Racial Attitudes and Vote Choice in National Canadian Elections

Abstract This article examines the effect of racial attitudes on the electoral performance of the New Democratic Party (NDP). Since 2017, the NDP has been led by Jagmeet Singh, the first non-white leader of a nationally competitive Canadian political party. Voters’ racial attitudes and the race of party leaders have a significant effect on vote choice in the United States. Less is known about whether similar effects exist in Canadian elections. I show that NDP vote choice polarized on the basis of racial attitudes following Singh's ascension to party leader. Voters with cold feelings toward racial minorities were less likely to vote for the NDP in 2019 and 2021 than in comparable historical elections. In contrast, there is no significant difference between 2019/2021 and prior elections in support for the Liberals and Conservatives among such voters. These results suggest that racial attitudes are salient in Canadian elections and that national parties may face an electoral penalty when selecting non-white party leaders.

While race has not historically been a major focus in the study of Canadian politics research (Bilodeau et al., 2012;Thompson, 2008;Gidengil et al., 2012) and has not historically been highly salient in Canadian politics and policy (Banting and Thompson, 2021), race has increasingly been at the forefront of both academic and popular discourse in Canada in recent years.While this scholarly disinterest might at one point have been tied to Canada's historical postwar ethnic homogeneity, when 96 per cent of the population was of European descent (Banting and Thompson, 2021), one in five Canadians today are people of colour (per the 2016 Census). 1 In addition to this increasing diversity, racial justice has risen to prominence in Canadian political discourse.There have been revelations in recent years about the mistreatment of Indigenous peoples in Canada, which Canada's Truth and Reconciliation Commission has dubbed a "cultural genocide," (BBC News, 2021), and anti-immigrant politics have gained new traction (Magesan, 2019;Newbold, 2020).In the 2019 election campaign, Liberal Prime Minister Justin Trudeau's use of blackface in his younger days was a highly publicized scandal (Besco and Matthews, 2022;Dobrowolsky and Leal-Iyoupe, 2022).In addition, a body of research has emerged examining multiple dimensions of race in Canadian politics, including electoral support for minority candidates (for example, Besco, 2015;Bird et al., 2016;Black and Erickson, 2006) and public support for permissive immigration policy (for example, Soroka and Roberton, 2010;Harell et al., 2012;Stolle et al., 2016).Despite this flourishing body of research, only a small body of work so far has attempted to assess the relationship between racial attitudes and voters' partisan voting preferences, though Blais (2005) and Gravelle (2018aGravelle ( , 2018b) are notable exceptions.
The 2019 and 2021 Canadian federal elections provide an ideal opportunity to contribute to this emerging area of Canadian politics literature.In these elections, Jagmeet Singh was the party leader for the New Democratic Party (NDP), which has been one of Canada's most electorally popular parties since its founding in 1961.Importantly, Singh was the first non-white leader to lead a nationally competitive political party. 2 Given the overwhelming body of evidence in the United States that Barack Obama and downticket Democrats received an electoral penalty among voters with negative attitudes toward racial minorities (for example, Lewis-Beck et al., 2010;Clarke et al., 2011;Piston, 2010;Knuckey and Kim, 2015), it is worth asking whether the NDP faced a similar electoral penalty in 2019 and 2021 under Singh's leadership.This question is particularly salient given that Singh is likely to once again be the only non-white national party leader in the next Canadian election (which will take place by 2025).Furthermore, the use of the first-past-the-post electoral system in Canadian parliamentary elections heightens the stakes, given the precarious third-party position of the NDP.A penalty of even a few points nationally has the potential to swing the party's seat share to a substantially greater extent than it might in a system of proportional representation.
There are two existing articles examining the effect of Singh's leadership of the NDP in the 2019 election on Canadian politics.Bouchard (2022) finds that coethnic Sikh Canadians were relatively more favourable toward Singh (though other voters of colour were not) and that Singh was viewed relatively unfavourably in Quebec. 3 Besco and Matthews (2022) find that non-racial policies have acquired racial valence by dint of their association with Singh-a "racial spillover" effect.While research by Bouchard (2022) and Besco and Matthews (2022) provides vital lessons about the role of race in policy and leader evaluations, I seek to expand on this nascent body of work by interrogating the relationship between racial attitudes and support for the NDP under Singh's leadership.
In order to assess whether racial attitudes took on new importance for NDP support under Singh's leadership, I provide a novel analysis of large-N data from the 2004-2021 Canadian Election Studies (CES). 4I find that Canadian voters with more negative attitudes toward racial minorities were less likely to support the NDP under Singh in 2019 and 2021 than they were to support the NDP in prior contests, even after accounting for potential confounding variables such as partisanship.No such effect occurs for either the Liberal or Conservative parties.This study provides strong evidence that the NDP paid an electoral penalty in 2019 and 2021 among Canadian voters with negative attitudes toward racial minorities.

Race and Voting Behaviour
Recent research on Canadian politics has sought to analyze public attitudes toward racial minorities and immigrants.In the aggregate, at least one-third of Canadians have clearly negative views of diversity, immigration, multiculturalism and racial minorities-a figure echoed in public polling data (Besco and Matthews, 2022).Breton (2015) finds in a survey experiment that, in contrast to the Netherlands, priming Canadians to national identity does not decrease their support for immigration-suggesting that immigration and multiculturalism may be positively associated with national identity in the Canadian context.Banting and Soroka (2020) find that while Canadian support for immigration is unusually high in a comparative context (see also Harell et al., 2017, andBilodeau et al., 2012), these attitudes are driven by the same factors as other countries (in contrast to Breton, 2015).As in other contexts, Canadian anti-immigrant sentiment is propelled by cultural anxiety (Breton, 2015) and concern about loss of control over personal social and economic conditions (Harell et al., 2017)-though both factors are offset in the aggregate by the widespread belief among Canadians that immigration aids the economy.
Beyond the national level, scholarship on Canadian politics has also examined variation in attitudes on race and immigration by province.Bilodeau et al. (2012) find that, as is the case nationwide, attitudes toward immigrants and racial minorities are net positive in every province and that these attitudes became substantially more positive between 1988 and 2008.Despite this overall positivity, some differences do emerge: Ontario and British Columbia residents are more likely than other Canadians to prefer a reduction in immigrants, while Quebecers are less comfortable with racial minorities.With regard to Quebec, this divide has parallels with traditional Canadian social cleavages between English and French speakers (Johnston, 2019).
New studies have also begun to home in on the relationship between party identification and Canadians' attitudes toward racial minorities and immigrants.Gravelle (2018a) finds that Liberal and NDP party identifiers have more positive feelings toward Muslims than do Bloc Québécois (BQ) and Conservative identifiers.Gravelle (2018b) also examines attitudes toward immigrants and finds that NDP supporters are less likely than Liberals to believe that there is too much immigration to Canada-and that Liberals are substantially less likely to hold this belief than Conservatives.A similar pattern emerges when partisans are asked about their support for accepting refugees.
Although this body of work is still relatively new, a number of scholars have examined the independent effect of racial attitudes in Canadian elections.Some of this research has examined support for minority candidates (for example, Murakami, 2014;Black and Erickson, 2006;Black and Hicks, 2006;Besco, 2015;Bird et al., 2016;Tossutti and Najem, 2002), while a smaller body of work has examined how minority identity and feelings toward racial minorities affect partisan vote choice among white voters.Research by Blais (2005) finds that white voters who favour immigration and aid to racial minorities are more likely to support the Liberals in national elections.
Despite the relative novelty of scholarship on the effect of racial attitudes on Canadian elections, there is ample reason to expect that racial attitudes affect Canadian voters.Like the United States, Canada, from its time as a British colony to the present day, has been greatly defined by settler-colonial expansion and both cultural and outright genocide against Indigenous peoples (Woolford, 2015;Preston, 2013).Though there were fewer forms of legal discrimination in Canada than in the United States, Canadian politics in the mid-twentieth century were also heavily influenced by minority demands for expanded civil rights (Calliste, 1995).As in the United States, there is both current and historically significant racial inequality in Canada in areas such as income, health and social integration (Reitz and Banerjee, 2007;Ramraj et al., 2016), and there is evidence that rates of racial discrimination in employment in Canada are similar to rates in the United States (Quillian et al., 2019).While Canadian voters may not have sorted into parties on the basis of race (Adams and Griffith, 2015) in the way that many ethnic groups in the United States have (Carmines and Stimson, 1989;Layman and Carsey, 2002;Valentino and Sears, 2005), there is still ample reason to expect that Canadian party politics, like US party politics, should be affected by voters' racial attitudes.While the two countries' demographic compositions and racial politics are clearly distinct (Harell et al., 2012;Soroka and Roberton, 2010), the overwhelming evidence that racial attitudes shape voter preferences in the United States raises questions about whether similar effects manifest in Canadian elections.
The US case provides an important theoretical basis for investigating electoral support for the Singh-led NDP in the 2019 and 2021 federal elections.Racial attitudes affect political behaviour in the United States both historically and in the modern era.The 1950s heralded the beginning of a multidecade process of racial realignment in American politics, with racially conservative whites increasingly affiliating with the Republican Party while black voters and racially liberal whites moved to the Democratic Party (Carmines and Stimson, 1989).During this period, there was a substantial shift in elite rhetoric on race, as explicit appeals to "old-fashioned racism" (OFR) became less socially desirable, as both parties publicly embraced the new norm of racial equality (for example, Mendelberg, 2001;Tesler, 2013;Hillygus and Shields, 2014).While racial resentment against ethnic minorities (that is, the belief that black people don't adhere to American cultural values) continued to drive vote choice in this period (Tesler and Sears, 2010), old-fashioned racist attitudes (such as belief in black intellectual inferiority or opposition to miscegenation) were not predictive of party preferences (Valentino and Sears, 2005;Tesler, 2013) until 2008.That year, the candidacy of Barack Obama, the United States' first non-white major party presidential candidate, harkened a return of OFR as a predictor of voting behaviour (for example, Tesler, 2013).
The return of OFR in predicting voter behaviour in 2008 coincided with a surge in the predictive power of race for voting behaviour in many contexts.Racial resentment made white voters less likely to support black candidates in the 2010-2016 elections (Hale, 2019;Petrow et al., 2018).Scholarship in the following years clearly identified that voters with negative attitudes toward racial minorities were less likely to support President Obama in both his historic 2008 run (for example, Lewis-Beck et al., 2010;Clarke et al., 2011;Piston, 2010) and his 2012 re-election (Knuckey and Kim, 2015).Negative attitudes toward racial minorities also depressed the vote shares of Democratic candidates in the 2009-2020 time period (for example, Abrajano and Hajnal, 2015;Algara andHale, 2019, 2020;Hale, 2019), by dint of the party's association with racial liberalism and racial minorities.
An important finding of the US literature is that Obama's role as the first non-white candidate to lead a major party was the catalyst for the increased salience of racial attitudes in voting behaviour.Several studies have found that Obama underperformed in 2008 (for example, Lewis-Beck et al., 2010;Piston, 2010;Tesler, 2013) and 2012(for example, Jardina, 2019;Knuckey and Kim, 2015) as a result of racial prejudice.The racial backlash against Obama also spilled over into congressional races.As Luttig and Motta (2017) and Abrajano and Hajnal (2015) find, perceptions of the 2014 midterm congressional election as a referendum on Obama were racialized, and those perceptions mediated the link between racial attitudes and 2014 vote choice.Racial backlash against Obama also spilled over into the 2016 presidential race, where Hillary Clinton's embrace of Obama and racially inclusive policy helped Donald Trump's explicit racial appeals resonate with white voters (Sides et al., 2017).More generally, public opinion in the United States has polarized on the basis of racial attitudes, in reaction to Obama's historic presidential campaigns and his presidency (for example, Tesler, 2013;Petrow et al., 2018;Luttig and Motta, 2017;Jardina, 2019).

Theory and Hypotheses
The overwhelming evidence that Obama's historic candidacy dramatically increased the salience of racial attitudes in US elections raises the question: Could a similar dynamic be at work in Canada?The 2019 and 2021 Canadian federal elections provide an ideal test, with NDP leader Jagmeet Singh making history as the first non-white national party leader of a party competitive throughout the country.In addition to his historical candidacy, Singh, as both a provincial and national party leader, vocally advocated for racial minorities (Besco and Matthews, 2022).This article's focus on party leadership, rather than the race of individual candidates for the House of Commons, aligns both with research on Obama in the United States and with research finding that Canadian voters, when deciding how to vote, tend to focus substantially more on a leader's image than on candidates in their riding (Blais et al., 2003;Gidengil et al., 2006;Clarke et al., 2019;Stevens et al., 2019;O'Neill, 1998).
Importantly, there are strong reasons to be skeptical that Singh's racially historical candidacy would engender the same electoral penalty for the NDP as Obama's did for the Democratic Party in the United States.As Banting and Thompson (2021) note, the Canadian party system is substantially less racially polarized and race is less salient in Canadian political discourse.Furthermore, Ambrose and Mudde (2015) point out that the Conservative Party, Liberal Party and NDP all publicly embrace multiculturalism, and in the first two cases, the parties officially embrace it as policy when in government.Finally, the Liberal Party has historically performed very well among racial and ethnic minorities and been perceived as the primary party allied with them (Blais, 2005).
My expectation is that opposition to the NDP in 2019 and 2021 under the leadership of Jagmeet Singh was racialized to a greater extent than opposition to the NDP in prior elections.I expect that racial animus toward minorities will spill over from Singh and affect voting for the NDP, as it did for Democratic congressional candidates during the Obama presidency.In other words, I predict that voters with warmer (colder) feelings toward racial minorities will support the NDP at higher (lower) rates in 2019/2021, all else equal, and that this differential will be greater than in elections prior to 2019.
H1a: More positive (negative) attitudes toward racial minorities among individual voters increased (decreased) their likelihood of supporting the NDP in the 2019 and 2021 elections.

H1b:
The effect of racial attitudes on NDP vote choice in 2019 and 2021 will be greater than in pre-2019 elections.
While racial attitudes may affect vote choice for other parties as well, I expect that the historical candidacy of Jagmeet Singh engenders a unique electoral penalty for the NDP among voters with conservative attitudes toward racial minorities.As such, I predict that there will be no similar statistically significant difference in support for the Liberals and Conservatives pre-2019 versus 2019/2021 among racially conservative voters.

Data and Methods
For this research, I employ data from the 2004-2021 CES, excluding 2011.The year 2011 is excluded from this analysis due to the highly atypical party vote shares, particularly with regard to the NDP and Liberals.The 2011 election saw the Liberals reduced to third-party status for the first time in the party's history, with the NDP forming the Official Opposition to the Conservative government following the election.While the NDP's vote share has fluctuated between 7 and 20 per cent for every other election from 1962 to 2021, the NDP won 31 per cent of the national popular vote in 2011.Ultimately the NDP surge did not result in a long-term realignment of the party system, with the NDP returning to its traditional third/fourth party status in the 2015, 2019 and 2021 elections.
Each of the six surveys included in this analysis (the 2004(the , 2006(the , 2008(the , 2015(the , 2019(the and 2021 CES) CES) employs both a campaign period survey (CPS) in the campaign period and a post-election survey (PES) in the weeks following the election.The 2004-2008 studies were conducted primarily by phone and also included a post-election mail-back survey.The 2015 CES was primarily web-based but also included a telephone and mail-back component.The 2019 CES was conducted primarily online, though a separate phone survey was also administered. 5The 2021 CES was conducted entirely online.
In order to effectively compare voter behaviour in the pre-2019 and 2019/2021 periods, data from these six surveys are pooled.For all analyses, weights are used to ensure a nationally representative sample.Weights are stratified by CES year in the pooled data to ensure proper application.The coding of all variables (for example, party identification, vote choice, income category, education, and so on) has been standardized across each CES prior to pooling.

Measuring vote choice
To test my hypotheses, I generate a standardized vote choice variable, with categories for the five largest parties by vote share in each election: Liberal (1), Conservative (2), NDP (3), BQ (4) and Green (5).Supporters for other parties are coded (6).In each year, vote choice is coded first based on PES response.If no party vote is reported in the PES, CPS vote choice intention responses are used.Finally, CPS questions asking voters which party they lean toward supporting are used to code vote choice if the aforementioned responses are missing.For a full description of the coding scheme, question wording and summary statistics for each variable, refer to the supplementary material.

Measuring racial attitudes
The main explanatory variable in this study is the respondent's feelings toward racial minorities.In each CES survey, respondents are asked to report their feelings toward racial minorities on a continuous 0-100 scale, with zero corresponding to "really dislike" and 100 to "really like."Respondents who decline to answer are coded as missing. 6To aid the ease of interpretability of the feeling thermometer in subsequent analyses, responses have been rescaled to range from 0-1.
Figure 1 shows the unweighted distribution of minority feeling thermometer scores pooled among all respondents and across partisans in the six CES surveys.Though responses are left-skewed for all respondents and for the three largest partisan subgroups, there is still considerable variation in each.Notably, the Canadian parties are not fully sorted on the basis of racial attitudes: there are substantial numbers of Conservative voters in the electorate with warm feelings toward racial minorities, and there are substantial numbers of Liberal and NDP voters with negative attitudes toward racial minorities.While social desirability bias may be contributing to a left skew in this variable, such a skew is likely to result in an underestimation of any effect of racial attitudes on vote choice rather than an exaggeration of it.
Direct measures of feelings toward minorities (also known as old-fashioned racism, or OFR), such as the feeling thermometer scores shown in Figure 1, have often been criticized as likely to dramatically underestimate effects in an era where such views are commonly perceived to be socially undesirable (Tesler, 2013;Mendelberg, 2001).In the United States before Barack Obama's 2008 presidential run, oldfashioned racist sentiments were unrelated to party preferences for decades (for example, Sears et al., 1997;Valentino and Sears, 2005).As such, a finding that a direct effect of OFR exists in support for the Singh-led NDP in 2019 and 2021 would be notable, as analyses using such measures are likely to underestimate rather than overestimate the effects of racial attitudes.Furthermore, a growing body of research in contexts outside of Canada shows that explicit measures of racial attitudes are increasingly as effective as implicit measures (for example, Valentino et al., 2018;Schaffner, 2020).
In addition to using a minority feeling thermometer, I test my hypotheses using an alternate explanatory variable.In each of the included CES surveys, respondents are asked "How much do you think should be done for racial minorities?" and allowed to respond on a 5-point Likert scale ranging from "much less" to "much more." Figure 2 shows the unweighted distribution of responses pooled among all respondents and across partisans in the six CES surveys.As with the minority feeling thermometer, the variable is left-skewed, though there is variation in the aggregate and within the three largest partisan subgroups.
While this survey item provides a valuable alternate explanatory variable, it has two major drawbacks.First, this question has been asked only in the PES, not the CPS, meaning that a large proportion of respondents in the pooled data did not answer this question.Furthermore, while the 2019 and 2021 CES provide PES-only survey weights (which are applied in this article in analyses using this question as the explanatory variable), prior CES instruments do not provide such an option.Second, while this alternative explanatory variable may raise fewer social desirability concerns than the minority feeling thermometer, it introduces a policy consideration that is distinct from racial attitudes.A "much less" response due to racial animus is observationally equivalent to a "much less" response due to ideological opposition to government intervention.This measurement problem has been identified in similar questions measuring "racial resentment" in the American National Election Study (ANES), which critics have argued conflate symbolic racism with conservative ideological views on economic redistribution (Feldman and Brook, 2005) and more generally measure attitudes other than affect toward racial minorities (Kam and Burge, 2018; Banks and Valentino, 2012).As a result of these caveats and a growing body of research in international contexts suggesting that the power of racial attitudes is largely consistent regardless of whether survey questions are racially explicit or implicit (for example, Valentino et al., 2018;Schaffner, 2020), the minority feeling thermometer is presented as the primary explanatory variable in this article, with the "done for racial minorities" item used as an alternative explanatory variable as a robustness check.

Control variables
In addition to my primary independent variable, my regression models also include a number of control variables commonly used in models of vote choice in legislative elections.I control for voter demographics, individual-level characteristics (such as party identification) and election context (such as region).Full question wording, summary statistics and coding details are provided in the supplementary material.
I include standard demographic controls in my predictive models, including education, age, income and gender.Education is an ordinal scale ranging from 1 to 11, with 1 signifying "no schooling" and 11 a professional degree.Age is coded as an ordinal variable, generated by subtracting the respondent's birth year from the year when the survey was administered.Income is coded as an ordinal scale ranging from 1 to 5, with 1 indicating household income below $30,000 and 5 indicating household income over $110,000. 7Gender is coded as a binary variable, with 0 for men and 1 for women. 8 I also account for individual and contextual characteristics identified in the Canadian vote choice literature. 9Party identification is included and coded with identical categories to the vote choice dependent variable.I also account for regional variation by including a binary indicator variable for Quebec, given that numerous studies have noted substantial differences in political behaviour between Quebec and English-speaking Canada (for example, Gidengil et al., 1999;Blais et al., 2003;Blais, 2002;Bilodeau et al., 2012;Wright et al., 2017;Johnston, 2019) and prior work by Bouchard (2022) finding more negative perceptions of Singh in Quebec.Not only are political and racial attitudes distinct in Quebec, but voters' choice set of parties is distinct as well, thanks to the BQ. 10

Specifying a model of voting
Let us now consider my models of vote choice in the 2004-2021 (excepting 2011) Canadian federal elections.In each case, the dependent variable indicates which party's candidate the respondent reports intending to vote for in their riding in the election.Because of the multiparty choice set in Canadian elections, a multinomial logistic regression is used.
Party Vote j = a + b 1 × (Minority Feeling Thermometer × 2019/2021 indicator) where Party Vote is the predicted likelihood of a voter casting a vote for party j in their riding, Minority Feeling Thermometer is the respondent's self-reported attitude toward racial and ethnic minorities, 2019/2021 indicator is an indicator variable reflecting whether the election is 2019 or 2021 (as opposed to an earlier election), X is a set of control variables and ε is the error term.The minority feeling thermometer variable is interacted with the 2019/2021 election indicator variable, since I expect that the effect of racial attitudes on vote choice will be conditioned by the presence of Jagmeet Singh as party leader (which was first the case in the 2019 election).vote choice for the NDP.Regression output for other parties is provided in the supplementary material.

Feelings toward minorities and vote choice
Each model treats a Liberal vote as the base outcome and Liberal Party ID as the baseline party ID, meaning that regression coefficients should be interpreted relative to those bases. 11Each coefficient represents the expected change in log odds of NDP vote choice for a unit change in the predictor variable.For each model, the minority thermometer coefficient indicates the effect of very warm feelings (1.0) toward racial minorities in federal elections from 2004 to 2015.This effect is positive and significant for the naive and demos only models but not statistically significant in the full model.The 2019/2021 indicator variable indicates the effect on the likelihood of support for the NDP at the minimum level of the minority feeling thermometer (0) in the 2019/2021 elections.This effect is insignificant for the naive and demos only models but negative and significant in the full model.The interaction term captures the effect on the likelihood of support for the NDP at the maximum feeling thermometer level (1.0) in 2019/2021.This effect is positive across all models but only statistically significant in the full model.
Across both the demos only and full models, higher levels of income and age are associated with decreased likelihood of voting for the NDP, whereas female gender identity is associated with greater likelihood of voting for the NDP.In addition, higher levels of education are associated with decreased likelihood of voting for the NDP in the demos only model, but the effect is not statistically significant in the full model.Quebec residence is negatively associated with NDP vote choice in the full model as well.Compared to Liberal Party partisans, Conservative Party, NDP, BQ and Green Party identifiers are more likely to vote NDP than Liberal.
To get a better understanding of how racial attitudes affected vote choice for the NDP in 2019 and 2021, we can examine Figure 3.Given that the coefficients in Table 1 present logistic regression coefficients relative to a baseline of vote choice for the Liberal Party, it can be challenging to assess the hypothesized relationship of interest without further examination of the model results.Figure 3 shows the predicted probability of voters supporting the NDP in their local riding in 2019/2021 versus 2004-2015 (excluding 2011). 12Unlike the coefficients reported in Table 1, the predicted probabilities shown in Figure 3 and in the subsequent figures in this article are not relative to a baseline of Liberal Party vote choice.In the 2019/2021 elections, as voters' feelings toward minorities become more positive, their likelihood of voting for the NDP increases (confirming H1a).In the earlier election period, there is no statistically significant difference between the likelihood of an NDP vote as feelings toward racial minorities go from the minimum to maximum value.
Importantly, there is a statistically significant 4.5 per cent gap in the likelihood of NDP vote choice between voters with very negative feelings toward racial minorities in in the 2019/2021 elections versus the 2004-2015 elections (confirming H1b).Voters with very warm feelings toward racial minorities were not more likely to support the NDP in 2019-2021 than in earlier elections (as evidenced by the overlapping confidence intervals at the maximum value of the minority feeling thermometer).
Output for regressions and predicted probability plots for models both excluding Quebec and exclusive to Quebec are presented in the supplementary material.National analysis excluding Quebec yields results with similar substantive implications. 13Analysis in Quebec alone (Table 9 and Figure 7 in the supplementary material) reveals a somewhat, though not altogether, distinct story.As in the national analysis shown in Figure 3, voters in Quebec were less (more) likely to support the NDP when they had very cold (warm) feelings toward racial minorities.Unlike in the national analysis, this effect is not contingent on period, with racial attitudes conditioning Quebec voters' NDP support in the 2004-2015 elections as well. 14While a comprehensive exploration of this distinct result in Quebec is beyond the scope of this article, it aligns with previous work by Bilodeau et al. (2012) showing less enthusiasm for racial diversity and immigration in Quebec and by Turgeon et al. (2019) showing that attitudes toward minority religious symbols are negatively associated with voter-level Liberal values in Quebec but not the rest of Canada.
The relationship presented in Figure 3 is not sensitive to model specification.Figure 4 shows corresponding predicted plots for the naive and demos only models.As we can see, the relationship between racial attitudes and vote choice in the 2004-2015 elections and the 2019-2021 elections is similar, though differences in NDP vote choice likelihood between election periods among those with the most negative sentiments toward racial minorities are not statistically distinct at the 95 per cent level when party identification and region are not accounted for (as they are in the full model shown in Figure 3).As we might expect, given the predictive power of party identification (see column three of Table 1), the effect of feelings toward racial minorities on vote choice is greater in both pre-and post-2019 periods when party identification is omitted.
To provide further validity for these findings, we can examine whether similar relationships exist for support of the Liberal Party and Conservative Party.Such a result would suggest that the shift observed in NDP support by racial attitudes pre-2019 versus post-2019 is reflective of a broader shift in the party system rather than a function of perceptions of the NDP. Figure 5 shows the predicted probabilities of support for the Liberals (a), Conservatives (b) and NDP (c).While the effects of feelings to racial attitudes on vote choice vary by party and by election period, the NDP is the only party for which support varies between election periods among voters with highly negative feelings toward racial minorities.The other notable shift between these two election periods is a 4 per cent decrease in support for Conservatives among voters with highly positive views of minorities (Figure 5, panel b).In other words, it appears that voters with very warm feelings toward racial minorities are decreasingly supportive of Conservatives, while voters with very cold feelings toward racial minorities are decreasingly supportive of the NDP.This mirror image between NDP and Conservative support suggests the emergence of polarization on racial attitudes, paralleling that observed in US elections since the 2008 Obama candidacy (for example, Sides et al., 2018;Algara and Hale, 2020).In sum, the results presented in Figure 5 suggest support for the theoretical expectation underlying H1b: that the shift in NDP support across periods among voters with cold feelings toward minorities is exceptional.

Assessing vote choice with an alternate measure of racial attitudes
It is important to assess whether this result is simply a measurement artifact.I address this concern by using the only alternate measure of racial attitudes present in every CES wave included in this analysis.As described and visualized in the section above on measuring racial attitudes, I leverage a PES question asking respondents "How much do you think should be done for racial minorities?" with responses in the form of a 5-point Likert scale ranging from "much less" to "much more."Though, as described earlier, this measure has drawbacks in comparison to the feeling thermometer, it is nevertheless worthwhile to assess whether a less different measure of racial attitudes yields similar results.
Figure 6 shows the relationship between level of agreement that "more should be done" for racial minorities with likelihood of vote choice for the Liberals (a), Conservatives (b) and NDP (c), by election period. 15Aside from the use of this new measure in lieu of the minority feeling thermometer, the control variables included to predict vote choice are specified identically and all included.As we can see across partisan vote choice categories, the results closely parallel the findings using the minority feeling thermometer presented in Figure 3 and Figure 5. Once again, the results suggest that the findings presented here are robust to alternate measures of attitudes toward racial minorities.

Conclusion
In this article, I find that racial attitudes affected support for the NDP in the 2019 and 2021 Canadian federal elections under the leadership of Jagmeet Singh.While feelings toward racial minorities appear to have had no independent effect on Canadians' likelihood of supporting the NDP in the 2004-2015 elections, colder feelings toward minorities were associated with a decreased likelihood of NDP support in 2019 and 2021.Other parties also saw effects of racial attitudes on vote choice.After accounting for standard predictors of vote choice, colder feelings toward racial minorities decreased support for the Liberals and increased support for the Conservatives in both the 2004-2015 and 2019-2021 periods.
Importantly, the NDP was the only major national party to see a change in support between these election periods among voters with the most negative feelings toward racial minorities.By contrast, voters with very warm feelings toward racial minorities were less likely to support the Conservatives in the 2019-2021 period than the 2004-2015 period.These shifts at the ideological poles of Canadian major party politics suggest (though do not conclusively prove) the emergence of polarization on the basis of racial attitudes in the Canadian party system.Further research in this area is crucial.
While recent research on American politics has consistently found an independent effect of racial attitudes on vote choice, this article is one of the first to do so in the Canadian context.The consensus in Canadian elections research has long been that racial attitudes are significantly less salient for vote choice in Canada than in the United States.The 2019 and 2021 elections provide a strong test of this theory, with the NDP's Jagmeet Singh standing as the first non-white leader of a nationally competitive political party.In line with research showing that the Democrats received an electoral penalty associated with Barack Obama's status as the first nonwhite party leader in the United States, my results suggest that the NDP paid a similar electoral price among voters with negative feelings toward racial minorities in 2019 and 2021.These results are particularly notable given that, unlike some prior studies, the analysis here is not limited to white voters.This article's findings on the interplay between racial attitudes and voter support for the NDP since 2019 join previous studies' findings by Bouchard (2022) and Besco and Matthews (2022) on the policy and public opinion ramifications of Singh's leadership.
The presence of an electoral penalty for the NDP among such voters is particularly challenging for the party, given the electoral system used in Canadian elections.Under a first-past-the-post electoral system, parties have two primary paths to receiving a large share of seats.The first is to be competitive in ridings nationwide (as is the strategy for the Liberals and Conservatives).The second is to be highly regionally concentrated (as is the case for the BQ).While the NDP achieved the former in 2011, becoming the Official Opposition in Parliament for the first time in Canadian history, it has not found a path to similar electoral success in subsequent national elections.Despite winning more than twice the popular vote of the BQ in both 2019 and 2021, the NDP nevertheless earned fewer seats in both contests.In short, if the NDP strategy for winning seats is through nationwide competition, any electoral penalty, even a minor one, is potentially a major obstacle.Indeed, despite surveys showing that Singh was perceived to be the most trustworthy and competent leader (as well as the most favourably viewed overall) in the 2021 election campaign (Rieti, 2021), the NDP ultimately gained only a single seat.
While this article is an important first step, more research should be done to assess the impact of racial attitudes on Canadian elections.Additional data collection is needed to determine whether the race of individual candidates in voters' ridings also distort their vote preferences.It is possible that Canadian politics are sufficiently nationalized that no such effect exists (for example, Stevens et al., 2019), but a candidate-level effect is still possible-especially given that research in the United States has found that black congressional candidates in the Obama era faced an additional electoral penalty among voters with negative attitudes toward racial minorities (Hale, 2019).
The findings in this article join an emerging body of research in showing that race is a more central feature in Canadian politics than previously believed.Earlier work has found evidence that racial minorities in Canada support racial in-group candidates at higher rates (for example, Landa et al., 1995;Besco, 2015) and that support for social welfare policies is affected by the racial group perceived to be benefiting from that policy (Stolle et al., 2016).This study finds evidence that these effects extend to the electoral arena as well, with troubling implications for Canadian politics.If non-white leaders present real barriers to partisan success in Canadian elections, it is perhaps unsurprising that it has taken so long for such a leader to emerge.Even though partisanship is by far the largest predictor of vote choice in Canadian national elections, this article demonstrates that the confluence of negative racial attitudes and a non-white party leader can nevertheless affect vote choice.
In summary, this study extends US research on the effect of racial attitudes on voting behaviour to recent Canadian elections.I find that following the historic major party leadership of Jagmeet Singh in the 2019 and 2021 elections, the NDP was the only party to suffer a novel electoral penalty among voters with strongly negative attitudes toward racial minorities.Furthermore, this shift may be occurring in the context of polarization in party support on the basis of racial attitudes.This article thus contributes to our knowledge of the relationship between racial attitudes, descriptive representation and voter behaviour in Canada.
Supplementary material.The supplementary material for this article can be found at https://doi.org/10.1017/S0008423923000367 13 A minor, though interesting, difference is that the increase in likelihood of support for NDP candidates among those with maximally warm feelings toward racial minorities is statistically significant once Quebec is omitted.See Figure 8 in the supplementary material for more detail.This may be a result of Quebec's exclusion or a result of the absence of survey weights (non-national analyses are unweighted).14 Analysis of Quebec alone does not use survey weights, as weights to generate a representative sample of Quebecers are not available across the included CES waves.15 Full multinomial logistic regression output in table form is provided in the supplementary material.

Figure 2 .
Figure 2. Distribution of responses, by party, to CES question asking "How much do you think should be done for racial minorities?"(a) All Voters, (b) Liberals, (c) Conservatives (d) NDP.

Figure 3 .
Figure 3. Predicted probability of NDP vote by minority feeling thermometer and time period (full model); 95% confidence intervals.

Figure 4 .
Figure 4. Predicted probability of NDP vote by minority feeling thermometer and time period (naive + demos only models); 95% confidence intervals.(a) Naive model, (b) Demos only model.

Table 1
Table 1 reports parameter estimates for my vote choice models for the NDP across the elections in this study.Each column displays the results for NDP vote choice.The first model (naive) is bivariate, with the interaction between minority feeling thermometer and the 2019/2021 election indicator as the sole right-hand term.The demos only model includes demographic characteristics but excludes the Quebec indicator and party identification.The full model includes all variables of interest, including party identification (the strongest predictor of vote choice).All analyses use nationally representative survey weights, stratified by election survey.Each model presented in Table 1 is a multinomial logistic regression predicting Three Models of NDP Vote Choice, Minority Thermometer IV (multinomial logit, Liberal baseline)