Local conditions and nationalist voting: Evidence from the 2011 Canadian Vote Compass

Transcription

1 Local conditions and nationalist voting: Evidence from the 2011 Canadian Vote Compass Peter John Loewen Department of Political Science, University of Toronto Maxime Héroux-Legault Department of Political Science, University of Toronto Carolina de Miguel Department of Political Science, University of Toronto July 1, Introduction As with many countries in the world, Canada witnessed a rise in support for nationalist parties 1 in the second half of the 20 th century, coinciding with the election of an official separatist government in the French-speaking province of Quebec in The Parti Québécois (PQ) would go on to hold referenda on secession from Canada in 1980 and again in In the interim, Canada also saw the rise of a French-Canadian nationalist party within its federal Parliament. In 1993, the Bloc Québécois (BQ) was successful in having MPs elected in twothirds of Quebec s constituencies. As a result, they held the status of the second largest party in the House of Commons (taking on the title of Her Majesty s Loyal Opposition ). The rise of these two parties Parti Québécois (PQ) and Bloc Québécois (BQ) has motivated substantial academic inquiry on the determinants of support for nationalist parties, which fits into a broader comparative literature on vote choice for nationalist options. Scholars have found that support for nationalist parties is driven by voters policy positions on issues of nationalism as well as by voters positions on socio-economic policy issues linked to the traditional left-right dimension. This work relies heavily on the 2011 federal election edition of Vote Compass Canada, a voting engagement application. The original research team on this project included Peter Loewen, Yannick Dufresne, Gregory Eady, Jennifer Hove, and most especially Clifton van der Linden, who is the Founder and Executive Director of Vote Compass Canada. More details on the project can be found at 1 In this paper we use the term nationalist parties to refer to a broad category of parties in which issues of nationalism are an important dimension of political competition. One of the goals of these parties is to defend the interests (cultural, linguistic, economic) of a certain geographically-defined nation. The comparative literature uses a variety of terms for these types of parties such as regional parties, ethno-regional parties, autonomist parties and separatist parties. We note that it can include parties which do not promote secession within a nation (e.g. United Kingdom Independence Party. 2 They were out of office between 1985 and

2 In addition, support for nationalist parties is correlated with feelings of attachment to one s nation and various individual-level demographic factors. Surprisingly there has been very little work that examines whether these individual-level determinants of nationalist vote vary across regions in different countries, across time within the same region, or across districts or localities within the same region. The assumption is presumably that individual predictors of nationalist support are constant across space and time, which has led to most studies of vote choice omitting an examination of context. This omission of context and the reliance on individualistic models is understandable for both theoretical and practical reasons. Rational choice theory argues that voting is an individual act [Downs, 1957, Enelow and Hinich, 1984] and as such the principal focus ought to be on individuallevel variations. Some scholars focus on sociological determinants of voting, but even then such determinants are measured and conceptualized as characteristics of the individual voter [Lazarsfeld et al., 1944, Berelson et al., 1986]. Practical concerns also prompt this focus on individual factors since the modal election study has a limited number of respondents (generally a few thousand) spread over hundreds of geographic units. Accordingly, estimating contextual effects at a local level is to invite problems of low statistical power. Put differently, it is hard to uncover meaningful local variation in the determinants of voting when only a few respondents live in each locale. 3 In this paper we question the assumption that the determinants of support for nationalist parties have a constant effect across localities, and we explore both the individual-level correlates of nationalist party vote choice and the extent to which these variables are conditioned by local context. To do so we focus on the case of support for the Bloc Québécois in the province of Quebec and we use data from the 2011 Canadian Vote Compass, a uniquely large survey of voters. 4 We estimate the contextual effects at the level of census divisions (roughly corresponding to cities and towns) using a two-step model recommended by Jusko and Shively [2005]. To demonstrate the robustness of our findings, we then test our arguments using a different dataset the 2011 Canadian Election Study as well as different statistical models. We first run an individual-level model of vote choice that shows that more nationalist policy positions and more left-wing ideology correlate with support for the Bloc Québécois across the province of Quebec. This is consistent with findings in the Canadian and broader comparative literatures that two main cleavages (nationalism and ideology) articulate support for nationalist parties. Our paper then turns to explore how two local contextual factors (percentage of English speakers and percentage of unemployed) condition the effect of the individual-level determinants of nationalist vote. Our main argument is that policy positions on nationalism should become more important in predicting vote for the Bloc Québécois in localities where voters feel threatened culturally or economically. In support of our argument, we find that voters policy positions on nationalism become more important in predicting a vote for the Bloc Québécois when the percentage of English speakers (our proxy for ethno-cultural threat) increases in their locality. Our findings have implications for how we think about nationalist parties electoral strategies, especially the trade-off these parties face between catering to identitarian policies or to classical ideological policies. If nationalist parties want to maximize electoral support across geographic constituencies they will have to tailor their message differently in different localities. More specifically, the emphasis they will place on nationalist policy positions should not be the same 3 We do note, however, that related and very promising work in small area estimation [Park et al., 2004, Jiang and Lahiri, 2006, Ghitza and Gelman, 2013] is presenting an easily applied solution to this problem. 4 Restricting our analysis to the province of Quebec in Canada, we still have more than thirty thousand respondents. 2

3 everywhere. Furthermore, our findings lend support to arguments made in the literature on the importance of an economic and ethno-cultural calculus among voters voting for nationalist parties. Our contribution within this literature is to empirically test how individual-level choices vary depending on these contextual factors. Finally, this paper seeks to contribute to the broader vote choice literature by examining the role that local context plays in individuals choice of parties. We assume that voters hold different calculi depending on the environment in which they are situated, and we try to show this by explicitly modelling how individual-level determinants of vote choice vary by local conditions that should plausibly affect voter s calculus. In what follows, we review the Canadian and comparative literatures on individual level determinants of vote choice for nationalist parties and apply their insights to our case of the Bloc Québécois (Section 2). In turn we explore how local context matters to the study of vote choice for nationalist parties and derive some empirical expectations of the ways in which contextual variables exacerbate or attenuate the effects of individual-level predictors of nationalist vote (Section 3). We then review our data and modelling strategy (Section 4), and we follow this with our analyses and discussion of results using two different datasets and methodologies (Section 5). We end with a discussion of the implications of our findings and further avenues of research (Section 6). 2 Individual-Level Determinants of Nationalist Voting What underlies support for nationalist parties? The existing literature suggests that support for nationalist parties is driven by voters policy positions on nationalism as well as by voters policy positions on the traditional left-right ideology. In addition, support for nationalist parties has been associated with feelings of attachment to one s nation (nationalist sentiment), and a series of individual-level demographic variables such as level of education, age, gender and religion. Most individual models of vote choice for nationalist parties include these variables, and we thus expect them to matter in predicting support for the Bloc Québécois in Canada. Policy positions on nationalism can vary from demands for cultural or linguistic recognition within a constitutional framework, to demands for greater political or fiscal autonomy to demands for outright independence. The expectation is that the more recognition or autonomy voters desire for their region the more likely they are to vote for nationalist parties. The fact that voters policy positions on nationalism explain vote for nationalist parties is not particularly surprising given that nationalist parties emerge in part to defend these nationalist positions and to push for policies of cultural recognition, autonomy and even political independence of their region. 5 For example, in line with the case of Canada scholars have found that political support for sovereignty translates into electoral support for the Parti Québécois [Bélanger and Nadeau, 2009]. We expect a similar relationship with support for the Bloc Québécois. Nationalist parties also compete for votes on the traditional left-right ideological dimension. In the abstract, nationalism and ideology are orthogonal to each other. In practice, however, political entrepreneurs will often frame their nationalist demands in the language of left and right politics. Noël and Thérien [2008], for instance, write: Still, in so far as they acquire a political expression, social identities almost always do so in the language of left and right. They detail that nationalism has been wielded equally by tenants of right-wing and left-wing discourses, including fascists, nazis as well as progressive proponents of decolonization and democratization. 5 What is less obvious, and will show in this paper, is that the importance of these nationalist positions is not constant across localities. 3

4 More directly relevant to the current research project, they explain that sub-state identities that sought recognition and autonomy in multinational countries were always tied either to the left or to the right, depending on the class and political alliances that defined them. On this topic, they agree with Guibernau [2013] that modern Québec nationalism is social-democrat. For this reason, we expect left-wing voters to be more likely to support the Bloc Québécois. This hypothesis is strengthened by previous works from Canadian scholars who have found that ideology, quite apart from policy positions on nationalism, is likewise thought to increase the probability of supporting a nationalist party in Quebec, whether at the federal or provincial level. [Blais, 2002, Gidengil et al., 2012, Bélanger and Nadeau, 2009]. In the case of the Bloc Québécois we expect left-wing voters to be more likely to vote for this party given its general policy orientation as a pro-labour, left-wing party and the background of its leader as a former trade union representative. Although we do not have a previous expectation of which dimension will be more important in the case of support for the Bloc Québécois, we posit that both factors will matter to explain vote choice for this party. 6 We expect that all these factors will be important predictors of vote for the BQ as they are for many nationalist parties in other developed democracies (mostly in Western Europe). The goal of this paper however is to go beyond this individual-level model of vote choice to explore whether these individual-level predictors of vote vary by local context. In other words, it is not particularly surprising that nationalist positions correlate with support for nationalist parties, but what is less obvious is that the importance of these nationalist positions might vary across local context, which poses particular challenges and opportunities for nationalist parties. In what follows we explore how the effect of policy positions on nationalism and ideology vary in concert with changes in local contextual factors within the province of Quebec. 3 The Role of Local Context Scholars that have approached the question of support for nationalist or secessionist options from a less individualistic perspective are more interested in the effect of regional and/or country-level variables on aggregate levels of support for nationalist and/or secessionist options [e.g. Sorens, 2005, Sambanis and Milanovic, 2011]. Arguments within this line of work are based on a rational choice approach that posits that individuals are more likely to pursue secession or autonomy for their region if they expect that the new arrangement will improve their community s welfare in cultural, political and/or economic grounds. Therefore individuals tend be more supportive of secessionist options if they live in regions that are very different than the rest of the country. Although these micro-foundations are based on an individual-level logic, the link between contextual variables and individual-level behaviour in most of these aggregate-level studies is assumed but oftentimes not directly tested. Instead scholars have mainly focused on testing aggregate correlations between regional or country-level factors and aggregate levels of support for nationalist or secessionist options, which have yielded very interesting findings. For example, Sambanis argues that regions with higher levels of income, resource-rich regions, or ethnoculturally distinct regions should all generate some measure of demand for autonomy [Sambanis and Milanovic, 2011, 6]. In a similar vein Sorens posits that economic and linguistic factors 6 The extent to which voters place more or less importance on the nationalist dimension versus the ideological dimension in voting for nationalist parties has become a source of scholarly debate and has important consequence for the nature of polices that nationalist parties pursue: identitarian policies versus classical ideological programs. In the literature on Spain for example research has consistently found that voters are more attentive to ideology than to nationalism (de la Calle, 2005; Balcells, 2007). (Rivero 2011). 4

5 explain support for secessionist parties, and finds that high-income regions are more prone to request independence than poorer regions and that regions with a region-specific minority language are also more likely to opt for secession [Sorens, 2005]. 7 In the Canadian context, a body of literature has relied on a similar rational choice approach to explain the attitudinal bases of support for sovereignty in Quebec. According to this literature, support for sovereignty is the product of a rational calculus on the part of Quebecers that has two main components: one linguistic and one economic. Quebecers are thought to ask themselves what gains or losses they would experience regarding the protection of French and the vitality of their economy in the case of Quebec independence. Findings suggest that Quebecers who expect gains in these two areas following independence are more likely to support independence than other voters [Mendelsohn, 2003, Blais et al., 1995, Blais and Nadeau, 1992, Nadeau and Fleury, 1995, Howe, 1998]. Others give more attention to regional differences. For instance, Gagné and Langlois [2002] find that support for sovereignty in Quebec is markedly lower in the regions of Outaouais than in the rest of the province. They surmise that the reason for this is that citizens of Outaouais are likelier to work for or have family working for the federal government, and thus would be worried about sovereignty for economic reasons. 8 However, because of limited data, they cannot actually test this hypothesis. (p.115) This work suggests the importance of economic and ethno-cultural contextual factors in explaining individual-level attitudes towards sovereigntist options. We hope to push this research forward by leveraging our large dataset to investigate the reasons behind regional variations in support for sovereignty. We build on these insights to argue that similar economic and ethno-cultural contextual factors should operate at the local level to affect individuals support for nationalist options and especially for nationalist parties. In line with this rational choice approach we argue that voters in Quebec are influenced by the local context in which they live. In particular if voters feel culturally or economically threatened in their locality then they should be more critical of the status quo and more inclined to think that an alternative scenario of autonomy (or even independence) for Quebec would improve their economic and cultural situation. If this is the case, voters that feel threatened culturally or economically should be more likely to value nationalist or secessionist options and to turn to parties that pursue these options. In order to test these predictions we look at how the influence of nationalist policy positions on vote for the Bloc Québécois vary according to two broad local conditions: the percentage of English speakers, which we use as a proxy of ethno-cultural threat, and average rates of unemployment, which we take as a proxy for general economic uncertainty. In particular we expect that the influence of nationalist policy positions on the vote for the Bloc Québécois will be greater in local contexts in which there are more English speakers. Feelings of threat and linguistic insecurity will be greater when individuals are exposed to more speakers of another language. This day-to-day reminder of a conflict over language should spill into politics, and make voters more likely to prioritize policy positions on nationalism when voting for the Bloc Québécois. It should likewise be a ready trope for enterprising politicians who wish to prime feelings of nationalism and link them to vote choice. In a similar fashion, we also expect that high unemployment will trigger the feeling of economic threat that would make nationalist policy positions a more salient dimension in voting for the Bloc Québécois. This leads to the following two hypotheses: 7 We note that Sorens finds that the linguistic effect is decidedly smaller than the economic effect. Sorens also devotes substantial attention to other variables. 8 The federal capital, Ottawa, is located close to the Outaouais region. 5

6 Hypothesis 1. The positive effect of nationalist policy positions on the vote for the Bloc Québécois should increase as the percent of English speakers increases in a locality. Hypothesis 2. The positive effect of nationalist policy positions on the vote for the Bloc Québécois should increase as the percent of unemployed people increases in a locality. In the previous section we posit that policy position on nationalism should have a positive (and significant effect) on the vote for the Bloc Québécois. This is not particularly surprising; nationalist parties emerge in part to defend these nationalist positions and to push for policies of cultural recognition, autonomy and even political independence of their region. Our argument however is that the importance of policy positions in predicting vote for the nationalist parties varies across local context. The implicit assumption in individual-level studies of support for nationalist parties has been that the effect of nationalism on vote is constant across space (i.e. electoral districts or other geographic units) in the regions that these parties compete in. In contrast we posit that there are reasons to expect heterogeneity across local context. In the sections that follow, we outline an empirical strategy that allows us to estimate the contextual conditionality of the factors associated with voting for nationalist parties. 4 Data and modelling strategy Our paper employs data collected in the 2011 Canadian federal election by Vote Compass, a voting advice application hosted by the Canadian Broadcasting Corporation website. 9 The application asks respondents thirty policy and attitudinal questions related to Canadian politics, in addition to a number of questions related to sociodemographics, voting behaviour, and evaluations of political parties and their leaders. The application received nearly two-million responses during the thirty-seven day campaign. The benefit of the Vote Compass data is that it allows us to estimate separate individual-level models within precise and low levels of geography. Whereas samples from standard election surveys range in the tens or hundreds of respondents per constituency, Vote Compass data number in the hundreds or thousands. These data should be treated with caution. Unlike person-to-person or telephone interviews, data input and respondent effort cannot be as closely monitored. Accordingly, we take some precaution with the data. We limit our set to those respondents who answered a complete schedule of sociodemographic questions. We make the assumption that while respondents may enter political positions on numerous occasions to see how their estimated political positions change, they are less likely to enter demographic information multiple times. A respondent who is clicking or speeding through the tool will skip non-mandatory questions. Once this filter is applied, and once we limit the data to French-speaking Canadians in Quebec, we are left with approximately 30,000 respondents. These respondents are nested within 98 census divisions. 10 This converts into an average of 306 respondents per census division. Our modelling strategy closely follows that of Jusko and Shively [2005]. In this two-step strategy an individual-level model is estimated in each geographic unit (which in this case is 9 This particular application, which is just one of a class of VAAs, has also been deployed in partnership with the Wall Street Journal and the Australian Broadcasting Corporation. As a matter of disclosure, Loewen has a passive financial interest in this application. 10 For purposes of comparison, we note that there are 75 federal constituencies in the province of Quebec. 6

7 a census division). 11 The estimation of separate models implies uniquely estimated variance and covariance structures. Coefficients of interest are then captured from each of these models and regressed against variables measured at the level of the geographic unit. The second level coefficients can then be inspected to estimate how much the influence of some individual-level variable varies by the levels of some contextual factor. The advantages of this approach are clearly articulated by Jusko and Shively [2005], and among them are straightforward interpretation of results, ease of visual inspection, and no loss in statistical efficiency in comparison to more standard HLM approaches [eg Snijders and Bosker]. 4.1 Individual-level Model The first model in our two-step procedure is a linear additive model of the probability to vote for the Bloc Québécois. Formally, we estimate the following individual-level model of vote choice: Y (V OT E) ij = α j + Nationalism ij β1 j + Ideology ij β2 j + Covariates ij β3 j + ε i j, (1) where, for example, β1 indicates the effect of a respondent i s policy positions on nationalism on a nationalist vote choice in census division j. As the subscripts imply, the model is estimated separately in every census division. For ease of interpretation and comparability across census divisions, we estimate an ordinary least squares regression at the first level. 12 The dependent variable is whether the respondent indicated a vote intention for the Bloc Québécois. It is a dummy variable. The key independent variable in our model is nationalism. Nationalism captures the voter s policy positions on nationalist issues and is an index based on the average response to three statements asked in the Vote Compass survey. Respondents were asked to indicate how much they agreed or disagreed with the following statements: The federal government should have a say when it comes to decisions about culture in Quebec, Quebec should be formally recognized as a nation in the Constitution and Quebec should become an independent state. The items exhibit reasonable internal consistency, returning a Cronbach s α=.70. The scale runs from 1 to 5 with higher values indicating more support for autonomy, cultural recognition or independence. The scale has a mean of 3.86 and a standard deviation of 1.01 Nationalist offerings are often bundled together with left-wing policies. To parcel out nationalism, we carefully construct a measure of left-wing preferences, as in a classic economic left-right dimension. To generate this dimension, we compile an index composed of three categories of questions about the economy, fiscal redistribution and social spending. 13 The index 11 We note an implicit assumption of this approach: the relevant level at which contextual effects occur is the local city or municipality, rather than a neighbourhood, for example. We return to a discussion of this in our conclusion. 12 This increases ease of interpretation because we can directly compare the size of coefficients across constituencies. Substantively, our results do not change if we use a logit estimation, but they prove more difficult to demonstrate. 13 The statements are: When there is an economic problem, government spending usually makes it worse. ; The federal budget deficit should be reduced, even if it leads to fewer public services. ; Canada should seek closer economic relations with the USA. ; How much of a role should the private sector have in health care? ; The government should fund daycare instead of giving money directly to parents. ; It should be easier to qualify for Employment Insurance. ; Workers should contribute more to their government pension plan (CPP/RRQ) so that it can offer bigger pensions. ; How much should wealthier people pay in taxes? ; How much tax should corporations pay? 7

8 has a Cronbach s α=.61. The index runs from one to five, with higher results indicating an ideologically left-leaning respondent. The mean of the index is 3.36, and its standard deviation is Our models also include controls for gender, age, and education. 4.2 Contextual-level Model Following the estimate of n = j different individual-level models, we then capture the coefficient on nationalism in each model and regress each separately on two aggregate-level characteristics in each census division, namely the log of long-term average unemployment (M=1.99, sd=.46) and the log of the percentage of English-speakers (M=0.65, sd=1.47). These measures are drawn from the most recently available census data. Estimates of the effect of policy positions on nationalism thus takes the form: β1j = α j + EnglishSpeakers j β1 j + Unemployment j β2 j + ε j, (2) The intercept in this model gives us the effect of nationalism on voting for a nationalist party when the share of English speakers is 0 and the long-term unemployment rate is also 0. We can then see how the effect of nationalism on nationalist voting increases or decreases as these two contextual factors change. 5 Results Table 1 presents the individual-level regression results. The model regresses vote intention for the Bloc Québécois on nationalism, ideology, age, gender, and education. The model is estimated in an OLS framework. All estimates are clearly distinguishable from zero, with the exception of education. Being male, a nationalist, and favouring leftist policy all increase the likelihood of voting for the Bloc. Only age decreases the likelihood. These results clearly accord with the picture of nationalist voting outlined at the start of our paper. Our variables point in the expected direction, and there is no statistical question of their influence. In the next section, however, we demonstrate that this influence varies by context much as we expected. In Table 2, we capture the Quebec nationalism coefficients from models estimated in 98 census divisions. We then regress these on two variables: the logged percentage of long-term unemployment in the division; and the logged percentage of English speakers in the division. What we find is that the effect of Quebec nationalism on vote choice is greater when the logged percentage of English speakers in the division is greater. Division of the coefficient by the constant tells us for each 1 point increase in the logged percentage of Anglophones in a district, the effect of nationalism increases by 3% (i.e /0.261). Given that the log of the percentage Anglophones in a district varies from -2.1 to 4.4 (as the percentage varies from 0.11 to 57), this is a sizeable effect. Indeed, an analytical inspection tells us that the likelihood that an individual with average levels of nationalism votes for the Bloc in the most French riding is.19, ceteris paribus. 14 However, an individual with an average level of nationalism living in the most English census division has markedly higher likelihood of In other words, the likelihood increases on the order of 100% (0.008*-2.15*3.86)= (0.008*4.4*3.86)=.40 8

9 By contrast, we do not find any conditional effects for the percentage of unemployed people in a locality. In other words, the influence of an individual s nationalist sentiments does not depend on the underlying economic hardship in a locale. This does not accord with our second hypothesis, but it does correspond to a certain logic. Rational choice accounts suggest that individuals are more likely to support secession when they reckon that their economic situation will be better off after secession[mendelsohn, 2003, Blais et al., 1995, Blais and Nadeau, 1992, Nadeau and Fleury, 1995, Howe, 1998]. This is an eminently logical claim. However, the place to find such an effect may be within individuals economic circumstances, i.e. we should search for a direct effect. There may be, in other words, little about living in an economically depressed locale that primes nationalism. Instead, it may simply prime the importance of beliefs about a better economic station in a new country. By contrast, there is a more logical link between living among anglophones - that is, others - and an amplified effect of nationalist sentiments. Living among others primes the importance of cultural difference, and provides a stronger link between this and nationalist vote choice. We next show that this effect can be corroborated using a different dataset and a different modelling approach Robustness: Reproducing the Analysis using Representative data The data that we use in the above analysis has the problem of having self-selected respondents. As such, there is no guarantee that such results can hold over a representative sample of the general population of French-speaking Quebecers. In order to improve the robustness of our findings we reproduce our analysis with data gathered during the 2011 Canadian Election Study. The CES is Canada s national election study and is conducted by random polling over the phone. Different questions were used in the Vote Compass and the CES surveys. For the CES replication the Quebec nationalism scale is the mean of five different questions, which ask respondents how much should be done for Quebec, whether they support sovereignty or not, how sovereignty will affect the state of the French language in Quebec and respondents standards of living, and finally whether there is a reason to have a sovereignist party in Ottawa or not. This scale has a Cronbach s α =.70. To measure ideology on the left-right scale we use a set of variables related to spending in various left-leaning sectors, such as healthcare, welfare, education, environment and immigration and minorities. This scale has a Cronbach s α =.52. We use ridings rather than census divisions as the geographical area of interest in these models, since the Canadian Election Study 2011 survey does not contain information that would allow us to pinpoint the census division in which a respondent lives. This adds to the stringency of the robustness test. We will have more faith in our results if they hold across both census divisions and electoral ridings than if they hold for only one such regional division. It deserves to be noted that census divisions and ridings will typically be of inversely proportional sizes, and thus solve a potential problem in the design. A possible concern would be that some census divisions are simply too big to accurately reflect neighborly or immediate experiences of voters. Since census divisions are drawn around urban lines, whereas ridings are delimited by population, using both levels of analysis solves this problem. While the city of Montreal is a single census division, it is divided in many ridings, and so studying ridings gives us a greater degree of precision in this region. The opposite relationship holds for census divisions that are smaller than ridings, which typically occurs in rural parts of the provincce. Given the number of respondents, it is impossible to run a separate model in every con- 16 And one which is arguably more appropriate to the data at hand. 9

10 stituency. While we had, on average, more than 300 respondents per census division when using the Vote Compass data, we have to work with an average of 14 French-speaking respondents per riding when using the 2011 CES. Using the two-step model in these conditions would not be appropriate since it is difficult to obtain reliable estimates in the first step of the analysis for most ridings. We thus opt for a somewhat different approach. We estimate simultaneous regressions for the ridings in the highest or lowest third of the variable of interest. We then compare the coefficients across these regressions and verify whether there are significant differences between them. We repeat this procedure with the ridings having the lowest and highest unemployment rate. Table 3 presents results on the conditioning effects of the share of English-speakers in a riding. Comparing the coefficients for nationalism across the two regressions, we find that the effect of nationalism is greater in constituencies with more anglophones. A χ 2 test suggests a significantly greater effect (χ 2 = 2.8, p =.09. The effect is substantively large and similar to that estimated with the Vote Compass data. Indeed, the increase in the likelihood of voting for the Bloc in the most versus the least anglophone constituencies is 55% greater, net all other effects. 17 Table 4 presents results for the conditioning effects of unemployment. As with the Vote Compass data, we find that the effect of nationalism does not vary according to the extent of unemployment in a constituency (χ2 = 0.08, p =.77). In sum, our main findings regarding the importance of contextual effects hold regardless of the dataset (Compass Data or CES data) and methodology used for the analysis. We find that nationalism matters more for French voters when they live amidst a higher proportion of Anglophones. Given the many differences between the two models, including the different geographical areas and different measures, this replication is a reasonable guarantee that the results hold across a variation of conditions. These results also hold over two other tests of robustness which are described in the Appendix. Despite this robustness, we acknowledge that other contextual factors that we have not studied might also matter. 6 Conclusion Our main finding is that the influence of nationalism on the decision to vote for the Bloc Québécois will be greater in a context where the Other (in our case, English Quebecers) is more prevalent. This finding may seem counter-intuitive to many students of Quebec politics. They will rightly point out that levels of support for nationalist parties is higher in remote regions of Quebec, where there are less English speakers, and lower in Montreal, where English Speakers are more prevalent. How can our results be reconciled with this general pattern? Two things need to be considered to solve this apparent riddle. First, our findings are only relevant for French-speaking Quebecers. It is natural, then, that ridings with a larger population of English Speakers would show lower aggregate support for nationalist parties since English- Speaking Quebecers almost never support these parties. In other words, even if French-speaking Quebecers are more likely to vote for nationalist parties in a context where there are more English speakers, this latter group s political preferences will also affect the aggregate vote in a given region and push support for Quebec nationalist parties downward. Second, our results show that French-Speaking Quebecers who are nationalists will be more likely to vote for the Bloc Québécois when living among English speakers. However, it is possible that levels of nationalism 17 i.e..454/.293=

11 vary across regions within Quebec. For these reasons, our findings are not sufficient to make inferences about aggregate levels of support in various ridings and regions of the province. Rather, they detail how context may affect the decision to vote at an individual level, especially across voters who hold similar nationalistic views. Our findings are important to scholars of nationalist politics. At an individual level we find that respondents are more likely to support a nationalist party when they hold more nationalist sentiments. This is unsurprising. However, we show that this effect is not always unconditional. Instead, we show that nationalism plays a stronger role in explaining nationalist voting among voters who live among citizens of another group - in our case, English Quebecers. But this effect does not vary according to the economic hardship in a voter s locale (which we capture through long-term unemployment). These findings seem to corroborate the expectation that ethnocultural threat (but not economic contextual threat) plays a role in individuals nationalist vote choice. These findings have implications for how we think about nationalist parties electoral strategies, especially the trade-off these parties face between catering to identitarian policies and catering to classical ideological policies. If nationalist parties want to maximize electoral support across geographic constituencies they will have to tailor their message differently in different localities. More specifically, the emphasis they will place on nationalist policy positions should not be the same everywhere. At some level the geographic variation allows nationalist parties to navigate both dimensions. It also allows us to better understand why nationalist parties also take pains to articulate other policy positions: nationalism will play better in some places than in others, and something else should fill out a voter s decision calculus. More broadly our findings shed light on some controversies regarding inter-ethnic conflict. The literature is unresolved regarding how contact should affect such conflict. Some think that more contact means more mutual understanding and greater tolerance between groups [Kalin and Berry, 1982]. Others argue exactly the opposite, that is, a majority will feel threatened when a minority becomes too strong and will vote to limit the powers of the minority [Tolbert and Grummel, 2003]. Others adopt a more nuanced view, and claim that while individual-level contact breeds tolerance and mutual understanding, more contact yields more conflict in the aggregate [Pettigrew et al., 2006]. Our results corroborate the second theory. We have found that when they live among more members of the lingusitic minority (English speakers), French Québécers are more likely to let nationalist positions determine their vote. In this sense, our results support the idea that contact breeds national conflict rather than mutual understanding. There are limitations to our findings. First, we have specified a rather broad locale, namely towns and municipalities. It may very well be the case that the relevant context is the much more localized and selected areas in which individuals transact their daily lives. It may also be the case that the effects of economic hardship do condition determinants of nationalist votes, but we have failed to capture the level at which this occurs. All of this suggests future research. Second, our results are obtained in an election in which the party in question was in a marked decline. Indeed, they found themselves outpaced by a left-wing but essentially non-nationalist rival. It may well be the case that the factors that condition support for a nationalist party change when the party is on the ascent rather than the descent. Third, our results rely largely on a new, untested source of data. However, we are able to replicate these results using a more conventional dataset. We think this is thus a small limitation. Going forward, our findings suggest three extensions. First, we can consider and estimate models at different levels of geography, including neighbourhoods and towns. Second, we can consider other contextual level factors, including employment types and local political factors, 11

12 for example the presence of incumbents. Third, we can consider different methods for estimating causal effects, in particular small-area estimation approaches [Park et al., 2004, Jiang and Lahiri, 2006, Ghitza and Gelman, 2013]. Individuals live among other individuals. In doing so, they engage in contact and they share political and non-political experiences. While we have well-established and tested models of vote choice, we know less about how they vary from one place to the next. Our principal contribution, then, is to show that the effects of well-known variables can sometimes vary, and sometimes substantially. Acknowledging and estimating this can help us both understand voters and evaluate theories of their behaviour. 12

13 References Éric Bélanger and Richard Nadeau. Le comportement électoral des Québécois. Cambridge Univ Press, Bernard R Berelson, Paul F Lazarsfeld, and William N McPhee. Voting: A study of opinion formation in a presidential campaign. University of Chicago Press, André Blais. Anatomy of a liberal victory: Making sense of the vote in the 2000 Canadian election. University of Toronto Press, André Blais and Richard Nadeau. To be or not to be sovereignist: Quebeckers perennial dilemma. Canadian Public Policy/Analyse de Politiques, pages , André Blais, Pierre Martin, and Richard Nadeau. Attentes économiques et linguistiques et appui à la souveraineté du québec: une analyse prospective et comparative. Canadian Journal of Political Science, 28(04): , Anthony Downs. An economic theory of democracy James M Enelow and Melvin J Hinich. The spatial theory of voting: An introduction. Cambridge University Press, Gilles Gagné and Simon Langlois. Les raisons fortes: nature et signification de l appui à la souveraineté du Québec. PUM, Yair Ghitza and Andrew Gelman. Deep interactions with mrp: Election turnout and voting patterns among small electoral subgroups. American Journal of Political Science, 57(3): , Elisabeth Gidengil, Neil Nevitte, Andre Blais, Joanna Everitt, and Patrick Fournier. Dominance & Decline: Making Sense of Recent Canadian Elections. University of Toronto PressHigher education, Montserrat Guibernau. Nations without states: political communities in a global age. John Wiley & Sons, Paul Howe. Rationality and sovereignty support in quebec. Canadian Journal of Political Science/Revue canadienne de science politique, 31(1), Jiming Jiang and P Lahiri. Mixed model prediction and small area estimation. Test, 15(1):1 96, Karen Long Jusko and W Phillips Shively. Applying a two-step strategy to the analysis of cross-national public opinion data. Political Analysis, 13(4): , Rudolf Kalin and John W Berry. The social ecology of ethnic attitudes in canada. Canadian Journal of Behavioural Science/Revue canadienne des sciences du comportement, 14(2):97, Paul F Lazarsfeld, Bernard Berelson, and Hazel Gaudet. The people\ s choice; how the voter makes up his mind in a presidential campaign

14 Matthew Mendelsohn. Rational choice and socio-psychological explanation for opinion on quebec sovereignty. Canadian Journal of Political Science, 36(3): , Richard Nadeau and Christopher J Fleury. gains linguistiques anticipés et appui à la souveraineté du québec. Canadian Journal of Political Science/Revue canadienne de science politique, 28(1), Alain Noël and Jean-Philippe Thérien. Left and right in global politics. Cambridge Univ Press, David K Park, Andrew Gelman, and Joseph Bafumi. Bayesian multilevel estimation with poststratification: state-level estimates from national polls. Political Analysis, 12(4): , Thomas F Pettigrew, Linda R Tropp, et al. A meta-analytic test of intergroup contact theory. Journal of personality and social psychology, 90(5):751, Nicholas Sambanis and Branko Milanovic. Explaining the demand for sovereignty T Snijders and R Bosker. Multilevel analysis: An introduction to basic and advanced multilevel modeling, Jason Sorens. The cross-sectional determinants of secessionism in advanced democracies. Comparative political studies, 38(3): , Caroline J Tolbert and John A Grummel. Revisiting the racial threat hypothesis: White voter support for california s proposition 209. State Politics & Policy Quarterly, pages ,

15 7 Tables Table 1: First-level Model Variable Coefficient (Std. Err.) Male (0.005) Age (0.000) Education (0.001) Nationalism (0.002) Ideology (Left) (0.004) Intercept (0.019) N R F (5,31602) Significance levels : : 10% : 5% : 1% 15

16 Table 2: Variation of the effect of Quebec nationalism Variable Coefficient (Std. Err.) Percentage of Anglophones (log) (0.004) Percentage of Unemployment (log) (0.012) Intercept (0.024) N 98 R F (2,95) Significance levels : : 10% : 5% : 1% This table shows whether contextual variables affect the strength of the effect of Quebec nationalism on the decision to vote for the Bloc Québécois. The first row shows that the logged percentage of Anglophones in a census division has a significant positive effect on the strength of Quebec nationalism. However, the same cannot be said of the logged percentage of unemployment in a census division. 16

17 Table 3: First Level Model depending on Proportion of English Speakers in the Riding Variable Coefficient (Std. Err.) Equation 1 : Low Proportion of Anglophones Nationalism (0.059) Ideology (Left) (0.111) Male (0.035) Education (0.009) Age (0.001) Intercept (0.105) Equation 2 : High Proportion of Anglophones Nationalism (0.076) Ideology (Left) (0.103) Male (0.041) Education (0.009) Age (0.001) Intercept (0.120) N 668 Log-likelihood. Significance levels : : 10% : 5% : 1% This table replicates the analysis using CES data. It shows that there is a strong difference between the coefficients for Nationalism in ridings with low and high proportions of Anglophones. 17

18 Table 4: First Level Model depending on Unemployment Rate in the Riding Variable Coefficient (Std. Err.) Equation 1 : Low Proportion of Unemployment Nationalism (0.058) Ideology (Left) (0.092) Male (0.031) Education (0.008) Age (0.001) Intercept (0.093) Equation 2 : High Proportion of Unemployment Nationalism (0.070) Ideology (Left) (0.131) Male (0.044) Education (0.010) Age (0.001) Intercept (0.152) N 746 Log-likelihood. Significance levels : : 10% : 5% : 1% This table replicates the analysis using CES data. It shows that there is only a small difference between the coefficients for Nationalism in ridings with low and high proportions of unemployment. 18

19 8 Appendix We ran further tests of robustness to ensure that the relationship we identity can be found through a variety of tests and specifications. In addition to the robustness check discussed in the test, we present two more tests in this Appendix. One possible problem with the research design is that our respondents are self-selected in the Vote Compass dataset. Even though our results hold when we run a similar analysis on CES data, which is characterized by telephone interviews and representative of the Canadian population, there may remain concerns regarding whether the relationship is truly valid in the Vote Compass data. To allay any such concerns, we apply weights to the analysis to make the data more representative. The weights are a function of the individual-level variables used in the first step of the analysis, that is, age, gender and education. The weights are based Census Canada information for each census division. Table 5 shows the results of this new analysis. It confirms that the relationship holds even when weights are added to the analysis. We conduct a final robustness test by trying to replicate our results using a multilevel model. The variables are the same. We still investigate the effects of age, gender, education, nationalism and ideology at the individual level, and the effects of the log of unemployment and the log of the percentage of Anglophones in the census division at the second level of analysis. However, we need additional variables to investigate the variation in the effect of nationalism on the decision to vote for the Bloc Québécois conditional. For this reason we add to the model an interactive variable that is the multiplication of Quebec nationalism and the logged percentage of Anglophones in the census division and a second interactive variable that is the multiplication of Quebec nationalism and the logged percentage of unemployment in the census division. The analysis confirms the results we obtained using the two-step model. Results can be found in Table 6. 19