Introduction
In a rich democracy such as Canada, minimum income protection—known in this country as welfare incomes—gives content to a basic social right (Bahle et al., Reference Bahle, Hubl and Pfeifer2011). This right to a minimum income, however, is implemented differently over time, across provinces and among different types of households. Concretely, the living conditions of the country's poorest citizens depend on yearly decisions made by provincial governments. Yet we know relatively little about these decisions. Anchored in the comparative literature on the politics of social assistance, this article considers the Canadian provinces from 1990 to 2017 to better identify the determinants of minimum income protection. The adequacy of welfare incomes can be associated, notably, with the strength of trade unions, the importance of redistribution, and a mechanism identified as a “welfare paradox” by Ivar Lødemel, which links the adequacy of social assistance to its salience as a program. When social assistance concerns more people, politicians pay attention.
Minimum income protection, or social assistance benefits plus associated transfers, is related in the Organisation for Economic Co-operation and Development (OECD) to welfare state generosity and the level of public debt (Noël, Reference Noël2019). Comparing minimum income protection across countries, however, remains difficult, because social assistance benefits must be estimated for a given family type, assuming full take-up of benefits, a value for housing and other in-kind benefits, and residency in a given city or region. In a federation such as Canada, where provincial governments have full jurisdiction over social assistance, it is possible to circumvent some of these difficulties because systematic data on minimum income benefits are available. The provinces are also similar in many respects, having identical parliamentary institutions, comparable party systems, close standards of living and fairly uniform political cultures. A most-similar systems comparison is thus possible.
These advantages come with problems of their own. Indeed, the similarity of Canadian provinces, their limited number and the relative stability of social assistance benefits over time make it difficult to establish the significance of explanatory factors in multivariate models. We can nevertheless consider four questions of interest:
1) Do political parties matter? Are leftist or centrist parties more likely to improve minimum income protection than parties of the right?
2) Do social actors, and in particular trade unions, encourage generous benefits for social assistance beneficiaries—the ultimate “outsiders”? Or do they concentrate their efforts on their members or, at least, on labour market “insiders” (Rueda, Reference Rueda2007)?
3) Does the adoption of poverty reduction strategies by provincial governments encourage an improvement in social assistance benefits? Or do these strategies remain mostly “window dressing” (Plante, Reference Plante2019)?
4) Is minimum income protection connected to welfare state generosity and redistribution (Noël, Reference Noël2019)? Or does it reflect instead the political importance of the program, as measured by the proportion of the population that is dependent on these benefits, following the logic identified by Lødemel (Reference Lødemel1997) as a welfare paradox?
To answer these questions, this article compares social assistance incomes for four household types in the 10 Canadian provinces between 1990 and 2017 and relates these incomes to a number of factors, including partisan dominance over time, trade union density, the presence or absence of poverty reduction strategies, provincial social expenditures, overall redistribution efforts and social assistance recipiency rates. Descriptive statistics, bivariate correlations and multivariate time-series cross-sectional models are then considered, to establish trends and relationships. The final results remain tentative because there are only 10 units and minimum income protection varies little across provinces and slowly in time, but they point in the following direction:
1) Parties of the left do not help and parties of the right do not hurt, but centrist, Liberal governments may have an effect over time, at least for the “most deserving” beneficiaries: persons with a disability or families with children. This effect, however, remains limited, in line with the international comparative evidence (Noël, Reference Noël2019).
2) Everything else being equal, provinces with a stronger labour movement, and presumably more organized social forces, tend to be more generous, as is the case across the OECD (Noël, Reference Noël2019). This effect is particularly evident for the “least deserving” single, able-to-work persons.
3) Poverty reduction strategies have inconsistent effects. In bivariate correlations, the relationship between the presence of a strategy and welfare income adequacy appears positive, but it becomes negative when we control for other factors. All in all, as the evidence on poverty outcomes suggests, these strategies do little for the income of social assistance recipients (Plante, Reference Plante2019).
4) A provincial government's propensity to redistribute, which is an indirect measure of welfare state generosity, tends to be favourable to social assistance beneficiaries, but so does the proportion of people claiming benefits. The idea that minimum income protection reflects welfare state development remains plausible, but so does the welfare paradox, whereby governments pay attention to social assistance incomes when they concern more people (Noël, Reference Noël2019; Lødemel, Reference Lødemel1997).
These results fit with the established representation of social assistance as a low-profile, incremental program that evolves over time with limited political input, except perhaps the diffuse support of what David Brady calls the “latent coalition for egalitarianism” (Reference Brady2009). Without ever producing a strong push, centrist, liberal parties favour benefit adequacy over time, whereas parties of the left and of the right hardly leave a mark; strong trade unions also tend to help. Poverty reduction strategies, however, make little difference. The overall generosity of the welfare state, as measured by redistribution efforts, and the importance of the population concerned by social assistance matter more for minimum income protection.
The article starts with a brief review of the literature on minimum income protection in the OECD and in Canada, followed by a theory section, and one on data and methodology. A fourth section presents descriptive statistics and trends, and a fifth considers bivariate and multivariate relationships. The final two sections wrap up and draw the implications.
Literature Review
In comparative studies of minimum income protection, Canada really is Ontario. To establish the adequacy of social assistance in Canada, the OECD takes the benefits offered by Ontario Works (Immervoll, Reference Immervoll2009: 13; Bachelet et al., Reference Bachelet, Rastrigina, Browne, Immervol, Neumann and Pacifico2018: 12). In the more reliable Social Assistance and Minimum Income Protection Interim Dataset (SaMip) developed by Kenneth Nelson, data for Canada are also drawn from Ontario (Nelson, Reference Nelson2008: 109; Reference Nelson2013). Provincial benefits, however, vary significantly. In 2017, for instance, a single person considered employable received $7,433 a year in Nova Scotia, compared to $11,379 in Newfoundland and Labrador. A person with a disability living in Calgary and not eligible for Alberta's special disability program obtained $10,225 a year, compared to $12,741 in Montreal. Income support for this person in Calgary stood at 50 per cent of the Market Basket Measure (MBM) low-income threshold; in Montreal, it amounted to 71 per cent of this threshold, which now constitutes Canada's official poverty line (Tweddle and Aldridge, Reference Tweddle and Aldridge2018: 54).
In Canada, the provinces and territories are responsible for social assistance, and they make their own decisions, with little influence from federal social transfers that are basically unconditional block grants. As Gerard Boychuk notes, the result is not a “single national system” but rather “13 distinct social assistance regimes” that evolve at their own pace, in different socio-political contexts (Reference Boychuk, Béland and Daigneault2015: 35).Footnote 1 Some authors expected, following the end in 1995 of the shared-cost arrangements associated with the Canada Assistance Plan (CAP), that benefit levels would decline and that diversity across provinces would increase (Banting and Myles, Reference Banting, Myles, Banting and Myles2013). Benefits did go down in real terms in the late 1990s, but the trend started before the end of cost-sharing, and it was later reversed. Provincial trajectories have their own determinants (Boychuk, Reference Boychuk, Béland and Daigneault2015).
These social assistance trajectories are likely to be associated with broader characteristics of welfare state institutions. Using various redistribution indicators, Rodney Haddow finds significant differences between provincial welfare states, with Quebec standing out as the most distinct. Inter-provincial differences, he argues, are on par with international ones among advanced welfare states, and they are likewise related to the usual politics of social policy, with left and centre party incumbency and trade union density as good predictors of redistribution efforts (Haddow, Reference Haddow2014, Reference Haddow2015).
We know from the comparative literature, however, that the politics of minimum income protection is not a perfect extension of that of the welfare state. In his seminal study of Britain and Norway, Lødemel (Reference Lødemel1997) concluded that the overall generosity of the welfare state could even work against minimum income protection by making social assistance more residual and less politically visible and relevant. This situation, he proposed, generated a welfare paradox, whereby an encompassing welfare state such as Norway ended up being less supportive of persons receiving social assistance than was a more restricted, liberal welfare state such as Britain. In the following years, scholars reassessed this argument with a larger set of cases, better data and more formal statistical tests. Nelson, in particular, built the best longitudinal and comparative dataset of social assistance incomes in OECD countries and put to test the power resources theory, which predicted, contrary to Lødemel, that parties of the left, strong trade unions and a developed welfare state would favour generous minimum incomes. Nelson's findings were not entirely conclusive, however. While his earlier results, for 18 countries at one point in time, appeared consistent with power resources theory (Reference Nelson2003: 125), more elaborate tests with time-series cross-sectional models proved less convincing (Reference Nelson2008: 114). The same was true for Natascha Van Mechelen, who did not find a strong connection between the welfare state and social assistance incomes (Reference Van Mechelen2009: 164). In recent work on the 1990–2010 period, I do find a significant relationship between welfare state and social assistance benefit adequacy, as well as a negative influence of public debt, with indirect effects from left parties and trade unions, in a pattern akin to that found by Brady for poverty reduction (Noël, Reference Noël2019: 243; Brady, Reference Brady2009). These results set out theoretical expectations for the study of Canadian provincial variations.
Theory
The natural starting point for a comparative investigation of the politics of social assistance adequacy is the standard power resources theory, which asserts that social policy is first and foremost a product of class conflicts, as expressed through partisan politics and collective action (Van Kersbergen and Vis, Reference Van Kersbergen and Vis2014: 48–49). In this perspective, the long-run presence in power of the left, the centre or the right should make a difference, as should the strength of trade unions. This is the logic used by Haddow to account for differences in redistribution across Canadian provinces (Reference Haddow2014). It is also the logic identified in more qualitative studies of Quebec's distinct social model within Canada (Noël, Reference Noël, Banting and Myles2013; Van den Berg et al., Reference Van den Berg, Plante, Raïq, Proulx and Faustmann2017; Arsenault, Reference Arsenault2018).
For minimum income protection, however, the power resources theory may not hold as well. Social assistance is a residual program that concerns a minority of persons, seldom heard in political debates, and often despised.Footnote 2 These benefits, observe Pierre-Marc Daigneault and Daniel Béland, have “a bad reputation among the public,” and they are rarely an object of electoral competition (Reference Daigneault, Béland, Béland and Daigneault2015: 1). Even social-democratic parties tend to see persons receiving social assistance as labour market “outsiders,” at a distance from their main working-class or middle-class electorate, who may have little interest in improving the lot of the poor (Rueda, Reference Rueda2007; Iversen and Soskice, Reference Iversen and Soskice2019: 3, 21). If the power resources theory works for social assistance, it would be less because the different actors relentlessly pursue their interest, as is assumed in the standard power resources model, than because parties of the left and trade unions are careful to craft orientations that can sustain a “latent coalition for egalitarianism” (Brady, Reference Brady2009: 103–4). In other words, ideological consistency matters as much, and perhaps more, than the interplay of interests.
In this perspective, we can hypothesize that the incumbency of parties of the left and of the centre should sustain generous welfare incomes, while the dominance of the right should depress them. Likewise, a strong trade union movement should support minimum income protection, even though such protection concerns primarily labour market “outsiders”:
H1a: The cumulative presence in power of leftist or centrist parties has a positive relationship with minimum income protection adequacy.
H1b: The cumulative presence in power of parties of the right has a negative relationship with minimum income protection adequacy.
H2: Trade union density has a positive relationship with minimum income protection adequacy.
If actors, interests and ideology play a role, so should policy ideas and institutions (Daigneault and Béland, Reference Daigneault, Béland, Béland and Daigneault2015). Consider, first, ideas. From the 1990s to the 2010s, social assistance was a recurrent object of reforms, driven by a preoccupation with incentives and activation (Cox, Reference Cox, Béland and Daigneault2015). The new orientations, captured by the notion of social investment, were often suspected of being oblivious and perhaps detrimental to the poor (Cantillon, Reference Cantillon2011). They could be implemented in various ways, however, and overall these new policies were not introduced at the expense of the poor (Noël, Reference Noël2018). Among Canadian provinces, it would be difficult to take a systematic measure of such a social investment turn, but there is another reform idea that is easy to track down: the adoption of poverty reduction strategies, in line with a growing concern for poverty in global policy debates (Plante, Reference Plante2019). This type of intervention allows us to consider a simple, almost natural, hypothesis about the power of ideas:
H3: The adoption of a poverty reduction strategy has a positive relationship with minimum income protection adequacy.
Finally, concerning institutions, there is the basic power resources expectation of a congruence between welfare state generosity and minimum income protection, in contrast to what could be called the welfare paradox hypothesis, where the adequacy of social assistance incomes evolves less with the welfare state as a whole than with the salience of the issue, indicated by the social assistance recipiency rate. We can measure welfare state generosity directly, by the level of social expenditures as a percentage of gross domestic product (GDP), or indirectly, through an estimate of a province's redistribution effort. If we keep in mind the welfare paradox argument about the salience of social assistance, we have the following three, somewhat rival, hypotheses:
H4a: The level of social expenditures as a percentage of GDP has a positive relationship with minimum income protection adequacy.
H4b: A province's overall redistribution effort has a positive relationship with minimum income protection adequacy.
H4c: The social assistance recipiency rate has a positive relationship with minimum income protection adequacy.
Relatedly, because it is a variable identified as significant in the comparative literature (Noël, Reference Noël2019), we can test whether the level of public debt acts as a constraint on adequacy.
H5: The level of public debt as a percentage of GDP has a negative relationship with minimum income protection adequacy.
As in any observational study, questions of endogeneity and reverse causation may arise. One could think, for instance, that minimum income protection adequacy actually boosts social assistance recipiency rates and social expenditures and makes the adoption of a poverty reduction strategy easier. There is no fireproof statistical test to settle such questions, and a full discussion of this possibility is beyond the scope of this paper, but some theoretical and empirical observations can be brought to bear on the matter. First, social assistance recipiency rates are notoriously tied to the employment rate and, possibly, to structural changes in the labour market (Kneebone and White, Reference Kneebone, White, Béland and Daigneault2015: 67–69). Changes in social assistance benefits are unlikely, for instance, to account for a drop in the recipiency rate as important as that observed in Canada between 1994 and 2008 (see Figure 3 below). Second, social assistance benefits account for less than 10 per cent of the country's social spending (9.5% in 2015; calculated from OECD, 2019). This relatively small envelope is unlikely to drive spending trends. Third, while the adoption of a poverty reduction strategy may influence welfare incomes, the reverse does not make much sense. Social assistance recipiency rates, social expenditures and poverty reduction strategies are unlikely to be endogenous.
Data and Methodology
Comparing the adequacy of minimum income protection across provinces is relatively straightforward because reliable data are provided by Maytree, a Toronto organization that took over this mission from the Caledon Institute of Social Policy in 2018, which itself undertook this task when the National Council of Welfare was abolished in 2012. In an annual publication entitled Welfare in Canada, Maytree collects provincial and territorial welfare incomes for four types of households: a single person considered employable; a single person with a disability; a single parent with one child age 2; and a couple with two children ages 10 and 15. Welfare incomes correspond to what is called “minimum income protection” in the comparative literature, and they include not only social assistance payments but also other transfers such as tax credits, child benefits or special-purpose payments.
In a country where incomes and costs of living vary significantly across provinces, raw annual amounts would not capture perfectly the adequacy of welfare incomes in a given province. To do so, at least four options are available: the first two are computed by Maytree; the third is proposed by Ronald Kneebone and Katherine White (Reference Kneebone, White, Béland and Daigneault2015); and the fourth is that adopted here.
Maytree provides two measures of adequacy: welfare income as a percentage of the Canada-wide Low-Income Measure (LIM; half the Canadian median income) threshold; and welfare income as a percentage of the Market Basket Measure (MBM; a Canadian measure based on the cost of basic necessities in a given region) threshold. The LIM option seems to make sense because it refers to a standard, international measure of poverty. To compare provinces, however, this option is deficient because it relies solely on the overall Canadian median income, and we know that provincial median incomes differ substantially (Noël, Reference Noël2017). The MBM, Canada's new official poverty measure, is more helpful in this respect. Its only drawback is that it is a purely Canadian measure, which is unhelpful in placing provincial incomes in a broad comparative perspective. A third option, proposed by Kneebone and White, is to use the “basic needs” index developed by Christopher Sarlo (Kneebone and White, Reference Kneebone, White, Béland and Daigneault2015: 77). This option is also solely Canadian, and even in this country, it does not appear very legitimate, as Sarlo's index considers only the needs “required to ensure physical survival,” an understanding of poverty widely considered too restrictive (Kneebone and White, Reference Kneebone, White, Béland and Daigneault2015: 77; CEPE, 2009). The fourth option, adopted here, consists in measuring the adequacy of a province's welfare incomes by their value as a percentage of the provincial median income, adjusted for household size. This option creates an adequacy measure identical to that used in international comparative work, and it is thus easily readable for non-Canadians. It also allows Canadians to situate provincial welfare incomes in a broader international context.
Concretely, this measure of adequacy is based on the provincial after-tax median income for persons not in an economic family (singles).Footnote 3 For households with more than one person, this median income is adjusted by multiplying it by the square root of the number of persons in the household.Footnote 4 Welfare income adequacy for a given household is then estimated as a percentage of provincial median income for this household.
In 2017, for instance, Alberta welfare incomes amounted to $8,027 for a single person and to $28,989 for a family of four. The Alberta median income for a single person stood at $36,500. Adequacy for a single-person household was thus 8,027/36,500, or 0.220. Adequacy for a family of four was based on the same median income multiplied by the square root of 4; that is 73,000, which gives 28,989/73,000, or 0.397 (for more details, see the online supplementary material).
One should note that the adequacy of social assistance is also affected by a host of administrative rules regarding a person's marital and health situation, assets or allowed earnings. These rules, however, are hard to capture with a single metric and cannot be considered in a comparative analysis.
Partisan incumbency is based on the party of the premier in a given year, as established in the Canada Guide (http://www.thecanadaguide.com/data/provincial-premiers/). Each year a party is in power, starting 10 years before the period under study, in 1980, an increment of 1 is added. For election years, the increment starts the following year. The left includes the New Democratic Party (NDP) and the Parti Québécois (PQ); the centre designates the Liberal party (except in British Columbia); and the right includes the Conservatives, the Saskatchewan Party, and the Social Credit and Liberals in British Columbia.
Trade union density combines data from two Statistics Canada series, one for 1997 to 2017 and one for earlier years.Footnote 5 Adjustments for the two series are made using Haddow's estimations (Reference Haddow2015: 43).
The existence of poverty reduction strategies is operationalized by a dummy variable, according to the list established by Charles Plante (Reference Plante2019).
Social expenditures (health plus social services) and debt service costs are drawn from a dataset compiled by Kneebone and Margarita Wilkins (Reference Kneebone and Wilkins2019). Expenditures are then calculated as a percentage of provincial GDP, using Statistics Canada's GDP at market, current prices.Footnote 6 Provincial redistribution efforts are estimated using the OECD “standard approach,” with Statistics Canada inequality data (Van Lancker et al., Reference Van Lancker, Ghysels and Cantillon2015): redistribution equals the Gini for market incomes minus the Gini for after-tax incomes, divided by the Gini for market incomes and multiplied by 100.Footnote 7
Social assistance recipiency rates represent the number of persons receiving social assistance in proportion of the total of persons eligible (that is, individuals between 0 and 64 years of age). The data used here are those compiled by David Deault Picard and me for the years 1990–2015, updated with the Maytree Social Assistance Summaries and recent Statistics Canada population estimates (Deault Picard and Noël, Reference Deault Picard and Noël2016; Maytree, 2019; Statistics Canada, Table 17-10-0005-01 [formerly CANSIM 051-0001]).
Descriptive statistics are presented in the online supplementary material (Table SM1). Because we have a small number of cases with modest variations and slowly changing independent and dependent variables, we seek to keep the statistical analysis simple, theoretically grounded and transparent, with an emphasis on descriptive statistics, bivariate correlations and multivariate models including a limited range of independent variables. Multivariate tests are conducted with time-series cross-sectional models for the 10 provinces, over the 1990–2017 period. The conventional fixed effect approach is not optimal in this case because it controls out variation across provinces, making it difficult to assess the impact of variables that change little over time, such as union density or debt service (Bartels, Reference Bartels and J2015; Bell and Jones, Reference Bell and Jones2015). Following a methodology used in other studies of welfare incomes, we opt instead for a random effect approach, explicitly modelling the between-case and within-case effects (Noël, Reference Noël2019; Parolin, Reference Parolin2019). Panel-corrected standard errors (PCSE) and fixed effects models are nevertheless tested as robustness checks.
Descriptive Statistics and Trends
Measuring minimum income protection adequacy in relation to the provincial median income allows us to compare the Canadian provinces to other advanced welfare states. Figure 1 ranks adequacy in the provinces and in a number of OECD countries for a single employable person in 2010, complementing Canadian data with benefits data from the Social Assistance and Minimum Income Protection Interim Dataset (SaMip), available in the Social Policy Indicators (SPIN) database (http://www.sofi.su.se/spin/), adjusted with OECD data for the equivalized median disposable income (OECD, 2016).
Figure 1 Minimum Income Protection Adequacy, Employable Single Person, OECD Countries and Canadian Provinces, 2010
This comparison between OECD countries and Canadian provinces remains imperfect because it relies on different sources and measurements, but it nevertheless helps us locate Canadian welfare incomes in a global context. With the exception of Newfoundland and Labrador, Canadian provinces all stand at the bottom, near the level of social protection offered in Spain. Wealthy Alberta is by far the least generous province. Only the United States, where there is hardly any income protection for single employable persons, ranks behind Canadian provinces.Footnote 8
Over time, as Figure 2 indicates, the mean adequacy of welfare incomes did not vary dramatically, and it displayed a constant hierarchy between more or less “deserving” households. Incomes nevertheless fluctuated over the years. As Boychuk points out, the decline started in the beginning of the 1990s (Boychuk, Reference Boychuk, Béland and Daigneault2015: 45). It was particularly pronounced for single employable persons and was possibly a reaction to the steep rise in the average social assistance recipiency rate, which went from 8 per cent in 1990 to a peak of 11 per cent in 1994, as Figure 3 shows. With the twenty-first century, social assistance recipiency rates declined steadily and then levelled off, while adequacy slowly improved, in the 2010s in particular. The composition of the population concerned evolved as well, with a growing proportion of persons with a disability (Kneebone and White, Reference Kneebone, White, Béland and Daigneault2015: 75).
Figure 2 Mean Adequacy of Minimum Income Protection, Four Household Types, Canadian Provinces, 1990–2017
Figure 3 Mean Social Assistance Recipiency Rate, Canadian Provinces, 1990−2017
Averages mask relatively similar provincial trajectories. As can be seen in Figure 4, single employable persons are always the least favoured, and overall one finds again the average pattern of retrenchment in the beginning of the 1990s, stability thereafter and progress in the 2010s. Newfoundland and Labrador stand out, however, with a major upward shift in the beginning of the century. There was also a marked improvement, from a relatively low point, in New Brunswick. Elsewhere, movements remained limited. Social assistance incomes were obviously not an object of much political attention.
Figure 4 Adequacy of Minimum Income Protection, Four Household Types, Canadian Provinces, 1990−2017
Can we find, nevertheless, social and political correlates of minimum income protection? To this question, we can now turn.
Bivariate and Multivariate Relationships
The previous sections suggest that a number of factors contribute to define the level of minimum income protection—namely, partisan incumbency, trade union density, the adoption of a poverty reduction strategy, welfare state generosity, redistribution effort, the social assistance recipiency rate and the importance of the public debt. For the sake of simplicity, and because these cases capture the bottom line of minimum income protection, unrelated to attitudes and policies about disability or children, the following discussion focuses on the welfare incomes of single employable persons. Table 1 reports correlations for the 10 cases over 28 years (N = 280).
Table 1 Correlations between the Adequacy of Minimum Income Protection for a Single Employable Person and Political and Institutional Variables, Canadian Provinces, 1990–2017
***p < 0.01, **p < 0.05, *p < 0.1
Simple correlations allow us to assess the relationship between the different explanatory factors and adequacy before turning to multivariate models. Most relationships are significant in the expected direction, except for public debt. First come the power resources variables: partisan cumulative power and trade union density. The left and the right have weak, hardly significant relationships, but the presence of the Liberal party in power is associated positively to welfare incomes. The relationship for trade union density is also consistent with our theory and with the evidence for OECD countries (Noël, Reference Noël2019). Ideas, operationalized by the presence of a poverty reduction strategy, and institutions—redistribution and the social assistance recipiency rate—also have the expected relationships. The positive effect of the social assistance recipiency rate is particularly interesting because, for lack of comparable data on recipiency rates, the same hypothesis cannot be tested for OECD welfare states (Noël, Reference Noël2019).
Surprisingly, the social expenditures budget has only a modest relationship with minimum income protection, and the public debt is not correlated significantly with social assistance—both results contrasting with what is found for the OECD (Noël, Reference Noël2019). One possible explanation may be that provincial social expenditures, being driven by health care spending, reflect less strongly a government's commitment to social justice than does the redistribution index. As for the counter-intuitive result for debt service, one must take into account the unique position of provincial governments as public borrowers.
Canadian provinces, explains Kyle Hanniman, are free to borrow on the domestic and international bond markets, and they do so significantly, given their share of the country's public expenditures (Reference Hanniman, Goodyear-Grant, Johnston, Kymlicka and Myles2018). As such, they are exposed to market pressures. At the same time, their very weight in terms of spending makes it obvious that the federal government would not let any of them default. Provincial governments are thus sheltered, to some extent, from market constraints. Unlike American states, they borrowed abundantly in recent years to pursue counter-cyclical policies and sustain social spending (Hanniman, Reference Hanniman, Goodyear-Grant, Johnston, Kymlicka and Myles2018). The size of a province's debt service as a percentage of GDP may constitute less a constraint than an indicator of a provincial government's willingness to spend.
If we leave aside variables that have weak or nonsignificant bivariate relationships with minimum income protection, we can identify the following equation for a time-series cross-sectional analysis:
Since many of the variables identified are slow-changing institutional factors that vary more across than within provinces, a conventional fixed effect model would prove insufficient, as it would obliterate variation between units and only identify relationships within provinces, over time. Results for such a model are reported in the online supplementary material (Table SM2), and they suggest that, over time, centrist cumulative power, social assistance recipiency rates and poverty reduction strategies have a positive relationship with adequacy.
To go beyond these findings and consider differences among provinces, as well, we must adopt a random effect regression model that estimates specifically the effects of variables between and within cases. This model is tested for our four household types, with two different approaches: one using a regression with a lagged dependent variable, to take into account the path-dependent nature of welfare incomes (past values influence current values; Bartels, Reference Bartels and J2015), the other with a panel-corrected standard errors (PCSE) procedure with first-order autocorrelation (AR1) (as in Haddow, Reference Haddow2016). The two approaches yield similar findings. Results for the first approach are displayed in Table 2; those for the second are presented in Table SM3 (online supplementary material).
Table 2 Random Effect Model Separating Between- and Within-Province Effects for the Determinants of Minimum Income Protection Adequacy, Four Household Types, Canadian Provinces, 1990–2017
***p < 0.01, **p < 0.05, *p < 0.1. Standard errors in parentheses.
Between effects capture differences among provinces, and within effects track change over time. Consider, first, differences among provinces. Most of the relationships we have identified as significant in bivariate correlations remain significant, in the expected direction. Provinces that have a strong trade union movement, a propensity to redistribute and a high social assistance recipiency rate tend to provide more generous welfare incomes. The recipiency rate is significant for all households except the four-person family; the redistribution effect matters for families with children, and the union effect is significant for single employable and single parents but not for the other two households.
As expected from the fixed effect model, the presence of the Liberal party in power and the adoption of a poverty reduction strategy have a positive relationship with adequacy over time (within). Intriguingly, the presence of a poverty reduction strategy yields a negative relationship between provinces. This is contrary to our expectations but possibly consistent with Plante's findings that poverty reduction strategies follow, rather than precede, poverty reduction efforts (Reference Plante2019). More fundamentally, these results are consistent with the view that provincial governments are keener to bring persons out of poverty, children in particular, than to alleviate the fate of those who are deep in poverty, such as single persons relying on social assistance, except perhaps by helping them integrate the workforce (Larocque, Reference Larocque, Pétry and Birch2018). Improving welfare incomes would alleviate poverty, but it would not reduce the poverty rate.
To weigh the relative importance of each factor, Table 3 displays the same model with standardized coefficients.
Table 3 Standardized Random Effect Model Separating Between- and Within-Province Effects for the Determinants of Minimum Income Protection Adequacy, Four Household Types, Canadian Provinces, 1990–2017
***p < 0.01, **p < 0.05, *p < 0.1. Standard errors in parentheses.
These results indicate that for single employable persons, the relationships for union density and for the social assistance recipiency rate are about equivalent. For persons with a disability, the main relationship is by far the recipiency rate. For single parents, the key is also the social assistance recipiency rate, along with provincial redistribution efforts and union density. With four-person families, the most important variable is the redistribution effort. In all four cases, poverty reduction strategies maintain a significant negative relationship with adequacy.
Discussion
Our first hypothesis (H1a) was that parties of the left and of the centre were associated with the adequacy of welfare incomes. This hypothesis is only partly supported. First, parties of the left have no significant effect. Consider, for instance, Manitoba. The NDP was in power for four terms between 1999 and 2016 but nevertheless acted conservatively on social assistance, keeping benefits lower than in most other provinces (Simpson, Reference Simpson, Béland and Daigneault2015: 206; see Figure 1 above). In British Columbia in the 1990s, an NDP government also implemented market-oriented, “get tough” social assistance reforms (Pulkingham, Reference Pulkingham, Béland and Daigneault2015: 144) and so did the Parti Québécois at different occasions (Noël, Reference Noël, Béland and Daigneault2015: 137). Surprisingly, Liberal governments had more of an impact, at least over time (within cases). In Ontario, for instance, adequacy for a single employable was at 24 per cent of median income in 2003 when Dalton McGuinty was elected, and it had risen to 31 per cent by 2017, after 14 years of Liberal government (Graefe, Reference Graefe, Béland and Daigneault2015: 115). Overall, conservative parties also proved more prudent than tough, as suggest our nonsignificant results for the second part of our first hypothesis, about the right (H1b).
Findings for trade union density are consistent with the second hypothesis (H2) and suggest that social actors matter more than political parties for welfare incomes. Trade union density may indeed be an indirect measure of collective action capacity in a province. Case studies point to the importance of active community and social rights organizations in relatively generous provinces such as Newfoundland and Labrador, Saskatchewan, and Quebec (Mondou, Reference Mondou, Béland and Daigneault2015: 241–42; August, Reference August, Béland and Daigneault2015: 187–88; Noël, Reference Noël, Béland and Daigneault2015: 136–37).
Negative results for poverty reduction strategies seem more problematic (H3). They may suggest, as mentioned above, that governments care more about the rate than about the depth of poverty. One should keep in mind, though, that these strategies are still relatively recent. Some of them also remain meagre, relatively shallow documents, designed mostly as public recognitions that poverty remains a problem.
Contrary to the comparative literature, social expenditures do not correlate strongly with minimum income protection (H4a), probably because spending does not vary all that much between provinces. Redistribution efforts, however, seem to matter, at least for families with children (H4b). Even more important is the salience of the issue. When the social assistance recipiency rate is high, income adequacy is also improved (H4c). Public debt, as mentioned above, does not appear to be a constraint (H5).
Overall, these findings make sense, with a weak partisan effect, a strong relationship for union density and the social assistance recipiency rate, a positive relationship for redistribution, and the unhelpful consequences of poverty reduction strategies. One should keep in mind that the adequacy of social assistance varies little across provinces and over time, as do many of our independent variables. Partisan incumbency variables and trade union density also constitute rough indicators of a province's politics, which leave aside a host of potentially relevant actors, institutions and ideas.
Consider, for instance, the possibility that welfare incomes embody a sense of solidarity that is more or less pronounced across provinces. There are good theoretical reasons to consider such a relationship, and the link between subnational identity and social solidarity is documented for different cases (Béland and Lecours, Reference Béland and Lecours2008; Singh, Reference Singh2015). To assess summarily this possibility, we can use a survey conducted by the Environics Institute between December 2018 and January 2019, which presents the proportion of Canadians who consider their region or province “very important or somewhat important to their sense of identity” (Environics Institute, 2019: 22). The scores range from 71 per cent in Ontario to 89 per cent in Newfoundland and Labrador. The relationship between these scores for 2018–2019 and adequacy for a single employable person in 2017 is positive and quite strong (R = 0.68), as is the relationship between identity and broader redistribution efforts (R = 0.71).Footnote 9Figure 5 locates the provinces according to the strength of their identity and welfare incomes adequacy. Newfoundland and Labrador stand out in the upper left corner, as does Ontario at the opposite end. If we take out these two outliers, the relationship between identity and adequacy remains positive (R = 0.58).
Figure 5 Relationship between Provincial Identity in 2018−2019 and the Adequacy of Minimum Income Protection for Single Employable Persons in 2017
Unfortunately, we do not have yearly provincial identity data that would allow a full integration of this variable in the model. This perspective on subnational identity, for one time point only, is nevertheless indicative of the need to know more, with additional concepts and indicators, about the politics of minimum income protection. In a recent article, for instance, Zachary Parolin (Reference Parolin2019) demonstrates how the presence of black families within an American state influences the distribution of Temporary Assistance to Needy Families funds, racial bias being stronger where black families are more numerous.
Conclusion
Comparative and Canadian research on the welfare state has focused largely on programs aimed at workers or the middle class, leaving aside last-recourse measures designed for the poorest. This article seeks to fill this gap with a comparative analysis of the determinants of provincial welfare incomes and finds that a number of factors contribute to define minimum income protection in Canada. While partisan incumbency has little effect, trade union density matters, as do an overall commitment to redistribution and a sense of provincial identity. The salience of welfare incomes, tied to higher social assistance recipiency rates, also contributes to boost welfare incomes. Poverty reduction strategies, however, do not seem to help. All in all, the balance of social forces (as indicated by trade union density, redistribution and a sense of provincial identity) and the saliency of the issue weight more than partisan politics.
The politics of minimum income protection in Canada can be summarized in seven points. First, and this is not a surprise, with the exception of Newfoundland and Labrador, Canadian provinces rank at the bottom compared to other advanced welfare states, at least for single persons deemed employable. Only the United States, where there is hardly any minimum income protection, remains behind Canadian provinces. On average, in 2017, single employable Canadians on social assistance received 30 per cent of the provincial median income, well below the standard poverty line of 50 per cent. Still, this meagre level of income security constituted progress, minimum income protection having increased throughout the 2000s and 2010s.
Second, in Canada as elsewhere in the OECD, social assistance adequacy hardly constitutes a partisan issue dividing the left and the right. Social-democratic and conservative parties have no significant impact on benefit levels, while centrist, Liberal parties have a modest one over time, within provinces. Elections are not won or lost over minimum income protection.
Third, contrary to the view that trade unions benefit exclusively their members—or labour market “insiders”—we find a significant union effect on minimum income protection, for the “least deserving” single employable persons in particular. This result is consistent with power resources theory and with a number of findings on union density and redistribution (Bradley et al., Reference Bradley, Huber, Moller, Nielsen and Stephens2003). Trade unions also help “outsiders,” possibly by reinforcing their members’ preferences for equality or by consolidating the possibilities of collective action in a province (Mosimann and Pontusson, Reference Mosimann and Pontusson2017; Brady et al., Reference Brady, Baker and Finnigan2013).
Fourth, poverty reduction strategies do not enhance the revenues of households receiving social assistance. The effect of these strategies is, in fact, negative. This sobering conclusion is consistent with Plante's (Reference Plante2019) conclusions about the lack of incidence of these strategies on the poverty rate. They also underline the fact that improving minimum income protection rarely was the road to poverty reduction favoured by governments, who insisted instead on labour market integration. In Quebec, for instance, which was the first province to adopt a poverty reduction strategy, little was done to improve social assistance benefits for single employable persons (Noël, Reference Noël, Banting and Myles2013: 270). Newfoundland and Labrador, the second province with a strategy, was perhaps the only one where improvements in social assistance incomes were explicitly part of the plan (Hudson and Close, Reference Hudson and Close2011: 85).
Fifth, the welfare state matters, not so much in terms of social expenditures but through the influence of a province's overall redistribution effort. This mechanism is consistent with power resources theory, and it accounts, in particular, for the adequacy of minimum income protection for households with children.
Sixth, in line with Lødemel's welfare paradox, the political salience of social assistance, measured by the social assistance recipiency rate, contributes to enhance welfare incomes—and for single adults, in particular (employable, with a disability, or with a child). Debt service costs, on the other hand, do not seem to act as a constraint for provincial governments.
Seventh, and more tentatively, a strong sense of provincial identity may have a positive impact on the adequacy of welfare incomes.
Among advanced welfare states, Canadian provinces stand near the bottom for minimum income protection. Provinces nevertheless differ, in ways that are basically consistent with the comparative literature. Without ever being at the centre of partisan debates, and hardly an object of poverty reduction strategies, welfare incomes benefit from trade union strength and from provincial redistribution efforts. They also appear more generous in provinces where more households rely on social assistance, a result that could not be established in the comparative literature, for lack of comparable data on social assistance recipiency rate. The welfare paradox thus meets power resources theory. Provincial identity may sustain, as well, redistribution, for all and for the poorest. In Canada, as elsewhere, politics matter for minimum income protection.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0008423920000098
Acknowledgments
An earlier version of this article was presented at the annual conference of the Canadian Political Science Association, University of British Columbia, Vancouver, June 5, 2019. I am grateful to Ronald Kneebone, who gracefully shared his provincial budgetary data, and to Keith Banting, Daniel Béland, Pierre-Marc Daigneault, Carmen Jacqueline Ho, Olivier Jacques, Charles Plante, Phillippe Scrimger, Axel van den Berg, and this journal's reviewers and assistant editor, Jack Lucas, for comments and suggestions. I owe a special thanks, as well, to David Deault Picard, who did the early work on the data and with whom I started working on welfare incomes in Canada. I also wish to acknowledge the financial support of the Social Sciences and Humanities Research Council of Canada.
