STATE CONTEXT AND WHITE VOTING

Viable Black candidates often have to compete with White opponents in attracting White voters’ support in order to win U.S. elections. Thus, the willingness of Whites to vote for Black candidates has drawn considerable attention from students of racial politics (Carsey Reference Carsey1995; Gillespie Reference Gillespie2012; Hajnal Reference Hajnal2007; King and Smith, Reference King and Smith2008; Kraus and Swanstrom, Reference Kraus and Swanstrom2005; Liu Reference Liu2006; Lublin Reference Lublin1997; Stein et al., Reference Stein, Ulbig and Post2005). It has been found that Americans often cast their votes along racial lines (Barreto Reference Barreto2007; Dawson Reference Dawson1994; Gay Reference Gay2004; Kinder and Sanders, Reference Kinder and Sanders1996; Liu and Vanderleeuw, Reference Liu and Vanderleeuw2007). Yet some election outcomes also showed that White voters are willing to vote for Black candidates, even when these Black candidates are competing with White candidate(s) in biracial elections. White crossover voting was instrumental, for example, in many Black victories in cites such as Newark, Los Angeles, Memphis, and New Orleans during the last four decades (Gillespie Reference Gillespie2012; Hajnal Reference Hajnal2007; Liu Reference Liu2006).

Many theories have been proposed to downplay the role of race. It has been argued, for example, that White voters evaluate Black candidates based on their “quality,” rather than on race. Like in any other job application process, Black electoral office-seekers need to have certain personal and professional qualifications in order to appeal to White voters (Thernstrom and Thernstrom, Reference Thernstrom and Thernstrom1999). Zoltan Hajnal (Reference Hajnal2007) suggests that Whites will show more willingness to support Black candidates once they have a chance to observe the quality of Black incumbents. The willingness of White liberal voters to support a “qualified” Black candidate perhaps is best represented by Thomas Bradley’s control of the mayor’s office of Los Angeles for two decades (Browning et al., Reference Browning, Marshall and Tabb1984, Reference Browning, Marshall and Tabb2003; Sonenshein Reference Sonenshein1993).

Some scholars discovered that the role of race can be played in a more “subtle” and “implicit” way. For example, in order to win as many White votes as possible a “deracialization” strategy very often is vital to Black candidates’ success in elections (Liu Reference Liu2003; Wright and Middleton, Reference Wright and Middleton2001). Especially when facing strong White opponents, a pragmatic campaign strategy for Black candidates is a deracialization strategy targeted at White liberal voters in White majority districts (Liu and Vanderleeuw, Reference Liu and Vanderleeuw2007; Persons Reference Persons1993). Deracialization strategies are reported to have been the key to such electoral victories as that of the first Black elected governor of Virginia, L. Douglas Wilder; the first African American woman elected to the U.S. Senate, Carol Moseley-Braun of Illinois; and the former mayor of New Orleans, Ray Nagin. More recently, Andra Gillespie (Reference Gillespie2012) suggested that deracialization may already run out of its “magic” in the postracial America. One “side effect” of deracialization is the loss of Black vote. Black candidates’ deracialized campaigns may also be “interrupted” or even “damaged” unexpectedly by both their White opponents and mass media through racially “coded words” that injected White fear (Mendelberg Reference Mendelberg2001). Obama faced arguably the greatest crisis of his 2008 campaign because of his connection with Jeremiah Wright, his long-time pastor who was repeatedly displayed on national media for his strong accusation of “White guilt” in African American suffering. As a result, Obama gave his critical “A More Perfect Union” speech in Philadelphia on March 18, 2008, in which he called for racial reconciliation rather than racial blaming. A month later Obama denounced his pastor altogether to show that he did not agree with Wright’s publicized anti-White rhetoric.

Arguably, White voters’ reactions to the news coverage of Jeremiah Wright were not identical partly due to their different levels of fear of Black threat. Overall, the success of Black candidates’ strategy to win White votes may depend on the racial context in which the election takes place. The concept of context can have many meanings. It often refers to a variety of characteristics of a specified geographic area. Context can also be based on “the distribution of a population characteristic” (Huckfeldt Reference Huckfeldt1986, p. 14). The population characteristic that receives the most attention perhaps is the relative percentage of Blacks within a certain area (i.e., Black density). It has been shown repeatedly in the political science literature that there is a negative relationship between Black density in an area and White racial tolerance (Giles and Buckner, Reference Giles and Buckner1993; Glaser Reference Glaser1994; Longoria Reference Longoria1999; Taylor Reference Taylor1998). Todd Donovan (Reference Donovan2010) directly linked the White support for Obama in the 2008 presidential election to Black threat. The Black threat theory, which originated from the classic study of southern politics by V. Key Jr. (1949), explains this relationship based on Whites’ group interests and the relative threats posed by Blacks in different contexts. According to Black threat theory, different contexts affect White perceptions of how their group interests are threatened by Blacks (Giles and Hertz, Reference Giles and Hertz1994; Taylor Reference Taylor1998). A higher level of Black population density may produce a higher level of White perception of Black threat, and therefore a lower level of White crossover voting.

Perceptions of threat may be reduced by civic engagement and interpersonal trust, or social capital. Robert Putnam (Reference Putnam2000) in his seminal work, Bowling Alone, assembled an array of empirical measures of social capital to demonstrate that individuals who interact with others in their communities possess both high levels of interpersonal trust and civic engagements. These individuals are the “social capitalists” (p. 403) who, based on Putnam’s state-level measure of social capital index, are happier psychologically and more successful socially and economically than those who are “hermits” (p. 403). Moreover, states reveal different contexts in terms of the level of collective social capital. Putnam also tried to control for the effect of racial composition of the state when he assessed the impact of social capital, and his conclusion was that social capital was far more important in explaining a variety of indicators of success. When he analyzed the effect of social capital on education and childcare, for example, Putnam insisted that “A state’s racial composition and rate of single-parent families also affect child well-being, though far less consistent or strongly than do poverty and low social capital . . . The beneficial effects of social capital persist even after accounting for a host of other factors” (pp. 298–299) such as racial composition.

Putnam (Reference Putnam1995) emphasizes social capital as the “features of social life-networks, norms and trust-that enable participants to act together more effectively to pursue shared objectives” (pp. 664–665). How did the Whites in a rich social capital states react to Obama’s historical candidacy, compared to other Whites who are from a poor social capital state? One expectation from the social identity theory and social psychology model is that social capital enhances Whites’ in-group identity at the expense of out-group members—a side effect of the so-called “bonding capital,” (Putnam Reference Putnam2000, p. 22) which may reinforce the negative stereotype that White social capitalists use as their cognitive shortcut in their information processing of the Obama candidacy. Young Mie Kim and Kelly Garrett’s (2012) recent study of the 2008 Democratic primary revealed that voters used more memory-based information processing on Obama’s candidacy, compared to the more affective on-line tally information processing used for Hillary Clinton.

On the other hand, Whites in rich social capital states, because of their high level of interpersonal trust and civic engagement, may be more likely to support a Black candidate who represented a change (at least racially) in a nation “divided by color” for a long history, rather than seeing him as a “Black threat.” Ryan Carlin and Gregory Love (2013) in their recent experimental study found that the previous view that interpersonal trust tends to produce the so-called “trust bias” (p. 44) (which favors trustors’ own partisan group at the cost of out-groups) failed to explain their experimental results. In fact, “the trust decision itself provides information to override partisan bias in reciprocity” (p. 44) because trustors are capable of dual process of decision making in “extending the trust” (p. 47) to obtain benefits and cooperation from out-groups. As Putnam emphasized, bonding (exclusive) capital is vital to “getting by,” but the bridging (inclusive) capital (which serves a greater role in trusting the outer-groups) is essential to “getting ahead” (Reference Putnam2000, p. 23).

The empirical evidence from the National Election Study (NES) also sheds important light on the relationship between White trust and their opinions toward the possibility of a first Black President. The 2008 NES asked whether people can be trusted, and the respondents could choose from the following answers: always, most of the time, about half the time, once in a while, and never. The Whites who answered that “people can always be trusted” provided the most positive and optimistic answers to the questions: Is the United States ready for an African American president? Do you hope that United States has an African American president? Will a Black president make you pleased? In contrast, the Whites who indicated that “people can never be trusted” provided the most negative answers to these same questions.^{Footnote 1} In short, the 2008 NES data seem to suggest that a higher level of trust based on Putnam’s social capital thesis might indeed have helped reduce Whites’ perception of Black threat caused by Obama’s run for the Presidency.

It is also important to note that the influential work of Putnam on the significance of social capital has always invited criticisms about its implications on race (Field Reference Field2003). In his recent book, Racial Diversity and Social Capital, Rodney Hero (Reference Hero2007) juxtaposed the social capital thesis and the racial diversity thesis. Derived from his own empirical analysis of the racial makeup of states which takes into consideration not only Black population, but also Whites and other minorities such as Latinos and Asians, Hero insisted that the racial diversity variable is the key to understanding political, social, and economic differences across American states. Hero attempted to put both social capital and racial diversity in the same empirical models and tested the relative effects of each variable. Based on his racial diversity thesis, Hero asserted that race is so fundamental that “when racial diversity is appropriately factored into the research design and political and policy indicators, the salutary effects of aggregate social capital are dramatically diminished” (Reference Hero2007, p. 17).

Hero (Reference Hero2007) positions his racial diversity argument and Putnam’s social capital thesis at two opposite ends of the spectrum in American political science. This is because, according to Hero, there are two theoretical traditions that distinctively emphasize either a pluralist society centered on a group approach (pluralism), or the unequal structural elements in American political institutions that have long suppressed minorities into a disadvantageous position. Hero argued that the social capital thesis belongs to the first approach that in the history of American political science has produced influential works from Alexis de Tocqueville’s Democracy in America (2004) to The Civic Culture of Gabriel Almond and Sidney Verba (1989), and Daniel Elazar’s American Federalism (1984).

Elazar’s (Reference Elazar1984) work on state political culture is especially important, because it is a major scholarly work on how states developed their own cultural identities throughout U.S. history, and how these identities shaped the nature of American federalism. The states in which White voters live, according to Elazar, may influence their vote choices. There are three major types of state cultures, which Elazar called moralistic, individualistic, and traditionalistic. The fundamental differences between these cultures are that individualistic culture views government as a market or means to respond efficiently to demand, while moralistic culture views government as a commonwealth or means to achieve the good community through positive action, and finally, the traditionalistic culture views government as a means of maintaining the existing order (Elazar Reference Elazar1984). It is also important to note that there have been many debates on Elazar’s state political culture measurement (Brown and Palmer, Reference Brown and Palmer2003). Elazar himself, in his later edition of American Federalism: A View from the States (1984), recognized the possibility of a synthesis of two subcultures as well as the existence of two separate sub-cultural communities with the first being dominant and the second secondary in the same states.

Were White voters in the 2008 and 2012 presidential elections influenced by these political cultures? Based on Elazar’s elaboration of political cultures, one can reasonably assume that the Whites from traditionalistic states would be most likely to oppose the change that Obama, a Black candidate, was trying to bring to America. The Whites in the individualist states would be skeptical of Obama because of the lack of understanding of Obama’s real ability to bring the necessary change to the political marketplace. Moralistic culture may generate the highest level of support from White voters for Obama because “politics, to the moralistic political culture, is considered one of the great activities of humanity in its search for the good society . . . [Good] government is considered a positive instrument with a responsibility to promote the general welfare” (Elazar Reference Elazar1984, p. 117).^{Footnote 2}

Furthermore, “[b]y virtue of its fundamental outlook, the moralistic political culture creates a greater commitment to active government intervention in the economic and social life of the community” (Elazar Reference Elazar1984, p. 118). In short, the moralistic states would most likely be the places where the White voters embraced Obama’s appeal to “the better angels of our nature” and the “active government intervention” such as the health care reform (later the Patient Protection and Affordable Care Act) that Obama campaigned for in 2008.

To summarize above discussions of the four contexts at the state level, we provide the following four competing hypotheses:

RACIAL TENSION AND POLITICAL DEVELOPMENT

As discussed above, many of the debates during the last three decades on the racial contexts of American states have involved four competing theories that examine racial and ethnic conflict from the perspectives of Black threat (Key Reference Key1949), racial diversity (Hero Reference Hero1998), political culture (Elazar Reference Elazar1984), and social capital (Putnam Reference Putnam2000). However, one major problem that has been reported by scholars of state contexts is that Black density, racial diversity, and social capital at the state level are in fact highly correlated (see below for an empirical test). Hero (Reference Hero1998), for example, noticed the correlation between his diversity measure and Elazar’s political cultures. Putnam (Reference Putnam2000) also reported that his social capital measures for 1980 and 1990 at the state level are highly correlated with Elazar’s state culture scores. The observed correlations not only make the empirical tests of the four competing hypotheses uncertain, but also demand a strong theory to explain the correlations among the four contextual variables (King et al., Reference King, Keohane and Verba1994).

As a whole, the theories of Black threat, racial diversity, political culture, and social capital imply that the four variables conceptually measure different state contexts. No previous studies, however, systematically analyzed why all the four variables were correlated with each other in the first place. This lack of understanding is certainly related to the tendency to treat Black density, racial diversity, and state capital as the “separated” explanatory variables in previous studies, which ignore the possibility that the four variables may simply reflect a deeper level of racial relation. One important question to ask, therefore, is what produced these clearly interconnected contexts at the state level in the first place. The answer may be found after one goes beyond these observed contextual variables to trace the deep-seated racial tension in states. To do this, it is especially important to draw insights from the recent literature of American political development.

Desmond King and Roger Smith (2005) argued that “racial orders” (p. 75) are the key components of American political development. More specifically, “American politics has historically been constituted in part by two evolving but linked racial institutional orders’: a set of White supremacist’ orders and a competing set of transformative egalitarian’ orders” (p. 75). Furthermore, King and Smith suggested that the interplay between these two racial orders have shaped how coalitions of “state institutions” and political actors are “responding to the tensions and opportunities generated by America’s racial orders” (p. 84). Thus, in order to find how different White voters responded to Obama’s historical candidacy differently, it is necessary to discuss “racial tension” generated by competing racial orders in different states. In this regard, however, King and Smith (Reference King and Smith2005) did not provide any empirical measure of state-level racial tension to test the impact of race on Obama’s White support. The reason for lacking an empirical measure of racial tension is understandable, as racial tension is better conceptualized as a latent variable, not just a theoretic construct. One may feel the impact of racial tension in her daily life, but it is hard to pinpoint a particular social phenomenon as racial tension.

To fill the gap in the literature, this article proposes a new theory of racial tension to go beyond the observed Black density, racial diversity, political culture, and social capital. Borrowing arguments from the recent American political development literature, this research suggests that voters make voting decisions in a context of racial tension. Racial tension reveals an overall racial relationship in a state (Novkov Reference Novkov2008). It shows the degree to which racial polarization may be materialized once a racially sensitive event, such as the 2008 presidential election, takes place (King and Smith, Reference King and Smith2008). The level of racial tension in a state can also be understood as the racial status of a state, which has a deep root in the history of racial orders in the state (King and Smith, Reference King and Smith2005, Reference King and Smith2008).

The origin of racial tension itself is a story of American racial relations that reflected American experiences concerning multiple racial groups (Marx Reference Marx1971). For instance, the Deep South had a long history of racial struggle between African American slaves and their White slave owners who benefited from a slave economy (Mulcare Reference Mulcare2008). The “White supremacist racial order” (p. 75) that King and Smith (Reference King and Smith2005) articulated ran deep in the Deep South, and historically the Deep South has had the highest level of racial tension in the country. In comparison, the upper Midwest states saw the influx of German and Scandinavian immigrants to participate in economic and territorial expansion. The “egalitarian racial order” (King and Smith, Reference King and Smith2005, p. 75) is more likely to be accepted in states such as Minnesota, Wisconsin, and the Dakotas, and racial tension is, relatively speaking, low there. The “Frontier West,” on the other hand, did provide multiple racial groups with more economic opportunities, but the early racial interactions there also pitted minority groups against each other and against White ethnic groups (such as Irish workers) for employment and job benefits (Novkov Reference Novkov2008). Thus, the West (especially California and Southwest) did attract various racial and ethnic groups traditionally, and the racial tension of the West is not as high as in the South, but not as low as in the Upper Midwest either.

These early political developments at the state level had a profound impact on the formation of long-term, and often enduring, racial tension of the states (King and Smith, Reference King and Smith2005). It is also possible that a large-scale change in a state, such as the new birth of the automobile industry in Michigan that provided the engine for the great migration of African Americans to Michigan, can enhance the racial tension in Michigan. Nevertheless, the overall geographic distribution of state-level racial tension (e.g., a high level of racial tension in the Deep South and a low level of racial tension in the Upper Midwest) is durable in the United States as a whole.

RACIAL TENSION AND THE FOUR CONTEXTUAL VARIABLES

How is racial tension related to Black density, racial diversity, state political culture, and social capital? First, racial tension should not be measured by just one of the four contextual variables. For example, West Virginia and Minnesota had almost the same level of low Black density (i.e., slightly over 3%), but arguably West Virginia has an overall higher level of racial tension than Minnesota. Second, to see the relationship between racial tension and the four contextual variables, it is important to emphasize that racial tension is the underlying latent factor. The states’ racial makeup is the “visual effect” of the states indicated by their levels of Black density and racial diversity. A low level of racial tension maintains White homogeneity. A higher level of racial tension, on the other hand, leads to more “White flight” and larger proportions of minorities, which then reinforces the racial tension of the state.

Moreover, a state’s racial tension also leads to the formation of specific community norms, through which members of the community interact with each other. The norms of states are exactly the subjects of Putnam’s (Reference Putnam2000) and Elazar’s (Reference Elazar1984) classic studies of social capital and political culture. For example, a moralistic culture is likely to appear and be sustained in states that have a low level of racial tension, while a traditionalistic political culture develops due to the high level of racial tension. Similarly, social capital is more likely to be accumulated in states with low levels of racial tension. In sum, it is the deep-seated racial tension that links all four contextual variables together and leads to the observed correlations among Black density, racial diversity, social capital, and political culture.

In sum, this paper further suggests that the higher the level of racial tension a state has, the smaller the likelihood that White voters may support Obama in his presidential elections (Hypothesis 5).

Rather than treating the theories of Black density, racial diversity, social capital, and state political culture as competing theories, the racial tension approach of this paper suggests that racial tension is the fundamental factor (i.e., a latent variable) that are revealed by states’ racial makeup (Black density and racial diversity) and community norms (social capital and political culture). More importantly, borrowing insights from the recent political development literature, this article suggests that it is the deep-seated racial tension that directly affected how Whites voted for Obama, our ultimate dependent variable (see Figure 1). The following sections empirically compare this new racial tension approach with the previous competing hypothesis approach. The goal is to demonstrate why empirically all four contextual variables are simultaneously linked to the underlying factor of racial tension, and furthermore why it is necessary to take consideration of racial tension to explain Obama’s White voter support.

Fig. 1.

Proposed Path Diagram for Obama’s White Support in States

THE PREVIOUS COMPETING HYPOTHESIS APPROACH AND ITS LIMIT

This section reports the findings based on the previous competing hypothesis approach, and discusses the limitation of this approach. We test Hypotheses 1 to 4 by using the state-level data. A state-level analysis is especially important because the U.S. presidential election outcome is based on the Electoral College votes that use states as the election units. The winner-takes-all electoral system forces both candidates and voters to be sensitive to the state contexts.

To test the four hypotheses, the data from the exit poll are used to measure our dependent variable, Obama’s White support in states.^{Footnote 3} The social capital data are directly from Putnam’s Reference Putnam2000 social capital index, which is his standardized factor score based on his fourteen indicators (range=-1.43 to 1.71, mean=.02, sd=.78).^{Footnote 4} The racial diversity measure is based on the 2006 census population data concerning the population shares of Whites, Blacks, Latinos, and Asians (range=.08 to .78, mean=.39, sd=.17). The census data also include the measure of Black density based on the percent non-Hispanic Black in the state population in 2006 (range=.37 to 36.95, mean=10.02, sd=9.56).^{Footnote 5} The state political culture measure is derived from Elazar (Reference Elazar1984) (range=1 to 8, mean=4.14, sd=2.52).^{Footnote 6}

Table 1 shows the results of four multiple regression models based on the competing hypothesis approach.^{Footnote 7} Using a two-tailed test and the model with the four independent variables (i.e., no other controls), state political culture is found to be the variable accounting for Obama’s White support at the state level in 2008 and 2012 (see Models 1 and 2), while Black density is not a statistically significant variable in all four models, which is contrary to Key (Reference Key1949) and our Hypothesis 1. In addition, contrary to Hero’s (Reference Hero1998, Reference Hero2007) claim about the greater explanatory power of racial diversity variable, Hypotheses 4 receives no empirical support in our four models. Hypothesis 3 regarding the positive effect of political culture appears to have some empirical support in Models 1 through 3, while Hypothesis 1 about social capital needs to be rejected. Upon a further examination of the findings presented of this table, however, two problems remain to limit our confidence in drawing a final conclusion.

Table 1.

Multiple OLS Regression of White Support for Obama in the 2008 and 2012 Presidential Elections: The Competing Hypotheses Approach

***p < .001; **p < .01; *p < .05

(two-tailed test)

First, the regression results may not be evaluated effectively due to the high correlations among the four contextual variables, which produced a high level of multicollinearity. The correlation coefficient is -.70 between Black density and social capital, and -.62 between social capital and diversity, .46 between Black density and racial diversity. These coefficients are all significant at .01 or .001 level. Thus, it is extremely difficult to assess the independent effect of each independent variable on Obama’s White voter support.

Second, to deal with the problem of multicollinearity, one method that has been used frequently in the literature is to drop one or a set of independent variables from the model in order to test the unique effects of the remaining independent variables. However, as Kevin Arceneaux and Gregory Huber (2007) convincingly demonstrated through their state-level simulation data analysis, omitting variables from a regression model in an effort to eliminate MC [multicollinearity] is a cure certainly worse than the disease because of the bias this method will certainly create. Moreover, “especially in small samples [state-level analyses], where even moderate MC can make finding statistically significant coefficients for individual covariates with a real effect on Y less likely, the lack of statistical significance of a coefficient estimate should not be taken as grounds for concluding that the variable has no effect” (Arceneaux and Huber, Reference Arceneaux and Huber2007, p. 97). Because of these problems, the previous competing hypothesis approach leads to an intractable question about the real impact of the four contextual variables on Obama’s White voter support. We now turn to the empirical solution to this multicollinearity problem based on our new racial tension approach discussed above.

THE NEW RACIAL TENSION APPROACH AND THE EMPIRICAL EVIDENCE

As stated, this research proposes a new racial tension approach, which suggests that a state’s racial tension, as a latent variable, is revealed through both the racial makeup (Black density and racial diversity) and community norms (political culture and social capital) of the state (see Figure 1). The first task empirically therefore is to show that indeed the observed high correlations among the four contextual variables are due to a deeper level of racial tension. Using principal component method, Table 2 shows that there is truly an underlying pattern reflected by factor one. All the four contextual variables measuring Black density, social capital, state political culture, and racial diversity are strongly clustered into factor one, which explains more than 62% of total variance. The loadings for the four contextual variables all have much higher values (i.e., more than .804 absolute values) than the conventional minimum values of significant factor loading of .5 (Guadagnoli et al., Reference Guadagnoli and Velicer1988; MacCallum et al., Reference Maccallum, Widaman, Preacher and Hong2001; Stevens Reference Stevens2002). Thus, an underlying latent variable does exist, and all four contextual variables are correlated due to their reflections on this underlying factor.

Table 2.

Latent Racial Tension in American States

* indicates that loading is greater than .5

The signs of the loadings provide more clues about the nature of this underlying factor. The four variables are linked to factor one in a way that reveals the racial tension of states. To see this, according to Table 2, factor one is negatively correlated with social capital and state political culture (in the order of from traditionalistic to individualistic and further to moralistic), which suggests that a higher level of racial tension (i.e., larger factor one score) will lead to less social capital (i.e., less interpersonal trust and civic engagement) and the tendency to adopt the traditionalist political culture (i.e., the existing political order). On the other hand, factor one is positively correlated with Black density and racial diversity, which shows that a higher level of racial tension (i.e. factor one score) will enhance Black density and racial diversity. Therefore, it is logical to interpret factor one as a state context measuring the underlying level of racial tension (see below for a further empirical test of state-level latent variables). The standardized factor one scores thus represent states’ various levels of racial tension (range=-1.39 to 2.49, mean=0, sd=1).

To prove that the latent factor is indeed racial tension, one can also trace the spatial distribution of racial tension to see whether it confirms qualitatively the findings suggested by the recent political development literature.^{Footnote 8} More specifically, the five Deep South states of South Carolina, Georgia, Alabama, Mississippi, and Louisiana exhibit the highest-level of racial tension (racial tension scores of 1.73, 1.82, 1.82, 2.49, and 1.85, respectively. Note that the greater the value of this factor score, the higher the racial tension). As expected, the four contextual variables indicate that these five states are indeed those with the greatest Black density, traditionalistic political culture, a high level of racial diversity, and low social capital. At the other end of racial tension spectrum are eleven states where Black density and racial diversity are low, social capital is high, and state political culture is oriented to moralistic direction, rather than traditionalistic existing order. These eleven states are mostly located in the northern part of the country including Wisconsin (-.98) and Minnesota (-1.39). Between the most racially confrontational states of the Deep South and the low racial tension states of North are the three layers of racial tension from the Southeast to the Northwest. These three layers of intermediate states reflect the complexity of hybrids of the four contextual variables.

It is important to note that our racial tension latent variable should not be considered as just a proxy for state location. Southern states, in particular, have a higher level of racial tension score, but it is not because they are located in the South. Our racial tension has no geographic attributes per se. The score is derived from no geographic variables. To be more specific, racial tension, as a latent variable, is constructed based on our four contextual variables on racial diversity, Black density, social capital, and political culture. It is also important to note that neighboring states do not always share the same racial tension level. For example, North Carolina (a racial tension score of 1.07), which shows a mixed political culture (traditional moralistic), has a lower level of racial tension than the other five Deep South states with traditionalist political culture. In the Mountain West, Wyoming (-.42) and Idaho (-.44), the two states with relatively low social capital and early racial antagonism within the immigrant labor force, are also the two states with a relatively higher level of racial tension than Colorado (-1.09), Montana (-.91), and homogenous and high-social capital Utah (-1.20).

Certainly, the value of racial tension is not just the classification of states that these factor scores can provide. To show the continuing effect of racial tension, Hypothesis 5 suggests that racial tension has a direct and negative impact on Obama’s White voter support. Based on this hypothesis, for instance, Obama is expected to have more White support in Colorado and Montana than in Wyoming and Idaho, though they are all Mountain West and red (Republican) states. The following will show the empirical tests of Hypothesis 5, and demonstrate why racial tension had a direct and negative impact on Obama’s White support in states in both presidential elections.

RACIAL TENSION AND OBAMA’S WHITE VOTE IN 2008 AND 2012

With Obama’s White vote as the dependent variable, robust regression analysis can offer much more convincing evidence about the effect of racial tension on Obama’s White support when other plausible variables are controlled for in the model. Robust regression is used here because our sample size at fifity is extremely small, which is especially sensitive to how errors are distributed. Any outliers or high-leverage observations may cause biased and inefficient estimates.^{Footnote 9} Two robust regression analyses of White racial voting based on the 2008 exit poll and the 2012 estimates of White support for Obama are performed to see whether a similar conclusion can be drawn. The models and findings are presented in Table 3.

Table 3.

Robust Regression of White Support for Obama in the 2008 and 2012 Presidential Elections: The Racial Tension Explanation

***p < .001; **p < .01; *p < .05; †p < .1

(two-tailed test)

Table 3 controlled for college education and age distribution of states, as Obama was reportedly able to draw significant support from young educated people at his record-breaking rallies.^{Footnote 10} During the 2008 campaign, the age of John McCain and his allegedly not conservative enough standings on issues also attracted much media attention. We therefore control for both states’ age group and ideological distributions, in addition to the political party and union influence in the states.^{Footnote 11} As shown in Table 3, the level of racial tension, derived from factor scores, is a robust explanation for Obama’s White vote in 2008 and 2012. Those states with higher levels of racial tension indeed provided a lower level of White support for Obama, controlling for other variables. Each unit increase in racial tension score, as indicated by the 2008 model, will reduce Obama’s White voter support by 7%, when other control variables are held constant. Among the controls, the senior resident ratio turned out to be a positive factor for Obama’s White voter support.

Model 2 of Table 3 lists the findings on Obama’s 2012 reelection. The racial tension latent variable continues to be statistically significant at the .01 level (two-tailed test). Each unit increase in racial tension score, as indicated by the 2012 model, will reduce Obama’s White voter support by more than 4%, when other control variables are held constant. Again, the finding is consistent with our Hypothesis 5. The control variables, however, present a more puzzling picture. No control variables come out of the robust model significant in 2012. Even the liberal distribution in states failed to predict White vote for Obama. This finding is striking, considering the fact that Whites were much less likely to support Obama in 2012 than in 2008. Figure 2, based on the exit poll data available for both presidential elections, shows that Obama’s 2012 White support declined in all states except Alabama which provided the lowest level of White support for Obama in 2008 (see the only observation in the Figure that is above the 45 degree diagonal).

Fig. 2.

Obama’s White Support in States based on the 2008 and 2012 Exit Polls

To find a better explanation for the decline in Obama’s White support in 2012, it is also necessary to reexamine the measurement construct of our latent variable, racial tension. Is it possible that this latent variable was discovered simply due to too few variables (five to be exact) included in our principle analysis shown above? Since Model 1 of Table 3 suggests the importance of ideology, it is important to include more ideology measures in our further latent-variable analysis to examine whether the four contextual variables still exhibit high loadings surrounding the same underlying latent variables. To do so, we borrow the idea from previous studies that highlighted the importance of not only ideological but also religious coalition in the past presidential elections (Putnam and Campbell, Reference Putnam and Campbell2012). In particular, we will include the religious measures into our latent analysis. As repeatedly suggested in the literature, “From 1968 to 2008, Republicans controlled the White House for more than twice as many years as Democrats, a dominance made possible by a movement toward the GOP by southern evangelical Protestants and, to a lesser extent, Roman Catholics. . .” (Wald and Calhoun-Brown, Reference Wald and Calhoun-Brown2011, p. 101).

A CONFIRMATORY FACTOR ANALYSIS

We further performed a confirmatory factor analysis (CFA) based on the maximum likelihood approach. The result is shown in Table 4. The eight observed variables are linked highly with the three latent factors. Again, the four contextual variables are attached to the first factor with very high level of factor loadings, while the second factor reveals the latent variable leading to religious and ideological indicators. The third factor also captures 20% of the variance, this factor, seemingly about religion, does not carry much unique information. We therefore focus on the first two factors. The most important clue to discover the nature of a latent variable is the size of factor loading as well as its sign.

Table 4.

Factor Analysis of American States based on Maximum Likelihood and Varimax Solution

Three factors are sufficient, p=.11.

Empty entries indicate |f actorloading|< .37.

Once again, the four contextual variables (Black density, diversity, political culture, and social capital) are simultaneously linked to the underlying latent variable, factor one, which provides additional support for our earlier findings. Racial tension existed even when ideology and religion were taken into consideration. To better understand Table 4, it is also important to examine the signs of the factor loadings for the indicators, which are shown as opposite to those in Table 2, once eight variables are entered into CFA simultaneously. If the factor one in Table 2 is about racial tension, then the factor 1 in Table 4 is about racial tolerance.^{Footnote 12} In other words, they measure the same underlying racial relation status at the state level, though they show them from different directions. To be more precise, here Black density and diversity are negatively related to racial tolerance, that is, a low level of racial diversity and Black density should reveal a racially tolerant state context, whereas a low level of social capital and a traditionalist-government political culture exhibit a racially intolerant state context. It is also important to note that given the signs of the factor loadings for factor two in Table 4, clearly it is a latent variable concerning how liberal (and nonreligious) a state context is in terms of its overall state’s ideological and religious distributions, as indictors of percent liberal and percent secularism are positive while the other two indicators are negative.

RACIAL TENSION, LIBERAL IDEOLOGY, AND OBAMA’S WHITE VOTE

Table 5 uses the factor scores derived from Table 4 and shows the relationships between the two newly constructed latent variables and Obama’s White vote in 2008 and 2012.^{Footnote 13} One clear advantage of using the latent variables in the model, as discussed above, is that it minimizes, if not completely eliminates, the muliticollienarity problem commonly seen in state-level analyses. Compared to Tables 3 and 1, many of the highly correlated variables now are represented by the two latent variables in Table 5. For the 2008 model, racial tension is a significant predictor of Obama’s White vote. For each unit increase in racial tension score in a state, controlling for other variables in the model, Obama lost about 3% votes from White electorate. In short, racial context mattered for Obama’s 2008 state-level White coalition.

Table 5.

Robust Regression of White Support for the Democratic Candidates in the Presidential Elections, 2004–2012

***p < .001; **p < .01; *p < .05

(one-tailed test)

Racial tension is also statistically significant in Obama’s 2012 reelection. The significance level is at .05 for a one-tailed test (in other words, it is assumed to be negative by our Hypothesis 5). Again, based on the magnitude of the regression coefficient racial tension (measured by per unit increase in its factor score) is shown to reduce the White vote by 3% in 2012. The second latent variable, which reveals how states are closely connected with the liberal ideological and (non)religious belief, turned out to be an enduring and constant factor in Obama’s White racial coalition. An unit increase in the liberal factor score will produce about 6% White votes for Obama in both 2008 and 2012. This is an important finding. It should be noted that our ideological measurement construct is built on indicators of both ideology and religious attachment (i.e., church attendance and secular belief). In 2012, Obama faced challenges from his Republican opponent, Mitt Romney, who was the first-ever Mormon Republican nominee in U.S. history. Obama would have been able to capture more White votes, if Romney’s religious affiliation became a liability for the GOP candidate to gain religious White conservatives. Model 2 of our Table 5 shows that Romney was able to maintain the White conservative coalition, in spite of news coverage on his religion. Obama, on the other hand, did not break into this White GOP conservative stronghold, as his White votes were mainly from the liberal bloc.

Table 5 also suggested that Obama’s White supporters were from those states with more union memberships, and more college-educated areas, consistent with previous findings (Mellow Reference Mellow and Nelson2014). The states with more senior citizens, however, did not support him as much in 2012 as in 2008. This finding shows that some of Obama’s policy victories in his first administration, such as health care reform, did not bring him more White votes in the areas where he had hoped to bring a change, especially among senior citizens.

RACIAL TENSION, SOUTHERN STATES, AND WHITE OPPOSITION TO THE DEMOCRATIC PARTY

Table 5 is constructed based on mainly our Hypothesis 5 that treats the four contextual variables as the indicators of an underlying latent variable racial tension, which as shown in Table 5 significantly reduced White support for Obama in both 2008 and 2012. This empirical finding is robust given the small N and a total of sixteen variables in the regression equation either through the latent factors or as covariates. Two more important questions remain, however. First, the effect of racial tension on White opposition to Obama might be in fact caused by Whites’ disapproval of the Democratic party concerning its racial and other social welfare policies, or its stand on abortion, gay marriage, and other social issues (Liu Reference Liu2010; Putnam and Campbell, Reference Putnam and Campbell2012). In other words, is it possible that the White opposition to Obama discovered so far in this article might be targeted at his party, not himself as the nation’s first “electable” Black candidate? Second, as indicated above the southern States, especially the Deep South, exhibited the highest level of racial tension based on our factor scores. Is it possible that the impact of racial tension on White opposition to the Obama candidacy is actually a product of southern Whites’ long-time history of racial discrimination and political institution against African Americans (Bartels Reference Bartels2008; Springer Reference Springer2012)?

Table 5 provided clearly a “No” answer to both of these questions. First, the 2004 Democratic John Kerry’s White support was regressed based on exactly the same regressors, and the result based on the third model in Table 5 showed that the racial tension latent variable is no longer statistically significant. This finding thus clearly suggests that Whites’ opposition to Obama in 2008 and 2012 went beyond their “usual” partisan positions, and the Obama candidacy indeed triggered the latent racial tension to negatively affect his overall level of White support. In contrast, racial tension did not play any role in John Kerry’s White support in the 2004 presidential election, which featured two White male candidates from the major political parties. Second, note that all three equations of Table 5 contain the “South” control variable. In all three presidential elections whether a state is located in the South had no effect on White support for the Democratic Party once we entered racial tension into the regression models. Put differently, our racial tension variable is a much more robust factor in affecting the level of White support than the “South” variable. To understand Southern states’ recent anti-Democratic Party voting record, one has to go beyond simple geopolitics, and pay attention to the candidate characteristics, especially their racial identities.