Geographic Contexts and the Religious Cleavage— Johnston's Argument

Johnston published two analyses of religion and voting, both of which emphasize the importance of geographic and social context for explaining the persistence of the religious cleavage in Canadian party politics (Johnston, 1985, 1991). In both pieces he contended that religious affiliations should be understood as signifying something more than the attributes of individual voters. Rather, such affiliations represent group memberships that involve individuals in networks of association, tradition and affiliation. In his critique of Irvine's (1974) earlier analysis, Johnston argued that the intergenerational reproduction of the religious cleavage in Canadian voting depended not simply on the intra-family processes of socialization (as Irvine had contended), but the efficacy of these processes was itself contingent on the larger context in which families were situated:

Taking up Johnston's suggestion, Jon Pammett (1991) looked at the impact of the religious composition of Federal Electoral Districts (FEDs) on the propensity of Catholics to vote Liberal. Specifically, he found that increases in the density of Catholics across FEDs strengthened the relationship between Catholicism and Liberal support. In addition, concentrations of Catholics in an FED contributed to overall levels of partisan instability by drawing non-Liberals to the party (Pammett, 1991: 410). Pammett's findings are suggestive of a contextual effect in which Catholics in non-Catholic environments are less likely to vote Liberal than are their counterparts in heavily Catholic FEDs. Moreover, in heavily Catholic environments individuals, regardless of their personal, political or religious affiliations, are drawn to conformity with a locally predominant Liberal/Catholic political ethos (on this type of explanation more generally, see Blake, 1978; Eagles, 1995). We will return to this type of explanation in the interpretation of our empirical results.

Johnston's argument in “The Geography of Class and Religion in Canadian Elections” (1991) departs from this type of social influence contextual mechanism. Instead, he explains the apparent effects of religious context on individual political behaviour as the result of strategic or tactical voting. Since we are testing and extending his analysis in this piece, it is worth summarizing his argument and his empirical analysis in some detail here. Party support in English Canada can be depicted as a function of two oppositions, namely those of religion and class. The former is operationalized simply as Catholics versus non-Catholics and the latter as union members versus non-union members. Where Catholics are locally strong, their well-known support for the Liberals assures the party of local victory. This in turn discourages non-Catholics from voting sincerely (on the basis of their class interests, for example). The result is that where Catholics are present in substantial numbers, other prospective cleavages are not activated and overall relationships between class and party support are attenuated. Conversely, where Catholics are locally sparse, voters (including Catholics) will be more likely to vote on the basis of their class position rather than their religious affiliation. Johnston offers the following analysis:

Johnston's test of this model came from the 1979 federal election (excluding Quebec, since its party system dynamics were obviously different and, as overwhelmingly Catholic, it constitutes a “massive outlier”; see Johnston, 1991: 114–15), and several provincial elections. Provincial elections were included since at the time of his study Catholics constituted 45 per cent of the 1979 National Election sample and he was concerned that this near-majority would impart a national-level strategic calculus for voters, along the lines suggested above, that would discourage non-Liberal supporters from voting sincerely. By 2000 this is no longer a concern, in large part because Quebec, long the pivot of Catholic-related Liberal governments, now gives most of its support to the Bloc Québécois. We have chosen to avoid the complexities of incorporating voting at the provincial level into our analysis (for an exploration of the complex and spatially variable relationships between provincial and federal voting, see Bélanger, 2002).

Parliamentary constituencies are the primary unit of political competition in Canadian politics—governments are built one seat at a time. Federal ridings therefore represent the most plausible site to look for evidence of the kind of tactical voting that Johnston's model posits. However, Johnston (1991: 119) chose to operationalize the contextual presence of Catholics at the provincial level, claiming that “[m]edia reporting of poll and other information is commonly for the province-wide or metropolitan-area wide results.” According to this logic, non-Liberal voters develop perceptions of the hegemony of the Catholics or the Liberals at the provincial level and cast their tactical votes accordingly. He notes that voters may have difficulty finding constituency-specific information to assist them in their strategic calculus.

We are not convinced. We believe that the province is simply too large, too heterogeneous, too remote and inadequately relevant for federal elections to impress a strategic voting logic upon would-be tactical voters. By contrast, the federal electoral district is a much more meaningful and accessible environment to support the development of perceptual and strategic cues (Carty and Eagles, 2005). Recent research by Blais and Turgeon (2003: 455–61) found that fully half of voters in the 1988 election could identify the party which would finish third in their constituency, suggesting that local information is not as difficult to come by as might generally be believed. Similarly, in their analysis of the 1997 election, Blais et al. (2001: 343–52) present evidence that the local constituency race, but not the national-level question of government-formation, influenced the vote, but they estimated the proportion of strategic voters at only 3 per cent. This low level certainly casts doubt on the plausibility of Johnston's claim that strategic voting is the causal mechanism responsible for the observed geography of party support, while simultaneously opening the possibility that his original argument (1985), that broader social influence mechanisms that vary across local contexts are indeed at work. In our empirical analysis, we report evidence that is consistent with the latter interpretation.

Notwithstanding these issues, Johnston's position is that, even armed with sufficient local information, we should not simply rule out the possibility that voters might respond strategically to higher levels of political competition such as the province. He concluded: “If, as I hypothesize, the province is a valid unit for voters' strategic calculations, then the constituency percentage Catholic should not be analysed other than with the province percentage controlled” (1991: 120). Fortunately, we are able to take up this recommendation. Employing multilevel modelling, we are able to simultaneously estimate the impact of varying Catholic presence at the riding and the provincial level on the behaviour of individual voters.

Using probit analyses to estimate the relationships between Catholicism (individual and in the provincial context), union membership and party support, Johnston confirmed his expectations for the patterns of voting for both the Liberals and the NDP. The relationship between Catholicism and Liberal voting was strongest in provinces where Catholics predominated, reflecting the political salience of the religious cleavage. In provinces with few Catholics, however, the Liberals were less dominant. As such, the class cleavage appeared to grow in salience and the NDP was able to attract the support of union members. However, Johnston predicted that in these areas, middle-class (non-union) voters would be more inclined to support their “natural” class party, the Progressive Conservatives. However, his results contradict this hypothesis, since this party's support appears to become more (rather than less) strongly structured by class as the provincial density of Catholics increases. For this party Johnston admitted that “the story makes little sense” (1991: 126).³

To explore the plausibility of Johnston's argument as an account of voting in the 2000 election, we employed the post-election wave of the CNES 2000 survey (N = 2860). Following Johnston, we kept only non-Quebec respondents who voted in the election (N = 1596). Table 1 explores the representativeness of our working sample. Liberal voting is underrepresented in our sample, as is Alliance support. Some of this discrepancy is the result of non-response to the vote question, since 13.5 per cent (N = 215) of respondents refused to disclose their vote choice. Assuming that the non-response is roughly evenly distributed between Liberal and Alliance supporters (the discrepancy is about the same for either party, and sums approximately to the non-response level), we have decided against compensatory weighting. In our working data file, then, are 31 per cent union members, as against a general civilian unionization rate for the country as a whole of approximately 26 per cent. Similarly, Catholics comprise 26 per cent of our sample, whereas outside of Quebec, the 2001 census reveals that Catholics comprise about 31 per cent of the non-Quebec population. These minor discrepancies may introduce some sampling error into our analysis of the foundations of party support outside Quebec in the 2000 election, but they are unlikely to significantly undermine our test of Johnston's argument.

Representativeness of Working Sample

To the individual-level survey data we have appended census data (from Statistics Canada) on the concentrations of Catholics in respondents' FEDs and their province of residence.

Religion and Party Politics in the Fourth Party System

Obviously, Canadian society and politics have been transformed in many ways in the more than twenty years that have elapsed since the 1979 election. Writing in the 1980s, Johnston (1985: 92) noted that there had been a decline in the strength of the religious cleavage in Canadian voting over the 1965–1979 period. Most obviously, this long-term secularization of the population has had a further two decades to proceed since Johnston's study. Moreover, the religious landscape of Canada has become more complicated as a result of the substantial waves of immigration that have occurred in the last several decades. However, according to the 2001 Canadian census, three-quarters of Canadians are still Catholics or Protestants (and a further 16% declined to profess any religious faith; see Bibby, 2000).

Even more dramatic have been the changes to the country's party system (see Carty et al., 2000). The Conservative landslide of 1984 was based in large measure on Brian Mulroney's appeal to Catholics in Quebec. The collapse of the Mulroney coalition in the 1993 election ushered in an entirely new party system characterized by strong Reform/Canadian Alliance roots in Western Canada and the Bloc Québécois dominance in Quebec. Have these social and political developments finally eroded the religious foundation of party support?

The simple answer to this question is no. A 1996 survey analysis inquiring into the religious, social and political (including partisan) orientations of a sample of 3,000 Canadians, for example, concluded that “… religious variables are still among the most useful in differentiating party adherents (Guth and Fraser, 2001: 61.) Similarly, according to the principal investigators of the Canadian National Election Survey (CNES), outside Quebec, in the 2000 election 54 per cent of Catholics voted Liberal—other things being equal, a Catholic was 14 per cent more likely to vote Liberal than a non-Catholic (Blais et al., 2002: 93). As such, the religious cleavage remains centrally important in explaining election outcomes. “It would be impossible to understand the Liberals' victory in the 2000 election without recognizing the extent to which their strength outside Quebec hinges on the support of Catholics and Canadians of non-European origin. These two groups constitute the core of Liberal support outside Quebec” (Blais et al., 2002: 96). Clearly, in regards to Canadian voting behaviour, religion remains “… the strikingly peculiar house-guest who has overstayed his welcome”(Irvine, 1974: 560).

A second empirical underpinning of Johnston's argument concerns the uneven spatial distribution of Catholics. Is there evidence that Catholics are now more evenly distributed over the Canadian landscape? While definitive comparisons with the 1970s are not possible with the data at hand, it is clear that wide geographic variations in the concentration of Catholics in Canada's federal electoral districts persist into the new millennium. Outside Quebec, the 2001 census reveals that the percentage of Catholics in federal ridings ranges from about 10 per cent to over 90 per cent. The standard deviation of 14 per cent for the proportion of Catholics is almost half of the mean (32%) for that variable, confirming the general spatial variability of Catholicism. Clearly, the intervention of more than two decades has done little to diminish the potential relevance of Johnston's interpretation. The raw materials upon which his interpretation is premised seem to be firmly in place in 2000. With this, then, we are now in a position to introduce our test to see if the explanatory potential is realized.

Multilevel Models of Religion, Union Membership and the Vote in 2000

Our analyses begin at the individual level, with the familiar logistic regression (logit) model in which the relationships between union membership, Catholicism and vote choice are evaluated. The dependent variables for these—and all other models—are the odds of voting for each of the four major political parties campaigning outside Quebec. We then take account of the nesting of individuals within both riding and provincial contexts. As a second step, we estimate two-level models in which the individual-level relationships are conditioned by the concentration of Catholics in their FEDs.⁴

Since there are no ridings or provinces in which the proportion of Catholics is zero, we choose to centre the riding and provincial measures of Catholicism around their respective means. Though we are primarily interested in the random variables measuring the slopes of Catholicism and union membership and voting behaviour—given the contextual effects associated with Catholic density—centring the contextual measures renders the intercepts readily interpretable as the odds of a non-Catholic, non-union member supporting the party in a district and a province of average Catholicism.

To summarize our hypotheses, we expect that, at the individual level, Catholics will be more likely to support Liberals than non-Catholics, and the reverse should be true for Catholic-Alliance and Catholic-Progressive Conservative relationships. We have no particular individual-level expectation regarding Catholicism and NDP voting, but we do expect that union members should be more likely than non-members to support candidates from this party. Our primary interest is in the contextual effects associated with concentrations of Catholics at the FED and the provincial levels, though we expect that these relationships will primarily be associated, in terms of their magnitude or strength, with the local (FED) as opposed to the provincial level of aggregation. As the density of Catholics increases in the local and provincial contexts, we expect that the propensity of Catholics to vote Liberal will be enhanced. As the religious cleavage intensifies with the increased proportion of Catholics in the context, we expect that those who might otherwise be drawn on the basis of their class position to the Conservatives, the Alliance or the NDP, would be more likely to vote for the Liberal party. Specifically, the lower individual-level odds of a union member supporting the Conservatives or the Alliance should be further attenuated as the proportion of Catholics in the environment increases, but non-union members would also be more likely to support Liberals under these conditions. As such, we might well expect that the effects associated with union membership on party support for the CA, PC and NDP parties will diminish in magnitude and statistical significance in relation to the increased environmental presence of Catholics. Finally, we expect that the individual-level relationship between union membership and NDP support should be depressed as the environmental presence of Catholics increases.

Tables 2, 3, 4, and 5 present the results of these three levels of analysis for voting for each of the main parties in the 2000 election. Beginning with the Liberals in Table 2, the first model identifies the expected strongly positive relationship between Liberal voting and membership in the Catholic Church. Other things being equal, Catholics outside Quebec are almost twice (odds ratio = 1.95) as likely to support Liberal candidates than are non-Catholics. Membership of an individual in a trade union, on the other hand, does not exert any significant effect on the likelihood of an individual's voting Liberal. Are these individual-level relationships conditioned by the density of Catholicism at higher levels of aggregation? According to Johnston's hypotheses, only the slope of the Catholic-Liberal relationship should vary as a function of the contextual presence of Catholics (Johnston, 1991: 121). And as our variant of Johnston's argument leads us to expect, the contextual influence of Catholicism is evident in models 2 and 3, but it is riding-level Catholicism that intensifies the individual-level relationship between Catholicism and Liberal support. Once the riding-level effect is accounted for, the provincial density of Catholics exerts no additional significant effect on the individual-level relationships. This suggests that the possibility, recognized by Johnston, that provincial-level effects might appear as the result of model misspecification, spuriously capturing the operation of sub-provincial contextual influences (Johnston, 1991: 120), was in fact correct, at least in the 2000 election. There is no evidence, however, to support Johnston's contention that increasing the density of Catholics at either the riding or provincial level leads union members to be more likely to vote Liberal.

Multilevel Models of Class, Religion and Liberal Voting, 2000

Multilevel Models of Class, Religion and Canadian Alliance Voting, 2000

Multilevel Models of Class, Religion and Progressive Conservative Voting, 2000

Multilevel Models of Class, Religion and NDP Voting, 2000

With the emergence of the Canadian Alliance as a major contender in the fourth party system and prior to their merger with the Progressive Conservatives in 2003, it is necessary to look at the support for both parties of the Canadian political right in the 2000 election. Table 3 presents parallel models to those for the Liberals for the Canadian Alliance (CA) in 2000. The individual-level relationship between Catholicism and CA voting is strongly negative—all things being equal, Catholics were almost half as likely as non-Catholics to vote for an Alliance candidate. This finding lends support to the view that the Christian fundamentalists who support the Alliance diminish the attractiveness of the party to Catholics. Individual union members also appear to be less likely than non-members to support the Alliance, but this expected relationship is not statistically significant. Interestingly, model 2 suggests that this relationship is conditioned by the riding-level concentration of Catholics such that increases in the latter further reduce the odds of a Catholic voting for the Alliance. In neither of the first- or the second-level models does union membership exert a significant effect on the odds of Alliance voting. To this point, the results are generally consistent with the logic of Johnston's argument.

However, the picture becomes more complex when the provincial concentration of Catholics is included in the estimation (level 3). In the three-level model, both individual-level Catholicism and union membership diminish the odds of Alliance support. Increases in the proportion of Catholics in FEDs intensifies this negative relationship. Interestingly, increasing provincial concentrations of Catholics do not significantly influence the individual-level religious relationship with the CA vote, but this factor does slightly (and significantly) attenuate the negative propensity of union members to support Alliance candidates. As such, this positive odds ratio from the provincial level serves to cancel out some of the disposition of individual-level union members to avoid this party (all things being equal, the conditioned probability of an individual union member voting CA increases slightly, to .917 (see the level 3 model in Table 3), indicating that the relation is still negative when provincial religious composition is controlled, but less so than when based on the individual attributes alone). In this outwardly puzzling finding, we have some evidence of the putative weakening of the class cleavage in the more heavily Catholic provinces.

Looking at the second party of the Canadian political right, Progressive Conservative support in 2000 was less structured than that of the Alliance by either religion or class. Model 1, incorporating individual-level effects only, suggests that neither cleavage is related to the likelihood of Tory voting. Once the contextual presence of Catholics at the riding level is accounted for, however, individual Catholics become significantly less likely than non-Catholics to support Tory candidates. No other individual or contextual effects from either the riding or provincial level are significant.

Recalling that Johnston's own findings for the Progressive Conservatives were puzzling to him and out of line with his expectations, we are tempted to suggest that the unexpected patterns on the right of the Canadian political spectrum continue to defy interpretation using this framework. There simply is no evidence for either the Alliance or the Tories to suggest that union members are less likely to support right-wing parties when Catholics are locally weak (and the class cleavage is therefore free to dominate political choice).

Table 5 presents the parallel analyses for NDP voting in 2000. The estimations suggest only the weakest of social structuring for NDP support, perhaps reflecting the diminished state of the party in the 2000 election generally. Though union members are more likely than non-members to vote NDP, none of these individual-level relationships is statistically significant at conventional levels (though they come close). Inexplicably, the expected positive individual-level relationship only reaches statistical significance when the concentration of Catholics at the provincial (but not the riding) level is controlled.

Conclusion

We set out to replicate, update and extend Johnston's arguments about the interaction across geography of class and religion in structuring vote choice in Canada. We extend his original analysis by employing multilevel modelling techniques that allow for the simultaneous assessment of contextual influences from a variety of levels on individual behaviour. And we update the empirical test by looking at the state of these relationships in the current (fourth) party system.

Our results are, like Johnston's, somewhat mixed. On the one hand, we find that there is a clear riding-level contextual effect intensifying the individual-level association between Catholicism and Liberalism to be found in 2000. This suggests that the politicization of religion in Canada does indeed spring from social, rather than simply individual-level, processes—a finding that is somewhat comforting, given the general secularization of Canadian society and the absence of explicitly religious issues and platforms at election time. In addition, we are able to resolve an issue that Johnston could not. By estimating a multilevel model incorporating measures of the density of Catholics at both the riding and the provincial level, we demonstrate that voters are responding to the local and (generally) not the provincial-level influences. This is an important finding, since it reinforces a variety of other research that suggests that the local riding is an important source of political stimuli for voters (Blake, 1978; Cutler, 2002; Carty and Eagles, 2005). Like most research that uncovers evidence of contextual effects, we cannot identify the precise causal mechanisms responsible for producing such contextual effects on religion and the vote. However, we at least have sharpened the geographic focus by identifying the riding as being closer in size than the province to the actual scale at which these contextual influence forces operate.

Also in keeping with our expectations based on Johnston's argument is our finding that the odds of Catholics or union members voting for CA candidates diminishes further as the density of Catholics at the riding level increases. However, as provinces become more Catholic, the probability of a union member voting CA was enhanced, thereby weakening the negative individual-level relationship slightly and diminishing the strength of the class cleavage.

On the other hand, however, we find that Progressive Conservative support is unresponsive to the density of Catholics at either the local or the provincial level. Individual Catholics are less likely to vote PC than others, but this relationship is invariant across variations in contextual Catholicism. Similarly, NDP support appears to be associated with an individual's union membership, although this effect only reaches statistical significance when the religious composition of both the FED and the province are controlled.

On the strength of this mixed evidence, we conclude (with others such as Blais, 2005) that religion remains as a significant determinant of the vote choices of Canadians in 2000. That is, there appears to be a relatively strong individual-level grounding of religion and party choice that operates for three of the four major pan-Canadian parties (not the NDP). Admittedly, our empirical model is highly stylized and simplistic. In doing this we attempted to replicate Johnston's original analysis as faithfully as possible. As suggested by Johnston, these individual-level relationships are strongly conditioned in important ways by the religious composition of riding environments (though not that of the province).

For its part, the NDP vote in 2000 was not at all structured by religion and only in the weakest of terms was its vote related to union membership. There is relatively little evidence suggesting that the politicization of class is conditioned by the local or provincial density of Catholics, and as such, little evidence for an interaction of these cleavages in determining the level of strategic and sincere voting. We are therefore unable to confirm this aspect of Johnston's (1991) argument. Rather, our findings are suggestive of a conventional “contextual effect” stemming from the religious cleavage, in that individual Catholics are more likely to vote Liberal when surrounded by other, like-minded Catholics (Pammett, 1991). The fourth party system may remain profoundly rooted in tribal loyalties, but religion appears to trump class as a determinant of vote choice, regardless of the religious composition of the voter's context.