming in fourth and fifth. In general, then, it is very unlikely that a randomly selected white person in the Northeast is a born-again Christian, while it is very unlikely a randomly selected white person in the South is not a born-again Christian. Born-again Christians believe that eternal salvation requires a born-again conversion experience. Until the mid-1970s, most Americans who did not themselves identify as bornagain had never heard of the term. Jimmy Carter s 1976 campaign for president brought the 4

7 term to mainstream discourse for the first time. During the campaign, Carter, a Baptist, casually mentioned to New York Times chief religion reporter Kenneth Briggs that he was a born-again Christian. Many in the mainstream news media were unfamiliar with the term; indeed, [n]o one was sure that a presidential candidate should be talking about such things as private born-again experiences and conversions. Yet the term could hardly have been more familiar or less controversial to evangelical Christians (Martin 1996: 149). While the Democratic Carter can be credited with bringing white Evangelicals into the political mainstream, the group trended strongly Republican starting with Ronald Reagan. Attention to the disproportionate concentration of born-again Christians in the southern states links the work of political scientists with sociologists of religion, a combination that might prove useful to the theoretical concerns of both disciplines. Hypotheses The last section hinted at potential inquiries, but here I state them as testable hypotheses: H1. White southerners are distinctive from their non-southern counterparts. Southern political history provides ample reasons to suspect the region might still be quite unique in American politics (Key 2006 [1949]; Black and Black 1987, 2002; Hillygus and Shields 2008; Kousser 2010), yet some have suggested economic development, migration from outside the region, and other factors have lead to a decline of southern distinctiveness (Shafer and Johnston 2006). Hypothesis 1 assesses the relevance of the region considered as a whole. H2. Whites in the Deep South are distinctive from their Peripheral South counterparts. The Deep South is poorer, more racially heterogeneous, and less economically developed than the Peripheral South. There is reason to believe subregional variation might exist, 5

8 but scholars disagree over whether it remains meaningful (Kousser 2010) or not (Shafer and Johnston 2006). Hypothesis 2 addresses this disagreement. H3. The traditional Deep/Peripheral pattern exists when examining the states individually. The division between Deep and Peripheral South states is standard in the study of the region, yet it is generally asserted as inductively plausible (Key 2006 [1949]: 669; Black and Black 1987: 14) rather than properly tested. Hypothesis 3 looks at the actual empirical patterns. H4. The effect of identifying as a born-again Christian is homogeneous nationally. If the effect is homogeneous, then exploring its influence on southern distinctiveness is a straightforward affair. If, however, it varies by state, then exploring the confluence of region and religion requires a more interactive approach. H5. Disproportionate concentration of born-again Christians in the southern states confounds assessments of regional distinctiveness when properly controlled for. Identification as a born-again Christian is not widely used as a standard demographic control. Hypothesis 5 assesses the implications of this methodological neglect. Definitions, Data, and Methods Before testing these hypotheses, I first address a few relevant methodological issues. Definitions The first question in the study of the South in American politics is how to define the region. I define the South as the eleven states of the former Confederacy: Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Texas, 6

9 and Virginia. This is the classic definition used by Key (2006 [1949]: 11). 2 To explore subregional variation, I distinguish between the Deep South (Alabama, Georgia, Louisiana, Mississippi, and South Carolina) and the Peripheral South (Arkansas, Florida, North Carolina, Tennessee, Texas, and Virginia). This is the traditional breakdown in the southern politics literature (Key 2006 [1949]: 669; Black and Black 1987: 14). Beyond its intellectual lineage, the breakdown represents significant historical differences. The Deep South states have much larger black populations than the relatively whiter Peripheral South, although both regions are well above the national average. Key noted the Peripheral South states were far more accepting of the 1944 Smith v. Allwright case outlawing the white primary (2006 [1949]: 669). The five Deep South states were Barry Goldwater s only victories in his 1964 presidential campaign other than his home state of Arizona. In general, the Peripheral South states have long been more influenced by the rest of the nation than the Deep South. I define born-again Christian status as solely a matter of self-identification. If a respondent answers yes to the question, Do you consider yourself an evangelical or bornagain Christian?, then I assume they are. This is similar to the conception of partisanship described by Green et al. (2002) in that it is grounded in emotional attachment (indeed, they even compare their definition of partisanship to religion (Ibid.: 6)). Similarly, Lewis and Huyser de Bernardo argue that being an evangelical may be a foundation social-psychological identity that, similar to party identity, influences them beyond their denominational affiliation (2010: 124). This identity is at least as strong and perhaps stronger than partisan identification. 2 The case for a broader definitions of the region can be compellingly made, especially for historical studies. See Farhang and Katznelson (2005) and Springer (2011). However, in this paper I adhere to the Key-inspired 11-state definition more common to contemporary public opinion studies. 7

10 Data I use the 2004 and 2008 National Annenberg Election Survey (NAES) data. 3 The 2004 study consists of a national rolling cross-section of 81,422 interviews conducted between October 2003 and November 2004; the 2008 study consists of 57,967 interviews conducted between December 2007 and November The sampling procedure is random digit dialing. In combination, these two aspects of the data RDD and large sample size mean the data is generally representative at the state level, unlike, for example, the ANES cluster sample. Since my interest is in state-level variation, this is a significant strength. Although the 2008 data is more timely, I also utilize the 2004 data for two reasons: first, certain questions of substantive interests were not asked in 2008; and second, a much smaller percentage of the 2008 sample was asked the born-again Christian identification question, making analysis of Hypotheses 4-5 more difficult at smaller subnational levels. I analyze the following dependent variables of interest to scholars of public opinion and political behavior: in 2004, vote choice (1=Bush, 0=Kerry), feeling thermometer assessments of Muslims, gay organizations, and feminist organizations (1-10 scale, where 1=very favorable and 10=very unfavorable), support for an anti-gay marriage constitutional amendment (1=support, 0=oppose or neutral), support for an abortion ban (1=support, 0=oppose or neutral), whether Iraq was worth it (1=yes, 0=no), and support for more free trade agreements (1=favor, 0=oppose or neutral); in 2008, vote choice (1=McCain, 0=Obama), preferences for abortion availability (1=no restrictions, 2=available but with more limits, 3=not available except for rape, incest, or health reasons, 4=never available), support for gay marriage or civil unions (1=full marriage rights, 2=civil unions or domestic partnerships, 3=no legal recognition), whether Iraq was worth it (1=yes, 0=no), and support for free trade agreements (1=favor, 0=oppose or neutral). Variables are coded so that conserva- 3 This dataset is available online from the Annenberg Public Policy Center of the University of Pennsylvania at 8

11 tive responses are higher. 4 Dependent variables were selected with certain criteria in mind. I include vote choice because voter behavior is itself a special concern for political scientists. I look at issues of interest to social conservatives, as these might be especially salient in the region. I also give preference to questions asked in both years. I finally try to utilize questions that were asked to as many respondents as possible. Many questions of interest were only asked for a limited time period, which does not leave a sufficiently large N for state-level analysis. While I would prefer to analyze a more general economic question, for example, the free trade question was asked for a much longer period than similar questions about the minimum wage or labor organizing (which do not come with a sufficient N for state-level analysis). Methods My main explanatory variables of interest for Hypothesis 1-3 are region- and state-level dummy variables. I estimate three models for each dependent variable with these terms specified in slightly different ways so as to provide assessments of my hypotheses: a single South dummy variable (model 1); Deep South and Peripheral South dummy variables (model 2); and dummy variables for each of the 11 southern states (model 3). For models 1-3, the base category is the non-south. To better isolate regional and state effects, I control for partisanship (dummy variables for Republicans and Democrats, where independents are the base category), ideology (a 5-point scale from very liberal to very conservative), age, gender (dummy variable for female), household income, and education (dummy variables for high school or less and college degree, where some college is the base category). I limit my analysis to whites because of my theoretical objectives. Doing so narrows the scope of the paper I am analyzing white Americans, not Americans as a whole but this allows me to hold race constant to better isolate the independent effects of region and religion. 4 I code support for free trade as conservative, as this is the position taken by the Republican Party 9

12 A significant southern dummy variable in model 1 supports Hypothesis 1. A comparison of models 1 and 2 tests Hypothesis 2 that the Deep South is significantly different than the Peripheral South. This subregional distinctiveness hypothesis is supported when the Deep South term is larger than the Peripheral South and general South terms. I also conduct a test of the difference between the Deep and Peripheral South terms. If they are significantly different, this provides statistical confirmation of subregional distinctiveness. I also use Akaike s Information Criterion (AIC) to assess model fit. The AIC formula penalizes the inclusion of additional parameters, so if the inclusion of two regional terms instead of one nonetheless produces a lower value, the subregional model can be considered a better fit. A comparison of models 2 and 3 assesses Hypothesis 3, whether the Deep vs. Peripheral division arbitrarily obscures the actual state-level variation by looking at effects for each individual state. I treat feeling thermometer placements as continuous variables and estimate these models using OLS regression. Binary dependent variables are estimated using logistic regression. For the gay marriage and abortion questions in 2008, I allow for the qualitative nature of the ordered categories by using ordered probit models. Tables 1-4 report the results of the models. To provide more substantive meanings when discussing the results, I also calculate marginal effects for the logic models. Each marginal effect is the change in probability associated with that variable when all other explanatory variables are held constant. 5 The variables of theoretical interest are dummy variables: the marginal effect is the discrete change from zero to one, making substantive interpretation rather straightforward. I calculate predicted probabilities for shifts to theoretically interesting categories for the ordered profit models. 5 The base categories for the dummy variables are: independent, some college, male; linear predictors are set at their means; and other dummy variables in the regional/state sets are set to 0 when computing each regional/state effect. One can always quibble with the specific values at which the controls are set, but the overall substantive interpretation of the marginal effects is not significantly changed by other specifications. 10

13 To assess Hypothesis 4, whether the effect of being a born-again Christian varies by state or is homogeneous nationally, I estimate multilevel models where the dependent variables used in the previous section (2004 only) are modeled as a function of born-again status and the intercepts and slopes are allowed to vary by state. The same controls are used as before. Multilevel models can be understood as extensions of regression in which data are structured in groups and coefficients can vary by group (Gelman and Hill 2007: 237). The group here is state, which allows for estimates in which the effect of born-again identification is allowed to have differential effects in each of the states. The 95 percent confidence intervals are generated by simulating the model 1,000 times and cutting off 2.5 percent at each end of the distribution (Ibid.: 142). For Hypothesis 5, I compare results from the 2004 models estimated with and without a born-again Christian control variable. Hypothesis 5 is supported if the regional coefficients are substantively different when the born-again Christian term is included in other words, if part of regional distinctiveness is simply driven by the region s greater share of a particular demographic characteristic: born-again Christian self-identification. However, if controlling for this does not meaningfully change the regional terms, then claims of regional distinctiveness are robust even controlling for differential levels of religious identification. Results I present the results in the order of the stated hypotheses. H1: Do South/non-South differences still exist? Assessment: Yes. The standard southern dummy variable is significant in every single area, except for two: the feminist organizations feeling thermometer in 2004 and the free trade deals question in

14 Overall, Hypothesis 1 receives very substantial support: southern whites are very much still different than their counterparts elsewhere in the nation. I discuss the precise estimates in the next subsection, as they are more theoretically useful in comparison to the coefficients obtained for Deep and Peripheral South variables. H2: Is the Deep South different than the Peripheral South? Assessment: Yes. In 2008, white southerners were 14 percent more likely to vote for McCain over Obama overall, but there is substantial variation. Those in the Deep South were 25 percent more likely to support McCain, compared to only 9 percent for Peripheral South residents, a marginal effect almost three times the size. In 2004, white southerners were 10 percent more likely to support Bush over Kerry, but when the South is divided into its Deep and Peripheral subregions the coefficients again diverge. Being from the Deep South is associated with a 16 percent increase in the probability of voting for Bush, while being from the Peripheral South is associated with an 8 percent increase. Both terms are statistically significant and substantively meaningful, but the Deep South term is double the size of the Peripheral South term, and is 6 percentage points higher than the effect estimated in a single-south model. These results clearly affirm the analytical value of distinguishing between the Deep and Peripheral South in studies of presidential voting behavior. 6 For social issues, this trend continues for issues related to gay rights and assessments of Muslims, but interestingly a bit less so for feminist issues. For the gay marriage question in 2008, southerners overall were 10 percent more likely to move from category 2 to 3 (from support for civil unions, a moderate stance, to no support for any legal recognition of gay and lesbian relationships). In the Deep South, the number is 15 percent, compared to 8 6 The AIC value is lower for model 2 and a test of the difference between the two subregional terms is highly statistically significant for both years. 12

15 percent in the Peripheral South. For the abortion question, the South is statistically but not especially substantively more conservative. Stronger resistance to abortion access emerges in the Deep South, but the difference is not huge. Turning to 2004, southerners overall are 4 percent more likely to support a constitutional amendment banning gay marriage, while the probabilities are 7 percent in the Deep South and 3 percent in the Peripheral South just over twice the size. For the abortion ban question that year, being from the South is associated with just a one percent increase in the probability of supporting a ban, while in model 2 being from the Deep South is associated with a 5 percent increase, whereas the Peripheral South effect is not statistically significant. The 2004 survey also includes feeling thermometer questions. Significant subregional variation exists in assessments of Muslims. The South coefficient is significant in model 1, but in model 2 the Deep South coefficient is notably larger. Whites in the Deep South are more negatively disposed toward Muslims than whites in the Peripheral South. A similar pattern emerges for assessments of gay organizations. The South coefficient in model 1 is significant, but the Deep South term in model 2 is larger. 7 However, the patterns abruptly stop for assessments of feminist groups. None of the regional terms are significant in either model specification. Because research on regional variation in foreign policy attitudes is limited and somewhat dated (Cotter, Shaffer, and Breaux 2006: 189; see Hero 1965 for an older analysis), I next consider attitudes toward the Iraq war. White southerners are 7 percent more likely to say Iraq was worth it in 2008, but the Deep South marginal effect is 11 percent compared to 6 percent in the Peripheral South. In 2004, the marginal effect of being from the South is a 6 percent increase in the probability of thinking the war was worth it, but the marginal effect of being from the Deep South is a 10 percent increase in probability, while the marginal effect in the Peripheral South is only 4 percent. The regional trends seem to be robust over 7 For all dependent variables mentioned in this paragraph so far, the AIC value is lower for model 2 and a test of the difference between the subregional terms is statistically significant. 13

16 time for the Iraq question, not just a function of the electoral climate. 8 I finally look at attitudes toward free trade, an economic issue supported by conservatives generally but often opposed by those with lower levels of education, something that could plausibly create an ideological tension among southerners. The South also has a complex history with trade policy. In Key s time period, the South was the area with the most intense attachment to free trade, at least among congressmen (2006 [1949]: 353). There is no regional or subregional southern distinctiveness for the free trade question in The 2004 results are not much stronger. In model 1, being from the South is associated with a statistically significant but substantively small 1 percent decrease in the probability of supporting more free trade deals. In model 2, the results suggest this is driven by the Deep South: its marginal effect is a 2 percent decrease, while the Peripheral South term is not statistically significant. However, the AIC value is actually higher in model 2 and a test of the difference between the subregional terms is not statistically significant. Hypotheses 1-2 are not supported when the dependent variable is preferences for more free trade deals. H3: Do state-by-state patterns match subregional division? Assessment: Yes, with a few interesting exceptions. The state dummy variables largely act according to the Deep/Peripheral division, providing general support for Hypothesis 3 overall. Space concerns prevent going through each model in detail, but I discuss a few interesting results. The largest state effects for 2008 presidential vote choice are in the Deep South, with the estimate for Louisiana at an especially large 37 percent. The smallest marginal effects are for North Carolina, at 10 percent, and Florida and Virginia, where there is not a statistically significant difference from the rest of the nation. This provides face validity, as these were 8 For both years, the AIC value is lower for model 2 and a test of the difference between the subregional terms is statistically significant. 14

17 the three southern states Barack Obama won in 2008, and North Carolina was the closest. The state-level effects for the abortion questions are complex. For the question in 2008, the strongest opposition is present in Mississippi and Louisiana. Being from Florida, interestingly, is associated with a more liberal attitude. The state-level variation for the abortion question in 2004 is especially revealing. There is a positive increase in probability of supporting a ban in Alabama, Arkansas, Louisiana, Mississippi, and Tennessee; a decrease in probability in Florida and Virginia; and no effect for Georgia, North Carolina, South Carolina, and Texas. For the abortion variable, then, the Deep/Peripheral breakdown does not quite capture the nuances. The Peripheral South coefficient is effectively pulled down by negative effects in Florida and Virginia and the lack of any significant effect in Texas, its largest state. By contrast, whites in Tennessee and Arkansas are closer to their Deep South neighbors. For the feminist organizations feeling thermometer question, only the state of Mississippi stands out in the analysis as being uniquely anti-feminist. None of the Peripheral South states are statistically significant in themselves in the models analyzing the 2004 Iraq question, with one exception. It is that the sizable effect in Texas alone makes the Peripheral South term significant in model 2. None of the other Peripheral South states are themselves individually distinguishable from the non-southern base category, meaning in 2004 there wasn t so much a Peripheral South effect as a Texas one. North Carolina joins Texas in the 2008 dataset, but overall the biggest effects are throughout the Deep South and Texas. The association of the Iraq war with Bush, the former Texas governor, probably plays some role in this. There is some interesting complexity in the free trade question that further nuances the results from the first two models. Rather than grouping together regionally, the estimated state coefficients vary widely in statistical significance and in sign. In the 2004 data, especially clear patterns emerge: In Alabama, Mississippi, North Carolina, and Tennessee, whites are less likely to support free trade deals than whites in the non-south. However, in Florida, 15

18 Georgia, and Texas, whites are slightly more likely to support more free trade deals than non-southerners. Finally, the effect in Arkansas, Louisiana, South Carolina, and Virginia is not statistically significant. There is, in other words, not a consistently conservative direction in the effect of being from a southern state on free trade attitudes, but rather a mix of no effect, a conservative effect, and what might be labeled a populist effect. In the 2008 data, North Carolina, Mississippi, and Tennessee remain in the more populist, anti-free trade camp, but point estimates for the other southern states are not statistically significant. 9 H4: Is born-again Christian identification consistent nationally? Assessment: Yes, with minor exceptions. Overall, Hypothesis 4 is very strongly supported. For 5 of the 8 dependent variables in 2004 vote choice, the gay organizations feeling thermometer, gay marriage, and Iraq there is a homogeneous effect that is constant across the states. There is typically a significant effect of being a born-again Christian, but the confidence intervals overlap. For these dependent variables, then, there is no regional variation in the effect of religious identification that would complicate the previous analysis. 10 The three remaining dependent variables present a small amount of variation. Graphs of the effect on these variables are presented in Figure 2, with the point estimates for the 11 southern states presented first, followed by the rest of the nation in alphabetical order. There is a heterogeneous effect of born-again Christianity that is stronger in the South in the Muslims feeling thermometer model. Born-again Christian self-identification has a bigger impact on whites in states like Alabama, Mississippi, South Carolina, and Tennessee than it does in California, New York, Vermont, and Wisconsin, although in most states in the 9 The number of observations available for this question in 2008 is about a third the size of what is available in the 2004 data, making it difficult to conjecture about change over time vs. sample size issues. 10 These null results are not shown, but are available upon request from the author. 16

19 middle the effect is statistically indistinguishable. There is also some state-level variation for the feminist groups and abortion ban models, but with no coherent regional or ideological pattern. In the feminist groups model, born-again Christian identification has a stronger effect in West Virginia than Vermont, but the effect is also bigger than Vermont in New York and New Jersey. What could possibly distinguish these three disparate states from Vermont is unclear. However, it is again the case that the effect in most of the states is statistically similar. The effect in the abortion ban model is actually mostly homogeneous, but the point estimate in Arizona is statistically distinguishable from smaller point estimates in Arkansas and Louisiana. Yet this is an exception from the general lack of differentiation. Overall, with a small number of exceptions, Hypothesis 5 is supported. The effect of bornagain Christian identification on white attitudes is largely consistent regardless of region. H5: Does the concentration of born-again Christians in the South confound the regional effect? Assessment: For social issues, yes. However, the impact is not substantial across other issue areas. The strongest support for Hypothesis 5 is generally in social issues especially on the policy questions related to abortion and gay marriage and changes are most noticeable at the state level. There is also some effect in the vote choice model, again especially at the state level. Other issue areas are generally more robust, suggesting the disproportionate concentration of born-again Christians in the South isn t substantially altering the southern effect in those issues overall. I use graphs to compare the point estimates in each of the model specifications (Kastellec and Leoni 2007). Figures 3-4 plot the changes in the OLS and logit estimates of the regional (models 1 and 2) and state (model 3) coefficients with the addition of the born-again control 17

20 variable. The black dots are the coefficients from the previous section; the white dots are the new coefficients. With the born-again Christian term included, the statistical patterns in the South and Deep/Peripheral models remain consistent for 4 of the 8 dependent variables. For vote choice, Iraq, and the Muslims and gay organizations feeling thermometers, the Deep South still has the largest coefficient, the AIC is still lower in model 2, and a test of the difference between the subregional terms remains statistically significant. For feminist organizations, none of these hold the same as in the previous section. Yet there are significant changes of the substantive interpretations at many points. Interesting regional changes present in the free trade models (the Deep South term loses its slight statistical significance), the gay marriage models (a test of the difference between the Deep and Peripheral South is no longer statistically significant), and the abortion ban models (now the Peripheral South term is larger and statistically significant, while the Deep South term is not statistically significant, making it an outlier from the general pattern). The most interesting changes occur at the state level. Whites in the ten southern states other than Florida were previously estimated to be more likely to vote for Bush or approve of his job performance than whites outside the region, but when a born-again Christian variable is included in the model, the effects in Arkansas, North Carolina, Tennessee, and Virginia become statistically insignificant as well. The Peripheral South term remains significant because Texas remains significant. There appears to be, then, a serious confounding influence of state effects versus religion effects in many of the Peripheral South states, something missed in the subregional model. This is significant for two reasons. First, it shows the potential of the subregional terms to obscure the precise pattern of state-level variation. Second, it shows the incredible potential of a single additional control variable born-again Christian self-identification to substantially alter the results. In the Iraq models, the coefficient on Louisiana loses its statistical significance, leaving only Alabama, Georgia, South Carolina, and Texas in the significant column. While the 18

21 non-texas Peripheral South states were originally statistically insignificant and remain so, the (insignificant) point estimates get smaller in some cases negative and this leads to the Peripheral South term in model 2 becoming statistically insignificant. The effect in Texas alone can no longer drive the Peripheral South term to significance at the subregional level. For the trade variable, the Alabama and Florida coefficients lose their significance, while South Carolina moves from insignificant to negative and significant, joining the populist category along with Mississippi, North Carolina, and Tennessee. Now only being from Georgia and Texas is associated with a more pro-free trade position. The most substantial changes of sign and interpretation occur on social issues. For the Muslims feeling thermometer variable, the Texas coefficient loses its statistical significance in the model with the born-again term, joining Florida and Louisiana. In the gay organizations model, the Virginia and Texas coefficients lose their significance, again joining Florida and Louisiana. There is, additionally, some significant change in interpretation of the coefficients in the gay organizations models among the terms that remain significant, with some coefficients being significantly altered by the presence of the born-again Christian control. At the regional level, the South coefficient loses about 43 percent of its original size, dropping from.49 to.28. This is reflected in the subregional coefficients as well, with the Deep South coefficient moving from.74 to.42 (a 41 percent decrease) and the Peripheral South coefficient moving from.40 to.22 (a 45 percent decrease). At the state level, similarly substantial changes in point estimates occur for every state except Florida. While some would focus on the consistency of statistical significance only, substantive interpretation matters as well: a reader trying to assess the impact of the regional variable on attitudes would make an inflated assessment of the regional term s impact without the born-again control variable. When adding a single additional variable causes the coefficient on the variable of theoretical interest to be cut by nearly half, model specification takes on additional significance. Finally, the two policy issues offer perhaps the most interesting changes. For the anti- 19

22 gay marriage amendment variable, the coefficients on the majority of states lose their initial statistical significance: whites in Georgia, Louisiana, North Carolina, South Caroline, Tennessee, and Texas are now not statistically different than the rest of the nation. The Virginia and Florida coefficients are not significant in either model. Only the effects in Alabama, Arkansas, and Mississippi remain robust with the addition of the born-again control variable. And even in these states, the logit coefficients decrease by 55 percent, 48 percent, and 45 percent, respectively. The marginal effects decrease from 9 percent to 4 percent in Alabama, 11 percent to 6 percent in Arkansas, and 14 percent to 8 percent in Mississippi. 11 State and regional effects on this issue appear to be heavily influenced by born-again Christianity. While the pattern of change from conservative to insignificant is consistent in the gay marriage models, the changes in the abortion ban models are far more complex. While all states move noticeably to the left with the exception of Florida, which barely changes the end result depends on where the state effect started in the previous model specification. The coefficients on Alabama, Mississippi, South Carolina, and Tennessee previously significant in the conservative direction become statistically insignificant. The coefficients on Georgia, North Carolina, and Texas previously statistically insignificant become significant in the liberal direction, with new estimated marginal effects of -5 percent, -4 percent, and -3 percent, respectively. Arkansas and Louisiana remain statistically significant in the conservative direction, but less so moving from +12 percent to + 7 percent and +8 percent to +7 percent, respectively. Finally, Florida and Virginia remain statistically significant in the liberal direction, with Virginia becoming significantly more liberal, moving from -3 percent to -8 percent, and Florida moving from -3 to -5 percent. One final result to note is that the born-again coefficient is statistically significant in every single model. White born-again Christians were more likely to vote for Bush; offer negative assessments of muslims, gay organizations, and feminist organizations; support bans on gay 11 Born-again is set to 1, as this is the most common category among southern respondents. 20

23 marriage and abortion; and think the Iraq war was worth it. They were less likely to support free trade. While some of these are fairly obvious, the relationship with Iraq and free trade are somewhat less so. The greater propensity of white born-again Christians to think the Iraq war was worth it fits well with with Barker et al. s (2008) analysis of the relationship between belief in biblical authority and the salience of foreign policy issues. However, it is unclear why born-again Christianity would be expected to correlate with attitudes toward free trade, especially since conventional variables related to trade attitudes (education and income) are controlled for in the models. Future research could examine the underlying reasons for such a relationship. Conclusion This paper demonstrates the continuing significance of the South for studies of white public opinion and voting behavior, the utility of subregional divisions within the South, and the complicated role of born-again Christian identification in this disproportionately evangelical region. To summarize the results, the support for Hypotheses 1-5 is as follows: The general southern distinctiveness hypothesis, H1, is strongly supported, as is the subregional variation hypothesis, H2. The state-by-state hypothesis, H3, is likewise supported overall, although there are a few interesting exceptions. The born-again homogeneity hypothesis, H4, is generally supported. The born-again regional concentration hypothesis, H5, is supported for social issues, but the impact of the additional control is smaller for most other issue areas. These results have numerous implications for the broader study of American politics. One implication is that the common inclusion of a South/non-South dummy variable is effectively a weighted average of extremely significant Deep South variation watered down by the far less distinct Peripheral South. One way of framing this is that if by the South analysts mean traditional southern areas like Mississippi or Alabama rather than, for instance, the 21

24 northern Virginia suburbs of DC or the retirement communities of Florida, a South dummy variable blends the very southern respondents of Biloxi with the perhaps southern-in-nameonly respondents of Arlington, downplaying the sizable differences in comparing the Deep South to the rest of the nation. The Deep/Peripheral South division is not perfect, but it is much better than a single-south term when datasets allow for it. Arguments that southern distinctiveness is irrelevant, or that meaningful variation no longer exists within the South, are misguided. Second, this paper also suggests political scientists should think more about born-again Christianity. Those interested in the South should ask whether they are more interested in a bivariate or multivariate definition of what constitutes distinctiveness. In other words, is the question whether the South is different overall, or rather whether it is different controlling for its greater concentration of particular demographic characteristics, like born-again selfidentification? This paper remains agnostic on this question, but the results suggest it requires consideration. Beyond this, considering the link between religious identification and other issues might also prove enlightening to American politics scholars more generally. More broadly, this analysis can serve as an empirical middle ground between two seemingly contradictory phenomena: the Democrats southern struggles in 2004 and the strong showing of Barack Obama in many of the Peripheral South states in The Democratic Party nominee for president will not be winning the Electoral College votes of Mississippi or Alabama anytime soon. But what Schaller (2006) and many journalistic accounts missed but the Obama campaign did not is the diversity of the South. Even if the Democrats are unlikely to start winning the Electoral College votes of the Deep South, the party is already winning them in the Peripheral South. Whether this continues to occur will be decided by an array of variables beyond the scope of this paragraph. But the differentiation of the Peripheral South from the Deep South provides an electoral opportunity for the Democratic Party that many overlooked. 22

25 References Abramowitz, Alan and Kyle L. Saunders. Ideological Realignment in the U.S. Electorate. Journal of Politics. Vol Is Polarization a Myth?. Journal of Politics. Vol. 70. No. 2. April Barker, David C., Jon Hurwitz, and Traci Nelson. Of Crusades and Culture Wars: Messianic Militarism and Political Conflict in the United States. Journal of Politics. Vol. 70. No. 2. April Black, Earl and Merle Black. Politics and Society in the South. Cambridge: Harvard University Press, The Rise of Southern Republicans. Cambridge: Harvard University Press, Black, Merle. The Transformation of the Southern Democratic Party. Journal of Politics. Vol 66. No. 4. November Claassen, Ryan L. and Andrew Povtak. The Christian Right Thesis: Explaining Longitudinal Change in Participation among Evangelical Christians. Journal of Politics. Vol. 72. No. 1. January Cotter, Patrick R, Stephen D. Shaffer, and David A. Breaux. Issues, Ideology, and Political Opinions in the South, Writing Southern Politics: Contemporary Interpretations and Future Directions. Eds: Robert P. Steed and Laurence W. Moreland. Lexington: The University Press of Kentucky, Farhang, Sean and Ira Katznelson. The Southern Imposition: Congress and Labor in the New Deal and Fair Deal. Studies in American Political Development. Vol. 19. April Fiorina, Morris P., with Samuel J. Abrams and Jeremy C. Pope. Culture War? The Myth of a Polarized America. New York: Pearson, Polarization in the American Public: Misconceptions and Misreadings. Journal of 23