Calls for solidarity and political co-operation between racial and ethnic minority groups have been made by various public figures and activists, generally on the left, for many years. Perhaps most famously, Jesse Jackson called for a “rainbow coalition” at the 1984 US Democratic Party Convention. Such a coalition is usually defined as a political coalition of multiple non-white racial groups. A rainbow coalition, moreover, typically implies mutual co-operation and support of members, rather than just a coincidental coalition of groups voting for the same party or candidate.
Academic studies, however, have generally been skeptical about rainbow coalitions. By far the most widely studied inter-minority relationship is between Black and Latino Americans, and while examples of co-operation do exist (Browning et al., Reference Browning, Marshall and Tabb1984), as Kaufmann (Reference Kaufmann2007: 80) puts it, Black-Latino coalitions are “the exception and not the rule.” Indeed, there is often considerable hostility and prejudice (McClain et al., Reference McClain, Carter, DeFrancesco Soto, Lyle, Grynaviski, Nunnally, Scotto, Kendrick, Lackey and Cotton2006). On the other hand, initial analysis of the 2008 and 2012 US presidential elections seemed to suggest that such a coalition might be in the offing. As is well known, African-Americans' support for Barack Obama was nearly unanimous in these elections. Less well known is the fact that Latino voters also strongly supported Obama (Pew Center, 2012) and, in the 2012 general election, a record 72 per cent of Asian-Americans did the same (Goldmacher, Reference Goldmacher2012). Clearly, the ethnicity of voters and candidates can play an important role in elections, but the extent to which these effects occur in other circumstances is unclear. Is this a phenomenon that might emerge elsewhere, and if so, under what conditions?
Activist and academic advocates of “rainbow coalitions” make the claim that minority communities often share similar experiences and needs, and argue that these should provide motivation for political co-operation. Some adopt a critical approach to identities, as in Spivak's notion of “strategic essentialism” (Reference Spivak1987), while others express similar ideas in the more prosaic language of policy gains and electoral calculi (Meier and Stewart, Reference Meier and Stewart1991). Interests arising from shared experiences might manifest in concerns related to racism and discrimination, immigration, recognition of foreign credentials, as well as economic inequality and social exclusion, among others. Certainly the rhetoric of political elites, such as Jesse Jackson, focuses on pragmatic coalitions for ideological reasons.
Nonetheless, given standard findings of low levels of knowledge about party and candidate platforms (see Converse, Reference Converse and Apter1964), and the surprisingly small effects of self-interest on citizen choices (see Sears and Funk, Reference Sears and Funk1990), the electoral success of such coalitions may rest more on the structure and influence of social identities than strategic actions over issues of common concern. That is, the likelihood of minority voters supporting minority candidates—a crucial part of the success of any rainbow coalition—may rest largely on the effects of identity. To that end, this study focuses on the behaviour of minority citizens, rather than elites, and on the role of identity, rather than strategic self-interest.
The plausibility of rainbow coalitions, so far as they rest on identities and not policy, hinges on how ethnocultural identities influence voters and how those identities are understood. But how are these identities defined and which identities matter? One possibility is that generalized racialized status matters: racialized voters may be more likely to support racialized candidates, whether or not a specific ethnocultural identity is shared.Footnote 1 On the other hand, racialized citizens may only show a preference for their own ethnocultural background and not for other minority ethnicities, in which case there will be little identity basis for a rainbow coalition.
The demographics of racialized Canadians present an especially interesting case for examining the plausibility of rainbow coalitions. As the racialized proportion of the Canadian population continues to increase in number and political importance, we should expect a growing number of racialized candidates and party leaders. However, while Canada's racialized population is large, it is also very diverse: nearly one in five Canadians is classified as belonging to a visible minority, but the largest sub-group, South Asians, comprises only 25 per cent of the total visible minority population (Statistics Canada, 2011). Does this diversity imply division?
If conflict rather than co-operation among ethnic minorities is the norm, there are a number of important implications. Most obviously, coalitions on shared issues will be much harder to build. In addition, while racialized candidates are more likely to be nominated in diverse areas (Tossutti and Najem, Reference Tossutti and Najem2002), their ethnocultural community is likely to constitute only a plurality of the population in these areas. Consequently, if racialized voters prefer white candidates to candidates from racialized communities other than their own, then this may have a systematically negative impact on candidates from all racialized communities. Perhaps most importantly, mutually negative group attitudes between ethnic minorities would make leadership of major parties by racialized candidates much more difficult, since a racialized leader would suffer from both the racism of (some) white Canadians and antipathy from other ethnic minority groups.
This study draws on data from a web-based survey of racialized Canadians. The analysis centres on the results of an experiment that aimed to manipulate the perceived ethnicity of a candidate by varying that candidate's name between stereotypically European (or “white”), Chinese, and South Asian alternatives. The article discusses existing research that suggests the prospects for rainbow coalitions are slim and that social and political conflict between minority ethnic groups is the norm. In an important departure from the existing literature, the possible role of strength of group identity as a moderator of affinity effects is also discussed. The experiment, data and results are then set out, including regression analyses addressed to specific hypotheses. Here, the focus is on a) the overall influence of ethnocultural background and b) the conditional influence of ethnocultural background, given respondents' strength of group identity. The results show that while racialized respondents are most likely to support a candidate of their own ethnocultural group, at least some racialized voters are more likely to support racialized candidates of a different ethnicity over white candidates. Strength of identification with one's ethnocultural group appears to be critical to the process. In particular, the affinity effects uncovered here are substantially larger for respondents expressing a strong sense of identification with their group.
1. Candidates and the Influence of Social Group Membership
Inferring a candidate's membership in a social group—a simple and easily obtainable piece of information when compared to forming inferences about policy commitments or performance in office—allows citizens to bring their opinions about social groups or political values to bear on the voting decision. When voters and candidates belong to the same social group and this relationship increases support, this is often referred to as “affinity.” This includes voting for candidates of a shared gender (Brians, Reference Brians2005; Goodyear-Grant and Croskill, Reference Goodyear-Grant and Croskill2011) and shared race, among black Americans (Philpot and Walton, Reference Philpot and Walton2007; Sigelman and Sigelman, Reference Sigelman and Sigelman1982) and Latino-Americans (Bareto, Reference Barreto2007; Stokes-Brown, Reference Stokes-Brown2006). A considerable number of findings confirm the political influence of ethno-cultural background and that voters are more likely to support candidates from their own social group.
There are two major categories of theoretical explanations for affinity effects: interest-based heuristic accounts, and identity-based accounts. Heuristics are decision rules or mental shortcuts to make choices with less information or effort and, as Cutler (Reference Cutler2002) suggests, candidate demographics are the “simplest shortcut.” Candidate demographics may allow voters to infer policy positions, issue competency or make self-interested judgments about which candidate will help their group the most. These kinds of accounts are by far the most common explanations for candidate demographic heuristics in general (Cutler, Reference Cutler2002; Popkin, Reference Popkin1991), and racial affinity effects in particular (McDermott, Reference McDermott1998; Philpot and Walton, Reference Philpot and Walton2007; Stokes-Brown, Reference Stokes-Brown2006).
However, a second possible explanation for affinity effects—identity based effects—has received little attention. Social identity theory (Tajfel, Reference Tajfel1981) argues that evaluations of fellow group members can have a direct impact on self-esteem, suggesting that citizens may be more likely to want a candidate belonging to their social group to win because it affects their own sense of self. In addition, such “ingroup bias” tends to make people rate their own group higher on evaluations in general, and we would expect these group attributions to be reflected in judgments about candidates. Naturally, positive candidate attributions increase the likelihood of support and negative attributions reduce it (Bittner, Reference Bittner2011; Stokes, Reference Stokes1966). Also, there may be indirect effects reflecting the psychology of persuasion. The effectiveness of messages depends on the credibility of their source (Iyengar and Valentino, Reference Iyengar, Valentino, Lupia, McCubbins and Popkin2000), and if the voter belongs to the same social group as the candidate, the persuasiveness of campaign messages may be increased. Given its neglect in existing research dispite these strong theoretical foundations, this study examines the role of identity in affinity voting.
2. Which Identity?
While the logic of affinity voting appears simple, its manifestation depends on the construction of group identities. Since group identities can exist along multiple dimensions and at multiple levels of specificity, whether a candidate is seen as part of the voter's ingroup might depend on how the ingroup is defined. Given the history of race as a polarized concept in the United States and other Western countries, the difference between white and non-white may be the most important group cleavage in these societies, especially when the electoral decision is constructed as a choice between a white and a racialized candidate. Similarly, we might think that, because racialized citizens may suffer from generic forms of discrimination or have common interests in immigration and other policies, these citizens will see members of other racialized groups as potential allies in a shared political struggle. Thus, one possibility is that ethnic minorities will see themselves as a common “racialized” identity group, distinct from white Canadians.
Alternatively, racialized citizens might see members of other ethnic minority groups as outgroup members. This is certainly implied by findings from Europe, and the United States which often show a preference for whites over other racialized minorities and political conflict between minority groups. The most explicitly political research on inter-minority relations has been conducted in the United States, with the general finding being conflict (Kaufmann, Reference Kaufmann2007). Among other circumstances, this includes conflict in urban politics (McClain and Karnig, Reference McClain and Karnig1990), school district elections (Meier and Stewart, Reference Meier and Stewart1991) and general negative stereotypes (McClain et al., Reference McClain, Carter, DeFrancesco Soto, Lyle, Grynaviski, Nunnally, Scotto, Kendrick, Lackey and Cotton2006). Related social attitudes research comes to a similar conclusion: that minority groups prefer to maintain greater social distance from each other than they do from the dominant majority. In the Netherlands, for example, Turkish immigrants preferred to associate with ethnically Dutch people as compared to Surinamese immigrants, and vice versa (Hagendoorn, Reference Hagedoorn1995).
In Canada, while there is little evidence of direct conflict, existing research does point to a preference for association with white Canadians over other minority ethnic groups. A study of social attitudes by Kalin and Berry (Reference Kalin and Berry1996) found that minorities tend to rate whites higher than other minority communities on a whole range of positive attributes, such as honesty and being well educated. This research also indicates that minority groups view themselves as “more similar” to white Anglophones than to other minorities. This is the same pattern of preference for the majority found in an earlier study which examined white British Canadians relative to Canadians of other European origins, such as German-Canadians, Ukrainian-Canadians, and so forth (Berry and Kalin, Reference Berry and Kalin1979). The only study in Canada to focus on inter-minority relations in electoral politics is a paper on municipal council elections for the Welcoming Communities project (Bird, Reference Bird2011). Subjects evaluated a single candidate with experimentally manipulated ethnicity, with one same-ethnicity pair (South Asian candidate, South Asian respondents) and one different-ethnicity pair (South Asian candidate, other racialized respondents).The results indicated that South Asian respondents were more likely to support the South Asian candidate, but other racialized respondents preferred the white candidate.
While existing research in Canada is consistent with findings of inter-minority conflict elsewhere, the evidence is clearly limited. Moreover, on many measures associated with inter-minority conflict, such as prejudice and economic marginalization (Gay, Reference Gay2006), Canada fares better than the United States and many European countries (OECD, 2012; Schleicher, Reference Schleicher2006). While the relationship between white/minority conflict and inter-minority conflict is unclear, broadly speaking there appears to be less racial conflict and tension in Canada than in many countries. Thus, there are some reasons to think that there will be at least less inter-minority conflict in Canada then than elsewhere.
In addition, the Common Intergroup Identity Model (Gaerdner et al., Reference Gaertner, Rust, Dovidio, Bachman and Anastasio1993) suggests that even if ethnic group identities are more important generally, the particular circumstances of a vote choice might result in a broader racialized group identity becoming more salient. Gaertner and colleagues (Reference Gaertner, Rust, Dovidio, Bachman and Anastasio1993) suggest that discrimination can be mitigated by re-categorisation, that is, membership in a super-ordinate group generalizes ingroup bias to people who would otherwise be outgroup members. Notably, this has been applied to ethnic group boundaries (Gaertner and Dovidio, Reference Gaertner and Dovidio2000). It might well be that framing a choice between a white and racialized candidate would make a more generalized racialized identity more salient and therefore the effects of ingroup bias would apply to the racialized different-ethnicity candidate. In sum, we should expect ingroup bias for members of the same ethnic group, but while the predominance of the evidence suggests inter-minority conflict, the expectations for candidates of different ethnic minority groups are less certain.
3. Strength of Ethnic Group Identity
The concept at the centre of this inquiry is ethnocultural identity. That is, we expect participants to support the candidate who shares with them a common group identity. At the same time, it is likely that individuals vary in the extent to which identification with a particular group is personally important to them. Research on ingroup bias suggests that stronger ethnic identities lead to more ingroup bias (Pfeifer et al., Reference Pfeifer, Ruble, Bachman, Alvarez, Cameron and Fuligni2007). In this case, I suggest, if the voter only weakly self-identifies with the ethnic group, then the performance of a candidate of that group is unlikely to affect the voter's self-esteem. Conversely, a voter with a strong ethnic identity is more likely to support a candidate of their ethnic group, because the success or failure of the candidate is tied to their own self-esteem, and their common group identity affects their perceptions of that candidate.
Although its use in political science has been limited, the measurement of self-identification has played an important role in sociology, social psychology, and organizational behaviour studies. One important measure is the identification with a psychological group scale—IDPG (Mael and Tetrick Reference Mael and Tetrick1992)—though, to date, the use of this measures in political science has been limited to party identity (Greene, Reference Greene1999, Reference Greene2004). The fruitful application of the IDPG scale to socio-political identities such as partisanship suggests that it may be very useful in evaluating how strength of identity moderates the effects of ethnocultural membership.
To summarize, political research from the United States and sociological and attitudinal research from Europe suggest that the key identity categories for affinity effects are specific ethnocultural groups, not racialized status in general. While there is reason to expect less inter-minority conflict in Canada, what evidence that does exist still shows a preference for affiliation with the white majority rather than other racialized ethnic groups. Affinity effects, therefore, should occur only in the presence of a shared ethnocultural background. In other words, rainbow coalitions seem unlikely. Furthermore, affinity effects should be related to the strength of group identity: identity-related considerations—whether coalitional or conflictual—should be more influential for those with stronger identities.
4. Hypotheses and Method
Based on previous research, this study tests five hypotheses. The first focuses on the rainbow coalition thesis, that, in general, racialized citizens are more likely to support racialized candidates than other candidates. The second and third hypotheses examine more specific affinity effects for same-ethnicity candidates, and different-ethnicity candidates.
H1:
Racialized voters will be more likely to choose a racialized candidate than a white candidate.
H2:
Racialized voters will be more likely to choose a candidate of their own ethnocultural background than a white candidate.
H3:
Racialized voters will be less likely to choose a candidate of a different racialized ethnicity than a white candidate.
The final two hypotheses repeat hypotheses two and three but suggest the effects will interact with the strength of ethnic group identity. Stronger identification with the ethnocultural group should increase these effects, whatever their direction.
H4:
Racialized voters who have a stronger ethnocultural identity will be more likely to choose a candidate of the same ethnocultural background than those with a weaker ethnocultural identity.
H5:
Racialized voters who have a stronger ethnocultural identity will be less likely to choose a candidate of a different racialized ethnicity than those with a weaker ethnocultural identity.
To test these hypotheses, a survey experiment was conducted. Participants were presented with short biographies of two candidates and asked which they would vote for. One benefit of using a choice between two candidates, rather than asking for an evaluation of a single person as some studies have done, is that this design presents a more realistic test for the question at hand. The two-candidate design replicates the structure of a real election where citizens are asked to choose between multiple options rather than to evaluate a single candidate. In addition, by allowing respondents to choose from alternative options, the two-candidate design used in this study does not require a “negative” action—explicit rejection of a candidate—which might be more likely to trigger socially desirable behaviour.
The candidate biographies themselves were not manipulated, and they were written to present more or less equally qualified candidates to be members of Parliament, balancing the need to have plausibly different candidates while not introducing factors that might interact with the treatments. The candidates were “independent,” rather than having party labels, since research suggests that party cues may suppress the effect of other heuristics (Kam, Reference Kam2007). No pictures were used, both to avoid confounding with the effects of candidate attractiveness and to reflect the reality of Canadian elections at the local level, where most voters see signs with the local candidate's name without a picture.Footnote 2 Finally, the order of presentation of the candidates' biographies was randomized across subjects.
One notable feature of the biographies is that each contained a “fault,” that is, some less-than-ideal (from the perspective of electoral politics) personal feature was attributed to each candidate. Since discrimination is often evident only when there is an “excuse” for such behaviour (such as “covering,” Crandall and Eshleman, Reference Crandall and Eshleman2003). When evaluating a “perfect” candidate, even highly prejudiced people might not discriminate because there is no socially acceptable justification available. Thus, the faults in the biographies provide both an external social justification and an internal psychological justification for participants who might have discriminatory tendencies. For reasons of plausibility these are different faults, specifically, candidate 1 is described as having been laid off twice, before going on to found his own company. Candidate 2, on the other hand, is described as having run and lost in a previous election. The candidate biographies themselves were not manipulated and, as a result, it is possible that there could be some interaction with non-randomly assigned variables of the respondent (such as ethnicity) or of the biography (type of fault, novelty of the candidate ethnicity). More generally, random assignment ensures a lack of bias among groups but not between groups (that is, between South Asian and Chinese respondents). Thus, caution should be taken in comparing affinity effects between groups, though none of the formal hypotheses that follow does so.
The primary treatment was a manipulation of candidate ethnicity. The control group was presented with two candidates, both of whom had stereotypically European names: John Hawkes (candidate 1) and Arthur Dorre (candidate 2). The two treatment groups were presented with the same two candidate biographies, but candidate 2 was given either a traditionally Chinese (Jun Zhang) or South Asian (Satveer Chaudhary) name. The names were chosen from a list of common Chinese and South Asian names in Ontario health care records (see Shah et al., Reference Shah, Chiu, Amin, Ramani, Sadry and Tu2010). Several of these were selected and discussed with members of the respective communities, as well as other researchers with relevant experience. Names that were the same as Canadian politicians or candidates were excluded. Those that had highly specific religious associations (such as Singh) or that might also be European (such as Lee) were also avoided.
Table 1 illustrates the factor structure and reports cell sizes by ethnic self-identification (candidate ethnicity was randomly assigned). The complete stimulus and candidate-preference question read as follows, with manipulated sections bolded:
Candidate 1
John Hawkes is an entrepreneur, and after being laid off twice he started the successful company, Allsort Inc. Despite a busy schedule Mr. Hawkes works with a number of organizations, including Kids Help Phone, and served as Vice Chair of the Municipal Safety Committee. John Hawkes is an independent.
Candidate 2
Arthur Dorre/Jun Zhang/Satveer Chaudhary is an active local businessman, who was recently honoured as “Businessman of the Year” for his many contributions. Mr. Dorre/Zhang/Chaudhary helps at the local community centre, and is the fundraising chair for the Hospital Foundation. A former provincial candidate, he lost in the most recent election. Arthur Dorre/Jun Zhang/Satveer Chaudhary is an independent.
Which candidate would you vote for?
Table 1 Factor Structure and Assignment
Note: Cell entries are frequencies.
To measure the strength of group identity, the survey used a modified set of questions from the identification battery, the IDPG scale (Mael and Tetrick Reference Mael and Tetrick1992) introduced earlier. The scale consists of a number of statements and a five-point response scale running from “strongly agree” to “strongly disagree.” Three statements were used in this survey.
When someone criticizes (group) people, it feels like a personal insult.
When I talk about (group) people I usually say “we” rather than “they.”
When someone praises (group) people it feels like a personal compliment.
The category selected in response to the Statistics Canada ethnicity question defined what was inserted into the question at group. So, for instance, the participant might read: “When someone criticises South Asian people, it feels like a personal insult.” To ensure comparability across groups, relatively broad ethnic identifications, rather than highly specific nationalities, were utilized. Specifically, Japanese, Chinese and Korean respondents were asked about “Asian” people and Filipino respondents were asked about “South-East Asian” people. The other categories provided by Statistics Canada, noted above, could be used without modification (that is, Arab, Latin American, and so forth). Responses to the three strength of identification items were combined into a 15-point additive index and scaled to the (0, 1) interval. To avoid priming group identity considerations, these questions were asked after the candidate experiment. It is possible that the racialized candidate treatment would affect the strength of identity questions, but in this case the difference between treatment groups on the mean value of the strength of identity scale is very small (.011) and statistically insignificant.
The sample is drawn from an opt-in web-panel, a somewhat new development in both mode and sample method, which have costs and benefits. There have been a number of comparisons between web panels and more traditional methods, and one important finding has been that opt-in web panels are generally reliable for examining relationships between variables, but less so for estimating population parameters (Sanders et al., 2007, Stephenson and Crête, Reference Stephenson and Crête2011). This means, for example, that we should be more confident in estimates of the relationship between ethnic self-identity and affinity voting than the estimates of the absolute level of ethnic self-identification. In addition, the online mode tends to have lower social desirability bias than telephone surveys (Holbrook and Krosnick, Reference Holbrook and Krosnick2010), an important benefit in studying race and ethnicity.
The crucial question is whether the sample differs from the population in ways that are likely to affect the analysis. A comparison of various demographic variables (age, income, education, gender and immigrant status) to the Canadian Election Study (CES) and National Household Survey (NHS) data shows that, at least on these variables, the sample is broadly representative. Moreover, deviations from the NHS are generally no greater than the difference between the NHS and the CES, although education is higher in the sample (see online appendix). One notable point is that there are more Chinese Canadians than the NHS estimates. The reasons for this are not clear and so care should be taken in comparing between ethnic groups, though none of the following hypotheses does so. Finally, the web sample offers considerable benefits; the web-survey enables longer questions, such as the candidate biographies used here, which would not be possible in a traditional telephone survey. Perhaps most importantly, it allows the relatively cost effective recruitment of a large, national, sample of racialized Canadians; crucially, this enables the analysis of specific ethnic groups.
The survey was conducted with Canadian residents, in English, with no respondents from Quebec. The sample was drawn from a web panel, selected using a demographic screening survey. Ethnicity was measured using a self-definition question: “Statistics Canada defines some Canadians as ‘visible minorities.’ Do you consider yourself to belong primarily to any of the following visible minority groups?” The categories used by Statistics Canada were provided, including Arab, Black, Chinese, Filipino, Japanese, Korean, Latin American, South Asian, South-East Asian and West Asian. Participants could also respond that they were Aboriginal or did not belong to any visible minority group (no aboriginal-identifying respondents are included in the analysis). The analysis conducted here uses 711 racialized respondents, including 305 subjects self-identified as having Chinese origins, 141 as having South Asian origins and 265 as having other visible minority backgrounds.
5. Data and Analysis
To test these hypotheses, logistic regression models were constructed using support for candidate 2 as the dependant variable, that is, support for the candidate whose name was manipulated. The first hypothesis is that racialized respondents in general will be more likely to support racialized candidates. In model 1, the independent variables are candidate racialized status (that is, the experimental treatment level) and participant racialized status, and an interaction between these two variables. Demographic control variables are also included (income, education, gender, age, religion—Christian, non-Christian, and non-religious—and immigrant status), which increases the precision with which certain model parameters can be estimated, and improves generalizability. The predicted probabilities reported below were generated using the Margins command in Stata.Footnote 3
Figure 1 compares the predicted probability of racialized respondents voting for candidate 2 under two conditions: a white candidate 2 and a racialized candidate 2. This allows us to consider the possibility of a rainbow coalition, broadly defined, that racialized respondents are more likely to support a racialized candidate than a white candidate. The results show that racialized respondents are 7 percentage points more likely to support a racialized candidate than an identical white candidate. The difference between the predicted probabilities falls just short of the conventional 95 per cent confidence level (p = .07). Contrary to most existing research on inter-minority conflict, we find affinity among racialized respondents in general; in other words, this is evidence for the plausibility of rainbow coalitions.
Figure 1 Racialized Affinity
To test the second hypothesis concerning affinity along specific ethnocultural lines, support for the white version of candidate 2 needs to be compared to support for candidate 2 when his ethnicity “matches” that of respondents. In model 2, candidate and participant ethnicity are broken down into dummy variables for specific ethnicities: for candidate ethnicity these are white, Chinese and South Asian; for participant ethnicity the categories are white, Chinese, South Asian and other minorities. The dependent variable remains vote for candidate 2. Predicted probabilities are generated for each combination of participant/candidate ethnicity and the first difference of the experimental manipulation is taken. This is the treatment effect, the difference in support between the white and minority versions of the candidates. Results are shown in Figure 2. As the first two bars show, both “same ethnicity” groups show evidence of affinity effects: South Asian-identifying participants are on average 17 percentage points more likely to vote for a South Asian candidate than for an otherwise identical white candidate (p = .09), while Chinese-identifying participants are 8 percentage points more likely to support an apparently Chinese candidate (p = .23). This partially confirms the second hypothesis; while there is some evidence for affinity effects among South Asian identifying respondents, given the statistical significance the evidence for affinity effects among Chinese respondents is equivocal.
Figure 2 Affinity for Ethnic Groups
The third hypothesis is about relations between different minority ethnocultural groups: whether participants are more likely to support a white candidate over a candidate of a different minority group. There are four groups of respondents who have the opportunity to support a different minority candidate: Chinese respondents and the South Asian candidate; South Asian respondents and the Chinese candidate; other minority respondents and the South Asian candidate; and other minority respondents and the Chinese candidate, again shown in Figure 2. Contrary to expectations, support is higher for the candidate of a different ethnocultural minority background than when the candidate is white. This suggests that racialized citizens have an affinity for racialized candidates generally, rather than just for candidates of their own specific ethnocultural group. While the effects are not significant, the point estimates of all combinations of ethnic groups are positive. The findings are certainly inconsistent with the third hypothesis, if not decisively demonstrating affinity between different ethnocultural groups.
The final two hypotheses are restatements of the previous two, with the addition of an interaction with strength of identity. The fourth hypothesis states that stronger ethnic group identity will strengthen affinity along ethnocultural lines. In this third model, the dependent variable is, again, support for candidate 2, with dummy variables for Chinese and South Asian versions of candidate 2, and for Chinese, South Asian, and other racialized respondent identification. Demographic controls are included (income, education, gender, age, religion and immigrant status). To evaluate the moderating effect of strength of ethnocultural identity, a three-way interaction between the strength of identity scale, participant ethnicity and candidate ethnicity is introduced, along with its constituent terms. Since the measurement of “white” identity is quite different conceptually, non-racialized respondents are dropped from the model. Then, predicted probabilities for candidate support are generated at various levels of strength of group identification.
Figure 3 compares the support for white and minority versions of candidate 2 and clearly shows that the affinity effects are considerably stronger for participants with stronger ethnocultural group identities, confirming hypothesis 4. Both Chinese and South Asian affinity effects are large and highly significant at high levels of strength of identity. Chinese respondents with high levels of ethnic group identification are much more likely to vote for the Chinese candidate than the white candidate: 27 percentage points at .75 on the strength of ID scale (p = .004), and 47 percentage points at the top of the strength of ID scale (p < .001). South Asian respondents show similar effects. At .75 on the strength of ethnic identification scale, South Asian respondents are 24 percentage points (p = .03) more likely to choose the South Asian candidate than the white candidate, and at the top of the strength of identification scale, South Asian respondents are 38 percentage points (p = .02) more likely to choose the South Asian candidate than the white candidate. Notably, these are not extreme values: .75 on the strength of ethnic identification scale is the 75th percentile, meaning that 25 per cent of the sample scored .75 or higher. A large portion of the sample seems to show high levels of ethnic affinity voting.
Figure 3 Affinity for Same Ethnic Group, by Strength of Identification
By contrast, both Chinese and South Asian at the 25th percentile of the strength of ethnic identification scale (.5), showed no significant affinity voting. At the bottom of the scale, respondents actually showed discrimination against candidates of their own ethnic group: Chinese respondents scoring .25 on the strength of identification scale are 32 percentage points less likely to select the Chinese candidate (p = .03). While the point estimates of low ethnic identifying South Asians are also negative, they are not statistically significant. However, these scores on the strength of ethnic identification, .25 and below, only compose 5 per cent of the sample. Thus it seems dis-identification with one's own ethnic group, and the corresponding rejection of same-ethnicity candidates, seems uncommon. Far more common, at least in this sample, are respondents who seem indifferent about their ethnic self-identification and who show neither affinity nor discrimination.
These results are a clear confirmation of the fourth hypothesis: participants who identify strongly with their ethnocultural group are more likely to vote for a candidate of their ethnocultural group, but those who identify only weakly with their ethnocultural group are not. These effects are large and highly significant statistically. Importantly, this shows that affinity effects apply broadly rather than being limited to any particular ethnic group; recall that without strength of identification (Figure 2) the results were not statistically significant, especially for Chinese respondents. Here, however, there is strong evidence for affinity voting among candidates and voters of the same ethnicity, for both Chinese and South Asian identifiers.
The fifth hypothesis states that participants who have a stronger ethnic group identity will be less likely to choose a candidate of a different ethnocultural background than those with a weaker ethnic group identity. To test this hypothesis, predicted probabilities are generated for the white and minority versions of candidate 2 for each of the four combinations of different-minority ethnicity participant and candidate pairs. Figure 4 presents the difference between the white and minority candidate—effectively, the size of the treatment effect—at different levels of strength of ethnic identification.
Figure 4 Affinity for Different Ethnic Groups, by Strength of Identification
The results are not consistent with hypothesis 5; in most cases, it seems that stronger ethnic identification makes racialized respondents more likely to support the racialized candidate, not less. Chinese respondents at the top of the strength of identity scale and are 29 percentage points more likely to support a South Asian candidate than a white candidate (p = .05), but at.75 on the strength of identity scale the effect is not significant (p = .16). Conversely, Chinese respondents at the bottom of the strength of identification scale are 39 percentage points less to support a South Asian candidate (p = . 05). South Asian respondents at the top of the strength of identity scale are 22 percentage points more likely to support a Chinese candidate, but the effect is not statistically significant (p = .2), nor are those at lower levels of strength of identity. None of the predicted probabilities for the “other racialized respondents” category is significant. Overall, these results suggest cross-minority affinity voting effects, although only effects for Chinese respondents were statistically significant and only at the top of the strength of identity scale. That said, the general lack of inter-minority discrimination provides decisive evidence against the hypothesis in question.
Finally, in an effort to test hypothesis 5 with more precise estimates, Model 4 groups all racialized respondents and candidates together, but drops those cases where respondents treated with a candidate of their own ethnicity. This allows predicted probabilities to be generated that compare racialized respondents with the white version of candidate 2, to those with a racialized candidate 2 who was of a different ethnicity than themselves. Effectively, this merges the four different-ethnicity respondent/candidate pairs in Figure 4. The independent variables are a white or racialized candidate 2, strength of ethnic identification, an interaction term, and demographic controls. The predicted probabilities in Figure 5 are encouraging. Racialized respondents at the top of the strength of identification scale were 17 percentage points more likely to support a different minority ethnicity candidate (p = .05), and those at.75 on the strength of identification measure are 8 percentage points more likely to support a different minority ethnicity candidate than a white candidate, (p = .1). In terms of proportion of the sample, 33 per cent of the sample show affinity effects at a 90 per cent confidence interval (.75 or above on the IDPG scale), but only 5 per cent at a 95 per cent confidence interval (1 on the IDPG scale). While these results clearly show evidence for affinity effects among the relatively small number of strong ethnic identifiers, the more broad-based effect of moderate identifiers is less certain. Nonetheless, these results provide stronger support for different-ethnicity affinity voting than the previous analysis of single ethnic groups. Racialized citizens who strongly identify with their own ethnic group are more likely to support candidates of other racialized ethnic groups, although the effects are smaller—roughly half the size—than those for same-ethnicity affinity voting.
Figure 5 Different-Ethnicity by Strength of Identification
6. Discussion and Conclusions
This paper explores the role that different kinds of ethnocultural identity play in affinity effects, and begins with the expectation that there will be affinity between candidates and voters of the same ethnocultural group but a preference for white candidates over candidates from other ethnocultural minorities. While the existence of ethnocultural affinity is confirmed, there is no evidence of discrimination between minority groups. In fact, affinity effects seem to extend to racialized candidates in general.
Of the five hypotheses, hypotheses 1, 2 and 4 were confirmed, and hypotheses 3 and 5 were not. The most decisive evidence was for hypothesis 4 and against hypothesis 3: same-ethnicity affinity effects were large and highly significant, and there is no evidence of inter-minority conflict. These findings suggest four broad conclusions. First, there is no discrimination among racialized respondents against other ethnocultural minorities, that is, there is no preference for white candidates over candidates of a different racialized background. Second, there are affinity effects for candidates of the same ethnocultural group. Third, there appears to be at least some affinity for candidates of different ethnic minority groups. Finally, the likelihood of affinity is positively correlated with the strength of ethnic affinity; respondents who identify strongly with their ethnocultural group are much more likely to support a racialized candidate, either of their own ethnicity or of a different ethnic minority group, than those who identify only weakly.
The first hypothesis is the simplest, but also focuses on the crux of the political issue: Do minority candidates on average attract less support from racialized citizens? Here, racialized respondents are 7 percentage points more likely to support a racialized candidate than an otherwise identical white candidate (p = .07). If we are looking for evidence of the plausibility of a “rainbow coalition,” this is a good start.
Conversely, unlike in other countries, there is no evidence that inter-minority relations in Canada are characterized by conflict. Hypotheses 3 and 5 look specifically at willingness to support other ethnic minority groups, and are both rejected: racialized respondents do not discriminate against candidates of different minority backgrounds. These effects are clearly concentrated among those who have a stronger identification with their ethnocultural group; as Figure 4 suggests and Model 4 confirms, stronger ethnocultural identity increases the likelihood of supporting the minority candidate, even when the candidate is of a different ethnic minority group. The breadth of different-ethnicity affinity is less clear; while only effects at the top of the identity scale are statistically significant at a 95 per cent interval, a much larger proportion of the sample showed effects at a 90 per cent confidence interval. What this means in electoral terms is harder to establish; even a 5 per cent swing among racialized citizens would be politically consequential given their location in battleground ridings (Marwah et al., Reference Marwah, Triadafilopoulos, White, Rayside and Farney2013). Unfortunately, while experiments are well suited to establishing causal mechanisms, they are less useful for estimating the size of these effects in real elections. The size and extent of affinity voting, particularly for candidates of a different minority group, would benefit from additional research.
Nonetheless, participants clearly differentiate between their own and other ethnocultural minorities; support for the candidate of the same ethnocultural group is still higher and much more statistically significant. For example, Chinese respondents at the top of the strength of identification scale are 38 percentage points more likely to support a Chinese candidate but only 23 percentage points more likely to support a South Asian candidate. Overall, this evidence suggests weaker affinity effects across ethnocultural lines, but certainly no evidence of inter-minority discrimination.
Clearly, affinity effects are strongly conditional on the strength of identification measure. “Objectively” belonging to a demographic category is not enough, it is self-identity that really matters here. This is an important consideration for research on the behaviour effects of ethnicity. The strength of identification also helps clarify the general applicability of affinity effects. Figure 3 shows no significant affinity for the Chinese candidate by Chinese-identifying participants. It might seem that affinity effects are limited to South Asian participants, a group that, in Canada, is well known as exceptional in terms of political participation (Bird, Reference Bird2005). As noted earlier, we should interpret between-group comparisons cautiously, given that experimental randomization does not control for the effect of the many possible differences between groups, such as cultural values or time of arrival in Canada. Nonetheless, it is notable that once we have modelled the conditioning effect of strength of ethnic group identification, the predicted probabilities show that at the 75th percentile of the strength of identification scale, Chinese respondents and South Asian respondents show essentially the same level of ethnic affinity voting. In addition, none of the predicted probabilities for other racialized respondents was significant. This could be a measurement issue: for these respondents, the strength of ethnic identification included questions referencing West Asians and Southeast Asians, terms which may have less resonance for social identities, an important area for future research.
As already mentioned, these findings of cross-ethnicity affinity and a lack of inter-minority discrimination are a striking contrast to previous research in other countries. Why might this be? One possibility is the relatively low levels of racial conflict in Canadian society. The (relatively) high levels of economic and educational achievement of immigrants in Canada (OECD, 2012, Schleicher, Reference Schleicher2006) and the broad acceptance of immigration and multiculturalism in the broader population (Focus Canada, 2011), may increase the plausibility of rainbow coalitions. A second potential explanation is that immigrants tend to adopt the values of the communities into which they integrate (Bilodeau et al., Reference Bilodeau, White and Nevitte2010). Since there seems to be little explicit discrimination against racialized political candidates by white Canadians (at least, on average) (Black and Erickson, Reference Black and Erickson2006), this behaviour may be adopted by racialized immigrant Canadians. In addition this study did not include a group that is clearly lower in social status. Research suggests that there is usually a widely accepted “consensual hierarchy” of ethnic groups (Hagendoorn, Reference Hagedoorn1995). Inter-minority conflict may, therefore, be partially a result of some groups trying to “distance” themselves from those low in the hierarchy. It is possible that neither Chinese nor South Asians are, in the eyes of one another, a clearly lower-status group. The respondents in this study, therefore, may not have felt a need to socially distance themselves from each other. While the first two potential explanations suggest that lower levels of racialized conflict and lower levels of discrimination are linked to greater likelihood of rainbow coalitions, this third explanation suggests that the composition of the coalition itself, and relationships between specific groups, may be crucial. Finally, the Common Inter-group Identity Model points to the importance of context defining the ingroup in more inclusive ways (Gaerdner et al., Reference Gaertner, Rust, Dovidio, Bachman and Anastasio1993). Here, the choice set may be responsible for affinity between ethnic minority groups. Even if racialized citizens generally identify in terms of their own ethnic group, being presented with a choice between a white and a racialized candidate may results in an increase in the salience of super-ordinate racialized identity group, leading to including candidates of other racialized ethnic groups in their perceived ingroup.
Finally, we might consider how broadly these dynamics might apply. In other words, is the situation used in the experiment, where respondents chose between one visible minority and a white candidate (or candidates), common in Canadian elections? Although visible minorities are underrepresented as candidates compared to the proportion of the population they represent, their overall numbers have been increasing (Black, Reference Black2009). In the 2011 Federal election, there were some 54 ridings (of 308) that had a minority candidate competing against white candidates (analysis by author). Clearly, these effects may apply to a considerable part of the political landscape.
According to the evidence here, the notion that minority candidates primarily appeal to their own ethnic community while white candidates are, in some sense, “neutral” and thus more successful in a highly diverse areas, is not just objectionable but also empirically incorrect. Certainly in the United States, ethnic heterogeneity has often benefited white candidates (see Meier, Reference Meier and Stewart1991). Yet these findings suggest that, at least in Canada, political parties would do well to nominate more minority candidates in ethnically diverse areas, regardless of their particular ethnic make-up.
Despite hopes and aspirations for rainbow coalitions among different minority groups, most research has drawn pessimistic conclusions. This suggests an inability to present a united front on common policy concerns, and negative consequences for racialized candidates. However, the data examined here present quite a different picture. The first preference is clearly one's own ethnocultural group but there appears to be no corresponding rejection of other minorities. Importantly, this result holds even at higher levels of ethnic identification, and in fact stronger identification with one's own ethnocultural minority group increases the likelihood of supporting a candidate from a different racialized group. This, if nothing else, suggests that the prospects for rainbow coalitions may be considerably brighter than previous research would lead us to expect.
