INTRODUCTION
Intersection theorists such as Kimberlé Crenshaw, bell hooks, and Patricia Hill Collins have long argued that women of color experience multiple oppressions, all of which are mutually constitutive (Crenshaw 1991; Hill Collins 2000; hooks 2000). Although many feminist theorists have accepted this formulation, intersectional analysis has rarely made its way into empirical work within political science (some exceptions are Cohen 1999; Dawson 1994, 2001; Hochschild 1995; Jones-Correa 1998; Leighley 2001; Tate 1993) for two reasons. The first impediment has been normative and, to some extent, epistemological. Many scholars adopting the intersection approach believe that issues of identity and marginalization cannot be “measured” and that to do so would constitute the reification of what are, in fact, contested and fluid categories (Hartsock 1990; Hill Collins 2000; hooks 2000; Spivak 1988).1
We do not mean to dismiss the validity of this argument, and acknowledge that the “quantification” of these categories has important theoretical and normative implications. We believe that the lack of these kinds of concerns in empirical work has led to the absence of the experiences of women of color from much of political science discourse. While any quantitative analysis of these categorizations is by definition imperfect, we believe the attempt is better than the alternative, which is ceding the empirical field.
Because of the diversity of the California electorate, the 2003 California recall election provides a unique opportunity to explore differences in vote choice among women and men from different racial groups. In addition, the recall itself constituted a unique political context that provides an especially rigorous test of intersection theory. In particular, race could have been expected to be less salient and gender issues more salient to voters, leading one to expect more similarities than differences in women's voting patterns. First, according to political commentators, the recall election had little to do with partisanship. Both incumbent Governor Gray Davis's inadequacies and Arnold Schwarzenegger's candidacy as Davis's replacement were cast as being more about the characteristics of the individuals themselves than about their party affiliations. Since party identification is also highly correlated with race in California (as it is across the United States), one could extrapolate from this that race could have been less salient in this election than it is normally in California elections. Second, because of the allegations of Schwarzenegger's sexual misconduct that were published in the Los Angeles Times shortly before election day, many expected gender to play a role in the election, with men and women reacting very differently both to Schwarzenegger's candidacy and to those allegations. Thus, a recall election of this kind could be expected to have few conventional cues for voters, making intersection potentially easier to observe. Our goal in this study is to use this particular electoral context as an empirical test case for intersection theory.
RACE, CLASS AND GENDER IN POLITICS
Previous studies have shown important racial and gender differences in American political behavior. In terms of race, studies have found significant racial polarization in black and white voting patterns (Collet 2004; Grofman, Handley, and Niemi 1992; Shaw 1997) and in public opinion (Dawson 2000). In California, specifically, research has shown that the state's electorate is consistently divided along racial lines in terms of voting for candidates and propositions, as well as in political party registration (Alvarez and García Bedolla 2004; Cain 1991; Hajnal and Louch 2001; Jackson 1991; Regalado and Martínez 1991; Tolbert and Hero 1996). In terms of gender, studies have shown differences between men and women in public opinion (Shapiro and Mahajan 1986; Wilcox, Ferrara, and Allsop 1993), partisanship (Box-Steffensmeier, De Boef, and Lin 2004, Kaufmann and Petrocik 1999), and presidential voting patterns (Kennedy Chaney, Alvarez, and Nagler 1998).
Yet largely due to small sample sizes, few studies have examined gender differences within racial groups, and their findings have varied significantly. Most studies identifying a gender gap in American politics have relied on all or mostly white samples. Susan Welch and Lee Sigelman's (1992) comparative analysis of the gender gap among Latinos, African Americans, and Anglos found that gender differences were similar across all three groups. The racial gap was greatest with respect to ideology, but minimal with regard to party identification and vote choice (ibid., 194). These findings differed from those of Welch and Sigelman (1989) in previous work. Other studies of gender and black party identification have had conflicting results (Dawson 1994; Mangum 2004; Tate 1993). Studies using the Latino National Political Survey showed no significant differences between Latino men and women in terms of their party identification, and a recent study using 1999 survey data had similar results (Alvarez and García Bedolla 2003; García, Falcón, and de la Garza 1996). In terms of gender and Latino public opinion, Lisa Montoya (1996: 261–62) conducted the most thorough analysis to date, and found gender differences varied among different Latino national origin groups. Across all Latino groups, the most significant “gender gap” she found was with regard to women's roles, with Latina women more supportive of nontraditional gender roles than Latino men.
Thus, we know about racial differences and gender differences in political behavior, but little about how race and gender interact (some exceptions are Gay and Tate 1998; Mansbridge and Tate 1992; Welch and Sigelman 1992). With regard to class, because the socioeconomic status (SES) model has been the dominant paradigm in studies of American political behavior since publication of The American Voter, the area of intersection we know most about is how race or gender relate to class (Campbell et al. 1960; Verba, Schlozman, and Brady 1995; Wolfinger and Rosenstone 1980). With regard to race, Tate (1993) found that education and income are only occasionally related to African-American participation, and studies of Latinos have found that SES can only explain part of the gap between Latino and Anglo electoral and nonelectoral participation (F. García, Falcón, and de la Garza 1996; J. García 1997). Regarding gender, Nancy Burns, Kay Schlozman and Sidney Verba (2001) argue that two of the six competing hypotheses to explain women's voting behavior emphasize socioeconomic issues, positing that women's generally lower levels of income, education, and occupational status are what decrease their participation. They test these hypotheses and find that due to differences in overall political engagement and access to politically relevant resources, men are more interested, informed, and efficacious about politics than women. In another study, Kay Schlozman, Nancy Burns, Sidney Verba, and Jesse Donahue (1995, 285) found the greatest gender differences in political activity among the poor, suggesting that class plays an important role in women's political behavior.
Yet these findings raise the theoretical question of how exactly scholars see group memberships, like race and gender, operating in American politics. Is it, as the SES model contends, simply that race and gender act as a proxy for socioeconomic indicators, and that those factors are in fact driving voting patterns? Or do we believe that race, gender, and experiences of marginalization as sociological phenomena affect how particular groups of people see and interpret the political world in a way that is fundamentally different from, for example, age, union membership, or ideology (Omi and Winant 1994)? The sociological approach is similar to Virginia Sapiro and Pamela Conover's (1997) distinction between “positional” and “structural” approaches to the study of gender and political behavior. They argue that the positional view assumes that women's and men's political beliefs, values, or issue positions differ, leading them to make different electoral choices. The structural explanation posits that gender will gain political significance through the structure of relationships among political attitudes, values, and judgments. In other words, even if men and women hold exactly the same policy positions, the value and weight they place on each issue, because of their life experiences, will vary, making it possible for men and women to vote similarly, but for very different reasons (Sapiro and Conover 1997, 498).
Our understanding of how race and gender operate in U.S. politics is further complicated if we attempt to incorporate the political effects of multiple group experiences, such as race, class, and gender, into our analysis. With an approach similar to the structural position, Kimberlé Crenshaw (1991) was among the first to define the idea of “intersection,” arguing that categories like race, class, and gender are socially constructed characteristics that differentially shape people's experiences, which in turn color their political attitudes, opinions, and behavior (Cohen 1999; Crenshaw 1991; Hill Collins 2000; hooks 2000). One of the core aspects of this theoretical framework is that these social constructions are not hierarchically ordered, but instead mutually constitutive. In other words, women's experiences as women result in political responses that are different from men's. Likewise, racial and ethnic experiences result in political responses that vary among races. As a result, the intersection of these identities within a single individual also should yield varied responses. As race, class, and gender interact, we can expect to witness behavior that cannot be accounted for simply by locating only one or the other of these characteristics. Instead, intersections theorists argue, we must consider the combination of gendered racial and class experiences.
If these scholars are correct, it is not sufficient simply to look at whether or not a race or gender “gap” exists; one must ask if different factors are driving differences, or similarities, in vote choice. By analyzing men and women from different racial groups using separate models for each, this study attempts to move closer to that type of analysis.
RACE, GENDER, CLASS, AND THE 2003 CALIFORNIA RECALL ELECTION
In early 2003, an antitax group called the People's Advocate spearheaded the campaign to oust then-Governor Gray Davis, a Democrat. Despite his having won reelection in November 2002, recall advocates believed that Davis's low approval ratings, spurred by the state's budget crisis and his handling of the electricity crisis, made the sitting governor vulnerable to removal. Three California political parties endorsed the recall effort—the Republican Party, the Libertarian Party, and the American Independent Party. The recall campaign received a boost in May 2003 when Congressman Darrell Issa (R-San Diego) launched his own recall effort, backed with substantial financial and organizational resources. As a result, by July 2003, recall proponents had gathered enough signatures to force a special election, which was set for October 7, 2003. In August 2003, actor Arnold Schwarzenegger, a Republican, added his name to the list of 135 possible replacement candidates, and quickly became the front-runner. In his campaign, he minimized his party affiliation and campaigned on a platform of changing “politics as usual” in Sacramento. In the end, 55.4% of California voters supported Davis's recall and 48.7% voted for Schwarzenegger. Lieutenant Governor Cruz Bustamante won the second highest number of votes, with 31.6%. As a result, Davis was removed from office and Arnold Schwarzenegger was sworn in as California's thirty-eighth governor on November 17, 2003.
California's 2003 recall election was unique on a number of levels. First, it was historic. Although the initiative process is a common part of California politics, no sitting governor had ever been removed from office using the recall process until Gray Davis. Its historic value makes the recall election an interesting focus of study, but also makes generalizability difficult. To our knowledge, no work has looked specifically at vote choice in recall elections per se; much work, however, has analyzed voting and the initiative process (Bowler and Donovan 1998; Bowler, Donovan, and Tolbert 1998; Gerber and Lupia 1995; Lupia 1994; Magleby 1984). Little of this work has looked at racial issues and initiative voting (some exceptions are Donovan and Bowler 1998; Gamble 1997; Hajnal, Gerber, and Louch 2002; Hajnal and Louch 2001; Tolbert and Hero 1996). Barbara Gamble (1997) found that members of racial groups vote similarly to whites on ballot propositions, except in cases where the proposition deals with civil rights questions, but her findings have been contested by others (Donovan and Bowler 1998; Frey and Goette 1998; Hajnal, Gerber, and Louch 2002), leaving the question open to debate. Even less of this scholarship has been concerned with gender issues, except in mainly a tangential way, such as in relation to voting on abortion rights or initiatives and women's right to vote (Banaszak 1996; Gerber 1996). Thus, it is difficult to hypothesize how we can expect the interaction of race, class, and gender to affect recall voting from the extant literature.
Nonetheless, what we know about vote choice in regular elections and initiative voting does allow us to make some educated guesses. Much of the work on initiative voting has focused on information and vote choice, particularly how voters address the lack of information and traditional political cues in making their choices on ballot propositions (Bowler and Donovan 1994; Gerber and Lupia 1995; Lupia 1992). These studies suggest that voting behavior on ballot propositions is similar to voting behavior in candidate elections, particularly in terms of the role of party identification (Branton 2003; Bowler and Donovan 1998; Smith and Tolbert 2001). What is unclear is the relationship between party identification and ideology, given a focus on voting on issue-specific propositions that cue particular liberal or conservative ideologies (Branton 2003, 368). Thus, scholars have found that voters' political propensities, as reflected in their ideological and partisan orientations, affect their vote choice on ballot propositions.
The recall vote, however, was different from a typical ballot initiative, insofar as a vote for the recall was not issue- or party-specific. In other words, voters could not look to a particular partisan or policy position to determine the direction in which they should vote; many Democrats supported recalling Gray Davis, and no one single policy issue or concern drove the movement to oust him from office. Thus, we believe that the recall election could have been motivated by a desire for a “time out” from traditional partisan politics in California. The race for a replacement candidate was multicandidate and multiparty, and support for the recall was related to party, but crossed party lines. Previous research has shown a strong relationship between party identification and race in California (Alvarez and García Bedolla 2004; Cain 1991; Hajnal and Louch 2001; Jackson 1991; Regalado and Martínez 1991; Tolbert and Hero 1996). Thus, an election that cues party identification also cues racial identification. If party identification was less salient in this election, then the 2003 California recall election provides a unique opportunity to explore the role of race in politics in an election that, on its face, minimized that role.
The unique context of the 2003 California recall also opened the possibility that the opposite was true with regard to the role of gender, namely, that gender would have been more salient in this election. Although we know little about the role gender plays in initiative voting in particular, work on gender and electoral behavior in general has suggested that gender gaps are largely the product of a particular political context (Sapiro and Conover 1997, 499). Virginia Sapiro and Pamela Conover (1997) argue that in these instances, candidates, campaigns, or the media can “prime” the influence of gender by framing specific issues and making them more salient to voters. In particular, they argue that gender differences in voting emerge most between men and feminist women (ibid., 500). The sexual harassment charges that surrounded the leading replacement candidate in the recall, Arnold Schwarzenegger, could easily have served this cuing function for the general public and feminist women, in particular. Shortly before the election, the charges were publicized in the state's largest newspaper, the Los Angeles Times, and they received considerable attention in print, television, and radio news. In addition, the nature of the charges, sexual harassment, meant that feminist organizations, such as Code Pink in Los Angeles, held marches and other events to bring attention to the issue, thus cuing the consciousness of feminist voters. Although this study does not look at candidate choice per se, given that Schwarzenegger was the front-runner, attitudes toward these charges likely affected attitudes toward voters' willingness to oust Gray Davis from office.
Therefore, the 2003 California recall election provides a unique electoral context where, on the one hand, the role of party identification and race could have been minimized while, on the other, that of gender could have been maximized. That given, one would assume that this was an election when we could expect more similarities across women in general than within the members of racial groups, making it an excellent test case for the theory of intersection. Our assumption is that intersection would be less likely to appear in this election than in other elections. To test that proposition, we formulated four hypotheses.
Hypothesis 1: Despite the uniqueness of the recall election, partisanship, and therefore race, as has been found for initiative voting in general, will remain the most salient factor in recall vote choice.
Hypothesis 2: Because of the gendered cues in this particular election, voting patterns will fall along gendered lines, with women across the racial groups voting more similarly than men and women within racial groups.
Hypothesis 3: In support of the SES model, class will be the most salient factor, with income and education operating similarly across groups to drive vote choice.
Hypothesis 4: None of these factors—race, class, or gender—will operate the same across all groups, suggesting that the intersection theorists are correct in positing that the meaning and effects of group memberships are fluid, varied, and likely contextually driven.
DATA AND METHODS
In order to test these hypotheses, we employ logistic regression analysis using data from the Los Angeles Times Recall Election Exit Poll conducted on October 7, 2003. The total sample consists of 5,205 registered voters from 74 polling locations across California, weighted using demographic data and actual returns to reflect absentee approximations. The sample included 3,367 white respondents (1,724 male, 1,643 female), 245 black respondents (130 male, 115 female), 509 Latino respondents (260 male, 249 female), and 244 Asian respondents (131 male, 113 female). The main strength of this data set is that the sample is large enough to cross-tabulate California voters by race and gender, something that is often impossible with smaller surveys. Exit polls do have important methodological problems, particularly when used to examine racial groups and individuals with low socioeconomic status. Given that exit polls include only a subset of voters at the polls, they tend to undersample the members of racial groups and oversample more affluent voters. Since the California electorate is more white and affluent than the state population, those responding to exit polls also are more white and affluent (Ramakrishnan and Baldassare 2004). Despite these important limitations, we believe that the size of the survey and its geographic scope across California mitigate these problems to some extent, making it appropriate for use in this analysis (Levy 1983).
Because our dependent variable is dichotomous, we use a logistic analysis to predict the probability of vote on the recall by first constructing a model that includes all groups and then creating models for each gender and racial group separately. In order to keep our models as parsimonious as possible, our independent variables include gender, race, education, income, age, party registration, and political ideology.2
Gender and vote on the recall were recoded as dummy variables for all models. Race was recoded to include only white, black, Latino, and Asian respondents. Dummy variables for race were constructed and used in all of the models. The original categories were kept for education but recoded as dummy variables: No High School, High School Degree, Some College, and College Degree and Above. “Age” was recoded into the following categories: 18–29, 30–44, 45–59, and Over 60. Each age category was then constructed into a dummy variable. For “Income,” we maintained the original categories but recoded them into dummy variables: Below $20,000, $20,000–39,999, $40,000–59,999, $60,000–74,999, $75,000–100,000, and Over $100,000. Dummies for party identification include Democrat, Independent, and Republican. Political ideology was recoded to include only those who identified as liberal (Very Liberal and Somewhat Liberal) and conservative (Very Conservative and Somewhat Conservative). We then created dummy variables for each, one liberal and one conservative.
Although it is true that many Latinos self-identify as “white,” we consider Latinos to be a racialized group in the United States. It is their historical experiences of discrimination and current social construction as members of a “race” that make it appropriate, in our estimation, to treat Latinos as a racial, rather than an ethnic, group (Omi and Winant 1994).
We acknowledge that our control variables, particularly party identification and SES, are highly correlated with race. This is a reflection of the difficulty of quantitatively “measuring” what are interrelated social factors. We think that a finding of racial difference with these controls only strengthens the assertion that race has a strong, and independent, effect on vote choice in this instance.
In order to avoid methodological problems, we choose to look only at the vote for the recall itself, rather than candidate choice. We believe support for the recall can be treated as a freestanding vote, whereas candidate choice must be considered in relation to the recall vote.
Logistical regression is performed first for all groups combined, and then models testing for the effects of race and gender, respectively.6
All the models contain measures of class. In other words, we examine factors driving the vote for the recall among the selected groups in separate models. To do this, we examine seven models. Model 1 represents the full model, yielding coefficients with all gender and racial groups included. Three interaction variables were created for Model 1: black women, Latinas, and Asian American women. These interactions allow us to test whether there is an added effect of being a woman of color in comparison to the reference group in the full model. Model 2 investigates patterns among women and Model 3 among men. This allows us to examine distinctions among women of dissimilar racial groups and distinctions among men in different racial groups. Model 4 predicts vote on the recall for whites, while Models 5, 6, and 7 predict vote on the recall for blacks, Latinos, and Asian Americans, respectively. Breaking out the different factors in this way allows us to see if our models of voting behavior vary significantly across gender and/or racial lines.RESULTS: VOTE ON THE RECALL
We see in Table 1 that there was significant variation in the extent to which different racial groups and women supported the recall. In terms of gender overall, there was a significant gap between men's and women's vote on the recall, with 59.4% of men in support compared to 51.4% of women. Yet once those numbers are broken down by the different racial groups, we see quite a bit of heterogeneity among women and men. In general, whites supported the recall at a much higher rate than any other racial group: 59.5%. Blacks, on the other hand, were the least likely to support the recall, with only 21.5% voting in favor. Latinos and Asians fell in between these two extremes, favoring the recall at 44.8% and 47%, respectively. In terms of education, we see the most support from groups with a high school degree (61.8%) and some college (63.5%). Vote choice does not vary much by income and age, with most groups supporting the recall by a moderate majority. Contrary to our expectations, partisanship and political ideology were strong determinants of recall vote choice. Republicans favored the recall by 88.2% compared to only 23.6% of Democrats. Conservatives voted yes by a margin of 85%, but only 24.7% of liberals cast “yes” votes. Thus, as has been found for initiative voting generally, partisanship affected vote choice in this election, despite the presence of multiple replacement candidates and a front-runner who tried to distance himself from a “party” label.
Vote on the recall by independent variable
Table 2 includes Models 1, 2, and 3 and reports the β and Exp(β) logistic coefficients for each variable with standard errors in parentheses.7
The reference group for the dependent variable is a “Yes” vote on the recall. The reference group for gender is male, for race is white, for education is college and above, for partisanship is Republican, for income is over $100,000, for age is over 60. The interactions are in reference to white males. Only valid cases are included in the models.
In regards to Model 2, which includes only women, we see a significant difference between support for the recall for black and Asian American women in reference to white women, with both less likely to support the recall by 50.3% and 47.3%, respectively. In Model 3, we detect statistical significance in support for the recall between black and Latino men, with both less likely than white men to cast a yes vote by 62.6% and 55.5%, respectively.8
In Models 2 through 7, education, party identification, and political ideology behave similarly to the full model, and are not directly discussed here.
Logistic regression results for voting Yes on the recall for all groups and gender models
Table 3 reports the results from the race models: Model 4 consists of whites, Model 5 of blacks, Model 6 of Latinos, and Model 7 of Asian Americans. We only see a statistical association between gender and vote choice among white men and women. After controlling for education, age, income, partisanship, and political ideology, white women were approximately 25% less likely than white men to vote yes on the recall. On the other hand, we do not detect a statistically significant difference among black, Latino, or Asian American men and women, even though men in these groups favored the recall slightly more than women. The absence of statistical significance for these groups could be due to the small sample size in each category. We would like to note that in Model 6, the Latino model, the direction of the coefficient is positive, indicating that Latinas were more likely than Latinos to vote yes on the recall. While this result is not significant, it does connect with the results of the interaction term included in the full model, as noted earlier. In general, the results of the disaggregated racial models are consistent with other research and offer partial support for our first hypothesis in that we may be observing racially polarized voting on the recall, even after controlling for partisanship. Race appears to have been quite relevant in this election, as has been found in previous California elections, suggesting that the structural explanation is more accurate and that racialized experiences affect how members of racial groups make political choices, even absent overtly “racial” issues in a particular election.
Logistic regression results for voting Yes on the recall for race models
Considering the evidence from all seven models, we find partial support for Hypotheses 1 and 2. Voting on the recall did vary significantly across racial groups, and we find evidence of racially polarized voting. Yet we also find important gender differences in voting, particularly among whites. Both race and gender were salient in this election. In terms of class, education and income seem to have had different effects across the different racial groups, and between women and men. Thus, although we find partial support for the first three hypotheses, we find the strongest support for Hypothesis 4: that the effects of race, class, and gender vary across groups in important ways, leading to differences in how those group members interpret their political choices. In other words, we find empirical evidence that intersection exists, and that it affects individual vote choice.
DISCUSSION AND CONCLUSION
The California recall election was unique in that the recall mechanism had never before been used to remove a sitting governor from office. The media and political pundits argued that the novelty of voters' having multiple replacement candidates on the ballot and a two-step voting process meant that the normal rules of elections, and conventional political cues, did not apply in this instance. This analysis suggests that was not the case. Even though California Democrats were slow to defend Gray Davis, and Arnold Schwarzenegger made efforts to disassociate himself from any party, voters clearly saw the election from a partisan standpoint, with Democrats much less likely than Republicans to vote to remove the governor. This suggests that traditional political cues remain important even in atypical elections. Additionally, our analysis shows that members of racial groups, male and female, saw and voted on the recall quite differently, suggesting that the “structural” interpretation of gendered experiences also applies to that of racial groups, as intersection theorists would lead us to expect. Thus, experiences of marginalization (for racial groups) and privilege (for whites) seem to affect how group members, male and female, make political choices.
These findings inform the literatures in both race and gender politics in important ways. The previous analysis shows that race, class, and gender interact, and affect vote choice in this case. In addition, we find their effects varied among women and among racial groups. As such, this analysis provides important empirical support for the idea of intersection. Intersection theorists have long argued that the experiences of women of color are qualitatively different from those of white women. Historically, these women have had very different experiences in the United States, and so it is intuitively logical that they would also have different attitudes toward politics. Yet this claim has been subjected to few empirical tests (some exceptions are Dawson 2000; Gay and Tate 1998; Tate 1993; Welch and Sigelman 1992). The fact that we find important differences across groups and across women provides rare empirical support for the contention of intersection and other theorists that multiple marginalizations vary across groups, time, and circumstance.
Additionally, our analysis provides an important methodological insight. Given that intersection theory presumes an “interaction” among multiple variables, we decided to test the proposition of using an interaction term to measure these effects, and included these terms in Model 1. The result is that these terms told us little about how gender was operating within each of these racial groups. To begin with, we had the statistically significant finding regarding Latina voting. Although this finding, at least from a directional standpoint, was supported in the Latino model, it is unlikely that the effect was as strong as the interaction term led us to expect. More problematic is the fact that the coefficient for the female/black interaction term in Model 1 is positive, suggesting that black women were more likely to vote in favor of the recall, yet the female coefficient is negative within the black model. These idiosyncrasies are likely a product of fairly small sample sizes. They suggest, however, that interaction terms may not be the best way, at least statistically, to get at finding intersection. We believe that looking at each group separately is a more fruitful avenue for these kinds of analyses.
Our work makes a theoretical contribution. Although scholars in general have accepted the idea of categories such as race, class, and gender as socially constructed concepts, sometimes our discussion of these categories solidifies them in ways that can preclude more nuanced analysis. One example is the case of class. Because our measures of class are easily quantifiable indicators, there is a tendency to see these categories as absolute. Yet as Jan Leighley and Arnold Vedlitz (1999) point out, it is not necessarily reasonable to assume that a year of education will have the same effect on white attitudes as it would on African-American attitudes. The individual's family environment, historical experience, and socialization process will likely mitigate those effects. In addition, equal levels of income may not result in similar feelings of class status or economic security between whites and blacks. As Dalton Conley (1999) shows, family wealth, rather than income, more accurately measures people's actual economic worth and resulting feelings of economic security. In this study, we see that educational level and income had different effects on how the members of different racial groups made their vote choice in the recall election. Those with no high school degree were more likely to vote in favor of the recall, but that phenomenon was most true among whites. Stratifying the sample by race allows us to go a step deeper and, potentially, explore why white high school dropouts, in particular, were more in favor of ousting Governor Davis. If we were only to analyze the results for the sample as a whole, we would miss that important bit of information. Although aggregation may be most appropriate when answering some political questions, these findings show that disaggregating across multiple dimensions can also provide fruitful insights into how these categories interact to affect political decision making within individuals and among groups.
As such, disaggregating among and across groups can lead scholars to important information about how experiences of race, class, and gender affect the lived experiences of the members of different groups within American society. In particular, it would open up more analysis of the variation and differences that exist within racial groups. As we have seen, the literature has shown important differences in group political behavior. The theoretical explanation often is that these differences are due to variation in groups' historical experiences, in how they are socialized into American politics, and therefore how they interpret the world around them. That said, how do we go about explaining differences within groups themselves? To what extent do middle-class African Americans or Latinos vary from poor members of those communities? This study and others suggest that, at least for now, the most salient line within American politics continues to be the color line. The models for the different racial groups look quite distinct from that for whites. That is most likely because of the legacy of race relations in the United States, rather than anything intrinsic to race as a social category. As the U.S. political context evolves, it is possible that other lines of distinction—gender, class, sexuality—will become more salient.
This analysis should push scholars to consider more deeply why we believe group membership and identities matter in politics. We must consider why exactly we believe that race, class, and gender affect an individual's group consciousness, as well as political ideology. Is it group identity, marginalization, socialization, or some combination? Does one trump the other in particular contexts (Gay and Tate 1998; Mansbridge and Tate 1992)? If so, why? And if not, why not? As Michael Dawson points out, a racially stratified society creates “systematically different patterns of outcomes [that] shape individual life chances as well as the perceptions of society, thereby providing the basis for the huge racial gulf in public opinion” (2001, 4). We need to know more about what those patterns look like and how they vary within and among racial groups, especially in terms of class and gender. As political scientists, we have just begun to examine and address these questions.
Acknowledging the need to expand our understanding of the effects of marginalized (and privileged) identities and experiences in American politics returns us to the normative concern raised at the outset—namely, the problems with reification and essentialization that, in some ways, are inherent in the quantification of these categories. Some would, rightly, argue that given the socially constructed and contingent nature of these categorizations, it is impossible to arrive at an accurate “measurement” of their existence. We acknowledge that the survey data and variables we used to conduct this analysis are, in fact, quite rudimentary in this regard. Although we agree that this is an important point, and one that scholars should keep in mind when gathering and analyzing data, we believe a greater sensitivity to these questions could lead to the development of research designs that would come closer to the theoretical ideal. For example, Mustafa Emirbayer (1997) suggests that scholars look to the relational aspects of individual experience. Social psychologists have developed collective identity scales that do just that, including questions regarding feelings of social stigma and the respondents' sense of linked fate and group consciousness (Ethier and Deaux 1994; Luhtanen and Crocker 1992; Phinney 1989). If such scales could be modified to include multiple potential group identifications, they could yield important information about how individuals experience marginalization and/or privilege across multiple dimensions, and how those experiences can vary within and among groups. Here, intersection theory could inform our methods in important ways, enhancing our understanding of group dynamics within American politics. Race, class, and gender are important in politics because they help shape our view of the world and our understanding of our relative place within that world. Getting a better handle on the relational effects of that experience could be the first step toward developing a better understanding of how group memberships and experiences help shape the American political landscape.
