Explaining the Turnout Decline in Canadian Federal Elections: A New Survey of Non-voters

3. The Correlates of Not Voting

To begin our examination of the correlates of not voting, we will assemble a variety of possible predictors, many of which were identified in our previous report. We will enter into our regression equations several socio-demographic characteristics of respondents:

• 1. Age, as measured by year of birth.
• 2. Education, as measured by the highest level of formal education achieved.
• 3. Income, as measured by total household income.
• 4. Gender.
• 5. Place of birth, whether inside or outside Canada.
• 6. Mobility, as measured by the length of residence in the current neighbourhood or community.

Most of these socio-demographic factors are expected to relate to not voting. Our hypothesis is that young people, as well as people with less education and lower incomes will be less likely to cast a ballot in any given election. We also expect that people who were born outside Canada will be less familiar with the country's politics, and less likely to vote, and that those who are more geographically mobile will also be less likely to vote, as they may be less familiar with the political situation of the area to which they have recently moved. We have no hypothesis about gender differences, but include that as it may be of interest.

We will pay particular attention to the variable of age. We know, of course, that age is related to not voting, and always has been: younger people vote at lower rates than older people. If we want to explain why some people vote and others do not, on any given occasion, the variable of age will be an important predictor of the phenomenon. However, we also want to pay attention to age in another form, that of cohort (age group). The youngest age group, under 25, many of them newly eligible, has been particularly likely to abstain from voting in recent elections. We will analyze age cohorts in detail in this report.

These cohorts are formed by grouping ages according to birth year, as shown in Table 13. This treatment allows us to consider patterns that are specific to particular elections, and/or to more broadly defined political eras. The election(s) in which a particular group first became eligible to vote is denoted by the column labelled "first eligibility". Respondents born between 1971 and 1975, for example, would have been between the ages of 25 and 29 at the time of the 2000 federal election, and would have first become eligible to vote in 1993. The number of cases for the total study (including the screening interview) falling into each group is shown in the column labelled "N", and the number of cases of confirmed non-voters is indicated in the column labelled "NV".

Table 13 Distribution of Cases by Age Cohorts

While a more extensive use of the age cohorts for analysis will occur later in the report, we present here the basic pattern of voting and not voting (Table 14). This table uses the full study, including the screening interview respondents, since "age" and "vote" were two of the very few questions asked of everyone contacted. Because this overall sample over-represents those voting in 2000, we have entered a corrective weight into Table 14 to put voters and non-voters in their correct proportions for that election.⁶

Table 14 Voting and Not Voting in 2000, by Age Cohort

The differences in voting among the age cohorts are extraordinary. The drop-off in electoral participation steadily increases as the cohorts get younger. There is even a slight difference in the voting rates of some of the older cohorts in the study, with those having entered the electorate during the early Trudeau years (aged 48 to 57 in 2000) participating at lower rates than those who entered at earlier periods of time. Those who entered the electorate during the later Trudeau period (aged 38 to 47) voted in 2000 at lower than a two-thirds rate. For those who entered the electorate during the Mulroney years (aged 30 to 37 in 2000) the overall percentage that cast a ballot in 2000 was only 54.2 percent. From that point forward, the voting rate slips to well below half, with the cohort entering the electorate in 1993 voting at 38.2 percent, the 1997 cohort at 27.5 percent, and the 2000 cohort voting at only a 22.4 percent rate.

Thus, the entry of cohorts of new electors who participated at particularly low rates in the last three elections has played a major role in the turnout decline during this period. Table 14 shows that this lessening of electoral participation with subsequent age groups is not a recent phenomenon, but dates back to those who entered the electorate in the 1970s, if not earlier. The life-cycle effects, which work to increase the voter turnout rates of initially-lower young cohorts, have not brought the Trudeau and Mulroney cohorts up to the levels of the King, St. Laurent, Diefenbaker and Pearson generations. The outlook is even worse for the Chrétien generation, entering from 1993 to 2000, since they are starting their voting rates at such low levels. If life-cycle effects continue their reduced impact on young citizens, the most likely outcome is that voting rates will continue to decline.

Another factor of potential importance is that of region. In particular, we know that turnout in the 2000 election in Ontario was only 58 percent (Results of the 2000 election, Elections Canada Web site, Table 4). Newfoundland (57.1 percent), and the Northwest Territories (52.2 percent) were also well below the average (61.2* percent), and Alberta (60.2 percent) was slightly below average. However, a preliminary breakdown of important factors by province does not show significant differences between Ontario (the area where we have sufficient cases to be confident of our results) and the national results.

Table 15 Factor Analysis of Variables Related to Interest, Civic Duty and Party Competition

Our second category of predictors of voting and not voting will be derived from two factor analyses, reported in tables 15 and 16. As we have mentioned previously, factor analysis explores the correlations among all the items in a group of variables, and identifies any common underlying commonalities, or factors, which lie behind them. The factor loadings (correlations of the individual variables with the underlying factors) are presented in tables 15 and 16, and factor scores will be used in the ensuing regression analyses.

Table 15 reports a factor analysis of variables including political interest. Our purpose in using this technique is to observe which variables load on the same factor as political interest, so as to use a more complex factor to predict voting. Political interest by itself raises as many questions as it answers, as we have mentioned before. We have included in the factor analysis a measure of "civic duty", namely the perceived importance of voting in elections. We also included two "political discussion" variables, one measuring socialization, that is, whether people discussed politics with their parents when they were growing up, and one measuring current frequency of discussing politics with other people. We are also interested here in the impact of a perceived competitive situation on voter turnout, on the hypothesis that if people feel their vote will matter more they will be more likely to vote, and that if the political parties in the country, and in the respondent's riding, are seen as more competitive, the vote will also seem to matter more and turnout will be higher. All of these hypotheses are consistent with a rational-choice approach to voter turnout, which would say that people are more likely to act when it is in their personal interest to engage in that activity rather than some possible competing activity.

Three factors are produced from the group of variables described in the previous paragraph. The first groups political interest with civic duty and with discussing politics, both in the past when growing up, and at present with family and friends. We might refer to this as an "engaged citizen" factor. The second also includes the civic duty variable of considering that it is important that people vote in elections, but groups it with the two questions about whether people felt their votes would make a difference, in the country as a whole, or in their electoral districts. These latter two variables have much higher loadings on this factor, and so we might label this a "vote matters" factor, hypothesizing that people are more prone to act in circumstances where their vote might make a difference or where it is important to the country that people take part in elections. It can be considered to be consistent with rational-choice theories in the sense that it will be in the elector's self-interest to vote in situations where that vote would make a difference, or "count more", since that would give more value to the choice of that action as opposed to some competing action. The third factor in Table 15 groups the two variables that ask the respondent to rate how competitive they found the political parties to be in the country as a whole, and in the electoral district. This variable is also consistent with the rational-choice approach, and we can call it a "party competitive" factor. It is interesting that this factor is distinct from factor 2, involving the questions about whether people felt their vote would matter. The fact that civic duty loads on the second factor and not on the third implies that other considerations than party competition are involved in people deciding whether their vote would matter or not.⁷

Table 16 Factor Analysis of Variables Related to Efficacy, Trust and Party Support

Table 16 presents the second factor analysis. Here we have included variables measuring the concept of political efficacy, the feeling that one can understand and potentially influence the political process. Also included are measures of trust, in politicians and the political system more generally. Finally, we have included a set of measures of attitudes toward political parties, the primary agents of political representation. A number of these variables measure the degree of positive or negative feelings people have toward parties, as, for example, representing regions or the country as a whole, confusing or clarifying the issues, and being influenced by "people with lots of money."

Three factors are produced in this analysis, as in Table 15. The first is one we will label "inefficacy, cynicism, negativity to parties". The questions with high loadings on this factor include such classic "low political efficacy" questions as "those elected to Parliament soon lose touch with the people," "people like me don't have any say about what the government does," and "I don't think that the government cares much what people like me think." Party-related measures which also load on this factor represent agreement with statements that the parties are all the same, that they confuse the issues, that they represent regions of the country to the detriment of national representation, and that they are too influenced by people with money. The second factor represents "political trust", including such items as "most of the time we can trust people in government to do what is right" and "those elected to Parliament represent the diversity of Canadian society." Factor three groups two of the "party support" questions, "political parties are the best way of representing people's interests," and "during election campaign periods, political parties and candidates discuss issues that really are of interest to voters."

The two factor analyses reported in tables 15 and 16 have produced six factors, and respondents' scores on these will be entered as additional predictor variables into the regressions to follow, joining the socio-demographic predictors already listed. They are:

• 7. Attitudes reflecting interest, discussion and civic duty
• 8. Sense of votes mattering and civic duty
• 9. Attitudes towards parties being competitive; since this variable relates to competitiveness in the 2000 election, it will be used in the regressions in tables 17 and 20 only
• 10. Feelings of inefficacy, cynicism and negativity about parties
• 11. Feelings of political trust and effective representation
• 12. Expressions of support for parties

In addition, we will include in the upcoming regressions one measure of an administrative nature, asked of both voters and non-voters. We will include this measure in all the regressions except that for 1993, since administrative problems in 2000 may be an indication of previous administrative problems as well. For example, the Register of Electors used in the 2000 election was a result of the 1997 list of electors prepared through a door-to-door enumeration. Someone not on the 1997 list would only be on the 2000 list if they took action to register themselves.

• 13. Whether a person's name was on the list of electors in 2000

And in the 2000 analysis only, we will use an additional variable, asking whether the respondent had received any contact from the parties or candidates:

• 14. Report of contact by parties or candidates in 2000

In the four regressions to follow, three of the dependent variables, votes in the 2000, 1997 and 1993 elections, are nominal, that is, have only the two categories: voted, or did not. Some analysts object to the use of OLS regression for nominal dependent variables, and other techniques have been developed for these situations, like probit and logistic regression. We prefer to report the OLS regressions, because it is the easiest technique to interpret, but before doing so, we have run confirmatory logistic regressions for the predictors of voting in 2000, 1997 and 1993. These results show the same variables as important and statistically significant predictors as those appearing in tables 17, 18, and 19. Table 20 has an ordinal dependent variable, and the use of OLS regression with such variables is commonly accepted.

Table 17 Predictors of Voting/Not Voting in 2000 (Multiple Regression)

Table 18 Predictors of Voting/Not Voting in 1997 (Multiple Regression)

Table 19 Predictors of Voting/Not Voting in 1993 (Multiple Regression)

Table 20 Predictors of Voting Frequency in Elections of 1993, 1997 and 2000 (Multiple Regression)

The series of regressions reported in tables 17 to 20 uses the predictors of voting or not voting we have developed above, in the 2000 federal election, the 1997 federal election and the 1993 federal election, and a composite index of the frequency of voting of respondents in all three of these elections. The categories of that last variable of voting frequency range from 3, for people who voted in all three, to 0 for those who did not vote in any. Only those respondents who were eligible to vote in all three were included.

There are many similarities among the results of the four regression analyses, allowing us to identify the most important factors in not voting in recent elections. In all four instances, age emerges as the number-one predictor of whether someone voted or did not. As measured by the Beta statistic, which standardizes the regression coefficients measuring the change in the dependent variable effected by one unit of change in the independent variable, age is usually a substantially larger coefficient than the second-highest. In Table 20, where the dependent variable is frequency of voting in the three most recent federal elections, its Beta coefficient is -.396, while the next factor, the attitudes of interest, discussion and civic duty, is -.239. (The minus signs, incidentally, are simply the result of the direction of the coding of the variables. Age is measured by year of birth, which goes from low to high. Frequency of voting goes from 0 to 3. Therefore, the minus sign of the coefficient means that the youngest people (born in the later years) are less likely to have voted.)

As has been mentioned, the two factors involving citizen duty have significant connections to voter turnout. The interest, discussion, civic duty factor (#7) is the second most important in predicting voting frequency, and third in 2000. The other related factor, vote matters, civic duty (#8), comes in fourth place in predicting voting frequency, is second in 2000, and fourth in 1997 and 1993. The other attitudinal factors derived from the factor analyses (#9 – 12) do not reach statistical significance, with the exception of the trust factor in 2000, which is a weak predictor of voter turnout.

Another predictive factor of importance is the one measuring administrative effects, namely the respondent having his/her name on the list of electors in 2000. This factor is fourth in importance in explaining voter turnout in the 2000 election, but also seems to be measuring effects that were important in previous elections. It is actually the second strongest predictor of not voting in the previous election of 1997, as well as third in predicting voting frequency in the three elections. Although it might appear puzzling that not being on the 2000 list is important in explaining not voting in previous elections, it must be remembered that the 2000 list was composed from an enumeration in 1997. Thus, this variable picks up administrative difficulties from a previous time period, and makes it appear that the same people may suffer from them consistently over time.

Some additional factors reach statistical significance in tables 17 to 20. In terms of socio-demographic variables, higher income is associated with higher voting frequency, and also with turnout in the specific elections of 1993, 1997 and 2000. Being new to Canada, as measured by whether respondents were born in this country or not, is associated with lower turnout. So is geographical mobility, as measured by the length of residence in one's current neighbourhood or community. Finally, in the 2000 election, where we were able to measure this factor, being contacted by the parties or candidates is correlated with higher voter turnout.

In re-running Table 17 for Ontario only, we determined that the order and approximate impact of the significant predictor variables of voter turnout was the same as shown in that table for the country as a whole. Therefore, the fact that Ontario had a lower than average turnout in 2000 was not due to the influence of any unusual factor specific to that province.

• Previous
• Table of Contents
• Next