PDF
Understanding Voter Turnout in Canada: What Data Do We Lack?
Christopher H. Achen

Voter turnout, particularly among youth, has been in decline over the past few decades. Federal officials have expressed concern about this trend. Although they have sought help from researchers to understand the reasons for the lack of participation in hopes of reversing it, scholars lack some of the information they need to confidently advise policymakers and their fellow citizens on how to get more ballots cast. In this article, the author outlines the main factors/variables which explain voter turnout. He then explains why researchers require supplementary information that only official government records can supply to properly consider these variables. Two sources of official information are highlighted as being particularly relevant—official turnout records and unemployment surveys with a voting supplement. The author concludes by offering three recommendations for how to make this information available to researchers while still taking steps to protect Canadians’ privacy.

Introduction¹

Like most democracies in recent decades, Canada has experienced a decline in turnout (see figure 1). Voting among Canadian youth has fallen particularly dramatically. When turnout falls, both the representativeness of the electorate and the legitimacy of election outcomes come under scrutiny. Federal officials have expressed concern, and for a decade and a half, Elections Canada has commissioned research on the topic, including repeated special surveys on youth turnout beginning with Pammett and LeDuc in 2003 and continuing to 2015.² Thus, turnout matters both as a research puzzle and as a policy issue. Yet understanding the decline, particularly among younger voters, continues to challenge scholars.³

At present, a lack of relevant data blocks researchers from confidently advising policymakers and fellow citizens on how to get more citizens to cast a vote. We simply do not have the information we need. This article reviews the problem, with an emphasis on Canada and to a lesser degree on the United States. However, the problem is familiar in the rest of the democratic world as well.

The Main Factors in Voter Turnout

The standard variables in use in turnout studies of individual voters fall into three broad categories:

1. The turnout decision itself. Did the citizen cast a ballot?

2. Demographic variables. Here we include the classics known to predict turnout, especially age and education, along with a variety of other factors such as residential location, income, gender, race and ethnicity, religious preference and church attendance, union membership, and other group affiliations.

3. Attitudinal variables. A citizen’s sense of civic duty and the strength of preference for candidates are the most powerful factors influencing turnout, a finding that dates to Riker and Ordeshook.⁴ Policy views, candidate evaluations, partisanship and partisan strength, media consumption, information levels, and a host of other variables all matter to some degree.

Academic election surveys, notably the Canadian Election Study, include all these variables. However, these surveys on their own are insufficient. They need supplementary information that only official government records can supply, as the next sections explain. Two sources of official information are particularly relevant—official turnout records and unemployment surveys with a voting supplement. The next two sections take them up in turn.

Figure 1. Canadian Federal Turnout since 1968⁵

Why Official Turnout Records Are Needed

In the great majority of academic studies, turnout is measured by asking the citizen in a post-election interview whether she voted (“reported vote”). In many internet surveys, finding people post-election is deemed too difficult, and the citizen’s pre-election “intention to vote” is used instead. Only a handful of studies have used the official government record of whether the citizen cast a ballot (“validated vote”).⁶

Vote intentions and reported votes each have well known problems. Good intentions (to lose weight, to quit smoking, and to get to the polls) often fail.⁷ Reported votes are also unreliable in every democracy.⁸ As many as one quarter of nonvoters falsely report that they voted (“misreport”), inducing substantial error in the turnout measure. Overreport – the combination of misreport plus the greater willingness of more politically engaged citizens to be interviewed – has grown worse, making reported turnout rates in the Canadian Election Study now more than 20 points higher than the actual rate. As recently as the 1970s and 1980s, reported vote was not too misleading,⁹ but trusting it has become more difficult in recent years.¹⁰ In consequence, Gidengil et al.¹¹ omitted a planned chapter on turnout from their book on recent Canadian elections.¹² Without knowing who in the survey had actually voted, the researchers were stymied.

Thus, validated vote is the gold standard, the only genuinely reliable source of turnout information. However, to make use of official vote records, scholars must have access to them. That is currently impossible in Canada.

Official Canadian eligible voter files are treated as confidential, almost as state secrets. In contrast to Britain and the United States, Canada does not make them available even to political parties, and certainly not to academic researchers, not even in redacted form with no identifying information. Moreover, the record of who voted is not recorded in the voter file itself, and turnout information is destroyed within one year after each election, as specified in the Canada Elections Act. Thus in Canada, even the voter files do not include validated turnout information. In consequence, there has never been a comprehensive voter turnout survey in Canada with validated votes. Even when Elections Canada, the agency responsible for conducting federal elections and for maintaining the federal electoral rolls, has commissioned surveys to help understand low youth turnout, reported vote was used.¹³ No vote validation was done, raising some questions about the findings.

Canadian rules are very different from their American equivalents. In the U.S., voter files are a state responsibility, and each citizen’s appearance at the polls (or casting of a mail ballot) is recorded at each election. The cumulative record is maintained so long as the citizen is resident at the same address. With some qualifications, the records are essentially public information.¹⁴ Thus with time and effort, American academic surveys can validate their turnout reports.

Maintaining U.S. voter turnout records is not thought to be onerous for the states. California, with a population larger than Canada’s, maintains a high-quality record of turnout for each citizen. Many advanced democracies, such as Germany, Sweden, and Japan do the same, though their records are not public. Even Britain, which has turnout recordkeeping laws like Canada’s, has permitted researchers to use validated turnout information for several British National Election Studies in the 80s and 90s.¹⁵ Thus, in its pursuit of voter privacy, Canada has become an outlier among advanced democracies in not maintaining key administrative records on the functioning of its democracy. Of necessity, therefore, Canadian scholarly studies of federal turnout have been forced to rely on self-reports from surveys, with all their attendant errors, if turnout is studied at all.

Elections Canada has done validated-vote studies internally after the last five federal elections, sampling from its own voting records and making use of occasional academic consultants.¹⁶ The sample sizes are very large—more than half a million voters in 2016, for example. These studies are very helpful and should be continued, as Canadian scholars have stressed.¹⁷ The surveys are not comprehensive: the turnout records include very few demographic variables (age, gender, and provincial residence, but not the powerful factor of education, for example) and no attitudinal data. Even so, it would be very helpful for researchers to have access to the data. However, those internal data files have not been released to scholars interested in extending the results, as has been done in Taiwan, for example, another democracy with strict privacy laws.¹⁸

Canadian provinces maintain their own voter rolls for provincial elections. In Québec, the voter file is updated with the voter’s actual turnout at each election, and the complete longitudinal record is kept in Québec City, just as American states do. While the files remain confidential, one researcher (François Gelineau of Laval University) has been given access to the entire file. Thus, at least in one province, the files themselves are maintained and made selectively available Hence, a follow-on survey with vote validation might be possible in Québec, though none has yet been carried out to my knowledge.

In light of Canadian privacy laws, it is important to understand that what researchers need and what identifies individuals are quite different. Scholars do not need names, exact addresses, or exact ages to study turnout. “Age 40-45, male, and lives in northern Manitoba” suffices for research purposes, and it certainly does not identify anyone uniquely nor threaten anyone’s privacy. Thus, releasing either the national vote file or Elections Canada’s internal samples, with validated turnout recorded but other information anonymized in this fashion, would not in any way violate the secrecy of individual turnout records.

Validating turnout in external academic surveys raises a different set of issues. In that case, survey respondents need to be linked to their official validated vote records. Doing so requires that researchers have access to the full national voter file with validated turnout recorded for each voter.¹⁹ At present, no such voter file exists in Canada. But if it did, it could be released on a restricted basis to scholars who could demonstrate a valid research need for it. And if even restricted release of the voter file is impossible under current interpretations of Canadian confidentiality laws, access could be provided in a “clean room” like those used in the U.S. for access to Census records. Statistics Canada already has a procedure of this kind, using Research Data Centres (RDCs) for some of its sensitive data.²⁰ Alternately, Statistics Canada might do the turnout validation themselves in return for a user fee. Then the full voter file itself would not need to be released. In all such cases, of course, the usual confidentiality rules would have to be observed, but that ethical norm has been virtually universally honored in academic survey research. A validated vote study would present no new obstacles.

Thus, the Québec precedent is an important one for Canadian turnout studies. Releasing a redacted version of Elections Canada’s internal studies, and creating a national voter file with turnout recorded for each citizen that could be used to validate self-reports from surveys, together would add considerably to our knowledge of Canadian turnout, why it has been falling, and why Canadian youth have been slow to learn to vote in recent years. Under current administrative and legal interpretations, however, these data releases have not occurred – only unvalidated turnout reports are available. As mentioned earlier, unvalidated reports have caused some of Canada’s most sophisticated scholars to abandon the study of turnout, making progress difficult at best.

Labour Force Surveys

In the mid-60s, the U.S. began adding a registration and voting supplement to its Current Population Survey in November of even-numbered years, the dates of presidential and congressional elections. The Canadian equivalent was carried out for the first time after the 2010 federal election. The work was done by Statistics Canada as part of their Labour Force Survey (LFS), paralleling the U.S. procedure. Elections Canada paid for the add-on, which is voluntary for respondents but has achieved a very good response rate. The series has been continued with each subsequent election. As in the U.S., the sample is large (currently more than 50,000 households), stratified by province. The sample size is far beyond that of any academic survey, and thus the LHS is extremely valuable for studying provinces individually. Reported vote is the turnout measure. A large number of demographic and economic variables are included, but political attitude data are not.

A few tables are released from each LHS study; for example, reported turnout by age and education, with some breakdowns by province.²¹ The descriptions reported are valuable, but there is a great deal more that outside researchers could do with the data file. For example, the discrepancies between Election Canada turnout rates by age (based on validated vote from their internal studies) versus those in the LHS data (using reported vote) could be addressed, and research done to reweight the data for comparability. However, a redacted version of the LHS file (preserving anonymity and privacy) is not freely available as it is in the U.S.

The American data have led to numerous, detailed, widely-cited studies that have greatly advanced knowledge of the American electorate, beginning with Wolfinger and Rosenstone.²² It would be enormously helpful to do the same in Canada. Unfortunately, the individual LHS data are available only to researchers whose institutions pay a $5,000 annual charge.²³ Only a handful of institutions have subscribed to the data, mostly Canadian federal or provincial government agencies. As of early autumn 2018, not a single Canadian university had access, and only two American universities did (including my own). It is an odd feature of current Canadian data dissemination rules that any researcher from anywhere in the world who has an affiliation with Princeton University can access the LHS data, while not a single Canadian academic researcher can do so via his or her home institution. In my view, the data should be made available to Canadian academic researchers at nominal cost.

Last, in both the Canadian and American cases, it would be enormously helpful if the turnout data from their respective labour force surveys (the LHS and the CPS) were validated. At present, they are not.

Overview

The availability of data for turnout studies in Canada and the U.S. is summarized in Table 3. What the table makes clear are two findings. First, only rarely do American researchers have what is needed – a full complement of demographics, attitudinal variables, and validated turnout. Canadians never have them. Second, Canadian governmental turnout data are much more restricted than in the U.S., due to privacy laws and electoral administrative procedures.

Summary and Conclusion

Voter turnout merits scholarly attention. Unlike much of what political scientists do, this kind of research interests ordinary engaged citizens. They are right to worry about non-voting. Low turnout reduces government legitimacy. It may also bias government policy by underrepresenting particular groups of citizens. Last but not least, large pools of inexperienced and disengaged citizens are available for recruitment by charismatic politicians, some of whom are naïve, some a little weird, and some downright dangerous. However, understanding why people do not vote and learning what can be done about it has largely stalled due to data limitations.

Where is the scientific bottleneck in Canada? What do we really need to make serious progress on understanding voter turnout? One short-run answer seems relatively simple to implement. Elections Canada’s internal surveys using their own validated turnout records could be suitably redacted and anonymized to protect privacy and then released to researchers. Though the explanatory variables included are few, the surveys are large, and they would be helpful to scholars, particularly in explorations of how age, gender, and provincial residence effect turnout across Canada.

Second, the reported votes in the big national LHS unemployment survey could be validated. If validating the full sample is too costly, then even a five per cent or 10 per cent validation would be quite valuable. In addition, if possible, just a few attitudinal questions should be added. Key variables would be duty, interest in the outcome, partisanship strength (not direction, for privacy reasons), and perhaps media usage to measure political engagement.²⁴ A few questions like these were successfully asked in the Irish Quarterly National Household Survey in 2002, and a shorter battery again in 2011 after the Ireland national election of that year. Thus, such questions have already been combined successfully with a national economic survey. The result for Ireland is a fully comprehensive battery of variables in a large national sample, released publicly to researchers, with full protection for anonymity and privacy. Nothing of the kind is available in either Canada or the U.S. at the moment.

Third, in the longer run the federal government could match Québec’s provincial recordkeeping on voter turnout. A longitudinal record of turnout could be maintained for each voter. Doing so might require legal changes and special protections for the data file but maintaining records of this kind is standard across most modern democracies. Such a file would allow researchers to validate their turnout reports or to have Elections Canada do so.

As elections succeeded each other, the national file would come to include turnout reports for the same citizens in multiple elections. In turn, that time series, especially if combined with academic panel surveys, would allow scholars to factor out the individual idiosyncrasies that damage inferences from one-time, purely cross-sectional studies. Because academic surveys include a far more comprehensive battery of question items than government turnout records, the addition to them of validated turnout information over time would supply researchers with powerful tools for the study of why people vote. Indeed, data sets of that kind would put Canada at the leading edge of international turnout research.

Doing all this will undoubtedly require some additional funding, as well as new inter-agency coordination and cooperation. For example, to validate votes in the LHS, Elections Canada would need identifying information from Statistics Canada so that they can match individuals in the LHS to their voter files. But the larger concern is likely to be legal. Restrictive Canadian privacy laws and rules have hobbled researchers. Privacy laws always deserve respect, but it seems clear that under their current interpretation, they hinder the understanding of key Canadian policy issues such as low youth turnout.

Are present Canadian rules and laws about managing voter rolls and federal survey results really necessary to protect citizen privacy? Statistics Canada already employs sophisticated procedures to protect citizens who are surveyed by the LHS.²⁵ Under their Data Liberation Initiative, many of their anonymized surveys have been released for scholarly use. Might not similar methods allow redacted and anonymized voter files and Elections Canada in-house studies to be released to researchers?

The U.S. Census Bureau recently empaneled a group of external scholars to suggest modifications in how the CPS voter supplement is carried out. Similarly, Elections Canada may wish to convene a small group of experienced turnout researchers, perhaps combined with administrative law experts, to advise how the Data Liberation Initiative might be extended to help researchers study turnout, including youth turnout. Initiatives of that kind may well be needed to unshackle scholars so that they can use their expertise to improve the health and strength of Canadian democracy.

Notes

1. I thank Amanda Bittner, my discussant, and several members of the audience at the 2018 annual meeting of the Atlantic Provinces Political Science Association, St. John’s, Newfoundland, for helpful comments and suggestions. I am also grateful to André Blais and to Aram Hur for many helpful discussions. Officials at the U.S. Census Bureau and at Elections Canada gave generously of their time to help me understand their data sets and sampling procedures. Miriam Lapp and Angelo Elias at Elections Canada, along with Nishanthy Chitravelu at Statistics Canada, responded promptly and professionally to email questions. A preliminary version of this paper was presented at a conference in honor of André Blais’s 65^th birthday, Montreal, Canada, January 20-21, 2012. Remaining errors of fact and interpretation are solely my responsibility.
2. Pammett, Jon H., and Lawrence LeDuc, Explaining the Turnout Decline in Canadian Federal Elections. Ottawa: Elections Canada, 2003; Blais, André, and Peter Loewen. Youth Electoral Engagement in Canada. Ottawa: Elections Canada, 2011. URL: http://inspirerlademocratie-inspiredemocracy.ca/rsch/yth/yeec/index-eng.asp. Accessed October 5, 2018.
3. Blais and Loewen 2011; Blais, André, and Daniel Rubenson. “The Source of Turnout Decline: New Values or New Contexts?” Comparative Political Studies 46, 1 (2013): 95-117.
4. Riker, William H., and Peter C. Ordeshook. “A Theory of the Calculus of Voting.” American Political Science Review 62, 1 (March 1968): 25-42; a recent discussion is Blais, André, and Christopher H. Achen. “Civic Duty and Voter Turnout,” Political Behavior, Forthcoming (2018).
5. Elections Canada. Voter Turnout at Federal Elections and Referendums, 2018. http://elections.ca/content.aspx?section=ele&dir=turn&document=index&lang=e . Accessed October 5, 2018.
6. Another source of election data is exit polls, based on interviewing voters as they leave the polling place. Combined with Census data, they can be used to estimate which categories of citizens voted. So far as I am aware, however, a national Canadian exit poll has never been conducted (see Brown, Stephen D., David Docherty, Ailsa Henderson, Barry Kay, and Kimberly Ellis-Hale. 2006. “Exit Polling in Canada: An Experiment.” Canadian Journal of Political Science 39, 4 (Dec.): 919-933.), and as in the U.S., the growing importance of advance voting and other forms of voting away from the polling place make exit polls increasingly less representative and thus less useful.
7. Achen, Christopher H., and André Blais. “Intention to Vote, Reported Vote, and Validated Vote.” In David Farrell, ed. The Act of Voting: Identities, Institutions and Locale. Abingdon: Taylor and Francis (Routledge), 2015.
8. Karp, Jeffrey A., and David Brockington. “Social Desirability and Response Validity: A Comparative Analysis of Overreporting Voter Turnout in Five Countries.” Journal of Politics 67, 3 (August 2005): 825-840.
9. Traugott, Michael W., and John P. Katosh. “Response Validity in Surveys of Voting Behavior.” Public Opinion Quarterly 43 (1979): 359–77; Achen and Blais.

10. See Ansolabehere, Stephen, and Eitan Hersh. “Validation: What Big Data Reveal About Survey Misreporting and the Real Electorate.” Political Analysis 20 (2012): 437–59, which contains an extensive bibliography.
11. Gidengil, Elisabeth, Neil Nevitte, André Blais, and Joanna Everitt. Dominance and Decline: Making Sense of Recent Canadian Elections. Toronto: University of Toronto Press, 2012.
12. André Blais, personal communication.
13. http://www.elections.ca/content.aspx?section=res&dir=rec/eval/pes2015/surv&document=p2&lang=e#a3. Accessed March 11, 2019.
14. McDonald, Michael. Voter List Information, 2018. URL: http://voterlist.electproject.org/full-list-purchase-facts-and-info. Accessed October 5, 2018.
15. Karp and Brockington, p. 838.
16. Elections Canada. Estimation of Voter Turnout by Age Group, 2018. http://elections.ca/content.aspx?section=res&dir=rec/part/estim&document=index&lang=e . Accessed October 5, 2018.
17. Blais and Loewen, p. 17.
18. Wang, T.Y., and Christopher H. Achen. “Declining Voter Turnout in Taiwan: A Generational Effect?” Electoral Studies. Forthcoming (2019).
19. This is not to say that linkage is easy or inexpensive. The U.S. experience with the American National Election Study (ANES) in the Eighties is set out in: Traugott, Santa. Validating Self-Reported Vote: 1964-1988. Institute for Social Research, University of Michigan. Prepared for the Annual Meeting of the American Statistical Association, Washington, D.C., August 7-10, 1989. No national validations with human coders have been done since. To reduce costs, automated validation using the U.S. states’ electronic voter lists has been tried recently, including a validation of the 2016 ANES, but with mixed results. Berent, Matthew K, Jon A. Krosnick, and Arthur Lupia. “The Quality of Government Records and ‘Over-estimation’ of Registration and Turnout in Surveys: Lessons from the 2008 ANES Panel Study’s Registration and Turnout Validation Exercises.” Working Paper no. nes012554. February 15, 2011 version. Ann Arbor, MI, and Palo Alto, CA: American National Election Studies. Available at http://www.electionstudies.org/resources/papers/nes012554.pdf; Ansolabehere, Stephen, and Eitan Hersh. “Validation: What Big Data Reveal About Survey Misreporting and the Real Electorate.” Political Analysis 20 (2012): 437–59; Enamorado, Ted. Active Learning for Probabilistic Record Linkage. Princeton: Princeton University, 2018. Manuscript.; Pew Research Center. Commercial Voter Files and the Study of U.S. Politics. 2018. URL: http://www.pewresearch.org/2018/02/15/commercial-voter-files-and-the-study-of-u-s-politics/ Accessed November 15, 2018.
20. See: https://www.statcan.gc.ca/eng/rdc/index
21. The most recent report is Elections Canada. Turnout and Reasons for Not Voting During the 42nd General Election: Results from the Labour Force Survey, 2016. http://elections.ca/content.aspx?section=res&dir=rec/eval/pes2015/lfs&document=index&lang=e. Accessed October 5, 2018.
22. Wolfinger, Raymond E. and Steven J. Rosenstone. Who Votes? New Haven: Yale University Press, 1980.
23. See: https://www150.statcan.gc.ca/n1/pub/11-625-x/11-625-x2010000-eng.htm.
24. Blais and Achen (2018) briefly review the many relevant citations.
25. http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=3701