Canadian Behavioural Scholarship on Race

Behavioural scholarship in Canada has centred on several major factors to explain national voting patterns and political partisanship over time. This includes religion (Johnston, Reference Johnston2017), regional divides (Gidengil et al., Reference Gidengil, Nevitte, Blais, Everitt and Fournier2012) and language (Lijphart, Reference Lijphart1979). The political importance of racial identity and racial attitudes traditionally received only limited attention across Canadian research (Thompson, Reference Thompson2008), but more recent work has begun to investigate the topic more closely.

Older theoretical work conceived the difference between French and English Canadians as one based on race (Siegfried, Reference Siegfried1907). Throughout most of the nineteenth century, it was argued that English Canadians did not consider French Canadians to be white (Scott, Reference Scott2015). However, this changed as French Canadians came to be viewed as “white folks,” who are a distinct Western national group of people and “solidly lodged in whiteness” (Scott, Reference Scott2015: 1280).

Beyond the divide between English and French Canadians, it has been shown that visible minorities in Canada are more likely to support the Liberals (Blais, Reference Blais2005; Blais et al., Reference Blais, Gidengil, Nadeau and Nevitte2002; Gidengil et al., Reference Gidengil, Nevitte, Blais, Everitt and Fournier2012; Nevitte et al., Reference Nevitte, Blais, Gidengil and Nadeau2000). Throughout the 2000s, those who favoured doing less for minorities were more inclined to support right-wing parties, and those who favoured doing more for minorities were more supportive of the NDP (Blais, Reference Blais2005; Gidengil et al., Reference Gidengil, Nevitte, Blais, Everitt and Fournier2012). Work by Besco finds that nonwhite voters show a strong affinity for supporting their own ethnocultural group and seem to show stronger support for other minority candidates (Besco, Reference Besco2015; Reference Besco2019). Ethnic minority candidates in Canada experience electoral discrimination when they run as members of right-wing parties, but not when they run as members of left-wing parties (Besco, Reference Besco2020).

Recent work by Hale (Reference Hale2023) analyzes the effect of racial attitudes on electoral support for the federal NDP under the leadership of Jagmeet Singh, finding that colder feelings towards ethnic minorities are associated with a decreased likelihood of supporting the NDP in both the 2019 and 2021 federal elections. Bouchard (Reference Bouchard2022) finds that Singh’s candidacy was more negatively perceived in Quebec, but that it also led to an increase in affinity-voting for the NDP by Sikh Canadians. Using Canadian Election Study data and social group feeling thermometers, Lachance and Beauvais (Reference Lachance and Beauvais2024) show how feelings towards others, including ethnic minorities and groups associated with Quebec, have become more important determinants of voting for the Conservatives outside of Quebec over time. Yet, Bilodeau et al. (Reference Bilodeau, Turgeon and Karakoç2012) show that attitudes among white Canadians towards immigrants and racial minorities are a net positive in every province, and that these attitudes became more positive from 1988 to 2008. Relatedly, Gravelle (Reference Gravelle2018) shows that Liberal and NDP partisans express more positive feelings towards Muslims than do either Bloc Québécois (BQ) or Conservative Party partisans.

The only empirical study of white identity in Canada is by Beauvais and Stolle (Reference Beauvais and Stolle2022). These authors measure white Canadians’ attachment to their racial in-group as well as their resentment towards Indigenous peoples. They find that white identity among Canadians increases support for public pensions, and that anti-Indigenous resentment lowers support for welfare policies (Beauvais and Stolle, Reference Beauvais and Stolle2022). Furthermore, Beauvais and Stolle show that white identity and anti-Indigenous resentment are associated with voting for the Conservative Party (Beauvais and Stolle, Reference Beauvais and Stolle2022).

This important recent work has begun to focus more attention on the politics of racial identity and racial attitudes in Canada. This project builds on this scholarship in several ways. I directly compare racial identity, racial attitudes and their political effects between white Canadians and Americans by posing virtually identical questions on temporally aligned surveys and employing an original experimental design to capture the causal effect of a candidate’s race.Footnote ¹ This analysis allows for the first direct empirical comparisons of white racial identity and racial attitudes between white Canadians and Americans.

Data

Almost thirty years ago, Nevitte (Reference Nevitte1996) argued that part of the reason for the lack of comparative scholarship on Canada and the United States was methodological difficulties. The same point can be made today. Past qualitative work is extensive and rich in context, but its generalizability is suspect (Nevitte, Reference Nevitte1996). On the quantitative side, there exist few publicly available surveys that pose identical questions to both Canadians and Americans. Surveys that ask similar questions are not always temporally aligned and are not always theoretically relevant. Both Canada and the United States have long-running national election studies, but these surveys are rarely conducted in the same year, contain different sets of questions or altered question wording, and are not always focused on the same theoretical avenues.Footnote ²

Correspondingly, original data is necessary to expand the scope of comparative analysis and the reliability of inference. On this front, this project leverages a cross-sectional public opinion survey of Americans and Canadians fielded from December 2022 to January 2023. Survey respondents were recruited by Leger, a Canadian-based marketing/polling firm hired to conduct the survey.Footnote ³ Leger is a well-regarded polling firm known for their accuracy. In both the 2019 and 2021 Canadian federal elections, Leger provided the most accurate polling relative to actual election results (Grenier, Reference Grenier2021). I used Leger for both the American and Canadian portions of the survey to avoid any potential bias from the use of different survey firms.

The survey was fielded to a nationally representative sample of 2,000 Americans and 2,000 Canadians, which are then subset in the analysis to only white individuals. Respondents were recruited to fill quotas and configured using American and Canadian census data, for age, gender and geographic region. Following data collection, national survey weights (for age, gender and region) were calculated and appended to the dataset.Footnote ⁴

To analyze the representativeness of the collected data on other important indicators, namely education and political leanings, in the Appendix, I present several tables comparing the collected survey samples to national election study surveys in Canada and the United States. In terms of education level, the collected samples contain almost the exact same percentage of respondents with bachelor’s degrees but a smaller proportion of graduate degrees and a slightly larger proportion of only high school graduates (Table 1A). In terms of ideological-lean, there are only marginal differences with the Canadian sample being 0.3 points (on a 0–10 scale) more left-leaning than the 2021 Canadian Election Study (CES) sample and the American sample being 0.07 points more right-leaning than the 2020 American National Election Study (ANES) sample (Table 2A). The collected Canadian sample contains a slightly larger proportion of Liberal voters than the 2021 CES (3%) but slightly fewer Liberal partisans (2%) (tables 3A and 4A). The collected American sample contains more Democratic partisans but almost the same number of Democratic and Republican voters as the 2020 ANES (tables 5A and 6A). Accordingly, the collected samples are slightly less educated and skew to the left compared to the samples in the most recent national election studies in Canada and the United States, but by only a marginal degree.

To maximize empirical comparability, the survey contains nearly identical questions for American and Canadian respondents. Question wording was only altered for country-specific contexts, such as the names of political parties. The survey contained a large number of questions to gauge respondents’ background demographics, their political preferences, and the social identities that are most important to them. This includes a number of variables measuring racial attitudes, resentment and white identity as described in the subsequent section.

The median survey completion time was just over 17 minutes for Canadians and 14 minutes for Americans. The survey included two attention checks to ensure data quality. The first was an easy attention check that almost all respondents passed. The respondents who did not pass this first attention check were removed from the survey and replaced with new respondents.Footnote ⁵ The harder attention check is the favourite colour question discussed in Berinsky et al. (Reference Berinsky, Margolis and Sances2014). A larger percentage of respondents failed this check (36%), similar to the results of Berinsky et al. (33%), but no respondents were removed based on answers to this question to avoid biasing the sample.Footnote ⁶ As a robustness check, the main regression models in the forthcoming analysis are also run amongst the sample that passed this harder attention check.Footnote ⁷

This project’s original quantitative design also included a paired-profile, candidate-choice conjoint experiment.Footnote ⁸ The advantage of a conjoint experiment is that it allows researchers to assess the effects of multiple experimental treatments simultaneously and identify respondents’ preferences within and across multiple tested dimensions (Bansak et al., Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman and Green2021). The intent of the experimental design is to mirror the relevant background information about political candidates that might determine a respondent’s actual choice at the ballot box, and to then compare the relative effects on vote share of these different pieces of information between Americans and Canadians. The design of the conjoint experiment was pre-registered at Open Science Framework in November 2022, prior to it being fielded to respondents.Footnote ⁹ Respondents were shown two side-by-side profiles of hypothetical political candidates and were asked to choose which candidate they would support.Footnote ¹⁰ The experiment provided six pieces of information about each candidate. These six attributes were the hypothetical candidate’s party, race, gender, major endorsement received, occupation history and their number one policy priority.

Measuring Racial Attitudes, Identity and Resentment

Feeling thermometers were first employed in the 1964 ANES in the United States to measure survey respondents’ affect towards groups of people and prominent political figures (Winter and Berinsky, Reference Winter and Berinsky1999). Feeling thermometers have been employed in the CES, which has asked survey respondents to rate groups of people, provinces and prominent politicians on a 0–100 scale.Footnote ¹¹ Because of their use in the CES, the most common empirical approach to assessing the importance of attitudes towards other social groups in Canadian politics has often been to utilize feeling thermometer scores (see Lachance and Beauvais (Reference Lachance and Beauvais2024) on the growing importance of group-based affect in the structure of the Canadian party system). The advantages of these measures are their conceptual clarity and logistical ease.

I utilized feeling thermometers in the Canadian and American surveys to capture general sentiments towards different racial groups. This set of questions is utilized in order to operationalize racial affect and to measure public attitudes towards nonwhite groups more generally (Winter and Berinsky, Reference Winter and Berinsky1999). I included a feeling thermometer measure for Indigenous peoples, which can helpfully capture sentiment towards this group among Canadians and Americans (see Appendix 1 for each question).

However, feeling thermometer scores only provide one piece of the puzzle. How individuals rate other groups of people on 0–100 scales, for example, says little about the extent to which their racial group is an important social attachment, or whether they believe other racial groups are undeserving of a hand-up from government. Feeling thermometers indicate the extent to which individuals have “warm” or “cold” feelings towards other groups of people. While this may be correlated with other attitudes, accurately measuring theoretical constructs such as white racial identity and racial resentment requires their own questions. And these types of questions have often not been included in the CES.Footnote ¹² Moreover, as the forthcoming analyses will show, the comparative picture of the behavioural effects of racial attitudes are markedly different when researchers go beyond measures of racial affect.

Therefore, this study employs several other measures of racial attitudes and identity to estimate the political importance of race more comprehensively.Footnote ¹³ Jardina bases her measure of white racial identity on five questions across multiple survey sources, including the ANES (Jardina, Reference Jardina2019).Footnote ¹⁴ Each question targets a particular aspect of white racial identity construction: the centrality of the identity, evaluations of the group and the degree of white consciousness. The original survey launched as part of this project replicated Jardina’s (Reference Jardina2019) measure of white racial identity, fielding each of the five questions simultaneously to nationally representative samples of Americans and Canadians (see Appendix 1 for each question). This measurement of white racial identity differs from Beauvais and Stolle (Reference Beauvais and Stolle2022), who construct their index measure of white identity using three of the survey questions originally operationalized by Jardina (Reference Jardina2019).

I also measure racial resentment using the traditional set of four survey questions used in the ANES (Kinder and Sanders, Reference Kinder and Sanders1996), recoded for appropriate attitudinal direction, and combined into an index measure for both the Canadian and American samples (see Appendix 1 for each question). This index, born out of American scholarship, captures anti-Black sentiment, moral traditionalism and is said to be a measure of symbolic or more subtle racism as opposed to older forms of explicit racism and bigotry that were based on claims of biological inferiority (Agadjanian et al., Reference Agadjanian, Michael Carey, Horiuchi and Ryan2023; Carmines et al., Reference Carmines, Sniderman and Easter2011; Kinder and Sanders, Reference Kinder and Sanders1996). Respondents who express high levels of racial resentment hold that Black people lack moral values of hard work, self-sufficiency and discipline; that they have received more than they have earned; and that they do not try hard enough to overcome difficulties (Carmines, et al. Reference Carmines, Sniderman and Easter2011; Kinder and Sanders, Reference Kinder and Sanders1996). Some critics argue that the racial resentment index, as it is commonly operationalized, is flawed because it can be conflated with attitudes towards racial policies and that it measures what some argue it predicts (Carmines et al., Reference Carmines, Sniderman and Easter2011; Schuman, Reference Schuman2000). Yet, others have shown the wide predictive power of varying scores on the racial resentment index for other racial attitudes, policy attitudes and voting behaviour in the United States (Agadjanian et al., Reference Agadjanian, Michael Carey, Horiuchi and Ryan2023; Hooghe and Dassonneville, Reference Hooghe and Dassonneville2018; Kinder and Sanders, Reference Kinder and Sanders1996).

But is racial resentment relevant in Canada? Beauvais and Stolle (Reference Beauvais and Stolle2022) have assessed the importance of racial resentment in Canada, tying evaluations of a nonwhite group to questions of deservingness and effort, but do not measure resentment towards Black people. Instead, these authors argue that the more appropriate out-group which white Canadians are more likely to express resentment towards are Indigenous peoples. While this is theoretically possible, this project departs from Beauvais and Stolle (Reference Beauvais and Stolle2022) and opts to measure racial resentment in the same way for Canadians and Americans for several reasons.

First, this is for the sake of comparison: unless identical constructions of racial resentment are utilized, comparisons between Canadians and Americans cannot be meaningfully conducted. Second, simply stated, anti-Black racism exists in Canada too (Calliste, Reference Calliste1995; Lewis, Reference Lewis1992; Thompson, Reference Thompson2008). The number of Black Canadians as a percentage of the population is not insignificant, totalling over 1.2 million as measured by the 2016 census. Given that empirical scholarship in Canada has not posed survey questions measuring resentment towards Black people in Canada, we cannot say with certainty whether some Canadians express out-group resentment towards Black people or how this compares to the United States. Third, it is not clear why the link between anti-Black attitudes and moral traditionalism, feelings of deservingness and the broader ideal of the Protestant Ethic should only exist in the United States (Kinder and Sander, Reference Kinder and Sanders1996). This set of moral attitudes, arguably based in part on religious doctrine, and its link to attitudes on work and capitalism are explicitly not limited to the United States and are said to have originated in Northern Europe (Weber, Reference Weber1930).Footnote ¹⁵ Furthermore, if we have yet to pose these questions to respondents in Canada, we cannot be certain whether these connections do not exist in the minds of some Canadians who may be drawn to more subtle forms of racial prejudice. Finally, consistent with literature on racial resentment in the United States, the index is strongly correlated with other measures of racial attitudes, policy preferences, derogatory racial stereotypes and opinion towards other minority groups (Kinder and Sanders, Reference Kinder and Sanders1996). The racial resentment index is an important measure of racial attitudes and worthy of investigation in the Canadian context.

Moreover, in the Appendix (section 8), I present the results of a principal component factor (PCF) analysis of the measures for racial resentment. I do this to analyze the statistical relevance of each item used in the index and to show whether the construct means the same thing between the American and Canadian samples of white respondents. In short, the PCF analysis shows that among white respondents the racial resentment index is quite comparable between the Canadian and American samples.

Public Attitudes in North America: The Weaker Racial Cleavage in Canada

How do white Canadians and white Americans compare in terms of white racial identity and racial attitudes? This section reports descriptive statistics and difference-in-means tests among white respondents from the nationally representative survey samples. Beginning with white racial identity, Figure 1 plots the raw percentages of Canadian and American survey respondents to the question “How important is being white to your identity?” which is the first and, arguably, most crucial question for the white identity index measure (Jardina, Reference Jardina2019).

Figure 1. White Identity in Canada and the United States.

Source: Original survey data (2023). National samples are nationally representative samples from the United States and Canada, white respondents only. X-axis is percentage of the respective sample. Y-axis indicates answer to question. Colour indicates which country sample.

Figure 1 shows that there are significant differences in the extent to which white Americans and Canadians identify with their racial group. Starkly, only 7 per cent of white Canadian respondents stated that being white was “extremely” important to their identity compared to almost 19 per cent of American respondents. Over 37 per cent of the sample of white Americans stated that being white was either “very” or “extremely” important to their identity, while only 20 per cent of the sample of white Canadians did the same. A much greater percentage of white Canadians (almost 52%) stated that being white was “a little” or “not at all” important compared to the sample of white Americans (36%).

Figure 2 repeats the comparison across the national samples for a survey question measuring racial resentment. It displays the frequencies corresponding to how white respondents answered the first question of the racial resentment scale. Figure 2 shows how there are again significant differences between white respondents across the border. Only 15 per cent of white Canadians stated that they strongly agree with the prompt that Black people should not receive any special favours, compared to almost 33 per cent of white Americans. Over 32 per cent of white Canadians disagree either “strongly” or “somewhat” that Black people should not receive any special favours, compared to less than 20 per cent of white Americans. A majority of white Americans (over 57%) “strongly” or “somewhat” agreed with the prompt, compared to 37 per cent of white Canadians. However, it is notable that more white Canadians agreed with the racially resentful prompt than disagreed, suggesting that resentment towards Black people in Canada, as measured with this survey question, is not insignificant. In the Appendix (figures 11A–17A), I present the relative frequencies of the samples of white Canadians and Americans to the other survey questions measuring white identity and racial resentment, which show a similar empirical pattern.

Figure 2. Racial Resentment in Canada and the United States.

Source: Original survey data (2023). National samples are nationally representative samples from the United States and Canada, white respondents only. X-axis is percentage of the respective sample. Y-axis indicates answer to question. Colour indicates which country sample.

Table 1 displays the results of a difference-in-means analysis of nine measures of white identity and racial attitudes between the national samples of white respondents. Each variable has been recoded between 0 and 1 to increase comparability. These variables include: the full white racial identity index, the single variable measure of white identity (“How important is being white to your identity?”), the full index measure of racial resentment, as well as feeling thermometer scores for different racial groups.Footnote ¹⁶ The feeling thermometer for nonwhites was calculated by averaging the scores respondents provided on each of the thermometers for nonwhite groups (Black people, Indigenous/Native peoples, Asian people, and Hispanic/Latino people).

Table 1. Difference-in-Means National Samples of White Respondents

Source: Original survey data (2023). Nationally representative samples from the United States and Canada. White respondents only. Each variable has been recoded between 0–1 to improve comparability. Means are calculated and rounded to four decimal points. Standard errors are reported in brackets. *Mean for nonwhites feeling thermometer is the average across each nonwhite group thermometer.

On average, white Americans display a greater degree of attachment to their racial identity than the average Canadian. On the full index measure, this difference is equal to 0.09 on a 0–1 scale. On the single-item measure of white identity, this difference is equal to 0.14 on the same scale. Evidently, there is also a considerable difference between the national samples in terms of racial resentment. On average, the national sample of white Americans express higher levels of racial resentment than the corresponding Canadian sample. This difference on the index measure of racial resentment is equal to 0.11 on a 0–1 scale.

However, the differences between the national samples are more modest when comparing scores on feeling thermometers. The average feeling thermometer score for white people and nonwhite people is virtually identical between the national samples of Americans and Canadians. These differences are small and not statistically significant. Yet, there are larger differences when the feeling thermometers are broken down by specific nonwhite racial groups. The American national sample had colder feelings towards Latinos and Black people relative to the Canadian national sample, while the Canadian national sample expressed colder feelings towards Indigenous peoples than the American sample. Feeling thermometers provide important pieces of information, especially as it pertains to the specific dimensions of racial affect. But it also shows the importance of not only using feeling thermometer scores to evaluate the political importance of race.

Going further, in the Appendix, I present three violin plots that display the distribution, the median, and the interquartile range of the respective samples of white Canadians and white Americans on the measures of white identity (Figure 8A), racial resentment (Figure 9A) and nonwhite feeling thermometer scores (Figure 10A). These figures confirm the key findings of the difference-in-means analysis and provide additional context on the attitudinal differences between Canadians and Americans. Consistent with Table 1, the median points on the white identity index and racial resentment index are considerably lower for white Canadians compared to white Americans, but the median point for nonwhite feeling thermometer scores is virtually identical between the two samples. The violin plots do not evidence highly polarized distributions on these measures among either the Canadian or American sample. Yet, they do show how the bulk of white Canadian respondents express lower levels of racial resentment and less attachment to their racial identity than most white American respondents, underscoring how the base level attitudes of these populations are different.

Experimental Evidence: The Comparative Effects of a Candidate’s Race

What do differences in white identity and racial resentment potentially signify for political behaviour and electoral outcomes? This section turns to experimental evidence to assess how the presence of nonwhite political candidates affects vote choice between white Americans and white Canadians. Examining socially sensitive topics, such as how the race of candidates influences vote choice, can be difficult because people may not want to admit how this affects their decisions. However, one inherent advantage to conjoint experiments is that they have been shown to reduce social desirability bias (Bansak et al., Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman and Green2021; Horiuchi et al., Reference Horiuchi, Markovich and Yamamoto2019). By posing several varied pieces of information about candidates, respondents may be less concerned that researchers will connect their choices to a specific attribute (Bansak et al., Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman and Green2021).

The setup of the experiment was identical for respondents from either country, but with some of the levels of the attributes adjusted for country context. For instance, the party attribute could be Liberal, NDP, or Conservative for Canadians, and Democrat or Republican for Americans. The race attribute could be white, Black, Asian, or Indigenous for Canadians, and white, Black, Asian, or Hispanic/Latino for Americans.Footnote ¹⁷ The gender attribute, major endorsement attribute, occupation attribute and policy priority attribute had the exact same possible levels for both Americans and Canadians.Footnote ¹⁸

The levels of each attribute that respondents saw in the experiment were determined randomly with one important exception. To better replicate political conditions in each country, the levels of the policy priority attribute were constrained to certain political parties. For example, it would make little sense for respondents to encounter a candidate from either the NDP or the Democrats campaigning primarily on a policy platform of reducing immigration and refugee settlement. Thus, to enhance realism and external validity, the levels of the policy priority attribute that had a clear ideological-lean were constrained to the corresponding political party on that end of the left-right spectrum.Footnote ¹⁹ Given the randomization constraint between a candidate’s policy priorities and their political party, I created an interaction variable (Party_Policy) that combined each possible pairing of the two attributes and estimated an AMCE for each level.

To first analyze the results of the conjoint experiment, I estimated Average Marginal Component Effects (AMCEs) for each possible level of the attributes with a given baseline category among a subset of only white respondents from each country.Footnote ²⁰ The AMCE represents the causal effect of an attribute level (that is, candidate race = Black) against another baseline value of the same attribute (that is, candidate race = White), while holding equal the joint distribution of the rest of the attributes in the design (Bansak et al., Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman and Green2021). In this experiment, the AMCEs represents the average effect of an experimental treatment on a candidate’s vote share, given all the information that respondents have seen about a candidate. This experimental approach models how individuals make a single choice within a multidimensional information environment, like voting in an election (Bansak et al., Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman and Green2021).

Figure 3 plots the AMCEs of each candidate attribute with confidence intervals from the conjoint experiment among white respondents, both French and English, from the national sample of Canadians.Footnote ²¹ There are numerous findings from this experiment, yet in this article I will focus on the estimated effects of a candidate’s race.Footnote ²² Among white Canadians, Figure 3 shows resoundingly that the race of a political candidate has no discernible negative effect on their vote share. Relative to white candidates, the baseline category, candidates randomly assigned to be Asian received a slightly lower average vote share (−1.3%), yet this effect was not statistically significant. Candidates randomly assigned to be Black (+2%) or Indigenous (+1.5%) received slightly higher average vote shares than white candidates, although these effects were also not statistically significant.

Figure 3. Canadian Conjoint Results.

Average marginal component effects (AMCEs) of candidate attributes. Outcome is a binary forced choice variable of which candidate you would vote for. Lines represent confidence intervals. Respondents are nationally representative sample of Canadians, whites only. N = 1,549. Effect sizes for race treatments: Asian = −0.013 (0.012); Black = 0.020 (0.012); Indigenous = 0.015 (0.013).

Figure 4 plots the corresponding results from the candidate-choice conjoint experiment among white respondents from the national sample of Americans. By contrast to the Canadian results, nonwhite political candidates do receive an electoral penalty from white Americans. Political candidates randomly assigned to be Asian (−2.9%), Black (−2.1%), and Hispanic/Latino (−3.6%) all received lower average vote shares relative to white candidates in the experiment. The negative effect corresponding to Black candidates is just below the conventional level of statistical significance, but the negative effects corresponding to both Asian and Hispanic/Latino candidates are statistically significant.

Figure 4. American Conjoint Results.

Average marginal component effects (AMCEs) of candidate attributes. Outcome is a binary forced choice variable of which candidate you would vote for. Lines represent confidence intervals. Respondents are nationally representative sample of Americans, whites only. N = 1,399. Effect sizes for race treatments: Asian = −0.029 (0.013); Black = −0.021 (0.013); Hispanic/Latino = −0.036 (0.013).

These comparative results are especially noteworthy given the consistency of the direction of effects. For the sample of white Americans, all nonwhite racial candidates received lower average vote shares than white candidates. For the sample of white Canadians, only Asian candidates received a lower average vote share than white candidates, but this penalty was both marginal in magnitude and smaller than the penalty that Asian candidates received from white Americans.

Following the approach advocated by Leeper et al. (Reference Leeper, Hobolt and Tilley2020), the Appendix presents the results of a subgroup analysis of the conjoint experiment. I investigate the marginal means of each candidate attribute among white Americans and Canadians conditioned by a respondent’s levels of racial resentment or white identity. First, I split each of the American and Canadian samples of white respondents into those who were above or below the mean of the racial resentment index and estimate the marginal means of each attribute level (see figures 4A and Figure 6A in the Appendix). Next, I do the same for white identity, splitting the American and Canadian samples of white respondents into those above or below the mean of the white identity index measure and estimating the marginal means of each candidate attribute (see Figure 5A and 7A in the Appendix). This subgroup analysis shows how the effects of a candidate’s race are highly responsive to the degree to which a respondent is racially resentful or identifies as white. White candidates perform markedly better among those who are high in racial resentment or high in white identity. Nonwhite candidates perform better among those who are lower in racial resentment or white identity. Accordingly, the lower average levels of racial resentment and white identity among white Canadians relative to white Americans uncovered in the previous section of this article is a key reason why nonwhite candidates perform better in the aggregate among white Canadian voters.

Importantly, while this subgroup analysis shows how the effects of a candidate’s race are conditioned by the underlying racial attitudes and identities of American and Canadian respondents, there are also differences between the populations. For white Americans with low levels of racial resentment or white identity, candidates randomly assigned to be white fare about the same as nonwhite candidates. Among the sample of white Americans with high levels of racial resentment or white identity, white candidates perform much better than nonwhite candidates. However, for white Canadians with low levels of racial resentment or white identity, nonwhite candidates outperform white candidates in the conjoint experiment. For instance, among the sample of white Canadians with below average levels of racial resentment, white candidates (marginal mean = 0.4403) performed more than 5 percentage points worse than Asian candidates (marginal mean = 0.4948), almost 9 percentage points worse than Indigenous candidates (marginal mean = 0.5289) and almost 10 percentage points worse than Black candidates (marginal mean = 0.5363). Among the sample of white Canadians with below average levels of white identity, white candidates (marginal mean = 0.4546) performed 2.4 percentage points worse than Asian candidates (marginal mean = 0.4786), more than 6 percentage points worse than Indigenous candidates (marginal mean = 0.5193) and 9 percentage points worse than Black candidates (marginal mean = 0.5470).

This provides some necessary intuition behind the results of the experiment among the total samples of white Americans and Canadians. For white Americans, nonwhite candidates perform worse than white candidates in part because respondents with higher levels of racial resentment and white identity strongly favour white candidates. For white Canadians, there is no net-negative effect for nonwhite candidates due in part to respondents with lower levels of racial resentment and white identity favouring nonwhite candidates, as well as the lower prevalence of racial resentment and white identity among white Canadians uncovered in the previous section.

Electoral Behaviour: White ID, Racial Resentment and Nonwhite Feeling Thermometers

To further investigate the comparative electoral importance of race among white Americans and Canadians, this section reports the results of multiple regression analyses estimating the political effects of white racial identity, racial resentment and feelings towards nonwhite people. While measures of white identity, racial resentment and feeling thermometer scores are complimentary and often correlated, I analyze these independent variables separately because they are distinct theoretical constructs and to better pinpoint differences between white Canadians and Americans across these measures. The key outcome variables for this set of analyses are past vote choice, future vote intention and a scaled index measure of right-wing partisan identity (five-point intensity scale).

A large selection of control variables is included in the analysis to limit any concern of omitted variable bias, yet for conciseness, only the estimates associated with white racial identity, racial resentment and feelings towards nonwhite people are shown below. The full regression models with the corresponding estimates for each control variable are shown in Appendix 2. These include standard demographic variables, such as age, income, gender, union membership and religious importance, as well as variables measuring a respondent’s level of national identity and their level of support for free market principles versus government intervention in the economy and society (termed “laissez-faire attitudes”).Footnote ²³ For the purpose of comparability, I merge the Canadian and American survey data and estimate regression models with identical specifications and identically constructed variables.Footnote ²⁴ The key difference in the models is an indicator variable conditioning whether respondents are American or Canadian. There are, of course, important limitations to cross-sectional observational analysis (Lenz, Reference Lenz2012). Hence, while the variables included in the regression analyses theoretically precede vote choice and partisanship (Miller and Shanks Reference Miller and Shanks1996), I avoid making causal claims or inferences pertaining to causal order.

Beginning with the effects of white identity, I estimate logit models for past vote choice in either the 2021 Canadian federal election or the 2020 American presidential election, as well as vote intention for the next Canadian federal election or American presidential election. These outcome variables are coded 0–1, where 0 indicates a past vote or future vote intention for a non-right-wing party and 1 indicates a past vote or future vote intention for a right-wing party (Republicans for Americans, Conservatives or People’s Party of Canada for Canadians). I also estimate ordinary least squares (OLS) models for right-wing party identity (Republican ID for Americans, Conservative ID for Canadians), which is measured on a five-point intensity scale and recoded between 0–1 for ease of analysis. Robust standard errors are employed to protect against heteroskedasticity. Figure 5 displays the average marginal effects of the white racial identity index derived from each of the six models among white Americans or white Canadians.

Figure 5. White ID and Electoral Behaviour.

Average marginal effects for White Racial Identity index derived from six regression models (full controls included, reported in the Appendix); confidence intervals denoted with lines. Nationally representative samples of white Canadians and white Americans. Size of estimated effects, with sample size from each population in brackets: 1) Right-Wing Vote 2020, US Sample = 0.188 (n = 1,092); 2) Right-Wing Vote 2021, Canadian Sample = 0.094 (n = 1,277); 3) Right-Wing Vote Intention, US Sample = 0.240 (n = 1,144); 4) Right-Wing Vote Intention, Canadian Sample = 0.171 (n = 1,390); 5) Right-Wing Party ID, US Sample = 0.198 (n = 1,357); 6) Right-Wing Party ID, Canadian Sample = 0.095 (n = 1,282).

Figure 5 shows that white racial identity is positively correlated with vote choice and party identity in each of the models. For white Canadians, white identity is statistically significant in the right-wing vote intention model and the right-wing party ID model. A one-unit increase in white racial identity translates to a 17 per cent increase in the likelihood a respondent intends to vote for the Conservative Party or the People’s Party (PPC) in the next federal election, and a 9.5 per cent increase in Conservative Party ID. For white American respondents, there is a consistently large, statistically significant effect of white racial identity across each model. A one unit increase on the white racial identity scale corresponds to a respondent being 19 per cent more likely to have voted for Trump in 2020 and 24 per cent more likely to state they would vote for a Republican for president in 2024. A one-unit increase in white racial identity translates to a respondent being almost 20 per cent higher in Republican Party ID.

Figure 6 moves to the regression models for the effects of racial resentment. The design of these models is identical to Figure 5, but with racial resentment as the main independent variable instead of white identity. An indicator variable is again used in the models to distinguish between Canadian and American respondents. I derived average marginal effects from each of the regression models for ease of interpretation.

Figure 6. Racial Resentment and Electoral Behaviour.

Average marginal effects for Racial Resentment index derived from six regression models (full controls included, reported in the Appendix); confidence intervals denoted with lines. Nationally representative sample of white Canadians and white Americans. Size of estimated effects, with sample size from each population in brackets: 1) Right-Wing Vote 2020, US Sample = 0.554 (n = 1,090); 2) Right-Wing Vote 2021, Canadian Sample = 0.275 (n = 1,280); 3) Right-Wing Vote Intention, US Sample = 0.532 (n = 1,143); 4) Right-Wing Vote Intention, Canadian Sample = 0.297 (n = 1,393); 5) Right-Wing Party ID, US Sample = 0.198 (n = 1,357); 6) Right-Wing Party ID, Canadian Sample = 0.188 (n = 1,285).

Figure 6 shows that racial resentment has a consistently stronger effect across each of the models relative to white identity and is statistically significant in each of the models of vote choice and party identity. Specifically, a one-unit increase on the racial resentment scale translates to white Canadians being 27 per cent more likely to have voted for the Conservatives or the PPC in the 2021 Canadian Federal Election, 30 per cent more likely to intend to vote Conservative or PPC in the next election and a 19 per cent increase in Conservative Party ID. The effect sizes for the American portion of the merged data are even greater in magnitude. For white Americans, a one-unit increase on the racial resentment scale translates to respondents being over 55 per cent more likely to have voted for Trump in 2020, 53 per cent more likely to state that they would vote for a Republican for President in 2024, and an almost 20 per cent increase in Republican Party ID.

Evidently, racial resentment and white racial identity are strong correlates of right-wing voting and party identity in both Canada and the United States. Crucially, however, the two preceding figures show that the magnitude of the average marginal effects of white racial identity and racial resentment are lower in each model for the Canadian sample compared to the American sample. The effect size of white racial identity, across each of the models for past vote, future vote intention and right-wing party identity, is at least 7 percentage points higher among white Americans relative to white Canadians. For racial resentment, the differences in effect size between the cross-national samples are even greater on the first two regression models for voting behaviour (over 23 percentage points higher for white Americans). The exception is the models for party identity, where racial resentment has only a slightly greater effect size among white Americans relative to white Canadians. Consequently, white racial identity and racial resentment have larger effects for white Americans, but both variables are important correlates of right-wing voting and partisanship among white Canadians.

To conclude this section, I also estimate regression models with feelings towards nonwhite people as the key independent variable. As stated above, this variable is measured as the average feeling thermometer score provided by respondents for each nonwhite racial group that the survey included. The setup of the regression models is otherwise the same as the prior models, but with the variables for average feelings towards nonwhite people in place of racial resentment or white identity. The cross-national samples again consist only of white respondents.

Figure 7 plots the average marginal effects derived from the regression models among the samples of white Canadians and Americans. As above, the models for past vote choice and future vote intention are estimated with logistic regression. The models for right-wing party ID are estimated with OLS. We expect to find that the warmer individuals feel towards nonwhite people is negatively associated with voting for and identifying with right-wing parties. Figure 7 shows that this is the direction of the relationship. However, unlike the models for racial resentment or white identity, these effects are almost entirely not statistically significant for both American and Canadian respondents. Only the models for right-wing vote intention reach the conventional level of statistical significance and the size of the coefficients are modest. Therefore, while racial affect is an important attitudinal measure, it has a weaker correlation with electoral behaviour compared to racial resentment and white identity. These results suggest how important it is for researchers to go beyond feeling thermometer scores and extend their analysis of racial attitudes to include other measures.

Figure 7. Nonwhite Feeling Thermometer and Electoral Behaviour.

Average marginal effects for Nonwhite Racial Feeling Thermometer derived from six regression models (full controls included, reported in the Appendix); confidence intervals denoted with lines. Nationally representative samples of white Canadians and white Americans. Size of estimated effects, with sample size of each population in brackets: 1) Right-Wing Vote 2020, US Sample = -0.057 (n = 1,085); 2) Right-Wing Vote 2021, Canadian Sample = −0.097 (n = 1,253); 3) Right-Wing Vote Intention, US Sample = −0.118 (n = 1,138); 4) Right-Wing Vote Intention, Canadian Sample = −0.128 (n = 1,364); 5) Right-Wing Party ID, US Sample = −0.020 (n = 1,353); 6) Right-Wing Party ID, Canadian Sample = −0.054 (n = 1,259).