The Canadian Municipal Election Study

Department of Politics and Public Administration, Ryerson University, Toronto, ON, Canada, Department of Political Science, Western University, London, ON, Canada, Department of Political Science, McGill University, Montreal, QC, Canada, Urbanisation Culture Société Research Centre, Institut National de La Recherche Scientifique, Montreal, QC, Canada, Department of Political Science, University of Calgary, Calgary, AB, Canada, Department of Political Science, Memorial University of Newfoundland, St. John’s, NL, Canada, School of Political Studies, University of Ottawa, Ottawa, ON, Canada, Department of Political Science, University of Winnipeg, Winnipeg, MB, Canada, Department of Political Science, Dalhousie University, Halifax, NS, Canada, Department of Political Science, Carleton University, Ottawa, ON, Canada


INTRODUCTION
Not long ago, research on Canadian local elections was "virtually non-existent" (Stanwick 2000) and municipal elections were referred to as the "poor cousins" in the study of voting behaviour (Cutler and Matthews 2005, 359). This scholarly oversight is perhaps somewhat understandable, given that local elections are considered by some to be of "lower rank" than national or provincial elections. As compared to elections at the federal and provincial levels, we know that Canadians are less interested in local elections, and they think that local government affects them less (Lucas and McGregor 2021). At the same time, there is so much variation between cities that the findings from studies of city elections are difficult to generalize. Local elections are of a "different kind" from federal and provincial election, but they also differ greatly from one another (see Oliver et al., 2012 on the United States, but the claim is as true, if not truer, in Canada). 1 In the past decade, however, the field of local political behaviour has experienced an unprecedented surge in attention in Canada. Two main streams in this growing literature are candidate studies (Tolley, 2011;Breux et al., 2019;Scott and Medeiros, 2021) and elector studies (Couture et al., 2014;Goodman 2014;Breux et. at. 2017;Kiss et al., 2020;Lucas and McGregor, 2021;McGregor et al., 2021). The latter is the largest and fastest growing segment of the field, to the effect that Eidelman and Taylor's (2010), 305) description of the field of Canadian urban politics as a "black hole" no longer rings true, at least as far as studies of municipal elections and political behaviour are concerned.
We have witnessed parallel trends elsewhere, as well. Writing about the United States just a decade ago, Marshall noted that "to say that a field of study on local elections exists would be a bit of an overstatement" (Marschall, 2011, pg. 97). Since then, however, there has been an immense growth in research on American local elections (Oliver et al., 2012;Hajnal and Trounstine 2014;Sances 2018;Warshaw 2019;Holman and Lay 2021). Growth during this period has also occurred in many other contexts, as scholars around the world have increasingly turned to the local level to understand the behaviour of their electorates (see Kang et al., 2018;Marien et al., 2015;Šaradín et al., 2021, as well as a recent special issue on comparing local elections and voting in Europe,; Gendźwiłł and Steyvers, 2021). Though the volume of scholarship on municipal elections remains much smaller than that focused on elections at upper levels of government, the field of local political behaviour is firmly established and growing. This growth has been motivated by a number of factors. First, the vast majority of elections and politicians are at the local level-for example, according to Taylor and Bradford, 2020, Canada has more than 3,700 municipalities. Municipalities thus provide a rich landscape for those interested in studying electors, candidates, and elections. Second, local elections have many features that do not exist at the federal or provincial/state levels, including non-partisan races, concurrent elections, and widely varying electoral institutions (such as district types, electoral systems, and voting procedures). Many important questions about how institutions affect electors, in Canada and elsewhere, can therefore only be answered by looking locally. Similarly, comparisons across municipalities enable researchers to exploit still more interesting variation. Cities vary from one another in numerous ways, including turnout rates, governance structures, socio-demographic diversity, economic conditions, and the presence or absence of local political parties. Municipalities therefore serve as important laboratories of political behaviour, providing an opportunity to consider questions that could not be answered using federal or provincial data. Enabling this growth is the fact that technological improvements mean that data collection and analysis are easier and more cost effective than ever before. We therefore concur with Marschall's description of the possibilities for research on local elections as "practically limitless" (2011:97).
It is in this context of rapid growth of the field of local political behaviour that we introduce the Canadian Municipal Election Study (CMES) dataset. With a total sample size of 14,458, the dataset includes survey responses from voters and non-voters in eight of Canada's largest cities: Vancouver, Calgary, Winnipeg, London, Mississauga, Toronto, Montreal, and Quebec City. Cities were chosen on the basis of size, contextual variation, and institutional characteristics. The dataset, which consists of panel survey data collected before and after each election, provides an unrivaled opportunity for researchers to study the motivations, attitudes, and behaviour of Canadians in local elections.
There are a number of reasons why we expect that the CMES will be of interest to scholars both within and outside of Canada, as well as to those who study elections other than at the local level. First, as noted above, the study of municipal elections is in a significant growth phase, both in and outside of Canada. While large comparative election studies have long been the norm for national elections, however, the CMES is the first such study focused exclusively upon the municipal level. The fact that the dataset includes very different cities (as we detail below) and an extremely large sample size, means that a great many important and novel research questions can now be explored, taking advantage of these and other features of the dataset. Researchers in contexts where local elections tend to be quite similar to those in Canada (including in the United States) should also find the dataset useful. The CMES should also be of interest to those who study elections other than at the local level, as the cases under study here have important features that may allow for generalization to other contexts. The dataset should therefore be of interest to scholars who conduct research into non-partisan elections, decision making in a low-information context, simultaneous elections, incumbency (rates of re-election in local elections are astonishingly high), and elections with low voter turnout.
After describing the CMES dataset, we discuss below some of the possibilities for research using this important research tool. For illustrative purposes, we provide an example of such research, and conduct a novel analysis that shows the importance of a sociodemographic indicator rarely employed in studies of national elections (home ownership) at the local level.

STUDY DESIGN
CMES data are drawn from online surveys of eligible voters in eight Canadian cities that held elections in 2017 and 2018. A total of 2.4 million votes were cast in these contests, representing 24.6% of the Canadian population. 2 Data were collected in two waves, with one questionnaire administered before election day (over the course of several weeks, with an eye towards identifying possible campaign dynamics), and one fielded shortly after the election. 3 The total sample size for the pre-election survey was 14,458; 9,409 respondents also completed the post-election questionnaire (return to sample rate 65.1%). 4 The pre-election survey was roughly 20 min in length and the post-election questionnaire was about 15 min. The average number of respondents per city is just over 1,800, though there is some variation (see Table 1).
With comparability with previous election studies in mind, many of the questions in the CMES were modelled on the Canadian Election Study (which itself shares many questions in common with other studies, such as the Making Electoral Democracy Project and the Comparative Study of Electoral Systems). This includes questions on interest, attentiveness, issue opinions, ideology, candidate ratings, partisanship, economic and performance evaluations, and political knowledge, among other common election study questions.
A variety of new questions were included to take into consideration the distinctive characteristics of municipal government. Thus while standard questions were asked about attitudes towards candidates, political orientations, and, sociodemographic characteristics and media consumption, so too were questions about council, borough and school board 2 As per the 2016 Census 3 A number of questions were asked in both waves, meaning that change over time can be considered 4 Surveys were fielded on the following dates: Calgary: Pre-election: Sep. 29 to Oct. 15, 2017, post-election: Oct. 17  Forum Research Incorporated recruited respondents and coded and administered the surveys. Participants were recruited in two ways. Most (roughly 74%) were recruited via random digit dialing (RDD) by telephone, in an attempt to collect a probability sample. After confirming survey eligibility, respondents were connected with a live operator who collected an email address to which a survey link was sent. Though it is increasingly common in the discipline to draw sample from existing panels, it is difficult to achieve a substantial sample size from relatively small populations, such as cities. As such, we chose to recruit new respondents via RDD. Samples were supplemented with respondents from an existing online panel when possible. The cooperation rate was 5.87%. 5 Quotas for age and gender, taken from 2016 census data, taken from 2016 census data, were used for panel respondents,. 6

POSSIBILITIES FOR RESEARCH
We see at least five ways in which the CMES may be of use to researchers. First, the dataset allows for comparison across very different municipalities. The cities in the sample differ from one another in several respects, and provide variation on dimensions that do not apply at other levels of government. For instance, it is only locally that we see variation in electoral systems (London used ranked ballots and Vancouver uses an at-large system for council elections) and the presence and absence of political parties (most elections in the study are non-partisan, but not Vancouver, Montreal and Quebec). 7 Table 1 shows just some of the many characteristics of the CMES cities that might interest researchers conducting comparative work.
Second, the CMES has a large enough sample size to conduct analyses of each election separately. Each of the cities in the study has unique characteristics (some of which are listed in the final column in Table 1), and all have a substantial number of respondents. Researchers who have interests in any of these cases therefore are able to conduct detailed analyses using data from residents of any one city. An edited volume by Lucas and McGregor (2021) provides examples of the single case studies that can be conducted using the data.
Third, the size of the dataset makes it possible to conduct subgroup analyses. Many groups that traditionally make up small segments of survey samples have sizable numbers of respondents in the dataset. For instance, the CMES dataset contains 269 respondents who identify as Indigenous. Though small as a share of the overall respondent pool, this figure is large enough to allow for meaningful analysis. There also similarly sizable subgroups of religious minorities (331 Muslims, 381 Jews, and 198 Buddhists) and sexual minorities (626 homosexuals, 393 bisexuals and 175 who identify as other). Though non-voters tend to be vastly underrepresented in election studies, the size of the CMES means that the dataset contains 1,496 respondents who did not vote. These and other relatively small groups often cannot be As is the case with existing election studies, the sample is unrepresentative of the population on some dimensions (for example, as is always the case in datasets of this nature, voters are overrepresented). Accordingly, the dataset includes weights to match to 2016 census data on the basis of age and gender. Separate weights were created for the RDD respondents, and for the pooled, RDD and panelist, data 7 Though NWT and Nunavut have non-partisan territorial elections, there are no election study data available from their elections 8 Laakso and Taagepera (1979) 9 This is the margin in the final round of counting. The gap in the first round was 12.0 points Frontiers in Political Science | www.frontiersin.org September 2021 | Volume 3 | Article 745331 studied in any meaningful way with datasets of only one or two thousand respondents. 10 The large overall size of the CMES dataset means it is possible to study some subgroups that make up a relatively small proportion of the electorate. 11 Fourth, the CMES allows researchers to ask a wide variety of questions about Canadian electors that simply cannot be answered without data from the local level. As already noted, municipalities have many characteristics that are absent in federal or provincial elections, such as concurrent elections (mayor and council), lower-interest "down-ballot" races (school and park board elections) and frequent concurrent plebiscites. Local governments also have different powers and profiles than their federal and provincial counterparts, and turnout tends to be comparatively low. Studying local elections provides the variation that makes identifying important insights about how these understudied circumstances affect electors possible.
Finally, we see potential for CMES data to be combined with local survey data from other settings. Recent local election studies in the Czech Republic (Šaradín et al., 2021), the Netherlands (Jansen and Denters 2019) and other Western-European countries (NCCR Democracy 2016) ask similar questions on turnout, vote choice, and attitudes towards politics that might be combined with CMES data in a comparative study. Researchers from around the world may be interested in using data from all CMES cities, or just a small number of cities. As noted above, there are many great differences between the cities included in the dataset (indeed, one of the exceptional features of Canadian municipalities that should be of interest to other researchers is the institutional variation that cities across the country provide). Some cities included in the study may be more appropriate to include in comparative studies than others, and the high number of survey respondents available from each city means that select cities can be used, if researchers wish.

USING CANADIAN MUNICIPAL ELECTION STUDY DATA-A LOCAL-FEDERAL COMPARISON
Another type of analysis that can be conducted with CMES data, and one we provide an example of below, is comparison across levels of government (the CMES includes a number of questions about attitudinal orientations towards multiple levels of government). Such comparisons provide insight into how the same electors might reason and act towards elections at different levels, and also to see what types of factors might be associated with observed differences. We know, for example, that some elector characteristics have a different importance locally as compared to federally or provincially. One such characteristic is homeownership, with previous research showing that the turnout gap between owners and renters is higher municipally than at other levels of government (Fischel 2001;McGregor and Spicer 2016;Jiang 2018). 12 An illustrative example of potential between-level research is shown in Figure 1, which includes a series of coefficient plots. These plots demonstrate the relationship between homeownership and several important attitudes and behaviours at the municipal and federal levels (all coded to range from 0 to 1). The results in the first column show the results from a series of OLS regressions, where five outcomes are regressed on a binary measure of home ownership (where the baseline is those who do not own their homes), as well as a series of sociodemographic controls (results not shown). 13 We ran models separately for each factor, for the federal (green) and municipal (orange) levels. Each dot therefore represents a separate regression model, and the labels on the left of the figure represent the outcome variable in each model. Results to the right of the zero line indicate that home ownership has a positive relationship with the outcome variables. Whiskers show 95% confidence intervals. We use multi-level models with varying intercepts for city. 14 Though the results in the first column allow us to eyeball differences between levels, it is difficult to infer whether these differences are statistically significant. As such, we show the second column of results, where the outcome variables are based upon a combination of the federal and municipal outcome variables. More specifically, the value for the federal variables are subtracted from the municipal (values therefore range from −1 to 1). In this setup, coefficients to the right of zero indicate that the effect of homeownership is greater, in the positive direction, at the municipal level than federally. If the whiskers do not cross zero, we can conclude that homeownership has different effects at the two levels of government (because homeownership leads to significantly different effectsat the municipal level compared to the federal level).
CMES data provide strong evidence that homeownership is distinctively important in municipal elections. As expected, the turnout gap between owners and renters is larger at the municipal scale. There are also gaps in a variety of attitudes, including interest in politics and the belief that voting is a duty (as opposed to a choice). Ownership-based differences in government impact and external efficacy are also greater locally. Home ownership, a variable that receives scant attention in studies of federal voting behaviour, is clearly important in Canadian municipal elections. 10 The dataset also includes information on forward sortation area, so respondents can be categorized on the basis or small geographic areas 11 Despite the limited number of cities in the dataset, multi-level analyses are also possible with the data, where individuals level-variables might be interacted with city-level characteristics 12 The argument tends to be that owners care more than renters about local politics because of the potential impact that local government can have upon property values 13 Though not all of the outcome variables considered here are continuous, we use OLS for the sake of consistency. The substantive conclusions of the analysis of the turnout and duty variables are unchanged if we use logistic regression models instead 14

CONCLUSION
The CMES dataset provides researchers with the opportunity to begin to map the many ways electors reason and behave in an understudied but important type of election. Coupled with possibilities for comparative research, single city studies, and subgroup analyses, the CMES allows researchers to consider a variety of questions about how attitudes are formed and decisions are made in municipal elections (and how this differs from other levels of government). We expect that the dataset will prove to be of great use to scholars of local political behaviour, both within and outside of Canada. 15 Many of the features of the elections present in the CMES, including non-partisanship, low turnout and low information races, simultaneous elections and very high rates of incumbency, should also attract scholars who might not ordinarily be interested in the study of local electoral contests. It is our hope that both established and emerging researchers will use the CMES extensively to uncover important insights about elections and electorates.

DATA AVAILABILITY STATEMENT
The dataset presented in this study can be found in an online repository at the following link: https://doi.org/10.7910/DVN/ HK9GJA.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the Ryerson University Research Ethics Board. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

AUTHOR CONTRIBUTIONS
RM is principal investigator on the CMES and primary author of the data note. All other authors contributed to the creation of the dataset and provided feedback on data note.

SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpos.2021.745331/ full#supplementary-material 15 Any user of the data should be cautioned against overgeneralizing from the dataset. The eight cities chosen here were included due to the important variation they provide. One dimension on which they do not provide variation, however, is size-all are among the most populated in the country. We encourage the creation of future datasets that focus on electors from municipalities of all sizes, as we suspect important variation on many fronts may exist in this dimension