Reconciling Sexism and Women’s Support for Republican Candidates: A Look at Gender, Class, and Whiteness in the 2012 and 2016 Presidential Races

Much of the gender gap literature focuses on women’s greater average liberalism relative to men. This approach masks considerable heterogeneity in political identity and behavior among women based on race, class, and other key socio-demographic characteristics. In the 2016 Presidential contest, political divisions among women were evident in exit polling, which demonstrated that a majority of white women voted for Donald Trump. This was not an anomaly but reflects a more long-standing distinction between white women and women of other racial and ethnic identifications. In this paper, we draw on intersectionality and system justification theory as frameworks for exploring the distinctive political behavior of white women. Using data from the 2012 and 2016 American National Election Studies, we evaluate the factors that attracted white women voters to the GOP and kept them in the fold in spite of expectations that sexism in the campaign would drive women away from the party during the 2016 Presidential race. Our analyses show that many white women endorse sexist beliefs, and that these beliefs were strong determinants of their vote choice in 2016, more so than in 2012. Our findings also point to important divisions among white women based on educational attainment and household income in terms of both the endorsement of sexism and vote choice. These results shed new light on white women’s political behavior and An earlier version of this manuscript was presented at the 3rd New Research on Gender in Political Psychology Conference in New Orleans, LA, October 22–24, 2017. Thanks to the panelists for their helpful feedback. Electronic supplementary material The online version of this article (https://doi.org/10.1007/s11109018-9468-2) contains supplementary material, which is available to authorized users. & Erin C. Cassese Erin.Cassese@mail.wvu.edu 1 Department of Political Science, West Virginia University, Morgantown, USA 2 Department of Political Science, University of Kentucky, Lexington, USA 123 Polit Behav https://doi.org/10.1007/s11109-018-9468-2 Author's personal copy

we compare the effects of income and educational attainment on the vote choice of white men and white women, we gain further insight into the 2016 election. Contrary to popular narratives about Trump's victory, which centered on his strong base of support among economically-dispossessed white men, low-income white women displayed stronger support for Donald Trump in 2016 compared to lowincome white men.
Collectively, our results point to the need to revise contemporary theories of the gender gap, which tend to emphasize women's attraction to the Democratic Party while failing to account for the forces that keep a subset of women firmly in the Republican camp. Our findings contribute to a growing body of scholarship that seeks to better understand Republican women (Barnes and Cassese 2017;Deckman 2016;Norrander and Wilcox 2008;Ondercin 2013;Schreiber 2014), by demonstrating that whiteness, educational attainment, and class are important factors accounting for women's attraction to the GOP and for understanding heterogeneity among women voters more generally. Our work also contributes to this literature by explicating the role that sexism plays in women's political thinking and behavior. In addition, our results provide new insights into the importance of campaign context for understanding the political implications of sexism, as hostile sexism and perceived discrimination prove to be critical determinants of vote choice in 2016 but not 2012. We situate our finding in the context of subsequent elections with the hope of better predicting women's political behavior in future races where gender and sexism are prominent facets of campaign rhetoric.

White Women Voters Challenge Expectations about the Gender Gap in 2016
On the surface, analysts were correct in predicting one of the largest gender gaps in American history going into the 2016 presidential election. Exit polling revealed an 11-point gap, with 52% of male voters casting their lot with Donald Trump compared to 41% of female voters (CNN 2016). The average gender gap in Presidential elections between 1980 and 2016 was 8.1%, and only once prior, in 1996, did the gap reach this size (CAWP 2017). This seemed to track the dominant narrative about the role of sexism in the campaign, which argued that Trump's highprofile feuds with female celebrities and journalists, his comments about Carly Fiorina's face, his explicit and implicit gender-based attacks on Clinton's suitability for office, and the release of the Access Hollywood tapes would alienate female voters. In light of these campaign events, analysts speculated that Republican women could not be expected to remain in the fold-they would either be demobilized and fail to turn out or cross party lines to cast sympathetic votes for Hillary Clinton.
However, Trump's missteps with women along the campaign trail did little to push GOP women toward the Clinton camp. In exit polling, 88% of Republican women voted for Trump, a figure indistinguishable from the 89% of Republican men of who did so (CNN 2016). One of the most striking figures from the polling data concerned white women, 52% of whom voted for Trump. By contrast, only 4% of Black women and 25% of Latinas voted for Trump. Indeed, white women's political preferences were much closer to those of white men (62% of whom voted for Trump) than to other groups of women. Turnout was also up among white women (66.8% in 2016 compared to 65.6% in 2012) counter to speculation that sexism would demobilize women voters (Pew 2016).
Many analysts expressed surprise about white women voters because the conventional wisdom about the gender gap is that women are more liberal and more closely identified with the Democratic Party than men (Kellstadt et al. 2010;Ondercin 2017). This focus on average differences between the sexes obscures significant sources of heterogeneity among women (Barnes and Cassese 2017). Surprise at the behavior of white women voters reflects implicit assumptions about cohesion among women-namely that women voters feel a sense of solidarity with women candidates, and that women would respond in a uniform fashion to campaign events that were widely interpreted as sexist. These expectations reflect stereotypical thinking about women voters and point to the limitations of the existing gender gap literature, much of which considers gender as an isolated category rather than one of many categories simultaneously influencing political cognition and behavior (Brown and Gershon 2016).
Although the outcome of the race ran counter to expectations, white women's behavior in 2016 was not an anomaly, but reflects a consistent historical tendency for white women to vote for Republican presidential candidates (Junn 2017). As an illustration, Fig. 1 plots the two-party vote choice over time for all voters (Panel A), for white voters (Panel B), and for nonwhite voters (Panel C) using time series data from the ANES.
Panel A demonstrates that in every election since the early 1990 s, a majority of women voters supported the Democratic presidential candidate. Nonetheless, this same trend does not hold for white women voters. Rather, Panel B demonstrates that with the exceptions of the 1992 (H.W. Bush v. Clinton) and 1996 (Clinton v. Dole) elections, a majority of white women voters have supported the Republican presidential candidate. The same is true for white men (Panel B), though they did not experience the reversal observed for white women in the 1990 s. In recent years, voting patterns for white women are quite similar to voting patterns for white men. After the 2000 election, the percentage of white men voting for the Republican candidate averaged 59%, and the percentage of white women voting for the Republican candidate hovered around 54%.
Why is Political Cohesion Among Women So Low? Figure 1 highlights that women as a broad social group lack political cohesion and suggests that race has a cross-cutting influence on women voters. The attraction of white women to the Republican Party is obscured by most academic and popular discussions of the gender gap, which focus on average differences between men and women. Because women are on average more liberal than men (Norrander and Wilcox 2008) and more likely to identify with the Democratic party (Box-Steffensmeier et al. 2004;Kellstadt et al. 2010;Ondercin 2017), most theories of gender difference seek to explain the origins of women's liberalism, ignoring the factors that push millions of American women into the Republican camp (but see Barnes and Cassese 2017;Deckman 2016;Kaufmann and Petrocik 1999). As a result, extant theories of the gender gap focus on large-scale factors that promote liberalism among women, such as motherhood, diffuse social roles, greater economic vulnerability, and employment concentration in the social service sector Polit Behav (Diekman and Schneider 2010;Norrander 2008). Much of this work treats Republican women as a deviation from the political ''norm'' for women. The gender gap literature's tendency to focus on average differences between men and women has led to an inflated sense of cohesion among women, which may explain why predictions about women voters in 2016 went awry. Much of the research investigating gender as a political identity in and of itself-as gender consciousness, linked fate, or feminist identity rather than the gender gap-shows women vary considerably in their thinking about the political consequences of their own gender (e.g., Cassese and Holman 2016). Much of this work reports little evidence of political cohesion among women, with cohesion defined as ''the existence of shared attitudes, beliefs, and behavior among group members that can be directly attributed to group membership'' (Huddy 2003, p. 516). Cohesion is typically measured by the extent to which a group's common interests motivate common political preferences and political behavior, but women do not seem to place a greater importance on group-linked issues than men (Huddy 2003), or to hold distinctive opinions on many women's issues including abortion. In addition, there is little evidence that women emphasize women's issues when making their vote choice (e.g., Mansbridge 1985;Chaney et al. 1998;Manza and Brooks 1998). This lack of cohesion among women stands in stark contrast to attitudinal and behavioral cohesion among other social groups, such as African-Americans (e.g., Seltzer et al. 1997;Sapiro and Conover 1997).
In truth, there is much more heterogeneity than cohesion among women, and we can better grapple with it by adopting an intersectional framework (Crenshaw 1989;Collins 2002;Junn 2017). Research on women voters needs to shift from an intercategorical approach to intra-categorical approach to understand sources of diversity among women (McCall 2005), including factors related to white women's attraction to the GOP. There is a tendency not to think of white as a racial or ethnic category in the U.S., particularly among whites themselves, who tend to treat ''whiteness'' as a cultural default (e.g., Frankenberg 1993;Sue 2004). As a result, the motives, attitudes, and behaviors of white women are under-theorized (Levine-Rasky 2011). Research on diversity among women tends to focus on women of color, who are conceptualized as distinctive from a white norm or baseline that is rarely defined explicitly or put to critical scrutiny. This notion of a white baseline also assumes white women are relatively homogeneous in their political preferences. Thus, there is a need to look critically at the political behavior of white women and the factors that might unite or divide them.
Whiteness did feature prominently in another popular narrative about the 2016 election: that Donald Trump's appeal was concentrated among economicallymarginal white men (e.g., Katz 2016). Race, class, and gender are often considered the ''holy trinity'' of intersectional analysis, and this explanation for the election outcome points to the influence of all three factors on American voters. But the close correspondence between the political behavior of white men and women observed in the exit polling raises questions about the gender-based assumptions in this account of Trump support. Was Trump's appeal concentrated among downwardly-mobile white men, or was an economically-disadvantaged subset of white women also attracted to his candidacy? If so, why did these women prove resistant to sexist aspects of the campaign?
Again, the gender gap literature provides little guidance here. Many economic explanations for women's political behavior focus on economic vulnerability as a factor that contributes to liberal orientations and behaviors. Given gender-based patterns of economic inequality (e.g., the feminization of poverty), women are more dependent on the welfare state and express greater pessimism about the economy, which translates to stronger preferences for social welfare programs and ultimately the Democratic Party (Box-Steffensmeier et al. 2004;Carroll 2006;Welch and Hibbing 1992). Women's employment also tends to be concentrated in the public sector, such that self-interest contributes to working women's support for Democratic candidates and liberal social policy. This work connects downward mobility or financial insecurity to Democratic Party support rather than Republican Party support, contrary to the narrative about Trump voters in 2016.
However, work by Iverson and Rosenbluth (2010) qualifies the relationship between economic vulnerability and Democratic identification in important ways. The authors note that public sector employment contributes to a liberal-Democratic gender gap because women with public sector jobs support Democratic candidates and policies as a way to protect their economic opportunity and autonomy, given that demand for female labor is lower in the private sector. Working women also gain economic power when government subsidizes family work-such as child care and elder care-and these policies are typically favored by the political left. By contrast, they find that women who do not work or who work sparingly favor Republican candidates and fiscally conservative policies that maximize the spending power of the male breadwinner in their homes. This work suggests economic factors may be an important source of heterogeneity among women and points to the need for further research into the way economic factors intersect with gender and race to shape voter behavior.

A Closer Look at Male and Female Voters in 2016
To evaluate the role of political divisions based on whiteness, educational attainment, and class, we compared demographic characteristics of all male and female Trump and Clinton voters, and then evaluated educational and class divisions among white voters only, using data from the 2016 ANES. Looking at all survey respondents, it is clear that Trump supporters are overwhelmingly white. 87% of men and 90% of women who voted for Trump were white, compared to 60% of men and 57% of women who voted for Clinton. Education was a major division among women voters, with 45% of female Clinton voters having a college degree compared to 29% of female Trump voters. Looking only at white women survey respondents, this difference is even more pronounced. 57% of white women Clinton voters had a college degree compared to 29% of white women Trump voters-a 28-point difference. Whereas white male Clinton voters were also more likely to have a college degree, the difference between Clinton and Trump support among white men with college degrees is smaller, about 13%. This suggests higher education is liberalizing for both white men and women, but more so for women.
Clear gender differences were also evident in terms of social class. Here class is measured in three increments based on household income. We created the measure following guidelines from the Pew Research Center (Fry and Kochhar 2016). Lower income corresponds to an annual household income of less than 40 k. Middle income is in the 40-125k range, and upper income corresponds to an annual household income of greater than 125k. Among Trump voters, women were much more likely to be in the lower income category compared to men, a difference of 13 points in the full sample and 14 points for white respondents only. By contrast, the proportion of male, upper-income Trump supporters is greater than the proportion of female, upper-income Trump supporters by about 9 percentage points in the full sample and among white voters only. These findings challenge a dominant narrative surrounding the election-rather than attracting downwardly-mobile white men, Trump's campaign disproportionately attracted and mobilized economically marginal white women.
To determine whether this pattern is unique to 2016, we compared these results with the profiles of Obama and Romney supporters in 2012. See Fig. 2 for the gender differences in support for Republican candidates among low and highincome white voters. White women in the lower income category were more prevalent among Obama voters (35%) than Romney voters (29%) in 2012meaning the voting pattern for low-income white women was reversed in 2016. 1 The numbers for lower income white men in 2012 also challenge this narrative. In 2016, about 18% of Trump supporters were white men in the low-income group. In 2012, 27% of Romney supporters fell into this category. Thus, the seismic electoral shift among low-income voters is better attributed to white women than to white men. The proportion of high-income voters shifted among men as well-about 15% of male Romney supporters were in this category compared to about 25% of Trump supporters. Thus, the primary story emerging from this analysis is one of education and class creating divisions among white women. Given that many of these sociodemographic factors are correlated, we estimated multivariate models of vote choice in order to obtain more precise estimates of how demographic factors influenced support for Trump v. Clinton. We estimated Seemingly Unrelated Logit models for all male and female respondents regardless of race, then re-estimated the same models for white men and women only-see Table 1. The Seemingly Unrelated Logit (SUL) method allowed us to determine which socio-demographic factors predict vote choice for both men and women and to compare the size of coefficients for men to the size of the coefficients to women. This offers insight into whether certain factors have a significantly stronger or weaker influence on vote choice for men compared to women. SUL jointly estimates models for men and women and provides a variance-covariance matrix for the jointly estimated coefficients, which is used to compute standard errors for the post-hoc tests that compare the size of coefficients across the two models. Wald post-hoc comparison tests were used to evaluate whether the effect of each variable on vote choice significantly differed for men and women. Italicized values in the table indicate the coefficients for men and women differ at the p \ .10 significance level or below. 2 The 2012 data are included for comparison purposes; all measurement information is provided in Table A1 of the Online Appendix.
The results confirm the class division among women described above. Whereas class is not a significant predictor of vote choice for men, women in the lower income category were significantly more likely to vote for Trump than women in the middle or upper income categories. This result was observed in the full sample models (p \ .05), and the gender difference was substantively larger in the models for white respondents only (p \ .05). This finding reiterates that class was divisive for women, but not for men, in 2016. By comparison, we observed no gender differences in the effect of educational attainment on vote choice. Having a college degree significantly reduced the likelihood of voting for Trump for both men and women. However, when we compare the effect of education in 2012 and 2016, we see that it had a substantively larger effect on all voters in 2016. This suggests that there was something about the campaign that divided Americans based on their education level, regardless of their sex.
A final gender difference that stands out from the analysis concerns Southern residence. Women residing in the South were significantly more likely to vote for Trump than women residing elsewhere. By contrast, Southern residence did not have an effect on men's vote choice that differed from zero (though the comparison across models falls just short of statistical significance). By comparison with 2012, this finding seems consistent with a broader pattern for women voters, rather than something idiosyncratic to the 2016 election (see Ondercin 2013).

Why Wasn't Trump Too Sexist for White Women?
This analysis demonstrates class and educational attainment explain much of white women's support for Republican presidential candidates and for Trump in particular. Counter to the narrative that dominated the pre-election period, Donald Trump was not ''too sexist'' for white women. Why were a majority of white women willing to tolerate Trump's sexism? We argue that many white women were more willing to support Trump, despite his overt displays of sexism, because they adopt sexist beliefs that reinforce the status quo. At first glance this may seem counterintuitive, because sexism strengthens men's privileged status in society over women. Nonetheless, in upholding the status quo, many white women emphasize their advantage over poor and minority women, thus reinforcing their privileged status in the social hierarchy relative to other women.

White Women and Sexist Beliefs
To understand women's incentives to reinforce the status quo, we draw on research on intersectionality and system justification. Intersectionality is an analytic framework for understanding group-based inequality focused on evaluating multiple intersecting group memberships rather than any single social category in isolation (e.g., Crenshaw 1989;Collins 2002). Although most scholarship focuses on how the intersection of multiple identities disadvantage people of color (Hancock 2016;Collins and Bilge 2016), the intersection of different identities can simultaneously reflect advantage and disadvantage (Choo and Ferree 2010). For instance, white middle-class women may experience advantage based on their race and class status, but disadvantage based on their gender. Junn (2017) refers to this positionality of white women as ''second in sex to men, but first in race to minorities (346).'' The relative advantage and disadvantage white women experience is contingent across contexts, making their social and political position permeable (e.g., Collins 1993; Levine-Rasky 2011). Although white women's political interests are rarely explicitly articulated as a function of their whiteness (Frankenberg 1993)-whiteness confers advantage on white women. At the same time, white women belong to a disadvantaged group, based on their gender. Because white women benefit more from the advantages associated with their racial and economic status, as opposed to the disadvantage associated with womanhood, they may be less likely to acknowledge shared interests with minority women or to develop a broader gender consciousness (Levine-Rasky 2011). Instead, in an effort to reinforce their privileged status in society, many white women engage in social positioning practices-such as adopting attitudes and behaviors that reinforce privilege through exclusion and social distancing (Wellman 1993)-that separate them from racial and economic inequality (Cassese and Barnes 2018;Hurtado and Stewart 2004).
One way that some white women justify their status in society is by adopting sexist beliefs that normalize and justify traditional gender relations and genderbased inequality (Jost et al. 2008). Modern sexism, for example, is an unwillingness to attribute gender inequality to discrimination or other structural obstacles facing women (Swim et al. 1995). Instead gender differences are attributed to women's preferences and choices, which differ from those of men by virtue of their traditional roles as mothers and caregivers. Sexism also manifests in more explicit ways. For example, hostile sexism involves ''antipathy toward women who are viewed as usurping men's power'' or making illegitimate claims on government to advance their own status, whereas benevolent sexism involves ''a subjectively favorable, chivalrous ideology that offers protection and affection to women who embrace conventional roles'' but not to those who embrace unconventional roles Fiske 1996, 2001, p. 109). This work suggests that one way white women can maintain their privilege is by protecting and prioritizing their relationships with white men. Women of color, by contrast, typically have little incentive to protect white men's privileged status or to endorse traditional gendered power arrangements (Cole 2009).
Thus, white women's interdependence with white men and their desire to maintain their privileged status relative to more socially distant racial and ethnic groups, may lead to a greater endorsement of hostile sexist beliefs and a reluctance to attribute gender-based inequality to discrimination. Importantly, uneducated white women, those in lower economic classes, and those economically dependent on men (e.g., homemakers) have the most to gain by reinforcing the status quo, as their primary ''advantage'' in society is tied to their race. As a result, we posit that more economically disadvantaged white women will be more likely to endorse hostile sexist beliefs and more reluctant to attribute gender-based inequality to discrimination (Hypothesis 1).

White Women and Support for Donald Trump
To the extent that white women endorse sexist beliefs, we should expect that theylike other voters who endorse sexist beliefs-will not be deterred by Trump's sexism. As a matter of fact, research on the 2016 election shows voters exhibiting high levels of hostile sexism were willing to support Trump, even after numerous allegations of sexual assault (Bock et al. 2017;Schaffner et al. 2018). Some contend that the symbolic implications of a Clinton presidency for upending status quo gender relations was sufficient to attract the ire of hostile sexists, who are less likely to support women in positions of power and authority (Christopher and Mull 2006;Hebl et al. 2007;Masser and Abrams 2004) and opted to throw in their lot for Trump (Bock et al 2017). Others argue that Trump's rhetoric contained themes central to hostile sexism, which activated latent hostile sexism among voters and attracted them to his candidacy (Cassese and Holman forthcoming). Hostile sexists are more likely to endorse a sexual double standard , believe rape myths (Begany and Milburn 2002), and show a proclivity to engage in acquaintance rape and sexual aggression (Masser et al. 2006). All of these factors might explain why the Access Hollywood tapes did not deter all women voters from Trump's candidacy.
Based on the previous discussion, we anticipate that as hostile sexism increases, the likelihood of voting for Trump will increase (Hypothesis 2). As perceptions of discrimination against women increase, the likelihood of voting for Trump will decline (Hypothesis 3). Drawing on the whiteness and system justification literatures, we expect these beliefs about gender will influence vote choice in a similar fashion for both white men and women (Hypothesis 4). To evaluate our expectations, we compare analysis of the 2012 and 2016 ANES surveys, with the expectation that hostile sexism and perceptions of discrimination had a larger impact on voters in 2016 due to the salience of sexism in the campaign (Hypothesis 5).

Data and Key Measures
To evaluate our hypothesis that disadvantaged white women will be most likely to endorse hostile sexist beliefs and more reluctant to attribute gender-based inequality to discrimination, we rely on the hostile sexism scale (Glick and Fiske 1996). The ANES included two questions from this scale: (1) Do women demanding equality seek special favors? and (2) Do women complaining about discrimination cause more problems than they solve? Items were combined to form a mean-centered scale. We also rely on a single survey item asking respondents how much discrimination women face in the United States. Responses were given on a 5-point Likert scale ranging from none to a great deal. This item taps modern sexism (Cassese et al. 2015). Whereas both surveys contain other items gauging gender attitudes (e.g., the 2016 survey contains a long battery of hostile sexism items), the items we use here are the only ones found in both surveys and thus facilitate direct comparisons, with accurate significance tests, between 2012 and 2016. 3 To gauge vote choice, we used a dummy variable that includes the two-party vote and is coded 1 if the respondent voted for the Republican candidate and 0 if they voted for the Democratic candidate. 4 The remaining control variables used in the analysis are described in Table A1 of the Online Appendix. 5 These variables include a traditional battery of demographic control variables, as well as other variables reflecting common explanations for Trump support, namely racial resentment, and authoritarianism, as well as core political values such as egalitarian attitudes and preferences regarding the scope of government. Our results are robust to each of these alternative explanations. 3 The measures for hostile sexism and perceived discrimination against women are correlated at -0.15 (-0.09 for women and -0.20 for men) in 2012 and -0.20 (-0.17 for women and -0.22 for men) in 2016. 4 We also used 101-point feeling thermometers for both Republican presidential candidates to determine whether candidate evaluations were more polarized for some groups of men and women. Tables A4 and A6 of the Online Appendix demonstrates that the results for vote choice reported in Table 3 are consistent with feelings towards Romney and Trump. 5 Replication materials are available at: https://doi.org/10.7910/DVN/QHMXAQ. Table 2 displays our analysis of the determinants of white women's attitudes about hostile sexism and discrimination against women. We focus exclusively on white women here in order to better understand how factors like educational attainment, class, partisanship ideology, and other core political values like egalitarianism and scope of government reflect important divisions among this group of women. 6 First, Models 1 and 2 show that white women with a college education are significantly less likely to display sexist attitudes in both 2012 and 2016 compared to white women without a college degree. With respect to class, the analysis indicates that there are not significant class differences in white women's tendencies to endorse hostile sexist attitudes. Neither lower-income women nor upper-income women are more likely than middle class women to display sexist attitudes. 7 In 2012, there is evidence that employed women are slightly less likely than unemployed women to endorse sexist attitudes. Being a homemaker, as opposed to a woman who works or is unemployed but does not identify as a homemaker, is not significantly associated with hostile sexism in 2012 or in 2016.

Explaining Sexism Among White Women
Thus, our findings for Hypothesis 1 are mixed. In support of Hypothesis 1, disadvantaged white women in society-those without a college degree-are more likely to display sexist attitudes. That said, with the one exception of employment in 2012, we do not find that white women are more likely to display sexist attitudes when they are disadvantaged in terms of income or employment status. It is worth noting that in 2016, party identification is a significant predictor of sexist attitudes among white women, with Republicans and independents being far more likely to express sexist attitudes than Democrats. This finding stands in sharp contrast to 2012, where party identification is not significantly associated with sexism.
In Models 3 and 4, we consider the factors that influence perceptions of gender discrimination among women, working from the perspective that denial of discrimination is a manifestation of modern sexism. To begin with, we observe that college education does not predict white women's likelihood of denying gender discrimination in 2012 but is positive and significant in 2016. The difference in effect sizes for 2012 and 2016 is significant at the p \ .10 level. That is, non-college educated women are more likely to say that women do not face discrimination in society and women with a college degree are more likely to acknowledge discrimination against women. This result largely reinforces the finding that educational attainment predicts sexism. That said, this finding only holds for 2016, suggesting that the context of the 2016 election may have increased the salience of sexism among women with a college education. Alternatively, the 2016 election may have stimulated women without a college degree to deny discrimination against women in an effort to protect the status quo. Regardless, this result suggests different processes are at work in the two elections.
A similar change in attitudes is observed between 2012 and 2016 among lower income women. In 2012, lower income white women are more likely than middle class women to perceive that women in the U.S. face gender-based discrimination. The same relationship does not hold in 2016. Here too, it is possible that the context of the 2016 lower-income white women became more likely to engage in system justification-rejecting the notion that American women face discrimination as a way of reinforcing the status quo and protecting their privileged racial status. Finally, we do not observe that either employment status of party identification predicts perceptions of discrimination against women in 2012 or 2016. Collectively, the results presented in Models 3 and 4 provide some additional evidence in support of Hypothesis 1-with educated white women being more likely to perceive that  American women face discrimination in 2016. We do not, however, find evidence that white women who are economically disadvantaged are more likely to deny that American women face discrimination when controlling for education.

Predicting Support for Republican Candidates among White Women Voters
Next, we consider how well hostile sexism and perceptions of discrimination against women predicts white women's vote choice and support for president Trump in the 2016 election. We include the same socio-demographic variables from the vote choice models provided in Table 1 and add our two measures of sexism, along with a series of other variables intended to rule out competing explanations for Trump support, including: racial resentment, authoritarianism, and political values including egalitarianism and scope of government. Again, we include the 2012 models as a benchmark to understand how the relationship between sexist attitudes and support for Republican presidential candidates has changed over time.
The results are presented in Table 3. Consistent with the expectations laid out in Hypotheses 2 and 3, sexism and the denial of discrimination against women are strong predictors of white women's vote choice in 2016. However, in line with Hypothesis 5, they are not predictive of voting for Romney in 2012. The difference in coefficients for hostile sexism in 2012 and 2016 are significant at the p \ .05 level, while the difference between the coefficients for perceptions of discrimination fall just short of the p \ .10 level.
This result is interesting in light of previous research demonstrating that although conservatives tend to be more sexist than liberals, sexism cuts across both partisanship and ideology, with both liberals, conservatives, Democrats, and Republicans harboring sexist attitudes (Cassese et al. 2015). Given this, it is not surprising that when we control for education, ideology, and partisanship in the 2012 model, all of which are strong predictors of sexism in Table 2, sexist attitudes do not predict support for Romney. In 2016, by contrast, even when controlling for education, ideology, partisanship, core values, racial resentment, and authoritarianism, we find that sexism still shapes support for Trump. 8 The strong significant relationship between both hostile sexism and denial of discrimination against women in a fully specified model suggests that these beliefs exert a robust and independent effect on candidate support. Taken together, the results presented in Model 2 of Table 3 provide strong support for our expectations that white women who display hostile sexist attitudes and who perceive low levels of gender discrimination in society are more likely to support Trump (Hypothesis 2 and 3). Further, the comparison of Models 1 and 2 provide evidence that hostile sexism (and to a lesser extent perceptions of discrimination) had a larger impact on voters in 2016 than 2012 due to the salience of sexism in the campaign (Hypothesis 5).
Not only are white women Trump supporters more likely to endorse sexist beliefs than Romney supporters, but they are also more likely to be lower class than middle class (p \ .001) and also more likely to be homemakers (p \ .10). That is, even after controlling for sexist attitudes, these important demographic characteristics still predict support for Trump. By comparison, of the demographic factors explaining support for Trump, only the lack of college education predicts white women's support for Romney in 2012. These findings further support our argument that the campaign narrative about the mobilization of economically disadvantaged white men overlooks the important role played by women with similar characteristics.

Comparing the Effects of Sexism on Vote Choice for White Men and Women
Having established that sexism played an important role in determining vote choice among white women in 2016, we next turn to the question of whether the effect of sexism on vote choice is comparable for white men and women. In Hypothesis 4, we posit that hostile sexism and perceptions of discrimination against women will influence both white men's and white women's vote choice similarly, given that some white women are motivated to endorse sexism as a way to emphasize the privilege associated with their race and de-emphasize the disadvantage associated with their gender. To test this hypothesis, we re-estimated the Table 3 models, including interaction terms between our two measures of sexism and respondent gender (see Table A5 in the Online Appendix), then calculated the difference in the predicted probability of voting for the Republican presidential candidate between individuals at the minimum and maxim values of our sexism measures. Figure 3 reports these first differences for men and women in the 2012 and 2016 elections for both measures of sexism. Panel A shows the difference in the predicted probability of voting for the Republican presidential candidate for participants who scored the minimum value on the hostile sexism scale (-1.5) compared to the probability of voting Republican for participants who scored the maximum value (2.8) on the hostile sexism scale. The differences in the predicted probability of voting Republican are surrounded by 90% confidence intervals. When the confidence intervals cross zero, it indicates that the difference between the two predicted probabilities (i.e., the probability for those exhibiting high levels of hostile sexism compared to those exhibiting low levels of hostile sexism) is not statistically different from zero. In 2012, hostile sexism is not significantly related to the likelihood of men or women voting for Romney. The confidence intervals for both groups cross zero indicating that the effect is not statistically significant. However in 2016, moving from the minimum value of hostile sexism to the maximum value of hostile sexism increased the probability that white women voted for Trump by about 14%. A similar effect is  Table A5 Models 1 and 2 and include white survey respondents only Polit Behav observed for men (also about 14%), though the relationship for men is less robust (it is significantly different at the 90% confidence level but not at the 95% level). The results are suggestive of our expectations in Hypothesis 4, that sexism operates in a similar fashion for both men and women. The significant relationship between hostile sexism and vote choice in 2016, in contrast to the lack of relationship in 2012, suggests that aspects of the 2016 campaign activated hostile sexism attitudes, particularly among white women, consistent with Hypothesis 5.
Panel B graphs the difference between respondents who perceive low levels of discrimination against women (1 on a 5-point scale) compared to those who perceive high levels of discrimination against women (5 on a 5-point scale), using the same approach described above. Consistent with Panel A, beliefs about gender are not significantly related to vote choice in 2012 for either men or women. That is, men and women who perceive high levels of discrimination against women are no more or less likely to vote for Romney than are men and women who perceive low levels of discrimination against women. Turning to 2016, we observe that women who perceive high levels of discrimination against women are 16% less likely to vote for Trump than are women who perceive low levels of discrimination against women. However, men who perceive low levels of discrimination against women are not more likely to vote for Trump than are men who perceive high levels of discrimination against women. Contrary to the expectations stated in Hypothesis 4, we do not find that the relationship between perceptions of discrimination against women and support for Trump operates in the same fashion for men and womeninstead, it seems to matter primarily for white women voters. The differences we observe across the 2012 and 2016 elections lend further support for Hypothesis 5, suggesting the 2016 campaign activated beliefs about discrimination among white women.

Conclusion
Our analysis of white women voters offers some new insights into their behavior in 2016. Looking at the demographic correlates of vote choice in our first set of analyses, we see that white women without a college degree and those with a household income of less than 40 k were disproportionately likely to vote for Trump. This trend represents a departure from the 2012 election, given that women in the lower income category were more likely to vote for Obama than Romney in 2012. The voting behavior of low-income men also ran counter to expectations. In 2016, about 18% of Trump supporters were white men in the low-income group compared to 27% of Romney supporters in 2012. These results upend the dominant narratives about Trump's electoral success-that his campaign successfully mobilized economically marginal white men. Instead, our results suggest that this popular narrative is mis-gendered, that is, Trump appears to have mobilized economically marginalized women-not men! Thus, our results highlight the need to further interrogate the intersection of whiteness, gender, and class in order to better understand support for GOP presidential candidates.
Our results also address analysts' incorrect expectations about women voters defecting from the GOP in response to Trump's campaign. We explain this discrepancy by illustrating that some white women-particularly those without a college education-endorse hostile sexism and have weaker perceptions of systemic gender discrimination. These beliefs are associated with an increased likelihood of voting for Trump-even when controlling for partisanship and ideology, both of which are also associated with endorsement of hostile sexism and perceived discrimination against women in American society. We offer a theoretical explanation for this finding linked to system justification theory, arguing that white women endorse traditional and even sexist views in order to maintain the privilege associated with whiteness. This argument dovetails with Junn's (2017) analysis of white women voters, offering empirical support for her arguments about white women's positionality. We build on Junn's work by evaluating the specific set of beliefs that undergird support for Trump's candidacy and by comparing the relationship between these beliefs and vote choice in two elections where the salience of sexism was markedly different.
We argue that some white women are motivated to distance themselves from the disadvantage associated with their gender. As a result, their belief systems emphasize and privilege their relationships with white men while underscoring their racial advantage over poor and minority women. Not surprisingly then, white women who harbored sexist attitudes were more likely to be tolerant of Trump's sexism and to cast their vote for him on election day. This finding contributes to a growing body of literature on sexism and political campaigns (e.g., Bracic et al. 2018;Schaffner et al. 2018;Valentino et al. 2018), by further investigating the origins and implications of hostile sexism from an intersectional viewpoint.
Collectively, our results point to the limitations of extant theories of the gender gap, which tend to emphasize explanations for women's liberalism and neglect the factors that encourage conservatism among women. One in four women identify with the Republican Party, and their political behavior has been subject to increasing scrutiny as voters and as they attain new leadership positions in the Republican Party (Barnes and Cassese 2017;Deckman 2016;Ondercin 2013;Norrander and Wilcox 2008;Schreiber 2014). Our results build on this past work by showing that whiteness, educational attainment, and class are important factors for understanding women's attraction to the GOP and for understanding heterogeneity among women voters more generally. Even among white women, class and educational attainment exert important cross pressures on political attitudes and behavior.
Whereas our insights stem from two presidential races, there is evidence to suggest these factors operate similarly for down-ballot races as well. For example, in the 2017 special election for the Alabama Senate seat vacated by Jeff Sessions, 63% of white women voted for Republican candidate Roy Moore, despite numerous accusations of sexual harassment against under-aged women and news coverage of a book written by Moore which argued against women's rights as voters and political officeholders. Here too, education was an important factor in shaping vote choice-52% of white women with college degrees voted for Moore compared to 73% of white women without them (Clement and Guskin 2017). This finding, coupled with the effects of higher education noted in our analysis, suggests that a college education is an important source of heterogeneity among white women. While it is not surprising to find evidence that a college education is liberalizing, it is striking that a majority of college-educated white women still voted for Moore. Our study points to the power of sexist beliefs and attitudes toward gender-based discrimination in shaping responses to sex scandals and accusations of male candidates' mistreatment of women. This example also highlights the power of motivated reasoning, consistent with past research demonstrating that it can cause voters to increase their support for a liked candidate upon learning something negative about them-essentially doubling down on their support for the attacked candidate (Redlawsk 2002).
Our study raises a number of important questions about sexism in American politics. For instance, what is the relationship between party identification and sexism? Is there any endogeneity between these concepts that may complicate our efforts to disentangle their unique effects on vote choice? Answering this question is beyond the scope of our paper, though it is an important one in light of scholarship that suggests gender plays an important role in party sorting mechanisms (Barnes and Cassese 2017;Gillion et al. forthcoming). Further research should evaluate the role of sexism in party sorting and party polarization using a longitudinal data source. Indeed, if sexism underlies party attachments among men and women, analysts should cease speculation that any amount of sexism in campaigns or among leaders will push Republican women toward the Democratic Party. A second set of questions warranting further study is: Which forms of sexism demand attention in campaign contexts and what is the relationship between them? Researchers have identified many manifestations of sexism and we explore only two here, but other forms of sexism (e.g., gender stereotypes and benevolent sexism) also have political consequences (e.g., Barnes et al. 2018;Bracic et al. 2018;Cassese and Holman forthcoming), and more work is needed to understand the full constellation of gender-based attitudes and their political implications. Although there is still much work to be done on this topic, the results of the 2016 Presidential race and the 2017 Alabama Senate race highlight the strength of women's attachment to the GOP, and our findings here show that traditional beliefs about gender and gender-based inequality undergird party loyalty among many white women. This suggests that popular characterizations of candidates as sexist or campaign narratives that invoke a ''Republican War on Women'' likely lack sufficient force to push white women out of the GOP camp, and will not prove to be a successful strategy for Democrats hoping to demobilize Republican women or lure them across the aisle.