Interprovincial Differences in Political Culture

There are two approaches to studying Canadian political culture. The longest standing perspective is qualitative in nature and focuses on the deep-rooted and progressive impact of historical phenomena. Hartz's (Reference Hartz1955, Reference Hartz1964) fragment theory, Innis’ (Reference Innis1930) staples theory and Lipset's (Reference Lipset1968) work on formative events are pioneering work in this line of research. A second perspective relies on survey data to assess political culture, building upon the work of Almond and Verba (Reference Almond and Verba1963). Rather than probing history, their work aggregates individual political attitudes and treats them as indicators of political culture.

The two perspectives both seek to identify, describe and account for patterns of political culture. Yet they adopt distinct epistemological perspectives that warrant the different theoretical and methodological approaches they use. Simeon and Elkins’ seminal pieces (Simeon and Elkins, Reference Simeon and Elkins1974; Elkins and Simeon, Reference Elkins and Simeon1980) relied on the civic culture approach pioneered by Almond and Verba, importing the framework in Canadian political science. This approach defines political culture as a macro-level phenomenon that appears through the aggregation of a group of individuals’ basic orientations, norms, values and assumptions about politics (Elkins and Simeon, Reference Elkins and Simeon1979). It embodies a set of orientations toward the political world widely shared among its group members (Almond and Verba, Reference Almond and Verba1963).Footnote ¹ Those who replicated Simeon and Elkins’ work also relied on the civic culture (quantitative) approach to studying political culture, and this study takes the same path.

Political culture is conceptualized as an enduring aspect of societies, which exerts a lifetime impact on individuals. Political culture certainly evolves over time (Almond and Verba, Reference Almond and Verba1980) but typically does so very slowly, constantly reproducing itself through each new generation of citizens (Wiseman, Reference Wiseman2007: 13). This sets it apart from public opinion: while the latter fluctuates rapidly in close connection to short-term political developments, political culture evolves only very slowly over time. Accordingly, although political attitudes can vary in reaction to salient events, a clear pattern should be discernable through the noise of short-term fluctuations. By looking for such long-term patterns, which wash away short-term fluctuations, we should be able to identify the deep-rooted differences in political culture across groups.

A variety of groups can be said to have a political culture of their own. In Canadian scholarship, the locus of attention has typically been placed on provincial and regional political cultures (Clarke et al., Reference Clarke, Pammett, Stewart, Archer and Young2002; Cooper, Reference Cooper, Archer and Young2002; Elkins and Simeon, Reference Elkins and Simeon1980; Fafard et al., Reference Fafard, Rocher and Côté2010; Gibbins, Reference Gibbins1980; Kornberg and Clarke, Reference Kornberg and Clarke1994; Leuprecht, Reference Leuprecht2003; McGrane and Berdahl, Reference McGrane and Berdahl2013; O'Neill and Erickson, Reference O'Neill and Erickson2003; Schwartz, Reference Schwartz1974; Simeon and Elkins, Reference Simeon and Elkins1974; Wesley, Reference Wesley2011; Wiseman, Reference Wiseman2007), although studies have also assessed political culture from a subcontinental (Baer et al., Reference Baer, Grabb and Johnston1993; Grabb et al., Reference Grabb, Anderson, Hwang and Milligan2009), national (Horowitz, Reference Horowitz1966; Lipset, Reference Lipset1990; Nevitte, Reference Nevitte1996) and gender (O'Neill, Reference O'Neill, Everitt and O'Neill2002) perspective. Some have even focused on geographically noncontiguous clusters of culture (Gidengil, Reference Gidengil1990; Henderson, Reference Henderson2004). This study builds on these important contributions but intends to provide empirical evidence to debates over provincial and regional differences in political culture. To even the most casual observer, such differences abound. Richard Simeon and David Elkins were among the first to try to precisely estimate the extent of those differences, which they grouped under the concept of “regional political cultures,” by using survey data to investigate the extent to which “the population of the Canadian provinces differ in some basic orientations to politics” (Simeon and Elkins, Reference Simeon and Elkins1974: 397; see also Schwartz, Reference Schwartz1974). Analyzing the 1965, 1968 and 1974 Canadian Election Studies (CES), the authors found wide differences across provinces with regard to their citizens’ sense of political efficacy and political trust. Both trust and efficacy were found to be particularly high in Ontario and British Columbia, whereas citizens of the Atlantic provinces exhibited much lower levels of trust and efficacy (Elkins and Simeon, Reference Elkins and Simeon1980; Simeon and Elkins, Reference Simeon and Elkins1974).

Simeon and Elkins’ work created a very sticky image of provincial political cultures. Decades later, research on the topic still tends to treat the authors’ conclusions as the baseline upon which to build empirical expectations or as the appropriate point of comparison for results (see Anderson, Reference Anderson2010; Henderson, Reference Henderson2004; Reference Henderson2010; Héroux-Legault, Reference Héroux-Legault2016; McGrane and Berdahl, Reference McGrane and Berdahl2013; O'Neill and Erickson, Reference O'Neill and Erickson2003; Wesley, Reference Wesley2015; Wiseman, Reference Wiseman2007). Yet the authors’ conclusions have been much debated, as many who addressed similar research questions using survey data found discrepant, sometimes even completely opposite, regional patterns to those identified by Simeon and Elkins (Clarke et al., Reference Clarke, Pammett, Stewart, Archer and Young2002; Henderson, Reference Henderson2004; McGrane and Berdahl, Reference McGrane and Berdahl2013; Ornstein et al., Reference Ornstein Michael, Michael Stevenson and Paul Williams1980; Wesley, Reference Wesley2015). Some therefore claim that the patterns identified by the authors are now outdated (Stewart, Reference Stewart1994). Others have discussed the possibility that provincial political cultures may have changed over time, in order to justify the discrepancy between their findings and those of Simeon and Elkins (Henderson, Reference Henderson2004; McGrane and Berdahl, Reference McGrane and Berdahl2013).

These inconsistencies raise questions about the ability of previous studies on the topic to identify long-term patterns of political culture. Importantly, these discrepancies could potentially be the result of extraneous factors confounding the results of those studies. Indeed, political attitudes can vary swiftly over time. As most studies rely on a short time period to conduct their analyses, they are vulnerable to mistaking contextual idiosyncrasies with enduring patterns of political culture. The first potential confounding factor that comes to mind is the economic condition of provinces, as many have identified this issue as a significant determinant of political attitudes (Cochrane and Perrella, Reference Cochrane and Perrella2012; Gidengil, Reference Gidengil1989, Reference Gidengil1990). Importantly, Gidengil's work claims that this is not merely a composition effect, as the location of one's province in the centre-periphery system is claimed to impact political attitudes independently of one's personal socio-economic status. Accordingly, while individual material status does matter in influencing political orientations, the provincial economic context also shapes citizens’ attitudes, thus fostering provincial patterns of political culture.

Partisanship constitutes another factor that could blur results across studies. Indeed, after receiving some initial pushback (see Clarke et al., Reference Clarke, Jane Jenson and Pammett1984; Meisel, Reference Meisel1975), the idea that partisanship exerts the determinant influence depicted by the authors of the Michigan model (Campbell et al., Reference Campbell, Converse, Miller and Stokes1960) has now made its way into Canadian politics (Gidengil et al., Reference Gidengil, Neil Nevitte, Everitt and Fournier2012). Accordingly, it appears likely that varying levels of support for federal and provincial governments could contribute to profound shifts in provinces’ levels of political trust and efficacy.

This study thus accounts for these two important contextual factors (economic conditions and partisanship) to better identify the underlying patterns of political culture prevailing within each province.

Data and Methods

Two issues with earlier studies of provincial political cultures potentially account for the discrepant findings in the literature: small sample sizes and short time frames. The first issue is the most straightforward. Focusing on provincial political cultures necessarily implies stratifying one's dataset in 10 smaller provincial groups, thus substantially reducing the sample size for each province, especially the smaller ones. For example, Simeon and Elkins’ (Reference Simeon and Elkins1974) study relied on only 48 respondents from Newfoundland. Studies thus become plagued by a substantial amount of sampling variance, which has often not been considered in the literature given its focus on descriptive comparisons of provincial averages.

Some recent investigations address the issue by relying on larger sample sizes allowing researchers to use statistical tools (regression or ANOVA) to infer about provincial differences (Cochrane and Perrella, Reference Cochrane and Perrella2012; Henderson, Reference Henderson2004; McGrane and Berdahl, Reference McGrane and Berdahl2013). Yet these studies fall prey to the second limitation: their datasets are collected over short time frames. This is especially worrisome given that political culture cannot be measured directly and can only be inferred through attitudes that, aggregated together, form political culture. Unfortunately, these individual attitudes tend to exhibit high volatility, whereas the underlying political culture is, in comparison, extremely stable. Political attitudes can thus vary differently from one province to the other, in reaction to political events that are salient in each province. Accordingly, by relying on short time frames, studies cannot distinguish between idiosyncratic fluctuations in political attitudes and underlying patterns of political culture. This study intends to overcome this challenge by relying on surveys collected over a lengthy time frame, which allows idiosyncratic fluctuations in indicators of provincial political cultures to average out. In doing so, our analysis provides a quantitative estimation of political culture prevailing within each province over the whole period. Yet it also needs to be underscored that such an approach is inherently unable to identify temporal changes, as it averages out those potential changes. Accordingly, results need to be interpreted as providing a snapshot of political culture within each province for the whole period, but these results cannot necessarily be extrapolated to any individual year included in the sample. Although we expect changes in political culture to be slow and moderate, their potential occurrence cannot simply be dismissed. They constitute an extension of this analysis that I encourage others to study in order to deepen our understanding of provincial political cultures.

The empirical analysis used here relies on three datasets: the Canadian Election Study (CES),Footnote ² the Comparative Provincial Election Project Survey (CPEPS)Footnote ³ and the Provincial Diversity Project (PDP).Footnote ⁴ Data for the CES cover the period 1974–2019, with a cross-sectional survey conducted for every federal election during the period, although inconsistencies in the questions included in each survey mean that some surveys are left aside for each model that is estimated.Footnote ⁵ The CPEPS, for its part, is a repeated cross-sectional survey covering a period from 2011 to 2015, surveying respondents of each province during one provincial election—or occasionally two provincial elections. Finally, the PDP is a cross-sectional survey that was fielded between December 2013 and February 2014 and was not tied to any election campaign. All three surveys are probabilistic and designed to be representative of their underlying population. Appendix A in the online supplementary information provides information on the sample size of each survey.

The outcome variables that are investigated were selected to reflect as closely as possible the original analysis of Simeon and Elkins (Reference Simeon and Elkins1974), in order to maximize the comparability of their results and those obtained in this re-examination of their work. I thus listed the nine survey items they use in their analysis and investigated how consistently they were asked in the datasets used in the analysis. Four items were selected based on the consistency of their presence in the surveys used. These were then grouped into three concepts: political cynicism, internal political efficacy and external political efficacy. Political cynicism is measured by combining answers to two questions, one asking respondents whether they believe that the government cares about what people like them think, and a second measuring their perception of members of Parliament losing touch with them after their election.Footnote ⁶ Internal political efficacy is measured by using a question asking respondents whether they think government and politics are too complicated for them to understand. Finally, external political efficacy is measured through a question asking respondents their agreement with the statement that people like them have no say about what the government does.

I slightly depart from Simeon and Elkins with regard to the conceptualization of the outcome variables. Whereas they combine their items to form two indicators of political culture (political trust and efficacy), I opted for a three-pronged approach (political cynicism, external efficacy and internal efficacy). The current analysis focuses on political cynicism rather than political trust, since the latter is conceptualized as a broad orientation toward government actions meeting (or not) citizens’ expectations (Hetherington, Reference Hetherington2005). Political cynicism, in contrast, is merely one ramification of political trust, capturing citizens’ evaluations of political officials’ integrity and individual actions (see Dancey, Reference Dancey2012: 412–13). The two indicators of political trust that were consistently asked in surveys used in the present analysis relate to politicians’ actions and integrity; thus I find it more appropriate to conceptualize the outcome variable as measuring the specific concept of political cynicism rather than the broader notion of political trust.

Finally, in contrast to Simeon and Elkins, I separate internal and external political efficacy, given that one indicator directly measured respondents’ perception of their ability to understand politics, whereas another indicator relates to their ability to have their grievances taken into consideration by elected officials. Considering the different target of the two survey items—respondents’ understanding of politics (internal efficacy) and influence on politics (external efficacy)—I find it more appropriate to keep the two items separate in order to clarify the interpretation of the results.

The main predictors of interest are the province dummy variables, which capture provincial differences in our three outcome variables of interest. I am also interested in the impact of economic conditions, as the literature suggests that poor economic conditions make citizens become more disaffected, and vice versa when conditions are good. Accordingly, measures of provinces’ unemployment and growth rate when each survey was fielded are included in the models (both measures taken from Statistics Canada's database). Socio-demographic indicators commonly used in Canadian political behaviour research are also included in the models as control variables (see Gidengil et al., Reference Gidengil, Neil Nevitte, Everitt and Fournier2012)—age, education, gender, native language and religion. Respondents’ party identification—both at the federal and provincial levels—is also accounted for, as it seems possible that those identifying with more successful parties could be less cynical toward politics and have a greater sense of efficacy. Given the importance attributed in the literature to individuals’ early socialization context and how attitudes developed during early adulthood persist over time, an indicator capturing whether respondents were born in Canada is also included.

The analysis proceeds in three steps, taking inspiration from the approach used by Cochrane and Perrella (Reference Cochrane and Perrella2012). For each of the three outcome variables, a first model—called the base model—incorporates all the variables listed above, except for provincial economic indicators and provincial partisanship. This base model thus provides a description of interprovincial variation on each of the three outcomes when accounting for individual-level socio-demographic indicators and federal partisanship. This specification controls for a composition effect—where interprovincial differences could merely be due to the distinctive socio-demographic makeup of each province.

The base model is estimated in the following way:

$$y_i = \alpha + {\bf Provinc}{\bf e}_{\bf i}{\bf \tau } + {\bf FedPI}{\bf D}_{\bf i}{\bf \theta } + {\bf X}_{\bf i}{\bf \kappa } + {\bf \lambda }_{\bf i} + {\bf \gamma }_{\bf i} + \varepsilon _i$$

where y _i is one of three outcome variables and α is a constant. ${\bf Provinc}{\bf e}_{\bf i}$ is a vector of province dummy variables, ${\bf FedPI}{\bf D}_{\bf i}$ is a vector of federal partisanship dummy variables (Liberal, Conservative, New Democrat, other or none) and ${\bf X}_{\bf i}$ is a vector of socio-demographic controls. ${\bf \lambda }_{\bf i}$ represents a vector of year fixed effects, while ${\bf \gamma }_{\bf i}$ constitutes a vector of survey fixed effects.

A second model specification keeps all the covariates included in the base model but also incorporates provincial economic indicators (unemployment and growth rate). Incorporating these economic indicators in a second model specification allows us to compare our point estimates for the province dummy variables before and after accounting for economic conditions in each province. In doing so, we can evaluate whether macro-level economic factors, which have largely been overlooked in the literature, contribute to decreasing or magnifying cross-provincial differences. This approach is similar to the block-recursive modelling strategy that was pioneered by Miller and Shanks (Reference Miller and Merrill Shanks1996) and that has become commonplace in voting behaviour analyses, including within Canadian political science (see Anderson and Stephenson, Reference Anderson, Stephenson, Anderson and Stephenson2010; Gidengil et al., Reference Gidengil, Neil Nevitte, Everitt and Fournier2012).

Accounting for provincial economic conditions is critical, given that there are systematic differences in the performance of provincial economies and such differences could potentially translate into differences in political outlooks. Yet political culture relates to deeply held beliefs and values independent of contextual circumstances. Considering how the literature on Canadian provincial political cultures puts a lot of emphasis on diverging economic conditions to account for interprovincial variation in political culture, it is worth investigating exactly how economic conditions contribute to such variation.

The second model takes the following form:

$$\eqalign{y_i & = \alpha + {\bf Provinc}{\bf e}_{\bf i}{\bf \tau } + {\bf FedPI}{\bf D}_{\bf i}{\bf \theta } + {\bf X}_{\bf i}{\bf \kappa } + {\bf \lambda }_{\bf i} + {\bf \gamma }_{\bf i} + \beta _1Unemp_i \cr& \quad+ \beta _2Growth_i + \varepsilon _i}$$

where provincial unemployment and growth rates are added to the base model.

Finally, a third model incorporates provincial party identification. While most of the observations used come from the CES, which is conducted during federal election campaigns and focuses mostly on national politics, the survey items used never refer explicitly to “federal” governments and politicians. It thus seems possible that citizens who identify with their provincial government might have a generally more optimistic outlook on politics that could carry over to the three outcome variables. Further, the CPEPS was fielded right after provincial elections, which might have primed respondents to think about provincial politics when answering it.

Provincial party identification (PID) is left aside until the third model, given that provincial party identification is asked inconsistently in the CES, and thus incorporating it considerably decreases the number of surveys included in the analysis, which makes the estimates less robust to idiosyncratic fluctuations.

The third model is estimated as follows:

$$y_i = \alpha + {\bf Provinc}{\bf e}_{\bf i}{\bf \tau } + {\bf FedPI}{\bf D}_{\bf i}{\bf \theta } + {\bf X}_{\bf i}{\bf \kappa } + {\bf \lambda }_{\bf i} + {\bf \gamma }_{\bf i} + \beta _1Unemp_i + \beta _2Growth_i + \beta _3ProvPID_i + \varepsilon _i$$

where provincial partisanship is added to the list of covariates. The latter variable is operationalized as a binary variable indicating whether the respondent identifies with the party that formed the government in their province when they were interviewed.

The inclusion of economic conditions and provincial partisanship in the models allows this study to expand upon the original analysis conducted by Simeon and Elkins (Reference Simeon and Elkins1974) and its subsequent replication by others (Henderson, Reference Henderson2010; McGrane and Berdahl, Reference McGrane and Berdahl2013). Using a much shorter time frame, the authors could not test the potential impact of these two important factors on provincial differences in political culture. In doing so, this study can paint a clearer picture of each province's underlying political culture, filtering away the potentially substantial impact of contextual factors that might bias the results.

Whereas the PDP and CPEPS surveyed respondents only once, most CES surveys used in the analysis incorporated three waves (campaign-period, post-election and mailback surveys), where respondents to the campaign-period survey were recontacted for both the post-election and mailback surveys. Some questions used were asked in the post-election or mailback surveys, which tend to have lower response rates. Given that answering the post-election and mailback surveys is unlikely to be randomly determined, the analysis cannot assume that observations are all independent and identically distributed, which invalidates some of the assumptions on which test statistics rely in regression analysis. To address this issue, bootstrapped standard errors are used instead of the conventional asymptotic standard errors.Footnote ⁷ This more conservative approach relaxes the assumption of random sampling while providing more robust estimates of the uncertainty surrounding our regression coefficients.

Finally, an important methodological caveat needs to be addressed. When pooling multiple surveys, one needs to be aware of the possibility that peculiarities of each survey (such as sampling design, question wording, answer choices or sample size) could be driving the results. Several precautions were taken to alleviate this risk in the present study. First, only surveys whose question wording and answer choices were directly comparable were used. The coding of the answers was also standardized to ensure comparability.Footnote ⁸ Further, I rely on a two-way fixed effects strategy using both survey and year fixed effects, which models heterogeneity across time and surveys to prevent it from driving the results. The survey fixed effects are designed to capture the possibility that each of the three surveys (CES, CPEPS and PDP) could yield systematically higher or lower measures of the outcome variables, which could bias our results if unaccounted for. The year fixed effects are intended to capture the fact that political cynicism and political efficacy could systematically vary over time across the country and also potentially bias the estimates of interest. The two-way fixed effects strategy thus allows the models to distinguish between federal, provincial and survey-level variations in each outcome variable. In doing so, the analysis can distinguish between long-term patterns and idiosyncratic fluctuations. Finally, a leave-one-out analysis (results not shown) was conducted, where the models presented in the main text were estimated leaving out one survey at a time, to determine how sensitive to the inclusion of a given survey are to the results presented. The results unsurprisingly suggested some sensitivity to the 2019 CES, given its much larger sample size than other surveys. To resolve the issue, the two models that use data from the 2019 CES (both political efficacy models, as questions measuring political cynicism that are used in the analysis were not asked in the 2019 CES) were re-estimated using only a random subset of a quarter of the full 2019 CES sample. In doing so, we substantially reduce the weight of the 2019 CES on the results, bringing it in line with the weight of most other surveys. The results of this alternative specification are presented in Appendix E in the online supplementary information and mentioned in the main text whenever their interpretation differs from that of the results using the full sample.