Introduction
As parties have sorted and the U.S. electorate has diversified, candidates running for office face different electoral incentives when signaling support from and for different minoritized groups. Voters can use a candidate’s associations—such as the endorsements they receive—as a heuristic for whom a candidate will work on behalf of if elected. Will voters, in turn, map onto candidates their attitudes toward the marginalized groups that candidates are associated with? Will affiliations with minoritized communities increase support for candidates among those who want to fight a given form of inequality while decreasing support among those invested in maintaining the social status quo? In 2020, for example, Joe Biden leaned heavily into his connections to former president Barack Obama and a crucial endorsement by Congressman Jim Clyburn as he sought to shore up his support among Black voters in a Democratic presidential primary of record candidate size (Caputo Reference Caputo2020). In contrast, Hillary Clinton distanced herself from President Obama in 2016 as Republicans used her connections with him to shore up support among conservative White voters (Mehta Reference Mehta2015; Scher Reference Scher2016).
Though political candidates clearly want to take positions that allow them to win, how precisely voters respond to signals of support for different marginalized groups, especially when they themselves do not belong to marginalized groups, is undertheorized. Research has shown that President Obama is a polarizing and “racializing” figure who can activate voters’ racial resentment attitudes when things are tied to him (Tesler Reference Tesler2016). Yet this phenomenon may play out otherwise for candidates with different ascribed identities. For instance, recent work suggests that voters’ evaluations of candidates are more complex than candidates’ ascribed identity alone. A growing body of research investigates how endorsements may signal coalitional support between groups and candidates (McDermott Reference McDermott2006; Rothschild Reference Rothschild2022), and strong ties or a demonstrated commitment to the Black community can influence how Black voters and racially resentful White voters evaluate candidates even if those candidates are not Black (Stephens-Dougan Reference Stephens-Dougan2016; Wamble Reference Wamble2018). Where racialization is gendered (Bell and Borelli Reference Bell and Borelli2023), might this phenomenon also generalize to other identities and dimensions of marginalization, such that an endorsement from a prominent women’s rights or LGBT organization negatively impacts evaluations of candidates among (hetero)sexist voters?
To answer these questions, we compare the effects of associations with different marginalized communities—which we operationalize using endorsements from advocacy organizations—on candidates with different race-gender-sexuality profiles. For instance, we test whether a straight White man endorsed by a group advocating for gender equality is evaluated differently by voters compared with a similarly endorsed straight Black man. Where candidates’ ascribed identities can be used as a heuristic for whether they will uphold or contest systems of power, associations with minoritized communities via endorsements can signal a candidate’s group commitments. Using an original conjoint experiment fielded on a nationally representative online sample recruited by the Qualtrics Panels Team, we asked respondents to evaluate six pairs of candidates in a hypothetical primary election that randomized candidates’ race, gender, and sexuality to test the effects of ascribed identity on voter evaluations. We also randomized whether the candidates were endorsed by a neutral group or an advocacy organization for people marginalized in terms of race, gender, or sexuality to signal candidate support for marginalized groups. The results show that these group endorsements are significant and often stronger predictors of voter behavior than candidates’ identities alone and that, in many cases, the effects of endorsements are conditional on the candidate’s demographics. Even for relatively privileged candidates, a marginalized community’s endorsement can signal the candidate’s commitment to contesting their marginalization. Consequently, those threatened by challenges to the status quo are less supportive of endorsed candidates, while those who support contesting the social hierarchy are more supportive.
We contribute to scholarship on voting behavior and identity in the following ways. First, we help bridge the literature on voting behavior and elections involving marginalized candidates (see, e.g., Carey and Lizotte Reference Carey and Lizotte2019; Haider-Markel et al. Reference Haider-Markel, Miller, Flores, Lewis, Tadlock and Taylor2017; Magni and Reynolds Reference Magni and Reynolds2021; Pearson and McGhee Reference Pearson and McGhee2013) and the literature on intersectionality (see, e.g., Cassese and Barnes Reference Cassese and Barnes2019; Junn Reference Junn2017). Previous work has often considered identities in isolation, while intersectional work questions this separation and details the shortcomings of such an approach. We apply an intersectional framework to consider how co-constitutive identities affect voters’ evaluations of candidates and how marginalized candidates running for office may be evaluated differently using endorsements to signal a candidate’s relationship to interlocking systems of power. In doing so, we show that while groups such as LGBT folks still face widespread scrutiny by voters, other historically marginalized groups are not penalized (and, in some instances, are rewarded) by voters when no other heuristics about their political positions and group loyalties are available. Second, we demonstrate that processes such as racialization are likely exclusive to neither President Obama nor racism. Our work thus suggests a broader “(de)marginalization by proxy” in which voters evaluate candidates with strong ties to and a willingness to form coalitions with historically marginalized groups similar to members of those groups, and this also applies to race, gender, and sexuality groups. Whether a candidate will uphold or contest existing interlocking hierarchies thus has a stronger effect on voters’ evaluations than identity alone.
Identity and Prejudice: How Voters View Candidates
To answer our questions of interest, we must first account for how candidates’ demographics currently influence voters’ perceptions of them. We must then determine how factors other than candidates’ identities influence voters’ evaluations of them, especially for candidates with close ties to marginalized groups. If a candidate comes from a comparatively privileged background but has close ties to marginalized communities, will voters with negative affect toward those groups evaluate them similarly to marginalized group members? In turn, we must address whether voters’ social attitudes and a candidate’s own identities interact to affect candidates differently. For example, will prejudiced voters evaluate White men and women endorsed by Black advocacy groups differently? We draw from two sets of research to inform our hypotheses.
Marginalized Candidates Facing Voting Skepticism
A rich set of work on elections and voting behavior suggests that voters often penalize political candidates from historically marginalized groups. A fairly robust literature suggests that LGBT and Black candidates running for office are penalized relative to non-LGBT and White ones (Carey and Lizotte Reference Carey and Lizotte2019; Haider-Markel et al. Reference Haider-Markel, Miller, Flores, Lewis, Tadlock and Taylor2017; Sigelman et al. Reference Sigelman, Sigelman, Walkosz and Nitz1995; Stephens-Dougan Reference Stephens-Dougan2020; Terkildsen Reference Terkildsen1993), although candidates can sometimes overcome these challenges by strategically choosing where and when to run (Haider-Markel Reference Haider-Markel2010). Women need more qualifications than men to win (Bauer Reference Bauer2020; Pearson and McGhee Reference Pearson and McGhee2013), and they are expected to conform to certain stereotypes when running (Huddy and Terkildsen Reference Huddy and Terkildsen1993; Teele, Kalla, and Rosenbluth Reference Teele, Kalla and Rosenbluth2018). Nonetheless, recent work has shown that—despite higher standards and expectations—voters may slightly prefer women to men (Schwarz and Coppock Reference Schwarz and Coppock2022). Still, these studies tend to look at identities in independently each other. They also focus on ascribed identities, failing to consider how voters with prejudice toward out-groups evaluate candidates associated with those out-groups.
Building a Theory beyond Ascriptive Identity
Voters can use identity as a heuristic when deciding how to vote. Social categories such as racial groups are understood as positional and relative to each other (Blumer Reference Blumer1958). Therefore, voters can use candidate identity as a signal for whose interests a candidate would be most likely to advance and whether they would reaffirm or challenge existing social hierarchies if elected (Frasure-Yokley Reference Frasure-Yokley2018; Strolovitch, Wong, and Proctor Reference Strolovitch, Wong and Proctor2017). However, this relationship between identity and vote choice is complex, and other factors may outweigh the use of identity as a heuristic in voting. Social categories are not fixed, and a candidate’s identity alone does not guarantee their beliefs or behaviors. Further, sharing one marginalized identity by no means ensures solidarity on other dimensions of marginalization.
Voters can use other information to update their beliefs about a candidate’s views and commitments to upholding or challenging status quo power hierarchies. While variables like candidate identity and partisanship can serve as a baseline, other information may become more important in shaping voter evaluations, particularly in low-information environments (Kirkland and Coppock Reference Kirkland and Coppock2018). For instance, White Democrats shown with a larger proportion of Black people serves as a heuristic for racist voters and decreases their support for such candidates (Stephens-Dougan Reference Stephens-Dougan2016). On the other hand, demonstrated sacrifices or endorsements by organizations like the NAACP can increase Black voters’ support for White candidates (Wamble Reference Wamble2018). To limit fallout among racist voters, Black candidates can attempt to use racial distancing to signal to White voters that they will not threaten the status quo racial order (Stephens-Dougan Reference Stephens-Dougan2020), with studies showing that the mere presence of information on a candidate’s identity does not necessarily activate negative prejudices among voters (Gonzalez Juenke and Sampaio Reference Gonzalez Juenke and Sampaio2010; McCormick and Jones Reference McCormick, Jones and Persons1993; McIlwain Reference McIlwain2011). In terms of upholding White supremacy, members of racist organizations do not view companionship with Black people as contradictory as long as they are not perceived to be challenging racial hierarchies (Blee Reference Blee1996). Other work shows that strong ties to figures such as President Obama can “racialize” policies and decrease racially resentful voters’ support for them (Tesler Reference Tesler2016).
Candidates who successfully demonstrate a commitment to working with marginalized communities may do so in a variety of ways. For example, interest group endorsements have been shown to serve as strong signals of potential coalitional relationships among interest groups, voters, and elected officials. A growing body of research has analyzed the relationship between voters and groups advocating on their behalf (see, e.g., Lau and Redlawsk Reference Lau and Redlawsk2001; McDermott Reference McDermott2006; Rapoport, Stone, and Abramowitz Reference Rapoport, Stone and Abramowitz1991). This research often emphasizes the value of such endorsements as a heuristic for low-information voters (Arceneaux and Kolodny Reference Arceneaux and Kolodny2009), and the coalitional component that interest groups bring to the table when choosing to endorse candidates cannot be understated (Grossmann and Dominguez Reference Grossmann and Casey2009). Candidate endorsements may provide strong signals of an organization’s confidence in a candidate’s agenda and relationship to various groups of voters (McDermott Reference McDermott2006; Rothschild Reference Rothschild2022), and organizations have incentives to deliver candidates who will be accessible in order to remain credible among their supporters (Heaney Reference Heaney2004). Still, while voters may infer whom candidates are likely to advocate on behalf of when organizations endorse them, how strongly they react is undertheorized in existing studies. Consequently, we argue that the mechanisms underlying phenomena such as “racialization” and how voters react to signals of support for members of specific social groups may be a subset of a more general phenomenon that applies to identity broadly. While presumptions of a candidate’s commitment to upholding or undoing racial inequality may influence voting behavior when race is made salient, the general phenomenon may interact with experiences of candidates at the intersections of identities. In turn, the number of and dimensions on which a candidate is marginalized may affect how associations with different marginalized communities affect voters’ evaluations of candidates.
Intersectional theories suggest that such a differential process may be both possible and likely. Political science research has historically approached questions of voting behavior using nonintersectional frameworks, looking at identities in isolation while focusing on the effect of being Black or a woman or LGBT. Although informative about how voters evaluate particular subgroups of candidates, some caution is needed in overgeneralizing which groups of candidates such findings apply to. In coining the term “intersectionality,” Crenshaw (Reference Crenshaw1989) notes the challenges that Black women—who were unable to bring lawsuits representative of all women and all Black folks based on their unique co-constitutive experiences as Black women—faced in relation to civil rights litigation. Similarly, Black women face unique stereotypes, different from those of Black men or White women (Crenshaw Reference Crenshaw1991). Analyzing identities in isolation can mask heterogeneity within social categories and nonadditive effects of forms of marginalization, at the risk of homogenizing groups’ experiences while missing how identities and systems of power intersect (Hancock Reference Hancock2007). It is therefore necessary to incorporate individuals’ complex relationships with and support for inseparable and intersecting systems of power (Collins Reference Collins2002; Combahee River Collective 1977). Ascribed identity may influence whether an individual struggles against oppression or seeks to uphold hierarchies, but an analysis of identity absent a focus on power is incomplete.
Our study jointly manipulates candidates’ demographics and ties to marginalized communities. This allows us to observe the precise and unique ways in which candidates’ relationships to power systems influence voter evaluations. We thus add to the growing number of studies using an intersectional approach in analyzing group behavior and attitudes (e.g., Bonilla and Tillery Reference Bonilla and Tillery2020; Cassese, Barnes, and Branton Reference Cassese, Barnes and Branton2015; Hutchings, Walton, and Benjamin Reference Hutchings, Walton and Benjamin2010) to test whether processes such as racialization play out for candidates differently conditional on their demographics.
Group Ties as a Proxy for Identity
We argue that a candidate’s perceived commitment to different social groups is equally important, if not more so, as identity alone. A candidate’s race may be used as a heuristic for whether they will support racial hierarchy or contest it, but other information can be used by voters to supplement identity alone. Voters may use such information to determine whether they think a candidate would push for policies that would challenge or maintain the status quo if elected. In turn, they will be more or less resistant to candidates depending on the degree to which they see candidates as rocking the boat in or out of their favor politically and socially.
We view endorsements as one source of this type of information. Endorsements by advocacy organizations can serve as signals to voters of potential coalitional relationships between candidates and (out-) groups that the organizations advocate on behalf of (Barreto Reference Barreto2007; Benjamin Reference Benjamin2017). In other words, while endorsements may signal to members of a group whom that group should vote for, they can also send signals to other groups as well—whether for the purposes of signaling a potential coalitional partner, future policy legislation, or otherwise. In turn, this may affect voter evaluations of the candidate depending on whether voters support the existing social order or feel threatened by changes to it (Craig and Richeson Reference Craig and Richeson2014). To that end, we argue that endorsements can have positive externalities for candidates by drawing some groups of voters into their camp while also having negative ones by pushing some other subsets of voters away from them given the threat voters perceive them as posing if they were to win election.
We propose a broader phenomenon than racialization, in which individuals’ social positional considerations are brought to bear on evaluations of policy and political candidates that we define as associational affect. For example, while by no means restricted to any particular identity, voters’ place in race-, gender-, and sexuality-based hierarchies may influence their perceptions of candidates who challenge their place in those hierarchies. This can lead to a sort of (de)marginalization by proxy, whereby an individual’s support for or opposition to a social hierarchy can negatively or positively affect their evaluations of people or policies tied to the social hierarchy. When a voter has negative affect toward some social category or feels threatened by challenges to the status quo, they will be less supportive of candidates associated with the social category as they infer that the candidates will support policies benefiting members of the category. Likewise, a voter with positive affect toward a group will become more favorable. This is especially likely to be the case during periods when social cleavages are particularly visible and salient and groups are seen as challenging the status quo, such as during a news period following a major protest at which a group is actively seeking change.
These associations become relevant when they are seen as signals of which groups a candidate supports and whether the candidate will work to preserve social hierarchies. An endorsement from an interest group advocating equality for a minoritized community can provide a strong signal of a candidate’s willingness to work with members of that group (Heaney Reference Heaney2004; Rothschild Reference Rothschild2022), and voters will evaluate these signals accordingly (whether in a positive direction among allies of the community or a negative direction among those hostile toward it). Particularly when voters have little other information about politicians and their positions, voters will likely rely heavily on heuristics such as endorsements in addition to a candidate’s identity in forming their evaluations.
We expect that other kinds of associations (compared to, for example, an endorsement from someone like Obama) can also serve as strong cues of a politician’s underlying attitudes and policy views than their identity alone—especially when voters can easily identify how that association relates to existing power hierarchies. Voters will likely view an endorsement from a group openly advocating for LGBT equality—a clear challenge to the status quo—differently than an endorsement from a group composed of LGBT individuals but with a more ambiguous or unknown mission. Moreover, endorsements for organizations are costly since they can usually make only one endorsement per race. Consequently, endorsements can provide powerful signals of which candidates organizations find likely to most benefit members of their group, even when they endorse someone other than a member of their group. Moreover, when the endorsing group advocates on behalf of a group to which a candidate belongs, the endorsement serves to reinforce assumptions about the candidate’s positioning. Consequently, a candidate with a marginalized identity and endorsement from an advocacy group working to address that marginalization may be more heavily penalized (rewarded) by voters threatened by (supportive of) challenges to the status quo relative to similarly marginalized candidates without such endorsements.
This first hypothesis is based on past work on prejudice and voting behavior with identity considered in isolation, which we will then build upon:
H 1 (Candidate Evaluations Hypothesis): Individuals who do (not) support the status quo hierarchy will evaluate Black, women, and LGBT candidates more (less) negatively than straight White men, and they will be less (more) likely to vote for them.
While we do not view prejudice as exclusive to these groups or categories, we suggest these groups as a starting point for exploration. In particular, given the historical dominance of the Black-White racial divide in the United States and its relationship to gender (Crenshaw Reference Crenshaw1989), we view these as critical to our study. Similarly, given the salience of sexuality in U.S. politics in recent years, we also view this as important to our investigation. Given the work on intersectionality, we suspect that candidates who are marginalized on multiple social dimensions may be evaluated more negatively compared to White men and candidates marginalized on just one dimension. We apply an intersectional framework to findings from previous studies showing that candidates from marginalized backgrounds are frequently penalized by voters:
H 2 (Intersectional Evaluations Hypothesis): Individuals who do (not) support the status quo hierarchy will evaluate candidates marginalized on multiple dimensions more (less) negatively than they would candidates marginalized on fewer or no dimensions and be less (more) likely to vote for them.
Our theory, however, takes this work a step further. As we argue, a candidate’s demographics alone are insufficient to fully account for how voters evaluate a candidate. Moreover, a candidate’s perceived group loyalties and what they would do to advance a group will be stronger predictors of how voters evaluate candidates; however, perceived group loyalties can combine with assumptions based on candidates’ ascribed identities. We thus anticipate that candidates from marginalized groups may stand to be penalized a greater degree than other candidates given the same ties a group. This, in turn, leads to the following two hypotheses:
H 3 (Associational Affect): Individuals who do (not) support status quo social hierarchies will view candidates with strong ties to marginalized groups—thereby presuming a commitment to fighting inequality—less (more) favorably than candidates without such associations.
H 4 (Associational Affect and Intersectional Marginalization): Individuals who do (not) support status quo social hierarchies will view single-dimensionally or intersectionally marginalized candidates with strong ties to marginalized groups less (more) favorably than a candidate with fewer marginalized identities with similar associations.
Although these ties may come in a variety forms—such as appearing alongside a high-profile member of that group (Bell and Borelli Reference Bell and Borelli2023) or a history of community commitment (Wamble Reference Wamble2018)—we operationalize these ties as interest group endorsements, which we describe in greater detail later.
Data, Methods, and Attitude Measurement
We use a conjoint experiment to test our theory and hypotheses, allowing us to jointly manipulate candidate demographics and endorsements to identify multidimensional effects and their interactions in order to determine whether evaluations of candidates are affected by their ascribed identity and endorsements from minoritized communities. Although some caution is warranted in inferring majority preferences from conjoint average marginal causal effects (AMCEs; see Abramson, Koçak, and Magazinnik Reference Abramson, Kocak and Magazinnik2022), we note similar challenges with individual-level inferences and average treatment effects in other settings (Bansak et al. Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman, Green, Druckman and Green2021). We conducted our preregistered survey in March 2020 using an online sample recruited by the Qualtrics Panels Team that was nationally representative in terms of gender, age, region, ethnicity, education, and income.Footnote 1 We did, however, have slightly more Democrats and LGBT respondents than the nation as a whole. For the full set of descriptive statistics, see Appendix 1.4. Our sample includes 648 complete respondent profiles, 60 of which failed a manipulation check.Footnote 2 The remaining sample of 588 respondents included in the analyses evaluated a total of 7,056 candidate profiles. Respondents evaluated the candidates on a feeling thermometer and selected which one they would prefer to vote for in a primary election.
Respondents were shown candidates from six hypothetical gubernatorial primary elections. Similar to Haider-Markel et al. (Reference Haider-Markel, Miller, Flores, Lewis, Tadlock and Taylor2017), we used primary elections to reduce the use of partisanship as a heuristic to imply that the candidates’ share the respondents’ ideology. The presentation of profiles was designed to approximate a Ballotpedia page and provide information that one would likely see in advertisements, fliers, debates, or through some other means during the campaign. This design was chosen to increase external validity by imitating the context in which an individual would obtain information about candidates and make judgments in both high- and low-information elections. Profiles included a photo (Ma, Correll, and Wittenbrink Reference Ma, Correll and Wittenbrink2015), name, spouse’s name, number of children, age, undergraduate education, two endorsements, and political experience.Footnote 3 We include a sample profile in Figure 1. A candidate’s race (Black or White) and gender (man or woman) were signaled using their name and corresponding photo, while sexuality was signaled by indicating their same- or different-gender spouse’s name. Race, gender, and sexuality have successfully been signaled using these methods in previous studies (Adida, Laitin, and Valfort Reference Adida, Laitin and Valfort2010; Bertrand and Mullainathan Reference Bertrand and Mullainathan2004; Mishel Reference Mishel2016; Sen and Wasow Reference Sen and Wasow2016).Footnote 4
Figure 1. Sample profile. Example of candidate pairs used for the manipulation check.
Each candidate had two endorsements: one that was always neutral and another that was either another neutral organization or one working on behalf of women, racial minorities, or LGBTQ people. The endorsement order was randomized. Candidate demographics and endorsements were assigned independently and with uniform probability except for sexuality, for which the probability that a candidate had a same-gender spouse was reduced to .25 given that same-gender couples are less common. In all, there were a total of eight possible race-gender-sexuality demographic profiles. Combined with the four different group associations, there were a total of 32 possible profiles of theoretical interest. All candidate profile matchups were possible; however, the same candidate and endorsing organization name and photos could not be used for both.
The names of the fictional endorsing organizations were designed to provide clear associations with minoritized groups and signal a candidates’ commitment or lack thereof to contesting social hierarchies. For example, interest groups included the National Women’s Policy Center, the National Racial Equality Policy Center, and the National LGBT Rights Policy Center. Although fictional, the names are modeled after groups such as the NAACP, and all specifically invoked references to equality or policy to give a strong signal of the endorsed candidate’s commitment to supporting a group. Thus, the policy-centric group endorsements provide a strong signal of what sorts of issues candidates would likely prioritize if elected to office and, by extension, how likely those candidates would be to maintain or alter the status quo social hierarchy by working with marginalized communities.
After respondents viewed and evaluated the candidates, we measured their levels of racial resentment, sexism, and heterosexism. We used the standard racial resentment battery (α = .823), five questions from the modern sexism battery (Swim et al. Reference Swim, Aikin, Hall and Hunter1995) (α = .755), and the amnestic heterosexism battery (Walls Reference Walls2008) (α = .775) for our measures. The 14 questions have a Cronbach’s α of .877 when combined. In order to come up with a single measure of social positional threat, we standardized the respective batteries and averaged them. We also performed as robustness checks latent class analysis on the batteries to identify respondents high or low in position threat, as well as to allow different questions or clusters to weigh more heavily in determining position threat. We made these decisions both given the intersectional nature of our hypotheses as well as our theory that these measures are more intricately and complexly linked than the measures individually could capture. We also performed principal component analysis (PCA) on the 14 questions (see Appendix 5). The questions all load heavily onto one component, suggesting that racial resentment, sexism, and heterosexism are connected and capture similar sentiments.
The Intersection of Prejudiced Attitudes
We provide the distribution of our position threat score along with racial resentment, modern sexism, and heterosexism by respondent race, gender, and sexuality in Figure 2. The measures varied in predictable ways. Non-LGBTQ White and non-Black respondents were more racially resentful than Black respondents. Likewise, non-LGBTQ respondents expressed higher levels of heterosexism than LGBTQ respondents. Men tended to register higher in sexism. However, the separation in attitudes is less stark than on racism and heterosexism. Men and White respondents in general tended to score higher across the batteries than women and non-White respondents. Depicted in Table 1, the three attitudes are moderately correlated with each other with coefficients from .478 to .598. The correlations between the attitudes are higher than their correlations with ideology and party identification, with a maximum of .374.
Figure 2. Average standardized position threat score and composite prejudice measure by respondent race-gender-sexuality. Respondents who did not indicate that White or Black was one of their primary racial identities were coded as non-Black men/women of color, or NB(M/W)OC.
Table 1. Attitude correlations
Note: Correlations between prejudice measures and political orientations.
The Limited Effects of Identity in Isolation
We turn to our first two hypotheses analyzing the effects of candidates’ race, gender, and sexuality independently ( H 1 ) and then combined into co-constitutive race-gender-sexuality demographic profiles ( H 2 ). We present the AMCE and the average causal interactive effect (ACIE) of candidate race, gender, and sexuality in Figure 3. The top two plots show the effects of being a Black, woman, or LGB candidate independent of one another relative to a White, man, or non-LGB candidate, respectively. The bottom two plots provide estimates of the effects of the race-gender-sexuality combinations relative to a candidate who is a straight White man. The left plots show the overall AMCE (the crossed line) and then the AMCE conditional on the respondent being in the high-threat class (circle bars) or the low-threat class (triangle bars), with significant differences highlighted in red and blue. The right plots show the coefficients of the demographics and their interaction with the continuous position threat score.
Figure 3. Candidate demographics and respondent position threat. The figure depicts the effects of candidate demographics on the probability of profile selection. Interaction with respondent social position threat class in (a) and (c) and with continuous threat score in (b) and (d) (from ordinary least squares regression). Figures (a) and (b) treat candidate race, gender, and sexuality separately, while figures (c) and (d) treat each demographic profile as categorically different. Estimates do not include controls. The sample includes only the 588 respondents who passed the manipulation check (N = 7,056; 2,436 in the low-threat class, 4,620 in the high-threat class).
We initially find little evidence in this context supporting overall penalties for candidates based on identity alone; however, the relationship between candidates’ intersecting identities and social position threat depicts a more complicated narrative and offers support for the first two hypotheses based on previous work. Minoritized candidates do receive some benefits in terms of extra support, but the penalties they face from those with high position threat are generally not enough to counteract the increase in support they receive from those who feel their social position is less threatened. The AMCEs in Figure 3(a) for Black candidates and women are significant and positive, indicating that, on average, they tend to be selected at higher rates than White or male candidates, while LGB candidates are chosen less often than straight ones. These effects are the result of an increase in support for Black and women candidates among those low in position threat (AMCE = 0.092, 0.079; SE = 0.023, 0.021), which is not matched by an equally strong counteracting negative effect among those higher in position threat. The opposite is true for LGB candidates, who receive a penalty without a corresponding increase in support (AMCE = –0.077; SE = 0.017). Shown in Figure 3(b), the relationship between minoritized status and position threat is negative, indicating that respondents high in prejudice or feeling their social position is threatened support minoritized candidates less; however, the effect is not strong enough to decrease the average support for minoritized candidates among those in the high-threat class. In sum, our findings break somewhat with conventional wisdom that prejudiced voters penalize marginalized candidates. Indeed, while LGBT candidates face a penalty—a finding that largely comports with existing scholarship—other marginalized candidates typically benefit from an endorsement among low-threat voters, providing mixed support for H 1 .
When looking at AMCEs for the intersections in Figure 3c, straight White women and straight Black men and women (AMCE = 0.065, 0.059, 0.071; SE = 0.02, 0.019, 0.021, respectively) are more likely to be chosen relative to straight White men. All profiles besides gay White and Black men are more likely to be selected than straight White men by low-threat respondents. Only lesbian White women are significantly less likely to be selected than candidates who are marginalized on fewer dimensions among high-threat respondents (AMCE = 0.096; SE = 0.034). The overall interaction between position threat and candidates’ intersecting demographics in Figure 3d provides additional insight. Women benefit relative to men; however, this benefit is not consistent across race and sexuality. White women are initially selected more than queer women, and they face a smaller penalty among those high in position threat relative to intersectionally marginalized women candidates. The penalty that straight White women face is significantly smaller than the penalty for Black women and queer White women. This provides clearer support for H 2 and the obstacles that intersectionally marginalized women face in running for office.
The same is true for race: straight Black men receive an initial benefit accompanied by a statistically significant smaller penalty relative to queer White women, queer Black men and women, and straight Black women candidates. While estimates based on threat class are not always significantly different from each other, the overall interactions between continuous position threat and a demographic profile (except for gay White men) is statistically significantly larger relative to the effect on straight White men. We also observe positive and significant effects among those scoring low in social position threat for all groups except gay White men and gay lesbian women. The “topline” AMCEs and estimates that do not account for intersecting identities in the top of Figure 3 mask heterogeneous effects among respondents with varying levels of social position threat, providing stronger support for H 2 that (intersectionally) marginalized candidates face greater penalties relative to comparatively privileged ones.
We also consider the effects of candidate profiles by respondent demographics in Figure 4. While not direct tests of our main hypotheses, these analyses can still provide insight into whether and how subgroups of respondents may be driving our specific results. Non-White respondents favor Black candidates regardless of position threat (AMCE for high/low = 0.0497, 0.0714; SE = 0.0249, 0.0354). White respondents low in position threat also show a preference for Black candidates (AMCE = 0.0821; SE = 0.0259). Men neither favor nor disfavor women, but women prefer them (AMCE = 0.0495; SE = 0.019). Non-LGBT people penalize LGB candidates. Thus, we again find only mixed evidence for our first set of hypotheses. Belonging to a marginalized group does not necessarily correspond to an electoral penalty, even among out-group respondents scoring high in social position threat. However, there are some gains in support among voters low in position threat.
Figure 4. Average marginal causal effects of a marginalized candidate identity by respondents’ race, gender, sexuality, and position through latent class. Plots provide the effect of a candidate belonging to a marginalized gender, race, or sexuality group among respondents who are (not) a member of the same marginalized group and by their position threat class. The bars with circles provide the change in support for a candidate with the given identity for all respondents in the given subgroup. The bars with triangles provide the effect among respondents in the high position threat class and the bars with a cross provide the effect among respondents in the low position threat class.
The results here consequently depict a complex narrative for the effects of identity alone even when considering the effects of prejudice. They do not serve to dismiss the racism, sexism, and heterosexism that marginalized candidates face when running for office, as voter choice and evaluations are but one means of capturing attitudes. Rather, they corroborate findings that candidates can overcome these obstacles in certain scenarios while suggesting that other factors—such as perceived or assumed commitments to marginalized communities and contesting inequality—can be more influential in shaping voter behavior than traditional studies often suggest.
Voter Perceptions of Candidates with Group Ties
To evaluate this possibility and test H 3 , we turn to the coefficient estimates for a race, gender, or sexuality advocacy group endorsement relative to a neutral group, provided in Figure 5.Footnote 5 The former provides the effects of any endorsement by an advocacy group relative to a neutral one, and the latter looks at the effect by type of group. We find some initial support for our associational affect hypothesis even before looking at the ACIEs. The results for the effect of an endorsement overall are clear in the top portion of Figure 5: candidates without a group endorsement are significantly less likely to be selected by respondents scoring low in social position threat (AMCE = –0.151, SE = 0.0711), and they are significantly more likely to be selected by candidates scoring high in social position threat (AMCE = 0.0711, SE = 0.0119). Those who recognize different forms of discrimination and do not feel that their own social position is threatened favor candidates who are endorsed by groups advocating for marginalized communities. The opposite is true for respondents who register high on our measure of social position threat, with marginalized community endorsements decreasing the likelihood that the candidate is chosen.
Figure 5. Effects of different endorsements and latent position threat class. The top portion provides the effects of being endorsed by any non-neutral organization relative to a candidate endorsed by a neutral organization. N = 7,056; 2,436 in the low-threat class, 4,620 in the high-threat class. The bottom portion provides estimates of the change in the probability of selecting a candidate with an endorsement from a racial, gender, or sexual minority rights organization relative to a candidate with a neutral endorsement.
When broken out by type of endorsement in the bottom portion of Figure 5, we find further evidence supporting our associational affect hypothesis. Driven by those with low position threat, candidates with support by organizations advocating for gender equality on average receive more support, and those with a gender or race endorsement receive a larger benefit among those in the low-threat class (AMCE = 0.049, 0.1222, and 0.153; SE = 0.0185, 0.0329, and 0.0358). Further, the gain of a gender or race endorsement relative to a neutral endorsement is larger than the effect of a candidate being a woman relative to a man and being Black relative to White, respectively. Those endorsed by an LGBTQ rights group are less likely to be selected than candidates with a neutral endorsement, which is driven by respondents scoring low and high in social position threat. The effect of an LGBT endorsement is also greater than the effect of a candidate themselves being LBG. Even though the endorsements do not register significant effects in every case, they all are registering effects in some way that match our hypothesized pattern. The endorsements alone thus are successful in priming voters to consider who the candidates would be likely to advocate on behalf of if elected to office, serving as a potential signal of candidates’ later policy support that voters may not otherwise expect based on demographics alone.
The Intersection of Endorsements and Candidate Demographics
We compare the overall effects of different endorsement types on a particular candidate profile to determine whether different demographic profiles are impacted in different ways by the same associations. We start by looking at candidate identities individually in Table 2, which provides the effects of group endorsements broken out by candidates’ gender (a), race (b), and sexuality (c) relative to a candidate with the same ascribed identity with a neutral endorsement. Although the candidate endorsements register effects for many of the profiles, we would like to draw attention to two key differences that provide some additional support for H 3 and H 4 .
Table 2. Endorsement effects by candidate demographics and respondent latent position threat class
p<.05 *; p<.0 **; p<.001 ***
Notes: Estimates by respondent race, gender, and sexuality of the change in the probability of selecting a candidate with the given endorsement relative to a candidate with a neutral endorsement. Includes the unconditional ACME and then the effect among respondents in the low and high latent position threat class for each endorsing group type. The sample includes only the 588 respondents who passed manipulation check (N = 7,056; 2,436 in the low-threat class, 4,620 in the high-threat class).
First, estimates using a candidate with the relatively privileged identity (man, White, non- LGBT) and a gender, race, and LGBT endorsement, respectively, as the baseline relative to women, Black, and LGB candidates reveals that relatively privileged candidates endorsed by marginalized communities are often penalized less than candidates of the marginalized communities with the same endorsements. Provided in Table C.20 in the Appendix, men endorsed by a gender equality advocacy organization and White candidates endorsed by a racial equality organization are affected by the continuous measure of respondent position threat significantly less than a woman with a gender endorsement and Black candidates with a race endorsement, respectively. Further, men with a race endorsement are favored on average and by those low in position threat relative to men without any endorsements (AMCE = 0.0695, 0.1991; SE = 0.0264, 0.0426), while women with a race endorsement are penalized relative to women with a neutral endorsement among high-threat respondents (AMCE = –0.065; SE = 0.0273). In the case of the LGBT endorsement, men are not rewarded, but women are significantly penalized in a way that men are not.
Second, candidates with an LGBT association are rarely rewarded for the endorsement, which is not the case among those with a race or gender endorsement. Every single subgroup here is penalized among those scoring higher in social position threat, and the penalty often outweighs any benefit from those low in social position threat. An LGBT endorsement for a LBG candidate is the only case in which support from a candidate’s own community significantly decreases support relative to a candidate without the support of their community (AMCE = –0.1266 among high-threat respondents). It is also the only endorsement type for which privileged subgroups face net penalties overall. We suspect that this may be a consequence of the comparatively “invisible” nature of sexuality, something not necessarily exclusive to this experiment (Cech and Rothwell Reference Cech and Rothwell2020). While respondents could often infer the race and gender of a candidate based on their name and photo in the experiment, the same was not necessarily true for sexuality—especially absent other information about a candidate’s behavior. Consequently, we suspect that many respondents may have interpreted the LGBT endorsement as the candidate being a closeted member of the LGBT community. Regardless, the findings are nonetheless informative about the nature of LGBT politics and effects of LGBT group endorsements, an often understudied area of scholarship in political science discourse.
Finally, we move to our intersectional analyses to test the specific effects of group endorsements on unique co-constitutive groups ( H 4 ). Although in the preceding paragraphs, we broke out the analyses by race, gender, and sexuality, we note that these effects can be misleading without considering the manner in which intersecting identities uniquely influence voter evaluations when interacted with the group endorsements. We depict these results in Figure 6 (coefficients in Table C.19 in the Appendix), which plots the ACIE by candidate profile and group endorsement type. Except for demographic profiles without an endorsement, nearly all race-gender-sexuality profiles with some endorsement receive a statistically significant penalty or benefit from those in the high and low position threat latent classes, respectively.
Figure 6. Candidate demographics and endorsement effects. The figure depicts the AMCE of a candidate demographic profile with each group endorsement on the probability of candidate selection relative to a straight White man. Interaction effects are depicted at the 25th and 75th percentiles of the position threat score. Estimates that are not significantly different from zero are included as black dashed bars. N = 7,056. See Table A.4 in the Appendix for the number of profiles evaluated per demographic association pair and number evaluated by respondents low or high in position threat.
Straight White men are generally not penalized if they have group endorsements, even though they do benefit from the race and gender endorsement among those scoring low in status group threat (AMCE = 0.154, 0.086; SE = 0.04, 0.042). We begin to see penalties among straight candidate profiles relative to White men with a neutral endorsement once we consider the other profiles that are endorsed by marginalized groups. Straight White women, straight Black men, and straight Black women are not necessarily penalized among respondents high in social position threat with a neutral endorsement, but each of these profiles is penalized when they have the LGBT group endorsement (AMCE [SE] = –0.164 [0.05], –0.148 [0.051], –0.231 [0.049], respectively). While we see only limited evidence that intersectionally marginalized candidates are penalized more than less intersectionally marginalized candidates, we do find that evaluations of them are affected more consistently, with more marginalized candidates facing more penalties than more privileged ones.
Robustness Checks to Account for Other Possible Explanations
To determine whether our results are the product of our measurement, we tested the robustness of the results to alternative constructions of the position threat variable.Footnote 6 First, we include analyses with the results broken out by racial resentment, sexism, and heterosexism to account for the possibility that one of the measures may be driving the results more than others. Second, we load on the first principal component from the PCA on the 14 questions. This component can account for nearly 40% of the variation in answers to the three batteries. When using this loading in place of the position threat score, the results are substantively the same and, in some cases, more precise. Third, we performed latent class analysis to categorize respondents into low, medium, or high levels of position threat. The findings remain largely unchanged across these measures. They are also substantively similar when broken out by attitude, which is depicted in Appendix 5. The same is true for other demographics, endorsements, and partisanship. That none of these variables is uniquely driving these results suggests that these attitudes are intricately linked in complex ways.
We also take into consideration that our sample is slightly more Democratic than the country as a whole and whether this may be driving our results. Still, we first observe that (when considering partisan leaners) our sample is comparable to the 2016 American National Election Study, only slightly overrepresenting Democrats and underrepresenting Republicans by about 6 points each. We do not expect heterogeneous effects by partisanship, and when breaking out the results by partisanship, our results hold.Footnote 7 As expected, the estimates for Democrats and Republicans are comparable and in the same direction, although the subsample of Republican respondents has less power and therefore is less able to consistently detect significant effects. For example, while the interactions between position threat and a sexuality or race endorsement overall are similar (for sexuality, point estimate = –0.178 [0.029] for Democrats, –0.165 [0.026] for non-Democrats; for race, –0.076 [0.028] for Democrats, –0.0124 [0.035] for non-Democrats), effect sizes are more difficult to detect for specific profile interactions. Additionally, a slightly more Democratic-leaning sample provides a more conservative test of our theory. Even as parties have increasingly sorted on social attitudes, our study shows that among Democrats, we still observe (1) a large subset of voters scoring high in social position threat (and, by extension, racial resentment, sexism, and heterosexism) and (2) large effects among a group of voters whom many would expect to be more favorable to out-groups than conservatives. Even among Democrats, high position threat can lead to penalties for candidates with an endorsement of an organization working on behalf of a marginalized community; our analysis strongly supports our discussion that those low in position threat should be more likely to favor candidates with an endorsement. The fewer respondents high in position threat reduces the power to detect penalties and provides a harder test, but our analysis is still able to consistently detect such penalties while enabling us to consistently identify significant rewards for candidates.
As a final check to account for the predisposition of some ideological, demographic, and partisan groups to have higher or lower position threat, we also conducted our main analyses with two matching analyses. In the first, respondents were matched on their predicted position threat class. In the second, respondents whose threat score is at or above the midpoint of three (indicating higher levels of prejudice and threat) were matched with those below. For both, respondents were matched on party ID (Democrat, independent, or Republican), ideology (liberal, moderate, or conservative), race (only White, non-Black POC, or Black), gender, sexuality, income, and education. We used the following matching algorithms: exact, genetic, random forest, and logistic propensity score. We achieved balance for both matching outcomes on all variables with exact matching and achieved balance on all but party identification and ideology with all other matching methods. Results and additional information using matched data are provided in Appendix 5, and they are substantively the same. We still detect significant effects on the interaction between position threat and other candidate demographics/endorsements as well, suggesting that the results are not based on respondent party or ideology and not merely a product of our model selection.
Discussion and Conclusion
Previous work has shown clear evidence that former president Obama was a powerful motivator in influencing racially voters’ attitudes of policies and things tangentially related to him (Tesler Reference Tesler2016), which other work has extended to consider gender (Bell and Borelli Reference Bell and Borelli2023). In this study, however, we demonstrate that this phenomenon is not necessarily limited to Obama, nor is it limited to race: endorsements by various race, gender, and LGBT organizations are also likely to prime racist, sexist, and heterosexist sentiments, respectively, resulting in situations in which affect toward a given marginalized group shapes attitudes toward those who are connected with the group. This is the case even for those who are not themselves a member of the group. We describe this more general process as associational affect, in which connections with communities that invoke considerations of one’s social position cause views of the community to shape views of the person or thing tied to it. Members of dominant groups in society who are endorsed by groups advocating on behalf of historically marginalized groups are viewed skeptically by people who feel threatened by those groups and viewed positively by people who either belong to or support such marginalized groups. Moreover, this phenomenon works in complex ways, being in large part conditional on the co-constitutive identities of the person who is endorsed.
Our findings have important implications for the study of identity and voting behavior. Marginalized candidates in our study are generally not penalized absent additional information, although this does not mean that they are not subject to various forms of discrimination or prejudice. Rather, our findings suggest that potential coalitional signals often far outweigh the effects of identity alone, a finding in line with work suggesting that (a) members of marginalized groups can win over support of members of dominant groups in choosing not to openly challenge existing power hierarchies (Blee Reference Blee1996); (b) dominant group members can signal commitments to challenging the status quo and win support of minoritized individuals (Wamble Reference Wamble2018). However, this is not to say that endorsements take away the privileges associated with being a member of a dominant group outside of an electoral context. Endorsements can serve as a strong coalitional signal of whether a candidate will challenge the status quo, but “(de)marginalization by proxy” by no means replicates the lived experiences of marginalized communities.
The clear exception to this group was gay and lesbian candidates, who tended to be penalized by voters absent any additional context. One possibility is that LGB politicians are still foreign to many respondents, who may have been surprised to see in such explicit terms male candidates with husbands and female candidates with wives. As scholars such as Haider-Markel (Reference Haider-Markel2010) note, LGB candidates can often overcome prejudice by strategically choosing when and where to run. And leading with information about spouses may have made prejudice toward the LGB candidate salient in a way that showing candidates’ race and gender did not. In any case, the particularly strong reaction toward gay and lesbian candidates—especially relative to Black candidates and women—merits further consideration in future research.
We note briefly here some of the challenges to using an intersectional framework for an experimental design, especially given that identities and voters’ attitudes toward them are not simply additive. While our results thus follow a general pattern, this pattern may not always be perfectly clear-cut given the complex ways in which identities interact. Given the lower probability that a candidate was LGB, there were fewer profiles evaluated, making it more difficult to accurately assess the effects of endorsements here. In terms of future extensions, we do not consider a number of potentially relevant identities that also may be primed through group endorsements. For example, ability, class, and religion can also be salient identities and other dimension on which candidates can be marginalized or privileged. Given the challenges that many marginalized people in these communities face (Sinno Reference Sinno and Sinno2009), we have no reason to expect that endorsements on behalf of these communities would not prime similar considerations among those more or less hostile toward these groups. However, we were not able to include them in the study because of power limitations. Other possible extensions include testing our theory using other non-White candidates and different racial advocacy groups. While we believe the same general pattern would hold—with endorsements on behalf of racial groups in closer proximity to Whiteness being perceived as greater threats by some and lesser among others—we were similarly constrained in our ability to test beyond White and Black candidates and broad racial advocacy organizations due to power concerns. Things other than group endorsements might also prime social position threat, which we did not include here. Finally, we only consider hypothetical progressive groups in our study, although we expect that conservative groups could have similar but opposite effects (e.g., a heterosexist group endorsement increasing support for candidates high in social position threat and decreasing it among those low) on voter behavior.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S1743923X23000612.
Acknowledgments
This project was funded by a grant by the American Studies program at Princeton University. We are very grateful for their support in making this project possible, as well as the generous feedback of Dara Strolovitch, LaFleur Stephens-Dougan, Andrew Flores, the students and faculty of Princeton’s American Political Behavior Group, and the reviewers and editors of Politics & Gender.
Open access
