One city, uneven participation

Neighbourhood-level data cannot tell us why particular people chose to vote or not to vote. But we know neighbourhood characteristics also do not vary independently: education overlaps with occupation, housing with income, immigration with age and language, and so on – and many of these characteristics are correlated to voting patterns.

To examine these relationships together, we used Partial Least Squares (PLS) regression with 70 measures of neighbourhood population, housing, employment, transportation, and accessibility. In simple terms, PLS finds a small number of components that capture characteristics that tend to occur together and are also related to voter turnout. Each neighbourhood census tract receives a score on each component, showing where it lies relative to other tracts.

The first component is much more closely associated with turnout than the others. Its higher-scoring side includes neighbourhoods with more university education, professional employment, working from home, higher employment rates, and tends to be more centrally located. Its lower-scoring side combines more trades, service and manufacturing work, long commutes, unemployment, larger and multigenerational households, lone-parent families, and larger immigrant and first-generation populations.

The important finding is the combination, not any one characteristic. The data and the model cannot tell us, for example, that immigrants living in a lower-turnout tract are the residents who did not vote. Nor can immigrant share, occupation, household structure or any other individual measure be isolated from the component and treated as a cause of lower turnout.

That distinction matters here. Our turnout denominator includes only Canadian citizens aged 18 and older, so neighbourhoods are not assigned lower turnout simply because they contain more residents who are ineligible to vote. Immigrant and first-generation shares instead describe the wider neighbourhoods in which eligible voters live, alongside differences in employment, education, household structure, commuting and geography.

The broader turnout literature also shows why these characteristics should be interpreted together. A major meta-analysis by Benny Geys reviewed 83 aggregate-level studies of turnout, while a later update by João Cancela and Geys expanded that evidence base to 185 studies and broadly confirmed the earlier findings. Both emphasize that turnout reflects a mix of socioeconomic, political and institutional conditions rather than a single demographic explanation.

Existing Canadian research gives us additional reason to resist treating immigration or racialization as simple explanations for voter turnout. A 2023 study by Stephen White, using Canadian election studies spanning 45 years, found that an apparent relationship between immigrants' length of residence and turnout largely disappeared once the historical period in which different cohorts arrived was taken into account. Even seemingly straightforward demographic relationships can therefore reflect political socialization and historical context rather than fixed characteristics of a group.

Likewise, recent Canadian research on racialized voters finds differences in turnout between several racialized groups and white Canadians, but also finds that socioeconomic circumstances and political engagement explain part of those gaps, while a substantial portion remains unexplained. Other work by Statistics Canada cautions us to note that many racialized Canadians still participate in politics beyond the ballot box: by engaging with news, going to meetings or demonstrations, signing a petition, or wearing signs in support of a cause.

The component is therefore better understood as a wider social geography involving class, employment, settlement, household responsibilities, commuting, and location. Across the 582 tracts with reliable data, its score correlates 0.72 with mean turnout, substantially more strongly than the other components identified by the model.

This component is not peculiar to one election. When we repeat the analysis separately for municipal, provincial and federal turnout, essentially the same underlying components reappear. Neighbourhood characteristics predict municipal turnout more strongly, but the broad pattern remains very similar across all three levels.

The model also identifies other recognizable components of Toronto's geography: relatively settled family neighbourhoods versus younger and more mobile renter areas; housing need and apartment living versus higher-income family ownership; and a weaker housing and employment contrast. These components are all much less closely related to turnout than the first.

None of this tells us that a particular policy would increase turnout. It does suggest a different way of framing the problem: lower participation may reflect barriers that institutions can reduce, rather than a lack of interest within particular neighbourhoods.

Time and convenience are part of that picture. In an Elections Canada survey following the 2025 federal election, 34% of voters said they chose their voting method because the date and time worked with their schedule, while 28% cited having a nearby voting location. Among non-voters, 24% cited reasons related to the electoral process, and 23% cited everyday-life or health reasons.

Access matters too. Observational research using Ontario polling locations found that greater distance or travel time to the polls was generally associated with lower turnout. And in the 2025 federal election, 44.2% of ballots were cast through advance or other alternative voting methods, showing how widely voters already use more flexible options.

Taken together, polling-place location, advance voting, voting hours, transportation, work schedules, and caregiving are useful areas for further study. Our model does not directly measure most of these mechanisms, so they are better treated as plausible ways to make participation easier, not as effects demonstrated by this analysis.

The component analysis summarizes how variables work together, but it is also useful to see how each relates to turnout on its own. The explorer below lets you select measures included in the model, including education, employment, age, housing, transportation, and access to services, and compare each with turnout.

Remember, however, that these are relationships between characteristics of neighbourhoods, not individual people. A census tract is not a voter, a correlation does not establish behaviour or causation, and turnout is not the sum of political participation.