Introduction

Why are some candidates perceived as more electable? While primary voters value candidate electability (Abramowitz, 1989; Rickershauser & Aldrich, 2007; Simas, 2017), how voters evaluate candidate traits when considering candidate electability is unstudied. Although scholars have examined the relationship between candidate traits and actual general election success (e.g., Hall, 2015; Terkildsen, 1993; Utych, 2020), whether voters perceive electability in the same way is unclear.

Despite extensive research on how various factors—most notably race (e.g., Sigelman et al., 1995; Terkildsen, 1993; Visalvanich, 2017b), gender (e.g., Dolan, 2004; Hershey, 1980; Sigelman & Sigelman, 1982), and ideology (e.g., Hall, 2015; Utych, 2020)—affect candidate vote shares in a general election (i.e. actual electability), we know less about what candidate characteristics voters believe make a candidate more electable and how those perceptions might vary across individuals.Footnote 1

To answer this question, we use observational and experimental evidence. Using polling data from 2019 and 2020 about perceptions about the electability of Democratic presidential candidates to motivate our work, we find evidence respondents view candidates whose ideology aligns with their own as more electable. However, observational data from the 2020 Democratic presidential primary field limits our ability to isolate the effect of certain candidate characteristics (e.g., ideology, race, and gender) on electability from other characteristics (e.g., fundraising or media attention).Footnote 2 This data also limits our ability to examine Republicans behaviors, which may also be different (Grossman & Hopkins, 2016).

As such, we use a conjoint survey experiment from a nationally representative sample to evaluate the importance of candidate characteristics on perceptions of electability. We ask respondents presented with a hypothetical congressional primary election match-up to indicate which candidate is more electable.Footnote 3 By randomizing candidate characteristics of interest (specifically, candidate race, gender, and ideology), we isolate their effects on voters’ perceptions of electability.

We find voters view minority and female candidates as less electable in the general election. The effect of candidate demographics, however, is much smaller than other factors such as candidate experience and candidate ideology. We find candidates who are ideologically extreme are perceived by voters to be more electable, a finding which runs counter to conventional wisdom (Masket & Noel, 2012) and actual general election success (Hall, 2015; Utych, 2020). However, these effects are driven by Republicans. While Republicans perceive candidate electability to increase with ideological extremism, Democrats perceive a slight general election electability penalty for ideologically extreme candidates.

However, across both parties, individuals view candidates who have ideological preferences aligned with their own to be more electable. Ideologically extreme respondents in both parties are more likely to view more ideologically extreme candidates as electable and partisan differences in perceptions of the electability of moderate candidates are driven by the presence of more ideologically extreme respondents among Republicans.

Similarly, although voters view minority candidates as slightly less electable, these effects are driven by white respondents’ perceptions. Minority respondents do not perceive differences in electability by candidate race. However, women are just as likely as men to perceive electoral penalties for female candidates.

Finally, in our survey experiment we find a strong relationship between voters’ perceptions of electability and vote choice. There are, however, some exceptions. While Democratic respondents viewed non-white candidates and female candidates as less electable than white candidates and male candidates, Democrats preferred non-white and female candidates. This result, however, is again driven by respondent race and gender, with women and minorities and more likely to support candidates of their own subgroup.

Actual Electability

While partisans want to nominate candidates who will advance their preferences in office, policy outcomes depend on their candidate winning office. As such, individuals may prefer to nominate candidates who are merely acceptable but who can win the general election. When given information about candidates’ general election chances, voters use that information (Abramowitz, 1989; Masket, 2020; Rickershauser & Aldrich, 2007; Simas, 2017; Stone & Abramowitz, 1983). What is less clear, however, is what types of candidates voters perceive as electable absent polling.Footnote 4

Research on actual (rather than perceived) electability has largely focused on candidate ideology. The median voter theorem predicts more ideologically centrist candidates should have greater appeal to the electorate (Downs, 1957). Observational research indicates ideologically extreme candidates suffer an electoral penalty in the general election (Hall, 2015; Utych, 2020), although the advantage appears muted in high-profile election campaigns (Cohen et al., 2016) and has declined in recent years in a more polarized environment (Canes-Wrone & Kistner, 2022; Utych, 2020).

Candidate traits also affect actual electability, with work indicating minority candidates (e.g., Sigelman et al., 1995; Terkildsen, 1993; Visalvanich, 2017b) and women (e.g., Dolan, 2004; Hershey, 1980; Karpowitz et al., Forthcoming; Sigelman & Sigelman, 1982, but see Schwarz and Coppock (2022)) suffer general election penalties.Footnote 5

Voters’ Perceptions of Electability

It is unclear, however, whether these factors also affect voters’ perceptions of candidate electability. Research on voters perceptions of electability has largely focused on single candidate traits (Sigelman et al. (1986); Williams (1990), but for an exception see Green et al. (2022).Footnote 6

Furthermore, scholarship has ignored the possibility perceptions of electability vary by voter characteristics.Footnote 7 Focusing on vote choice, voters’ evaluations of candidates, including evaluations of same party candidates, are linked to their own predispositions and grow over the campaign (Bartels, 1988) and party leaders view their own ideological preferences as more electable (Broockman et al., 2021).

In addition, perceptions of electability might vary by party. Republicans care more about ideological purity (Grossman & Hopkins, 2016). As such, we might expect penalties to perceptions of electability for ideological extremism to be smaller among Republicans. Similarly, work on political elites indicates Democrats are more supportive of minorities and women (Crowder-Meyer & Cooperman, 2018; Doherty et al., 2019; Hassell & Visalvanich, 2019) suggesting race and gender penalties on perceptions of electability could be smaller for Democrats.

Lastly, voter assessments of candidate electability, while influencing vote choice, are distinct from vote choice. Voters may compromise if it will increase the likelihood of success in the general election (Mirhosseini, 2015; Simas, 2017) but, some individuals with strong preferences (either ideological or descriptive) may view compromise as undesirable even if it reduces the likelihood of winning slightly (Bawn et al., 2019; Blum, 2020; Crowder-Meyer & Cooperman, 2018). As such, we also examine whether voters support candidates with attributes perceived as less electable.

Real World Evidence of Perceptions of Electability

We begin by looking for evidence of relationships between voter characteristics and perceptions of candidate electability in real campaign situations. The large and diverse 2020 Democratic presidential field provides an opportunity to see whether perceptions of electability vary systematically with individual respondent characteristics. In 2019 and in the early months of 2020, ABC and the Washington Post sponsored four nationally representative surveys where they asked Democrats and Democratic leaners (n = 1751 for all four surveys) to indicate which of the Democratic candidates for president “has the best chance to defeat Donald Trump in the general election?”Footnote 8 We use those surveys to examine how respondent characteristics affect the likelihood respondents chose candidates with similar characteristics as the most electable.Footnote 9

Figure 1 shows the marginal effect of sharing a trait on the likelihood of identifying a candidate with that trait as the candidate with “the best chance to defeat Donald Trump in the general election” using a linear probability model. In addition to variables for characteristics respondents shared with candidates (race, gender, and ideology), we also control for respondent income, education, age, and the specific survey. Full models are available in online appendix Table A1.

Fig. 1
figure 1

In-group perceptions of ‘most electable’ 2020 democratic candidate effect of in-group respondent (relative to outgroup candidate) identifying in-group candidate as the candidate most likely to defeat the Donald Trump in the 2020 presidential election. Full results including controls are available in online appendix Table A1

Full size image

Panel (a) in Fig. 1 shows the relationship between sharing a political ideology and identifying a candidate with a similar ideology as the most likely to defeat Donald Trump. The first estimate in Panel (a) shows the effect of being moderate (relative to identifying as very liberal) on the likelihood of choosing a moderate candidate as the most likely to defeat Donald Trump. The second (right hand) point estimate in Panel (a) shows the effect of identifying as very liberal (relative to being moderate) on the likelihood of selecting a very liberal candidate as the most likely to defeat Donald Trump. Respondents were coded as having selected a liberal candidate if they identified Bernie Sanders or Elizabeth Warren as the candidate most likely to beat Donald Trump and coded as having selected a moderate candidate if they chose Joe Biden, Steve Bullock, Pete Buttigieg, or Amy Klobuchar.

As shown in Panel (a) in Fig. 1, very liberal (moderate) voters were 17.6% (14.5%) points more likely to view one of the liberal (moderate) candidates as the most likely candidate to beat Donald Trump in the 2020 presidential election. Results for individual candidates are available in the online appendix in Table A2. In separating the results by candidate, we find the effect of support for a moderate candidate is driven by moderate voters’ perceptions of Joe Biden.Footnote 10 However, both Warren and Sanders were viewed as more electable by very liberal respondents.

Panel (b) shows the relationship between respondent race and selecting a candidate of a similar race as most likely to defeat Trump. Respondents were coded as having selected a Black candidate if they selected Cory Booker or Kamala Harris. Respondents selecting Julian Castro were classified as having chosen a Latino candidate. Respondents selecting any of those three candidates as the most likely to defeat Trump in the general election were identified as having selected a minority candidate.Footnote 11 Because all minority candidates had withdrawn from the race by the third poll, this analysis only includes the first two polls taken in 2019 (n = 897).

Latino respondents were 1.3 percentage points more likely to indicate the Latino candidate, Julian Castro, was the candidate with the best chance of defeating Donald Trump in the general election compared to non-Latino respondents (p < 0.1, two-tailed test).Footnote 12 We do not, however, find any evidence Black respondents were more likely to indicate one of the African-American candidates was the most electable, or minority respondents in general were more likely to select a minority candidate as the most likely to defeat the incumbent president. Moreover, as shown in the appendix, Black respondents were much more likely to view Joe Biden as the most electable, likely because of his role as the Vice President to Barack Obama. This emphasizes the difficulty in identifying causal relationships between individual candidate characteristics in a dynamic campaign environment and when candidate characteristics are not randomized.

Panel (c) shows the relationship between respondent gender and selecting a female candidate (Tulsi Gabbard, Kamala Harris, Amy Klobuchar, Elizabeth Warren, Kirsten Gillibrand, or Marianne Williamson) as the most electable candidate. We find no evidence female candidates were more likely to be viewed as the most electable candidate in the 2020 Democratic primary field by female respondents.Footnote 13

We note, however, that these effects do not capture how the dynamics of the campaign affect perceptions of electability. While we focus here on the association between individual characteristics and the propensity to identify similar candidates as the most electable candidate, these propensities are shaped by the nature of the campaign (Bartels, 1988). Indeed, we should note the overall perceptions of electability for candidates of different types change over the course of the campaign. For example, the perception of a liberal (moderate) candidate as most electable ranges from 21 to 27% (29–47%) and perceptions of minority or female candidates as the most electable range from 2 to 12% and 8 to 24% respectively depending on the survey.Footnote 14

In short, it is impossible to draw definitive conclusions about how shared characteristics affect perceptions of electability using this data because candidate characteristics are strongly related to other characteristics that may also influence perceptions of electability (e.g., candidate characteristics affect support from party leaders (Doherty et al., 2019; Butler & Preece, 2016), fundraising dynamics (Crowder-Meyer & Cooperman, 2018), and media coverage (Kahn, 1994)). Moreover, candidates recognize their strengths and weaknesses and design campaigns to mitigate weaknesses and emphasize strengths. While these results suggest the possibility perceptions of electability may vary systematically by respondent characteristics, it lacks the internal validity necessary to permit clear identification of relationships between candidate characteristics and voters’ perceptions of electability.

Experimental Data

Given our primary interest in identifying these relationships and the limitations of our real world observational data outlined above, we turn to a survey experiment implemented as part of a nationally representative survey (n = 2051) collected through Lucid Marketplace in August of 2020.Footnote 15Footnote 16 After completing demographic questions, including questions related to a respondent’s racial and ethnic identity, age, household income, partisan identification, and political ideology (see Table 1),Footnote 17 respondents completed a series of forced choice experiments asking them to evaluate two candidates in a hypothetical primary.Footnote 18 Respondents were shown a primary matchup between two candidates corresponding to their partisan identification (leaners included) and individuals identifying as pure independents were randomly assigned to partisan treatments.Footnote 19 Upon viewing the profiles, respondents were asked which candidate would be more likely to win in a general election and which candidate they would vote for in the primary. Respondents evaluated eight candidate pairings resulting in 16,408 choice tasks.Footnote 20

Table 1 Sample demographics
Full size table

Table 2 shows the universe of candidate attributes with the baseline category in parentheses.Footnote 21 Our primary interest is candidate race, gender, and ideology.Footnote 22 For ideology, we vary treatments by party. For Democratic candidates, ideological values include ‘Moderate,’ ‘Liberal,’ and ‘Extremely Liberal.’ For Republican candidates, ideological values include ‘Moderate,’ ‘Conservative,’ and ‘Extremely Conservative.’Footnote 23

Table 2 Treatments (baseline in brackets)
Full size table

Finally, we also include a number of other candidate characteristics to provide a more rounded perspective of the potential candidates. This additional information “should reduce social desirability biases by providing alternative justifications for sensitive choices” (Hainmueller et al., 2014, p. 19). We provided information about two aspects of candidate charisma—candidate speaking ability and the personableness of the candidate—and information about age and previous office-holding experience.Footnote 24

Conjoint Experiment Results

We estimate average marginal component effects (AMCEs) using a linear model (Hainmueller et al., 2014), clustering standard errors by respondent. The AMCE estimates the average extent to which a particular set of characteristics affects the probability a respondent will select a particular candidate. We rely on advice from Bansak et al. (2022) and treat AMCEs as the effect of changing an attribute from its baseline level on the probability of viewing that candidate as more electable.Footnote 25

Overall Perceptions of Electability

Before looking at whether perceptions of electability vary systematically by respondent characteristics, we first examine overall trends. Figure 2 presents the effects of candidate characteristics on perceptions of candidate electability for all respondents and for Democratic and Republican respondents separately. We focus on race, gender, and ideology but we also include the effects of candidate experience because it is an easily and clearly quantifiable characteristic previous work has highlighted as affecting actual general election success (Jacobson & Kernell, 1981, but see Porter & Treul, 2018), thus providing a reasonable gauge of the effects. We do note, however, the effects of speaking ability and personableness are larger than the effects we discuss here. Full models with all variables and models for only individuals who voted in the primary are in the online appendix.

Fig. 2
figure 2

Perceptions of candidate electability. Effect of candidate characteristics on the likelihood a respondent chose candidate as most electable. Bars are 95% confidence intervals. Full results are available in online appendix Table A3 and Fig. A1

Full size image

Because preferences can diverge in the baseline level across groups, we exercise some caution in interpreting differences-in-AMCEs across party as differences in preferences because AMCE differences could be affected by each party’s underlying preferences (Leeper et al., 2020). As such, differences in AMCEs do not necessarily represent overall preference differences across Republicans and Democrats.

Race and Gender

As shown in Fig. 2, compared to white candidates, we find Black and Hispanic candidates were 6% less likely to be viewed as electable and we find similar effect sizes for both Democrats and Republicans. While effect sizes would be classified as small using Cohen’s (1969) rule of thumb and recent standards outlined by Hartman and Hidalgo (2018) using equivalence testing, they are, at the same time, similar in substance to the effect of candidate experience (Jacobson & Kernell, 1981).

Similarly, the results in Fig. 2 show voters view female candidates as less electable. Women are about 3% less likely to be viewed as electable compared to men and these effects are similar for both Democrats and Republicans. On the whole, these findings confirm previously conjectures perceptions of electability are at least influenced by candidate race and gender (Green et al., 2022; Masket & Noel, 2019).

Ideology

We do not, however, find a positive relationship between candidate ideological moderation and perceptions of electability among the full sample. Candidates described as moderate were 3% less likely to be viewed as electable relative to ideological candidates (conservative for Republicans and liberal for Democrats). There is also no perceived electability penalty for more ideologically extreme candidates. Candidates described as very liberal (very conservative) were just as likely be viewed as electable relative to candidates who were described as liberal (conservative) and roughly 3% more likely to be viewed as electable relative to moderate candidates. While not large effects, they are in the opposite direction of effects of ideological moderation on actual vote share (Hall, 2015; Utych, 2020).

This is one area where there are distinct differences by party. For Republicans, as shown in Fig. 2, moderate candidates are viewed as less electable. There is, however, no similar penalty for moderate candidates (relative to more ideologically extreme candidates) among Democrats, and candidates classified as very liberal were viewed by Democratic respondents as slightly less electable (p < 0.10).

Individual Attributes Affect Electability Perceptions

We move now to the question of whether perceptions of electability vary systematically with respondent characteristics. We test for this possibility by re-estimating the AMCEs while interacting respondent characteristics (specifically race, gender, and ideology) with candidate characteristics. We present the full set of coefficients from the models in the online appendix in Fig. A3 and Table A4. Below we focus specifically on how respondent race, gender, and ideology affect evaluations of the electability of candidates with similar and dissimilar characteristics. Before continuing, we first want to make a note about statistical power. Because each respondent completed eight tasks (evaluating 16 candidates), we have almost 8,000 observations from minority respondents who make up roughly a quarter of the sample, providing us sufficient power to detect effects as small as 3%.Footnote 26

Race and Gender

While overall Black and Hispanic candidates are less likely to be considered electable, Fig. 3 indicates these effects are driven by white voters. Figure 3 plots the differences in the likelihood of choosing a minority candidate as more electable by the minority status of the respondent. While white respondents consistently view minority candidates as less electable than white candidates, non-white voters do not, an effect consistent across parties (see online appendix Fig. A3).Footnote 27

Fig. 3
figure 3

Candidate electability by candidate and respondent race. Effect of candidate race on the likelihood a respondent chose candidate as most electable by respondent race. Bars are 95% confidence intervals. Predicted likelihood derived from the results from online appendix Table A4

Full size image

We do not, however, see any differences in how male and female respondents evaluate female candidates (consistent with the observational results presented previously). Instead, both men and women view women as less electable (see online appendix Fig. A5).

Ideology

Figure 4 shows even larger differences in evaluations of electability of moderate and ideologically extreme candidates by the ideology of the respondent. Ideologically extreme (moderate) respondents are roughly 11 (7) percentage points more likely to believe ideologically extreme (moderate) candidates are more electable than moderates (extremists), effects comparable to or larger than those of candidate experience.

Fig. 4
figure 4

Candidate electability by candidate and respondent ideology. Effect of candidate ideology on the likelihood a respondent chose candidate as most electable by respondent ideology. Bars are 95% confidence intervals. Predicted likelihood derived from the results from online appendix Table A4

Full size image

However, we do not find, as shown in Fig. 5, any substantive differences in the patterns across parties. Both Republican and Democratic respondents view candidates with similar ideologies to their own as more electable (although the penalty for moderates among extremist respondents for Republicans is marginally larger). Overall partisan differences in perceptions of electability of ideologues (see Fig. 2) seem to be driven by the ideological makeup of the parties (with there being more (fewer) ideologically extreme Republicans (Democrats) and fewer (more) moderate Republicans (Democrats)).Footnote 28

Fig. 5
figure 5

Candidate electability by candidate and respondent ideology by party. Effect of candidate ideology on the likelihood a respondent chose a candidate as most electable by respondent ideology by party. Bars are 95% confidence intervals. Predicted likelihood derived from the results from online appendix Table A4

Full size image

Differences in Vote Choice and Electability

Lastly, we examine differences between perceptions of electability and vote choice to understand where individual preferences and perceptions of electability deviate sufficiently to where voters are willing to support less electable candidates. However, this test of the differences between who voters perceive as electable and who they support is a hard test. Respondent vote choice was asked immediately after the question about candidate electability. Thus, electability concerns would be foremost in respondents’ minds. We would expect differences between perceptions about electability and vote choice to be stronger if respondents were not primed about electability first.

Figure 6 shows effects of candidate characteristics on respondents’ vote preferences. As before, we present the results for race, gender, and ideology along with candidate experience (full results are available in online appendix Fig. A6 and Table A6).Footnote 29 Comparing Fig. 6 to Fig. 2, in most cases respondents’ vote choices are similar to perceptions of electability. There are, however, some differences which we explore below.

Fig. 6
figure 6

Effect of candidate characteristics on vote preference. Effect of candidate characteristics on the likelihood a respondent indicated the candidate would be preferred vote choice. Bars are 95% confidence intervals. Predicted likelihood derived from the results from online appendix Table A7

Full size image

Race

One difference between perceptions of electability and vote choice is the area of race. As noted in Fig. 2, minority candidates are perceived as less electable by about 6 percentage points for both Democrats and Republicans. In contrast, as shown in Fig. 6, among all voters, changing the racial identity of the candidate from white to Black had no effect on the likelihood a respondent would indicate they would vote for that candidate. We do find the penalty for changing the candidate’s race from white to Hispanic was still negative and significant (3%).

However, these effects again mask differences in two important ways. First, there are significant heterogeneities by respondent characteristics. Although minority respondents view white and minority candidates roughly as equally electable (as shown previously in Fig. 3), Fig. 7 shows minority respondents are more likely to indicate they would vote for a minority candidate than for a white candidate.

Fig. 7
figure 7

Vote preference by respondent race. Effect of candidate race on the likelihood a respondent indicated the candidate would be preferred vote choice by respondent race. Bars are 95% confidence intervals. Predicted likelihood derived from the results from online appendix Table A6

Full size image

These racial differences partially help explain the second heterogeneity: differences by party, as noted Fig. 6. Minorities make up almost 36% of Democrats in the sample compared to only 15% of Republicans in the sample. Although white Democrats are also slightly more likely to support minority candidates, minority respondents are much more likely to indicate they would vote for a minority candidate.Footnote 30 While both Democrats and Republicans view minority candidates as less electable, their willingness to vote for these candidates is fundamentally different. Whereas Democrats perceive Blacks and Hispanics as slightly less electable, they are, at the same time, actually more likely to vote for Black and Hispanic candidates. In contrast, while Republicans also perceive Black and Hispanic candidates as slightly less electable they are less likely to vote for those candidates prefer black and Hispanic candidates.

Previous work has shown perceptions of electability affect vote choice (Simas, 2017). Using our data we can examine the mediating effect of electability on vote choice and specifically whether voters would exhibit even stronger first-order preferences for racial minorities absent concerns about their electability following the procedures used by Green et al. (2022). In the online appendix, we use the Mediation package in Stata (Hicks & Tingley, 2011) which implements the procedures described by Imai et al. (2010) and Imai et al. (2011). In this case, we look at both the direct relationship between candidate race and vote choice and the indirect relationship between those two variable through perceptions of electability. Previous work has looked at this relationship but only for Democrats (Green et al., 2022).Footnote 31

Online appendix Table A11 shows the direct and indirect effects of candidate race on vote choice mediated by electability. The average direct effects of the candidate being Black (or Hispanic) is 2.4 (1.2) percentage points, indicating that, absent electability concerns, voters would prefer minority candidates by that margin. However, the average causal mediation effect is − 2.9 (− 3.2) percentage points, meaning electability concerns reduced overall support for Black candidates and were a net liability for Hispanic candidates.

Moreover, Tables A12 and A13 show these effects are not uniform by party. Whereas the average direct effect for Democrats is positive and significant (7.3 percentage points and 5.0 percentage points for Blacks and Hispanics, respectively), it is negative and significant for Republicans (− 2.6 and − 3.0 percentage points for Black and Hispanic candidates, respectively). The average causal mediation effect, however is negative for both.

Gender

Similarly, whereas women were perceived as less electable overall, we see respondents are slightly more likely to indicate they would vote for a female candidates overall. While not a large effect, changing the gender of a candidate from a man to a woman increases the likelihood of that candidate being chosen as the respondent’s vote choice preference by about 2%. As with race, however, these overall effects mask important heterogeneities by gender and party.

In Fig. 6, we find the effect of changing a candidate’s gender from male to female on the likelihood of being selected as a respondents’ primary vote choice to be positive for Democrats (5% increase), but negative (but not significant) for Republicans (1% decrease). Whereas both Democrats and Republicans view female candidates as slightly less electable, Democrats are slightly more likely to vote for a female candidate in the primary while for Republicans there is not statistically discernable differences in vote choice by candidate gender.Footnote 32

The differences by party in the choice to support female candidates despite concerns about electabilty is driven both by the demographics of the parties and by different patterns of preferences across parties. As shown in Fig. 8, women are slightly more likely to indicate they will vote for a female candidate (and women are slightly more likely to be Democrats). However, in online appendix Fig. A9, we show the relative preference for female candidates is stronger among female Democrats than for female Republicans.

Fig. 8
figure 8

Vote preference by respondent gender. Effect of candidate gender on the likelihood a respondent indicated the candidate would be preferred vote choice by respondent gender. Bars are 95% confidence intervals. Predicted likelihood derived from the results from online appendix Table A7

Full size image

Lastly, we can also look at the direct and indirect effects of candidate gender using the mediation analysis outlined above to look at how gender shaped concerns about electability affect candidate support. Online appendix Table A9 shows the overall direct and indirect relationships between candidate gender and vote choice through electability and Table A10 breaks out the result by party. Consistent with what we have presented previously (and with previous work (Green et al., 2022) who examine this relationship only among Democrats), we find both a direct (and positive) effect of candidate gender on vote choice and also an indirect (and negative) effect of those candidate characteristics on on vote choice mediated by electability. Combining these positive and negative effects, results in a small positive effect overall on vote choice. These effects, however, are not uniform across parties. As shown in online appendix Table A10, Democrats exhibit patterns consistent with what we describe above, albeit with larger direct effects. Republicans, however, show no significant positive direct effects, and small negative mediating effects for an overall effect that is negative (but not statistically significant).

Ideology

Lastly, previously we found moderate candidates were perceived as less electable, but that these differences were driven by more ideological voters regardless of party (see Figs. 4 and 5). In contrast to our findings for race and gender, we do not find any differences between the relationship between ideology and electability and ideology and vote choice, nor do we find any differences in the patterns across Democrats and Republicans (see Fig. A8 in the online appendix). While overall Democrats perceive very ideological candidates as less electable and are marginally less likely to vote for them and Republicans perceive very ideological candidates more electable and are more likely to vote for them, these effects are driven entirely by the percentage of Democrats and Republicans who identify as extreme ideologues. The underlying relationships between respondent ideological characteristics and their support for candidates of a particular ideological typology are not fundamentally different across parties (as shown in online appendix Fig. A8). Differences across parties are entirely due to differences in the numbers of very ideological voters within each party.

Online appendix Table A14 shows the direct and indirect effects of candidate ideology on vote choice mediated by electability and Tables A16 and A15 break those results out by party. Table A14 shows overall there is neither direct effects of candidate ideology (the average direct effect is − 0.7 percentage points) nor mediation effects (the average causal mediation effect is − 0.4 percentage points). However, as might be expected, these effects mask strong heterogeneities by respondent ideology with moderate respondents imposing both direct and indirect penalties and extreme respondents boosting their support of ideologically extreme candidates both directly and through electability.

Discussion and Conclusion

Previous work has provided an incomplete picture of how voters use candidate traits in evaluating electability and how those perceptions vary across individuals. Using data from real world campaigns and from a survey experiment, we find what voters perceive to be electable characteristics depends on those voters’ political preferences and characteristics.

Overall, we find individuals’ identity and preferences influence how they evaluate candidate electability. Using data from the 2020 Democratic presidential primary, we find evidence that moderate (liberal) Democratic respondents are more likely to identify a moderate (liberal) candidates as the most electable. However, identifying the nature of these effects is difficult because of our inability to decouple individual candidate characteristics from other characteristics (e.g., fundraising) that may also influence perceptions of electability or to control for campaign actions.

However, our primary purpose in this work is to identify clear causal relationships between candidate characteristics, something the messy nature of real campaigns makes difficult to do. As Rose McDerrmott notes, scholars knowing “the conclusions result from the manipulations imposed [can next try] to extrapolate those findings to other contexts” (McDermott, 2011, p. 43). To provide for better internal validity we rely on a conjoint survey experiment.

First, the conjoint experiment results provide no evidence voters systematically associate ideological moderation with electability. Instead, as with the real world data, we find evidence voters’ perceptions of the relationship between ideology and electability align closely with their own ideological preference.

We do find significant partisan differences in the relationship between candidate ideology and perceptions of electability, with Republican respondents seemingly finding more ideologically extreme candidates more electable and Democratic respondents finding more moderate candidates as more electable. While differences could be a function of the 2020 electoral environment where Republicans were re-nominating a president viewed as ideologically extreme (Hopkins & Noel, 2022), our results suggest these findings are a function of differences in the ideological composition of the parties (Grossman & Hopkins, 2016). Both Republicans and Democrats view candidates with ideological identities aligning with their own stated ideological preferences as more electable. Because there are more moderate Democrats in the Democratic Party than there are moderate Republicans in the Republican Party, Democrats are more disposed to view moderates as more electable while Republicans are more likely to perceive more extreme candidates as more electable.

Second, respondents, on the whole, perceived women and minority candidates to be less electable when compared to male and white candidates. We also do not find differences by party on the effect of race and gender on perceptions of electability. However, there are differences by race in perceptions of the electability of minorities, with minority respondents perceiving minority candidates as just as electable as white candidates while white respondents view minority candidates as less electable.

Finally, while the traits respondents found electable closely matched traits they favored when choosing to vote for a candidate, there were two notable exceptions. Democratic respondents saw non-white candidates and women candidates as less electable than white male candidates, but indicated they would support these candidates anyways. This finding aligns with recent research that has found a liberalization in racial attitudes among Democratic partisans who have come to value descriptive representation of minorities and women, a contributing to a trend of increased representation for these groups in the party in elected office (Crowder-Meyer & Cooperman, 2018; Hassell & Visalvanich, 2019) and align with discussions within the Democratic party that were ongoing during the 2020 election (Masket, 2020). However, these patterns are also a product of minorities and women (who are more likely to identify as Democrats) being more willing to vote for candidates like themselves despite electability concerns.

Despite the importance of our results which show the impact that candidate race, gender, and ideology have on perceptions of electability, it is important to note that our study does not exhaust the number of potential factors that might contribute to those perceptions. There are an endless multitude of candidate traits and campaign tactics (or interactions between traits and tactics) that likely have an influence on perceptions of candidate electability. While this study has focused on candidate qualities and their effect on electability, it is also true that how candidates appeal to different parts of their political coalitions (Grossman & Hopkins, 2016; Masket, 2020) and how those appeals interact with candidate characteristics (Karpowitz et al., 2020) could also factor into their broader appeal. Thus, while our efforts here have greatly expanded our understanding of electability, there remains much to be studied about primary elections and voters’ perceptions of electability.

In addition, while the results from our survey experiment that follow provide clear internal validity of the relationship between individual characteristics and voters’ perceptions of electability, the real world results also highlight the need for future work to better understand how campaign actions (and the actions of other political actors) extenuate or mitigate these relationships.

On the whole, however, our findings present a clearer picture of perceptions of how candidate electability is shaped by candidate race, gender, and ideology and the interaction of those characteristics with the characteristics of the individuals making those evaluations. Although years of scholarly research has provided clear evidence of the relationship between many candidate characteristics and actual electability, our findings show voters’ perceptions of how those components affect a candidate’s electability are connected to an individual’s own ideological preferences and identity.