Official voter turnout rates in Canada have suddenly declined in the last decade. In three of the four national elections held between 1979 and 1988, three of every four Canadians who were enumerated and included on the voter list actually cast a ballot in the federal election, but that turnout rate has fallen precipitously in subsequent federal elections. The turnout rate dropped to 67.0 per cent of the electors on the official voter list in 1997, to 61.2 per cent in the election of 2000, and to 60.9 per cent in the federal election of 2004 (Blais et al., Reference Blais, Gidengil, Nadeau and Nevitte2002; Elections Canada, 2003). This decline is by no means unique to Canadian federal elections, as similar patterns have been evident in provincial elections and in other advanced industrial democracies (Gray and Caul, Reference Gray and Caul2000; LeDuc and Pammett, Reference LeDuc and Pammett2003). A number of politicians, public commentators and academics have lamented Canada's low turnout rates relative to other democracies (Franklin, Reference Franklin, LeDuc, Niemi and Norris1996; Milner, Reference Milner1997), and concern about the recent decline spurred Elections Canada to commission a study of non-voters by two leading scholars of voter behaviour to determine the sources of this apparently troubling trend (Pammett and LeDuc, Reference Pammett and LeDuc2003). While the first-past-the-post electoral system and a relatively low population density accounts for Canada's low turnout in comparison to other democracies (Blais and Dobrzynska, Reference Blais and Dobrzynska1998; Martinez, Reference Martinez and Thomas2000), recent analyses have attributed a major portion of the decline to the extraordinarily low turnout of younger citizens, especially in comparison to the participation rates of preceding cohorts in the Canadian electorate (Blais et al., Reference Blais, Gidengil, Nevitte and Nadeau2004; LeDuc and Pammett, Reference LeDuc and Pammett2003; Milner, Reference Milner and Milner2004; for a parallel argument about the decline of turnout in the United States, see Lyons and Alexander, Reference Lyons and Alexander2000). The fact that this generational effect might persist well into the future only exacerbates the concern held by many that low and decreasing voter turnout could be an indicator of the degrading health of civic society in Canada (Blais et al., Reference Blais, Gidengil, Nadeau and Nevitte2002: 10).Footnote 1
Low and declining turnout in democracies often raises anxiety about the levels and trends in political inequality. In the absence of political institutions to entice, mobilize or coerce voters to the polls, the cleavages between voters and non-voters are likely to reinforce, rather than mitigate, the patterns of power and wealth associated with contacting public officials, campaign contributions, community participation, lobbying and running for office (see Verba et al., Reference Verba, Schlozman and Brady1995). Indeed, Lijphart's extensive review concluded that not only is there a clear link between turnout and inequality, but low turnout and its associated bias “have important consequences for who gets elected and for the content of public policies” (Reference Lijphart1997: 4). Specifically, higher turnout tends to be related to more redistributive policies (see, for examples, Hicks and Swank, Reference Hicks and Swank1992; Hill and Leighley, Reference Hill and Leighley1992; Martinez, Reference Martinez1997) and, consequently, slower economic growth (Mueller and Stratmann, Reference Mueller and Stratmann2003). Normatively, if liberal democracy can be thought of as a blend of pluralism (in which public policies are developed through bargaining among the “haves” who invest their political resources) and majoritarianism (in which policies reflect the preferences of the majority), low and biased electoral turnout might be seen as upsetting that balance by ceding control of the most majoritarian of institutions (elections) to those who already benefit from the inequalities inherent in patterns of candidate recruitment, the interest group system, control of the media, and other “pluralist” institutions.
However, scant attention has been paid to the consequences of the recent drop in voter turnout rates in Canada. In a partisan sense, the conventional wisdom believed by most casual observers of liberal democracies predicts that parties of the left should benefit from higher levels of turnout and parties of the right should pray for rain on election day. Echoing Lijphart's observations about the class bias in turnout, the conventional wisdom holds that the socio-economic factors that promote turnout also increase the likelihood of preferring a conservative party or candidate. Thus, in low turnout elections, disproportionate numbers of less educated, poor people in less prestigious occupations will abstain, leaving the election to be decided by a relatively better educated and wealthier electorate, which would presumably be more sympathetic to conservative parties. Higher turnout elections bring a larger share of the socio-economically disadvantaged to the polls, thereby increasing the prospects of parties of the left. There is a fair amount of research that supports that view within the US setting (Radcliff, Reference Radcliff1994; Tucker and Vedlitz, Reference Tucker and Vedlitz1986) and across observations of several countries (Pacek and Radcliff, Reference Pacek and Radcliff1995).
However, that pattern is not always evident. DeNardo (Reference DeNardo1980) proposes that the majority party is actually more likely to suffer in high turnout elections, due to high voter partisan defections associated with a high-interest election. From this revisionist perspective, low turnout elections are contests between the “true believers,” but high turnout elections include a greater number of voters whose preferences are more susceptible to persuasion. Thus, in areas with Democratic majorities in the United States, high turnout increases the number of potential Democratic defectors who will vote Republican.
Recent investigations have found that the link between turnout and partisan advantage varies. In some US states, Republicans appear to have benefitted from higher turnout, and in others, Democrats benefit (Citrin et al., Reference Citrin, Schickler and Sides2003; Nagel and McNulty, Reference Nagel and McNulty1996). Cross-national analyses have also concluded that advantages of higher turnout do not invariably accrue to the parties of the left (Bernhagen and Marsh, Reference Bernhagen and Marsh2004; Tóka, Reference Tóka2000). An examination of US presidential elections from 1960 to 2000 demonstrates that higher turnout clearly benefitted Democratic presidential candidates throughout the period, but that the probable effects of turnout on partisan choice declined substantially over time and were only barely still tilted in the direction of the Democrats by the year 2000 (Martinez and Gill, Reference Martinez and Gill2005). This decline in the relationship between turnout and partisan outcomes was concomitant with a decline in the party/class cleavage in the United States, and suggests that the relationship between turnout and partisan advantages may vary with the strength of the party/class cleavage.
In this article, we simulate varying turnout levels in the 1997 Canadian federal election in an attempt to determine which parties might have most plausibly benefitted from higher turnout in that election, and which parties might have benefitted from lower turnout in the same election. Canada in 1997 is an intriguing case, with a new partisan alignment reflecting regional, linguistic and religious cleavages in society, but a weak class cleavage. In that context, we might expect that the likelihood that “left” parties would benefit from higher levels of turnout would be diminished, and that the relationship between turnout and partisan outcomes might be shaped by other relevant cleavages.
Data and Method
Determining the likely effects of hypothetically different conditions of any specific election poses an interesting epistemological dilemma. We cannot rerun an election multiple times as if we are in a laboratory with the hundreds of candidates and campaign workers as research confederates and millions of citizens as experimental subjects. We can, however, use existing survey data on the characteristics and preferences of the electorate at that point in time to simulate different levels of turnout based on different probabilities of voting among individuals in the population. We can also simulate the likely partisan choices of actual non-voters, based on what we know about the decisions of those who actually did participate in the election. Thus, our general approach is to estimate a model of voter choice, derive predicted probabilities of each respondent's behaviour from those estimated models, and estimate the distribution of partisan preferences across different sets of respondents based on their estimated likelihood of voting in the election. Our estimates are based on analyses of the 1997 Canadian Election Study, which we obtained from the York University Institute for Social Research website.Footnote 2 These panel data are from telephone interviews conducted before and after the election, and provide us with a wide range of information about individual respondents' demographic characteristics, short- and long-term political attitudes and, of course, voting behaviour in the 1997 national election.
The 1997 federal election in Canada followed a significant change in the national electorate, which produced one dominant national party (the Liberals) and two powerful regional parties (the Bloc Québécois in Quebec and the Reform party west of Ontario). The preceding election in 1993 reduced the Progressive Conservative party to a shell of its former self, and its diminished popular support across the nation was exaggerated in the House of Commons by the first-past-the-post electoral system (Milner, Reference Milner and Milner2004: 21). The social democratic NDP also suffered significant losses (Nevitte et al., Reference Nevitte, Johnston, Blais, Brady and Gidengil1995). Because the resulting alignment left Canadians in different regions with different sets of electoral choices, we provide separate analyses for Quebec and the remaining provinces.
While many analyses consider the question of voter participation distinct from the question of party choice among voters, we base our model on the assumption that the citizen has the choice of abstaining or voting for each of the parties. Based on respondent self-reports, we classified Quebec respondents as abstainers, Liberal voters, Progressive Conservative voters or Bloc Québécois voters, and we classified respondents in other provinces as abstainers, Liberal voters, Progressive Conservative voters, NDP voters or Reform voters. We excluded respondents who voted for minor party candidates,Footnote 3 as well as those who did not know or refused to report if or for whom they voted.
We address the pervasive problem of missing data in analyses of survey research by using a multiple imputation process. King et al. (Reference King, Honaker, Joseph and Scheve2001) provide a general discussion of multiple imputation routines, which essentially allow us to make informed guesses about the unobserved values of the missing data from random draws from distributions constructed from the valid data that we do have about each case. Multiple imputation not only preserves cases that would otherwise be lost in listwise deletion models but, more importantly, it reduces the possible bias in estimated coefficients that might result from patterns of missing-ness in the data.Footnote 4 The iterative process is based on maximum likelihood estimates of each variable in the model as a function of all other variables in the model (plus a few additional variables used to maximize the predictive capacity of the process). Thus, if we are missing a respondent's score on one of our predictor variables (say, campaign interest), we can construct a distribution of her likely score on that variable based on everything else that we know about her, and take multiple (here, five) random draws from that distribution (Little and Rubin, Reference Little and Rubin1983; Rubin, Reference Rubin1987).Footnote 5 Using this procedure, we created and analyzed five replicate datasets, each of which contains all the valid data for all respondents, as well as imputed values for each of the missing values in the variables in the model. The results that we report below are the average results of the analyses on the five replicate datasets.Footnote 6
The CES sample distribution of abstentions and party votes in Quebec and in the rest of Canada are compared to distributions in the Canadian voting-age population in Table 1. As in election surveys in the United States, a combination of differential selection, differential panel mortality, panel conditioning and respondent misreports results in an underestimate of abstainers (see Bartels, Reference Bartels2000; Bernstein et al., Reference Bernstein, Chadha and Montjoy2001; Burden, Reference Burden2000; Cassel, Reference Cassel2004; Martinez, Reference Martinez2003). The overreport of voting appears to be more significant outside of Quebec, though that impression may be the result of greater proportions of non-citizens residing in the rest of Canada.Footnote 7 Among voters, the CES sample exaggerates the Bloc's lead over the Liberals in Quebec, and understates the Liberals' lead over Reform outside of Quebec.
Table 1 Comparison of Actual Voting Behaviour in 1997 Election to CES Sample Reports

Source: Actual votes from Elections Canada. 1997. Thirty-sixth General Election 1997: Official Voting Results. Voting-age population calculated from Statistics Canada. CES data calculated by authors from 1997 Canadian Election Study. Quebec data weighted by cpshhwgt. Rest of Canada data weighted by cpsnwgt2.
Our goal is to determine how higher or lower turnout in the 1997 Canadian federal election would have affected the distribution of votes received among the major parties. The first step in that process is to estimate vote choice (including the abstention category) as an unordered multinomial logit function of standard variables associated with both party preference and the likelihood of voting. From that estimation, we derive probabilities for each respondent's selection of each of the choices (which, again, are, for Quebec voters, abstain, vote Liberal, vote Progressive Conservative or vote Bloc Québécois, and for other voters are abstain, vote Liberal, vote Progressive Conservative, vote NDP or vote Reform). From those probabilities, we simulate several levels of turnout. Higher turnout is simulated by progressively adding to the pool of voters actual abstainers with the lowest probability of abstaining of those remaining in the pool of abstainers. Lower turnout is simulated by progressively subtracting from the electorate actual voters with the highest probability of abstaining.
Because the outcome variable represents both the decision to vote or not to vote and the party choice for voters, we chose explanatory variables that we expect to be associated with turnout or candidate preference, or both. We expected turnout to be positively associated with usual demographic correlates (age, education, importance attached to religion, income), as well as campaign interest and campaign knowledge (the latter operationalized as knowledge of party positions). We expected voters' preferences between parties to reflect a number of factors, including federal party identification (represented by dummy variables for each party, with non-identifiers as the excluded category), short-term evaluations of the parties (as measured by feeling thermometers), demographics (specifically, French language, Catholicism and gender), retrospective evaluations of the Liberal government's performance in office, and issue preferences. The issue variables in our Quebec model are indices of attitudes toward sovereignty and attitudes toward the United States, and the issue variables in the model outside Quebec are indices of Quebec support, attitudes toward free enterprise and moral traditionalism. (See Appendix A on our websiteFootnote 8 for our coding of these variables and the construction of indices.)
We also include a variable to represent the strategic choice that confronted voters on the right outside Quebec in the 1997 election. The Progressive Conservatives and the still relatively new Reform party had been unable to “unite the right” by 1997, so non-Quebeckers looking for a rightist alternative to the governing Liberals were forced to make a choice. A voter who truly preferred anyone but Chrétien (or the NDP leader, Alexa McDonough) might consider which of the “right” party candidates in his or her riding had the best opportunity to win. To incorporate the possibility of strategic voting into our model, we include the difference between the respondent's adjusted subjective probability that Reform would win the riding and his or her subjective probability that the PC would win the riding, as well as interactions between that variable and identification with either Reform or PC. In general, we expect that an individual's likelihood of choosing either Reform or PC (but not the Liberals or NDP) will be affected by the relative difference, and that effect should be greater for identifiers of one of the rightist parties.
The primary purpose of these models is to generate predicted probabilities of abstention and voter preference, and less to provide an elegant explanatory model of voter choice. We would be dismayed if some of the coefficients in the model failed to reach conventional levels of statistical significance, but we would not be surprised if the collinearity between party identification, party feeling thermometers, evaluations of the government's performance and positions on salient issues (among other things) partially obscured the independent contribution of each in a statistical sense. While that would be a concern in an explanatory model, it is not in these models since they are only used to generate predicted probabilities for later analysis.
Quebec: Model Estimation and Results
Quebec voters essentially had four choices (abstain, vote Liberal, vote Progressive Conservative and vote Bloc Québécois), so our multinomial logit function produces three vectors of coefficients, representing the effects of each variable on the likelihood of selecting each of the parties relative to the likelihood of abstention. Thus, the significant coefficient for “female” in the first column in Table 2 shows that Quebecois women were significantly more likely to vote for the Liberals than to abstain. The significant “female” coefficient in the Liberal vs. abstention vector and the insignificant “female” coefficients in the Progressive Conservative vs. abstention and the Bloc vs. abstention vectors provide evidence of a gender gap in Quebec in 1997. Other things being equal, Quebec women were more likely than Quebec men to stick with the incumbent government.
Table 2 Multinomial Logit Model of Vote Choice in Quebec, 1997

Entries are unstandardized multinomial logit coefficients (and associated confidence intervals) based on multiple imputations.
Again, in this step of our analysis, we value predictive accuracy over explanation, but the expected effects of other variables provided some assurance of our model's verisimilitude. Identification with each party (at the federal level) increased the likelihood of voting for that party over abstention, but had no noticeable effect on the likelihood of voting for another party relative to abstention. Similarly, short-term evaluation of each party (as measured by the feeling thermometer) was positively associated with the likelihood of voting for that party, and in the case of evaluations of the Liberals, was negatively associated with the likelihood of voting for the Bloc. Francophones and sovereignists were more likely to vote for the Bloc than abstain, while Quebecois who were more sympathetic (or less hostile) to the United States were more likely to vote for the Conservatives. Campaign interest and knowledge of the party positions were each associated with higher turnout, as both had strong, positive effects on the likelihood of voting for all three parties relative to abstention.
Multinomial logit assumes the independence of irrelevant alternatives (IIA), which states that the coefficients predicting any choice should not be significantly affected by the addition or deletion of any other choices. Hausman and McFadden (Reference Hausman and McFadden1984) propose a test to determine whether the null hypothesis of IIA must be rejected, in which the equation is reestimated several times, successively eliminating each choice alternative. The vectors of coefficients in the restricted models are then compared to the vectors of coefficients in the full model, and the resulting chi square tests (with degrees of freedom equal to the numbers of coefficients in each vector) indicate the probability that the IIA (null) hypothesis must be rejected. In the Quebec model, we are well below the critical values at the .95 probability that would require a rejection of the null.Footnote 9
We use the estimated coefficients in this model to derive the predicted probabilities of each respondent selecting each choice alternative. Since the product of the estimated coefficients and the explanatory variables for the i th individual is equal to the log of the odds of i selecting choice j (either Liberal, Progressive Conservative or Bloc) divided by the odds of abstaining, as in

we can algebraically rearrange to produce the probabilities of abstaining, voting Liberal, voting Progressive Conservative and voting Bloc, based on the fact that the sum the four probabilities equals one for every respondent. Then, for each respondent, we calculated conditional probabilities of voting for each party, excluding the probability of abstention, by

As a validation check, we calculated the correlations between the estimated probabilities of selecting each choice and dummy variables for each actual outcome. Those correlations were .78 for the Liberals, .81 for the Bloc, .60 for the Progressive Conservatives and .49 for abstention. Thus, while we generally did better in predicting voter choice between the parties than we did in predicting abstention, our model still suggests that the decision to abstain is partially explainable and is far from random (contrary to the conclusion reached by Matsusaka and Palda, Reference Matsusaka and Palda1999).
To simulate the effects of varying levels of turnout on the aggregate vote choice, we sum the conditional probabilities, p(Liberal|vote), p(PC|vote) and p(Bloc|vote), across different sets of respondents. We simulate higher turnout by including actual voters as well as non-voters who had the lowest probabilities of abstaining. We simulate lower turnout by excluding actual voters who had the highest probabilities of abstaining. Results for 20 simulated turnout levels in Quebec (from 9.11 per cent to 100 per cent) are displayed in Figure 1.

Figure 1 Simulated Turnout and Partisan Outcomes: Quebec
In CES's Quebec sample, the Bloc had a nearly five-point advantage over the Liberals among the 82.2 per cent of respondents who indicated that they had voted. Figure 1 shows that as simulated turnout increases, support for the Bloc increases at the expense of the Liberals. We estimate that if turnout had been nine points above the baseline (91.2 per cent of respondents), the Bloc's popular vote margin would have increased to about 6 percentage points, and at 98.1 per cent of respondents, it would have surged to 6.6 percentage points, which we would expect to be magnified in the distribution of the Quebec delegation seats in the House of Commons. Similarly, as turnout decreases, the Liberals' fortunes in Quebec appear to improve. If turnout had been about eight points lower than it was, the Bloc's popular vote margin among CES respondents would have been trimmed from 5 percentage points to about 3.5 percentage points. If only half of respondents had reported voting, our results suggest that the Liberals would have had a commanding win in Quebec. Turnout matters, and heavier turnout in Quebec appears to help the Bloc Québécois and hurt the Liberals. Quebeckers' support for the Progressive Conservatives is not significantly affected by large variations in turnout.
Outside Quebec: Model Estimation and Results
Voters outside Quebec generally had five choices: abstention, Liberal, Progressive Conservative, New Democratic party (or NDP, a democratic socialist party) and Reform, a western-based party that supplanted the Progressive Conservatives on the right end of the ideological spectrum (Nevitte et al., Reference Nevitte, Johnston, Blais, Brady and Gidengil1995). As in the Quebec case, our estimated model (in Table 3) reflects the expected effects of party identification and short-term partisan evaluations, issues and demographics on the vote choice.Footnote 10
Table 3 Multinomial Logit Model of Vote Choice in Rest of Canada, 1997

Entries are unstandardized multinomial logit coefficients (and associated confidence intervals) based on multiple imputations
The party identification dummy variables for Liberal, Progressive Conservative and NDP identifiers were strongly associated with votes for each of those parties. Though the coefficient for Reform party identification is not significant at conventional levels, short-term evaluations of the Reform party (as reflected in the feeling thermometer) were quite influential in shaping the vote outside Quebec. “Warm” temperatures on the Reform thermometer were significantly associated with voting for Reform relative to abstaining, and were negatively associated with voting for each of the other parties, relative to abstention. Similarly, evaluations of the NDP were positively associated with the probability of voting NDP and negatively associated with the probabilities of voting Reform and Progressive Conservative, relative to abstention. The importance of issues is also evident, as Quebec support is negatively associated with the probability of a vote for Reform, free enterprise is negatively associated with the probability of voting NDP, and moral traditionalism is positively associated with voting for Reform and negatively associated with voting NDP. Francophones and Catholics outside Quebec were more likely to vote Liberal than to abstain. Older people and those with more interest in the campaign were significantly more likely to vote for each of the four parties than to abstain. Education was positively related to voting Liberal, PC or NDP, and knowledge of party positions was positively related to voting PC, NDP or Reform, controlling for the effects of other variables in the model.
We also see in Table 3 that some voters on the right outside Quebec were strategic in casting their votes. Voters who estimated that the Reform party's candidate had a better chance of winning the riding than the PC candidate were significantly more likely to vote for the newer party. Moreover, Reform party identifiers were especially calculating. Strategic effects on voting for the Progressive Conservative candidate were trivial, except among Reform identifiers. The very large negative coefficient on the interaction between the relative chance of winning and Reform identification suggests that many Reform identifiers were willing to vote for the traditional PCs when their own party's prospects in the riding were dim.Footnote 11
As in our Quebec model, the estimated coefficients were used to generate probabilities of abstention, voting Liberal, voting Progressive Conservative, voting NDP or voting Reform. The correlations between the reported vote and the estimated probabilities were again high for Liberal (.79), Progressive Conservative (.63), NDP (.67) and Reform (.77), but lower for abstention (.49).
Figure 2 presents the effects of simulating increases and decreases in turnout in the rest of Canada, and shows almost a mirror image of the results in Quebec. The overall picture is that higher turnout is associated with a greater Liberal lead over Reform, but the effects are less steep than those observed in Quebec. At the baseline of the actual vote (79.4 per cent of respondents reporting voting), the Liberals hold a decisive 9.7 percentage point lead over Reform. At a higher simulated turnout rate of 89.7 per cent, Reform support decreases, and results in a slightly increased margin for the Liberals of 10.1 percentage points. Similarly, as turnout decreases from the baseline, Reform support waxes and Liberal support wanes. At 63.5 per cent (a rate comparable to reported turnout rates in the United States), the Liberal advantage drops to 8.2 points. Reform draws even with the Liberals only at a very low turnout rate of just below 30 per cent of the CES respondents. As predicted by the conventional model, support for the NDP increases as turnout increases, but the slope is not particularly steep. If turnout was about half of the actual rate, we estimate that support for the NDP would have slipped about one percentage point. If turnout was about 15 per cent higher, NDP support would have increased about a half point. Support for the Progressive Conservatives hardly budges over a wide range of simulated turnout.

Figure 2 Simulated Turnout and Partisan Outcomes: Outside Quebec
Discussion
Turnout matters, but a comparison of the figures for Quebec and the remaining provinces illustrates the strategic dilemma that the Liberals faced in the late 1990s. As the majority government with the greatest influence in setting electoral laws, the Liberal party had a limited ability to affect the level of turnout in national elections. Electoral reforms that could have raised turnout (or mitigated its decline) would have likely hurt the Liberals in Quebec, but would have helped marginally in other areas. While the net effect nationally may have been minimal, those changes could still be important in affecting the composition of the Liberal base. Lower turnout would have shifted the basis of Liberal support toward Quebec, while higher turnout probably would have tilted it away from the majority francophone province. Other parties did not face the same strategic dilemma, even if they had been in a position to affect turnout. Support for the NDP and the Progressive Conservatives did not vary as much by turnout level, and the Bloc Québécois (which would benefit from higher turnout) and the Reform party (which would benefit from lower turnout) had regionally confined constituencies.
The apparent relationship between turnout and partisan outcomes is something of a puzzle, as the class/party cleavage that provides a basis for both the conventional model and the defection model was relatively weak in Canada in 1997. As the conventional model predicts, we see some tendency for Reform, the new anchor of the right, to suffer from higher turnout, and a milder tendency for the leftist NDP to benefit from higher turnout. If we regard the Bloc Québécois as left of the traditional federal parties on economic issues (Nevitte et al., Reference Nevitte, Blais, Gidengil and Nadeau2000: 50), the advantages it receives from high levels of turnout would also fit the predictions of the conventional model. Yet, the alignments both in Quebec and elsewhere belie that logic. In Quebec, the Liberal vote share is highest among those who reported less than $30,000 in income, yet they benefitted from lower turnout. Support for the Bloc was highest among middle-income earners (between $30,000 and $60,000) and among high school graduates without a college education. Outside Quebec, Reform's support was weakest among the most educated, and highest among middle-income earners. The Liberals' vote share was highest among those in the high-income bracket, yet they benefited from higher levels of turnout. Clearly, the class/party cleavage assumptions of the conventional model do not fit the Canadian case well.
We can understand the relationship between turnout and partisan outcomes through the lens of generational politics. As we noted at the outset, the declining rates of turnout in Canada have been largely the result of especially low rates of participation among the young, and age is also correlated with attitudes toward the new party system. The dominant issue in Quebec at the time, of course, was sovereignty, and in our sample, age is negatively correlated with attitudes toward sovereignty (r = −.15). Among Quebec voters under the age of 30, the Bloc had a significant lead over the Liberals (46 per cent to 33 per cent), but among voters over the age of 60, the Liberals led the Bloc (50 per cent to 28 per cent). Thus, in our simulations in Quebec, lower turnout electorates include a disproportionate number of older people who are more sympathetic to continued federation and the traditional federal parties. As turnout increases, they are joined by younger people more sympathetic to sovereignty and less tied to the old party system. Outside Quebec, the new party system was not as closely linked to generational politics. The Liberal party vote share is greater than the Reform vote share across age cohorts, but is greatest among the young. In our simulations, low turnout elections disproportionately attract older voters who are the most sympathetic to Reform, and high turnout elections bring in younger voters who are the least enthralled by Reform. The weaker generational cleavage (on top of a weak class cleavage) results in flatter slopes than those observed in Quebec, but as in Quebec, the party with the greatest advantage among the youth benefits from higher levels of turnout.Footnote 12
This analysis of the 1997 Canadian case provides additional evidence to the emerging consensus that the relationship between turnout and partisan outcomes is context dependent. Previous research has found that the parties of the left generally benefit from higher turnout when the party system is defined by class, but that is less consistent when the class cleavage is undercut by other crosscutting social cleavages. To date, theory has been underdeveloped about which party or parties should benefit from higher turnout in the latter cases. This case suggests that, in some circumstances, an appreciation of generational politics can help us understand the likely partisan consequences of variations in turnout.