Political Partisanship and Economic Outcomes: Canada, 1870–2020

Abstract

1. Partisanship at the Municipal Level and Its Overall Presence in the Canadian Electorate

Political parties compete to govern by offering distinctive party platforms. For this reason, political parties can be expected to have distinctive influences on the economy. In this paper, we examine the extent to which left-right partisan political differences have played a role in influencing the aggregate economic environment or the economic policies of Canada’s municipal, provincial, and federal governments.

We begin by noting that while political parties play a central role in organizing the operation of Canadian provincial and federal governments, political parties are largely absent from the local level. In only two of Canada’s cities—Vancouver and Montreal—do parties contest municipal elections and while other large cities, such as Toronto and Ottawa, have had loose coalitions of council members forming around individuals with similar ideological positions, these are exceptions to the general rule. As Lucas (2022, p.104) writes, ‘most local elections are both formally and informally non-partisan, and voters often know nothing about the partisanship of their municipal candidates’. By and large the broader social concerns of the largest cities diverge from the administrative issues that are the focus of community governance and ‘important policy issues like housing, social policy and sustainability…appear not to have “trickled down” into policy debates in small and medium sized communities” (Goodman and Lucas 2016, p.46).

While nonpartisanship remains an important characteristic of municipal politics, Lucas (2022) finds clear support for a match between the partisan character of a (provincial or federal) district and the partisanship of its municipal representatives. Even with such ideological connections across leaders, however, most researchers find that the attachment of Canadian voters to a specific party is somewhat loose. In their classic study of the characteristics of the Canadian electorate, Leduc et al. (1980, p. 403) find that ‘the data on partisanship clearly demonstrates that many Canadians have partisan attachments that can be characterized as flexible in nature. These voters are either unstable in their partisanship over time, inconsistent in their partisan attachments between the national and provincial levels of the federal system, or weak in the intensity of partisanship… This is not to deny the presence of a sizeable group of individuals who are firmly attached to a single party at both levels of government’.

The existence of a relatively stable partisan voting core surrounded by a flexible fringe has been used to explain, on the one hand, the existence of competing ‘brokerage parties’ in Canada—broad-based liberal and conservative policy platforms that span the centre-left and centre-right of the ideological spectrum.¹ On the other hand, the sizeable flexible fringe has led Leduc et al. (1980, p. 415) to hypothesize that ‘flexible partisans will be more easily influenced by the types of short-term factors that are commonly stressed in campaigns-issues…indicat[ing] that large-scale reversals in parties’ electoral fortunes from one election to the next remain significant possibility’. Such reasoning has helped to explain the wide swings in federal seat outcomes associated with the 1958 Diefenbacker, 1984 Mulroney, and 1993 Chr$e’$tien one-sided election victories. For our purposes, the finding of significant partisan flexibility underscores the fact that Canadian voters can be influenced by policy and/or circumstance despite having a partisan preference.

In common with that has been happening across other OECD countries (see Potrafke 2017), many Canadian political scientists have registered a recent growth in partisan polarization (Johnston 2023). This is perhaps most noticeable in the growth of partisan negativity (see Caruana et al. 2015; McGregor et al. 2015) and is often characterized as becoming more ‘Americanized’ as implied in the title of the recent book by Levendusky (2009), The Partisan sort: How Liberals Became Democrats and Conservatives Became Republicans. The consensus view of the recent trend in voter partisanship in Canada is perhaps best summarized by Kevins and Soroka (2018, p.118) who write in their history of political party rivalry that ‘party competition … has historically been characterized not by competition between left(/labour)-and right-oriented parties but rather by strong, ideologically flexible centrist parties…put in power by broad-based regional coalitions. Yet the party system has been marked by increased polarization since the 1980s, [supporting] the conjecture that Canadian parties are currently more effective at capturing (and perhaps enhancing) ideological divisions in society’.² This in turn may have contributed to greater vote volatility since, as Daoust and Bol (2020) suggest, growing polarization increases the proportion of voters who become strategic in their voting choices.

2. Partisanship in the Canadian provinces: 1976–2019

While formally absent from the municipal level, partisan party variety is at its most diverse at the provincial level. Unlike the federal level, where only two partisan party types have governed since Confederation (1867), parties at the provincial level reflect the importance of significant cultural and ideological differences across regions. At present, for example, Canada’s 10 provinces are governed by premiers from 6 different political parties: 5 Progressive Conservative, 1 Liberal, 1 New Democratic, 1 United Conservative, 1 Saskatchewan Party, and 1 Coalition Avenir Qu$e’$bec. Moreover, the provincial and federal versions of the same nominal party often differ in both tone and structure.³ Despite such nominal variety, however, the winning party in each provincial election can be characterized as either relatively left or relatively right of its primary election rival, and in virtually all provinces the winning left-oriented party is typically labelled as either a provincial Liberal or New Democratic Party.⁴ This forms the basis of the test below.

Partisan theories of the business cycle (Hibbs 1977; Alesina 1987) hypothesize that left-leaning governing parties either spend more or tax less in comparison to their more conservative rivals such that the economy will experience periods of greater expansion under their governance. This prediction is usually tested by regressing either Gross Domestic Product (GDP) growth rates or unemployment rates over time periods of left versus right-leaning party government. In Table 1, we undertake this test on a panel of Canada’s 10 provincial governments over the years 1976–2019. We use as our measure of economic expansion—contraction the provincial unemployment rate and test the prediction that unemployment rates will be lower in periods governed by the relatively left-leaning party.⁵ We begin with a fixed-effects regression test, where the public choice hypotheses of partisanship and a political business cycle are related to provincial unemployment rates and where the degree of unemployment is influenced by the percentage of young people in the population. We then investigate the extent to which changes in the Canadian economy overall significantly impact provincial unemployment rates. Finally, the lack of support for either public choice theory leads us to explore the applicability of a rational theory of partisanship where uncertainty surrounding election outcomes may temporarily affect the economy.

The results of our series of fixed-effects regression tests are presented in Table 1. In column (1), the years in which the party in power was either the Liberal or the New Democratic Party were regressed against provincial unemployment rates while controlling for provincial fixed effects and for differences in the percentage of the population that is young (typically having a higher unemployment rate).⁶ In addition, we allow for the presence of an electoral cycle by adding for each province a dummy variable indicating the year in which a provincial election was held. The lagged unemployment rate is included to capture the persistence present in unemployment rates over time.

The results in column (1) imply that shocks to provincial unemployment rates converge slowly onto a long-run value that, as expected, is highly (positively) sensitive to the percentage of the provincial population that is young.⁷ In terms of the two public choice cycle hypotheses, the data give no support for the existence of a provincial electoral cycle or for a Liberal/New Democratic Party (NDP) partisan cycle effect. Provincial unemployment rates across Canada are unrelated to whether a left or right-leaning provincial political party has been in office.

In column (2), we add a control for the spillover onto provincial unemployment of changes in the economic conditions arising elsewhere in the country. This external effect is measured by the lagged value of the national growth rate of real GDP per capita.⁸ The result shown in the third last row of Table 1 implies that some significant portion of the change in provincial unemployment rates is driven by what is happening in the overall Canadian economy. In effect, a faster growth rate in the outside Canadian economy stimulates the demand for provincial products which in turn encourages faster provincial growth and reduces the province’s unemployment rate. But while its inclusion improves the equation’s overall fit with the data, it is apparent that its inclusion has had no significant effect on any of the other coefficient estimates. That is, the data remain unsupportive of the hypotheses of a partisan cycle in unemployment rates or its presence in an election cycle.

Columns (1) and (2) test what Alesina (1987) and Alesina et al. (1997) have called traditional partisan theory. In column (3), we test for the presence of what they categorize as rational partisanship by examining the pattern of unemployment rates that follow the election of a left-leaning government. The argument here is that uncertainty over an election outcome allows the policies of an elected left-leaning government to have only a temporary surprise effect on the economy and thus on unemployment. As private agents’ policy expectations adjust to the election result, unemployment is expected to decrease in the years following as unemployment reverts to its natural rate. In column (3), we interact periods of left-leaning government with the successive years that follow an election victory and see that while the election itself has no immediate effect on the unemployment rate, there is a significant negative effect on the provincial unemployment rate in the first year of left-wing governance, together with a disappearance of that stimulating effect in the years that follow. Although evidence of a partisan unemployment rate effect is relatively weak (significant only at the 10% level), there is support given to the existence of a rational partisan electoral cycle in the Canadian provincial data.

In his meta survey of partisanship across OECD countries, Potrafke (2017) finds strong evidence of partisanship in the period before 1990 and a diminishing (but not disappearance) of partisanship thereafter. To see whether there is evidence of a similar sub-division in our sample, we re-estimated the model of column (2) when the time series was divided about 1990, 1995, and 2000. None of these cases provide evidence of a significant partisan or electoral business cycle difference across time.

The conceptual link that connects the partisan ideology of a party to the prediction of a partisan effect on unemployment is the type of policy set adopted by the government when that party is in office. From this it follows that for left-leaning parties to have reduced unemployment or boosted provincial output, expansionary policies must have been tried. While these policies may or may not have been successful, the necessity of this policy connection has led researchers to search for a partisan budget rather than a business cycle.⁹ Here, the search for partisanship has had mixed success.

For example, at the general level, the meta-analysis of 43 partisanship studies by Imbeau et al. (2001) finds that the correlation between the left-right composition of the governing party and the presence of expansionary policies is not significantly different from zero. Nevertheless, a number of mediating factors are found to allow them to conclude that partisan insignificance is an oversimplification. In the more specific Canadian context, the search for partisanship in provincial fiscal policies exhibits the same amount of mixed success. Blais and Nadeau (1992), for example, find that right-leaning governments tend to spend less and alter the composition of spending away from social expenditures and roads. Tellier (2006, pp. 379–380) uses a tri-part partisan distinction to find that the partisan left spends more than the centre and the right, but only as long as fiscal pressures do not force the left-wing government to cut expenditures.¹⁰Kneebone and McKenzie (2001) present perhaps the most detailed study of provincial budget data. Examining the period between 1966 and 1997, they find that both left and right party types significantly increased both spending and revenues when in office with parties on the left spending significantly more than parties on the right. The details of their budget analysis indicate that while all parties spent more on health, parties on the left favoured social services and industrial development while parties on the right spent more on education, recreation, and security—suggesting that budget composition may matter more than spending levels. In their pooled analysis of provincial budgets, Tellier and Imbeau (2004) find no evidence of a partisan cycle. However, this they attribute not to a general case of insignificance but to the presence of significant offsetting effects among six of the provinces. In this they concur with Petry et al. (1999) who earlier found evidence of partisanship in some but not all provinces. In concluding their study, Tellier and Imbeau state that ‘[c]learly our provincial time series show that there is a partisan cycle in budget deficits in the Canadian provinces. But this cycle is not the classical one suggested in the literature (the right is fiscally disciplined)… The meaningful distinction seems not to be between parties of the left and parties of the right but on some other dimension not yet identified’ (p.12).

In summary while evidence of partisan differences in the policy choices of political parties is widely present at the provincial level, these differences appear most often in the composition of services provided, differences that did not always translate into predictable left-right effects on aggregate provincial unemployment or output.

3. Partisanship at the Federal Level

In this section we test the partisanship hypotheses at the federal level in Canada over the post-Confederation period (1870–2020). It parallels the analysis undertaken earlier at the provincial level but uses real GDP per capita as the measure of aggregate economic performance (since unemployment data is available only from 1919 onwards). The study aims to determine whether the time periods in which the more left leaning of the two dominant federal political parties (the Liberal Party of Canada) are associated with higher rates of per capita growth. In doing so the analysis controls for demographic changes, urbanization, changes in composition of aggregate output and the growth rate of US industrial production. We begin by presenting in Figure 1 a look at the data by plotting the relationships among the real growth rate of GDP per capita (pc_growth), the growth rate of government size (gov_growth), and periods of Liberal Party governance (liberal = 1). The figure shows considerable variation in all three series but suggests no strong positive relationship among them. The simple correlation between periods of Liberal Party governance and the growth rate of government size is found to be negative (−0.065) and while its correlation with per capita GDP growth is positive but small (0.240).¹¹

Figure 1.

Growth rates of income per capita and Federal Government size with periods of Liberal Government Canada: 1870–2015.

Open in new tabDownload slide

In what follows we test more formally for the presence of traditional partisanship and follow this with Alesina’s test for rational partisanship. We then explore the possibility of a partisan budget cycle using the work of Ferris and Voia (2021). This questions whether left-leaning partisanship is observable in fiscal instruments and whether there is support for the hypothesis that the Liberal Party influenced the Bank of Canada through its final output.

Table 2 presents our test of whether the two partisanship hypotheses apply to Canada over the entire post-Confederation period (1870–2020). The tests regress the years in which the more left-leaning political party (the Liberal Party) formed the government against the growth rate of real GDP per capita while controlling for changes in the percentage of the population that is young, the degree of urbanization and changes in the composition of aggregate output—the share of agriculture, the degree of trade openness, and the size of the federal government. Because the Canadian economy is highly integrated with the USA, we also use the growth rate of the US Index of Industrial Production to account for productivity changes common to both countries. The prediction is that periods of Liberal government will be associated with higher rates of growth.

In column (1), we test for traditional partisanship by using a dummy variable for all years of Liberal government rule as our partisanship variable. The equation controls for the possible presence of a political business cycle by asking whether the aggregate output has responded systematically to federal election years. In column (2), we test for the presence of rational partisanship by including, as separate dummy variables, successive years of Liberal government. The prediction here is that the election of a Liberal government will lead only to an initial stimulus to aggregate output (arising from a response to the difference between the expectation and realization of a Liberal election victory).

For the entire post-Confederation period, the model in column (1) can be seen as providing no clear support for either traditional partisanship or a political business cycle. Although the coefficient estimates on Liberal and Election Year are both positive as predicted, neither is found to be significantly different from zero. At best the results are suggestive of the presence of partisanship somewhere within this overall post-Confederation period. The results of the test for rational partisanship in column (2), however, are more positive. The election victory year of the Liberal Party is the only year of Liberal governance for which the coefficient estimate is significantly positive, albeit at the 10% significance level. The results for the entire period at the federal level then concur with those found at the provincial level in rejecting the hypothesis of traditional partisanship and weakly supportive of rational partisanship.

We note, however, that while the link tests for model specification imply that models (1) and (2) are unlikely to contain specification error, the equations for the period as a whole are somewhat disappointing in their inability to explain more than half of the deviations in the growth rate over time. Moreover, not all of the control variables that would be expected to be significant are found to be so. The possibility that a single model of partisanship may not best represent the role of partisan differences in Canada then underlies the extension of this analysis in the section that follows.

While the data give no strong evidence of a partisan cycle in output, there remains the possibility of a partisan budget cycle in policy. To answer this question, we rely on the recent work of Ferris and Voia (2021) who examine the economic and political factors underlying three dimensions of federal fiscal policy and two dimensions of monetary policy in Canada for the years 1870–2015. Their findings are presented in Table 3.¹²

The model used in Table 3 to estimate the different dimensions of fiscal policy (the growth rates of federal government spending, taxes, and overall deficit) and monetary policy recognizes that policy choices are not independent but are tied together through the government budget constraint. For this reason, they are estimated together as a seemingly unrelated regression (SUR). Because missing variables may lead the equation errors to be correlated, a control function approach was used to model possible endogeneity in the error term through a two-stage estimation procedure that generates a proxy for the omitted variable.¹³ More specifically, the residuals from a first-stage regression of the US Index of Industrial Production and a set of political variables regressed on the growth rate of real GDP per capita are used as a covariate (bias correction term) that corrects for the inconsistency of the feasible generalized least squares used to estimate the SUR.¹⁴

For our purposes, the key feature of Table 3 is the highlighted row in the middle of the table that reports the relationship between increases in federal spending, taxes, and deficits and with periods of Liberal Party governance. The results in columns (1) and (2) indicate that although the estimated direction of change is consistent with partisanship in spending and taxation, neither change is significantly different from zero. Moreover, the coefficient estimate of the change in the deficit during periods of Liberal tenure (in column (3)) is perversely negative and significantly different from zero (at 10%). The data then provide no support for the hypothesis that Liberal government fiscal policy is sufficiently expansionary (as compared to the more right-leaning opposition) to have produced the differential outcomes observed for the growth rate of real GDP per capita.¹⁵

Table 3 also presents estimates of the differential effects of two elements of monetary policy (the rate of growth of the money base and its variance) during periods of Liberal Party government. That is, even though the Bank of Canada is nominally independent of partisan influence, monetary policy is another potential route by which the Liberal Party could have exercised influence on the Bank and through this onto output. The data, however, provide no support for this hypothesis.¹⁶ The coefficients in columns (4) and (5) are both insignificantly negative, opposite in sign to what would be needed to produce a traditional partisan cycle.

4. Diversity in Nature and Strength of Partisan Influence

The preponderance of different findings as to the presence and/or strength of partisanship across countries, party structures, and/or time intervals in meta studies such as those by Imbeau et al. (2001) and Potrafke (2017) and cross-country studies such as Cahan et al. (2019) suggest that a test that imposes a single model of partisanship for any extended time period may suppress important sub-periods when the strength or weakness of partisanship may be of particular importance. For example, in concluding his meta-analysis of 100 panel empirical studies of OECD countries, Potrafke finds, ‘[l]eftwing and rightwing governments pursued different economic policies in OECD countries prior to the 1990s and the size and scope of government was larger when left-wing governments were in power. Partisan politics did not disappear till the 1990s, but certainly became less pronounced’. The idea that the form and/or strength of partisanship may change across time allows for the possibility that the suggestion (rather than significance) of partisanship found earlier at the federal level in Canada may hide particular sub-periods when traditional partisanship was more apparent.

In Canada’s case, there is a second reason for suspecting that hypothesizing partisanship as only the interplay between a left/expansionary and right/contractionary political dynamic may be inappropriate. That is, while the 20th century may have featured the broadening of the voting franchise and a persistent left-right partisan divide over the need for redistribution and the size of the welfare state, the early years of Canada’s existence featured a left-right political divide for or against the National Policy—the set of policies supporting the creation of the country now known as Canada. That is, following Confederation in 1867 and the unification of former British colonies on the east coast and continental centre, the federal conservative party (then called the Liberal-Conservative Party under Sir John A. MacDonald) promoted a three-part plan to consolidate Canada as a country integrated economically from sea to sea. The first part of the conservative policy was an accelerated program of investment in a national railway to facilitate the absorption of Manitoba in the prairies and British Columbia on the west coast into Canada. The other two parts consisted of policies to encourage greater immigration and a series of industrial taxes to encourage the development of infant industries. The opposition Liberal Party favoured free trade with the United States (but could never achieve it).¹⁷ That is, while in later years, it was partisan ideological differences that led to the Liberal Party becoming the party promoting enhanced spending on redistribution and welfare, in the early years, it was the Liberal-Conservative government adopting protectionism, higher infrastructure spending and a larger-sized government as part of its policy of aggressive nation-building.

The model is estimated in two steps. In the first step, the threshold parameter (⁠$γ$⁠) is estimated by minimizing the sum of squared residuals from the segmented regression. In the second step, the coefficients of the model are estimated using standard regression techniques, given the estimated threshold. Finally, an information criterion, in our case the Bayesian Information Criterion (BIC), is used to guide the selection process of determining the number of thresholds and the specification of the threshold.

In Table 4, we present the results of using threshold analysis to indicate whether or not there exist up to three different thresholds of partisan difference corresponding to up to four periods of distinctive partisan behaviour. In column (1), we allow the data to indicate the date at which the nature of partisan influence changes discretely. In column (2), we include the change in government size to test whether changes in the nature of partisan influence are associated temporally with a change in the effect of government size on per capita growth.

Before dealing with the implications for partisanship, it is important to note that allowing the model to incorporate distinctive partisan regions leads to a significant improvement in the model’s ability to explain per capita growth rates over the post-Confederation time period. Unlike the findings in Table 2, where only some of the expected control variables were found to be significantly different from zero, all coefficient estimates now have their expected sign and are significantly different from zero. The distinguishing of distinctive partisan regions then appears to have removed a potential source of bias, allowing for a better representation of the underlying growth model. More formally, the BICs can be used to compare the goodness of fit of different two regression models with the model generating the lower BIC indicating a better balance between explaining the data and avoiding overfitting. Using this criterion, then, the dramatically lower BIC value of 411.77 for the threshold regressions of Table 4 compared to the BIC value of 852.8 for the simple regression verifies the earlier conjecture that the threshold regression is the better-fitting model.

The results presented in Table 4 indicate that there are a maximum of two distinct thresholds where the nature of partisanship has changed distinctly, so dividing partisanship in the post-Confederation period into three distinct regions. The first threshold appeared in 1885 (nearing the end of John A MacDonald’s tenure) and the second in 1933 (at the depth of the Great Depression in Canada). In region 1 (1870–1885), increases in the growth rate of per capita output and increases in the growth rate of federal government size are both found to be associated with the more conservative party holding office. This finding is the opposite of what would be expected under traditional partisanship but is consistent with the National Policy of protectionist infrastructural support and nation-building introduced by the conservative MacDonald administrations and with the expansionary effect of higher industrial spending on Canadian productivity. The data is then consistent with political challenges and policies of this period producing the effects of traditional partisanship coming through the relatively right-leaning rather than left-leaning party. In region 2, the period between 1886 and the Depression in 1933, the data suggest that traditional partisanship held sway. Higher growth rates were experienced when the Liberal Party was in power even though the expenditure increases associated with larger government size arose at the cost of somewhat lower growth rates. In the third region (the years following the Great Depression through the present), there appears to be no significant difference in outcomes arising from the partisanship of either party. It may be that the early adoption by all parties of Keynesian countercyclical policy and the acceptance of a slowly growing welfare state reduced the distinctiveness of both left-right policy differences and hence their differential effects on aggregate outcomes.¹⁸

5. Conclusion

In this paper, we have argued for the lack of traditional partisanship in the economic outcomes and/or policies adopted by Canada’s dominant political parties. Most often this is attributed to the central positioning of Quebec in Canada’s Confederation as ‘a French-speaking nation within a larger English-speaking nation’ (see Johnston 2017). Because of its preoccupation with cultural survival, the Quebec electorate has tended to vote as a block and because of its relative size in Canada’s majoritarian electoral system Quebec’s vote alone can put one party within easy reach of a parliamentary majority. For much of Canada’s history, the Liberal Party has been Quebec’s party of choice, and this secure base has allowed it to maintain a centralist position in the ideological spectrum and champion national unity. Successful competing parties have then entered on the other sides of the ideological spectrum, the Conservatives to the centre-right and the Labour, then New Democratic, parties further left. The centralist positioning of the dominant parties has led to policy convergence and regional party concentrations have often led to more competing electoral issues—Conservatives concentrated in the richer oil—producing regions of the west and the NDP concentrated in the Maritimes in the east and British Columbia in the far west. That is, it is not that Canadian political parties do not espouse traditional partisan policies but that ‘brokerage’ incentives and issues deriving from differences in culture, language, resource allocation, and regional industrial specializations have played at least as important a role in particular elections and during each governing tenure.

Here, we have examined empirically the role of traditional left-right partisanship, at both the provincial and federal government levels. Our most general finding is that a simple left-right delineation of partisanship does not have strong explanatory power in accounting for the longer-run evolution of aggregate economic outcomes at either the provincial or federal levels in Canada. While many studies document important differences in the composition of services favoured by the different parties, few studies find a consistent partisan effect on output and/or unemployment. In our own analysis of the Canadian provinces between 1976 and 2019 and the aggregate economy between 1870 and 2020, we find no evidence of a consistent left-right partisanship difference in aggregate economic outcomes but evidence weakly consistent with a ‘rational’ theory (based on the uncertainty in predicting a successful left-leaning party electoral victory).

Perhaps the most interesting result of our analysis is the finding that the nature of left-right partisanship has changed discretely over time in response to the challenges facing the parties at the federal level in Canada. In its earliest years of existence, partisan differences in how to proceed with nation-building meant that expansionary spending on infrastructure and immigration leading to both higher growth and larger government size was driven by the more conservative (rather than liberal) of two dominant parties. With policy challenges reverting to those associated with the expansion of the voting franchise through the early 20th century, greater spending and the growth of government now came under the auspices of the left-leaning party generating outcomes consistent with traditional partisanship. The data also suggest that in the period since the Great Depression no strong differences have arisen in the policy prescriptions, actions, and outcomes arising under left versus right-leaning party governments. This is perhaps consistent with the evolution of party structure in Canada towards what has been argued to be the dominance of two brokerage-type parties, parties that seek to preserve national unity by accommodating (rather than highlighting) differences among disparate communities (Carty and Cross 2010). In this environment competition across broadly overlapping constituencies would lead to the convergence of partisan policy differences and through this to similar induced effects on economic outcomes.

Footnotes

References

Appendix