Youth Electoral Engagement in Canada
4. A cohort analysis of voter turnout in Canadian federal elections
In this section we revisit and update a cohort analysis of voter turnout in Canada. We refine a methodology initially employed by Blais et al. (2004) and inspired by Johnston (1989, 1992) to sort out life-cycle, generation (cohort), and period effects on turnout, and we update the analysis.Footnote 15 The basic idea is simple. We pool together all the available CES data sets, covering all federal electionsFootnote 16 from 1965 to 2008 inclusively (except for 1972, when no election study was conducted). We have a total of 13 election studies.Footnote 17
We then proceed to a multivariate estimation that includes life-cycle, cohort, and period effects. Life cycle effects simply correspond to the impact of age. The idea is that as people grow older, as they get more involved in their social milieu and develop stronger preferences over time, their propensity to vote increases. The relationship, however, is curvilinear. At the end of the life cycle, turnout decreases slightly, most especially for health reasons. Cohorts are defined in terms of the first election in which people had the right to vote. For instance, those born between 1942 and 1944 had the right to vote for the first time in 1965 (voting age was then 21). We estimate a model with 14 cohorts, each defined on the basis of the first election in which they were eligible to vote. We have also created variables for each election, which measure the impact of whatever specific contexts of an election that made the election more or less "exciting".
On the basis of the results of the multivariate model (Table A2 in the Appendix) we are able to estimate the turnout rate of the various cohorts at each federal general election.Footnote 18These estimations are presented in Table 5. A number of interesting patterns emerge. Reading horizontally, we can see the life-cycle effect: the turnout rate of a given cohort tends to rise over time, as people grow older. However, the rate at which it increases declines over time. This is the general trend but there are specific "period" effects, that is, turnout is particularly low or high in some elections. We can observe, for instance, that turnout decreases among all cohorts from 1979 to 1980 and increases substantially from 1980 to 1984, among all cohorts.
Generation effects can best be appreciated by comparing the initial turnout rates of the various cohorts. In the 1960s, about 70% of the members of a new cohort would vote in the first election in which they were eligible to participate; by 2004 it was only slightly over 30%. At least two-thirds of new voters would cast a ballot in the 1960s; by 2004 it was about one third.
As shown in previous analyses (Blais et al. 2004) this is the major reason why turnout has been going down. Turnout decline among new cohorts started in the 1970s, and has proved to be quite steady. The turnout rate of new cohorts had already declined to about 50% in the 1980s and into the 40% range in the 1990s. There seems to be a persistent downward trend in the turnout rate of new cohorts. The consequence of this is that despite the fact that young voters are more likely to vote as they get older, they are beginning at such a low level of participation that overall turnout can only be expected to decline.
Data are drawn from Canadian Election Studies conducted for every election between 1965 and 2008, except 1972. Each cell presents the probability of an individual voting given their cohort and the election. The estimates are drawn from the model reported in Table A2.
Return to source of Footnote 15 Life-cycle effects are those that take place as people get older, generation effects refer to differences between groups of people born at different times, and period effects refer to over-time changes that affect all individuals. Period effects may point to specific events, such as pre and post World War II, or to specific characteristics of a particular election. We have made two refinements to the methodology used by Blais et al. (2004). Generations are here defined in terms of the first election in which a group of individuals had the right to vote instead of the rather crude distinction between pre-baby-boomers, baby-boomer, born in the 60s, and born in the 70s. And we create a separate variable to tap the specific peculiarities of each election instead of a simple contrast between elections held before and after 1990.
Return to source of Footnote 16 The 1965 election does not appear because it is used as the reference group. We are thus measuring how more (or less) likely people were to vote in each election, compared to 1965, everything else being equal, that is controlling for life-cycle and generation effects.
Return to source of Footnote 17 Blais et al. (2004) examined nine elections, starting in 1968 and ending in 2000.
Return to source of Footnote 18 These are only estimations and they are valid only to the extent to which the assumptions that we have to make are plausible. We have to assume that the over-reporting bias is relatively constant across cohorts and age groups and also that the life-cycle effects are relatively constant across cohorts. Our intuition is that these assumptions are not strictly true but that the biases are not large and that they partially cancel out. Note that while we include variables for those who were eligible to vote for the first time in the 2006 election and the 2008 election, (see Table A2), we do not provide an estimation of the turnout rate for these voters, as the number of voters is too small (69 and 122 respectively).