Voting Behaviour in Canada: The State of the Discipline

Abstract This review surveys the literature on vote choice in Canada. It highlights key findings regarding a variety of factors that influence Canadians’ vote choice, while also suggesting future avenues of research. The focus is on studies conducted on vote choice at the federal level, with an emphasis on studies that have been published since 2000.


Introduction
Elections are fundamental to our system of representative democracy, offering citizens the opportunity to authorize future policy directions and to hold governments accountable for past performance. Understanding what motivates vote choice and shapes electoral outcomes is vital for assessing the working of representative democracy in Canada. The Canadian literature can also throw light on how voters decide in other post-industrial democracies. So much of what is known about voting behaviour is based on the American experience, but researchers learned early on that concepts and theories developed in the United States do not always travel well. This is hardly surprising, especially given the nature of the party system and the rules governing elections. The concept of party identification is a case in point: a foundational concept in studies of American voting behaviour, its applicability in Canada and elsewhere has been contested (see below).
The literature on voting behaviour in Canada is vast. To keep this review manageable, it is limited to studies on vote choice at the federal level. 1 The review is organized around various factors that influence Canadians' choice of party. The ordering follows the logic of the bloc-recursive model. 2 Commonly used to explain vote choice in studies of particular elections Gidengil et al., 2012), this multistage model assumes that explanatory factors differ in their proximity to the vote. Longer-term factors, such as social background characteristics and ideological orientations, are assumed to influence shorter-term factors, such as issue attitudes and leader evaluations. What this approach lacks in parsimony, it makes up for in comprehensiveness (Roy and McGrane, 2015). This does not mean that every voter takes all of these factors into account, but these factors figure to varying degrees in every voter's decision calculus. Beginning with social background characteristics, this review covers the role of ideological orientations, party identification, economic evaluations, issues, leader evaluations, and local candidates, ending with strategic calculations as the most proximate factor.
While this review is comprehensive, it is not intended to be exhaustive. First, space constraints precluded a discussion of the large literature on voter turnout in Canada as well as studies of campaign dynamics and the impact of the media, polls and political information on vote intentions. Second, it was impossible to do justice to all of the work of the many scholars who have contributed to the study of Canadian voting behaviour. Instead, the review highlights some of their key contributions.
Social Background Characteristics: "A Cornucopia of Intriguing Anomalies" 3 Some scholars have questioned the importance of Canadians' social background characteristics for understanding their vote. When included in a vote model along with voters' partisanship, leader evaluations, issue attitudes and performance evaluations, these characteristics may appear not to matter, but such a model misses the role of social background characteristics in shaping those factors. As Blais (2005: 834) concluded, "We miss something important if we do not examine the group bases of party support." However, some findings have proved puzzling and difficult to explain.

Social class
One debate that long animated the study of vote choice in Canada concerned the apparent absence of class voting. Multiple studies tried to refute Alford's (1963) conclusion that Canada was a case of "pure non-class politics," attributing his findings to misclassification of parties' class positions and flawed measurement of voters' social class, but to little avail (Gidengil, 2002). The most plausible explanation points to a lack of class consciousness (Pammett, 1987), which has been variously attributed to the formation of the Canadian working class, the nature of Canada's unions, the lack of a "true" class party, and the electoral system. Interest in this question appears to have abated, but the changes wrought by globalization and automation suggest a need to revisit the influence of material interests (Milner, 2021). So, too, does the increasing link between income and redistributive preferences (Kevins and Soroka, 2018).

Region
The weakness of the class cleavage has also been linked to "the domination of Canada's cleavage structure by cultural forces" (Johnston, 2012: 155). Class is eclipsed by regional and religious cleavages . That region matters to vote choice in Canada seems like a truism. We only have to think of how much party vote shares vary across the country. Research on regional differences in vote choice has typically defined regions in terms of provinces or, more commonly, groups of provinces such as Atlantic Canada or the West. 4 A key finding is that these differences cannot be explained by differences in their social make-up: people sharing the same social background characteristics vote differently depending on where they live, and they do so because of differences in their vote calculus (Gidengil et al., 1999;Stephenson et al., 2019). In other words, voters in different parts of the country differ in how they weigh various considerations.

Type of community
Type of community also matters. Using riding-level data and a novel measure of urbanity, Armstrong et al. (2022) conclude that the rural-urban divide has assumed a new importance in recent elections, with the Liberals' urban advantage contributing to their electoral success. However, individual-level studies have underlined the importance of intra-urban divides, documenting how inner-city residents have been shifting left while residents of the suburbs and especially the outer suburbs of Canada's three largest urban regions have been moving right (Walks, 2004).
Meanwhile, little appears to distinguish suburbanites from residents of small cities and towns or from rural residents (McGrane et al., 2016;Turcotte, 2001). It remains to be explored whether the distinctiveness of inner-city residents reflects residential choices, material interests, different modes of consumption, social interactions or day-to-day experiences and cultural practices (Walks, 2004). The notion of "cultural scenes" (Silver and Miller, 2014) may be useful. Using information on local amenities from art galleries to body-piercing studios to infer "the lifestyle of a place and the values it affirms or opposes," Silver and Miller (2014: 425) show how parties' vote shares vary systematically depending on different dimensions of a riding's cultural scene.

Religion and religiosity
The persistence of the religious cleavage has posed a puzzle. Viewed as anomalous, it has been characterized as an "unwanted intruder" or a "moderately interesting, but strikingly peculiar, houseguest who has overstayed his welcome" (Irvine, 1974: 560). The propensity of Catholics to vote Liberal has been observed since the earliest studies of voting behaviour in Canada, and yet a convincing explanation has eluded scholars. It is not an artifact of demographic differences between Catholics and non-Catholics, and it has persisted in the absence of explicitly religious issues. Irvine (1974) suggested an ingenious explanation: parents pass on both their religious affiliations and their partisanship, even in the absence of any conscious connection between the two. However, the reproduction of any cleavage requires the "intervention of live social forces outside the family" (Johnston, 1985: 108). Johnston's (1985: 109) suggestion that there is "a distinctive Catholic ethos" has received only mixed support, and so has Bélanger and Eagles' (2006: 607) argument that the Catholic-Liberal connection depends on the presence of "other, likeminded Catholics" (Blais, 2005). It is not a matter of Catholics preferring a party led by a Catholic or voting for Catholic candidates. Nor is it a matter of issue preferences. Blais (2005) observed statistically significant differences between Catholics and non-Catholics on only nine of 30 issues, and on only one did the difference exceed 10 points. Tellingly, when Catholics were deserting the Liberal Party in 2004 and 2006, anger over the sponsorship scandal mattered much more than disagreement with the party's support of same-sex marriage (Stephenson, 2010;Gidengil et al., 2012). Two factors have been found to narrow the gap between Catholics and non-Catholics: political information (Bittner, 2007) and exposure to the news media (Mendelsohn and Nadeau, 1997).
When Reform emerged in 1993, moral issues assumed a new importance. The contours of the religious cleavage appeared to be shifting, with evangelical Protestants leaving the Progressive Conservatives in favour of the party's new rival on the right (Guth and Fraser, 2001). Then, when many Catholics began switching their allegiance in the wake of the Sponsorship scandal, scholars detected a new cleavage, with religious Canadians supporting the Conservative Party and secular Canadians often attracted to the New Democratic Party (NDP) (Wilkins-Laflamme, 2016). The motive force appeared to be differences over moral issues that cut across the traditional denominational divisions (Ang and Petrocik, 2012), echoing, albeit weakly, the culture wars polarizing voters south of the border (Guth and Fraser, 2001).
As Wilkins-Laflamme (2016: 503) notes, "Two main dimensions of the religious vote, the effects of religious affiliation and level of religiosity, appear to currently coexist." Attention has focused overwhelmingly on Catholics, mainline and evangelical Protestants, and secular Canadians. We know much less about the many Canadians who adhere to non-Christian religions. There have been two challenges: their small numbers in the typical nationally representative survey and the overlaps with both racialized Canadians and immigrants from nontraditional source countries. There is clearly a propensity to support the Liberals, but whether this is for religious reasons remains unclear. Blais (2005) highlighted the role of racialized Canadians in the Liberals' electoral success but had difficulty coming up with a convincing explanation. The fact that the Liberals were in power when many racialized immigrants arrived was not the answer: these voters were not simply opting for the incumbent party. It was not a matter of shared policy preferences: taking account of their support for doing more for racial minorities, admitting more immigrants or increasing aid to developing countries explained little, and many of these voters were at odds with the party on abortion and same-sex marriage.

Racialized and new Canadians
Subsequent research confirmed that neither social background characteristics nor ideological considerations could explain why racialized Canadians are so much more likely to prefer the Liberals (Harell, 2013). The same was true of belonging to a religious or ethnic association, undermining the notion that ethnic mobilization is key (see also White and Bilodeau, 2014). The one significant factor was strongly identifying as a Liberal, but then, of course, the question becomes why racialized Canadians are more likely to be Liberal identifiers. Bilodeau and Kanji (2010) have invoked Fiorina's (1981: 84) conception of party identification as "a running tally of retrospective evaluations of party promises and performance" to answer this question. If that running tally was in the Liberals' favour, what began as an instrumental attachment may, with time, have been transformed into an affective one. Several Liberal initiatives may have contributed to a positive running tally: opening up immigration to the Global South, official multiculturalism and the Charter. While this is a plausible explanation, focus groups are needed to gain more insight into the Liberals' appeal to racialized Canadians.
The Conservatives and NDP have both tried to attract more support from immigrant and racialized voters. One strategy has involved nominating more candidates from these groups, especially in ridings with high concentrations of immigrants and racialized minorities. This presumes affinity-based voting. A survey experiment that manipulated the ethnicity of fictional candidates shows evidence of this. Not only were racialized respondents more likely to support a candidate from their own ethnocultural group, but they also showed some affinity for candidates from other minority groups (Besco, 2015). However, Besco's notion of "rainbow coalitions" may not extend to party leaders: unlike Sikhs, other racialized Canadians evaluated NDP leader Jagmeet Singh no more positively than other Canadians did (Bouchard, 2022). Moreover, his selection as party leader did not induce Sikhs or other Indian Canadians to shift to the party, challenging the assumption of affinity-based voting. Looking at Chinese Canadians, Goodyear-Grant and Tolley (2019) conclude that affinity voting is motivated by identity rather than interests.
Immigrants from traditional, predominantly white, source countries are much less likely than immigrants from the Global South to support the Liberals (White and Bilodeau, 2014). Given the concentration of more recent waves of immigrants in neighbourhoods where they are disproportionately likely to encounter Liberal supporters, more study of interpersonal influence is needed to understand why newcomers have become more Liberal with each successive wave (White, 2017).
Research on racialized Canadians has been hampered by their small numbers in nationally representative surveys. This has resulted in the homogenization of racialized voters. Moreover, researchers have had to pool data across multiple elections, which is problematic when voting behaviour is in flux. This is no longer the case, as the switch to online surveys allows for much larger samples. Indeed, the samples are large enough to permit the examination of intersecting identities. The Canadian Election Studies (CES) approach to measuring racial identity has also been problematic. First, it used to mix cultural and geographic labels (Taylor, 2021), leading studies to use non-European origin as a proxy for belonging to a racialized minority. Second, it failed to recognize that identity should be "what the individual says it is rather than a product of external assignment" (Thompson, 2012(Thompson, : 1414. However, the most recent CES allows for both racial selfidentification, using categories such as Arab and Black, and "multiraciality" (Thompson, 2012).
When it comes to women, evidence of affinity voting is mixed. In the 1993 election, not only did women evaluate both Audrey McLaughlin and Kim Campbell more favourably than men did, but these favourable ratings were more likely to translate into voting for their parties. Liking for Campbell may have kept some women in the Progressive Conservative fold (Banducci and Karp, 2000) and may even have attracted some women to the party (O'Neill, 1998). However, there is little evidence of affinity voting for women candidates, even among those lacking strong party loyalties, except when women are concerned about women's underrepresentation in Parliament (Goodyear-Grant and Croskill, 2011). There are even elections where men are more likely to vote for women candidates.
Sex is the "fault line of maximum potential cleavage" (Jennings, 1988: 9): even small sex differences in political preferences can affect the outcome of a close election. Canada is clearly a case of "gender realignment" (Inglehart and Norris, 2003). Until the late 1970s, women were less likely than men to vote NDP, but since 1997, this has reversed (Gidengil et al., 2013). However, it is unclear whether Canada ever had a "traditional gender gap" (Inglehart and Norris, 2003): even in the first CES, the sex difference in Progressive Conservative voting was not statistically significant. How early we date the emergence of a "modern gender gap" depends on how we classify the Liberals: in every election between 1965 and 1993 (save 1968), women were significantly more likely than men to vote for the party that oversaw the greatest expansion of the welfare state. It is no coincidence that the gap closed in 1997 when the Liberals focused on eliminating the deficit.
Women care more about social welfare issues, but not because of self-interest (Everitt, 2002): their greater reliance on the state, whether as clients or service providers, cannot explain the "modern gender gap." Indeed, structural and situational explanations more generally have fared poorly. If anything, the gap widens when taking them into account. What matters are cultural values and beliefs (Erickson and O'Neill, 2002;Gidengil et al., 2003Gidengil et al., , 2005Gidengil et al., , 2013. Women are more skeptical than men of the virtues of free enterprise, more open to government intervention and more supportive of redistribution and a strong social safety net, regardless of their employment sector or material circumstances. However, religiosity is an important countervailing force, offsetting the effects of feminist consciousness and pulling many women in a socially conservative direction (O'Neill, 2001). But for religiosity, sex differences in vote choice would be bigger. Indeed, religiosity is critical to understanding the gender gap phenomenon. It explains an overlooked aspect of the "modern gender gap" in Canada: women are more likely than men to vote for the NDP, but they are more likely to vote Conservative than NDP. It can also account for the "gender-generation gap" (Shorrocks, 2018). Older women are more likely than older men to vote Conservative-despite sharing younger women's support for economic equality and state intervention-primarily because they are more religious than younger women. This suggests that cohort replacement helps explain the emergence of the modern gender gap.
Research on the gender gap phenomenon has its pitfalls: "the risk of categorical thinking, reinforcing gender stereotypes, inviting normative comparisons and creating unrealistic expectations about the emergence of a 'women's voting bloc'" (Gidengil, 2007: 815). An undue focus on women's changing behaviour has too often obscured the role of men's changing behaviour, best exemplified by their support for Reform and the Canadian Alliance. This has variously been attributed to an anti-welfare-state backlash, anti-feminist backlash and/or an anti-postmaterialist backlash . Often overlooked is the fact that the differences among women often exceed the differences between women and men. Much more attention is needed to women's (and men's) intersecting identities (Harell, 2017).
Too often, studies have conflated sex and gender and treated both as dichotomous. Sex refers to "the division of bodies into categories based on genitals, chromosomes, and/or hormone levels," whereas gender refers to "behaviors associated with membership in a sex category" (Westbrook and Saperstein, 2015: 537). Bittner and Goodyear-Grant's (2017) development of a novel scale measuring how feminine or masculine men and women feel is an important step forward in differentiating these concepts. 5

Sexual orientation
Sexual orientation has received much less attention. Two clear conclusions emerged from the pioneering studies. First, LGB voters were much more likely to place themselves on the left of the ideological spectrum. This was especially the case for lesbian and bisexual females, replicating the gender gap in the electorate at large. Second, LGB voters were "decidedly and distinctively averse to supporting the Conservative Party" (Perrella et al., 2019: 64; see also Perrella et al., 2012). How they divided their support between the other parties varied from election to election, though gay and bisexual females were consistently more likely than their male counterparts to vote NDP. However, small sample sizes made it impossible for these studies to assess differences in ideological leanings and vote choice within the LGB community.
This changed with the 2019 CES. Using these data, Guntermann and Beauvais (2022) revealed that bisexual women were distinctively left-leaning, whereas bisexual men were more right-leaning than gay men. When it came to vote intention, bisexual women were more likely to opt for the NDP, whereas lesbian women preferred the Liberals. Gay and bisexual men alike opted for the Liberals. The Conservatives fared poorly in all four groups. Despite greater tolerance of sexual minorities and greater recognition of gay rights over the past 15 years, sexuality gaps between heterosexuals and non-heterosexuals did not differ across generations. Using matching methods to provide a stronger basis for inferring causality, the authors suggest that this is because marriage has a countervailing effect by making LGB voters more conservative.
While we are learning more about LGB voters, we know nothing about the political preferences of Canadians with other sexual identities (such as pansexual or Two-Spirit).

Indigenous peoples
Very little is known about Indigenous voting preferences. Many Indigenous peoples abstain from voting, often viewing it as part of a colonial imposition that conflicts with traditional Indigenous forms of governance. Among those who do vote, there is evidence of affinity-based voting, with parties faring better in ridings with high concentrations of Indigenous voters when their candidate is Indigenous (Dabin et al., 2019). However, we need to go beyond trying to fit Indigenous voters into pre-existing theoretical categories and develop theories that fit these voters.

Ideological Orientations: Left-Right and Beyond
When it comes to politics, Canadians are typically not very good at telling left from right. Early research showed that "most Canadians have but a scant understanding of the words 'left' and 'right.' That they do not position the parties correctly in left/ right space, and that, even if they do know what these words mean, they almost never report thinking about politics by using the labels 'left' and 'right'" (Cochrane, 2015: 158). However, as the party system became more clearly oriented along left-right lines, more Canadians were able to place the parties (Cochrane, 2015), and while Canadians are not becoming more ideologically extreme, their ideological consistency has increased (Merkley, forthcoming). Moreover, there is evidence of increasing "partisan sorting," with Canadians increasingly aligning themselves with the party that shares their redistributive preferences (Kevins and Soroka, 2018) or their ideological orientation (Merkley, forthcoming).
Focusing on Canadians' ability to define left and right or to locate the parties on a left-right spectrum risks understating the extent to which they think about politics in ideologically coherent ways. As Cochrane (2015: 159) aptly notes, we would not assess cognitive dissonance by asking people whether they have experienced it. Canadians' fundamental values and beliefs are often structured in ideologically meaningful ways, and where they score on a market liberalism scale (capturing the "old" left-right dimension) or a moral traditionalism scale (capturing the "new" left-right dimension) has a significant effect on their vote (see, for example, Gidengil et al., 2012), with moral traditionalism being key to the emergence of a new right in Canada (Lusztig and Wilson, 2005). 6 Lucas and Armstrong (2021) have taken a novel approach, estimating the latent ideological position of every 2019 CES respondent and then using these positions to estimate riding-level ideology. The authors have made both sets of estimates publicly available, opening up opportunities for new research on the impact of left-right ideology.
The focus on left-right thinking has been too constraining, however. There has been little study of the impact of environmentalism as an ideological dimension (Anderson and Stephenson, 2011). Feminism has figured in analyses of the gender gap (Erickson and O'Neill, 2002;O'Neill, 2001) and affinity voting (Goodyear-Grant and Croskill, 2011), but hostile and benevolent sexism have been neglected and so has social dominance orientation. Medeiros (2021: 929) has shown that "Canada is not, in fact, a bastion against populist attitudes" and that populist attitudes affect vote choice, but given electoral dynamics in other post-industrial democracies, more research is needed into the impact of populism, nativism and authoritarianism.
Party Identification: "Tethered Partisans"? 7 The concept of party identification has been highly contested in Canada, with most debate centring on whether Canadians have enduring psychological attachments to a political party. 8 The social-psychological model of voting behaviour developed at the University of Michigan in the 1950s and 1960s viewed party identification as a longterm component of the vote that affects vote choice both directly and indirectly (Campbell et al., 1964). Some partisans simply vote for "their" party out of a sense of loyalty, but others do so because their partisanship influences their issue attitudes and candidate evaluations. Importantly, party identification was not seen as determining vote choice: short-term factors might induce partisans to vote for another party. Critically, though, their party identification would remain unchanged. Meisel (1975: 67) concluded that a Michigan-style conception of party identification was "almost inapplicable in Canada": party identification appeared to be "as volatile in Canada as the vote itself." That judgment has not stood the test of time. Fewer Canadians may identify with a political party, but party identification "operates the same way for Canadian voters as it does for American voters" (Bélanger and Stephenson, 2014: 117). his colleagues (1991, 2019) would temper this assessment. They distinguished between "durable partisans" and "flexible partisans." Durable partisans identified very strongly or fairly strongly, identified with the same party at both the federal and provincial levels and reported always identifying with the same party. Anyone failing to satisfy one or more of these conditions was deemed a flexible partisan. Their number included people with no identification who were assumed to be transitioning between parties. Flexible partisans far outnumbered durable partisans, leading to the conclusion that "the keynote of partisanship in Canada was its flexibility" (Clarke et al., 1984: 56).
The flexibility of Canadians' partisan ties may have been overstated though. The consistency criterion is questionable. While it is plausible that identifying with the same party at both levels would reinforce people's sense of partisan attachment, provincial partisans living in provinces where the same parties compete at both levels are no more likely to be federal partisans, nor do they have stronger federal party attachments (Roy and Esselment, 2016). 9 Meanwhile, some of the apparent instability may have reflected the lack of an explicit "none" option in pre-1988 CES. When adapting the traditional American question to the Canadian context, the early CES substituted "or what" for Independent. Using a question wording experiment, Johnston (1992) showed that the absence of a "none" option may have induced some people to name the party they were voting for, despite lacking any meaningful psychological attachment, making it appear that party identification travelled with the vote. Since 1988, the party identification question has ended with "or none of these." The 2004-2006-2008 CES panel offered a critical opportunity to assess the stability of Canadians' party ties (Gidengil and Nevitte, 2020). Despite the scandal rocking the party, Liberal partisanship proved relatively resilient. Many Liberal identifiers registered their unhappiness by voting for another party, but relatively few changed their party identification along with their vote. Many Canadians seem to fit the profile of the "tethered partisan" (Green et al., 2002: 57); disaffected partisans are more likely to say they do not identify with any party than to name another party (Gidengil et al., 2012).
Still, assessments of the degree of stability vary. Green et al. (2002) attributed much of the apparent instability to random measurement error, but mixed Markov latent class mover-stayer models that account for random measurement error by treating partisanship as a latent variable have identified a sizeable group of movers (Clarke and McCutcheon, 2009; see also Clarke et al., 2019). However, these models do not differentiate between moving to another party and becoming a nonpartisan.
The Michigan team saw party identification as "a supplier of cues" (Campbell et al., 1964: 128), but there is mixed support for this in Canada. In the case of Liberal partisans, an experiment-based study found no evidence that strong partisans are the most likely to use party labels as cues and only limited evidence for Conservative and NDP partisans (Merolla et al., 2008). Two explanations have been advanced for the weak findings. Guntermann and Lachapelle (2020) find that attitudes toward the cue-giver moderate the effects of elite cues (Guntermann and Lachapelle, 2020), while Merkley (2021: 277) shows that party cues only affect partisans' policy attitudes when partisan motivation is primed by informing them of "a stark divide between parties in Parliament" on the policy.
According to The American Voter, "Identification with a party raises a perceptual screen through which the individual tends to see what is favorable to his partisan orientation" (Campbell et al., 1964: 133). There is evidence of this in Canadians' subjective political judgments but not in their perceptions of objective facts (Blais et al., 2010). Canadians are not "impervious partisans" (Matthews and Pickup, 2019): increased media coverage of the economy and employment can diminish bias in partisans' economic perceptions, but contingent on the prevailing economic and political context. Note that selective exposure plays a role in partisans' acquisition of information: a conjoint experiment varying news headlines indicates that partisans search out politically congenial news (but not news sources) (Merkley, 2021).
With most discussions in Canada understanding partisanship as a positive attachment to a political party, its "oft-forgotten relative" (Caruana et al., 2015: 771) has been neglected. Negative partisanship denotes "an affective repulsion from [a] party, one that is more stable than a current dislike and more strongly held than a passing opinion" (772). Measured by asking survey respondents whether there is a party they would never vote for (Medeiros and Noël, 2014), 10 negative partisanship is not simply the opposite of positive partisanship. It has an independent impact, decreasing the probability of voting for the disliked party by anywhere from 10 to 25 percentage points, depending on the party and the election (Caruana et al., 2015).
Lacking in the Canadian literature is a consideration of partisanship as a form of social identity. The central idea here is that people seek a positive social identity by comparing their group to relevant outgroups. This process of social comparison, and the need for positive differentiation that motivates it, heightens perceived ingroup/outgroup differences. Viewing party identification as a form of social identity provides a richer theoretical foundation to the classic Michigan conception. Originally developed in the United States, partisan identity measures have been adapted to the multiparty context. These measures "capture a subjective sense of group belonging, the affective importance of group membership, and the affective consequences of lowered group status" (Huddy and Bankert, 2017: 6).

Economic Voting: Reward and Punishment
Among the short-term factors that can induce partisans to vote for another party is the state of the economy or their own personal finances. According to the economic voting model, voters engage in a reward-and-punish calculus, voting to re-elect the incumbent when things are going well and voting them out when there is a downturn. However, Canadians are more apt to punish incumbents for bad times than reward them for good times (Gidengil et al., 2012). Voters do not just look to the recent past; prospective evaluations also matter, especially if the incumbent party has a new leader or when choosing among the opposition parties (Anderson, 2010). Whether based on past or future performance, voters' evaluations of the state of the national economy matter much more than how they rate their personal circumstances (Anderson, 2010). However, which matters more differs between "have" and "have-not" regions, with the latter being more likely to vote their pocketbooks (Godbout and Bélanger, 2002). Local economic conditions and the localized impact of national policy changes can also have an independent effect on incumbent support (Cutler, 2002a).
Several factors may diminish economic voting in Canada. Federalism makes it harder for voters to attribute responsibility for economic performance (Anderson, 2008;Cutler, 2004). The actions of one level of government can affect economic performance at the other level, and the different levels of government may engage in credit-taking or blame-shifting. This makes attributing responsibility cognitively demanding, as evidenced by the role of political knowledge in strengthening the effect of jurisdiction-specific evaluations on voting for the incumbent at each level (Anderson, 2008). The presence of more than one opposition party has also weakened economic voting by making it harder for voters to choose an alternative to the incumbent (Anderson, 2010). Canada's heavy dependence on the US market may inhibit economic voting as well, since blame for a downturn can be deflected (Gidengil et al., 2012). The regionalization of the party system has been cited as another factor. There is evidence of this in the case of Quebec with the advent of the Bloc, but not the West (Godbout and Bélanger, 2002). However, when the time frame is expanded, neither the presence of the Bloc nor the salience of the independence issue appears to diminish economic voting in the province (Daoust and Dassonneville, 2018). A final factor is misperception: in the 1997 election, the Liberals were not rewarded for the decrease in unemployment because most voters failed to recognize the drop (Nadeau et al., 2000). One source of misperception is the tendency of voters who are less informed, less attentive to the media and less educated to assess national economic conditions based on the local unemployment rate (Anderson and Roy, 2011).
The studies discussed so far relied on survey data. An alternative is to relate vote shares to economic indicators. Gélineau and Bélanger (2005), for example, regress incumbent vote shares at the federal and provincial levels on changes in the consumer price index, unemployment and personal disposable income at both levels. They conclude that improved provincial conditions do not offset the effects of deteriorating national conditions and that provincial incumbents suffer for worsening national conditions if their party is in power nationally. Aggregate analyses have also revealed how long-term economic hardship impacts third-party support (Bélanger and Nadeau, 2010;Perrella, 2005).

Issue Voting: Do Issues Matter?
Issue attitudes typically have more impact on individual vote choice than do economic evaluations. In the 1997 election, issues proved decisive for 9 per cent of voters while the economy was decisive for only 4 per cent . In other words, people were more than twice as likely to have voted differently if the issues had not mattered as would have had the economy not mattered. Across the four elections held between 1988 and 1993, the median figure was very similar, with issues being decisive for one voter in 10 and the economy being decisive for only one in 15 (Blais et al., 2004).
Of course, how much issues matter varies depending on the issue and the election. Moreover, issues can have an impact on individual vote choice without necessarily mattering much to the election outcome. For issues to affect parties' electoral fortunes, parties have to take clear stands on opposing sides, voters have to be aware of those stands and the balance of opinion has to favour one party over the others. Those conditions are not always going to be met. Indeed, the median change in party vote shares would have been only 3.2 points across the four elections held between 1988 and 2000 had issues not mattered (Blais et al., 2004).
According to the classic Downsian proximity model (Downs, 1957), voters seek to maximize their expected utility by voting for the party closest to their own position in a given policy domain. This model has been challenged by an alternative spatial model. The directional model predicts that voters prefer the party taking the strongest position on their side of an issue because of greater awareness of the party's position and greater confidence that the party will act on its promise (Rabinowitz and Macdonald, 1989). However, in Canada, the proximity model dominates (Blais et al., 2001;Johnston et al., 2000). The tests of these models are among the few studies of issue voting in Canada that take seriously the problems of rationalization and projection that occur when people assume that their preferred party shares their position.
According to the theory of issue ownership, some parties are viewed as being more competent than others in dealing with particular issues. These perceptions can be biased by partisanship, but Bélanger (2003) found that partisans are able to distinguish among the parties. He also found that perceptions change in response to parties' performance in office. Perceived party competence can also change when new parties appear on the electoral scene. Perceptions of issue ownership affect vote choice but only when the issue is important to voters (Bélanger and Meguid, 2008).
The conditioning effect of issue salience is also evident in performance-based voting: the more voters care about an issue, the more heavily they weight their evaluations of the government's handling of the issue when deciding how to vote (Fournier et al., 2003). Performance-based voting is also contingent on attributions of responsibility. As noted above, attributing responsibility is challenging for voters given the nature of Canadian federalism. Canadians have difficulty differentiating the roles of the provincial and federal governments on various issues and are apt to attribute responsibility to both levels (Cutler, 2008). Attributions of responsibility for health care have a significant effect on vote choice but only when voters assign primary responsibility to one level of government (Cutler, 2004). For responsibility judgments to affect performance voting, certain conditions must apply: voters care about the policy area, one level of government is clearly dominant, there is little intergovernmental conflict and media coverage publicizes each level's contribution to the outcome (Cutler, 2017). The fact that these conditions are often not met has important implications for Canadians' ability to hold their governments accountable.
Party Leaders: The "Superstars of Canadian Politics" It is easy to understand why voters' evaluations of the party leaders exercise a strong influence on their vote choice. After all, the leader of the winning party becomes prime minister. Party leaders' status as the "superstars of Canadian politics" (Clarke et al., 1991: 89) is reinforced by the way the media cover elections and by televised debates that put the spotlight on the leaders. Flexible partisan ties and lack of ideological polarization can also enhance party leaders' importance. Especially for low-information voters, feelings about the party leaders can serve as a useful shortcut, with demographic similarity being the "simplest shortcut of all" (Cutler, 2002b). The weight of leader evaluations in individuals' vote calculus is uncontested, but they are not necessarily as important to election outcomes (Johnston, 2002). Only if one leader is markedly more popular or less popular than the others will leader evaluations have much impact on party vote shares. The effects will also wash out if as many voters dislike as like a leader.
The notion of a prime ministerial prototype has proved useful in understanding what influences voters' leader evaluations (Brown et al., 1988). People judge party leaders based on how well they fit their mental image of what a prime minister should be like. These images are not idiosyncratic. The evaluation criteria are not specific to a given leader or to a given election. They represent something much more enduring, remaining stable from election to election and leader to leader. Rather than relating to personal style, the key leadership attributes proved to be task-relevant: competence, integrity, dynamism, empathy and responsibility. Bittner (2010) has focused on two overarching dimensions. Competence encompasses traits such as intellect and leadership ability, while character reflects traits such as honesty, compassion and trustworthiness. She finds that voters differentiate between leaders and their parties; they do not attribute the same trait to leaders simply because they are from the same party. And when voters evaluate a leader, they do not rate them similarly on all the traits or rate the same leader similarly from election to election. As voters learn more about a leader, their trait ratings can change. Voters evaluate leaders' traits relative to one another. When it comes to vote choice, perceived character seems to matter more than perceived competence.
According to the presidentialization of parliamentary elections thesis, voters' assessments of party leaders have assumed an increasingly important place in their vote calculus. This has been linked to changing patterns of media coverage as leadership becomes one of the dominant frames. The increasing electoral salience of party leaders has also been linked to the supposed erosion of traditional electoral cleavages: if voters' choice of party is less anchored in their social group memberships, leaders could plausibly have become more important. The final factor that is typically cited is partisan dealignment: if partisan ties are weakening, voters will no longer evaluate leaders through a partisan lens and they will have to look for other cues to guide their vote choice. However, these explanations are flawed (Bittner, 2018;Gidengil and Blais, 2007). The media focus on the leaders is nothing new, social background characteristics continue to help structure vote choice, and the proportion of nonpartisans has fluctuated. Whether we track the impact of leaders' perceived competence and character (Bittner, 2018) or leader evaluations (Gidengil and Blais, 2007), the verdict is the same: party leaders matter to vote choice but no more than they did 40 or 50 years ago.

Does the Local Candidate Matter?
The media spotlight may be on the party leaders, but the local candidate can be a decisive factor for 4 to 8 per cent of voters outside Quebec and 2 to 6 per cent of Quebec voters, depending on the election (Sevi et al., 2021;Stevens et al., 2019), and candidate evaluations may have caused as many as 10 per cent of seats to change hands in the 2015 election (Stevens et al., 2019). When they preferred a local candidate from a different party, two out of five voters opted for the candidate over their preferred party (Blais and Daoust, 2017). This tendency was weakest for voters who preferred the Liberals or Conservatives.
There has been little study of what drives these local candidate effects. A surveybased experiment suggests that one factor may be the quality of the candidate (Roy and Alcantara, 2015). Respondents were more likely to support a strong local candidate, as indicated by the candidate's previous experiences, education and employment history. However, the effect was conditional on the strength of partisanship and attention to campaign information.
This area is ripe for further study. We need to know whether some types of voters are more susceptible to candidate effects, which characteristics of candidates matter, and the role of the local campaign. Recent work using geospatial methods to measure riding effects on vote shares could be useful for examining the role of campaign intensity (Bodet et al., 2022), while taking advantage of natural experiments, such as changing electoral boundaries, could enhance confidence that local candidate effects are causal (Stevens et al., 2019).

Are Canadians Strategic Voters?
Elections are fought and won riding by riding. Party standings in a riding may induce strategic defections. A good deal of media speculation focuses on strategic voting, but numerous studies suggest that it is much less prevalent than media reports suggest. This research was motivated by Canada's being an apparent exception to Duverger's law that first-past-the-post plurality electoral systems result in two-party systems. In addition to a "mechanical effect" (systematically distorting the translation of votes into seats), such systems have a "psychological effect": voters should want to avoid "wasting their vote" on parties that have no chance of winning (Duverger, 1954).
Strategic voting occurs when a voter votes for a party that is not their first choice because their preferred party is perceived to have little or no chance of winning the seat. Instead, a strategic voter votes for a party with a better chance of winning, in hopes of defeating the party they like least. Thus, a strategic vote is based on both a voter's preferences among the parties and their expectations about their chances. Despite the incentives offered by Canada's electoral system, studies have detected relatively little strategic voting. In the 1988 election, voters opposed to the Canada-US Free Trade Agreement had to decide whether the Liberal or the NDP candidate was most likely to defeat the Progressive Conservative one, yet only one in eight of those preferring a third-place party opted for one of the top two candidates (Blais, 2002). Other studies estimate that somewhere between 2 per cent and 7 per cent vote strategically, depending on the election and how strategic voting is assessed (Blais et al., 2001;Cutler et al., 2022;Merolla and Stephenson, 2007).
So why do relatively few voters cast a strategic vote? The first answer is that many voters have no incentive to defect because their preferred party stands a very good chance of winning the seat, and so they vote "sincerely" for their first-choice party. 11 Accordingly, a more appropriate way to assess whether Duverger's psychological mechanism operates in Canada is to focus on supporters of nonviable parties (Blais and Nadeau, 1996). From 30 per cent to as many as 50 per cent of these voters vote strategically (Blais and Nadeau, 1996;Daoust and Bol, 2020). Two factors explain why the numbers are not higher: many supporters of nonviable parties have a strong preference for their first-choice party and may lack an acceptable second choice, and many overestimate their preferred party's chances of winning the seat. If they have only a weak preference for their first-choice party, realize that their party has no chance of winning, and perceive the race between the top two parties to be tight, supporters of nonviable parties are very likely to act as Duverger expected and vote strategically (Blais and Nadeau, 1996). However, these conditions are rarely satisfied (Blais, 2002). Daoust (2018) has added a new twist by asking whether voters factor a preference for a minority or a majority government into their vote calculus. If they expect their party to lose, major-party supporters might prefer a minority government to a majority government controlled by the other party. Third-party supporters, meanwhile, might want a minority government because it could enhance their party's influence in Parliament. Daoust estimates that about 5 per cent of voters may have deserted their preferred party in the 2015 election in hopes of electing a minority government.
Affluent Canadians and university graduates are more likely to be strategic voters and so, to a lesser extent, are older Canadians and women (Eggers et al., 2022). In other words, "the same types of voters who are more likely to turn out are also more likely to make their votes count" (Eggers et al., 2022(Eggers et al., : 1862. On the other hand, strong partisans are less likely to vote strategically because strategic defection carries a psychological cost (Daoust, 2018) and because strong partisans are more apt to engage in wishful thinking when it comes to their party's chance of winning the seat. However, minor party supporters are more likely to vote strategically: in 2019, 60 per cent of strategic voters were Green Party or People's Party supporters (Cutler et al., 2021). Given that strategic voting presupposes some understanding of how the electoral system works, it is unsurprising that political sophistication (as proxied by educational attainment) increases the likelihood of strategic voting (Daoust, 2018;Merolla and Stephenson, 2007). Loewen et al.'s (2015) experiments provide deeper insights into the role of cognitive abilities. People with the ability to reason strategically are no more likely to vote strategically, but they are less likely to base their decision on the strength of their preferences and more likely to rely on probabilistic information about their preferred party's chances of winning. Experimental evidence also highlights the role of emotion. Dumitrescu and Blais (2014: 453) conclude that "moderate levels of anxiety increase the probability of defection, but at high levels, anxiety has a paralyzing effect, making voters less likely to abandon their preferred choice." Studies have typically focused on the level of strategic voting in Canada. The consensus is that the overall level is low, but an important question remains: How often does strategic voting change the outcome of a local race? As Stevens et al. (2019) conclude with respect to local candidate effects, effects may be modest overall but they may be decisive in some ridings. Their estimation approach could usefully be applied to strategic voting.

Going Forward
This review has aimed to convey the richness of the Canadian voting behaviour literature. Inter alia, this literature has challenged the conventional wisdom about the (un)importance of the local candidate, the prevalence of strategic voting, the dominance of economic considerations and the increased impact of leader evaluations. It has demonstrated that Canadians structure their political thinking in ideologically coherent ways and that this influences their vote choice, and it has highlighted that when conditions are right, issues do matter.
At the same time, there are many aspects that warrant attention. For example, we know little about the voting behaviour of the losers of globalization: how have deindustrialization, globalization and automation affected vote choice? What might an investigation of populist attitudes, authoritarian predispositions and nativism tell us about the potential for a radical right populist party to have an electoral breakthrough? What makes people like or dislike their local candidate? How much impact does strategic voting have on election outcomes? How might an understanding of party identification as a social identity cast new light on the stability of Canadians' partisan ties?
The field is on the cusp of exciting new advances. The advent of online surveys means that researchers do not need major funding to field adequately powered studies. Importantly, projects like the CES can now have samples that are large enough to draw reliable inferences about small population groups. This will enable researchers to dig deeper into the vote choices of racialized Canadians, new Canadians, sexual minorities and religious minorities. Most crucially, it will be possible to take intersectionality seriously. Large sample sizes also allow for the use of split samples to maximize the number of questions asked.
The ease of embedding experiments is another major advantage of online surveys. 12 Enabling researchers to maximize both internal and external validity, survey experiments can help to address one of the major challenges in studying voting behaviour. Much of what we know about vote choice is based on cross-sectional studies where even the direction of the causal arrow may be uncertain. Experiments are the gold standard for demonstrating causality: random assignment reduces the threat of omitted variable bias, and the time order is unequivocal since the researcher controls who is exposed to the treatment and when. Studies reviewed here relied on experiments to investigate affinity voting (Besco, 2015;Goodyear-Grant and Tolley, 2019), the impact of local candidate quality (Roy and Alcantara, 2015), partisan cues (Guntermann and Lachapelle, 2020;Merkley, 2021;Merolla et al., 2008) and question wording (Johnston, 1992).
Some topics, however, do not lend themselves to the use of experimental designs. For example, we obviously cannot randomly assign some people to be Conservatives and others to be Liberals. One option is to use matching methods to mimic the logic of the classic experiment, as Guntermann and Beauvais (2022) do when examining the impact of marriage on the vote choices of LGB voters. However, matching is limited by data availability and cannot rule out reverse causation. 13 Studies of vote choice in Canada have yet to use statistical approaches, such as regression discontinuity and difference in difference designs, that allow for greater confidence in causal inferences. Unexpected Events during Survey designs and natural experiments can offer similar advantages. Meanwhile, mediation analyses can strengthen confidence in causal inferences by testing hypothesized mechanisms. There are potential pitfalls, of course: in all these cases, causal inferences are only warranted if strict conditions are satisfied. Another option is to combine panel data with an instrumental variable approach (Pickup and Evans, 2013). This approach has the potential to address the endogeneity challenges inherent in assessing the impact of partisanship, performance evaluations and leader evaluations on vote choice.
Finally, researchers need to consider expanding the methods used to study vote choice in Canada. Apart from studies of economic voting, there has been very little use of aggregate-level data. Dabin et al.'s (2019) study of Indigenous peoples and affinity voting is a rare exception (see also Johnston, 1985). Similar methods could be used to complement individual-level analyses of affinity voting on the part of immigrants, racialized Canadians and women. More generally, aggregate data can be used to examine how context affects vote choice. Geospatial methods can also provide more understanding of spatial variations in voting behaviour and prompt greater attention to contextual effects. Combining "big data" and artificial intelligence (AI) has the potential to prompt novel theorizing by revealing unexpected patterns.
However, in the rush to embrace new technologies, we should not overlook the potential benefits of sitting down and listening at length to what people have to say about why they vote as they do. With so many potential factors to weigh, the seemingly simple act of marking a ballot can involve a complicated decision calculus.
4 For a more theoretically motivated approach to defining regions, see Cochrane and Perrella (2012). 5 However, because feminine and masculine are treated as polar opposites, their scale cannot differentiate people who feel both masculine and feminine or neither (Gidengil and Stolle, 2021). 6 For Quebec, see Gauvin et al. (2016). 7 Green et al. (2002: 57). 8 See Gidengil (1992) for a discussion of early contributions to the debate about party identification in Canada. An alternative conception of party identification as "a running tally" (Fiorina, 1981) has not figured much in the Canadian literature (but see, for example, Kanji, 2010) 9 Clough (2007) showed that provincial party identification affects federal vote choice but did not examine whether it reinforced the effect of federal party identification. 10 Caruana et al. (2015) add the stipulation that the party score below the midpoint on the party feeling thermometer. 11 Strategic defection is not the only reason why voters vote insincerely: some do so as a way of signalling their discontent (Schimpf, 2019). 12 Taking advantage of crowdsourcing via MTurk provides an even lower cost way of running experiments, albeit with convenience samples skewed young. 13 Reverse causation is unlikely to be a challenge to Guntermann and Beauvais' (2022) conclusions.