Ethnoracial Minority Candidates and Campaign Contributions

Are there ways to increase the representation of people of color in campaign finance? Activists and researchers often recommend that the parties recruit and support minority candidates as a solution to unequal participation. If members of the public use shared ethnorace as a meaningful signal of shared experience or attitudes, or feel a sense of empowerment in the presence of coethnic candidates—an “ethnic-candidate paradigm”—the presence of minority candidates may encourage coethnic participation (e.g., Barreto Reference Barreto2007, Reference Barreto2010; Keele et al. Reference Keele, Shah, White and Kay2017; Rocha et al. Reference Rocha, Tolbert, Bowen and Clark2010; Shah Reference Shah2014).^{Footnote 5} Such an increase could lead to greater equality of participation.

Other studies have found minimal or even perverse effects of the presence of minority candidates on increasing coethnic participation or creating a more representative electorate (Fraga Reference Fraga2016a; Gay Reference Gay2001; Henderson, Sekhon, and Titiunik Reference Henderson, Sekhon and Titiunik2016). For instance, the presence of black candidates appears to increase white turnout more than black turnout, resulting in an overall whiter electorate and reducing Democratic vote shares on average (Washington Reference Washington2006). Evidence of the effect of candidate ethnorace on producing more ethnoracially representative electorates is mixed.

We argue that coethnic empowerment and the ethnic-candidate paradigm are especially likely to occur in campaign finance. We theorize “push” and “pull” factors that may generate patterns of coethnic contributing. In the aggregate, we are likely to observe coethnic contributing because ethnorace is a strong predictor of party identification, ideology, and geography; we aim to avoid these confounders with our research designs. We focus here on reasons that shared ethnorace, all else equal, may influence contributing.

A principal “push” factor, driven by contributors, is based on potential donors’ feelings of linked fate. Linked fate is the belief that one’s individual experience is tied to the collective experience of the ethnoracial group, and greater feelings of linked fate predict support for coethnic candidates among African American and Latino voters (Dawson Reference Dawson1995; McConnaughy et al. Reference McConnaughy, White, Leal and Casellas2010; Wallace Reference Wallace2014). Individuals with strong feelings of linked fate may, in pursuit of self-interest, donate to coethnic politicians. Factors that are positively correlated with contributing, such as education and socioeconomic status, have been found to be positively correlated with feelings of linked fate among African Americans (Dawson Reference Dawson1995; Gay Reference Gay2004; Simien Reference Simien2005; Tate Reference Tate1994). There is also evidence that politicians from marginalized identity groups tend to exert greater effort to represent and improve the standing of their group in society (e.g., Broockman Reference Broockman2013; Dawson Reference Dawson1995; Logan Reference Logan2018).

Linked fate may complement perceptions of ethnoracial group competition among potential donors. In local and national contexts, individuals perceive their ethnoracial group to be engaged in competition with other groups for economic and political resources (e.g., Gay Reference Gay2006; Kim Reference Kim2000; McClain et al. Reference McClain, Carter, DeFrancesco Soto, Lyle, Grynaviski, Nunnally, Scotto, Kendrick, Lackey and Cotton2006; Sanchez Reference Sanchez2008). Perceptions of out-groups as competitive threats, whether based on stereotyping or feelings of collective alienation (Bobo and Hutchings Reference Bobo and Hutchings1996), can increase incentives to support coethnic candidates over candidates from other ethnoracial groups.

Contributions to coethnic candidates may also serve expressive, rather than self-interested, motivations. Contributing, like voting, may “serve as a positive affirmation of identity group membership or as an expression of group solidarity and support, both of which convey psychological benefits” (Horowitz Reference Horowitz1985; Jackson Reference Jackson2011; Valenzuela and Michelson Reference Valenzuela and Michelson2016, 618). Donors often explain their motivations for contributing in terms of linked fate, identity expression, and empowerment. In 2016, for instance, the Los Angeles Times interviewed Lily Lee Chen, one of many Asian Americans who contributed large sums to John Chiang’s campaign in California. Chen explained her contribution with empowerment theory: “He would serve as a model for all the Chinese American young people who have political aspirations and want to be good public servants” (quoted in Willon Reference Willon2016).

Campaigns are also likely to create “pull” factors by appealing to potential donors’ identities in solicitations and appeals for contributions. Campaign appeals to coethnicity may prime feelings of linked fate or increase identity strength by “selectively reinforc[ing] the preexisting identity” (Jackson Reference Jackson2011; Valenzuela and Michelson Reference Valenzuela and Michelson2016; Rogers, Fox, and Gerber Reference Rogers, Fox, Gerber and Shafir2013, 100).^{Footnote 6} Prior studies have investigated the role of “ethnically angled advertisements” in campaigns, such as, in the case of the Latino community, television commercials featuring Latino narrators, pictures of Latinos, and descriptions of a candidate’s connections to the Latino community (e.g., Abrajano Reference Abrajano2010; Soto and Merolla Reference Soto and Merolla2006)—appeals that are likely to be more effective when the candidate is also Latino (Barreto Reference Barreto2007, Reference Barreto2010). This kind of appeal may extend to campaign finance. Political action committees (PACs) further facilitate coethnic fundraising. According to the mission statement of Latino Victory Fund PAC, the organization “identifies, recruits, and develops candidates for public office while building a permanent base of Latino donors to support them.”

Candidates also contact and solicit contributions from individuals in their social, educational, and professional networks (Bonica Reference Bonica2017a, Reference Bonica2017b). Such networks, including at the most elite levels of business and education, are shaped by ethnoracial identity (Allen, Epps, and Haniff Reference Allen, Epps and Haniff1991; McPherson, Smith-Lovin, and Cook Reference McPherson, Smith-Lovin and Cook2001). Lawyers have long been the majority of American political candidates and officeholders, which may be explained in part by lawyer candidates’ unique ability to tap their professional networks for funds and other support (Bonica Reference Bonica2017b). Only in recent decades, law schools have been open to black Americans (Gellhorn Reference Gellhorn1968), and de facto barriers for people of color in the legal profession persist even as de jure barriers have declined (Kornhauser and Revesz Reference Kornhauser and Revesz1995; Nussbaumer Reference Nussbaumer2006). In turn, during the 1960s and 70s lawyers of color created a plethora of professional organizations to facilitate development and networking, such as the National Conference of Black Lawyers and the Hispanic National Bar Association. Similar coethnic organizations exist in business, such as Latino chambers of commerce, and education, such as alumni networks of historically black colleges and universities. These organizations connect relatively financially well-off individuals who share an ethnoracial identity and are thus ideal locations for candidates to solicit contributions from coethnic donors.

Finally, campaign finance institutions may facilitate greater coethnic behavior than do electoral institutions. Whereas an individual’s ability to vote for coethnic candidates is largely determined by his or her electoral district, he or she can choose whom to donate to among hundreds of candidates in any given cycle. This allows donors to support “surrogate” descriptive representation from coethnic politicians who do not represent their districts (Mansbridge Reference Mansbridge1999). It also provides the opportunity for campaigns to develop national networks of coethnic contributors. For example, Bill Carrick, advisor to Loretta Sánchez (D-CA), described plans to solicit contributions from Latino individuals and organizations across the country: “We’re certainly going to reach out to the Latino community in California and all across the country, as well as the groups, but we’re going to have to work really hard to raise the money, and it will have to come from sources all over the place” (quoted in Drusch Reference Drusch2015).

On the other hand, there are reasons to expect that donors are less likely to be influenced by candidate’s ethnoracial identity compared with other types of political actors. Outside party affiliation, candidate ethnorace is the most powerful and widely used heuristic for candidates’ ideological and policy positions (Banducci et al. Reference Banducci, Karp, Thrasher and Rallings2008; Manzano and Sanchez Reference Manzano and Sanchez2010), but visual cues such as a candidate’s ethnorace appear to be most influential among individuals of low education and information (Lenz and Lawson Reference Lenz and Lawson2011). Donors tend to be politically engaged and have high political knowledge and policy sophistication (Barber, Canes-Wrone, and Thrower Reference Barber, Canes-Wrone and Thrower2017), all of which reduce the importance of heuristics for political decision making. In addition, the number of ethnoracial minority donors may be so minimal that there is simply too little variation across elections to observe coethnic contribution behavior. We might therefore expect the contributor class to decide to contribute based on candidates’ ideological and policy positions in ways that are only weakly associated with candidates’ ethnoracial identities.

We are not only interested in coethnic contributing, but also its potential to create a more racially representative contributor class. The key outcome measure for this test is the proportion of contributions from whites in an election. Black candidates may increase black contributions, but decrease Latino and Asian contributions such that the overall proportion of contributions from donors of color remains unchanged.

In addition to contribution shares by ethnorace, we also examined contribution amounts. This analysis investigates the potential for a white donor backlash to candidates of color, which could occur in two different ways: white copartisan donors being demobilized by a candidate of color or white outpartisan donors being mobilized. Research shows that the presence of a black Democratic nominee is associated with decreased Democratic vote shares and a whiter overall electorate in just such a dynamic (Washington Reference Washington2006). White backlash behaviors have also been associated with the presidency of Barack Obama (Tesler Reference Tesler2012, Reference Tesler2016). In our study, increased funds from coethnic donors may be offset by decreased white contributions to the candidate or an increase in white contributions to opponents of minority candidates. Minority candidates may receive greater contributions from donors of color, but this could come at the fundraising competitiveness.^{Footnote 7} Correspondingly, we estimated the effect of a nominee’s ethnorace on the amount of contributions the nominee receives by donor ethnorace, and separately, the amount of contributions to opponent of the nominee by donor ethnorace. We also tested whether, on average, ethnoracial minority candidates raise funds as competitively as their opponents.^{Footnote 8}

To our knowledge, Cho’s (Reference Cho2001, Reference Cho2002) studies of Asian American contributors are the only prior analyses of the association between candidate and donor ethnorace. Cho (Reference Cho2001) found that Asian American candidates receive greater amounts from Asian American contributors. However, Cho (Reference Cho2001) focused on the distinctions between contributions based on shared pan-ethnic identity and those based shared in national origin. Our current study focuses on pan-ethnic identities and, in contrast to prior research, identifies the causal effect of candidate ethnorace on the ethnoracial distribution of contributions and white contributions to the electoral opponents of candidates of color.

In this study, we use pan-ethnic conceptualizations of Asian, black, and Latino identity. Race and ethnicity are social constructs, and there are differences in the distribution of social capital and economic resources across these groups, as well as differences in the “racial formation” of Asian, black, and Latino ethnoracial identity in the United States (e.g., Omi and Winant Reference Omi and Winant2014). These distinctions may generate differences in contributing. In Section A.1.1 in the Online Appendix, we apply the “ethnic-candidate paradigm” to Asian, black, and Latino Americans to generate expectations about intergroup differences in coethnic contributing. In short, Asian Americans hold greater economic capital on average, which may increase the likelihood of coethnic contributing. However, pan-ethnic identity may also lead to greater coethnic contribution patterns. Pan-ethnic identity is more crystallized among Latino and especially among black Americans (e.g., Junn and Masuoka Reference Junn and Masuoka2008; Mora Reference Mora2014; Waters Reference Waters1994).

Obtaining the Race and Ethnicity of Contributors

We use an increasingly popular strategy to obtain the racial identities of individuals, extrapolation from the geographic distribution of names and ethnoracial groups in the US Census (e.g., Barreto, Segura, and Woods Reference Barreto, Segura and Woods2004; Henderson, Sekhon, and Titiunik Reference Henderson, Sekhon and Titiunik2016). In short, the method uses Bayes’ Rule to calculate the probability that an individual identifies as African American, Asian, Latino, or white conditional on his or her name, based on the Census Bureau’s Surname List.^{Footnote 9} We then assign each individual the ethnoracial category with the greatest posterior probability.^{Footnote 10} We implement this procedure with the wru package in R (Imai and Khanna Reference Imai and Khanna2016).

The measurement strategy assumes that surnames are independent of geographic location within racial groups.^{Footnote 11} Although there are plausible ways in which this assumption could be violated,^{Footnote 12} Imai and Khanna (Reference Imai and Khanna2016) validated the method by predicting voters’ self-reported race with high precision and demonstrating minimal association between geography and surname after conditioning on race (see also Fiscella and Fremont Reference Fiscella and Fremont2006).

Our estimates of individuals’ racial identity are quite precise. Figure 1 shows the probability of each individual i’s race estimate R _i, conditional on name S _i and geography G _i, or Pr(R _i|S _i, G _i). The median probability is 0.887. Note that 1 − Pr(R _i|S _i, G _i) is the probability that an individual is of any other ethnoracial identity (i.e., an individual with a 0.6 probability of being Asian American also has a 0.4 probability of being non-Asian). Noise in our measurement of contributor ethnorace may produce attenuation bias, such that our analyses underestimate coethnic contributions (e.g., Gustafson Reference Gustafson2003), but we do not expect sources of bias that would inflate estimates of coethnic contributions. The legacy of slavery, where slaves were often given the surnames of slaveholders (Inscoe Reference Inscoe1983), makes ethnoracial identity estimates less precise for African Americans (see Online Appendix Figure B.1). This greater error may produce downward bias in our estimates of black coethnic contributing compared with other ethnoracial groups.

FIGURE 1.

Precision of Race and Ethnicity Estimates

Note: Plot shows the probability that each contributor’s racial identity is correctly coded, conditional on name and geographic location.

Our main analyses use data from US House primary and general elections between 1980 and 2012. For candidate ethnorace, we use a variety of data sources. We obtain data on the ethnorace of members of the US House from membership in the Congressional Black Caucus and Congressional Hispanic Caucus. Data on general election losers for the US House during the 2008–12 period are from Goggin (Reference Goggin2017), who coded candidate ethnorace from statements of self-identification, other publicly available and verified campaign sources, and, if no other information sources were available, wru estimates validated with candidate photos. Finally, our paid research assistants replicated the Goggin (Reference Goggin2017) coding strategy for all House candidates from 1980 to 2008 and for primary election losers from 2008 to 2012.^{Footnote 13}

Our measures of contributions come from the Dataset on Ideology and Money in Elections (DIME) from Bonica (Reference Bonica2013), which compiles data from the Federal Election Commission (FEC), the Sunlight Foundation, and the National Institute for Money in State Politics.^{Footnote 14} To construct our dataset, we aggregate individual contributions by candidate for the periods before and after the primary election. US House district demographic data are from the 1980, 1990, and 2000 US Censuses (ICPSR 8091, 8903), and the Census American Community Survey (ACS) for 2005 through 2012.

Difference-in-Difference Design

We mitigate the potential for confounding by implementing a difference-in-difference design at the district level. The difference-in-difference models take the following form, for district i in year t. Y _it represents the share or log amount of donations from individuals of ethnoracial group r, with separate models for each ethnoracial group of donors r ∈ {Asian, black, Latino, white}:

(1)$${Y_{it}} = \,{\alpha _i} + {\delta _t} + {\beta _1}Asian.Candidat{e_{it}} + {\beta _2}Black.Candidat{e_{it}} + {\beta _3}Latino.Candidat{e_{it}} + \beta {X_{it}} + {e_{it}}.$$

In the equation above, Y _it represents the share or log amount of general election individual contributions that come from individuals of ethnorace r. Asian.Candidate _it, Black.Candidate _it, and Latino.Candidate _it are dummy variables for the presence of a nominee of the ethnorace in the general election. District and year fixed effects are represented by α _i and δ _t,^{Footnote 15} respectively, which eliminate time-invariant confounders across districts. X _it is a vector of controls for district characteristics, which include district ethnoracial demographics, as well as measures of the proportion of residents in the district who are over age 65, employed in “blue collar” occupations,^{Footnote 16} employed in farming, employed by the federal government, active military, veterans of the military, unemployed, union members, and urban. β ₁ through β ₃, the quantities of interest, are the within-district effects of a candidate’s ethnorace on the share or amount of contributions from donors of ethnorace r in the district-year general election it.

Regression Discontinuity Design

Although the identifying assumptions for the difference-in-difference design are plausible, endogeneity concerns remain. The presence of a candidate with ethnoracial identity r may be endogenous to the existence and enthusiasm of coethnic contributors—potentially biasing estimates of the causal relationship between candidate ethnorace and the composition of the contributor class. We thus use an RDD to identify the causal effect of candidate ethnorace by exploiting the “as if random” assignment of primary candidates to general elections in close primary elections (Hall Reference Hall2015). In particular, the quantity of interest is the effect of the presence of a general election candidate of ethnoracial identity r on the share or amount of total general election contributions from donors of ethnoracial identity r in US House general election it. This local average treatment effect (LATE) is the difference in contributions relative to the counterfactual general election candidate of a different ethnorace (in the Appendix we report additional RDD results of subsetting to only elections in which the candidate of ethnorace r faced a white opponent). In the following model, the LATE is represented by β ₁, with separate models for each racial group r ∈ {Asian, black, Latino, white}:

(2)$${Y_{it}} = \alpha + {\beta _1}Cand.of.Race.r.Primary.Wi{n_{it}} + f\left( {{V_{it}}} \right) + {e_{it}}.$$

Cand.of.Race.r.Primary.Win _it is an indicator of whether a candidate of ethnorace r won against a second-place primary finisher of a different ethnorace in a major party primary election in district i in year t. Y _it represents the percentage or log amount of campaign funds from donors of ethnorace r in the general election in district i and year t. Substantively, then, β ₁ is the causal effect of the “as if random” assignment of a candidate of ethnorace r on the share of contributions from coethnic donors.

f(V _it) is a function of the forcing variable, the primary vote margin. We primarily use bandwidths from Imbens and Kalyanaraman (Reference Imbens and Kalyanaraman2012) (hereafter the IK bandwidth), Calonico, Cattaneo, and Titiunik (Reference Calonico, Cattaneo and Titiunik2014) (hereafter the CCT bandwidth), and local linear specifications of f, which we use to estimate the LATE of candidate ethnorace at the RDD cut point (Imbens and Lemieux Reference Imbens and Lemieux2008). We provide additional analysis in the Appendix using additional specifications of f. Table 1 provides further details on our difference-in-difference and RDD designs.

TABLE 1.

Data and Estimation Strategies

Note: 53 districts in our data have at least one Asian primary candidate from 1982 to 2010; 156 have at least one black candidate; 107 have at least one Latino candidate. Difference-in-difference model specifications have 6,070 observations (1980 and 2012 are dropped because there is only one observation in these redistricting periods). For RDD analyses, N varies based on bandwidth specification. The N statistics for the RDD design displayed above are from specifications using the IK bandwidth. Additional specifications are provided in the Appendix.

This design relies on the assumption that potential outcomes are smooth across the RDD cut point, that is, that the race of close-winners and close-losers in the primary election are independent of the ethnoracial distribution general election contributions. This assumption may not hold for close House general elections (Caughey and Sekhon Reference Caughey and Sekhon2011), but it is likely to hold in primary elections and other contexts (e.g., Eggers et al. Reference Eggers, Fowler, Hainmueller, Hall and Snyder2015; Hall Reference Hall2015).

Although this assumption is not directly testable, we execute placebo tests of the effect of a narrow primary win on the contributions in the primary election. If a narrow primary victory by a candidate of ethnorace r is predicted by primary election fundraising, then the assignment of candidate ethnorace in the general election is unlikely to be “as if random” at the RDD cut point. Indeed, Figures A.6 and A.7 in the Online Appendix confirm that, by ethnorace and across all candidates of color, bare-winners and bare-losers of primary elections do not significantly differ in primary election fundraising. This increases our confidence that our RDD is not biased by systematic differences in candidate quality correlated with candidate ethnorace.

Candidate Ethnorace as a Treatment Assignment

We are confident that our research designs mitigate observed and unobserved confounders across districts (Henderson, Sekhon, and Titiunik Reference Henderson, Sekhon and Titiunik2016), such as the ethnoracial distribution of potential donors. We are less confident about potential candidate-level confounders. Politically relevant characteristics vary across ethnoracial groups at the population level such that, even within district or within a close primary election, potential nominees of different ethnoracial groups are likely to vary on additional unobserved dimensions. The appearance of coethnic contributing may occur without an “ethnic-candidate paradigm” if individuals contribute money based on these unobserved dimensions in ways that are only correlated with, not caused by shared ethnorace.^{Footnote 17}

This issue speaks to the challenge of operationalizing ethnorace in a constructivist framework (Sen and Wasow Reference Sen and Wasow2016), a challenge common to previous observational analyses of empowerment theory and the “ethnic-candidate paradigm” (e.g., Bobo and Gilliam Reference Bobo and Gilliam1990; Barreto Reference Barreto2007; Fraga Reference Fraga2016a, Reference Fraga2016b). The social constructs of race and ethnicity are made up of—and causally intertwined with—component parts, including characteristics of appearance as well as geography, social status, religion, and culture. These components contribute to and are affected by self-identification at the individual level (Davenport Reference Davenport2016; Thomas and Speight Reference Thomas and Speight1999) and the social construct at the macro level (Omi and Winant Reference Omi and Winant2014). Religion serves as a useful example. African Americans are more likely to attend Protestant churches than average, and coethnic contributing among African Americans could be driven by shared religious affiliation, not shared ethnorace. However, religion in the United States is highly racialized,^{Footnote 18} in that it shapes and is shaped by ethnoracial categorization. In such a context, it is difficult to effectively conceptualize religious identity absent ethnorace.

We are unable to definitively rule out “non-racial” mechanisms that could lead to patterns of coethnic contributions, such as shared social class, religion, culture, ideology, and other politically relevant characteristics. We are able, however, to test for systematic differences in candidate characteristics associated with the running variable in the RDD—the candidate of color’s primary vote margin. Figures A.8, A.9, and A.10 conduct these tests for sets of covariates on candidate ideology and election characteristics, career background, and religion, respectively. If contributors were choosing to donate to candidates based on ideology, for instance, we would expect to find significant differences in nominee ideology when the candidate of ethnorace r barely wins or barely loses the primary. Yet we find little systematic difference in nominee ideology, as measured by DW-NOMINATE, or other candidate characteristics such as prior military service, or for election characteristics, such as CQ race forecasts or the likelihood of running in an open seat general election. By contrast, we do find significant differences, as expected, in nominee religion. Overall, despite the difficulty of isolating ethnorace as a treatment assignment, this analysis mitigates our concern about confounding from prominently studied candidate and election characteristics.

Descriptive Results

Figure 4 plots average contributions by candidate and contributor ethnorace. Descriptively, we find strong evidence of coethnic contributing. Asian, black, and Latino candidates receive greater contributions from Asian, black, and Latino donors, respectively. Black and Latino candidates receive less money from white donors, but Asian American candidates receive equal amounts as white candidates. Black candidates appear to get a much smaller increase of black donors, but as described earlier, we are less confident in our identification of black donors in the data.

FIGURE 4.

Average Contributions by Ethnorace

Note: Panels correspond to contributor ethnorace. The x-axis represents candidate ethnorace.

We also see in Figure 4 that candidate ethnorace is correlated with overall fundraising from individual donors. Asian American candidates receive the most, followed by Latino, white, and black candidates. However, we urge caution in drawing causal conclusions from this figure because the relationship is confounded by time, geography, and, more speculatively, fundraising from nonindividual sources. Campaigns grew more expensive in recent decades as increasing numbers of Asian and Latino candidates ran for office. Campaigns in the US South, which have greater numbers of black candidates, tended to be less expensive than those in other regions during this period.^{Footnote 20} Finally, our analysis focuses only on individual contributions; nonindividual contributions, such as PACs and party organizations, may substitute for individual contributions.

In the aggregate, we observe that candidates count on coethnic contributors for large amounts of money. We find similar results with descriptive regression models (bivariate and controlling for district demographics) in Figure A.2 in the Online Appendix. Importantly, these descriptive regressions show that the association between candidate ethnorace and the ethnoracial distribution of contributions is only minimally affected by the inclusion of district demographic covariates. In the following section, we examine whether this relationship between candidate and contributor ethnorace is causal.

Difference-in-Difference Results

Recall that the difference-in-difference design exploits variation within districts across time. Controlling for district racial demographics, we estimate the effect of nominating a Democratic candidate of ethnorace r on contributions from donors of ethnorace r in a given House district-election cycle, relative to a counterfactual nominee of a different ethnorace.^{Footnote 21}

Our difference-in-difference analysis first shows that the nomination of ethnoracial minority candidates leads to a more diverse set of donors. Figure 5 shows the effect of the presence of a general election candidate of ethnorace r on the share of contributions from individuals of the various ethnoracial identities.

FIGURE 5.

Effect of Candidate Ethnorace on Share of Contributions by Ethnorace

Note: The presence of a candidate of ethnorace r increases the proportion of general election contributions from individuals of ethnorace r in a district-year election. The omitted category is white candidate ethnorace. Models include district and year fixed effects. Estimates shown in black also control for district ethnoracial demographics. Error bars represent 95% confidence intervals. Robust standard errors are clustered by district.

The nomination of an Asian American or a Latino candidate increases the district-election’s share of contributions from Asian and Latino donors by about 10 percentage points, respectively. Black nominees increase the share of black contributions by about three percentage points.^{Footnote 22}

We find that the increase in coethnic contribution shares from Asian, black, or Latino donors substitutes for white contribution shares. These results suggest that the nomination of any nonwhite Democrat produces a less white—and, given population demographics, a more representative—contributor class.^{Footnote 23}

We find no difference in fundraising competitiveness between white and ethnoracial minority candidates. Fundraising gaps between candidates and their electoral opponents are equivalent for white and minority candidates. Candidates of color are equally likely to achieve fundraising parity with their opponents (see Appendix Table A1). Democrats of color are slightly more competitive in their fundraising against opponents, whereas Republicans of color are slightly less competitive. Latino Democrats raise significantly more against opponents than white Democrats, but otherwise these differences are statistically indistinguishable.

Figure 6 reports the specific trends in fundraising competitiveness by candidate ethnorace. Panel (a) reports that Asian and Latino Democratic nominees increase the amount of funds from coethnic donors by more than fivefold. Although Table A.1 shows that ethnoracial minority Democrats are no less competitive against opponents in overall fundraising, we see in Figure 6 that black and Latino Democrats receive lower total amounts from white donors. This reduction of white contributions is offset by increased coethnic contributions, although not fully for black candidates.

FIGURE 6.

Effect of Democratic Candidate Ethnorace on Log Total of Contributions by Ethnorace

Panel (a): The nomination of a Democratic candidate of (nonwhite) ethnorace r increases the amount of contributions to the Democratic nominee by donors of ethnorace r but does not significantly decrease the amount from white donors. Panel (b): The nomination of a Democratic candidate of ethnorace r may decrease Republican contributions from donors of ethnorace r , but it has no effect on white contributions to the Republican opponent. Models include district and year fixed effects. Estimates shown in black also control for district ethnoracial demographics. Error bars represent 95% confidence intervals. Robust standard errors are clustered by district.

Moreover, in Panel (b), which shows the effect of Democratic candidate ethnorace on contributions to the Republican opponent, we find that nonwhite Democrats lead to no increase of white contributions to the Republican opponent. Relative to white Democratic nominees, ethnoracial minority Democrats often decrease the amount of contributions going to the Republican opponent from the various groups of donors. Interestingly, the presence of Asian and Latino Democrats appears to modestly increase Latino contributions to the Republican opponent (although this increase is significantly smaller than the increase for Latino Democrats). Taken together, Figure 6 and Table A.1 suggest that there are minimal to no fundraising costs for nominating Democratic ethnoracial minority candidates. This is in contrast to Washington (Reference Washington2006), who finds that white voters are less likely to vote for a black Democratic candidates and more likely to vote against her, leading to lower Democratic vote share and a whiter electorate.

As reported in the Online Appendix (Table A.1 and Figure A.5), the results are mostly consistent with respect to Republican candidates. Relative to white Republicans, Republican candidates of color receive marginally lower fundraising totals compared with their Democratic opponents. Black and Latino Republicans see reductions in white contributions compared with white Republicans, but see little difference in white contributions to their Democratic opponents. Unexpectedly, however, the nomination of Latino Republican candidates is also associated with decreased black contributions, and the nomination of black Republicans with decreased Asian contributions.

Regression Discontinuity Results

Figure 7 plots the RDD results for the effect of nominating an Asian, black, or Latino candidate (relative to a counterfactual candidate of a different ethnorace) on contributions. The x-axis shows the running variable, the primary election vote margin between the ethnoracial minority candidate and his or her primary opponent. Observations on the right side of the cut point receive the “treatment” of an ethnoracial minority nominee. The plots on the left show contributions made to Democratic candidates, whereas the plots on the right show contributions made to Republicans.

FIGURE 7.

Regression Discontinuity Results

Note: Plots show LATE of nominee’s ethnorace on the proportion of contributions from coethnic donors. Left column is for Democrats; right column is for Republicans. Shading represents 95% confidence intervals. Some data points used in estimates are outside the plot y limits.

The RDD estimates are remarkably consistent with the difference-in-difference estimates presented earlier. For both Democrats and Republicans, the share of general election contributions from donors of ethnorace r is about 10 percentage-points higher when the nominee is also of ethnorace r.^{Footnote 24}

However, there are two LATE estimates that are notably distinct. The first is the estimate for black Democrats, who only modestly increase the share of contributions from black donors relative to a counterfactual non-black Democratic nominee. The second is the estimate for Asian Republicans, who see a massive (approximately 45 percentage-point) increase in Asian contribution shares compared with a counterfactual non-Asian Republican nominee.

As we described earlier, we check the robustness of the RDD findings with multiple specifications of the function of the forcing variable, the primary election vote margin. The Appendix reports additional LATE estimates using the CCT bandwidth, the IK bandwidth, and local linear models with 5 and 10% bandwidths. The coethnic contribution findings are quite consistent across specifications. The nomination of candidates of color causes an increase in contributions from donors of color.