Numerous observers have noted increasing opinion polarization in the US. There is some debate as to whether polarization is occurring among partisans alone or among the population at large (see Abramowitz, Reference Abramowitz2010; Fiorina, Reference Fiorina, Shea and Fiorina2013); but where partisans are concerned, the consensus question is “not whether, but how much?” (Fiorina and Abrams, Reference Fiorina and Abrams2008: 578). One of the key components of this process has been a surge in “partisan sorting,” that is, the extent to which like-minded citizens have increasingly clustered into different parties. Centrist Democrats and Republicans appear to have become more and more rare, and there seems to be less and less overlap between the two groups. This dynamic is observed not just in academic work; it is a regular feature of public discussion as well.
Is Canada any different than its increasingly divided neighbour to the south? Despite common perceptions of a relatively liberal consensus, are there increasingly stark divisions between supporters of different Canadian parties? These questions are clearly relevant for scholars of Canadian politics, and they are of particular significance in the wake of Donald Trump's election as US president. “Could Donald Trump happen here?” has become a frequent refrain in Canadian media outlets (for example, Gillis, Reference Gillis2016; Levitz, Reference Levitz2017), with the question particularly pertinent in light of the 2017 Conservative Party of Canada leadership race. And perhaps spurred by polls suggesting that just over three-quarters of Canadians would consider voting for a candidate with a Trump-like platform (Russell, Reference Russell2016), several of the leadership candidates clearly attempted to take up that mantle; Kellie Leitch, who famously proposed a “Canadian values test” for potential immigrants, greeted Trump's election as “an exciting message and one that we need delivered in Canada as well” (Graham, Reference Graham2016); while Kevin O'Leary, who led the pack in favourability polls before dropping out to support his libertarian-leaning rival Maxime Bernier, is a reality TV star, business man and political outsider that many likened to Trump (for example, Levitz, Reference Levitz2017). Given recent marked changes in party support, first in 2011 and then in 2015, yet another dramatic electoral shift seems firmly within the realm of the possible. Questions about the nature and extent of partisan polarization in Canada have therefore become especially relevant.
The analyses that follow have important implications for the academic literature as well—a literature which is as-yet unclear about whether increased partisan sorting, particularly over the past ten years, has been a uniquely American phenomenon. Indeed, a small body of work on polarization in several European countries has suggested that US trends have not been reflected elsewhere, at least as of the early 2000s (for example, Adams et al., Reference Adams, Green and Milazzo2012a; Adams et al., Reference Adams, De Vries and Leiter2012b). Canadian political institutions, including a multi-party system, may put it more in line with some European countries. But Canada's cultural, political and economic proximity to the US, not to mention exposure to American media, make it a particularly interesting case. Indeed, there is some preliminary evidence, published in the Washington Post's Monkey Cage, to suggest that partisan polarization has indeed been occurring in Canada (Johnston, Reference Johnston2014; discussed further below).
Partisan sorting matters not just because it speaks to the nature of political preferences and the party system but also because it has the potential to disconnect voters from policy makers (Hacker and Pierson, Reference Hacker and Pierson2005; Kirkland, Reference Kirkland2014). Scholars of Canadian politics have nevertheless paid relatively little attention to this topic. We suspect that this has been motivated in part by a Canadian perception—acurate or not—that American politics is more dysfunctional than Canadian politics. We also suspect that current concerns in Canada have been lessened somewhat by the election of the Liberals in 2015, which is particularly striking when contrasted to the election of Donald Trump in 2016. Yet, as we show below, the 2015 election results are not the product of reduced partisan divisions. This paper thus sets out to correct this misperception, partly as an exploration of the Canadian political landscape, but also as a test of the generalizability of a trend that has been so prominent in recent studies of the US. We aim, in short, to assess the extent to which Canada has experienced increased partisan sorting around policy preferences.
We do so by focusing on preferences for redistributive policy. Our interest in policy preferences is driven in part by the recognition that public preferences play an important role in shaping policy outcomes, both generally and with regard to redistributive policy in particular (Brooks and Manza, Reference Brooks and Manza2007; Soroka and Wlezien, Reference Soroka and Wlezien2010), alongside recent work (discussed below) suggesting that partisan opinion polarization can be a complicating factor in this relationship. But redistribution is also a salient and long-standing policy domain with a relatively straightforward relationship to the left-right ideological spectrum. We intend for our results to speak both to the state of redistributive policy preferences and to the status of partisan sorting on policy issues more generally.
We explore partisan sorting vis-à-vis policy preferences using data drawn from the new 2015 Canadian Election Study, alongside similar studies for every election back to 1992. We focus separately on (a) Canada outside of Quebec, and (b) Quebec, given the province's distinct political cleavages and party system, with the pro-sovereigntist Bloc Québécois attracting a large proportion of votes since the early 1990s (for further discussion, see Nadeau and Bélanger, Reference Nadeau, Bélanger, Kanji, Bilodeau and Scotto2012). In doing so, we follow the established current practice in Canadian election studies (see, for example, Fournier et al., Reference Fournier, Cutler, Soroka, Stolle and Bélanger2013; Medeiros and Noël, Reference Medeiros and Noël2014). And we pay particular attention to the last two elections, each of which involved major changes in party support, as we explore the relationship between Canadians’ redistributive preferences and party identification since the 1990s.
Polarization South of the Border
There is a burgeoning literature on polarization in the US, portraying politicians and party identifiers as increasingly divided in their positions and posturing. We would characterize this research field roughly as follows: Congressional Democrats have become more liberal while Republicans have become more conservative (see Fiorina and Abrams, Reference Fiorina and Abrams2008; Lee, Reference Lee2009; Poole, Reference Poole2007); and this development, referred to in the literature as “elite polarization,” has pulled public policy away from the preferences of the general public (see Bonica et al., Reference Bonica, McCarty, Poole and Rosenthal2013; McCarty et al., Reference McCarty, Poole and Rosenthal2006).
Underlying this argument about a growing disconnect between citizens and their representatives is the assumption that elite polarization has not been reflected in mass polarization (that is, across society as a whole). Indeed, for both Republican and Democratic representatives, elite polarization appears to be disconnected from the collective preferences of citizens, whether at the district or the national level (see, for example, Bafumi and Herron, Reference Bafumi and Herron2010; Levendusky et al., Reference Levendusky, Pope and Jackman2008). Yet it would be a mistake to conclude that polarization has taken place only among elites; while the citizenry as a whole might not be polarizing, there is an abundance of evidence to suggest that citizens who hold partisan attachments to the Democratic and Republican parties are becoming increasingly distinct (see, for example, Fiorina, Reference Fiorina, Shea and Fiorina2013; Hetherington, Reference Hetherington2009; Lelkes, Reference Lelkes2016).
The precise drivers of partisan polarization have been much debated. On the one hand, some scholars point to the impact that elite polarization may have on partisan preferences, with increasingly distinct party elites helping to create increasingly distinct partisans (for example, Druckman et al., Reference Druckman, Peterson and Slothuus2013; Zingher and Flynn, Reference Zingher and Flynn2016). A key part of this top-down process has been a greater degree of mass partisanship (see Carmines et al., Reference Carmines, Ensley and Wagner2012; Lupu, Reference Lupu2015), driven by the easier partisan sorting of voters: distinct party positions make it simpler for (politically interested) citizens to figure out which party is closest to them, while at the same time making rival parties more unpalatable (Davis and Dunaway, Reference Davis and Dunaway2016; Hetherington, Reference Hetherington2001). This is reflected, for example, in studies suggesting that even independents, at least in their vote choice, have sorted more and more consistently over time (Smidt, Reference Smidt2017).
Other scholars, however, have highlighted the possibility that existing partisans may simply have changed their preferences, perhaps even for reasons unconnected to elite polarization. While there is no consensus on the exact drivers of such a process, researchers have pointed to various explanatory factors, including increased income inequality (Garand, Reference Garand2010); increasingly partisan media and media consumption habits, whether via television (Prior, Reference Prior2013) or the internet (Lee et al., Reference Lee, Choi, Kim and Kim2014); and mistaken beliefs about the extremism of political opponents (Ahler, Reference Ahler2014). As a consequence, partisan opinion polarization—that is, increasingly consistent preferences among party supporters and growing gaps between supporters of different parties—may in fact occur without any substantial changes in voters’ partisan allegiances.
Note that these discussions are reflected in debates over how best to label these developments. Certain authors prefer the term “partisan sorting” rather than “partisan polarization” (for example, Abrams and Fiorina, Reference Abrams, Fiorina, Thurber and Yoshinaka2016; Levendusky, Reference Levendusky2009b), especially given that party supporters have not clustered around the poles (as implied by the term polarization). Yet even these authors tend to find a shift in partisan preferences away from the centre (for example, Levendusky, Reference Levendusky2009b: 6). We can thus distinguish between two separate but related processes: partisan sorting, wherein partisans correctly identify and align themselves with the party that best matches their preferences, thereby creating more uniform preference sets among a given party's supporters; and partisan opinion polarization, wherein the preferences of partisans shift away from the centre (though not necessarily to the ideological extremes). In practice, however, evidence from the US suggests that these changes have typically gone hand in hand (see Bafumi and Shapiro, Reference Bafumi and Shapiro2009; Levendusky, Reference Levendusky2009a).
In light of these developments, the literature suggests that alongside the growing disconnect between politicians and the public as a whole there is a closer connection between partisan identifiers’ parties and preferences. Researchers have found evidence of this on a wide range of issues, such as abortion (Adams, Reference Adams1997), foreign policy (Shapiro and Bloch-Elkon, Reference Shapiro, Bloch-Elkon, Halperin, Laurenti, Rundlet and Boyer2007), the environment (McCright et al., Reference McCright, Xiao and Dunlap2014) and Israel (Cavari, Reference Cavari2013). And while this effect may be limited to “hot-button issues” (Baldassarri and Gelman, Reference Baldassarri and Gelman2008), the attitudinal changes seem clear.
These developments are interesting not only in and of themselves, but also for their consequences. At the societal level, some accounts suggest that partisan sorting and opinion polarization have fostered animosity that goes well beyond what one would expect on the basis of the actual extent of disagreement (Iyengar and Westwood, Reference Iyengar and Westwood2015; Mason, Reference Mason2015); the result, as Mason puts it, is “a nation that may agree on many things but is bitterly divided nonetheless” (2015: 142). At the political level, partisan polarization appears to negatively impact opinion representation (see, for example, Hacker and Pierson, Reference Hacker and Pierson2005), with legislators less likely to represent median voter preferences when they represent electoral districts in which preferences are more polarized (see Kirkland, Reference Kirkland2014). Add to these effects the importance of public opinion for government policy, and it becomes clear that partisan sorting and opinion polarization are important subjects of study outside of the US as well.
Polarization in Canada
Are these American developments reflected in Canada? The tendency in the literature comparing public opinion in Canada and the US has been to emphasize difference (for example, Adams, Reference Adams1998, Reference Adams2003; Lipset, Reference Lipset1990; though see Nevitte, Reference Nevitte1996, and Banting et al., Reference Banting, Hoberg and Simeon1997, for exceptions to this rule). There is nevertheless a good deal of similarity on a wide range of attitudes on both sides of the border. Broadly cross-national work makes this clear, and recent research has found greater similarity than difference on a range of attitudes, even those related to policies on which the two countries are thought to differ, such as health care (Nadeau et al., Reference Nadeau, Bélanger, Pétry, Soroka and Maioni2015) and immigration (Harell et al., Reference Harell, Soroka, Iyengar and Valentino2012).
A similar conclusion can be drawn regarding the nature of partisanship in Canada and the US. Although much of the early scholarship on partisanship in Canada described it as far more unstable than American partisanship (for example, Elkins, Reference Elkins1978; LeDuc et al., Reference LeDuc, Clarke, Jenson and Pammett1984; though see also Sniderman et al., Reference Sniderman, Forbes and Melzer1974), subsequent work has challenged the methodological underpinnings of that conclusion (see Blais et al., Reference Blais, Gidengil, Nadeau and Nevitte2001; Johnston, Reference Johnston1992). Many authors have highlighted the importance of partisanship for vote choice (for example, Clarke and McCutcheon, Reference Clarke and McCutcheon2009; Medeiros and Noël, Reference Medeiros and Noël2014; Nevitte et al., Reference Nevitte, Blais, Gidengil and Nadeau2000), suggesting that early contrasts with the US may have been exaggerated (for further discussion, see Anderson and Stephenson, Reference Anderson, Stephenson, Anderson and Stephenson2010: 20–21). The extent to which Canadian partisans take information shortcuts from their parties is less well-established: the only evidence comes from two experimental studies conducted in the mid-2000s, which suggest that while cue taking does occur, it is limited to more politically sophisticated partisans and/or supporters of the NDP (Merolla et al., Reference Merolla, Stephenson and Zechmeister2008, Reference Merolla, Stephenson, Zechmeister, Blais, Laslier and Van der Straeten2016). Overall, however, existing research points to broad similarities in the nature of partisanship in the US and Canada. In a recent direct comparison of partisanship in these two countries by Bélanger and Stephenson, for example, the authors conclude that “holding a party identification in Canada has the same implications for developing partisan preferences in favour of one's own party and away from other parties as it does in the United States” (Reference Bélanger, Stephenson, Turgeon, Papillon, Wallner and White2014: 117).
The literature on party polarization in Canada suggests further reasons to suspect similarities with the US. Despite the fact that Canadian parties have historically been ideologically incoherent (Clarke et al., Reference Clarke, Jenson, LeDuc and Pammett1996), recent decades have witnessed shifts in the party system that have increased the importance of ideology (see, for example, Carty et al., Reference Carty, Cross and Young2000). Both party manifesto data and expert assessments of parties point toward meaningful and expanding ideological distances between parties, especially since the 1980s (Cochrane, Reference Cochrane2010; Klingemann et al., Reference Klingemann, Volkens, Bara, Budge and McDonald2006). Within a comparative perspective, Kim and colleagues (Reference Kim, Powell and Fording2010) describe Canada as having a modestly polarized party system—though their analysis ends in the 1990s—while Dalton's measure (2008) of polarization points to a trajectory of party polarization that broadly mirrors that in the US over the 1960–2011 period (see Han, Reference Han2015: 583). Moreover, these between-party differences appear to be reflected in the policy preferences of the various party memberships, with evidence of consistency within parties and substantial differences across them (Cross and Young, Reference Cross and Young2002).
Is this pattern also reflected in partisan opinion polarization? Most existing work on Canadian politics touches only tangentially on the issue: a variety of studies point to the existence of regional divisions in voting patterns and political preferences (Anderson, Reference Anderson2010; Gidengil et al., Reference Gidengil, Blais, Nadeau and Nevitte1999; Ornstein and Stevenson, Reference Ornstein and Stevenson1999). Walks (Reference Walks2005) suggests that there is additional low-level geographical polarization between city and suburban voters in Canada; Johnston (Reference Johnston2008) concludes that party supporters are more polarized on the left-right scale outside of Quebec than in it; and Wesley (Reference Wesley2009) finds evidence in Manitoba of polarized opinions on various issues among both elites and party identifiers. Yet even in instances where research suggests opinion polarization (rather than simple cleavages), it neither tracks trends over time nor permits a comparison with other countries.
There are a few, albeit limited, exceptions. Dragojlovic and Einsiedel (Reference Dragojlovic and Einsiedel2014) compare public attitudes in the US and Canada, but they focus exclusively on opinions toward biofuels, finding less opinion polarization in Canada than the US. Pinard and Hamilton (Reference Pinard and Hamilton1977) and Smiley (Reference Smiley1978) find that the independence issue polarized opinion in Quebec in the 1960s and 70s, but the contemporary relevance of this conclusion is questionable. Indeed, it is telling that the most suggestive study is based on survey data from over 30 years ago. Comparing university undergraduates in the US and Canada, Gibbins and Nevitte (Reference Gibbins and Nevitte1985) conclude that Americans—regardless of whether or not they are party identifiers—are more polarized than their Canadian counterparts, while Anglophone Canadians are more polarized than Francophones.
The most relevant recent test comes from preliminary work by Richard Johnston (2014), which suggests that trends in party polarization and partisan sorting in Canada may well reflect those in the US. Johnston's work (2014) is to our knowledge the first systematic attempt to explore this dynamic in the Canadian context. Based on 11-point left/right placements, Johnston finds little evidence of widening gaps in the ideological positions of partisans since 2004. But the number of Conservative and NDP identifiers—parties on the right and left of the political spectrum—has increased. This marks a significant development, leading Johnston to conclude that “Canadians’ ideological locations are probably better sorted by party than they were 30 years ago, especially on the right.” The end result is an electorate that, as in the US, appears to be more clearly divided across partisan lines.
If this is indeed the case, Canada would stand apart from most other (non-US) cases. Studies looking at the Netherlands (Adams et al., Reference Adams, Green and Milazzo2012a) and the UK (Adams et al., Reference Adams, De Vries and Leiter2012b) find that parties there have depolarized, that partisan sorting has decreased, and that the public on the whole is less polarized than before. Similar research on Germany finds that issue polarization has decreased within the German public (Munzert and Bauer, Reference Munzert and Bauer2013). Yet only the last of these studies looks beyond the early 2000s (and none go beyond 2010), so it is impossible to say whether other countries simply lagged behind the US. An analysis of Canadian data up to 2015 thus offers both an additional case for study and the ability to investigate more recent non-US trends.
Methods
Note that we focus here on partisan sorting, that is, the degree to which citizens’ attitudes on policy issues are clustered according to partisanship. This includes an investigation of sorting both across partisan identification and current vote choice, where the latter speaks to opinion trends that extend beyond those with strong, durable attachments to parties.
We carry out this analysis using all Canadian Election Studies (CES) from 1992 to the present, including the new 2015 Canadian Election Study. We rely on a merged dataset, pulling together demographic, voting and policy preference questions since 1992. Demographic and voting questions are of course relatively simple; merging policy questions is rather more complex, since question wording and/or response categories invariably change in small (and sometimes large) ways over time. The analysis that follows relies on CES data exactly as they are distributed via the Canadian Opinion Research Archive; and a script to replicate recoding, merging and analyzing the data is available at the Harvard Dataverse.
We focus in particular on trends in three questions on redistributive policy, selected in part because they are asked in very similar ways over the past several decades.
Standard of Living (from the post-election mail-back wave)
The government should: See to it that everyone has a decent standard of living; Leave people to get ahead on their own; Not sure.
Reduce the Gap (from the post-election telephone wave)
How much do you think should be done to reduce the gap between the rich and the poor in Canada? Much more, somewhat more, about the same as now, somewhat less, or much less?
Get Ahead (from the post-election telephone wave)
Do you strongly agree, somewhat agree, somewhat disagree, or strongly disagree with the following statements?…People who don't get ahead should blame themselves, not the system.
These questions are not exclusive to the CES; they have been used regularly in other surveys as well. They each capture somewhat different elements in support for redistributive policy; each also has some limitations for what we are trying to do here. Standard of Living is a measure of absolute preferences, that is, preferences that are not measured in relation to current policy levels. It is asked in a perfectly consistent way from 1992 to 2011; it is however not a part of the 2015 survey and so cannot speak to the most recent trends in partisan sorting. Reduce the Gap is in contrast a relative preference measure; it asks about doing more than is currently done, and in so doing should shift due to movement in both (or either) the public's preferred level of policy and actual policy levels. (For a discussion of absolute versus relative preferences, see Soroka and Wlezien, Reference Soroka and Wlezien2010.) It was asked consistently from 2000 to 2015; it was also asked in 1992 and 1997, though as a five-point agree-disagree question, so we combine the two five-point scales here. Get Ahead captures some combination of preferences for redistribution and attitudes towards the recipients of social assistance and unemployment. It is thus a less direct measure of redistributive preferences per se, though attitudes towards recipients may of course be an equally relevant a dimension of opinion polarization. Get Ahead is asked consistently from 1997 to 2015. (It is asked before that, but not with the same five response categories.)
These three measures of redistributive preferences are related, but also show a good deal of independent variance. Table 1 shows simple bivariate correlations between the measures across our entire sample.Footnote 1
Table 1 Bivariate Correlations between Measures of Redistributive Preferences
Based on unweighted data, 1992–2015, all provinces combined. All correlations are significant at p < .01.
We do not combine these measures into an index below but rather analyze each separately, looking at the extent to which variation can be explained by each of two measures of partisanship: respondents’ party identification and respondents’ vote, where we use the actual vote drawn from the post-election wave. Both partisanship and self-reported vote use identical wording throughout our analysis, respectively as follows: “In federal politics, do you usually think of yourself as a…”, with the major parties then listed alongside other/none; and “Which party did you vote for?”, with options for the major parties as well as “other party,” “spoiled ballot” and “don't know.”
In both cases, we allow the number of parties to vary from year to year, so that any and all changes in the party system (including the appearance and/or disappearance of Reform, the Alliance and the Green party) are accurately reflected in our results. We include “other/none” as a category in our estimates of the impact of party identification and vote; another option is to exclude these cases, but doing so makes no significant difference to the results. We also examine the relationship between each of our measures and respondents’ income tercile. This measure is based on more detailed income variables, but given that the response categories vary slightly from year to year, we use the distribution of income in each annual CES file to determine terciles. We include income mainly to provide a baseline with which to compare the impact of partisan variables. Given that income should matter to individuals’ preferences about redistribution, the degree to which partisanship matters more or less than income, and the extent to which that changes over time, may be a useful metric.
Analysis
We focus on a simple but telling analysis, with an eye on the degree to which preferences for redistribution are structured by respondents’ own income (that is, self-interest), partisanship, and time. We do so using analyses of variance (ANOVAs), which estimate, within each election study, the degree to which variation in each of our dependent variables is captured by (1) party identification, (2) vote, or (3) income tercile. ANOVAs are equivalent to OLS regression with categorical variables, where the only critical difference is in the reporting of results. We regard the omnibus test of partisanship in the ANOVA table as a major advantage here, in addition to the fact that ANOVA results provide information about how much variance is explained by each variable (categorical or otherwise). The other advantage of ANOVAs, rather than a simple comparison of mean redistributive preferences across parties, is that while the latter is straightforward in the two-party US system, it is less illuminating in the shifting, multiparty Canadian system. ANOVAs offer a simple approach, with a single measure of overall variance explained by all parties (no matter how many there are).Footnote 2
Note that all our independent variables are included as categorical variables, necessary for party ID and vote, and useful for income since it allows us to relax the assumption that effects are linear. Note also that we do not include all three variables simultaneously, since we are not interested in the impact of each independent variable controlling for the others but rather the degree to which variance in the dependent variables is related to party ID, vote or income (even as these variables overlap). All results are included in the appendix, where tables show the partial sum of squares for each variable, alongside the degrees of freedom, an F-test of statistical significance and the proportion of variance explained by each variable.
In each case, the proportions of variance explained by the entire model make clear the degree to which variation in policy attitudes is related to party ID, vote, or income tercile. That said, there is one ANOVA for every combination of (a) a dependent variable (Standard of Living, Reduce the Gap and Get Ahead), (b) an independent variable (party ID, vote and income), and (c) an election year. Appendix results are thus difficult to sift through, but the critical findings are illustrated in Figures 1 through 3. In each case, the top row shows the trend in opinion over time, and the bottom row shows the proportion of within-year variance that is accounted for by party ID, vote and income. Over-time trends in the bottom row give us a very clear sense of the extent to which there has been increasing partisan sorting over time. The left column shows results outside of Quebec and the right show results in Quebec alone.
Figure 1 Standard of Living, Over time
Figure 1 presents results for the Standard of Living variable. The upper panels indicate that support for government spending is higher in Quebec over the entire time period; an upward trend is more evident there than in the rest of Canada (ROC) as well. In both regions, however, there is clear majority support for a strong welfare state. (Of course, the figure offers information only about the mean, not the distribution around that mean. The latter changes relatively little given the use of variables with a limited number of response categories, however.)
How do responses to the Standard of Living variable line up with partisanship, vote and income over time? Note first that percentage of variance explained by any single measure never exceeds 10 per cent. There is of course a lot of individual-level variation not accounted for by these simple models; even so, where noisy survey data are concerned, explaining 10 per cent of the variance with a single categorical variable is relatively striking. There is thus clear evidence in this and subsequent figures that Canadians’ attidues towards redistributive policy are at times powerfully affected by income and partisanship.
How do income and partisanship compare? Note in Figure 1 that at no point in time do income terciles account for more variance than the partisanship variables. Partisanship is related to income, of course, so it should capture at least some income-related variation, alongside variation in other drivers of partisanship as well. But the gap between income and partisanship seems here to be particularly wide, especially for an attitude that we expect to be closely related to economic circumstance. On this general measure of support for the welfare state, it is clear that partisanship, measured either by identification or vote, matters more than income.
Moreover, partisanship accounts for a growing proportion of variation in attitudes over time. There is some noise in year-to-year variation; the trend from 1992 to 2011 is not a simple, straight line. Even so, by 2011 the relationship between Standard of Living and either vote or party ID is higher than in any previous year. This is clearest in the ROC; it is evident for vote in Quebec as well, though party ID there matters as much as in 1992, the election in which the Bloc Québécois swept francophone ridings in the province.
Are similar dynamics evident for Reduce the Gap and Get Ahead? For the most part, yes. Analysis of the Reduce the Gap variable in Figure 2 shows, again, somewhat higher support for the welfare state in Quebec vis-à-vis the ROC. It also suggests higher degrees of sorting—not just across partisan groups, but income groups as well—in 2015. (Note that the jump in the impact of income in 2015 is absent in Figure 1 not because income does not matter to Standard of Living attitudes, but because we lack 2015 data for that variable.) The same is true of Get Ahead in Figure 3, though here the differences in overall levels of support are roughly similar in the ROC and Quebec. In this instance we note an even more dramatic increase in the importance of income in 2015 than with Reduce the Gap, as the largest overall difference is not party ID or vote but income. There has been increased sorting on this question, to be sure; but the nature of that sorting points as much to income as to partisanship.
Figure 2 Reduce the Gap, Over time
Figure 3 Get Ahead, Over time
Note also that, across the three questions, the percentage of variance explained by party ID tends to lag behind that explained by vote choice, although this trend is less clear in Quebec (with its added sovereigntist dimension in politics) than the ROC. It may be that polarization is reflected in voting decisions before generating party identifications. This would be in keeping with research suggesting that party polarization gives rise to more consistent ideological voting and a gradual expansion of partisan identities (see, for example, Dalton, Reference Dalton2008; Lupu, Reference Lupu2015). Evidence from the US, for example, highlights that even voters with no self-declared partisan identity have become increasingly sorted in their voting behaviour (Smidt, Reference Smidt2017). Further research is of course required to determine whether this is, in fact, what is happening in Canada as well.
Even so, over time and across various measures of redistributive preferences we note a broad increase in the proportion of variance explained by party ID and vote choice. This suggests growing partisan sorting; like-minded citizens have increasingly clustered into parties. Before 2015, this trend pointed to a growing divergence between the proportion of variance explained by partisanship variables. The year 2015 saw a sharp increase in the importance of income as well, however. Redistributive preferences appear to be more closely aligned with economic self-interest than in the past, though it is unclear whether 2015 will serve in this regard as a watershed moment or an outlier. Either way, the general trend in the data is clear: redistributive preferences in Canada are increasingly divided across both economic and partisan lines.
Discussion
This paper has used data from Canadian Election Studies to examine trends in partisan sorting of redistributive preferences from 1992 to 2015. Results suggest that Canada has experienced a surge in partisan sorting, indeed, one that may be comparable to what has been identified in the US. This is marginally truer when we look at partisans by vote rather than party ID; it may be that changes in political parties are reflected in voting decisions before they are consolidated into party identifications. Regardless, the overall trend is clear.
We readily acknowledge that there is plenty of work left to do. This study offers only a starting point for analyses of both redistributive preferences and partisan polarization. Our intention here has not been to offer a definitive account. Rather, our aim has been to trace out, for what to our knowledge is the first time, trends in partisan attitudes up to the most recent election. ANOVAs offer only a partial glimpse of what is happening in Canadian politics. But our sense is that this glimpse is at odds with what some might believe. And our results point towards the importance of further work, modelling redistributive preferences more completely and accounting more fully for what we have shown is an increasing importance of partisanship and income over time.
A few caveats are in order. First, we have looked here only at partisan sorting vis-à-vis attitudes about redistribution. We regard redistribution as a central, salient policy domain, of course, but increased partisan sorting on this issue will most powerfuly condition Canadian electoral politics only if it is correlated with sorting on other policy preferences. If other policy preferences matter to voting behaviour and pull voters in various directions, sorting on redistributive preference may matter only a little. This is one area for further research.
Another is an effort to explore the relationship between the two (interrelated) phenomena of partisan polarization and sorting. Our findings suggest that, over the last several elections, party ID and vote choice have come to explain a growing amount of variance in redistributive preferences. That these political factors have often mattered more than pure self-interest (that is, income) is striking. Yet our present study does not allow us to examine partisan polarization directly. There are strong a priori reasons to believe that a combination of elite polarization and partisan sorting is a recipe for (at least modest) partisan polarization, as has been the case in the US (see, for example, Bafumi and Shapiro, Reference Bafumi and Shapiro2009; Levendusky, Reference Levendusky2009a), yet further analysis is nevertheless needed to sort out these effects. Panel data would provide one potential way forward, but the most recent CES panel data only cover the 2004–2008 period, thus preceding recent trends. Original survey data collection could therefore be especially fruitful for this question.
In the meantime, we note that our findings have interesting implications for students of Canadian politics. Party competition in Canada has historically been characterized not by competition between left(/labour)- and right-oriented parties but rather by strong, ideologically flexible centrist parties (and mostly, just one party), put in power by broad-based regional coalitions (see Carty et al., Reference Carty, Cross and Young2000; Johnston, Reference Johnston and Perlin1988). Yet the party system has been marked by increased polarization since the 1980s (see, for example, Cochrane, Reference Cochrane2010), and our analysis offers support for the conjecture that Canadian parties are currently more effective at capturing (and perhaps enhancing) ideological divisions in society. It also serves as a useful reminder that the 2015 election was not the product of Canadians coming together in their support for redistribution, bur rather the electoral success of a pro-redistribution plurality. Whether this distribution of preferences, and the way in which it is captured by the party system, will withstand increasingly polarized and/or populist elite politics remains to be seen. For now, this much is clear: increased partisan sorting, at least regarding attitudes about redistribution, is readily evident in Canada.
Appendix
Table A1 ANOVA Results, by year, for each DV-IV combination
