Race, class, or both? Responses to candidate characteristics in Canada, the UK, and the US

ABSTRACT

Research suggests that voters use identity markers to infer information about candidates for office. Yet politicians have various markers that often point in conflicting directions, and it is unclear how citizens respond to competing signals – especially outside of a few highly stigmatized groups in the US. Given the relevance of these issues for electoral behavior and patterns of representation, this article examines the impact of intersectional identities and less intensely stigmatized markers in Canada, the UK, and the US. It does so using a survey experiment that varies the race (white/East Asian) and class background (higher/lower) of a candidate for office. I then compare results across the cases, examining willingness to vote for the candidate as well as assumptions about his ideological proximity, relatability, and potential contributions. In doing so, I build from past research suggesting that voter ideology likely shapes reactions to candidates from disadvantaged backgrounds. Results suggest that marginalized identity markers have relatively widespread effects among leftists and (to a lesser extent) centrists, but that, outside of the Canadian left, class seems to matter more than race. Overlapping marginalized identities, in turn, had little impact, with the lower-class white and East-Asian profiles eliciting similar reactions.

KEYWORDS:

• Candidate assessments
• race
• class
• survey experiment
• Canada
• United Kingdom
• United States

Voters use a variety of shortcuts to infer information about candidates for office (e.g. Popkin 1994; Bartels 1996). One such set of heuristics is centered around identity markers, such as race and class, that appear to be relevant considerations for many citizens (e.g. Citrin, Green, and Sears 1990; Carnes and Lupu 2016). By shaping presumed policy stances, perceptions of relatability, and/or assumptions about potential contributions, candidate markers may thus have important knock-on effects: trait-based influences on vote choice are liable to impact both the descriptive representation of marginalized groups and the prominence of their concerns in the political arena (e.g. Cowley 2013; Carnes and Sadin 2014; Jacobson and Carson 2015; O’Grady 2019). Yet candidates are not defined by a single marker and, most of the time, not every facet of a candidate’s identity will send the same set of signals. So how do citizens react in the face of conflicting markers?

When presented, for example, with equivalent working class white and upper-middle class racialized candidates, voters may well give pride of place to class or race. Similarly, they may or may not give special weight to intersecting marginalized identities (Crenshaw 1989) – perhaps viewing a racialized working-class candidate as more progressive or relatable than either a lower-class white candidate or a higher-class racialized one. Most existing research, however, does not allow us to assess the interactive influence of competing and complementary identity markers. What is more, even our understanding of the effects of individual markers is limited to a narrow subset of cases: past studies have concentrated on perceptions of African-American and female candidates, and have been disproportionately focused on the US (e.g. Jacobsmeier 2015; Lerman and Sadin 2016).

Each of these limitations is problematic. First, the effects of candidate identity markers may well be multiplicative rather than additive. As Philpot and Walton (2007, 49) argued in one of the few studies to consider the potential importance of intersectionality, “voters do not necessarily use one identity at the expense of the other when making political decisions. Rather, multiple identities can interact to create a separate single identity that can be used to evaluate candidates.” Second, while there are good reasons to be interested in how citizens react to candidates from particularly stigmatized groups, this focus leaves us with only a partial picture of identity-based effects: we know surprisingly little about reactions to less negatively stereotyped racialized groups, such as East Asians (see Visalvanich 2017a); and scholars have only recently begun to study the potential impact of candidates’ class backgrounds (e.g. Carnes and Sadin 2014; Carnes and Lupu 2016). Third, there are good reasons to think that past findings in the American context may not reflect dynamics elsewhere. To the extent that voters use class and race as cues, the effects of identity markers (and their interactions) are likely shaped by the salience and meaning of those markers within the broader political context (e.g. Coffé and Theiss-Morse 2016; Konitzer et al. 2019).

Given the relevance of these issues for electoral behavior and patterns of representation, this article uses a survey experiment fielded in Canada, the UK, and the US to study (1) the impact of intersectional identities, (2) the relevance of markers from less intensely stigmatized social groups, and (3) the uniformity of these effects cross-nationally. Although there are countless parameters of disadvantage that one could explore (e.g. gender, sexuality, disability, etc.), this first cut at the issue focuses on the intersection of candidate class and racial markers. The choice of countries allows us to compare results from the (highly-studied) American context to those in two cases where class and race are likely to have divergent levels of salience: on the one hand, social class has tended to play a much more prominent political role in the UK (e.g. Lipset 1983; Bélanger and Eagles 2006; Evans and Tilley 2012); on the other, Canada’s long history of multiculturalism has helped to generate a particularly acute focus on ethnic inclusion and diversity, including on the political right (Winter 2007; Harell 2009; Bird, Saalfeld, and Wüst 2010).

The experiment employs a two-by-two, between-subjects factorial design, varying the race (white/East Asian) and class background markers (higher/lower) of a candidate for office. I then investigate treatment effects across the three countries, examining respondents’ willingness to vote for the candidate as well as assumptions about the candidate’s ideological proximity, relatability, and potential contributions. In doing so, I also build from past research suggesting that respondent ideology likely shapes reactions to candidates from disadvantaged backgrounds (e.g. Weaver 2012; Campbell and Cowley 2014a) and consider potential divergences between leftist, centrist, and rightist respondents.

The results of the study suggest considerable cross-national variation, but several patterns nevertheless emerge. First, the impact of the marginalized identity markers is almost always positive in nature, but these effects are largely concentrated among centrist and (especially) leftist respondents. Second, in the face of candidate identity markers that send conflicting signals, class background mattered more than race everywhere but on the Canadian left. Finally, overlapping marginalized identities do not appear to have shaped reactions, as the lower-class white and East-Asian treatments had broadly similar effects.

Heuristics, biases, and candidate assessment

While the keenest of citizens may collect information on candidates via campaign material, newspaper articles, and debates, most voters also employ shortcuts that help to compensate for limited time and knowledge (e.g. Popkin 1994; Bartels 1996). These heuristics include not only straightforward political markers such as party labels (e.g. Tomz and Van Houweling 2009; Tessier and Blanchet 2017), but also candidate identity markers that are assumed to provide relevant information. Indeed, this is an incredibly widespread phenomenon, ranging from gender (e.g. Anderson, Lewis, and Baird 2011) to sexuality (e.g. Doan and Haider-Markel 2010) to religious background (e.g. Campbell and Cowley 2014b). Here I focus on two relatively less stigmatized markers – East-Asian and/or working-class backgrounds – that have only recently started to garner scholarly attention. A brief overview of the literatures on race and social class allows us to tease out how these markers might matter both independently and interactively.

Research indicating that race shapes voter assessments of candidates dates back decades (e.g. Brady and Sniderman 1985; Citrin, Green, and Sears 1990) – yet it is also surprisingly narrow, as the bulk of its focus has been on the perception of black candidates in the US. In broad strokes, this work suggests that African-American candidates are perceived to be more liberal, compassionate, and committed to minority rights issues than white candidates (e.g. Jacobsmeier 2015; Lerman and Sadin 2016), but that they may also be the victims of anti-black prejudices at the ballot box (c.f. Hutchings 2009; Schaffner 2011). Yet it is an open question whether these dynamics extend to politicians from other racial groups: while research on Latino candidates typically suggests similar effects to those found with African Americans, Asian-American candidates may be an exception to this pattern (c.f. Visalvanich 2017b; Kirkland and Coppock 2018). Nevertheless, analysis of electoral behavior in other Western democracies (including Britain) suggests that minority candidates typically suffer from vote penalties, especially from citizens on the political right (Fisher et al. 2015; Portmann and Stojanović 2019).

Our understanding of the impact of candidate class background is relatively less developed. Although studies have long highlighted that class stereotypes are ubiquitous (e.g. Cozzarelli, Wilkinson, and Tagler 2001; Lott and Saxon 2002), most existing research has only tangentially addressed the impact of class background on candidate assessments. Research on the effect of occupational background, for example, has looked at differences within the (upper) middle-class, but has not incorporated working-class occupations (e.g. Campbell and Cowley 2014b; Coffé and Theiss-Morse 2016). Yet several recent studies buck this trend. Carnes and Sadin (2014) find that American voters assume that politicians with working-class backgrounds are more economically progressive than other candidates, even when party labels are specified. In a study of the UK, the US, and Argentina, Carnes and Lupu (2016) conclude: (1) that respondents from all countries view working-class candidates as more relatable and equally qualified, and that they are just as likely to vote for them; and (2) that Britons (though not Americans or Argentinians) see working-class candidates as more left-wing. And though their focus is not expressly on class per se, Campbell and Cowley (2014a) report, in a study examining how citizens respond to relative candidate wealth, that richer candidates are particularly disliked by centrist and left-wing respondents in the UK.

Previous research is most lacking, however, when it comes to assessing the effect of overlapping class and racial identities. Notwithstanding the prevalence of psychological research highlighting that identity markers interact to shape stereotype formation (e.g. Landrine 1985; Lott and Saxon 2002), the closest studies to this one are focused on the intersection of candidate gender with other markers (e.g. Carey and Lizotte 2019). Yet, the varied conclusions of past research suggest that the effect of interacting candidate markers will differ depending on the characteristics under consideration.

On the one hand, there are those scholars who stress the importance of intersectionality (e.g. Crenshaw 1989; Doan and Haider-Markel 2010; Badas and Stauffer 2017). Most relevantly, Philpot and Walton (2007) argue that in the US, candidate race and gender cannot be considered independently: finding that black female candidates are treated differently than both black men and white women, they conclude that “race trumps gender but the intersection of the two trumps both” (Philpot and Walton 2007, 49). Yet other scholars suggest that certain identities may crowd out others. Coffé and Theiss-Morse’s (2016) study of candidate gender and occupational profiles, for example, suggests that candidate occupation, not gender, shapes competency perceptions in the US and New Zealand. For them, the key consideration is the ease with which a given set of stereotypes can be applied to politics: since gender-based assumptions are comparatively broad, they are more easily overridden by assumptions about occupational profiles (Coffé and Theiss-Morse 2016; see also Irmen 2006; Anderson, Lewis, and Baird 2011). But although this logic is easy to apply when one of the two markers is work experience, it is not obvious how one can make a similar a priori assessment in cases where both markers are identity-based.

Finally, note that there are strong reasons to believe that all of these effects should vary according to respondent ideology. Previous research suggests that voters may assume that racialized minorities (e.g. Brady and Sniderman 1985; Jacobsmeier 2015) and working-class candidates (c.f. Carnes and Sadin 2014; Carnes and Lupu 2016) are more progressive, all else being equal. Based on this heuristic, we would expect leftists to be more likely to support such candidates, whereas conservatives may shy away from them (e.g. Fisher et al. 2015; Lerman and Sadin 2016). Similarly, right-wing voters may be more biased than others against racialized and lower-class politicians, while those on the left might even “positively discriminate” in favor of less advantaged candidates (e.g. Weaver 2012; Campbell and Cowley 2014a; Portmann and Stojanović 2019). Regardless of the precise balance of heuristics and prejudice, however, existing scholarship suggests that ideology is likely to structure voter responses to class- and race-based markers.

Taken as a whole, the above research leads us to assume that voters’ attitudes toward candidates for office are likely influenced by the candidate’s race and class background. Three issues remain unclear, however: the impact of intersectional identities; the relevance of identity markers from less intensely stigmatized social groups; and the generalizability of findings from the US. The remainder of this article provides one attempt to address these limitations.

Data and case selection

Respondents were recruited via Qualtrics LLC’s Internet panel, with national samples constructed to reflect census demographics on the gender and age distribution of their respective adult populations. The surveys in Canada and the UK were conducted in December 2017, while the US variant was fielded in July 2018. After removing individuals who responded “Don’t know” to any of the questions used in the analysis,¹ we are left with a total of 1103 Canadians (outside of Quebec),² 1129 Britons, and 1740 Americans.³

The three cases all have sizeable and relatively long-standing immigrant and visible minority populations, and their shared use of English allows me to avoid introducing any translation-related variation. Elected representatives in all three countries are also disproportionately likely to be white and of a higher class background (Bloemraad 2013; Lamprinakou et al. 2016; O’Grady 2019) – though a lack of comparative work on the latter dimension makes it difficult to pin down exact patterns of class representation. Most importantly, however, this case selection allows us to compare results from the highly-studied American context to those in two cases where class and race are likely to have divergent levels of salience.

Addressing these expectations in turn, I begin by noting that class is likely to have a greater impact on attitudes towards candidates in the UK than in Canada and the US. Social class has tended to play a comparatively prominent societal role in Britain, with class dynamics both shaped by and shaping British politics (e.g. Evans 1999; Evans and Tilley 2012). Indeed, the decreasing presence of working-class candidates in Britain has been highlighted as a source of reduced working-class policy preference representation and increased alienation among working-class voters (e.g. Evans and Tilley 2017; O’Grady 2019). Class consciousness and class politics in North America, by contrast, have been relatively muted, with class voting comparatively uncommon and regional and religious cleavages far more dominant (e.g. Lipset 1983; Bélanger and Eagles 2006).

At the same time, Canada’s long history of multiculturalism has given ethnic diversity a particularly central role in (at least Anglophone) Canadian identity (e.g. Winter 2007; Harell 2009). Comparing Canada to the UK and the US, this distinction seems to manifest in a variety of relevant ways: more extensive efforts on the political right to attract visible minority candidates, potentially shifting ideological assumptions tied to racial markers (Bird, Saalfeld, and Wüst 2010); a somewhat smaller overall gap in racial descriptive representation (Bloemraad 2013); and a unique disconnect between on the one hand, white voters and the right, and on the other, visible-minority voters and the left (Medeiros and Noël 2014, 1036–1037). And while Canada is by no means “post-racial,” xenophobia levels and biases against (non-Aboriginal) visible minorities appear to nevertheless be weaker, on average, that in the UK or the US (Ariely 2012; Harell, Soroka, and Iyengar 2016). As a consequence, we would expect Canadian voters’ reactions to visible-minority candidates to stand out from those of their British and American counterparts.

Past research, then, leads us to expect that Canada and the US will share low levels of class salience (relative to the UK) while the UK and the US will share higher levels of racial salience (relative to Canada). Yet given the lack of comparative studies on the attitudinal effects of candidate identity markers and the perils of drawing out country-level determinants with only three cases, we proceed to the analysis in a broadly exploratory manner.

Experimental design

The survey experiment uses a two by two, between-subjects factorial design – with varying racial (white/East Asian) and class (higher/lower) backgrounds. In doing so, it follows past work (Lerman and Sadin 2016; Carnes and Lupu 2016) in presenting respondents with a single candidate profile. Rather than studying how respondents actively compare white/East-Asian candidates from lower/higher-class backgrounds, the goal is thus to examine the simple difference generated by varying candidate identity markers.

At the start of the experimental vignette, respondents were informed that they would be presented with “a few questions about a man who is thinking of running as a candidate for the [House of Commons/Congress]” and were asked to “Please carefully read his short biography below and answer the questions that follow.” They were then presented with the following text (T1 denotes the race-based treatment, while T2 denotes the class-based one):

John [T1: Kavanagh/Kim] is 40 years old and was born and raised in [Canada/the US/the UK], though his parents immigrated to the country from [T1: Ireland/South Korea]. John’s father was [Canada & US [T2: a factory worker/an orthopedic surgeon] / UK [T2: a factory worker/an orthopaedic surgeon]], and his parents sent him to [Canada & US [T2: public/private] / UK [T2: state/private]] schools as a child and teen. John is proud of his [T1: Irish/Korean] roots, but he considers himself [Canadian/American/British] to the core, and he wants to make a real contribution to his country through public service. A self-defined “complete centrist,” John’s political stances have been widely described as squarely in the middle of the ideological spectrum, both on social and economic issues.

A few considerations are worth noting at this point. First, the text focuses on the candidate’s class background, highlighting his family situation while he was growing up. This is in keeping with common practices among candidates campaigning for office (Carnes and Sadin 2014), and also allows me to avoid introducing variation relative to related yet distinct factors, such as post-secondary education, occupation, or wealth (Campbell and Cowley 2014a; Coffé and Theiss-Morse 2016). As described below, however, I also follow up with respondents to confirm an effect on the candidate’s perceived class status. Second, I wanted to ensure that the race-based treatments did not also induce different connotations regarding immigrant status. I therefore defined both the white and non-white candidates as the children of immigrants, while also highlighting the candidates’ identification with Canada/Britain/America.

Third, I sought to avoid selecting white and non-white ethnic backgrounds that were associated were overly divergent levels of stigmatization. For this reason, I followed recent research (Visalvanich 2017a) and opted to give the non-white candidate an East-Asian background – a so-called “model minority” group (Maddux et al. 2008); this status as an “acceptable” minority group is broadly common across the cases, with past work suggesting that whites are generally less racist toward East Asians than toward other visible minority groups (e.g. Blacks, South Asians) in all three of the countries under study (Harell, Soroka, and Iyengar 2016). The specific choice of Korean and Irish was, in turn, guided by two goals: to reduce the possibility that respondents would call to mind a prominent real-world candidate when presented with the vignette; and to increase the likelihood that respondents would infer white and East-Asian racial backgrounds. And while the Irish were of course historically discriminated against in all three of these countries (e.g. See 2000), it is clear that they are now deemed “white” by their broader host populations. Whether Irish immigrants to the UK are presently considered more similar to the native population (given the closer relationship between the nations) than in Canada and the US, however, is a question that existing research does not allow us to address.

Finally, the candidate is described (and self-defines) as a centrist, but I refrain from giving him a party label. This reduces respondent reliance on partisan heuristics (see, e.g. Campbell and Cowley 2018; Kirkland and Coppock 2018), while at the same time holding constant a common ideological baseline. The result is better isolation of candidate-trait effects and increased experimental power, as we avoid a multiplication of treatment groups (a particular issue outside of the bipartisan American context). The trade-off, however, is a certain loss of realism: despite the presence of non-partisan elections and “independent” elected officials at lower levels of government, party labels are of course central to national elections in all three countries under study; and these labels have been shown to matter for political behavior across these cases – reducing, for example, the extent to which voters project their own attitudes onto candidates (e.g. Banducci et al. 2008; Tomz and Van Houweling 2009; Tessier and Blanchet 2017). The specific choice of a centrist baseline, in turn, allows for a symmetrical comparison of left- and right-wing responses. This decision has the consequence, however, of creating an asymmetry between centrist respondents – who were presented with a candidate who reflects their ideological position – and left- and right-wing respondents – who were not. It also generates a harder test of ideological heuristics than if I had provided no information whatsoever about the candidate’s ideological position.

After being presented with the treatment, respondents were asked a series of questions about the candidate. The wording of these questions was as follows:

Without knowing which party John will be a candidate for, to what extent do you agree or disagree that… 

• [Vote:] You would consider voting for John.

• [Similar:] John’s political stances are likely similar to your own.

• [Understand:] Someone like John can understand the problems facing ordinary [Canadians/Americans/Britons].

• [More:] We need more people like John in [Parliament/Congress].

These questions were intended to capture the influence of candidate race and class background on: (1) willingness to vote for the candidate (vote); (2) assumptions about the candidate’s ideological proximity (similar); (3) the candidate’s perceived relatability (understand); and (4) the potential contribution of a candidate to parliament, whether driven by representative concerns of a substantive or descriptive nature (more).⁴ Note that while the vote question is meant to serve as a summary assessment, it is a rather blunt instrument for doing so – especially given the lack of competing candidates in the experimental prompt. The remaining three survey items are thus useful for exploring specific dimensions of candidate assessments. Potential responses to all questions were on a five-point scale of “strongly disagree,” “somewhat disagree,” “neither agree nor disagree,” “somewhat agree” and “strongly agree.” Answers were then rescaled from between 0 to 1 for the analysis. Online Appendix Table 1 provides the mean and standard deviation of responses to these questions in Canada, the UK, and the US.

On the screen following these questions, respondents were asked about the class background of the candidate they had just read about, thereby allowing us to assess the presence of a class treatment effect: in Canada and the UK respondents were asked “Would you describe John as … Working class; Middle class; Upper-middle class: Upper class; Don’t know”; while in the US, the equivalent question was divided across two survey items, allowing respondents to separately define the candidate as working class (“Yes; No; Don’t know”) and to place him on a more refined vertical class hierarchy (“Lower class; Lower-middle class; Middle class; Upper-middle class; Upper class; Don’t know”).⁵

Self-placement on the ideological spectrum, by contrast, was recorded considerably prior to the experiment, so as to avoid excessive priming effects. Specifically, respondents were asked to position themselves on a left-right political scale ranging from 0 (“Most left”) to 10 (“Most right”), with an additional option for “Don’t know.” I use answers to this question to divide respondents into leftists (0–3), centrists (4–6), and rightists (7–10).⁶

Findings

The main analysis is based on a series of OLS regressions⁷ carried out in two steps: first, I assess broad similarities and differences across the full samples using three-way interactions between candidate race, candidate class, and country; and second, I use a four-way interaction (adding in political stance) to examine the effects of respondent ideology. In each case, I run the same model on responses to vote, similar, more, and understand while using a consistent underlying sample (i.e. excluding respondents with missing data on any other item in the analysis). Taken together, these four questions help us to determine how candidates’ racial and class backgrounds shape citizens’ reactions to them.

As assessing direct and interactive effects via regression tables is complex, findings are illustrated via figures⁸ showing the marginal effect of each treatment relative to the higher-class white candidate. This approach makes it easy for us to tease out the impact of the two markers of comparative disadvantage, and it also has the merit of reflecting the real-world prevalence of higher-class (and male) white representatives (see, for example, Bloemraad 2013; Lamprinakou et al. 2016). The figures include both 95% (the thin line) and 90% confidence intervals (the thick line), thereby allowing us to note the treatment groups where average responses to a given question were significantly different from those in the higher-class white baseline group. In each instance, effects in Canada are illustrated with a red square, those in the UK with a yellow diamond, and those in the US with a blue circle. Finally, recall that the total response range for each of the dependent variables is 1.

Presenting the first stage of the analysis, Figure 1 shows the full sample treatment effects for each of the four questions (see Appendix Table A1 for regression results).⁹ Beginning with responses to vote, we note clear cross-country differences. Relative to the higher-class white baseline: Canadians were more likely to state they would consider voting for the candidate only when he had the lower-class Asian profile; Britons were more likely to do so only with the lower-class white profile (though this difference narrowly misses conventional significance levels, at p = 0.052); and Americans were more likely to agree whenever the candidate had a lower-class background, regardless of race. Strikingly, responses to similar are affected by candidate identity markers in precisely the same manner. Treatment effects for understand, by contrast, are stable across the three countries and suggest an impact of class, but not race: the two lower-class candidate profiles receive a similar relatability bonus relative to either of the two higher-class candidate profiles. Finally, we find comparable, though more muted, effects on more in Canada and the US – with lower-class background again the key determinant – but no evidence that candidate identity markers mattered in the UK.

Figure 1. Treatment effects (relative to higher-class white profile baseline), by country.

These overall effects should vary considerably across the political spectrum, however, and the remainder of our analysis thus includes an additional interaction with respondent ideology (see Appendix Table A2 for regression results). Figure 2 presents our first insight into this potential variation by illustrating treatment effects among leftist respondents, once again relative to the higher-class white treatment baseline.¹⁰ Results are broadly similar to those found in Figure 1, though with some important exceptions. On vote, we find no evidence of effects in the US: it is only in Canada and the UK that the candidate profiles clearly shaped agreement on this question. Among Canadian leftists, we note a higher likelihood of voting for the East-Asian candidate regardless of his class background, alongside a near-significant effect for the lower-class white candidate; effect sizes are relatively sizeable as well, suggesting a shift in responses ranging from a third to a half of a standard deviation.¹¹ Among British leftists, in turn, we find a positive effect only in the case of the lower-class white treatment.¹²

Figure 2. Treatment effects (relative to higher-class white profile baseline) among left-wing respondents, by country.

We find greater evidence of cross-country overlap, however, when it comes to left-wing respondents’ assumptions about shared political stances: a lower class background led to more positive responses to similar in all three countries (at significant or near-significant levels), with the sole exception of the lower-class white profile in Canada. Effects sizes here are comparable across the cases, though their magnitudes are more modest than those found on vote.¹³ A comparable pattern, with cross-case alignment alongside Canadian exceptionalism, is also present for understand. Leftists in all three countries were more likely to say that the candidate could understand the sorts of problems facing ordinary citizens when he had a lower-class background, regardless of class; yet Canadian leftists were also more likely to agree when the candidate was higher-class and Asian. What is more, the magnitude of these effect sizes is substantial, ranging from 53% of a standard deviation (US, lower-class white treatment) to 86% of one (Canada, lower-class Asian treatment). Finally, treatment effects on leftist responses to more are less widespread, particularly in the UK and the US. In Canada, results mirror those for vote: respondents are more likely to agree with this statement only when the candidate was Asian, regardless of his class background. On the British and American left, by contrast results indicate that solely in the case of the lower-class Asian candidate is there any difference from the baseline, higher-class white treatment. The magnitude of these effects is nevertheless relatively large, with variation ranging from 32% of a standard deviation (US, lower-class Asian treatment) to 57% of one (UK, lower-class Asian treatment).

Figure 3 repeats the exercise, but for centrist respondents – a group that, it should be recalled, was presented with a candidate ostensibly aligned with their ideological position.¹⁴ Unsurprisingly, the result is a reduced set of significant treatment effects with weaker impacts, save for in the US. Indeed, for both vote and similar, the only treatment effects that even near significance are found among Americans: for vote, we note modestly more positive reactions to the lower-class Asian profile¹⁵ as well as a smaller effect (straddling the 90% confidence level, with p = 0.10), for the lower-class white treatment; while for similar, all three of our alternative candidate profiles increased assumed ideological proximity. In this latter case, however, the marginal effect size of the higher-class Asian treatment on perceived ideological proximity is markedly smaller than for either the lower-class white (with a t-statistic of 2.64 and a two-tailed p-value of 0.01) or Asian profile (t = −2.56, p = 0.01).

Figure 3. Treatment effects (relative to higher-class white profile baseline) among centrist respondents, by country.

Treatment effects for understand are more comparable to those found among leftists, but several differences nevertheless stand out. On the one hand, their magnitude is somewhat more modest: whereas on the left these significant treatments were associated with agreement increases between 53% and 86% of a standard deviation, here the corresponding spread is between 27% (UK, lower-class white treatment) and 54% (Canada, lower-class Asian treatment). On the other hand, we also find that several of the Asian treatment effects that were significant among leftists drop out – namely, the higher-class variant in Canada, and the lower-class one in the UK. Indeed, these two race-based effects lose significance in the exact same cases for more as well. Overall, effects on more are found exclusively with the two North American samples and entirely reflect class distinctions.

Finally, Figure 4 rounds out our analysis by presenting treatment effects among right-wing respondents.¹⁶ To begin, we note no impact of any of the candidate profile variations on agreement with vote. Treatment effects on similar are comparably sparse, though we do see two near-significant effects for the Asian candidate treatments: with decreased assumed similarity in the US when the Asian candidate had a higher class background¹⁷ and increased assumed similarity in Canada when he had a lower-class one.¹⁸ Treatment effects on understand, by contrast, are larger and more generalized, with greater agreement among Canadians presented with either of the lower-class background candidates, among Britons presented with any of the three alternative candidate profiles, and among Americans presented with the lower-class Asian one.¹⁹ Note that we thus find positive, race-based treatment effects among right-wing Britons that were absent for their centrist counterparts – though a greater similarity to higher-class candidates is of course to be expected from this group. By contrast, we find no evidence in any of the countries to suggest that identity markers affected right-wing responses to more.

Figure 4. Treatment effects (relative to higher-class white profile baseline) among right-wing respondents, by country.

Bringing these findings together points to several conclusions. First, when lower-class and non-white identity markers mattered, it was by and large for respondents on the left and, to a lesser extent, the center; it is only with candidate relatability (understand) that we see clear indications of an effect among rightists. Yet even on the right, whenever the lower-class and/or East-Asian markers had a statistically significant effect (relative to the higher-class white baseline), it was a positive one. Notwithstanding a near-significant negative effect among American rightists’ perceptions of ideological proximity (similar), we thus find little evidence to suggest more negative reactions to marginalized identity markers on the right; instead, ideological differences are tied to the rarer presence of (positive) treatment effects among conservatives.

Second, in this comparatively hard test of racial effects, class background by and large appears to have mattered more than race. This finding is easiest to see if we consider reactions to the higher-class Asian treatment. On the left and center – the two groups that past research suggests would be more likely to react positively to racialized candidates – this profile was generally associated with either no treatment effect (relative to the higher-class white baseline) or a weaker effect than either of the lower-class background treatments; the major exception to this rule, however, is on the Canadian left, where this profile solicited more positive reactions on three of the four questions (vote, understand, and more). On the right, by contrast, the only effects present suggest greater perceived relatability in Britain (understand) but lower perceived ideological proximity in the US (similar).

Third, the intersection of marginalized identities seems to have had little impact. Reactions to the lower-class Asian and white candidates typically mirrored one another, and where they did not, the direction of the difference was just as likely to favor the white candidate as the Asian one. On the one hand, there are a few instances where the lower-class Asian treatment had a (positive) effect while the equivalent white profile did not – namely, on the Canadian left and right with similar, and on the left cross-nationally with more; yet, only with British leftist responses to more was this difference large enough to be statistically distinguishable (with a t-statistic of −2.15 and a two-tailed p-value of 0.03). On the other hand, the lower-class background Asian candidate was given modestly worse scores than his white counterpart on similar (t = 1.83, p = 0.07) and understand (t = 1.73, p = 0.08) by British centrists.

Finally, we also saw evidence of substantial cross-country variation. Arguably the most marked difference in this regard has already been mentioned, as the higher-class Asian profile increased agreement with vote, understand, and more among Canadian leftists but had no effect on the British or American left; given the weaker connection between visible minorities and the left in Canada, both in terms of voting behavior and candidate race, this is a particularly striking result. Among centrists, treatment effects were especially widespread among Americans (with eight significant effects), more modest in Canada (with two significant and two near-significant effects), and particularly muted in the UK (with only one near-significant effect). Cross-country differences were much less visible on the right, however – largely because the treatments simply had little impact on rightist responses to the questions.

I confirm the robustness of these conclusions in several ways: varying the sample size across the regressions (by including respondents who answered “Don’t know” to questions not required for a given model) has minimal impact on the results; and the same is true of using an alternative breakdown for ideological affiliation (namely, leftists (0–4), centrists (5), and rightists (6–10)) or limiting the sample to white respondents. Most importantly, I also ensure that respondents noticed a class difference between the upper- and lower-class background treatments, irrespective of the candidate’s race. Examining the marginal effects²⁰ of the lower-class treatment on the candidate’s perceived class (see Online Appendix Figure 5 for an illustration of the effects and Appendix Table A2 for the underlying regression results) suggests that the class treatment shifted assigned class in the expected direction. There is, however, one point of distinction between the North American and British samples. In both Canada and the US, the lower-class treatment decreased the probability that a respondent would declare the candidate upper-middle or upper class while increasing the likelihood of any other label. In the UK, by contrast, that class treatment effect was concentrated in a move toward labeling the candidate working class: the likelihood of saying he was middle class remained unchanged for the white candidate and even decreased slightly for the Asian one. Despite this difference, the race treatment was in no case itself associated with a shift in class assignment (see Online Appendix Figure 6, based on the same set of regression results).

Conclusion

This study set out to explore the relative impact of race, class, and their intersection on voter assessments of candidates for office. It did so using a survey experiment in which Canadian, British, and American respondents were asked a series of questions about a fictional candidate, with varying (white/East Asian) race and class (higher/lower) markers included in the biographical profile. I then compared results across the three cases, examining willingness to vote for the candidate as well as assumptions about his ideological proximity, relatability, and potential contributions. In doing so, I also built from past research suggesting that ideology likely shapes reactions to candidates from disadvantaged backgrounds and considered possible divergences between leftist, centrist, and rightist respondents.

Notwithstanding the meaningful cross-national variation that was uncovered, the findings of the study suggest several general conclusions tied to the broader literatures on heuristics, voting behavior, and patterns of representation. First, the effects of the two marginalized identity markers were relatively common among centrist and (especially) leftist respondents, and almost always induced more positive reactions, even among conservatives. Second, we noted widespread class effects, regardless of whether we look at Britons or (more surprisingly) Americans and Canadians. Third, race effects are notably more muted than one might assume based on American studies looking at more stigmatized racial groups – the only exception being on the Canadian left. Finally, overlapping marginalized identities evidently had little impact, with the lower-class white and East-Asian profiles eliciting broadly similar reactions.

Overall, the findings highlight the value of examining candidate identity markers in tandem, focusing on the ways in which their potentially conflicting effects might vary ideologically and cross-nationally. In doing so, this study has attempted to compensate for limited work on non-American and non-black racialized candidates, a relative dearth of research on the impact of class background, and a general tendency to neglect the effect of intersecting identity markers. The two-by-two, between-subjects factorial setup allowed us to tease out the relative impact of racial and class backgrounds, and I attempted to minimize the likelihood that the results might simply reflect large divergences in societal stigmatization, rather than race per se, via a focus on a so-called “model minority” group (Maddux et al. 2008). As a result of this approach, however, the findings here necessarily represent a more conservative test of biases against racialized candidates.

Moving forward, several limitations of the present study point toward potentially valuable avenues for future research. First, this study is restricted to one specific variant of intersecting class and racial markers, and it is of course possible that other overlapping identities have completely different effects (cf. Philpot and Walton 2007; Coffé and Theiss-Morse 2016). Insofar as Asian backgrounds may even elicit more positive assessments than white ones (e.g. Visalvanich 2017a), this is a particularly relevant consideration; future studies would thus be especially well-served by the inclusion of a broader range of candidate backgrounds, including from more denigrated racialized groups (such as blacks). Second, in highlighting the importance of country-level marker salience on voter responses (see Konitzer et al. 2019), I necessarily constrain myself to drawing observations about differences between Canada, the UK, and the US that may not be reflected elsewhere. Third, and relatedly, I assume that citizen reactions vary across different ethnic markers (see, for example, Kirkland and Coppock 2018), and this is likely true even once we extrapolate away from race; “Irish” and “Korean” are not perfect substitutes for “white” and “non-white,” even setting aside especially stigmatized minority groups – and this differentiation itself is a source of potential cross-national variation. Finally, the present study cannot assess how voters might respond to race and class markers when party labels are attached to candidates, nor can it tell us how voters might compare sets of candidates with other marginalized identity markers. More realistic, multi-candidate scenarios would therefore be an especially useful step forward, including to help minimize the risk of social desirability bias. Would left-wing voters care, for example, if a Conservative candidate were a working-class black man? Would they react differently to him if he were a Liberal Democrat, or if he were campaigning against an upper-class white Labour candidate? Incorporating other relevant traits (e.g. gender, sexuality, disability) further complicates this picture. Research seeking to address these sorts of questions would allow us to deepen our understanding of how citizens react to candidates’ overlapping markers of privilege and disadvantage.

Acknowledgement

This study was made possible by Seonghui Lee, who generously allowed me to embed the initial Canadian and British experiments within her broader survey project, and the EU’s Horizon 2020 programme. The article also benefitted from helpful feedback provided by Seonghui Lee, Joshua Robison, Suzanna Tomor, Barbara Vis, and three anonymous reviewers.

Disclosure statement

No potential conflict of interest was reported by the author.

Additional information

Funding

This work was supported by H2020 Marie Skłodowska-Curie Actions: [Grant Number 750556].

Notes

1 The total number of (at least partially) completed surveys was 1352 in Canada, 1403 in the UK, and 2236 in the US.

2 Quebec is excluded from the Canadian sample given its distinct political cleavages and context (see, e.g. Medeiros and Noël 2014).

3 Of these respondents, a substantial proportion identify as non-white (23% in Canada, 7% in the UK, and 22% in the US) and a non-negligible proportion identify as Asian (8% in Canada, 3% in the UK, and 2% in the US). The female/male gender divide, in turn, is 48/52 in Canada and 49/51 in the UK and the US, while the respective age breakdowns in Canada, the UK, and the US are as follows: 8%, 9%, and 11% between 18 and 24; 16%, 20%, and 16% between 25 and 34; 17%, 22%, and 17% between 35 and 44; 18%, 14%, and 18% between 45 and 54; 23%, 18%, and 17% between 55 and 64; and 19%, 16%, and 20% 65 and over.

4 Wordings of vote and understand are adapted from similar survey experiments (e.g. Carnes and Lupu 2016; Ostfeld 2018).

5 This variation is due to differences in the available background data from the surveys in which these experiments were embedded, as I aligned the item on the candidate’s class with the general item on the respondent’s class identification.

6 The left/center/right breakdown is respectively as follows: 288, 445, and 370 in Canada; 322, 438, and 369 in the UK; and 400, 712, and 628 in the US.

7 Having confirmed that these results are equivalent to those from ordered logistic regressions, I present the OLS findings for ease of interpretation and presentation. Also note that as randomization was effective across the treatments, I exclude controls from the models (see Mutz 2011).

8 Figures illustrated using coefplot (Jann 2014) and the 538 scheme (Bischof 2017).

9 See Online Appendix Figure 1 for an overview of mean treatment group responses (with corresponding 83.5% confidence intervals). Note that a lack of overlap between the 83.5% confidence intervals allows us to roughly visualize statistically significant differences at the p < 0.05 level (see Bolsen and Thornton 2014).

10 See Online Appendix Figure 2 for corresponding mean responses by treatment group.

11 Cohen’s d statistics are: 0.33 for the lower class white treatment; 0.44 for the higher class background Asian one; and 0.50 for the lower class Asian one.

12 Cohen’s d = 0.38.

13 Cohen’s d ranges from 0.26 (US, lower-class Asian treatment) to 0.36 (UK, lower-class Asian treatment).

14 See Online Appendix Figure 3 for corresponding mean responses by treatment group.

15 Cohen’s d = 0.32.

16 See Online Appendix Figure 4 for corresponding mean responses by treatment group.

17 Cohen’s d = 0.19.

18 Cohen’s d = 0.25.

19 Cohen’s d ranges from 0.23 in the US with the lower-class Asian treatment to 0.39 in Canada with the lower-class white one.

20 I carry out this analysis using ordered logistic regressions in each case, save for with the binary “working class” question in the US (where I employ a standard logistic regression). Note that I examine responses to the two US questions separately.

Appendix

Table A1. OLS regression results, interacting treatment conditions and country.

	Vote	Understand	Similar	More
Lower class background	0.0376* (0.016)	0.0884*** (0.017)	0.0695*** (0.017)	0.0336* (0.017)
Asian	0.00493 (0.016)	0.0116 (0.017)	0.00448 (0.017)	−0.00134 (0.017)
Lower class background # Asian	−0.00460 (0.023)	0.00231 (0.024)	−0.00974 (0.024)	0.0330 (0.024)
Country baseline: US
Canada	−0.0556** (0.019)	−0.0504* (0.020)	−0.0259 (0.020)	−0.0519** (0.020)
UK	−0.0861*** (0.019)	−0.0359+ (0.019)	−0.0490* (0.019)	−0.0390* (0.019)
Lower class background # Canada	−0.0131 (0.027)	0.0324 (0.027)	−0.0506+ (0.028)	0.00852 (0.027)
Lower class background # UK	0.00463 (0.027)	0.00324 (0.028)	−0.0334 (0.028)	−0.0175 (0.027)
Asian # Canada	0.0151 (0.027)	0.0203 (0.028)	0.00414 (0.028)	0.0323 (0.027)
Asian # UK	0.00345 (0.026)	−0.00856 (0.027)	0.00697 (0.027)	−0.0110 (0.026)
Lower class background # Asian # Canada	0.0192 (0.037)	−0.0169 (0.038)	0.0294 (0.039)	−0.0343 (0.038)
Lower class background # Asian # UK	−0.0346 (0.037)	−0.0207 (0.038)	−0.0204 (0.038)	−0.0119 (0.038)
Constant	0.634*** (0.012)	0.605*** (0.012)	0.569*** (0.012)	0.639*** (0.012)
Observations	3966	3966	3966	3966
R^2	0.030	0.041	0.022	0.019
Adjusted R^2	0.027	0.039	0.019	0.016

Note: Standard errors in parentheses.

⁺p < 0.10.

*p < 0.05.

**p < 0.01.

***p < 0.001.

Table A2. OLS regression results, interacting treatment conditions, country, and ideology.

	Vote	Understand	Similar	More
Lower class background	0.0377 (0.025)	0.101*** (0.026)	0.116*** (0.026)	0.0515* (0.026)
Asian	−0.00643 (0.025)	−0.00172 (0.026)	0.0557* (0.026)	0.00873 (0.026)
Lower class background # Asian	0.0391 (0.036)	0.00920 (0.037)	−0.0592 (0.037)	0.0230 (0.037)
Country Baseline: US
Canada	−0.0103 (0.030)	−0.0335 (0.031)	0.0378 (0.031)	−0.0228 (0.031)
UK	−0.0671* (0.029)	0.00569 (0.030)	0.0233 (0.030)	−0.0130 (0.030)
Lower class background # Canada	−0.0433 (0.042)	−0.00284 (0.043)	−0.125** (0.043)	0.0102 (0.043)
Lower class background # UK	0.00220 (0.043)	−0.0353 (0.044)	−0.0937* (0.044)	−0.0492 (0.043)
Asian # Canada	−0.0132 (0.041)	−0.00197 (0.042)	−0.0762+ (0.042)	0.00327 (0.042)
Asian # UK	−0.0116 (0.041)	−0.0405 (0.042)	−0.0937* (0.042)	−0.0396 (0.041)
Lower class background # Asian # Canada	0.0152 (0.058)	0.0283 (0.060)	0.101+ (0.060)	−0.0368 (0.059)
Lower Class Background # Asian # UK	−0.0399 (0.059)	−0.0254 (0.060)	0.0381 (0.060)	0.0135 (0.060)
Ideology baseline: center
Left	−0.0570+ (0.030)	−0.0606+ (0.031)	−0.0667* (0.031)	−0.0359 (0.031)
Right	−0.00435 (0.026)	0.0228 (0.027)	0.0401 (0.027)	0.0105 (0.026)
Lower class background # Left	0.00884 (0.043)	0.0407 (0.045)	−0.0236 (0.045)	0.00187 (0.044)
Lower class background # right	−0.00671 (0.037)	−0.0574 (0.038)	−0.114** (0.038)	−0.0495 (0.037)
Asian # left	0.0547 (0.042)	0.0412 (0.044)	−0.0511 (0.043)	−0.00252 (0.043)
Asian # right	−0.00170 (0.037)	0.0143 (0.038)	−0.106** (0.038)	−0.0246 (0.038)
Lower class background # Asian # left	−0.0876 (0.060)	−0.0302 (0.062)	0.0334 (0.062)	−0.00243 (0.061)
Lower class background # Asian # right	−0.0601 (0.053)	−0.00452 (0.054)	0.122* (0.054)	0.0295 (0.054)
Canada # Left	−0.0929+ (0.049)	−0.0442 (0.051)	−0.0888+ (0.051)	−0.0679 (0.050)
Canada # right	−0.0572 (0.044)	−0.00783 (0.046)	−0.111* (0.046)	−0.0298 (0.045)
UK # left	−0.0458 (0.048)	−0.0515 (0.049)	−0.120* (0.049)	−0.0829+ (0.049)
UK # Right	−0.0155 (0.043)	−0.0747+ (0.044)	−0.111* (0.044)	−0.00977 (0.043)
Lower class background # Canada # Left	0.0747 (0.069)	0.0540 (0.071)	0.0669 (0.071)	0.00253 (0.070)
Lower class background # Canada # Right	0.0366 (0.061)	0.0526 (0.063)	0.165** (0.063)	−0.00303 (0.062)
Lower class background # UK # Left	0.0450 (0.069)	0.0492 (0.071)	0.0836 (0.071)	0.0570 (0.070)
Lower class background # UK # Right	−0.0225 (0.062)	0.0688 (0.064)	0.116+ (0.063)	0.0491 (0.063)
Asian # Canada # Left	0.0704 (0.067)	0.0941 (0.069)	0.120+ (0.069)	0.0877 (0.069)
Asian # Canada # Right	0.0146 (0.063)	−0.00835 (0.065)	0.135* (0.064)	0.00922 (0.064)
Asian # UK # Left	−0.00248 (0.065)	−0.0000960 (0.067)	0.143* (0.067)	0.0725 (0.066)
Asian # UK # Right	0.0523 (0.060)	0.107+ (0.062)	0.201** (0.062)	0.0371 (0.061)
Lower class background # Asian # Canada # Left	−0.0212 (0.095)	−0.103 (0.098)	−0.0774 (0.097)	−0.00991 (0.096)
Lower Class Background # Asian # Canada # Right	0.0337 (0.086)	−0.0478 (0.089)	−0.153+ (0.089)	0.0178 (0.088)
Lower class background # Asian # UK # Left	0.0167 (0.093)	0.0752 (0.096)	−0.0522 (0.096)	0.0129 (0.095)
Lower class background # Asian # UK # Right	−0.0164 (0.086)	−0.0636 (0.089)	−0.157+ (0.089)	−0.104 (0.088)
Constant	0.648*** (0.018)	0.610*** (0.018)	0.569*** (0.018)	0.643*** (0.018)
Observations	3966	3966	3966	3966
R^2	0.045	0.054	0.055	0.032
Adjusted R^2	0.037	0.046	0.047	0.023

Note: Standard errors in parentheses.

⁺p < 0.10.

*p < 0.05.

**p < 0.01.

***p < 0.001.