If, God forbid, you are so annoyed with the major political parties that you are thinking of not voting at all, think again. Because if you vote, your money will talk, no matter who ends up in Ottawa on your behalf.
John Ibbitson, The Globe and Mail, May 25, 2004
Introduction
The 2004 Canadian federal election was contested under substantially different campaign finance rules than the previous election. Bill C-24 severely limited corporate and union donations to political parties, and also limited the size of individual donations. As a result, it made Canadian political parties more dependent on state funding than ever before. The bill also changed the determinant of the level of state funding for national parties, tying it to absolute votes, rather than incurred election expenses. This combination of measures could be expected to have at least two effects, one of which we put to test in this paper.
First, by seriously limiting corporate donations, the bill was seen to decrease parties' reliance on such donations and the quid pro quo that may have followed. We leave this claim to others to investigate. Second, by tying the level of state support to a party's raw vote total, the bill could be a tool to arrest declining voter turnout. If every vote counted—indeed, each vote was worth CDN$1.75 to a party—then parties would seek more voters, regardless of their own position vis-à-vis their competitors. Similarly, individuals would face a greater incentive to vote rather than abstain, as a ballot cast for a third place or worse party would not be wasted, but would instead endow that party with future financing.
We wish to test the impact of Bill C-24 on turnout through four simple but critical tests. We aim for an accumulation of several pieces of evidence, to ascertain the possibility that C-24 increased turnout. Specifically, we present two aggregate-level and two individual-level tests. Each examines the possibility that C-24 increased turnout. We find no evidence that C-24 increased turnout in the 2004 election.1
To put it more explicitly, we have not excluded any test results that suggest C-24 increased turnout.
Aggregate-Level Tests
We begin by considering evidence at the constituency or aggregate level. We create a test of the effect of C-24 by considering the effect of the closeness between the first- and second-place parties on turnout at the constituency level. Following Cox (1997), we know that individuals who prefer a party in third place have little incentive to vote for that party when its place is clear. Indeed, an individual whose preferred party is in third place should defect to her preferred party among the top two. If the individual is indifferent between the top two parties, or if the margin between the top two parties is wide, she should abstain. It follows from this that in more competitive ridings, where the distance between first and second is smaller, we should witness higher turnout. That is, closeness should drive turnout. Conversely, in ridings where the margin between first and second is larger, turnout should be lower.
How should C-24 change this relationship? If C-24 increases an individual's incentive to vote regardless of the closeness of the race, then turnout should be influenced less by closeness. To test if this is the case, we specify a simple regression between turnout and closeness with an interaction term for 2004 and closeness.2
This model may seem too simple, as it ignores any number of other possible factors, including riding-level demographics or provincial variations in turnout. We have two responses. First, we have specified models with provincial dummies, and the substantive findings of our model do not change. Second, we take seriously Clarke's (2005) argument that if we do not know the true model of the phenomenon we are specifying, then any addition of variables can increase bias, even if they are a part of the true model. In short, we side here with simplicity.
Closeness and turnout in 2000 and 2004
Three results are of note. First, closeness does predict turnout as expected, though rather weakly. That is, in a race where the margin of victory increases by 10 percentage points, turnout would on average decrease by 0.97 percentage points. Moreover, on average and at the same level of closeness, turnout was −1.55 percentage points lower in 2004. More importantly, the interaction term is not significant, which indicates that closeness had the same impact on turnout in 2004 as it had in 2000.
There is a second possible test at the aggregate level. As C-24 rewards parties for each vote obtained, it makes no distinction between the ridings from which votes are drawn. Parties should thus have an incentive to seek votes in non-competitive constituencies where they have little chance of winning, but can likely increase their absolute number of votes. In other words, parties and candidates should be less concerned in 2004 about the competitiveness of a constituency when seeking votes. We can test this through a similar regression as above.
To conduct this test, we regress a candidate's spending in a riding on the margin of victory in that riding, a dummy for 2004, and an interaction between 2004 and the margin of victory. The less close the race, the less spending which should occur, on average, as trailing candidates will curtail their efforts, and winning candidates may similarly let up. Accordingly, the margin should produce a negative coefficient. If C-24 has the expected effect, then the interaction term should be significant, and the coefficient should be positive. A positive coefficient indicates that the closeness of a race between first and second has a weaker effect on candidate spending in 2004 than in 2000.3
It can be argued that we should employ national party spending, rather than local candidate spending, as it is parties that benefit from the increases in vote count, not local candidates. We agree but we note two measurement issues that make this impossible. First, it is extremely difficult to track national spending geographically in Canada. National party spending reports are not itemized geographically. Rather, they are categorized by spending type. Accordingly, we cannot locate national party spending down to the local level. Second, even when national parties do make transfers to local candidates, the differing accounting practices and financing schemes of the various parties makes it very difficult if not impossible to develop comparative measures. Nevertheless, C-24 provides an incentive for parties to transfer money to candidates irrespective of the closeness of the race in the constituency. If this is the case, holding all else equal, total candidate spending should be more weakly related to the closeness of the race in 2004 than in 2000.
Table 2 presents our results. The results are opposite to expectations. Indeed, margin of victory exerts more of a negative effect on party expenditures in 2004 than in 2000.4
We performed one additional test, focusing on the spending of NDP candidates in Quebec in 2004. The NDP traditionally runs weak and poorly financed campaigns in this province, and it receives low vote totals as a result. While the ceiling on their support in the province is low, their vote is likely not maximized. Accordingly, the party should have a greater incentive to spend in weak ridings in 2004, where the returns for spending are greatest. However, regressing NDP candidate spending in Quebec on closeness, a dummy for 2004, and an 2004*closeness interaction shows the same negative sign for the interaction term as above (b = −184.1, s.e. = 113.3).
The dummy variable for 2004 is significant and substantively large. Accordingly, candidates spent more on average in 2004 than in 2000. However, the key fact remains that, as margins increased, candidates reduced their spending faster in 2004 than in 2000. Moreover, there are good reasons why average spending should have increased in 2004: the NDP was richer in 2004 than 2000, and the folding of the PC and the CA into the Conservative Party eliminated low-spending PC candidates.
Competitiveness and candidate spending in 2000 and 2004
Individual-Level Tests
Having found little evidence at the constituency level, what evidence can we locate at the individual level? We conduct two tests, both using the 2000 and 2004 Canadian Election Studies. Our first test follows from our earlier logic. Our second follows an earlier test of the determinants of individual-level turnout (Blais et al., 2002), with a focus on the individual-level factors on which C-24 should have an effect.
For our first test, we apply Cox's (1997) logic (laid out above), and we examine the behaviour of those individuals whose first preference was for a party that finished in third place or worse. We define as preferred the party that receives the highest rating in the pre-election wave of the CES (we exclude ties). These individuals had three options: to abstain, to vote for their preferred third-place party, or to desert and vote for one of the top two parties. We would expect to see more “sincere” votes for the third-place party in 2004 than in 2000, and less abstention or defection to a second-most-preferred party.
For a voter who prefers a third-place or worse party, Bill C-24 should increase her likelihood of voting for that party, as she can endow her preferred party with $1.75 per year, regardless of how far behind it lags.
Accordingly, we expect to see less abstention or desertion in 2004 than in 2000. We classify as deserters those voters whose preferred party finished in third place and who voted for the first- or second-place party. This party must be their second preferred choice. If C-24 is effective, it should decrease incentives to desert, as voters will be faced with a choice between endowing a continuing benefit to their preferred party, and playing a likely very marginal role in choosing between the top two parties.
Finally, we should expect to see more third-party voters. We regard third-party voters as those who prefer and vote for a party that finished third or worse. Put another way, if C-24 was effective in increasing turnout, it should render less effective the individual-level mechanism of Duverger's law (1954). Did C-24 achieve these goals?
Table 3 indicates that there is no significant difference in the propensity of individuals with a third-place preference to sincerely support their preferred party, desert to one of the top two parties, or abstain. There is no evidence that things were different in 2004 from what they had been in 2000.
Vote choice of voters with a preference for a third-place party in 2000 and 2004
We conducted a final test of the individual-level determinants of the decision to turn out, taking into consideration all voters. We largely follow the test of Blais et al. (2002). Our model regresses the respondent's reported decision to turn out on age, income, education, political interest and party identification (variable descriptions are included in the appendix).
Added to these standard variables are three key variables for testing the effect of C-24. The first is Closeness. As above, this measures the distance between the winning and second-place candidates in the respondent's constituency. We should expect a negative coefficient, as the larger the margin, the less incentive the individual has to vote, as her ballot is less likely to be pivotal. We add to this a dummy for the 2004 election, and an interaction between 2004 and Closeness. If C-24 reduced the importance of strategic considerations in 2004, and increased the incentive for individuals to turn out in 2004, then the interaction term between Closeness and 2004 should be positive, lessening the marginal effect of the competitiveness of a race in 2004.
As Table 4 demonstrates, the interaction term between Closeness and 2004 is not different from zero. Moreover, Closeness alone is not significant. Rather than strategic considerations carrying the day, it seems that standard demographic and psychological explanations are driving the turnout decision. To wit, age, university education, high income, political interest and party identification all predict turnout in the expected direction, and do so significantly.
Individual-level correlates of turnout
Conclusion
Whether Bill C-24 was a “good” bill or not is a question we leave to others. Whatever its theoretical merits, we are interested in understanding its effects. As it relates to voter turnout, the legislation seems to have achieved little. With four simple tests, we have shown that turnout was not differently affected by closeness in 2004 than in 2000; that parties' efforts were not greater in 2004 in more marginal ridings; that there was no difference in the likelihood of abstaining or deserting a preferred third-place party in 2004 and 2000; and that at the individual level, the decision to turn out was not affected differently by strategic considerations in 2004. While another election may change these findings, the current evidence indicates that Bill C-24 has not made measurable strides in arresting declining turnout.
Why did Bill C-24 not exhibit these possible effects? We think at least three explanations have some merit. First, the provisions of the bill were not well known to voters. Indeed, by 2005, only one-in-four Canadians were aware of the $1.75 provision (SES Research, 2005). Second, parties may not have responded to the new incentive structure. They were certainly aware of the legislation, as it was passed into law in June 2003, a year before the next election. Moreover, the parliamentary testimony of both large and small parties prior to the bill's adoption suggests that they understood the logic of the new funding formula (House of Commons, 2003). However, parties and candidates are not perfectly rational actors with the ability to take advantage of every opportunity in an election. The distribution of resources is as likely dominated by past practices, intuition and instincts as by calculation of future costs and benefits. Third, the provisions of the bill created a principal-agent problem. The bill provided incentives to the national party, but mobilization generally occurs at the local campaign level. Accordingly, national parties were faced with the difficult task of convincing local candidates to find and spend more resources. This is at worst impossible and at best difficult to implement.
Acknowledgments
We thank SSHRC for funding in the form of a doctoral grant. We also acknowledge the support of the Canada Research Chairs programme. We are indebted to Munroe Eagles and Lisa Young, who provided very useful comments, as did our anonymous (at the time) reviewers. All remaining errors are our own, randomly distributed between the authors. Replication code and data are available upon request.
Appendix A—Description of Variables
Constituency turnout: rate of participation in a given federal constituency.
Closeness: distance between the first- and second-place party in each federal constituency in the 2000 or 2004 election, measured in percentage points from 0 to 100.
2004: dummy variable indicating that the observation is from 2004 (1) or 2000 (0).
Closeness*2004: multiplicative interaction between 2004 and Closeness.
Campaign spending: total spending in dollars of major party candidates in the 2000 and 2004 elections, as reported to Elections Canada.
Abstainers: respondents who indicated a preference for a third-place or worse party and chose not to vote.
Deserters: respondents who indicated a preference for a third-place or worse party yet voted for a party that finished in first or second place, which represented their second-most-preferred choice.
Third-place voters: respondents who indicated a preference for a third-place or worse party and voted for their preferred party.
Turnout: dummy variable indicating that a respondent voted (1) or abstained (0).
Age: age in years, from 18 to 102. Don't knows and refused are classified as missing.
Income: dummy variable indicating a household income of $100,000 or greater. Don't knows and refused are classified as missing.
University: dummy variable indicating the respondent graduated from university. Don't knows and refused are classified as missing.
Interest: 0 to 1 scale measuring interest in politics from “no interest at all” to “extremely interested.” Don't knows and refused are classified as missing.
Party ID: dummy variable indicating the respondent is a strong or moderate identifier with one of the major parties.
