When women and people of color run for office in the United States, they do well. Female candidates win as often as male candidates (Smith and Fox Reference Smith and Fox2001; Lawless and Pearson Reference Lawless and Pearson2008; Dolan Reference Dolan2014; Anastasopoulos Reference Anastasopoulos2016),Footnote 1 and racial bias appears not to play a decisive role in most modern elections (Highton Reference Highton2004; Abrajano and Alvarez Reference Abrajano and Alvarez2005; Mas and Moretti Reference Mas and Moretti2009; Juenke and Shah Reference Juenke and Shah2016).Footnote 2 Indeed, in the 2018 midterms, female and nonwhite candidates won at rates that equaled or exceeded their white male counterparts (Reflective Democracy Campaign 2019).
So why do women and people of color remain underrepresented in U.S. politics? The candidate emergence literature suggests this disparity may originate in the pre-primary period, when prospective candidates test the waters, decide to run, and establish their viability (e.g., Shah Reference Shah2014; Shah, Scott, and Juenke Reference Shah, Scott and Juenke2019; Doherty, Dowling, and Miller Reference Doherty, Dowling and Miller2019). During this critical time, even slight headwinds can derail a nascent campaign—and compared to white men, women and people of colorFootnote 3 must navigate a rockier path to candidacy, with more bumps and off ramps along the way. Female and nonwhite candidates have to deal with overt harassmentFootnote 4 and disparities in financial resources (Crowder-Meyer Reference Crowder-Meyer2013), party recruitment (Niven Reference Niven1998; Lawless and Fox Reference Lawless and Fox2010; Crowder-Meyer Reference Crowder-Meyer2013; Butler and Preece Reference Butler and Preece2016), personal and professional networks (Fox and Lawless Reference Fox and Lawless2008; Carroll and Sanbonmatsu Reference Carroll and Sanbonmatsu2013; Crowder-Meyer Reference Crowder-Meyer2013), and political ambition and self-efficacy (Lawless and Fox Reference Lawless and Fox2010; Fox and Lawless Reference Fox and Lawless2011).
On top of these well-documented challenges, I identify another obstacle facing diverse candidates in the pre-primary period: strategic discrimination. Strategic discrimination occurs when an individual hesitates to support a candidate out of concern that others will object to some aspect of the candidate’s identity. The problem is not animus toward the candidate. In contrast to direct bias, strategic discrimination is motivated by the belief that a candidate’s identity will cause other people not to donate, volunteer, or vote for him or her.
Strategic discrimination is closely related to the idea of electability. In the run-up to a primary election, party leaders, donors, and activists want to recruit and support a well-qualified candidate who shares their policy preferences. But they also need a candidate who will be capable of winning the general election. So party gatekeepers and primary voters attempt to guess how others will react to prospective candidates. Will a candidate be able to raise the money necessary to run an aggressive campaign? Will he or she generate positive media coverage? Will enough general election voters support a candidate, or will they refuse to vote for him or her?
In this “futures market” of politics (Bai Reference Bai2004), female and nonwhite candidates are at a disadvantage. If party leaders, donors, and primary voters think a candidate could face discrimination later in the campaign season, they may hesitate to place their bets on him or her. Strategic discrimination thus forces female and nonwhite candidates to work doubly hard to establish themselves as real contenders. In addition to running a “traditional campaign,” they also have to run a parallel “‘campaign of belief’ convincing people that it is possible for them to win” (Barbara Lee Family Foundation 2019, 6).
After developing a theory of strategic discrimination, I present a series of three survey experiments that provide initial evidence of strategic discrimination. The first experiment finds that when presented with profiles of hypothetical gubernatorial candidates, Americans consider white men more electable than equally qualified Black women, white women, and less significantly, Black men.
The second experiment was conducted amidst the 2020 Democratic presidential primary. The results suggest that anti-Trump subjects consider candidates’ racial and gender identities when making strategic calculations about who is most likely to beat Donald Trump in 2020. When anti-Trump subjects are primed to think about the strategic importance of male and white voters, they evaluate female and Black Democratic presidential candidates, especially Elizabeth Warren and Kamala Harris, as significantly less likely to beat Trump.
The third experiment investigates possible strategies for combatting strategic discrimination. It finds that informing subjects about the true, low levels of bias against female and Black candidates has no effect. Neither does identifying strategic discrimination as a problem and discouraging subjects from engaging in it. More promisingly, when anti-Trump subjects read a message emphasizing the strategic importance of Black voters, they see Black Democratic presidential candidates as more competitive vis-à-vis Trump. A priming message about the success of a Black female congressional candidate in a majority-white Trump-leaning district has similar if smaller effects for both female and Black candidates. Rather than attempting to change misperceptions of others’ biases, diverse candidates may be better served by emphasizing their own strategic advantages.
Taken in combination, these three experiments suggest that strategic discrimination exists in the abstract, it can affect perceptions of real-world candidates, and it can be tricky to combat. While more research remains to be done, these preliminary results show that strategic discrimination can complicate the road to candidacy for women and people of color. When it comes to candidate emergence, the rules of the game are both raced and gendered.Footnote 5
A New Theory of Discrimination in Politics
Shortly after Abdul El-Sayed began running in the 2018 Democratic primary for governor of Michigan, “very powerful people who call a lot of the shots in the party” sat him down for a little chat. According to El-Sayed, these party insiders told him, “We think you’re great. You just, you know, it’s not that we’re racist. It’s just that we think that people outside of Southeast Michigan are racist, and so you can’t win. See? It makes sense” (quoted in Culham Reference Culham2018).
Variants of this conversation occurred across the United States throughout the 2018 campaign cycle. When former Representative Katie Hill started her campaign in California’s 25th district, key gatekeepers–including a member of the House Democratic leadership–told her they didn’t think a woman could beat incumbent Steve Knight (Kitchener Reference Kitchener2019). A few districts over, California Democratic Party delegates told congressional candidate Omar Siddiqui he was “too brown to win” (Fox News 2018). Similarly, in Alabama, a Democratic party official told congressional candidate Adia Winfrey, “You can’t win because you’re Black” (Gontcharova Reference Gontcharova2018). In Georgia, some longtime allies of Stacey Abrams would not support her gubernatorial campaign because “they did not believe a Black woman could win” (Chira Reference Chira2019).
Suneel Gupta encountered similar concerns during his primary campaign in Michigan’s 11th congressional district. Reflecting on his experience, Gupta concluded that there are
two types of biases. One is the type of bias that you face with [a] person directly. We talk about the type of bias that person has towards you. Then there’s another bias that we don’t talk about enough, which is the bias of, “I’m not racist, but my neighbor is racist, right, and therefore I don’t think you would be a strong candidate, not because I wouldn’t vote for you, but because my neighbor would have a tough time voting for you.” And I think that the second is much harder to address, because it’s not talked about enough. And that is ultimately the thing that I think holds a lot of candidates down. (First We Marched Reference Marched and Ran2019)
I call this second type of bias strategic discrimination. Strategic discrimination occurs when an individual discriminates against someone out of concern that others will object to some aspect of that person’s identity. Even individuals who value diversity may consciously or unconsciously engage in strategic discrimination if they believe that other people are biased.
As in strategic voting, individuals engaged in strategic discrimination support candidates for strategic reasons, rather than according to their true preferences. Yet strategic discrimination also involves behaviors other than voting, such as donating to candidates, volunteering, and making endorsements. These actions shape the field early in a primary, determining who appears on the ballot come election day.
Theoretically, strategic discrimination can affect any candidate who is outside the norm due to his or her sexual orientation, class, age, religion, national origin, parental status, or other dimensions of their identity. However, I focus on gender and race because these are especially salient characteristics, and there is a robust literature on racial and gender discrimination in politics.
Canonical works (e.g., Blank, Dabady, and Citro Reference Blank, Dabady and Citro2004, 56-65) commonly delineate four types of discrimination: intentional or explicit discrimination; subtle or unconscious discrimination; statistical discrimination (also known as profiling); and structural or institutional discrimination. Strategic discrimination is fundamentally different from all these forms of discrimination. The first three types involve an individual directly discriminating against another individual (whether consciously or not); the fourth type identifies structures, institutions, and procedures that unfairly disadvantage some groups while privileging others. Strategic discrimination, by contrast, occurs when an individual makes a judgment or takes an action in anticipation of discrimination by other people.
Intriguingly, strategic discrimination has some parallels to customer-driven discrimination in the labor market. Becker (Reference Becker1971) proposed that taste-based discrimination could originate with employers, coworkers, or customers. Becker’s model implies that customer discrimination should be the most difficult for the market to eradicate, and indeed, racial discrimination is most significant in hiring for jobs requiring direct contact with customers (Nunley et al. Reference Nunley, Pugh, Romero and Seals2015), especially when a business’s customers are of a different race than an applicant (Holzer and Ihlanfeldt Reference Holzer and Ihlanfeldt1998). However, customer discrimination is driven by the actual actions of customers. Strategic discrimination, by contrast, is more centrally motivated by beliefs about the inferred biases of other people, whether or not those biases really exist.Footnote 6
Because strategic discrimination in contemporary U.S. politics is based on incorrect beliefs about others’ willingness to vote for diverse candidates, it has strong similarities to the concept of pluralistic ignorance (Weisz Reference Weisz2020). Pluralistic ignorance occurs when individuals privately hold a belief, but they incorrectly assume that others think differently, misperceiving the aggregate norm (Allport Reference Allport1924; Katz and Allport Reference Katz and Allport1931; O’Gorman Reference O’Gorman1986; Miller and Prentice Reference Miller and Prentice1994). For example, individuals may personally oppose racial segregation while erroneously thinking others in their group support it (O’Gorman Reference O’Gorman1975; Fields and Schuman Reference Fields and Schuman1976; O’Gorman and Garry Reference O’Gorman and Garry1976). Such misperceptions can shape individuals’ behavior and perpetuate unpopular norms. Though seldom referenced in political science,Footnote 7 social psychologists have found evidence of pluralistic ignorance in realms as varied as alcohol consumption (Prentice and Miller Reference Prentice and Miller1993), sexual behavior (Lambert, Kahn, and Apple Reference Lambert, Kahn and Apple2003), and use of paternity leave policies (Miyajima and Yamaguchi Reference Miyajima and Yamaguchi2017).
When it comes to attitudes on race and gender, Americans typically over-estimate others’ levels of intolerance (O’Gorman Reference O’Gorman1975; Fields and Schuman Reference Fields and Schuman1976; O’Gorman and Garry Reference O’Gorman and Garry1976; Williams Reference Williams1990; Do et al. Reference Do, Samuels, Adkins, Clinard and Koveleskie2013; Sobotka Reference Sobotka2020). This reflects the “conservative lag” of pluralistic ignorance: even after individuals have changed their beliefs, they may not realize that others have also updated their attitudes (Miller and Prentice Reference Miller and Prentice1994, 543). Pluralistic ignorance can thus act as a “brake on social change” (ibid.), anchoring decision-making in the prejudices of the past.
This dynamic explains why concerns about the electability of women and people of color are so persistent. Even as large majorities of Americans are themselves comfortable with the idea of a female or Black president (Burden, Ono, and Yamada Reference Burden, Ono and Yamada2017; McCarthy Reference McCarthy2019; Gallup 2019), they doubt that others feel the same way (e.g., King, Elbeshbishi, and della Cava Reference King, Elbeshbishi and Cava2019). In Study 1, for example, a national sample of U.S. adults estimates that on average 47% of other Americans would not vote for a woman for president, and 42% of other Americans would not vote for a Black person for president. Though not precise measures of U.S. public opinion, these estimates are notable because they far exceed recorded levels of bias against female and Black presidential candidates, as illustrated in figure 1.Footnote 8 With such a high degree of skepticism about others’ willingness to support diverse candidates, conditions are ripe for strategic discrimination.
Figure 1 Perceptions versus reality
Study 1: Strategic Discrimination in the Abstract
In 1971, leading presidential candidate Senator Edmund Muskie was asked whether he would consider selecting a Black running mate, should he be the Democratic nominee. Muskie said he would not, because “in [his] judgment such a ticket was not electable” (quoted in Naughton Reference Naughton1971).
Nearly fifty years later, some commentators argue that electability is still code for “white and male” (e.g., Bacon Reference Bacon2018; Zhou Reference Zhou2019). To evaluate this claim, I designed a survey experiment in the tradition of the Goldberg paradigm (Reference Goldberg1968).Footnote 9 Study 1 investigates whether Americans consider white male candidates more electable than equally qualified female and Black candidates.
Methodology
Study 1 was conducted with a nationally representative sample of 1,947 U.S. adults on May 23–7, 2019. The implementing vendor (Lucid) constructed the sample to match the census on key demographics. While not the same as probabilistic sampling, Lucid samples have been shown to return experimental results that correspond closely to results from random samples (Coppock and McClellan Reference Coppock and McClellan2019).
The experiment was part of a collaborative survey fielded by MIT’s Political Experiments Research Lab (PERL). After answering demographic questions, an attention check question, and questions about political ideology and knowledge, subjects were asked to evaluate a series of three profiles of hypotheticalFootnote 10 gubernatorial candidates.Footnote 11 The profiles appeared in random order, one at a time, on three successive screens. As described in table 1, each profile listed the candidate’s current position, prior offices held, education, profession, age, race [Black/white], and gender [male/female]. Race and gender were randomized so that 25% of the profiles were white female candidates, 25% were white male candidates, 25% were Black female candidates, and 25% were Black male candidates.
Table 1 Candidate profiles for Study 1
Dependent Variables
Below each profile, the subjects were asked, “If this candidate ran for governor in your state, how electable would [he/she] be?” with a 4-point response scale ranging from very electable (4) to very unelectable (1). Based on this question, I construct two dependent variables: an electability score (Electability) and a binary variable indicating whether each candidate profile is considered “very electable” (VeryElectable).
Hypothesis
Study 1 tests a single hypothesis:
H1: White male candidates will be evaluated as more electable than otherwise identical white female, Black female, and Black male candidates.
Results
Table 2 reports the results of Study 1.Footnote 12 On average, when candidate profiles are labeled as Black women, white women, and Black men, they receive lower electability scores than when the same profiles are labeled as white men. This effect is statistically significant for Black female (p<0.001) and white female candidates (p<.05), but not for Black male candidates.
Table 2 Variation in perceived electability by candidate race and gender
All models are OLS. Robust standard errors clustered by respondent are in parentheses.
* = p<0.05, ** = p<0.01, *** = p<0.001
Table 3 Democratic presidential candidates in Study 2
Similarly, as compared to white male candidates, subjects are less likely to consider Black women, white women, and Black men “very electable.” Candidate profiles identified as white men are rated “very electable” 37% of the time. For Black male candidates, this number is 35%; for white women, 32.5%; and for Black women, 30.4%. The differences between white men and white women and white men and Black women are statistically significant (p<0.05 and p<0.001, respectively), but the difference between white men and Black men is not statistically significant (p=.283).
Robustness Checks and Quality Control
The results reported in table 2 are substantively the same if estimated with ordered probit (Electability) or probit (VeryElectable), and they are robust to the inclusion of profile fixed effects (refer to online appendix tables 1.1-1.3). Dropping subjects who failed an attention-check question increases the magnitude and significance of the results, and the difference in electability scores for Black male candidates becomes statistically significant (p<0.05; online appendix table 1.4). Similarly, the results of Study 1 are not driven by politically disengaged respondents who would be unlikely to participate in a primary election (online appendix tables 1.12-1.13). To the contrary, subjects with higher levels of political knowledge tend to show stronger responses to the experimental manipulation (online appendix tables 1.5-1.10).
Table 4 Average treatment effects, male voters treatment
Strategic Discrimination and Direct Discrimination as Possible Mechanisms
Theoretically, subjects may be rating white male candidates as more electable than other types of candidates due to both the subjects’ own biases (direct discrimination) and their estimates of others’ biases (strategic discrimination). It is difficult to adjudicate between these two mechanisms.Footnote 13 Direct discrimination probably plays some role in influencing how subjects rate candidates’ electability. Yet at the same time sub-group analysis suggests that direct discrimination cannot fully explain the results of Study 1.Footnote 14
In the United States, sexism and racism are strongly correlated with older age and lower educational attainment (Heerwig and McCabe Reference Heerwig and McCabe2009; Parker, Graf, and Igielnik Reference Parker, Graf and Igielnik2019). However, the results of Study 1 do not show any clear generational patterns (online appendix tables 1.16-1.17), and the effects of Study 1 are largest among the most educated subjects (online appendix tables 1.14-1.15). These sub-group results are puzzling and inconsistent with the idea that direct discrimination is the only mechanism at work in Study I.
So is strategic discrimination driving the results of Study 1? Perhaps. In addition to completing the candidate evaluation exercise, Study 1 asked subjects to estimate the percentage of other Americans who would not vote for a woman for president and the percentage who would not vote for a Black person for president. Among subjects who over-estimate others’ biases, Study 1’s main findings have greater magnitude and statistical significance (online appendix tables 1.19-1.29), with the results for Black male candidates reaching conventional levels of statistical significance (online appendix tables 1.24-1.29). By contrast, among subjects who have accurate or low estimates of others’ levels of bias, Study 1 generally produces null effects (online appendix tables 1.19-29). In some model specifications, subjects with accurate or low estimates of others’ racism actually rate Black men as significantly more electable than white men (online appendix tables 1.24 and 1.27).
Though not a smoking gun, these patterns are consistent with the notion that concern about others’ biases could be causing individuals to doubt the electability of diverse candidates. But are subjects’ estimates of others’ levels of racism and sexism simply a reflection of their own beliefs? Research on pluralistic ignorance finds that individuals’ estimates of others’ views are shaped by two biases: “looking glass bias” and “conservative bias” (Fields and Schuman Reference Fields and Schuman1976). It is true that individuals project their own views onto others. Yet at the same time they also tend to assume that others’ beliefs are more conservative than they really are. Taken in combination, these two biases typically produce a weak positive correlation between subjects’ own beliefs and their estimates of others’ beliefs, as in Mildenberger and Tingley (Reference Mildenberger and Tingley2019) and Sobotka (Reference Sobotka2020).Footnote 15 In studies of racism and sexism, even highly tolerant subjects have been shown to over-estimate others’ levels of intolerance (Fields and Schuman Reference Fields and Schuman1976; O’Gorman Reference O’Gorman1975; O’Gorman and Garry Reference O’Gorman and Garry1976; Do et al. Reference Do, Samuels, Adkins, Clinard and Koveleskie2013; Sobotka Reference Sobotka2020). This may explain why in Study 1, subjects’ estimates of others’ levels of bias are not correlated with known predictors of racism and sexism (online appendix table 1.33).
Partisan Cues as an Alternative Mechanism?
Do subjects consider female and Black candidates less electable because of their racial and gender identities, or because race and gender are cues for partisanship? To evaluate this possibility, I analyzed Study 1 for two sub-groups: subjects in states with Democratic governors, and subjects in states with GOP governors. For white female and Black female candidates, the results are largely consistent across the two sub-groups. Black male candidates, however, are rated less electable only by subjects whose states have GOP governors (online appendix table 1.18). This suggests that inferred partisanship could be driving Study 1’s limited findings regarding Black male candidates. By contrast, Black and white female gubernatorial candidates are seen as less electable even among subjects from Democratic-leaning states.
Discussion and Context
Study 1 finds that perceptions of electability vary according to candidates’ racial-gender identities. The perceived electability gap is especially severe for Black women. Compared to an identical white man, a hypothetical Black female gubernatorial candidate is about 20% less likely to be rated “very electable.” The numbers are even worse when considering only responses from the types of people most likely to participate in candidate selection; subjects who are attentive, politically knowledgeable, and ideological rate Black female candidates “very electable” 27.5% of the time, as compared to 37.7% for white male candidates. Yet with the notable exception of Philpot and Walton (Reference Philpot and Walton2007), Black female candidates are largely ignored in studies of public opinion toward candidates.Footnote 16 The results of Study 1 cry out for greater attention to intersectionality in future work in this field.
Sub-group analysis suggests that in addition to direct discrimination, strategic discrimination is a plausible mechanism for the results of Study 1. However, Study 1 is based on evaluations of hypothetical candidate profiles in an artificial, abstract context. In actual elections, people form opinions about candidates based on many different factors. The next experiment assesses whether and how strategic discrimination can influence the perceived competitiveness of real-world candidates.
Study 2: Strategic Discrimination in the 2020 Democratic Presidential Primary
In the 2020 Democratic presidential primary, electability was a top concern for voters (Quinnipiac 2019)—and the concept was often linked to race and gender. For example, some activists worried that “after the experience of 2016” their party “might need to flee to the safety of a white, male candidate” (Weigel Reference Weigel2019).
Yet even as they referenced candidates’ racial-gender identities, debates about electability also invoked candidates’ policy positions and qualifications. Consider this statement from South Carolina State Senator Dick Harpootlian, a prominent supporter of former vice-president Joe Biden:
This is do-or-die, and Joe Biden is the best candidate to go against Trump in November. Would Joe Biden be running if he thought any of these other folks could beat Donald Trump? No way. We can’t risk this thing with someone who has not done this before, who is unchallenged, who is untested. There is something to be said for two old white guys going at it. The African Americans in the State Senate with me are going to be with him overwhelmingly. Because this is a pragmatic year. This isn’t a battle of ideologies or identity or Medicare for All or Green New Whatever. It’s all about who can stop this juvenile narcissist from getting a second term. (quoted in Hamby Reference Hamby2019)
Harpootlian weaves together multiple arguments, ranging from Biden’s experience to the apparent desirability of seeing “two old white guys going at it.” How can we disentangle these factors?
To deal with this challenge, I conducted a survey experiment that exposes some subjects to priming messages designed to cue the strategic importance of white or male voters. If as compared to the control group, the subjects exposed to these treatments consider Black and female candidates less competitive, that would show how strategic concerns about race and gender can play a role in shaping assessments of real-world candidates’ competitiveness.Footnote 17 To clarify, Study 2 does not measure actual levels of bias present in the 2020 Democratic primary,Footnote 18 nor is it a study of voting behavior.Footnote 19 Rather, this experiment seeks to evaluate whether a candidate’s race and gender can influence perceptions of their competitiveness, even in the noisy, multi-dimensional context of an ongoing election.
Methodology
Study 2 was fielded on MTurk from May 6–11, 2019. High-quality MTurk workers who live in the United States were eligible to participate. Out of 3,386 people who took the initial screening questions, 1,702 subjects met the criteria for inclusion: they stated that they voted in the 2016 presidential election, they did not vote for Donald Trump, and they do not support Donald Trump’s reelection in 2020.
The subjects who completed the survey experiment are evenly divided between women and men. In 2016, the Democratic primary electorate was 58% female (Brownstein Reference Brownstein2019a), so women are under-represented in Study 2. The subjects are also younger (modal age range 25–34 years) and less racially diverse (70% white) than Democratic primary voters, who were 62% white in 2016 (Brownstein Reference Brownstein2019a). Online appendix table 1.34 contains a full demographic profile of the subjects.
Control and Treatment Groups
The subjects who were screened into the full survey were randomly divided into four equal groups. All subjects were told that a large number of Democrats are competing to run against Donald Trump in 2020. Then they saw the names, titles, and photos of the top eight Democratic presidential contenders, based on current polls at the time of the experiment. The candidates appeared one at a time in random order.
After viewing the candidates, the control group proceeded directly to a screen where they were asked which candidates had the best chance of beating Donald Trump in 2020. They were given a randomly ordered list of the eight candidates, and they were asked to drag their top three candidates into a box on the screen. The instructions specified that the candidate in the #1 position should be the person with the best chance of beating Trump, the candidate in the #2 position should have the second-best chance, and the candidate in the #3 position should have the third-best chance. After this exercise, the control group proceeded to a concluding module with demographic questions.
Before doing the ranking exercise, the subjects randomized into the “Male Voters” treatment group read a priming message emphasizing the strategic importance of winning male voters in 2020. Another treatment group (“White Voters”) read a priming message emphasizing the strategic importance of winning white voters in 2020. Both messages were condensed from actual narratives circulating in late 2018 and early 2019 (e.g., Hohman Reference Hohman2018; Brownstein Reference Brownstein2019b; Riccardi Reference Riccardi2019). The full text of the messages is in online appendix table 1.36.
The third treatment group (“Estimate Bias”) was informed that to beat Donald Trump in 2020, the Democratic presidential nominee needed to be able to win key swing states. These subjects were then asked to estimate the percentage of swing-state voters who would not vote for a woman for president and the percentage who would not vote for a Black person for president. As in Study 1, most respondents overestimated others’ biases. On average, they estimated that 38.5% of swing state-voters would not vote for a woman for president, and 37.4% would not vote for a Black candidate.
Dependent Variables
The main dependent variables are binary measures of whether each subject’s list of the top three candidates with the best chances of beating Trump includes at least one woman (IncludeWoman) or at least one Black candidate (IncludeBlack). Because there are multiple ways of interpreting the ranking exercise, I also code four additional dependent variables. Two are binary measures of whether a female (TopWoman) or Black (TopBlack) candidate occupies the #1 position in a subject’s list. The others are the total number of Black (TotalBlack) and female (TotalWomen) candidates included in a subject’s list of the top three most competitive candidates.
In some extensions of the analysis, I also code dependent variables that measure outcomes for specific candidates, including binary measures of whether each candidate was included among the top three most competitive candidates and whether each candidate occupied the top position.
Hypotheses
Study 2 tests three hypotheses:
H2: When subjects are told that winning the support of male voters is key to victory in 2020, they will evaluate female candidates as being less capable of beating Trump.
H3: When subjects are told that winning the support of white voters is key to victory in 2020, they will evaluate Black candidates as being less capable of beating Trump.
H4: When subjects are asked to estimate the percentages of swing state voters who will not vote for female and Black presidential candidates, they will evaluate female and Black candidates as being less capable of beating Trump.
Results—Strategic Messaging Treatments
All results are average treatment effects (ATEs). Each ATE is the difference in the means of the control group and a treatment group, estimated using Welch’s t-test.
Male Voters Treatment
When subjects are told that winning male votes is the path to victory in 2020, they are less likely to say female candidates are well-positioned to beat Donald Trump. In the control group, 70.5% of subjects include at least one woman in their list of the top three most competitive candidates, compared to 56.4% in the male voters treatment group. This effect is statistically significant (p<0.001). Similarly, subjects in the male voters treatment group include fewer women in their top three list, and they are less likely to list a female candidate as having the best chance of beating Trump (7.7% versus 15.6% in the control group, p<0.001).
White Voters Treatment
In the control group, 49% of subjects include at least one Black candidate in their list of the top three most competitive candidates. Among subjects told that white voters are the key to beating Trump, 41% do so (p<0.05). Subjects in the white voters treatment group are also less likely to say that a Black candidate has the best chance of beating Trump (4% versus 9.2%, p<0.01), and they include fewer Black candidates in their top three lists (p<0.05).
Candidate-Specific Results
Moving beyond the main results of Study 2, figures 2 and 3 show the average treatment effects by candidate. Each arrow shows the differences between the control group and the treatment group. Black arrows have p-values < 0.1; gray arrows are statistically insignificant.
Figure 2 Average treatment effects by candidate, male voters treatment
Figure 3 Average treatment effects by candidate, white voters treatment
Table 5 Average treatment effects, white voters treatment
Compared to the control group, the male voters treatment group is markedly less optimistic about Elizabeth Warren’s and Kamala Harris’s chances of beating Donald Trump.Footnote 20 In the control group, 7.5% of subjects say Harris has the best chance of beating Trump, compared with 3.5% in the male voters treatment group (p<0.05). Warren experiences a similar decline from 7.1% to 3.3% (p<0.05). Both Warren and Harris also see their chances of being considered among the top three most competitive candidates decrease by about ten percentage points.
Highlighting the importance of intersectionality, Kamala Harris is penalized again by the message about the strategic importance of white voters. As illustrated in figure 3, Harris’s probability of being rated most competitive falls from 7.5% in the control group to 2.6% in the white voters treatment group (p<0.001).Footnote 21
On average, white candidates benefit from the white voters treatment, and male candidates benefit from the male voters treatment. But intriguingly, figures 2 and 3 show that these increases in perceived competitiveness are not distributed evenly across all male and white candidates. This unexpected result may be due to the specific dynamics of the 2020 Democratic presidential primary, or it could reflect a broader pattern of some kind. Future researchers may want to explore the question of who benefits most from strategic discrimination, and under what circumstances.
Results—Estimate Bias Treatment
Before they rated the candidates’ competitiveness, a third treatment group was asked to estimate the percentage of swing-state voters who would not vote for a woman for president, and the percentage of swing-state voters who would not vote for a Black person for president. As reported in table 6, the effects of this treatment are statistically insignificant—though the negative effects for TotalBlack and IncludeBlack are close to conventional levels of statistical significance (p=0.116 and p=0.132, respectively).
Table 6 Average treatment effects, estimate others’ biases treatment
These results may be statistically insignificant because subjects responded to the treatment heterogeneously. Most subjects over-estimate others’ biases, while about one-quarter of subjects have accurate or low perceptions of others’ racism and sexism. Compared to the subjects who have low estimates, subjects who over-estimate others’ biases are more likely to construct all-male or all-white lists of candidates best able to beat Trump, and on average they include fewer Black and female candidates. These correlations are consistent with the theory of strategic discrimination, though it is important to note that they are correlations, not evidence of a causal effect.Footnote 22
Discussion and Context
Study 2 is best understood as a proof-of-concept experiment.Footnote 23 The results show that under certain circumstances, subjects can be induced to consider primary candidates’ racial-gender identities as they decide who would be most competitive in a general election. Study 2 also suggests that media coverage and analysis from pundits may affect the perceived competitiveness of candidates of different races and genders.
That said, Study 2 has some limitations. In particular, readers should be cautious about generalizing too broadly from these results. Study 2 was conducted at a unique moment in history, in the aftermath of Hillary Clinton’s 2016 loss to Donald Trump and within recent memory of the Obama presidency. Furthermore, Study 2 was conducted with a specific set of candidates. Pre-existing beliefs about those candidates likely interacted with the experimental treatments. There is no way to know if the same results would have been obtained had a different mix of candidates decided to seek the 2020 Democratic presidential nomination.
Study 3: Combatting Strategic Discrimination
When candidates encounter strategic discrimination, how should they respond? Early in 1996, this question was front of mind for former Charlotte mayor Harvey Gantt, Jr., who is Black. Gantt was embroiled in a tough Democratic primary for U.S. Senate, competing against white pharmaceutical executive Charlie Sanders.
Sanders ran on the slogan “the one Democrat who can beat Jesse Helms” (Germond Reference Germond1996), and he “made electability the primary rationale for his campaign” (Sack Reference Sack1996). Electability was widely understood as “subtle code for race” (ibid.)—“the idea that a Black candidate could not defeat Helms” (Ahearn and Alexander Reference Ahearn and Alexander1996). Even among Black voters, “there [was] a strong feeling that white North Carolinians [would] not let a Black man beat somebody like Jesse Helms” (Sack Reference Sack1996). As one pastor said, “Harvey can’t beat Jesse [Helms]. No Black can. It’s sad. This is a great country, but it’s not perfect” (quoted in Germond Reference Germond1996).
Gantt responded to these doubts by acting “firmly to bring the racial issue into the open so he [could] combat it on his own terms” (Sack Reference Sack1996). As he told one majority-Black audience, “I think it’s wrong for anybody, four years from the 21st century, whether from my opponent or on their own, to suggest that we can’t win because of the pigmentation of our skin. That is a corrosive and damaging argument” (quoted in Sack Reference Sack1996).
Gantt ultimately prevailed in his primary, so this approach appears to have worked for him. Should other candidates facing strategic discrimination deploy similar moral arguments? Or might other responses work better?
To evaluate strategies for combatting strategic discrimination, I designed a follow-up experiment (Study 3) based on Study 2.
Methodology
The structure of Study 3 is the same as Study 2, except instead of treatments designed to induce strategic discrimination, Study 3 evaluates four treatments intended to mitigate strategic discrimination. Study 3 was fielded on MTurk from May 28–June 2, 2019. MTurk workers who had already participated in Study 2 were ineligible for Study 3; 4,561 subjects took the screening questions for Study 3, and 2,219 completed the full experiment. Full subject demographics are reported in online appendix table 1.35.
Dependent Variables and Control and Treatment Groups
The dependent variables for Study 3 are identical to those in Study 2, and Study 3 includes the same candidates. As in Study 2, the control group subjects proceeded directly to ranking the top three Democratic presidential candidates with the best chances of beating Donald Trump in 2020.
Before evaluating the Democratic presidential candidates, the first treatment group (“Correct Information”) was told that levels of bias against female and Black candidates are at historically low levels, and the vast majority of Americans are willing to vote for a female or Black president.
The second treatment group (“Naming and Shaming”) was told that some Democrats think they need to nominate a white man to be able to win in 2020. They were told that this type of thinking is called strategic discrimination, and it unfairly advantages white male candidates. They were also told that even people who value diversity can unintentionally engage in strategic discrimination.
The third treatment group (“Role Model”) was told that to win in 2020, Democrats should consider what worked for their party in 2018. They were then provided with a short vignette about Representative Lauren Underwood, a Black woman who beat a white male GOP incumbent in a majority white district that voted for Trump in 2016. The vignette did not explicitly mention Underwood’s race, but it was accompanied by her official portrait.
The fourth treatment group (“Black Voters”) saw a message emphasizing the strategic importance of Black voters. The text closely paralleled the messages about white and male voters used in Study 2. The full text of all these treatments can be found in online appendix table 1.37.
Hypotheses
Study 3 tests four hypotheses:
H5: When subjects are informed of the true low levels of bias facing female and Black candidates, they will evaluate female and Black candidates as being more capable of beating Trump.
H6: When subjects are informed that strategic discrimination advantages white men and unfairly harms female and Black candidates, they will evaluate female and Black candidates as being more capable of beating Trump.
H7: When subjects are primed with a vignette about a successful female African American congressional candidate, they will evaluate female and Black candidates as being more capable of beating Trump.
H8: When subjects are told that high Black turnout is the key to beating Trump, they will evaluate Black candidates as being more capable of beating Trump.
Results
All results reported in tables 7–10 are average treatment effects (ATEs) estimated using Welch’s t-test. I find support for H7 and H8, but not H5 or H6.
Table 7 Average treatment effects, correct information treatment
Table 8 Average treatment effects, naming and shaming treatment
Table 9 Average treatment effects, role model treatment
Table 10 Average treatment effects, Black voters treatment
Correct Information Treatment
For the correct information treatment, we cannot reject the null hypothesis of no effect. Contrary to Dowling and Miller (Reference Dowling and Miller2015), I find no evidence that facts change subjects’ beliefs about the competitiveness of female or Black candidates. Instead, my findings are consistent with a broader literature suggesting that when it comes to politics, misperceptions are sticky (Nyhan and Reifler Reference Nyhan and Reifler2010; Berinksy Reference Berinksy2017).
Naming and Shaming Treatment
The results of the naming and shaming treatment are also statistically insignificant. Even when subjects are explicitly told that strategic discrimination is a problem that unfairly disadvantages female and Black candidates, they do not meaningfully increase their assessments of female and Black candidates’ competitiveness against Donald Trump. This null effect suggests that merely calling out strategic discrimination is not a promising strategy for combatting it.
Role Model Treatment
When subjects are primed with the vignette about Representative Underwood, they rate Black and female presidential candidates as significantly more competitive against Donald Trump. Subjects in the role model treatment group include more female and Black candidates in their top three lists. They are more likely to put a female candidate in their #1 position (21.7% versus 12.8% in the control group; p<0.001), and they are more likely to say that a Black candidate has the best chance of beating Donald Trump (p<0.001).
These results are largely driven by subjects’ perceptions of Kamala Harris—7.8% of the subjects in the role model treatment group list Kamala Harris as having the best chance of beating Trump (versus 2% in the control group; p<0.001). Similarly, 39.7% of the treatment group includes Harris in their top three lists, as compared to 27.9% of the control group (p<0.001).
Black Voters Treatment
When subjects are told that Black voters are the key to victory in 2020, they rate Black candidates as more competitive against Donald Trump. As compared to the control group, subjects in the Black voters treatment group are more likely to list a Black candidate as having the best the chance of beating Trump in 2020 (4.9% versus 15.9%; p<0.001). They are about fourteen percentage points more likely to include at least one Black candidate in their list of the top three most competitive candidates (p<0.001), and on average they include more Black candidates on their lists (p<0.001).
Both Kamala Harris and Cory Booker benefit significantly from this treatment. As compared to the control group, Harris’s chances of being rated most competitive against Trump increase from 2% to 8.1% (p<0.001), while Booker’s chances increase from 2.9% to 7.8% (p=0.01). Similarly, both candidates see large jumps in their chances of being included among subjects’ top three most competitive candidates, from 27.9% to 42.1% for Harris (p<0.001) and from 18% to 26.8% for Booker (p<0.01).
Discussion and Context
Study 3 suggests that for candidates trying to overcome strategic discrimination, it is most productive to make the case that fielding diverse candidates advances the goal of winning elections. By contrast, messages that call out strategic discrimination and attempt to correct subjects’ misperceptions about others’ biases have no statistically significant effect. However, it is important to note that Study 3 used written priming messages, rather than videos or interactive exercises that might be more compelling. A different strategy for communicating these messages could yield different results.
Additionally, in real elections, candidates seeking to overcome strategic discrimination often use a strategy that was not possible to evaluate in Study 3: they out-perform the competition, proving viability by shattering fundraising records, winning debates, and notching up victories in straw polls, caucuses, and primaries. To be clear, out-performing the competition is an imperfect, individual-level approach to deal with strategic discrimination, not a means of solving the broader problem. It is neither reasonable nor equitable to say that if women and people of color want to be taken seriously as candidates, they must be exceptional. Put simply, not everyone can be Barack Obama.
Nonetheless, it is worth learning from Obama’s path to the White House. In 2007 and 2008, many people—including African AmericansFootnote 24—were skeptical of Obama’s candidacy. They wondered, “Is America ready to elect a Black president?” (e.g., Crowley and Johnson Reference Crowley and Johnson2007; 60 Minutes 2008, min 4:40 and 6:48), and they worried that the answer might be “no.” To quote one young Black voter in South Carolina, “Personally, I don’t think he has a chance in hell. All those white people? Come on!” (quoted in Helman Reference Helman2007).
Obama eventually overcame these concerns by showing he could win over the very voters expected to be most biased against him. The Obama campaign deployed white surrogates (Zengerle Reference Zengerle2008), put “diverse but mostly white faces” on the risers at campaign events (ibid.), and produced campaign videos that intentionally featured footage of white audiences applauding enthusiastically (Zeleny Reference Zeleny2008).Footnote 25 Finally, Obama gained crucial momentum when he won the Iowa caucuses, proving that yes, a Black man could win even in the whitest corners of America.
Conclusion
Strategic discrimination is a subtle yet consequential form of discrimination in politics. All else being equal, Americans see hypothetical white male candidates as more electable than Black women, white women, and to a lesser extent, Black men. Real-world candidates’ racial-gender identities can also affect their perceived competitiveness. These dynamics are strongly intersectional. While problematic for white women and Black men, my experiments show that strategic discrimination poses particularly steep challenges for Black women. Though somewhat limited in their generalizability, these initial findings should mark strategic discrimination as an important phenomenon worthy of further study.
But why is strategic discrimination so salient at this particular moment in U.S. history? The answer may lie in three parallel trends: increasing diversity among candidates, increasing awareness of racism and sexism in society, and increasing political polarization.
First, a historic surge of women and people of color is flooding into politics, particularly on the Democratic side. Their presence may be prompting more active consideration of questions that were purely theoretical in the past. Without many Black candidates running in majority-white districts, and without many women running, period, there was little reason to contemplate whether diverse candidates were electable. Now such concerns are more immediate and tangible.
Second, from Black Lives Matter to the 2016 election to #MeToo to the Women’s March to “shithole countries” to Charlottesville to the Kavanaugh hearings, issues of race and gender are front and center in America’s national conversation. The percentages of Americans naming racism and sexism “very big” national problems have increased markedly in recent years, especially since 2016 (Neal Reference Neal2017; Hartig and Doherty Reference Hartig and Doherty2018). These changes in public opinion are generally framed as good for diversity, but strategic discrimination could be an unexpected side effect. If Americans are increasingly concerned about the prevalence and severity of racism and sexism, they may worry that sexism and racism will keep female and nonwhite candidates from winning general elections.
Finally, political polarization may fuel strategic discrimination. Hayes and Lawless (Reference Hayes and Lawless2016, 7) contend that polarization has “significantly leveled the playing field” for female candidates—and when it comes to general elections, they are right. Under conditions of extreme polarization, general election votes are cast largely according to party affiliation, with little room for a candidate’s identity to matter. But at the same time, as politics devolves into partisan warfare, each side becomes ever more desperate to “just win, baby!”Footnote 26 As a result, primaries may hinge more and more on electability, which is a raced and gendered concept.Footnote 27 Even if women and people of color objectively win their elections at the same rates as white men, they are perceived as less electable. So donors, party activists, and primary voters may gravitate toward white male candidates who feel like a safe bet, rather than taking a risk on a woman or a person of color. That’s strategic discrimination in action.
Supplementary Materials
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S153759272000242X.
The online appendix includes additional tables and information about the experiments.
