Testing Provincial Welfare Generosity and Leftist Politics As Macro-Social Determinants of Population Health

TITLE PAGE

The Politics of Population Health in Canada: Testing Provincial Welfare Generosity and Leftist Politics as Macro-social Determinants of Population Health

Edwin Yee-Hong Ng

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy

Graduate Department of Public Health Sciences Dalla Lana School of Public Health University of Toronto

The Politics of Population Health in Canada: Testing Provincial Welfare Generosity and Leftist Politics as Macro-social Determinants of Population Health

Edwin Yee-Hong Ng

Doctor of Philosophy

Graduate Department of Public Health Sciences Dalla Lana School of Public Health University of Toronto

2013

ABSTRACT

This dissertation pools time-series and cross-section data among Canadian provinces to examine: (1) whether provincial welfare generosity (health, social services, and education expenditures), power resources and political parties (unions and left, centre, and right political parties), and political democracy (voter turnout and women in government) affects population health; (2) whether the effect of leftist politics channels through or combines with provincial welfare generosity to affect population health; and

(3) whether provinces cluster into distinct political regimes which are predictive of population health.

Data is retrieved from the Canadian Socio-Economic Information Management

System (CANSIM) II Tables from 1976 to 2008 and Canadian Parliamentary Guides from various years. Population health is measured using total, male, and female age- standardized mortality rates. Estimation techniques include Prais-Winsten regressions with panel-corrected standard errors (PCSE), a first-order autocorrelation correction model (AR1), and fixed unit effects. Hierarchical cluster analysis is used to identify how

ii provinces cluster into distinct regimes.

Primary findings are three-fold. First, provincial welfare generosity has a significant impact in lowering mortality rates, net of other factors, such as demographic and economic variables. Second, the political power of left and centre political parties and women in government have significant negative effects on mortality rates. Whereas left political parties and women in government combine with provincial welfare generosity to improve population health, the effect of centre political parties is channeled through provincial expenditures. Third, provinces cluster into three distinct regimes based on left political party power and women in government: 1) leftist

(Saskatchewan, British Columbia, and Manitoba); 2) centre-left (Ontario and Quebec); and 3) conservative (Alberta, Nova Scotia, New Brunswick, Prince Edward Island, and

Newfoundland). Compared to the leftist regime, centre-left and conservative provinces have significantly higher mortality rates; however, provincial welfare generosity accounts for most of the observed inter-provincial differences in population health.

iii ACKNOWLEDGEMENTS

This dissertation represents the support, guidance, and instruction of many important people. I would like to express my sincere appreciation to past instructors, supervisors, and mentors: Drs. Michael Bagby, Gerald Booth, Janice DuMont, Jim Dunn, Wilfred

Gallant, Kevin Gorey, Michael Holosko, Peggy McDonough, Blake Poland, and Ann

Robertson. Invaluable assistance was provided by former and current staff and faculty within the Dalla Lana School of Public Health: Karin Domnick, Bessie Gorospe, Joan

Kwan, Sandra Lang, Ted Myers, Andrea Sass-Kortsak, Ellen Sokoloff, Diane Tang,

Mariana Vardaei, and Rachel Zulla. Special thanks to the helpful staff at the School of

Graduate Studies: Lisa Fannin, Jong Mi Sze, and Pat Singh, as well as to fellow Social

Science and Health classmates: Dana Howse, Krista Maxwell, and Sandra Moll. I am appreciative for the doctoral supervision provided by Drs. Joan Benach and Haejoo

Chung, and for the collegial support from Drs. Carme Borrell and Patricia O’Campo.

Special thanks to members of my final oral examination committee: Drs. Paul Hamel

(Chair), Andrea Madarasová Gecková (External Examiner), Jim Dunn (Voting Member), and Arjumand Siddiqi (Voting Member).

I am deeply grateful for the support and encouragement of family members: Lai-

Yin and Fai-Sheung Ng (parents), Mabel and Ken Tung (parent-in-laws), Anthony Tung

(brother-in-law), Howie Tran (future brother-in-law), and Julia Chan (super aunt). Last but not least, I am most grateful to my thesis supervisor, sister, and wife. Special thanks to Dr. Carles Muntaner for serving as my thesis supervisor, giving me the chance to contribute to his research program, and having full confidence in my academic abilities.

Second, extraordinary thanks to Karina Ng for being my ever-supportive sister,

iv encouraging me in all my endeavors, and raising my spirits whenever I needed a confidence boost. Third, I owe a very important debt to my wife, Jenn Tung, who has supported me and my studies in countless ways and who always believed that I could earn my doctorate. Without the support and encouragement of Carles, Karina, and Jenn, this dissertation would not have been possible. For this, this work is dedicated to you all;

I am and will forever be in your debt.

v TABLE OF CONTENTS

TITLE PAGE ...... i

ABSTRACT ...... ii

ACKNOWLEDGEMENTS ...... iv

TABLE OF CONTENTS ...... vi

LIST OF TABLES ...... viii

LIST OF FIGURES...... xi

LIST OF APPENDICES ...... i

CHAPTER ONE. INTRODUCTION ...... 1 Theoretical Location ...... 9 Methodological Approach ...... 12 What this Dissertation Adds ...... 14 Outline of Chapters ...... 15

CHAPTER TWO. LITERATURE REVIEW ...... 17 Social Epidemiological Approaches ...... 17 Political Sociological Theories ...... 25 Empirics: Political Determinants of Population Health ...... 39

CHAPTER THREE. CONCEPTUAL MODEL ...... 75 Politics of Population Health in Canada ...... 75 Conceptual Model: Welfare Generosity, Leftist Politics, and Population Health ...... 79

CHAPTER FOUR. METHODS ...... 90 Data Sources ...... 90 Measures of Dependent, Independent, and Control Variables ...... 91 Estimation and Post-Estimation Techniques ...... 115

CHAPTER FIVE. PROVINCIAL WELFARE GENEROSITY AND POPULATION HEALTH ...... 125 Introduction ...... 125 Data and Methods ...... 129 Results ...... 133 Discussion...... 141

CHAPTER SIX. LEFTIST POLITICS, PROVINCIAL WELFARE GENEROSITY, AND POPULATION HEALTH ...... 169 Introduction ...... 169

vi Data and Methods ...... 177 Results ...... 181 Discussion...... 187

CHAPTER SEVEN. HIERARCHICAL CLUSTER OF LEFTIST POLITICS AND POPULATION HEALTH: A TAXONOMY OF PROVINCIAL REGIMES ...... 211 Introduction ...... 211 Data and Methods ...... 217 Results ...... 220 Discussion...... 227

CHAPTER EIGHT. DISCUSSION ...... 244 Revision of Conceptual Model and Summary of Findings ...... 244 Dissertation Limitations ...... 249 Suggestions for Future Research ...... 254

CHAPTER NINE. CONCLUSION ...... 266 Welfare State Policies versus Macro-Level Politics ...... 266 Political Strategies ...... 272 Policy Proposals ...... 273

REFERENCES ...... 276

vii LIST OF TABLES

Table 2.1. Conceptualizing Politics as a Macro-social Determinant of Population Health: Variables, Claims, and Hypotheses ...... 60

Table 2.2. Summary Characteristics of Empirical Studies on Welfare Generosity, Welfare Regimes, and Leftist Politics (N = 46) ...... 62

Table 2.3. Empirical Studies on Welfare Generosity and Population Health (n = 29) .... 63

Table 2.4. Empirical Studies on Welfare Regimes and Population Health (n = 12) ...... 69

Table 2.5. Empirical Studies on Leftist Politics and Population Health (n = 14) ...... 71

Table 4.1. Im, Pesaran, and Shin (2003) Unit Root Tests for Dependent Variables ... 124

Table 5.1. Summary Statistics, 1989-2008 ...... 148

Table 5.2. Total Age-Standardized Mortality Rates, Aggregate Expenditures, and Provincial Welfare Generosity Index by Provinces, 1989-2008a ...... 150

Table 5.3. Pairwise Correlation Matrix for Main Variables in Analyses ...... 151

Table 5.4a. PW-PCSE Models of Aggregate Expenditures on Total Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 153

Table 5.4b. PW-PCSE Models of Aggregate Expenditures on Male Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 154

Table 5.4c. PW-PCSE Models of Aggregate Expenditures on Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 155

Table 5.5a. PW-PCSE Models of Disaggregated Health Expenditures on Total Age- Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 156

Table 5.5b. PW-PCSE Models of Disaggregated Health Expenditures on Male Age- Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 157

Table 5.5c. PW-PCSE Models of Disaggregated Health Expenditures on Female Age- Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 158

Table 5.6a. PW-PCSE Models of Disaggregated Social Services Expenditure on Total Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 159

Table 5.6b. PW-PCSE Models of Disaggregated Social Services Expenditure on Male Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 160

viii Table 5.6c. PW-PCSE Models of Disaggregated Social Services Expenditure on Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 161

Table 5.7a. PW-PCSE Models of Disaggregated Education Expenditures on Total Age- Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 162

Table 5.7b. PW-PCSE Models of Disaggregated Education Expenditures on Male Age- Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 163

Table 5.7c. PW-PCSE Models of Disaggregated Education Expenditures on Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 ...... 164

Table 5.8. PW-PCSE Models of Provincial Welfare Generosity Index on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 ... 165

Table 6.1. Summary Statistics: 1976-2008 ...... 195

Table 6.2. Power Resources, Political Parties, and Democratic Politics by Province, 1976-2008 ...... 196

Table 6.3. Pairwise Correlation Matrix for Main Variables in Analyses ...... 197

Table 6.4a. PW-PCSE Models of Power Resources and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ...... 199

Table 6.4b. PW-PCSE Models of Power Resources and Provincial Welfare Generosity on Male Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ...... 200

Table 6.4c. PW-PCSE Models of Power Resources and Provincial Welfare Generosity on Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ..... 201

Table 6.5a. PW-PCSE of Models Centre and Right Political Parties and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ...... 202

Table 6.5b. PW-PCSE Models of Centre and Right Political Parties and Provincial Welfare Generosity Male Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ...... 203

Table 6.5c. PW-PCSE Models of Centre and Right Political Parties and Provincial Welfare Generosity Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ...... 204

Table 6.6a. PW-PCSE Models of Voter Turnout, Women in Government and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ...... 205

ix Table 6.6b. PW-PCSE Models of Voter Turnout, Women in Government and Provincial Welfare Generosity on Male Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ...... 206

Table 6.6c. PW-PCSE Models of Voter Turnout, Women in Government and Provincial Welfare Generosity on Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 ...... 207

Figure 6.1. Standardized Effects of Political Variables and Provincial Welfare Generosity Index for Total Age-Standardized Mortality Rates ...... 208

Table 7.1. Summary Statistics of and Correlations between Leftist Politics and Total, Male, and Female Mortality Rates ...... 230

Table 7.2. Hierarchical Cluster Analysis Proximity Matrix across Canadian Provinces (squared Euclidian distance) ...... 231

Table 7.3a. Summary of One-way Analysis of Variance: Provincial Regimes and Total Age-Standardized Mortality Rates ...... 232

Table 7.3b. Summary of One-way Analysis of Variance: Provincial Regimes and Male Age-Standardized Mortality Rates ...... 232

Table 7.3c. Summary of One-way Analysis of Variance: Provincial Regimes and Female Age-Standardized Mortality Rates ...... 232

Table 7.4a. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008...... 233

Table 7.4b. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Male Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008...... 235

Table 7.4c. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008. ... 237

x LIST OF FIGURES

Figure 3.1. Total Age-Adjusted Mortality Rates (all causes of death per 1,000 population) by Province, 1975 and 2009 ...... 86

Figure 3.2. Male Age-Adjusted Mortality Rates (all causes of death per 1,000 population) by Province, 1975 and 2009 ...... 87

Figure 3.3. Female Age-Adjusted Mortality Rates (all causes of death per 1,000 population) by Province, 1975 and 2009 ...... 88

Figure 3.4. Conceptual Model: Combining Power Resources, Political Parties, Democratic Politics, Provincial Welfare Generosity, and Population Health ...... 89

Figure 5.1. Standardized Effects of Aggregate and Disaggregated Provincial Welfare Generosity Expenditures for Total Age-Standardized Mortality Rates ...... 166

Figure 5.2. Standardized Effects of Aggregate and Disaggregated Provincial Welfare Generosity Expenditures for Male Age-Standardized Mortality Rates ...... 167

Figure 5.3. Standardized Effects of Aggregate and Disaggregated Provincial Welfare Generosity Expenditures for Female Age-Standardized Mortality Rates...... 168

Figure 6.1. Standardized Effects of Political Variables and Provincial Welfare Generosity Index for Total Age-Standardized Mortality Rates ...... 208

Figure 6.2. Standardized Effects of Political Variables and Provincial Welfare Generosity Index for Male Age-Standardized Mortality Rates ...... 209

Figure 6.3. Standardized Effects of Political Variables and Provincial Welfare Generosity Index for Female Age-Standardized Mortality Rates ...... 210

Figure 7.1. Dendrogram - Hierarchical Cluster Analysis of Left Political Party Power and Women in Government among Canadian Provinces...... 239

Figure 7.2. Box Plots of Total, Male, and Female Age-Standardized Mortality Rates across Provincial Regimes ...... 240

Figure 7.3. Semi-Standardized and Standardized Effects of Provincial Regimes and Provincial Welfare Generosity Expenditures for Total Age-Standardized Mortality Rates ...... 241

Figure 7.4. Semi-Standardized and Standardized Effects of Provincial Regimes and Provincial Welfare Generosity Expenditures for Male Age-Standardized Mortality Rates ...... 242

xi Figure 7.5. Semi-Standardized and Standardized Effects of Provincial Regimes and Provincial Welfare Generosity Expenditures for Female Age-Standardized Mortality Rates ...... 243

Figure 8.1. Revised Conceptual Model: Power Constellations, Politics of Presence, Provincial Welfare Generosity, and Age-Standardized Mortality Rates ...... 265

xii LIST OF APPENDICES

Appendix A. Data Sources ...... 326

Appendix B. Jackknife Results ...... 342

CHAPTER ONE. INTRODUCTION

Over the past two decades, the idea that politics affects population health and health inequalities has reached a zenith (Bambra, Fox, & Scott-Samuel, 2005, 2007; Franco,

Álvarez-Dardet, & Ruiz, 2004; Gil-González, Ruiz-Cantero, & Álvarez-Dardet, 2009;

Mackenbach, 2009, 2013; Muntaner et al., 2011, 2012; Pega, Kawachi, Rasanathan, &

Lundberg, 2013). Inspired by the adages of Virchow (“Medicine is a social science and politics is nothing but medicine at a larger scale”) (Mackenbach, 2009) and Rose

(“Medicine and politics cannot and should not be kept apart”) (Rose, Khaw, & Marmot,

2008), emergent research has examined the impact of various political forces, processes, and institutions on population health and health inequalities (Beckfield &

Krieger, 2009; Brennenstuhl, Quesnel-Vallée, & McDonough, 2012; Muntaner et al.,

2011, 2012). As a point of reference, I searched PubMed from 1970 to 2012 using the keywords “politic*” and “health” to detect any publication trends. Search results produced 15,467 articles and revealed two basic trends. First, the vast majority of this work has been recent – 83.4% of studies have been published between 1992 and 2012.

Second, the number of articles published per year has steadily increased from a low of

12 in 1970 to 360 in 1992 to a high of 1052 in 2012. This volume of scholarship is clear evidence of the (re-)emerging awareness within public health of the political dimensions of health (Bambra et al., 2005, 2007; Doyal, 1979; Eikemo & Bambra, 2008; Muntaner &

Lynch, 1999; Muntaner et al., 2002, 2011, 2012; Navarro & Shi, 2001).

1 As a general statement, this growing work finds that politics, defined broadly as the art of government, the activities of the state, and the process through which desired outcomes are achieved in the production, distribution, and use of valuable resources

(Bambra et al., 2005, 2007; Hague & Harrop, 2010), affects population health through unequal power relationships (Tesh, 1988), different government ideologies (Bryant,

2006, 2009; Coburn, 2004, 2006; Navarro, 2000, 2007; Navarro et al., 2003, 2006) and various social institutions (Bambra et al., 2005; Navarro & Muntaner, 2004). Existing studies cluster around three research streams: (1) welfare generosity; (2) welfare regimes; and (3) leftist politics (Beckfield & Krieger, 2009; Brennenstuhl et al., 2012;

Muntaner et al., 2011, 2012).

The first stream, welfare generosity, uses information on government expenditures to measure welfare state effort (Wilensky, 1974, 1976, 2002; Wilensky &

Lebaux, 1965). This approach conceptualizes welfare effort as an independent variable using quantitative measures such as social welfare expenditures to explain population health or health inequalities (Bradley, Elkins, Herrin, & Elbel, 2011; Conley & Springer,

2001; Dahl & van der Wel, 2013; Fritzell & Lundberg, 2007; Lundberg, Fritzell, Yngwe,

& Kölegård, 2010; Lundberg et al., 2008; Kangas, 2010). The key argument is that welfare generosity are akin to health-promoting resources, or social determinants of health (SDOH), that ensure acceptable standards of living, compensate for market failures, and minimize negative exposures to health (Fitzell & Lundberg, 2007; Lundberg et al., 2008). Generous provisions of welfare services and social transfers are hypothesized to result in better health outcomes (Dahl & van der Wel, 2013). To date, welfare generosity studies find that higher levels of public spending in wealthy countries

2 are associated with improved health outcomes (e.g., higher life expectancy and lower infant mortality rates; Kangas, 2010; Lundberg et al., 2008) and narrower health inequalities (e.g., educational inequalities in health; Dahl & van der Wel, 2013).

According to Kangas (2010), “Bigger seems to be better” (p. s57), as welfare states devote a larger share of their economy to welfare spending, population health outcomes tend to improve likewise. An exception to this rule is the United States (US). In 2005, the US spent 7% more of their GDP on health care compared to other OECD countries

(16% vs. 9%); however, the US still ranked 25th in life expectancy and 29th in infant mortality among 30 OECD countries (OECD, 2009).

The second research stream conceives politics in terms of welfare regimes, and contends that the quality of welfare state programs is more salient than the quantity of welfare effort (Esping-Andersen, 1990, 1999). According to Esping-Andersen (1990),

“The existence of a social program and the amount of money spent on it may be less important than what it does,” (p. 2) and “Welfare states may be equally large or comprehensive, but with entirely different effects on social structure.” (p. 58). Given this limitation, Esping-Andersen (1990) developed a decommodification index, which measured the extent to which workers did not have to sell their labour as commodities

(e.g., receiving unemployment, sickness, and old age benefits), and classified countries into three welfare regimes: (1) liberal (United States, Ireland, Australia, Canada, and the

United Kingdom), (2) conservative (Germany, Switzerland, Austria, and France), (3) and social democratic (Sweden, Denmark, Norway, Finland). The welfare regime approach profoundly reoriented welfare state scholarship, and inspired various new typologies

(Castles & Mitchell, 1992; Ferrera, 1996; Scruggs & Allan, 2006a) and institutional

3 clusters (Korpi & Palme, 1998; Huber & Stephens, 2001; Hall & Soskice, 2001a, 2001b).

Applied to population health, the dominant hypothesis contends that welfare regimes with high levels of decommodification are the most advantaged given that individuals and families are allowed to maintain a socially acceptable standard of living regardless of their market performance or familial supports (Bambra, 2005, 2006, 2007b; Eikemo &

Bambra, 2008). Studies using a tripartite or four- or five-category welfare state model have reached the general consensus that social democratic countries, which feature universal welfare programs guaranteed to all citizens, extensive income maintenance programs, and generous family leave policies, have better population health outcomes and narrower health inequalities compared to conservative, liberal, and southern/ex- dictatorial regimes (Bambra, 2005; Bambra & Eikemo, 2009; Chung & Muntaner, 2007;

Eikemo & Bambra, 2008; Eikemo, Bambra, Judge, & Ringdal, 2008; Eikemo, Huisman,

Bambra, & Kunst, 2008; Muntaner, Borrell, Kunst, Chung, Benach, & Ibrahim, 2006;

Popham, Dibben, & Bambra, 2013; Zambon et al., 2006).

The third stream focuses on leftist politics, which refers to organizations and institutions committed to reducing social, economic, and gender equalities in capitalist democracies (Allan & Scruggs, 2004; Brady, 2003, 2009; Lenski, 1966). Labour unions and left political parties are the most well known examples given their ideological commitments to economic security and social democracy (Hicks, 1999; Huber, Ragin, &

Stephens, 1993; Korpi, 1978, 1983; Stephens, 1979; Swank, 2002). Leftist politics also includes voter turnout and female politicians. It is argued that increases in voter turnout and women in government tend to shift the political ideology of governing parties toward the left (Hicks & Swank, 1992; Iversen, 2001; Piven & Cloward, 1994; Pontusson &

4 Rueda, 2010). The underlying argument is that countries with strong histories of leftist politics are more likely to implement universal and generous welfare programs and policies, which in turn, result in better population health (Borrell, Espelt, Rodríguez-Sanz,

& Navarro, 2007). Strong labour unions can also have a direct and positive effect on workers’ health through the negotiation of higher wages, promotion of safer working conditions, and assurance of long-term employment (Bach, 2003; International Labour

Organization, 2007; Kawakami, Kogi, Toyama, & Yoshikawa, 2004; Muntaner et al.,

2002). Compared to the volume of research on welfare generosity and welfare regimes, research on leftist politics and population has received less attention (Muntaner et al.,

2011, 2012). Despite this, previous studies find that political parties with egalitarian ideologies (e.g., Social Democratic, Labour, Socialist) tend to implement redistributive policies, which has the effect of reducing social inequalities and improving health outcomes such as infant mortality and life expectancy (Chung & Muntaner, 2006;

Navarro et al., 2003, 2006; Navarro & Shi, 2001). Other work finds that countries with greater trade union membership and political representation by women have better health profiles with respect to child mortality (Lynch et al., 2001) and injury mortality

(Muntaner et al., 2002).

As argued here, political forces, processes, and institutions are significant determinants of population health, and its impact operates through proximate (e.g., welfare generosity) and distal (leftist politics) pathways. However, there are important limitations which warrant further attention. First, most studies on the political determinants of population health have exclusively used comparative methods to analyze data from the same set of wealthy democracies (e.g., North America, Europe,

5 and Australia). Thus, new data sources are needed to determine whether the associations found between politics and health between nations are also applicable within nations. Second, the most intriguing evidence on the impact of leftist politics on population health have used descriptive and bivariate methods, thus leaving other potential explanations unaccounted for (Coburn, 2004; Lynch et al., 2001; Muntaner et al., 2002; Navarro et al., 2003, 2006; Navarro & Shi, 2001). To seriously advance these arguments, more rigorous tests are needed, which control for alternative explanations.

An important exception is Chung and Muntaner (2006), who used robust panel data techniques to find that increases in the percentage of left votes are associated with improvements in infant mortality, under-5 mortality, and low birth weight.

Given these limitations, this dissertation conceptualizes politics as a macro-social determinant of health to examine the impact of provincial welfare generosity (aggregate and disaggregated-levels of provincial expenditures), power resources and political parties (union density and left, centre, and right political parties), and democratic politics

(voter turnout and women in government) on total, male, and female age-standardized mortality rates among Canadian provinces. Age-standardized mortality rates are tested become they allow for consistent comparisons across provinces and over time. This involves a time-series cross-sectional (TSCS) study of ten provinces from 1976 to 2008, including Newfoundland, Prince Edward Island, Nova Scotia, New Brunswick, Quebec,

Ontario, Manitoba, Saskatchewan, Alberta, and British Columbia.

Canada is often considered a liberal welfare state, similar to the US and the

United Kingdom, with minimal levels of state intervention in the market, low levels of unionization, and weak federal support for left political parties (Esping-Andersen, 1990,

6 1999; O’Connor, 1989). Although the development of Canada’s welfare state has been heavily influenced by free market and individualist ideologies, qualitative evidence suggests that labour unions and left political parties played essential roles in increasing federal redistributive programs (e.g., unemployment and family benefits) and implementing a universal government-funded healthcare system (Mulé, 2001).

For the purposes of this dissertation, Canada serves as an attractive case study to advance theory and research on the link between politics and health within a single country (Gattinger, Saint-Pierre, & Gagnon, 2008; Imbeau et al., 2000; Kellermann,

2005). First, Canadian provinces are sub-national units that possess sufficient political autonomy to achieve different levels of population health. Under Canada’s Constitution, provincial governments are responsible key welfare services, social transfers, and revenue sources, which include the provision of welfare services (e.g., health, social services, and education), the allocation of funds to municipal governments, and the legislation of income and capital taxes and minimum wages (Dyck, 2000). Without minimizing the importance of the federal government, Canadian provinces possess various instruments with which to influence social inequalities and population health

(Osberg & Xu, 1999). Second, this dissertation’s political variables of interest (e.g., welfare generosity, union density, left political parties) vary considerably across provinces. For example, Quebec and Nova Scotia spent the most and least on social services in 2007 (24.5% versus 8.1% of program expenditures) (Statistics Canada,

2007). In 2011, union density coverage ranged from a high of 39.6% in Quebec to a low of 23.9% in Alberta (Uppal, 2011). Also, the provincial success of Canada’s left political parties, the New Democratic Party (NDP) and Parti Quebecois, ranges from having a

7 strong presence in Saskatchewan, Manitoba, British Columbia, and Quebec to comparatively weak support in Alberta and the Atlantic provinces (Imbeau et al., 2000).

Third, there is some justification to conceptualize provinces not only as sub-national units but also as provincial regimes based on similar policy preferences or institutional characteristics. Guided by welfare regime theory (Esping-Andersen, 1990, 1999), existing studies have shown that provinces cluster into distinct groups based on different social assistance schemes (Boychuk, 1998), welfare policies (Bernard & Saint-

Arnaud, 2004; Saint-Arnaud & Bernard, 2003), and childcare programs (McGrane,

2010). For example, Bernard and Saint-Arnaud (2004) find that while Canada’s welfare regime as a whole qualifies as a liberal state, Quebec leans towards the social democratic regimes of Europe and Alberta leans towards the “ultra-liberal” regime of the

US. This raises the possibility that provinces may cluster together based on similar political histories, which are predictive of population health.

This dissertation’s main findings are three-fold. First, provincial welfare generosity is significantly associated with lower mortality rates, net of other province- specific factors. As provinces devote a larger share of their government expenditures to medical care, other social services, and post-secondary education, mortality rates tend to decline. Second, the political power of left and centre political parties and women in government have significant negative effects on mortality rates. Population health tends to improve when New Democratic, Parti Quebecois, and Liberal party members as well as female representatives are elected to provincial governments for long periods of time. Whereas left political parties and women in government combine with provincial welfare generosity to improve population health (left parties and female politicians

8 reduce mortality rates net of welfare generosity), the effect of centre political parties channels through provincial expenditures (provincial welfare generosity mediates the effect of centre parties on mortality). Third, provinces cluster together to form three distinct regimes based on leftist politics: (1) leftist (British Columbia, Manitoba, and

Saskatchewan), centre-left (Ontario and Quebec), and conservative (Alberta, New

Brunswick, Newfoundland, Nova Scotia, and Prince Edward Island). Compared to leftist provinces (reference category), centre-left and conservative clusters have significantly higher mortality rates; however, differences in provincial welfare generosity account for most of the mortality differences across provincial regimes.

Theoretical Location

How do politics affect population levels of health? To answer this question, this dissertation applies Rose’s ([1985] 2001) “population perspective” to conceptualize politics, measured as provincial welfare generosity, power resources, political parties, and political democracy, as macro-social determinants of population health (Beckfield &

Krieger, 2009; Muntaner et al., 2011, 2012; Navarro et al., 2003, 2006).

The advantage of Rose’s ([1985] 2001) ‘population perspective’ is that it makes a clear distinction between the causes of incidence (e.g., why do some populations experience an illness?) and the causes of cases (e.g., why do some individuals experience an illness?). The core argument contends that an individual’s risk of illness cannot be understood in isolation from the disease risk of the population (Doyle, Furey,

& Flowers, 2006; Marmot, 1998; Rose, 1992). In his classic example, Rose ([1985]

2001) compared the systolic blood pressures of middle-aged men in London and Kenya

9 and found no individual differences even though hypertension is significantly more common among London civil servants. Although a person living in London is more likely to experience hypertension compared to someone living in Kenya, the explanation is not because any particular individual in London happened to have high blood pressure, but because the population distribution of hypertension in London society as a whole is shifted to the right of the Kenya distribution (Berkman & Kawachi, 2000). Marmot (1998) augmented this idea by showing that the age and sex distribution of homicide perpetrators did not significantly differ between Chicago and England/Wales, even though homicide rates in Chicago are 30 times that of England/Wales. Having confirmed that “sick individuals” exist alongside “sick populations”, Rose ([1985] 2001) argued that “to find the determinants of prevalence and incidence rates, we need to study characteristics of populations, not characteristics of individuals” (p. 428). It is also advocated that political contexts be incorporated into explanations on why some people stay healthy and others get sick (Dunn, Frohlich, Ross, Curtis, & Sanmartin, 2005; Rose et al., 2008).

Hence, this dissertation applies Rose’s (1992) population perspective to study macro-level associations between politics and population health among Canadian provinces. By extension, this implies that not all causes of population health can be reduced to mid- or individual-level determinants (e.g., socioeconomic position, behavioural risk factors). Some determinants of health are best understood, measured, and tested as collective in nature and at the macro-level (Putnam & Galea, 2008). This view is consistent with Wright, Levine, and Sober’s (1992) argument that micro- foundational accounts are insufficient to capture the explanatory power of macro-level

10 causes and effects. Taking a macro approach also avoids what Hawley (1992) refers to as reductionist fallacy, or the presumption that societal patterns like population health can be explained by lower units of analysis.

In this respect, Canadian provinces are macro-level entities that are more than the sum of their individual parts (e.g., individuals, families, neighbourhoods, cities). I argue that provinces exhibit emergent political properties and health outcomes that cannot be adequately explained with lower units of analysis. Existing studies confirm that macro-level units like welfare states, political parties, and labor unions matter with respect to population health (Navarro et al., 2003, 2006). The advantage of focusing on macro-level determinants, or more “distal” and “upstream” causes, is that greater improvements in population health may be gained given that the majority of cases of illness arise within the bulk of the population who are outside the tail of high risk (Rose,

1992).

Within a macro-level context, I examine the connections between leftist politics, provincial welfare generosity, and population health among Canadian provinces. My theoretical argument is two-fold. First, power resources (labour unions and left parties), political parties (centre and right parties), political democracy (voter turnout and women in government), and provincial welfare generosity (e.g., health-promoting resources) are macro-level determinants of population health. Second, leftist politics may maintain an association with population health despite changes in provincial welfare generosity or net of provincial welfare generosity. In other words, leftist politics may affect population health along two potential pathways: channeled (e.g., leftist politics affects health by triggering welfare generosity) or combined (e.g., leftist politics combines with welfare

11 generosity to affect health). These pathways between politics, welfare generosity, and distributional outcomes such as poverty and income inequality have been established

(Brady, 2003; 2009; Kelly, 2011). The pathways between politics, welfare generosity, and population health are not.

Methodological Approach

This dissertation combines TSCS data to test politics as a macro-level determinant of population health. My methodological approach attempts to strike a balance with respect to the theory-method nexus (Plümper, Troeger, & Manow, 2005), which means that model specifications are guided by econometric theory (e.g., satisfying ordinary least squares (OLS) assumptions) as well as substantive grounds (e.g., modeling the long-effects of politics on health). Data analyses are based on cross-provincial comparisons from 1976 to 2008. All data is measured at the provincial level and the unit of analysis is “province-year”. The sampling frame consists of the total number of province-years based on data availability. Analyses on provincial welfare generosity and leftist politics included 200 (1989-2008) and 330 (1976-2008) observations, respectively.

Estimation techniques examined whether aggregate and disaggregated welfare expenditures, power resources and political parties, and political democracy affect population health, measured as total, male, and female age-standardized mortality rates, net of other province-level factors such as transfers, debt, dependency ratio, female labour force participation, unemployment, low income, and GDP per capita.

By using TSCS data, this dissertation overcomes the ‘small N’ problem, which arises when the number of macro-level units of analysis is inherently limited. In studies

12 where individuals are the units, entire populations can be sampled to increase sample size and power; however, when provinces are the units, the final sample is limited to ten cases. Statistical problems also arise when the final sample is not much greater than the total number of variables, leaving too few degrees of freedom and testing “over- determined” multivariate models, both of which contribute to biased results (Goldthorpe,

1997). This dissertation overcomes these methodological challenges by constructing a dataset of N x T observations, consisting of cross-sectional data on N (province-units) and T (time units). This significantly increases the number of comparisons that can be made across provinces, and allows for the statistical testing of causal inferences, multiple causation, and interaction effects (Goldthorpe, 1997; Ragin, 1994). Although

TSCS methods have been the standard approach in studies of comparative political economy for the past two decades (Beck, 2001; Beck & Katz, 2001, 2005, 2006), its application in public health research has been severely under-utilized (Chung &

Muntaner, 2006; Lundberg et al., 2008; Muntaner et al., 2011, 2012).

While the TSCS design solves the ‘small N’ problem, using N x T observations presents a number of statistical concerns that may affect the precision of estimation results (Podestà, 2002, 2006). The most common problem concerns the violation of standard OLS assumptions about the error process. The residuals for regression equations estimated from TSCS data using OLS procedures tend to violate three assumptions: 1) serial correlation of errors (i.e., errors are not independent from one time period to the other); 2) panel heteroscedasticity (i.e., errors have different variances across province units); and 3) contemporaneous correlation of errors (i.e., errors are correlated across provinces due to common exogenous shocks). In this

13 dissertation, I address contemporaneous correlation and heteroscedasticity by using

OLS parameter estimators with panel-corrected standard errors (PCSEs), as recommended by Beck and Katz (1995). Serial correlation is remedied by using a first- order autocorrelation correction model (AR1). Fixed effects are also used to capture unit effects in the error term. Although this estimation strategy is arguably the most defensible to test this dissertation’s hypotheses (Beck, 2001; Plümper et al., 2005), I used several alternative estimation techniques to demonstrate the robustness of my results (e.g., fixed- and random-effects, feasible generalized least squares, Driscoll Kray standard errors). Conclusions are generally consistent with these alternative estimations.

What this Dissertation Adds

This dissertation makes two new contributions to the extant literature. First, I confirm that political sociology theories can be effectively applied within a Canadian context to understand population health. Most studies on the political determinants of health have used comparative methods to analyze data from the same set of wealthy nations (e.g.,

North America, Europe, and Australia). For example, cross-national welfare generosity studies find that increases in public spending are associated with improvements in population health (Kangas, 2010; Lundberg et al., 2008). A growing set of results in the leftist politics literature finds positive associations between labour unions, left political parties, and population health among wealthy nations (Chung & Muntaner, 2006; Lynch et al., 2001; Muntaner et al., 2002; Navarro & Shi, 2001; Navarro et al., 2003, 2006).

Also, studies using a welfare regime approach have conceptualized countries as the primary unit of analysis, thus overlooking the possible the health effects of sub-national

14 regimes. Despite the fact that Canada is a liberal welfare state, making it a challenging context to analyze welfare generosity, leftist politics, and welfare regimes, this dissertation finds that political dynamics within Canada have a robust and powerful negative influence on mortality rates.

Second, this dissertation increases the methodological rigor of past studies on politics and health. To date, some of the most intriguing studies on the political determinants of health have relied on descriptive and correlational methods (Coburn,

2004; Lynch et al., 2001; Muntaner et al., 2002; Navarro & Shi, 2001; Navarro et al.,

2003, 2006). For example, Muntaner et al. (2002) used bivariate tests that adjusted for

GDP to find that increases in voter turnout, social pact (a measure of pact between labour and employers), percentage of left votes and seats, women in government are significantly associated with decreases in infant mortality rates in both males and females. In another key study, Navarro et al. (2006) also used correlational methods to find that political parties with egalitarian ideologies tend to implement redistributive policies, which in turn, has a salutary effect on infant mortality and life expectancy. The implication is that one cannot rely on bivariate methods to conclude that politics affects population between- and within-countries. Following the lead of Chung and Muntaner’s

(2006) work, this dissertation increases our confidence on the causal links between politics and health with the use of TSCS data and methods.

Outline of Chapters

This dissertation’s remaining chapters are organized as follows. In chapter 2, I provide the scholarly background and justification for this dissertation’s research questions and

15 hypotheses. The aim is to critically appraise previous work, identify current advances, and make clear how this dissertation makes an original contribution in theory and empirics to the extant literature. Chapter 3 summarizes this dissertation’s conceptual model and outlines the relationships between key concepts.

Chapter 4 reviews methodological details on data sources, independent and dependent variables, and estimation techniques. Chapter 5 examines the impact of aggregate and disaggregated forms of provincial welfare generosity on population health. I conceptualize Canadian provinces as sub-national welfare states and test which specific measures of welfare generosity influence population health. Chapter 6 examines the impact of leftist politics on population health. I test the health effects of power resources, political parties, and political democracy, and determine the pathways through which leftist politics and provincial welfare generosity affect population health.

In chapter 7, I conduct exploratory analyses to examine whether provinces cluster together to form distinct provincial regimes, and if so, whether these regimes are predictive of mortality rates. Chapter 8 summarizes major findings by revising this dissertation’s conceptual model, outlines study limitations, and offers suggestions for future research. In the concluding chapter, I outline a set of public health strategies and policies informed by this dissertation’s major findings.

16 CHAPTER TWO. LITERATURE REVIEW

This literature review consists of four sections that provide the theoretical and empirical contexts to guide this dissertation’s research questions and hypotheses. In section one,

I review contemporary social epidemiological approaches to population health, trace their sociological traditions, and note their limitations. In section two, I summarize dominant theoretical frameworks used in political sociology that elucidate the political contexts and power relations between state and society, and their hypothesized associations with population health. In section three, I conduct a systematic review of the empirical literature on welfare generosity, welfare regimes, leftist politics, and population health, summarizing our current state of knowledge. In the final section, I conclude with final observations and indicate how this dissertation contributes new knowledge to our knowledge base.

Social Epidemiological Approaches

This dissertation’s political approach to population health is intended to augment prevailing SDOH frameworks used in social epidemiology (O’Campo & Dunn, 2011).

Instead of devoting much attention to politics, prevailing social epidemiological approaches are primarily concerned with how social determinants give rise to systematic patterns of health between groups and populations (Berkman & Kawachi,

2000; CSDH, 2008). The major contribution of this work has been to move beyond traditional risk factors (e.g., diet, smoking, and exercise) and specific disease outcomes

(e.g., general susceptibility to disease, (Berkman & Syme, 1979) to focus on the living

17 conditions in which people are born, grow, live, work and age (CSDH, 2008). Recent reviews of the Canadian literature identify 17 key SDOH: income and income distribution, education, unemployment and job security, employment and working conditions, early childhood development, food insecurity, housing, social exclusion, social safety net, health services, Aboriginal status, gender, race, and disability

(Mikkonen & Raphael, 2010; Muntaner, Ng, & Chung, 2012).

On balance, SDOH research has made significant contributions by clarifying which types of individuals and groups are more likely to experience poor health (e.g., the poor, disabled and homeless, less educated, unemployed, women, immigrants, migrants, racial/ethnic groups, Aboriginals) and how health inequalities tend to follow a social gradient (e.g., health inequalities follow a clear pattern that runs from top to bottom along various social indicators). For example, low, middle, and high socioeconomic positions translate into poor, moderate, and good health, respectively.

Given that social gradients in health have been found across time periods using different social indicators and health outcomes, it constitutes one of the most consistent findings in social epidemiology. Social determinants not only account for general improvements in population health over time (McKinlay & McKinlay, 1977; Evans, 2002;

Evans, Barer, & Marmor, 1994; Raphael, 2006) but also for health differences between and within wealthy nations (CSDH, 2008; Marmot, 2005; Marmot & Wilkinson, 2009;

Raphael, 2004; Navarro & Muntaner, 2004).

Two dominant frameworks have emerged to explain the mechanisms that connect social causes to health effects. Each framework is associated with a different strand of sociological theory (Gehlert et al., 2008; Macintryre, 1997; Muntaner, Ng,

18 Vanroelen, Christ, & Eaton, 2013; Raphael, 2006). The first reflects a materialist and neo-materialist approach that emphasizes material life conditions as the primary determinants of health (Evans, Hertzman, & Morgan, 2007; Lynch & Kaplan, 2000;

Shaw, 1999; Shaw, Dorling, & Mitchell, 2002; Tarasuk, 2009; Wilkins, 2007) as well as public infrastructures in mediating or moderating health outcomes (Bryant, 2009; Dorling et al., 2005; Hall & Taylor, 2009; Ross et al., 2000; Sanmartin et al., 2003). With respect to sociological theory, this approach relies on a social stratification perspective, and in particular Weber’s notion of “life chances”, to rank individuals along multidimensional continuums (e.g., socioeconomic position) and to assess the link between “life chances” and health outcomes (Galobardes, Shaw, Lawlor, Lynch, & Smith, 2006; Giddens,

Ociepka, & Zujewicz, 1973; Lynch & Kaplan, 2000). From a materialist perspective, it is argued that social stratification exposes individuals to various forms of disadvantaged material conditions, such as poverty, poor housing, and poor working conditions, which accumulate over the life-course to produce poor health (Lynch, Smith, Kaplan, & House,

2000; Lynch, Due, Muntaner, & Smith, 2000; Muntaner & Lynch, 1999; Shaw, 1999;

Smith, 2003). Hence, material conditions reflect the impact of various social determinants and affect health by influencing the quality of individual development, family life and interaction, and community environments (Gordon & Townsend, 2000;

Shaw, 1999; Shaw et al., 2002). A convincing body of research supports the idea that material conditions are significant predictors of poor health, which operate through a series of connected pathways (Gordon, 1999b; Gordon & Townsend, 2000; Keating &

Hertzman, 1999; Mitchell, Dorling, & Shaw, 2000; Shaw, 1999; Shaw et al., 2002).

These pathways include the experience of absolute deprivation (e.g., lacking the

19 material necessities of life) and increased exposure to material hazards, both which lead to differences in psychosocial stress (e.g., “fight or flight” responses) (Brunner &

Marmot, 1999; Stansfeld & Marmot, 2002) and the adoption of health-threatening behaviours (e.g., poor diets, use of tobacco, physical inactivity) (Marmot, 2004).

In addition to acknowledging the links between material conditions and health, the neo-materialist approach argues that population health outcomes are also the result of how health-promoting resources are distributed within the population (Lynch et al.,

2000). Material conditions are significant predictors of population health; however, public infrastructures, which determine and distribute the quality and availability of

SDOH, also matter because of their important role in mediating and moderating health outcomes (CSDH, 2008; Kaplan, Pamuk, Lynch, Cohen, & Balfour, 1996; Lynch et al.,

2004; Ross & Wolfson, 1999; Ross, Wolfson, Berthlot, & Dunn, 1999; Ross et al.,

2000). For example, compared to the US and Canada, Sweden’s public infrastructure and distribution of resources is much more egalitarian and generous, which has translated into lower levels of poverty, narrower income gaps between rich and poor, and superior health outcomes (Raphael, 2003). It has been suggested that Canada enjoys better health than the US as measured by infant mortality rates, life expectancy, and mortality from childhood injuries (Sanmartin & Ng, 2004) because the former nation spends relatively more on strengthening its public infrastructures (Ross et al., 2000).

This seems to indicate that if welfare states distribute resources more evenly among its citizens, SDOH are strengthened and health outcomes are positively affected.

Moreover, this finding has empirical support even within jurisdictions within nations.

Kaplan and colleagues (1996) found that US states and cities with weaker public

20 infrastructures are less likely to invest in key SDOH (e.g., health, social services, and education), experience higher levels of poverty and income inequality, and disadvantaged health profiles. Thus, the neo-materialist approach emphasizes both the effects of material conditions on individuals’ health and the societal factors that determine the quality of SDOH.

The second approach, in which income inequality is theorized as the most important explanatory factor in determining population health, focuses on the impact of hierarchical standings, social capital, and social cohesion (Muntaner & Lynch, 1999;

Muntaner et al., 2013; Wilkinson, 2002, 2005; Wilkinson & Pickett, 2006, 2009).

Scholars have labeled this approach as “neo-Durkheimian” given that people experience and confront income inequality and social hierarchies as social facts, or objective realities that are beyond themselves (Muntaner & Lynch, 1999; Muntaner, Rai,

Ng, & Chung, 2012). The income inequality hypothesis argues that after wealthy nations make sufficient inroads toward economic development (approximately US $25,000 per capita), what matters most in determining population levels of health is the degree to which income is distributed among the poor and affluent (Auger & Alix, 2009; Auger,

Hamel, Martinez, & Ross, 2012; Kawachi, Kennedy, Lochner & Prothrow-Stith, 1997;

Kawachi, Kennedy, & Wilkinson, 1999; Wilkinson, 2002, 2005; Wilkinson & Pickett,

2006, 2009). Empirically, income inequality research has produced strong and consistent findings, which primarily involve between-country and within-nation (e.g., US states) correlations between various measures of income inequality (e.g., Gini coefficient, percent share of total household income received by the least well-off 50 percent of the population) and indicators of social problems and population health such

21 as infant mortality, self-rated health, life expectancy, teenage pregnancy, obesity, mental health, drug use. Existing studies find that more equal societies (e.g.,

Scandinavian countries, Japan, New Hampshire) are significantly healthier than their unequal counterparts (e.g., Liberal nations, United States, and Louisiana) (Wilkinson &

Pickett, 2009).

How do income gaps between the affluent and the poor “get under the skin” to produce population health differences across jurisdictions? The underlying mechanisms begin with psychosocial comparisons along two related constructs: social hierarchies and social cohesion (Kawachi, Kennedy, & Glass, 1999; Wilkinson, 2002, 2005). First, the perception and experience of social hierarchies in unequal societies are theorized to intensify poor self-esteem among lower status individuals, which in turn, causes poor health through psycho-neurobiological pathways. Second, steeper levels of hierarchy lead to a breakdown in social cohesion (e.g., cooperation, reciprocity, trust, civic participation) – a concept closely associated with Emile Durkheim and his work on social integration and suicide (Schwartz & Diez-Roux, 2001). As levels of social cohesion decline, population health is affected directly through psychological stress, poor self-esteem and increased mistrust and suspicion and indirectly through the weakening of public infrastructures (Ross et al., 2000).

Taken together, a wealth of evidence supports the arguments that social determinants operate through material pathways to influence population health (CSDH,

2008; Lynch et al., 2004), public infrastructures distribute social and economic resources to mediate and moderate the impact of SDOH (Ross et al., 2000), and higher levels income of inequality are associated with poorer population health outcomes

22 between and within national jurisdictions (Wilkinson & Pickett, 2009). Despite these contributions, population health theory and research runs the risk of becoming stagnant and repetitive if current frameworks remain focused on materialist, neo-materialist, and psychosocial frameworks. In order for the study of population health to advance in new ways, this dissertation acknowledges and addresses two limitations of prevailing frameworks.

Limitations of Prevailing Frameworks

Prevailing frameworks in social epidemiology are limited in two respects: 1) exclusive focus on mid-level determinants; and 2) neglect of political determinants of health

(Muntaner & Chung, 2009). First, the conventional approach in population health studies is to focus on mid-level social determinants and to compare the characteristics of disadvantaged groups (e.g., low socioeconomic position) to advantaged groups (e.g., high socioeconomic position). It is not an exaggeration to say that the vast majority of materialist and neo-materialist studies explain why one group of people are more likely to be unhealthy, or why some individuals are healthy while others are not (Muntaner,

Rai, Ng, & Chung, 2012). The implication is that mid-level factors are often conceptualized as the fundamental causes of health (Link & Phelan, 1995). As a consequence, materialist and neo-materialist explanations amount to what Merton

(1957) refers to “theories of the middle range”, which favour testing ‘minor working hypotheses’ (e.g., what is the degree of social gradients in health in a given population?) rather than ‘grand social theories’ (e.g., what are political and economic processes create social gradients in health in the first place?). Thus, a shortcoming of current

23 research has been the failure to consider “a master conceptual scheme” that sheds light on how macro-level forces may affect the quality and availability of SDOH (Raphael,

2006).

Second, prevailing population health perspectives tend to overlook the political contexts of population health. For example, most income inequality studies pay minimal attention to how distributional and health outcomes may be affected by political and economic factors (Muntaner & Lynch, 1999; Muntaner, Rai, Ng, & Chung, 2012). Given that income inequality is an emergent property of populations, income distribution studies use countries or sub-national jurisdictions as units of analysis. At the same time, however, these studies treat income inequality as a perfectly exogenous construct, which implies a conflict-free vision of society (e.g., as opposed to class conflict being an inherent feature of capitalist economies) (Navarro & Muntaner, 2004). The consequence of theorizing income inequality as the social fact to be transformed is problematic because it reduces the debate to a distributional conflict over income (e.g., increase minimum wages, increase taxes, increase income transfers) rather a societal conflict over the political and economic relations of production that create income inequalities in the first place (Muntaner & Lynch, 1999). The implication of focusing on income inequality and not considering the broader political context is that it serves to depoliticize the determinants of population health (Bambra et al., 2005; Muntaner, 2004).

Given these limitations, prevailing population health frameworks can benefit from broadening its substantive foci. To this end, political sociology offers a wealth of conceptual and analytical models to guide social theory on how political determinants

24 shape population levels of health (Beckfield & Krieger, 2009).

Political Sociological Theories

At the intersection of sociology and political science, political sociology has developed several useful conceptual models to understand the politics of population health

(Beckfield & Krieger, 2009; Navarro & Shi, 2001). These models provide constructive insights on the various intersections between the state and civil society (Janoski, 2005;

Svallfors, 2007), and on the causal links between political and welfare state determinants and population health. Building on an established body of social science literature, three political sociology frameworks are particularly relevant for the purposes of this dissertation: 1) welfare states, understood as generosity and regimes; 2) power resources and political parties, and 3) political democracy. Below and in Table 2.1

(located at end of chapter), I describe the key features of these frameworks and outline testable hypotheses.

Welfare States: Generosity and Regimes

The modern understanding of the welfare state denotes an industrial capitalist society in which state power is “deliberately used (through politics and administration) in an effort to modify the play of market forces” (Heclo, 1974; Kerr, Dunlop, & Harbison, 1960; Pryor,

1968; Rimlinger, 1971; Wilensky & Lebeaux, 1965). From the late 1950s to the early

1970s, many industrialized countries, including Canada, accelerated the expansion of their welfare states, establishing new social insurance schemes and national service programs in welfare, education, and health care. Whereas the average expenditures on

25 social transfers represented 7.5% of the GDP among affluent democracies in 1960, this figure had nearly doubled to 14% in 1980 (OECD, 1994). The sheer scale and scope of welfare state expansion of industrialized nations made them the focus of concentrated attention as a defining feature of capitalist democracies (Myles & Quadagno, 2002).

Given this concentrated focus, sociologists and political scientists over the past three decades have conceptualized the welfare state as both an independent and dependent variable to explain their development and effects on various outcomes such as income inequality and poverty (Bradley, Huber, Moller, Nielsen, & Stephens, 2003; Goodin,

1999; Headey, Goodin, Muffels, & Dirven, 1997; Hicks & Swank, 1984; McCrate, 1997;

Kenworthy, 1999; Korpi & Palme, 1998; Brady, 2005; Moller, Huber, Stephens, Bradley,

& Nielsen, 2003; Scruggs & Allan, 2006b; Scruggs, 2007, 2008). More recently, social epidemiologists and medical sociologists have extended welfare state scholarship by examining welfare state performance in improving population health (Bambra, 2005;

Chung & Muntaner, 2007; Chung et al., 2013; Conley & Springer, 2001, Eikemo et al.,

2008a, 2008b; Witvliet, Arah, Stronks, & Kunst, 2011; Witvliet, Kunst, Stronks, & Arah,

2012) and reducing health inequalities (Avendano, Juerges, & Mackenbach, 2009;

Bambra & Eikemo, 2009; Borrell et al., 2007; Dahl et al., 2006; Espelt et al., 2008;

Popham, Dibben, & Bambra, 2013).

In public health, the welfare state is commonly understood as the complex of social policies and programs that provide key welfare services (e.g., publicly funded services such as health care, education, and social services) and social transfers (e.g., income maintenance programs, social security or cash benefits such as unemployment, pensions and sickness and disability benefits (Eikemo & Bambra, 2008). The provision

26 of such services and transfers are designed to ensure social inclusion and economic capability as well as relieving citizens from being forced to exclusively depend on the private market for economic resources. Given that welfare states differ systematically in how they approach the provision of supports and security across the lifespan (Esping-

Andersen 1990, 1999), public health researchers have applied these ideas to better understand how welfare services and social transfers and linked to social inequalities and population health (Bambra, 2005, 2006, 2007b, 2011; Kaplan, 2007).

Recent theoretical thinking suggest that welfare states perform at least two major functions that affect population levels of health: 1) ensure equal access to an agreed range of publicly funded services; and 2) protect against certain universal health risks.

The first argument contends that welfare states affect population health through the provision of essential welfare services such as education, health care, and housing.

Welfare services provided by governments are considered “impure” or “quasi-public” goods because they are non-rivalrous (an individual's use does not reduce availability to others) yet excludable (individuals can be prevented from using them) (Bird & Slack,

1993). Applied to health, publicly funded services operate as proximate SDOH by building the human capital of the population (e.g., provision of public education and job skills training), preventing the onset of and treating disease (e.g., provision of high quality and universal health care services), and guaranteeing minimum standards of living conditions (e.g., provision of social housing). Second, welfare states are charged with protecting its citizens against certain universal risks that are known to affect health.

According to this argument, welfare states operate as public insurance programs to provide social transfers to people who experience trigger events such as becoming poor,

27 sick, disabled, retired, or unemployed. For example, since unemployment is a systematic feature of advanced capitalism, welfare states prepare for this inevitability by facilitating the saving of money to protect against this risk, which in turn, alleviates the economic and health consequences of unemployment (Gangl, 2006). Whether welfare states are minimal and strict or generous and universal in their provision of social transfers determines the degree to which vulnerable individuals are forced to depend on labour market earnings, familial supports, or private insurance schemes to maintain acceptable standards of living and optimal levels of population health.

To conceptualize and measure the welfare state’s effects on population health, researchers rely on two conceptually different approaches. Whereas the first quantitatively gauges the extensiveness of “welfare effort” or “welfare generosity”

(Wilensky, 1974, 1976, 2002), the second approach uses both qualitative and quantitative indicators to classify countries into distinct welfare regimes (Esping-

Andersen, 1990, 1999; Scruggs & Allan, 2006a). The welfare generosity approach to conceptualizing and measuring welfare state performance involves using public data on government spending, expenditures, and transfers. Government expenditure data is either collected as a percentage of the gross domestic product (GDP) or in adjusted per capita dollars. According to Wilensky (1975), these measures quantitatively gauge the extensiveness of social policy or what is often called “welfare effort”. The hypothesized impact of welfare generosity on population health is straightforward: greater and lower levels of welfare generosity should be associated with advantaged and disadvantaged health outcomes, respectively (per hypotheses 1.1 and 1.2 of Table 2.1). The key argument is that collectively provided resources in the form of essential welfare services

28 (e.g., health care, education, housing) and social transfers (e.g., income maintenance programs) improves population health by ensuring that individuals and families maintain a socially acceptable standard of living despite potential failures in market performance or family supports (Dahl & van der Wel, 2013; Eikemo & Bambra, 2008; Fritzell &

Lundberg, 2007; Lundberg et al., 2008). Representative examples of welfare generosity studies in population health include the works of Chung and Muntaner (2006), Conley and Springer (2001), Dahl & van der Wel (2013), Kangas (2010), Lundberg and colleagues (2008).

In contrast, the welfare regime approach contends that the quality of welfare programs is more salient than welfare generosity (Esping-Andersen, 1990, 1999).

Rather than focusing on welfare effort, Esping-Andersen (1990) developed an index of decommodification, which measured the extent to which workers did not have to sell their labour as commodities (e.g., unemployment benefits, sickness benefits, and pensions), and proposed a tripartite model of welfare state regimes: social democratic, liberal, and conservative. The social democratic welfare regime is epitomized by

Sweden, and typically includes Denmark, Norway and Finland. This regime features universal welfare programs guaranteed to all citizens, extensive public employment systems, and generous family leave policies. With their high rates unionization and female labour participation, the social democratic regime is uniquely collectivist and egalitarian. Australia, Canada, Ireland, New Zealand, the United Kingdom, and the

United States are considered liberal welfare states. These countries are based on free market and individualist ideologies and often rely on means-tested benefits to meet the needs of vulnerable individuals. Lastly, conservative welfare states such as Austria,

29 France, Germany, Italy, and Switzerland favour social insurance for unemployment, sickness, and old age for male breadwinners and their families. This regime developed during authoritarianism and reflects a tradition of corporatism, Catholicism, and familialism.

Since Esping-Andersen (1990) proposed his original tripartite typology, welfare state scholars have considered and classified other countries into distinct regimes, including Southern (e.g., Greece, Portugal, Spain) (Ferrera, 1996), Eastern European

(e.g., former Communist nations) (Esping-Anderson, 1999), East Asian (e.g., Hong

Kong, Japan, Singapore, Republic of Korea, Taiwan) (Walker & Wong, 2005),

Southeast Asian (e.g., Indonesia, Laos, Malaysia, Philippines, Thailand, Vietnam)

(Haggard & Kaufman, 2008) regimes as well as informal-security (e.g., Latin-America and South Asian nations) and insecurity (e.g., Sub-Saharan African countries) regimes

(Gough & Wood, 2006). Dominant hypotheses based on welfare regime theory contend that absolute and relative health outcomes should be better in social democratic countries given their generous and universal welfare provisions and social transfers (per hypothesis 2.1 of Table 2.1). Conversely, regimes that decommodify labour the least, such as liberal and southern nations, are hypothesized to experience worst health outcomes compared to regimes that ease the pressure on workers to sell their labour as commodities as social democratic and conservative nations (per hypothesis 2.2 of Table

2.1). Leading researchers on the health effects of welfare regimes include Bambra

(2005), Borrell and colleagues (2009), Chung and Muntaner (2007), Eikemo and colleagues (2008a, 2008b), Sanders and colleagues (2009), and Zambon and colleagues (2006).

Leftist Politics

Because the welfare state have been shown to be significant predictor of distributional and health outcomes, it is also useful to consider the political determinants of welfare states, and investigate their possible impact on population health. In the comparative literature, the political determinants of welfare state development are also referred as

‘leftist politics’. Broadly, the term ‘leftist politics’ refers to processes, organizations, and institutions that are committed to a more equal distribution of a society’s resources

(Bobbio, 1996; Brady, 2009; Thompson, 1996). Among the more prominent leftist political actors are labour unions and left political parties (e.g., power resources). Other key leftist political actors include centre-left or centre-wing political parties as well as various types of democratic politics (e.g., voter turnout and women in government). At its core, a general theory of leftist politics claims that where labour unions, left and centre-left political parties, and women in government have been historically strong and voter turnout has been high, welfare states are more expansive and generous. Applying this politically-oriented theory to population health suggests that where leftist politics is strong, welfare states are more generous, and better population health outcomes are achieved (per hypotheses 3.1 to 6.2 of Table 2.1) (Borrell et al., 2007; Espelt et al.,

2008; Navarro et al., 2003, 2006).

Power Resources and Political Parties

The three leading theories of welfare state development include the state-centric model

(Amenta, 2000; Skocpol & Amenta, 1986), the functionalist logic of industrialism

31 (Pampel & Williamson, 1988; Wilensky, 1974), and power resources theory (Huber et al.,

1993; Huber & Stephens, 2001; Korpi, 1978, 1983; Stephens, 1979). Whereas the state-centric model focuses on institutional structure and bureaucratic preferences, the functionalist logic of industrialism stresses changes in demography and economic development, and power resources emphasizes the strength of working class power in the form of labour unions and left political parties. Of these three theories, power resources theory has been cited in the extant comparative literature as the most dominant explanation for welfare state development (Bradley et al., 2003; Orloff, 1996;

Hicks, 1999; Huber & Stephens, 2001; Korpi, 1978, 1983, 2006). Guided by a neo-

Marxian understanding of social conflict and class-based politics, power resources theory is rooted in the idea that capital (business, the upper class) and labour (working and lower classes) struggle over valued resources, levels of social inequality, and welfare state generosity. Korpi (1983) defines power resources as “characteristics which provide actors .... with the ability to punish or reward other actors” (p. 15). Given that owners of capital control the means of production, they also control the balance of power resources in capitalist societies. In contrast, working and lower classes have fewer available options or abilities to reward or punish.

Since capital owners possess greater wealth and economic power, they also have more resources to deploy in politics (Burris, 2001; Domhoff, 1998). As a result, capital owners have at their disposal vast resources and effective strategies to shape and influence policy and distributional outcomes, including for example, tilting elections through political contributions, supporting pro-business political parties, and lobbying on behalf of pro-business policies. The unequal distribution of power resources in capitalist

32 democracies means that the default organization of markets skews in favour of business. Two implications follow. First, it enables and legitimizes the exploitation of workers and results in the subsequent economic insecurity for the broader population; and second, given that business has an interest in maintaining this default organization, it exerts its influence to maintain a minimalist welfare state. In this default position, the working class possesses lower levels of labour and political power.

However, capitalist societies also tend to be democratic ones, and freedom of association allows the working class and other leftist political actors to mobilize, organize, and alter power resources. As Huber and Stephens (2001) explain, “The struggle over welfare states is a struggle over distribution, and thus the organizational power of those standing to benefit from redistribution, the working and lower middle classes, is crucial” (p. 17). Working-class organization often involves two power resources. First, workers organize in labour unions, which are the immediate manifestation of working-class mobilization. Successful organizing leads to greater union centralization, bargaining coordination, and strengthens the resolve of workers to strike and interrupt the ability of business to make profits (Korpi, 1978; 1983). Second, the working class and other allied political actors can support left political parties that are pro-labour and pro-distribution. Political parties are essential because they perform the crucial mediating role between working class power and egalitarian outcomes.

When in office, these parties can push for an expansion of the welfare state to protect workers and guard against the economic insecurity that is inherent in capitalism.

According to Korpi (1983), “the variations in the differences between these two basic types of power resources – control over the means of production and the organization of

33 wage-earners into unions and political parties – are thus assumed to be of major importance for the distributive processes in capitalist democracies and for their final result; the extent of inequality” (p. 187). Taken together, the political sociology literature confirms that power resources consist of unions and left political parties, operate as leftist political actors, represent the interests of the working class, and pressure the welfare state to manage risk, distribute economic resources, and decommodify labour

(Bradley et al., 2003; Hicks & Swank, 1984, Huber & Stephens, 1993; Kittel & Obinger,

2003; Korpi, 1983, Moller et al. 2003; Stephens, 1979;).

In recent years, power resources theory has been revised and amended by

Huber and Stephens (2001). In their influential book Development and Crisis of the

Welfare State, these authors proposed power constellations theory, which argues that political parties are the central determinant of social welfare policies (Huber & Stephens,

2001; Moller et al., 2003; Huber, Mustillo, & Stephens, 2008; Huber, Nielsen, Pribble, &

Stephens. 2006). Power constellations theory downplays the role of labour unions to reinforce the importance of left- and centre-wing political parties (e.g., social democratic and Christian democratic) as the key drivers of welfare state development since the

1960s. Using a mix of qualitative and quantitative methods, Huber and Stephens (2001) found strong support that political party incumbency directly and indirectly influence the characteristics of welfare states and the resulting levels and types of social inequality.

Other research confirms the importance and distinctive role of centre-wing political parties, such as Social Catholicism and Christian Democratic parties, in generating high levels of welfare social spending (Kersbergen, 1995). The major contribution of this work is that centre political parties as well as left-wing ones are important to translating

34 class- and ascription-based social cleavages into generous welfare policies that reduce inequality and promote egalitarianism.

Compared to the amount of theoretical thinking devoted to welfare states and population health, considerably less attention has been paid to how labour unions and political parties are associated with population health (Lynch et al. 2001; Muntaner et al.,

2002; Navarro et al., 2003, 2006). One possible causal pathway between power resources and political parties is that unions and left and centre political parties trigger welfare state generosity, which subsequently improves population health (per hypotheses 3.2, 3.3, and 4.1 of Table 2.1). In this version, the impact of power resources and political parties is indirect and channeled through the welfare state to affect population health. Under this scenario, the welfare state mediates the effects of unions and political parties on health. A second plausible causal pathway suggests that the impact of unions and political parties combines with the welfare state to affect population health. In this scenario, power resources and political parties have an additional effect on population health net of the welfare state. If unions and political parties as well as the welfare state are significant predictors of population health, their effects would not mediate each other. For example, trade unions may improve health through higher wages and improved working conditions before and separately from the welfare state (per hypothesis 3.1 of Table 2.1). Major proponents of power resources and partisan political approach to population health include Navarro, Muntaner, Borrell,

Benach, and Chung (Borrell et al., 2007; Navarro et al., 2003, 2006; Chung & Muntaner,

2006; Muntaner et al., 2002).

35 Voter Turnout and Women in Government

Whereas power resources and partisan party theorists emphasize the determining role of labour unions and political parties in triggering welfare state development, theories of political democracy emphasize the importance of other types of leftist politics such as political participation (e.g., voter turnout) (Brym, Gillespie, & Lenton, 1989; Hicks &

Swank, 1992; Iversen, 2001; Piven & Cloward, 1994; Pontusson & Rueda, 2010) and gender equality in politics (e.g., elected women in government) (Lorber, 2010; Matland

& Studlar, 1998; McAllister & Studlar, 2002; Phillips, 1995; Studlar & Matland, 1996;

Wängnerud, 2009).

Political participation can assume a variety of forms including protesting, voting, and engaging more actively in campaign activities. Among industrial democracies, however, more people vote than engage in any other type of routinized mass political behavior (Jackman, 1987). At the same time, there is considerable variation across industrial democracies in voter turnout rates, which has stimulated political democracy scholars to investigate the possible associations between voter turnout rates and welfare state development and generosity (Dye, 1979; Hicks & Swank, 1992; Iverson,

2001; Pampel & Williamson, 1989; Piven & Cloward, 1994). Pampel & Williams (1989) argue that higher levels of voter turnout should be associated with higher levels of welfare generosity to support the underprivileged. This argument is based on three assumptions: 1) low voter turnout reflects low rates of voting among disadvantaged groups (e.g., the poor, racialized groups, unemployed) (Huber et al., 1993); 2) higher rates of voter turnout reflect increases of new, previously excluded, and disadvantaged voters into electoral politics (Iverson, 2001), which has the effect of pressuring

36 governmental incumbents to increase welfare spending; and 3) high voter turnout typically changes the class content of elections, shifting the political centre of gravity to the left (Lijphart, 1997; Piven & Cloward, 1994). Existing studies have confirmed that high voter turnout does operate as a leftist political process that shifts the political spectrum toward left political parties and pressures incumbents to increase welfare spending (Hicks & Swank, 1992).

According to gender equality theories, the historical strength of women in government can also be conceptualized as an indicator of leftist politics (Lorber, 2010;

Phillips, 1995; Wängnerud, 2009). The theoretical premise is that as more women are elected into parliaments, the relative position of women’s interests strengthens, which in turn, exerts a positive effect on welfare generosity, and possibly population health

(Muntaner et al., 2002). Although it cannot be assumed that women politicians are necessarily more committed to women’s interests or egalitarian outcomes, feminist scholars have theorized that increased gender equality in politics makes real contributions to the development, capacity, and policy preferences of modern welfare states (Phillips, 1995; Kelber, 1994; Tremblay, 1998; Wängnerud, 2009). Furthermore, these scholars go on to argue that achieving gender equality in politics is more than simply an issue of equity or human rights - the election of more female representatives makes real and substantive contributions in the policy sphere (Phillips, 1995). The idea that women make a difference in politics implies that as more women enter politics, the attitudes, opinions, and behaviour of men in political parties shift leftward and the overall political agenda becomes more receptive of women-friendly policies (Klein, 1984; Sigel,

1996). Regarding the latter point, it follows that female politicians substantively

37 represent women if, by their opinions and actions, they sustain the wishes, needs, and interests of the female population, which are deemed ‘women’s issues’. According to

Dodson and Carroll (1991), women's issues include both: 1) women's rights bills (e.g., those that are feminist in intent and that deal with issues having a direct impact on women such as pay equity, violence against women, and free choice in matters of reproduction); and 2) women's traditional areas of interest (e.g., bills that reflect women's roles as caregivers both in the family and society and thus that address issues in health care, care of the elderly, education, housing, and the environment). The theoretical connections between women in government, promotion of women’s issues, and increased welfare effort are well-established; however, the empirical rigor of existing studies to verify these claims needs strengthening (Wängnerud, 2009).

Similar to the amount of theoretical thinking on power resources, political parties, and health, research on the plausible pathways through which voter turnout and women in government may affect population health remains in its infancy (Smith & Dorling,

1996, 1997; Kawachi, Kennedy, Gupta, & Prothrow-Stith, 1999; Muntaner et al., 2002;

Navarro et al., 2003, 2006). Existing studies provide preliminary evidence, at best, that high voter turnout rates and more women in government has the effect of triggering welfare spending and improving population health (per hypotheses 5.1 and 6.1 of Table

2.1) (Muntaner et al., 2002; Navarro et al., 2003, 2006). The possible pathways through which political participation and gender equality in politics affect population health are the same as the proposed pathways involving power resources and political parties.

38 Empirics: Political Determinants of Population Health

As argued above, political sociology themes such as welfare states and leftist politics offer new and interesting theories to augment our current understanding on the political context of population health. Next, I review the empirical literature on politics and health to assess the extent to which political sociology theories are supported with empirical evidence.

Sample Selection for Literature Review

A two-step approach is used to locate articles that investigated the empirical link between welfare generosity, welfare regimes, leftist politics and population health: (1) review of electronic databases; and (2) hand search of reference lists. First, I searched

PubMed (1948 - May, 2013) to locate articles that empirically tested the association between political factors and population health in May 2013. Searching the keyword terms: population health AND welfare effort OR welfare generosity OR welfare/social/public/government spending OR welfare/social/public/government expenditure OR welfare state OR welfare regimes OR power resources OR unions OR union density OR trade union membership OR political part* OR power constellations

OR political democracy OR democratic politics OR voter turnout OR voting patterns OR electoral turnout OR women in government OR political participation by women OR political representation by women in the Title/Abstract yielded a total of 5,118 potentially relevant studies. The abstracts of these studies are reviewed and 319 are identified as potentially relevant. The full-text of potentially relevant studies are retrieved and evaluated against the following inclusion criteria: 1) published in English; 2) explicitly

39 formed a priori hypothesis on the independent, mediating, or moderating impact of political variables on population health using cross-national or within-country comparative methods (e.g., ecological or multi-level studies), 3) tested these hypotheses using empirical methods; and 4) focused on overall population health outcomes (excluding studies on health inequalities). A total of 35 studies met the full inclusion criteria. Second, the reference lists of these 35 studies are hand-searched for additional studies. This process identified an additional 11 studies that met the full inclusion criteria, resulting in a final sample of 46 core publications.

40 Consistent with my inclusion criteria, I excluded three types of studies that have appeared with some regularity in the emergent politics-health literature: 1) descriptive studies that did not use quantitative methods to test politically-oriented hypotheses

(Borrell et al., 2007; Burstrom et al., 2010; Coburn, 2004; Chung & Muntaner, 2008;

Farfan-Portet et al., 2010; Lahelma & Arber, 1994; Navarro & Shi, 2001; Raphael &

Bryant, 2004; Sacker, Worts, & McDonough, 2011; Siddiqi & Hertzman, 2007;

Granados, 2010); 2) empirical studies that used individual-level data and methods to examine politics and health inequalities (Avendano et al., 2009; Bambra & Eikemo,

2009; Bambra, Netuveli, & Eikemo, 2010; Borrell et al., 2009; Dragano, Siegrist, &

Wahrendorf, 2011; Eikemo et al., 2008b; Espelt et al., 2008; Frie, Eikemo, & von dem

Knesebeck, 2010; Lahelma, Arber, Rahkonen, & Silventoinen, 2000; Lahelma et al.,

2002; Nordenmark, Strandh, & Layte, 2006; Sanders et al., 2009; Zambon et al., 2006); and 3) systematic reviews of existing studies (Beckfield & Krieger, 2009; Brennenstuhl et al., 2012; Muntaner et al. 2011, 2012). However, I did draw on these studies to inform the analysis of included studies, identify additional papers, and inform my conceptual model (presented in chapter 3).

Studies are not treated as mutually exclusive. A total of nine studies examined more than one political theme and are coded along multiple themes (Chung &

Muntaner, 2006; Conley & Springer, 2001; Kangas, 2010; Lena & London, 1993;

Muntaner et al., 2002; Navarro et al., 2003, 2006, Rodríguez-Sanz et al., 2003; Ronzio,

Pamuk, & Squires, 2004). A pro forma is used to extract the following information: year of publication, sample size, study period, level and unit of analysis, data analysis procedures, political variables, health outcomes, and statistical findings. To broadly

41 summarize the statistical findings, a total of 1,614 independent study outcomes (welfare generosity = 639; welfare regimes = 117; leftist politics = 859) are extracted and coded a positive (e.g., greater levels of welfare spending, egalitarian welfare regimes, and strong leftist politics are associated with better health outcomes), negative (e.g., increases in welfare generosity, egalitarian welfare regimes, and leftist politics are associated with negative health outcomes), or no association (e.g., relationship between political variable and population health is non-significant). All data is entered in Stata/SE

12 (StataCorp., 2011) and summarized using descriptive and cross-tabulated statistics.

Results

Major study design characteristics of the 46 included studies are displayed in Table 2.2

(located at end of chapter). A total of 55 political themes are reviewed, with more than half of the themes devoted to welfare generosity (n = 29, 52.6%), followed by leftist politics (n = 14, 25.5%) and welfare regimes (n = 12, 21.8%). The majority of politically oriented studies have been published since 2000 (n = 41, 89.2%), used ecological designs (n = 37, 80.4%), sampled OECD countries (n = 35, 76.1%), and made comparisons between countries or welfare regimes (n = 33, 71.7%). The number of units compared primarily involve 10-19 (n = 19, 41.3%) or over 30 (n = 22, 47.8%). In terms of methodological rigor, most included studies have used some form of statistical modeling to control for alternative explanations (n = 36, 78.3%). With respect to health outcomes, 373 of the 1656 independent tests (22.5%) focused on all-cause mortality rates, 238 (14.4%) on life expectancy, 172 (10.4) on child health outcomes, 190 (11.5%) on unintentional injuries, and 375 (22.6%) on ‘other’ dependent variables (see Tables

42 2.3, 2.4, and 2.5 for other health outcomes tested). Of the 1,656 independent tests, 489

(29.5%) found a positive association between politics and health, 106 (6.4%) had a negative impact, and 1061 (64.1%) are non-significant. It is worth noting that close to half of the independent tests (793 out of 1,656, 47.9%) are extracted from a single exploratory study conducted by Muntaner and colleagues (2002).

Welfare Generosity

Studies concerned with the impact of welfare generosity on population health are presented in Table 2.3 (located at end of chapter). Existing studies focused on three major types of government expenditures. First, studies have examined the health effects of total expenditures, which accounts for all forms of government spending

(Dunn, Burgess, & Ross, 2005; Lena & London, 1993; Xu, 2006; Veenhoven &

Ouweneel, 1995). Second, studies have tested the impact of health expenditures, conceptualized as spending on total health (Arah & Westert, 2005; Arah, Westert,

Delnoij, & Klazinga, 2005; Bradley et al., 2011; Elgar, 2010; Fayissa, 2001; Filmer &

Pritchett, 1999; Laporte & Ferguson, 2003; Navarro et al., 2003, 2006; Novignon,

Olakojo, & Nonvignon, 2012; Rodríguez-Sanz et al., 2003; Schell, Reilly, Rosling,

Peteron, & Ekström, 2007), healthcare coverage (Arah et al., 2005; Chung & Muntaner,

2006; Navarro et al., 2003, 2006), hospitals (Dunn et al., 2005), and pharmaceuticals

(Liu, Cline, Schondelmeyer, & Schommer, 2008). Third, welfare generosity studies also investigated the association between population health and social expenditures, expressed as total social spending (Bradley et al., 2011, Dahl & van der Wel, 2013;

Kangas, 2010; Muntaner et al., 2002; Navarro et al., 2003, 2006), social security

43 transfers (Chung & Muntaner, 2006; Veenhoven & Ouweneel, 1995; Veenhoven, 2000), public services (Dunn et al., 2005; Ronzio, 2003; Ronzio et al., 2004); family benefits

(Ferrarini & Norström, 2010; Lundberg et al., 2008), income replacements (Kangas,

2010; Lundberg et al., 2008; Norström & Palme, 2010; Rodríguez-Sanz et al., 2003), coverage of insurance schemes (Kangas, 2010), redistributive systems of the state

(Muntaner et al., 2002), and active labour market policies (Stuckler, Basu, Suhrcke,

Coutts, & McKee, 2009).

Overall, these studies provide strong evidence that government expenditures are associated with improved population health outcomes (24 out of the 29 studies, 82.8%, and 216 out of the 639 of the independent hypotheses, 33.8%, found positive associations). Positive associations are consistently observed across four different contexts: 1) OECD countries (Arah et al., 2005; Bradley et al., 2011; Chung &

Muntaner, 2006; Conley & Springer, 2001; Dahl & van der Wel, 2013; Kangas, 2010;

Liu et al., 2008; Lundberg et al., 2008; Stuckler et al., 2009); 2) Sub-Saharan African nations (Fayissa, 2001; Novignon et al., 2012); 3) global nations, which included low-, middle-, and high-income countries (Schell et al., 2007); and 4) political jurisdictions within a single nation (Dunn et al., 2005; Laporte & Ferguson, 2003; Rodríguez-Sanz et al., 2003; Ronzio, 2003; Xu, 2006).

Supportive evidence for welfare generosity hypotheses are observed most often between total health spending and all-cause mortality rates (Arah et al., 2005; Laporte &

Ferguson, 2003), life expectancy (Bradley et al., 2011; Novignon et al., 2012), potential years of lost life (Arah et al., 2005; Bradley et al., 2011), child health outcomes such as infant mortality and low birth weight (Conley & Springer, 2001; Fayissa, 2001; Novignon

44 et al., 2012; Schell et al., 2007). Health care coverage, measured as the percentage of the population with access to medical care, had a significant impact with improving potential years of lost life (Arah et al., 2005) and lowering all-cause mortality (Arah et al., 2005) and child health outcomes (Chung & Muntaner, 2006). In terms of specific types of health expenditures, Liu and colleagues (2008) find beneficial gains in life expectancy at ages 65 and 80 as well as potential years of lost life are associated with increased pharmaceutical expenditures among OECD nations.

Studies testing the health effects of general social expenditures found positive associations with life expectancy (Bradley et al., 2011; Kangas, 2010), maternal mortality (Bradley et al., 2011), potential years of lost life (Bradley et al., 2011), and educational inequalities in self-rated health (Dahl & van der Wel, 2013). Empirical evidence also confirmed the health effects of disaggregated measures of social spending, including for example, income replacement levels, basic pensions, and social insurance coverage extended life expectancy (Kangas, 2010); the generosity of family policies and basic security reduced infant mortality and old-age excess mortality rates, respectively (Lundberg et al., 2008); and public spending on active labour policies protected unemployed individuals from suicide (Stuckler et al. 2009) . Two studies conducted among US states find that all-cause mortality rates are strongly related to public service expenditures by state and local governments, and in particular, spending on education (Dunn et al., 2005) and roads (Ronzio, 2003).

Five out of the 29 studies failed to find even a single positive association between government spending and population health (Elgar, 2010; Filmer & Pritchett,

1999; Lena & London, 1993; Veenhoven & Ouweneel, 1995; Veenhoven, 2000). The

45 reason for non-supportive findings might be due to two reasons. First, the impact of welfare generosity on population health is not an immutable parameter, and is likely to vary widely from country to country, and hence the results, will be sensitive to the sample used (Filmer & Pritchett, 1999). All five of these studies used a global sample of un-stratified countries, comprised of some combination of low- or middle-, and high- income countries, with varying final samples sizes (N = 33, Elgar, 2010; N = 98-119,

Filmer & Pritchett, 1999; N = 50-84, Lena & London, 1993; N = 97, Veenhoven &

Ouweneel, 1995; N = 22-35, Veenhoven, 2000). This suggests that government expenditures are likely to have a differential impact on population health across contexts and countries (e.g., a dollar spent in a poor country for example may improve health while in another context, a non-poor nation, the same dollar may create an expensive service with no positive health effect) (Filmer, Hammer, & Pritchett, 1998).

Second, the links between government spending and population health may be affected by differing definitions of welfare generosity and data analytical methods. Two of the five studies measured welfare generosity as central government expenditures, which might be too broad to capture the health-promoting features of welfare states

(Lena & London, 1993; Veenhoven & Ouweneel, 1995). Studies finding strong and positive associations are more likely to use aggregate indicators such as health or social expenditures (Bradley et al., 2011), or disaggregated measures, including public education dollars (Dunn et al., 2005), family benefit polices (Lundberg et al., 2008), and pharmaceutical spending (Liu et al., 2008). Third, non-positive findings might be due to the fact that these studies used analytical methods that produced over-determined results (e.g., running correlations and regressions on a limited number of cross-

46 sectional observations, also known as the small N problem). None of the four studies that are conducted at the ecological level took advantage of TSCS methods (e.g., fixed- effects models) to increase sample size and control for unobserved-variable bias (OLS and two-stage least squares regressions, Filmer & Pritchett, 1999; OLS regression,

Lena & London, 1993; correlation, Veenhoven & Ouweneel, 1995; correlation,

Veenhoven, 2000).

Welfare Regimes

Table 2.4 shows 12 studies that tested hypotheses regarding the impact of welfare regimes on population health (located at end of chapter). Guided by the welfare and institutional typologies of Esping-Andersen (1990, 1999), Huber and Stephens (2001),

Ferrera (1996), and Wood and Gough (2006), welfare regime researchers have classified OECD and non-OECD countries into several typologies, which include tripartite (Bambra, 2005; Conley & Springer, 2001; Kangas, 2010), four-category

(Chung & Muntaner, 2007; Richter et al., 2012), five-category (Eikemo et al., 2008a;

Rostila, 2007; van der Wel, Dahl, & Thielen, 2012), six-category (Karim, Eikemo, &

Bambra, 2010), and seven-category (Chung et al., 2013; Witvliet & al., 2011; Witvliet et al., 2012) welfare state models.

Eleven studies demonstrate significant health differences between welfare regimes, and only one failed to do so (Kangas, 2010). Overall, 47 out of the 117

(40.2%) independent hypotheses yielded positive associations. Welfare regime studies can be further divided into three sub-themes: 1) ecological, 2) multi-level, and 3) global.

First, studies using ecological designs find general support that social democratic

47 nations are the most advantaged regime with respect to infant mortality (Bambra, 2005;

Chung & Muntaner, 2007; Karim et al., 2010) and low birth weight (Chung & Muntaner,

2007). Interestingly, the effects of welfare regimes on child health appear to be modified by levels of welfare generosity. For example, in social democratic and liberal regimes, within-country changes in per capita health spending do not appear to be significantly related to infant mortality and low birth weight, however, the health effects of health spending are significant and strong for corporatist countries (Conley & Springer, 2001).

The inclusion of different regimes and choice of health outcomes appear to modify hypotheses based on welfare regime theory. For instance, Karim, Eikemo, and Bambra

(2010) conceptualize Japan, Korea, Hong Kong, Singapore, and Taiwan as a distinct

East Asian regime, conduct one-way ANOVA analyses along with European regimes

(e.g., Bismarckian, Scandinavian, Anglo-Saxon, Southern European, and Eastern

European), and find a significant effect of welfare regimes on life expectancy (F(5,24) =

6.42, p = 0.000).

Second, studies using multi-level designs with individual-level data find partial support for hypotheses based welfare regime theory (Eikemo et al. 2008a; Chung et al.,

2013; Conley & Springer, 2001; Richter, 2012; Rostila, 2007; van der Wel et al., 2012;

Witvliet et al. 2011, 2012). Supportive evidence includes van der Wel and colleagues’

(2012) finding that non-employment rates are lowest (most advantaged) among people reporting limiting longstanding illness in Scandinavian countries. However, the remaining studies fail to find similar health advantages for Nordic nations (Eikemo et al.

2008; Rostila, 2007; Richter, 2012). At the same time, none of these studies found that the social democratic regime to be the most disadvantaged health-wise. In fact, the

48 comparative health literature has come to the general consensus that liberal/ residual/minimalist welfare regimes, perhaps due to their strong ideological preference for marketization and decentralization, are associated with poorer health outcomes.

Depending on which regimes are included, the Anglo-Saxon/liberal or Eastern

European/Post-Socialist regimes are significantly and consistently associated with worse health outcomes, including self-rated health (Eikemo et al. 2008a, Rostila, 2007), health complaints (Richter et al., 2012), and life expectancy (Rostila, 2007).

Third, significant health differences are also observed among the three studies using a global approach to welfare regimes (Chung et al., 2013; Witvliet et al. 2011;

Witvliet et al. 2012). Chung and colleagues (2013) find that the odds of experiencing a brief depressive episode in the last 12 months are significantly higher for Southern/Ex- dictatorship countries than for Southeast Asian (odds ratio (OR) = 0.12) and Eastern

European (OR = 0.36) regimes even after controlling for gender, age, education, marital status, and economic development. Using Wood and Gough’s global typology of welfare regimes, Witvliet (2011) find that informal-security (South Asia) and insecure regimes

(Sub-Saharan Africa) have significantly higher levels of disability compared to

European-conservative nations. In another study, these same authors (2012) compare the prevalence of disability of low educated individuals in conservative and informal- security regimes, and find that the odds of disability are three-times higher (OR = 3.16) among the latter regime (South Asia). The possible explanations for counterintuitive findings, or why social democratic countries are most advantaged across absolute and relative health outcomes, are the subject of continued debate (Bambra, 2011;

Brennenstuhl et al., 2012; Muntaner et al., 2011, 2012; Popham et al., 2013).

Leftist Politics

Table 2.5 shows the 14 studies that tested hypotheses regarding the health impact of leftist politics (located at end of chapter). Population health studies focused on voting dynamics are conceptualized as voter turnout (Chung & Muntaner, 2006; Muntaner et al., 2002; Navarro et al., 2003, 2006; Reitan, 2003), voting partisanship (Chung &

Muntaner, 2006; Kelleher, Timoney, Friel, & McKeown, 2002; Muntaner et al., 2002;

Navarro et al., 2003, 2006; Smith & Dorling, 1996, 1997), and socioeconomic inequality in voter turnout (Blakely, Kennedy, & Kawachi, 2001). Eight studies tested the impact of power resources on population health - two on union density (Lynch et al., 2001;

Muntaner et al., 2002), and six on left political politics (Lena & London, 1993; Muntaner et al., 2002; Navarro et al., 2003, 2006; Rodríguez-Sanz, 2003; Ronzio et al., 2004).

Three studies focused on the association between women in government and population health (Kawachi et al., 1999; Lynch et al., 2001; Muntaner et al., 2002).

Compared to the other two political themes, independent hypotheses between leftist politics and health had the smallest percentage of positive associations (226 out of 900, 25.1%). Studies on voter turnout and population health produced mixed results.

Higher levels of voter turnout in OECD countries are associated lower rates of low birth weight (Muntaner et al. 2002), infant deaths from all causes (Muntaner et al. 2002), and infant mortality during the 1970s (Navarro et al., 2003, 2006). In Russia, voter turnout is positively and significantly linked to improvements in male and female life expectancy during the 1990s (Reitan, 2003). On the other hand, negative associations are observed between voter turnout and infant mortality and under-5 mortality rate (Chung &

50 Muntaner, 2006). Among US states, socioeconomic inequalities in voter turnout are associated with poor self-rated health, independent of income inequality and household income (Blakely et al., 2001).

Investigations on voting partisanship compared OECD countries (Chung &

Muntaner, 2006; Muntaner et al., 2002; Navarro et al., 2003, 2006) or political jurisdictions within a single country (Kelleher et al., 2002; Smith & Dorling, 1996, 1997).

Among OECD countries, the percentage of vote obtained by left parties is associated with lower infant mortality rates (Chung & Muntaner, 2006; Navarro et al., 2003, 2006), low birth weight, and infant deaths from all causes (Muntaner et al., 2002). Three studies investigated the association between voting partisan patterns and health

(Kelleher et al., 2002; Smith & Dorling, 1996, 1997). British general elections in the

1980s and 1990s found negative associations between voting Conservative or Liberal

Democrat and mortality, and positive links between voting Labour or abstention and mortality (Smith & Dorling, 1996, 1997). In Ireland, similar patterns between voting and health are observed; increases in left wing voting are significantly related with increases in mortality (r = 0.45, p < 0.05) and smoking rates (r = 0.47, p < 0.05) (Kelleher et al.,

2002).

In terms of power resources theory, countries with greater trade union membership had significantly lower rates of infant mortality (female, r = -0.56, p < 0.04; male, r = -0·58, p < 0.04), low birth weight (both sexes, r = -0.57, p < 0.05), unintentional injuries less than 1 year of age (female, r = -0.59, p < 0.03; male, r = -0.64, p < 0.02), and mortality among males aged 1–14 years (r = -0.57, p < 0.04) (Lynch et al., 2001). For the most part, Muntaner and colleagues (2002) replicated these positive

51 findings, and further demonstrated the health benefits of increased union density to include lower rates of infectious disease (female, r = , -0.72, p < 0.01; male, r = -0.75, p

< 0.01) and homicide (female, r = , -0.66, p < 0.05; male, r = -0.56, p < 0.05). At the same time, these authors also find negative results; increases in labour union power are associated with increases in mortality rates among males over 65 (r = 0.57, p < 0.05) and strokes among males (r = 0.57, p < 0.05). As for left political parties, the second major component of power resources, existing studies find that left parties are significantly related to lower rates of infant mortality (Muntaner et al., 2002; Navarro et al., 2003, 2006), mortality for various age groups (Muntaner et al., 2002), and premature mortality (Rodríguez-Sanz et al., 2003) as well as fewer deaths from infectious diseases

(Muntaner et al., 2002) and longer life expectancies (Lena & London, 1993; Navarro et al., 2003, 2006). One study among US cities found no evidence that left political parties, measured as Democratic mayors, improved health outcomes (Ronzio et al., 2004).

Population health studies concerned with non-left political parties find preliminary evidence that Christian Democratic governments have mixed effects on health (e.g., reductions in lung cancer among women, r = -0.59, p < 0.05; and increases in self-rated poor health among both sexes, r = 0.66, p < 0.05, Muntaner et al., 2002) and that strong right-wing regimes have populations with lower life expectancies and higher levels of various indicators of mortality (Lena & London, 1993).

The cross-national link between women in government and population health echoed the associations between labour unions and health. Countries that had greater political representation by women also had better child mortality profiles and lower unintentional injury deaths, especially among the young (Lynch et al., 2001; Muntaner et

52 al., 2002). In the one study conducted among US states, women’s representation in elected office, measured at the state-level as representatives, senators, and governors and weighted according to political influence, is inversely and significantly related to female mortality (r = -0.38, p < 0.05), male mortality (r = -0.55, p < 0.05), black female mortality (r = -0.33, p < 0.05), and mean days of activity limitations among females (r = -

0.34, p < 0.05) and males (r = -0.35, p < 0.05) (Kawachi et al., 1999).

Discussion and Research Needs

The preceding literature review has summarized the theoretical and empirical contexts that inform this dissertation’s research questions and hypotheses. Based on a careful reading of the extant literature, five conclusions are warranted.

First, political sociology theories that emphasize welfare states and leftist politics have much to offer prevailing approaches in social epidemiology (O’Campo & Dunn,

2011). Respectively, materialist, neo-materialist, and psychosocial explanations of population health are compelling accounts on the importance of living conditions, public infrastructures, and psychosocial comparisons as SDOH. The explanatory power of these population health explanations are bolstered given their theoretical associations with social stratification, Weber’s ‘life chances’, and Durkheim’s ‘social facts’. With this said, social epidemiological approaches to population health runs the real risk of becoming repetitive and inconsequential if the substantive foci remains dedicated to replicating social gradients in health (e.g., the wealthier are healthier) and testing income inequality hypotheses (e.g., high trust begets good health). In contrast, political sociology theories provide critical ideas (e.g., Neo-Marxism) and macro-level insights on

53 the political forces (e.g., leftist politics) and institutions (e.g., welfare states) that influence the degree of social inequalities and the availability and quality of SDOH, which in turn, affect population of health. Current population health research in social epidemiology and political sociology are certainly related, and have begun to cross- fertilize to produce new and exciting results (Beckfield & Krieger, 2009; Navarro et al.,

2006; Chung & Muntaner, 2007; McLeod et al., 2012a, 2012b; Stuckler, King, & McKee,

2009). This dissertation aims to contribute to this small but growing body of politically oriented scholarship (see conceptual model in chapter 3).

Second, the majority of existing studies on welfare generosity, leftist politics, and population health have been conducted at the national level either between countries or regimes. As revealed in the systematic review, only 13 out of the 46 studies (28.3%) investigated hypotheses regarding the impact of politics on health within a single nation

(welfare generosity: Arah & Westert, 2005; Laporte & Ferguson; Dunn et al., 2005;

Ronzio, 2003; Ronzio et al., 2004; Xu, 2006; Rodríguez-Sanz et al., 2003; leftist politics:

Blakely et al., 2001; Kawachi et al. 1999, Smith & Dorling, 1996, 1997; Kelleher et al.,

2002; Reitan, 2003; Rodríguez-Sanz et al., 2003). Because most comparative studies have developed and tested theories to understand cross-national or regime differences in health (Navarro et al., 2003, 2006), more theoretical and empirical work is needed to determine whether the major arguments of welfare generosity, leftist politics, and welfare regimes are also applicable within sub-national contexts. To date, only two studies, both of which tested welfare generosity as the independent variable, have examined the association between politics and health among Canadian provinces (Arah

& Westert, 2005; Laporte & Ferguson, 2003). This dissertation contributes to the extant

54 literature by testing the overall explanatory power of welfare generosity, leftist politics, and welfare regime theories in relation to population health among Canadian provinces.

Third, the dominant approach to measuring welfare generosity can be refined in ways to demonstrate a more nuanced understanding between spending and health. The conventional approach is to rely on aggregate spending measures such as ‘total expenditures’, ‘health expenditures’, or ‘social spending’. Studies using aggregate measures find general support for the idea that “bigger is better” (Kangas, 2010), or in other words, as health or social spending increases, population health outcomes also improve (Brennenstuhl et al., 2012). On one hand, this body of evidence has been instrumental in demonstrating the overall importance of public spending for the public’s health. On the other, the consequence of relying on aggregate spending measures is that specific features of the welfare state are often overlooked and unaccounted for. For example, Bradley and colleagues (2011) test variations in social services expenditures across OECD countries, assess their associations with multiple population health outcomes, and find that social expenditures are positively and significantly related to life expectancy, infant mortality, and maternal mortality. While this finding is informative and encouraging, the determination that social services expenditures are significant predictors of health represents the very beginning of this line of inquiry. More research is needed to understand which aspects of aggregate spending measures exert the greatest, least, or no effect at all on health. In the study undertaken by Bradley et al.

(2011), social services expenditures can be further disaggregated into specific priorities such as old-age pensions, survivors’ benefits, disability and sickness cash benefits, family support, employment programs, unemployment benefits, and housing support.

55 Given that only 10 out of the 29 (34.5%) welfare generosity studies in the systematic review tested the health effects of disaggregated expenditures (Dunn et al., 2005;

Ferrarini & Norstrom, 2010; Kangas, 2010; Liu et al., 2008; Lundberg et al., 2008;

Norstrom & Palme, 2010; Rodríguez-Sanz et al., 2003; Ronzio, 2003; Ronzio et al.,

2004; Stuckler et al., 2009), this dissertation complements these studies by testing the predictive value of both aggregate and disaggregated expenditures. Moreover, this dissertation will be the first to test the impact of disaggregated measures among

Canadian provinces. The two previous welfare generosity studies in Canada both relied on aggregate indicators (e.g., total health expenditure (per capita) and public sector health expenditure (per capita), Arah & Westert, 2005; and total health spending (per capita, Laporte & Ferguson, 2003).

Fourth, the welfare regime literature on the political context of population health has primarily focused on how national welfare states cluster into different institutional regimes, which are qualitatively different and not necessarily comparable. Common to these works is the presumption that countries cluster into distinct regimes that reflect the state’s historical willingness to intervene in market affairs and to provide welfare services and social transfers (Bambra, 2005; Chung et al., 2013). Because research on welfare typologies and population health has been so productive, this dissertation can benefit from this literature’s insights by testing ‘welfare regime theory’ within Canada and its provinces. As revealed in the systematic review, no studies in the comparative literature have tested the hypothesis that sub-national jurisdictions cluster together in meaningful ways that are associated with population health (see Table 2.4). The closest example is Blakely et al.’s (2001) decision to group US states by average voter turnout

56 into four distinct groups: high (7 states), medium-high (19 states), medium-low (15 states), and low (9 states), and to examine voter turnout at the state-level with individual self-rated health. Compared to states with high levels of voter turnout (reference category), individuals living in states with low levels of voter turnout had significantly higher odds of reporting fair/poor health (OR = 1.62, CI: 1.39, 1.90) even after adjusting for income inequality and median household income (Blakely et al., 2001). This dissertation makes a new contribution on this front by exploring whether Canadian provinces cluster together based on leftist political factors and by testing whether these regimes are predictive of population health (see chapter 7).

Fifth, the methodological rigor of existing politically oriented health studies has been uneven, and falls into three levels of evidential strength. First, the most robust and persuasive evidence has been produced with methods first developed in political science and international relations that overcome the classic ‘small N’ problem (e.g., limited number of countries, large number of explanatory variables) with TSCS designs

(Beck, 2001; Beck & Katz, 1995, 1996, 2001; Plümper et al., 2005; Podestà, 2002,

2006). By arraying cross-sectional data on N country-units and T time periods, these studies have been able to produce a strong body of evidence given their increased statistical power (e.g., sample sizes based on N x T) and increased data modeling options (e.g., capturing space and time variations simultaneously) (Bradley et al., 2011;

Chung & Muntaner, 2006, 2007; Conley & Springer, 2001; Ferrarini & Norstrom, 2010;

Kangas, 2010; Laporte & Ferguson, 2003; Liu et al., 2008; Lundberg et al., 2008;

Norstrom & Palme, 2010; Novignon et al., 2012; Rodríguez-Sanz et al., 2003; Stuckler et al., 2009). The second level of evidential strength includes studies using cross-

57 sectional regressions. Even though these studies are able to control for one or more independent variables, the potential sample size of these studies represents an important limitation (e.g., the number of countries is finite). As a result, researchers often find themselves running regressions on observations ranging from a low of 30 countries (Karim et al., 2010) to a high of 152 (Schell et al., 2007). Among studies using sub-national jurisdictions, the number of observations is higher and range from 48 US states (Dunn et al., 2005) to 561 electoral constituencies in England and Wales (Smith

& Dorling, 1996, 1997). Given that it is often not possible to add more political jurisdictions to increase the overall sample size, cross-sectional regressions are vulnerable to a small number of influential observations such as the US over- determining the results.

The third stream of studies, and weakest, in terms of methodological rigor include those using exploratory methods such as univariate and Pearson correlation coefficients

(Arah & Westert, 2005; Kelleher et al., 2002; Lynch et al., 2001; Muntaner et al., 2002;

Navarro et al., 2003, 2006). For example, studies investigating the differential effects of power resources on population health have primarily relied on bivariate associations to establish the link, thus leaving alternative explanations unaccounted for and raising concerns about reverse causality. More rigorous methods are needed, especially among leftist political studies, to confirm or refute the associations found by Navarro et al. (2003, 2006) and Rodríguez-Sanz et al. (2003). To increase the methodological rigor of the extant literature, this dissertation constructs a comparative database among

Canadian provinces that ranges from 1976 to 2008, allowing a maximum of 330 observations, and uses TSCS and hierarchical cluster data analysis methods.

58 In the next chapter, I present the conceptual model that guides this dissertation and further elaborates on the contributions that this dissertation adds to our current knowledge base.

59 Table 2.1. Conceptualizing Politics as a Macro-social Determinant of Population Health: Variables, Claims, and Hypotheses Political Variable Central Claim(s) Population Health Hypotheses 1. Welfare generosity Welfare generosity reflects the degree to which welfare states provide 1.1. Greater levels of welfare generosity should be key welfare services (health care, education, social services) and social associated with improved population health outcomes. transfers (income maintenance programs). 1.2. Lower levels of welfare spending should be correlated with disadvantaged health outcomes. 2. Welfare regimes Welfare regimes are systems of stratification that decommodifies labour 2.1. If social democratic welfare states decommodify labour and reinforces other social inequalities; welfare states cluster into to the greatest extent, social inequalities should be less distinct regimes (Social Democratic, Conservative, Liberal, Southern). pronounced and levels of population health should be favourable compared to conservative and liberal regimes. 2.2. If liberal and southern welfare states decommodify labour the least, social inequalities should be most pronounced and population health outcomes should be disadvantaged compared to the social democratic regime. 3. Power resources Trade unions and left political parties are power resources that tend to 3.1. Strong trade unions should improve the health of favour stronger welfare states and more egalitarian outcomes. workers through workplace safety measures. 3.2. Strong trade unions should improve population health by supporting left political parties, which in turn, increase welfare generosity. 3.3. If left political parties contribute to more egalitarian outcomes, then the cumulative power of left political powers should be positively associated with welfare generosity and population health outcomes. 4. Partisan Politics Left and centre-left political parties are the central determinants for 4.1. If left and centre-left political parties contribute to more welfare state development and translate class- and ascription-based egalitarian outcomes, then the cumulative power of left and social cleavages into social policy. centre-left political parties should be positively associated with welfare generosity and population health. 4.2. Conversely, if right political parties contribute to higher levels of social inequality, then the cumulative power of right-wing parties should be negatively associated with welfare generosity and population health.

(Continued on next page)

60 Table 2.1. Conceptualizing Politics as a Macro-social Determinant of Population Health: Variables, Claims, and Hypotheses (Continued) 5. Voter turnout High voter turnout typically changes the class content of elections, 5.1. Higher levels of voter turnout should be associated with shifting the political centre of gravity to the left; and reflects increases in improved population health outcomes through greater participation by previously excluded lower status groups. support for left political parties, which in turn, increase welfare generosity. 5.2. Low voter turnout increases the likelihood of right party incumbency which is associated with lower population health outcomes. 6. Women in Greater levels of female representation in legislative assemblies 6.1. The political mobilization and representation of women government advance the interests of children and women and increase levels of in institutionalized politics should be associated with welfare generosity. improved health outcomes. 6.2. Conversely, the lack of women in government should be associated with poorer population health.

61 Table 2.2. Summary Characteristics of Empirical Studies on Welfare Generosity, Welfare Regimes, and Leftist Politics (N = 46) N % Political Themea Welfare Generosity 29 52.7 Welfare Regimes 12 21.8 Leftist Politics 14 25.5 Year of Publication 1990-99 5 10.9 2000-09 28 60.9 2010-13 13 28.3 Level of Analysis Ecological 37 19.6 Multi-Level 9 80.4 Number of Units <9 1 2.3 10-19 19 41.3 20-29 4 8.7 >30 22 47.8 Sample OECD 35 76.1 Non-OECD 2 4.4 Global 9 19.6 Comparison Groups Between countries/welfare regimes 33 71.7 Within-country 13 28.3 Data Analysis Bivariate 10 21.7 Multivariate 36 78.3 Health Outcomesb Cancer 121 7.3 Infant mortality/low birth weight 172 10.4 Life expectancy 238 14.4 Mental health 60 3.6 Mortality rate 373 22.5 Self-rated health 127 7.7 Unintentional injuries 190 11.5 Other 375 22.6 Statistical Findings Positivec 489 29.5 Negatived 106 6.4 No associatione 1061 64.1 Note: OECD = Organization for Economic Cooperation and Development. a Nine articles conducted analyses on more than one political theme, and hence, the number of political themes (n = 55) exceeds the number of reviewed studies (n = 46). b Given that most studies investigated multiple health outcomes, the number of health outcomes represents the total number of independent tests between politics and health (n = 1,656). c Political theme exerts a positive, direct, or indirect effect on population health outcome. d Political theme exerts a negative, direct, or indirect effect on population health outcome. e Political theme is unrelated to population health outcome.

Table 2.3. Empirical Studies on Welfare Generosity and Population Health (n = 29) Author(s)/ Year N, Period, Political Variable(s) Health Outcome(s) Findings Level, Data Analysis # + - ≠ Arah & Westert -N = 13 Provinces and territories -Total health expenditure (per capita) -Self-rated health 12 3 2 7 (2005) in Canada -Public sector health expenditure (per capita) -Body mass index -2001 -Asthma rate -Ecological -Diabetes rate -Univariate correlations -Functional health -Life expectancy at birth Arah et al. -N = 18 OECD countries -Collective health expenditure (% of GDP) -All-cause mortality 4 4 0 0 (2005) -1970-1999 -Healthcare coverage (% of population) -Potential years of life lost -Ecological -Fixed effects regression Bradley et al. -N = 30 OECD countries -Health expenditures (% of GDP) -Life expectancy 15 8 2 5 (2011) -1995-2005 -Social expenditures (% of GDP) -Maternal mortality -Ecological -Ratio of social to health expenditures (% of GDP) -Infant mortality -Mixed effects regression -Maternal mortality -Low birth weight

Chung & -N = 19 OECD countries -Social security transfers (% of GDP) -Infant mortality rate 12 6 2 4 Muntaner (2006) -1960-1994 -Total public medical care (% of population with -Under 5 mortality rate -Ecological total medical coverage) -Low birth weight -Robust-cluster variance estimator Conley & -N = 19 OECD countries -Health care spending (per capita) -Infant mortality rate 8 4 0 4 Springer -1960-1982 -Low birth weight (2001) -Ecological -Fixed effects regression

(Continued on next page)

Table 2.3. Empirical Studies on Welfare Generosity and Population Health (n = 29) (Continued) Dahl & van der -N = 244,875 individuals -N = 18 -Gross public social expenditure (% of GDP & -Educational inequalities in self-rated 24 11 0 13 Wel (2013) European countries PPP) health* -2005 -Net total social expenditure (% of GDP & PPP) -Multi-level -Logistic multilevel random intercept Dunn et al. -N = 48 US states -Total expenditures/services (per capita) -All age mortality* 33 19 0 14 (2005) -1987 -Primary/secondary education (per capita) -Working-age mortality* -Ecological -Higher education (per capita) -OLS regression -Public welfare (per capita) -Health and hospitals (per capita) -Highways (per capita) -Police and corrections (per capita) -Environment & housing (per capita) -General government administration (per capita) -All other expenditures (per capita) Elgar (2010) -N = 33 global countries -Public health expenditure (% of GDP) -Life expectancy 2 0 0 2 -2005-2008 -Adult mortality -Ecological -Two-level linear regression Fayissa (2001) -N = 34 Sub-Saharan African -Government expenditures on health and nutrition -Infant mortality rate 2 2 0 0 countries (% of health expenditures) -Child mortality rate -1993 -Ecological -Two stage least squares (Continued on next page)

Table 2.3. Empirical Studies on Welfare Generosity and Population Health (n = 29) (Continued) Ferrarini & -N = 18 OECD countries -Family benefits: earning-related parental -Infant mortality rate 3 1 0 2 Norstrom -1950-2000 insurance, child leave benefits, child (2010) -Ecological allowances, maternity grants, and tax -Antoregressive integrated reductions for worker with a dependent moving average spouse Filmer & -N = 98-119 -Public expenditures on health (% of GDP) -Under-5 mortality rate 2 0 0 2 Pritchett (1999) -1990 -Under-1 mortality rate -Ecological -OLS regression & Two stage least squares Kangas (2010) -N = 17 OECD countries -Social spending (% of GDP) -Life expectancy at birth. 10 4 0 6 -1900-2000 -Generosity (income replacement levels through -Changes in life expectancy at birth - Ecological major social insurance schemes) -Prais-Winsten -Universalism (proportion of the population regressions and first-order covered under insurance schemes) differences -Basic pensions Laporte & -N = 9 provinces in Canada -Total health spending (per capita) -Total mortality rate 1 1 0 0 Ferguson (2003) -1980-1997 -Ecological -OLS and autoregressive distributed lag Lena & London -N = 50-84 -Central government expenditures (% of GNP) -Infant mortality 7 0 1 6 (1993) -1965-1970, 1983 -Life expectancy -Ecological -Child death rate -OLS regression

(Continued on next page)

Table 2.3. Empirical Studies on Welfare Generosity and Population Health (n = 29) (Continued) Liu et al. (2008) -N = 14 OECD countries -Expenditures on pharmaceuticals (per capita -Life expectancy at age 65* 6 3 0 3 -1985-2001 spending) -Life expectancy at age 80* -Ecological -Potential years of life lost* -Fixed effects regression Lundberg et al. -N = 18 OECD countries -Family policy generosity -Infant mortality rate 7 3 0 4 (2008) -1950-2000 -Dual earner support policies -Old-age excess mortality* -Ecological -General family support -Fixed effects regression -Basic security -Income security Muntaner et al. -N = 12-15 OECD countries -Social security expenditures (% of GDP) -Total mortality rate* 288 88 11 189 (2002) -1960-1990 -Redistributive effect of state (% of reduction in -Cause-specific mortality rate* -Ecological income inequality due to direct taxation and -Suicide* -Correlation transfers) -Homicide* -Total public medical care, 1960, 1970, 1980, -Life expectancy* 1990 (% of the population with medical care -Low birth weight* provided by public funds) -Self-rated health* Navarro et al. -N = 17 OECD countries -Public health expenditure (% of GDP) -Infant mortality rate 36 10 0 26 (2003) -1950-1998 -Public healthcare coverage (% of population) -Life expectancy at birth* -Ecological -Public social protection expenditures (% of GDP) -Correlation Navarro et al. -N = 17 OECD countries -Total public health expenditures (% of GDP) -Infant mortality rate 90 19 0 71 (2006) -1972-1996 -Public health coverage (% of population) -Life expectancy at birth* -Ecological -Correlation

(Continued on next page)

Table 2.3. Empirical Studies on Welfare Generosity and Population Health (n = 29) (Continued) Norstrom & -N = 18 OECD countries -Pension benefits (basic security index & income -Old-age excess mortality rate* 9 2 0 7 Palme (2010) -1950-2000 security index) -Ecological -Autoregressive integrated moving average Novignon et al. -N = 44 Sub-Saharan countries -Total health expenditure (% of GDP) -Life expectancy 9 9 0 0 (2012) -1995-2010 -Public health expenditure (% of GDP) -Death rate -Ecological -Private health expenditure (% of GDP) -Infant mortality rate -Random and fixed effects Rodríguez-Sanz -N = 17 communities in Spain -Public health expenditure (per inhabitant) -Total premature mortality* 30 7 6 17 et al. (2003) -1989-1998 -Income support (% of households) -Premature mortality by -Multi-level cerebrovascular disease*, cirrhosis*, -Hierarchical linear regression ischemic heart disease*, lung cancer* Ronzio (2003) -N = 138 US cities -Police expenditure (per capita) -Premature mortality 2 2 0 2 -1980 & 1990 -Roads expenditure (per capita) -Ecological -Generalized linear regression Ronzio et al. -N = 141 US cities -Social support expenditure (per capita) -All-cause death rate 8 3 2 3 (2004) -1980 & 1990 -Police expenditure (per capita) -Preventable death rate -Ecological -Roads expenditure (per capita) -OLS regression -Fire expenditure (per capita) -Welfare expenditure (per capita)

(Continued on next page)

67 Table 2.3. Empirical Studies on Welfare Generosity and Population Health (n = 29) (Continued) Schell et al. -N = 152 global countries -Public spending on health (per capita) -Infant mortality rate, stratified by all, 4 4 0 0 (2007) -2003 high-, middle-, and low-income -Ecological countries -Linear regression Stuckler et al. -N = 17 OECD countries -Active labour market social spending -Suicide rate 2 1 0 1 (2009) -1980-2003 -Active labour Market social spending * -Ecological unemployment -Fixed effects regression

Veenhoven & -N = 97 global countries -Social security expenditure (% of GDP) -Life expectancy 8 0 0 8 Ouweneel (1995) -1965-1985 -Government expenditure (% of GDP) -Changes in life expectancy -Ecological -Correlation Veenhoven -N = 22-35 global countries -Social security expenditure (% of GDP) -Life expectancy 3 0 0 3 (2000) -1980-1990 -Changes in life expectancy -Ecological -Gini life expectancy -Correlation Xu (2006) -N = 50 US states -State and local government expenditure (per -Health achievement index 2 2 0 0 -2001 capita) -Ecological -Public health spending (% of government -Interval regression expenditure) Notes: OECD = Organization for Economic Co-operation and Development; GDP = Gross domestic product; GNP = Gross national product; PPP = Purchasing power parities; OLS = Ordinary least squares; US = United States; (#) = number of independent hypotheses tested; (+) = positive association; (-) = negative association; (≠) = no association; * = health outcome stratified by gender

Table 2.4. Empirical Studies on Welfare Regimes and Population Health (n = 12) Author(s)/ Year N, Period, Political Variable(s) Health Outcome(s) Findings Level, Design # + - ≠ Bambra (2005) -N = 18 OECD countries Three welfare regimes: -Infant mortality rate 3 3 0 0 -1980 &1999 Social democratic, Corporatist & Liberal -Ecological -Cross-sectional Chung & -N = 18 OECD countries Four welfare regimes: -Infant mortality rate 13 7 0 6 Muntaner (2007) -N = 39 years Social democratic, Conservative, Liberal, Wage -Low birth weight -1960-2004 Earner -Multi-level Chung et al. -N = 93,505 individuals Seven welfare regimes: -Brief depression episode* 19 5 0 14 (2013) -N = 26 Global countries Social democratic, Conservative, Liberal, -2002-2004 Southern, East European, -Multi-level South Asian, Latin America -Cross-sectional Conley & -N = 19 OECD countries Three welfare regimes: -Infant mortality rate 8 4 0 4 Springer -1960-1982 Social democratic, Corporatist, Liberal -Low birth weight (2001) -Ecological -Country, fixed-effects Eikemo et al. -N = 65,065 individuals Five welfare regimes: -Poor self-rated health* 4 2 0 2 (2008) -N = 23 European countries Social democratic, Corporatist, Liberal, Southern, -Limiting longstanding illness* -2002-2004 Eastern -Multi-level (Continued on next page)

69 Table 2.4. Empirical Studies on Welfare Regimes and Population Health (n = 12) (Continued) Kangas (2010) -N = 18 OECD countries Three welfare regimes: -Life expectancy at birth 6 0 0 6 -1900-2000 Nordic/Scandinavian/Social democratic, Central- -Change in life expectancy -Ecological European/ Corporatist/Conservative, Liberal/ at birth -TSCS Anglo-American Karim et al. -N = 30 Global countries Six welfare regimes: Scandinavian, -Infant mortality rate 2 1 0 1 (2010) -2003 Anglo-Saxon, Bismarckian, Southern, -Life expectancy at birth -Ecological Eastern European, East Asian -Cross-sectional Richter et al. -N = 141,091 individuals Four welfare regimes: Scandinavian, -Self-rated health 8 3 0 8 (2012) -N = 32 OECD countries Bismarckian, Anglo-Saxon, Southern -Health complaints -2005/2006 -Multi-level -Cross-sectional Rostila (2007) -N = 36,489 individuals Five welfare regimes: Social democratic, Liberal, -Life expectancy at birth 4 1 0 3 -N = 20 European countries Conservative/corporatist, Mediterranean, Post- -Self-rated health -2002 & 3003 socialist -Multi-level -Cross-sectional van der Wel et -N = 213,585 individuals Five welfare regimes: Scandinavian, Bismarckian, -Limiting longstanding illness, 8 5 0 3 al. (2012) -N = 26 European countries Anglo-Saxon, Southern, Eastern stratified by educational level* -Multi-level -Cross-sectional Witvliet & al. -N = 207,818 individuals Seven welfare regimes: Social democratic, -Self-reported disability, stratified by 36 15 0 21 (2011) -N = 49 Global countries Conservative, Liberal, Liberal-informal, employment status and educational -Multi-level Productivist, South Asia, Sub-Saharan Africa attainment -Cross-sectional Witvliet & al. -N = 199,595 individuals Seven welfare regimes: Social democratic, -Self-reported disability 6 1 0 5 (2012) -N = 46 Global countries Conservative, Liberal, Liberal-informal, -Multi-level Productivist, South Asia, Sub-Saharan Africa -Cross-sectional Notes: OECD = Organization for Economic Co-operation and Development; (#) = number of independent hypotheses tested; (+) = positive association; (-) = negative association; (≠) = no association; * = health outcome stratified by gender

70 Table 2.5. Empirical Studies on Leftist Politics and Population Health (n = 14) Author(s)/ Year N, Period, Political Variable(s) Health Outcome(s) Findings Level, Design # + - ≠ Blakely et al. -N = 279,066 individuals -Voter inequality by SES (states coded as high, -Self-rated health 6 3 0 3 (2001) -N = 50 US states medium-high, medium-low, and low) -1990, 1992, 1994 and 1996 -Voter turnout (% of electorate that voted) -Multi-level -Logistic random effects regression Chung & -N = 19 OECD countries -Voter turnout (% of electorate that voted) -Infant mortality rate 6 3 2 1 Muntaner (2006) -1960-1994 -Left vote (% of total votes for left parties) -Under 5 mortality rate -Ecological -Low birth weight -Robust-cluster variance estimator Kawachi et al. -N = 50 US states -Women in elected office (weighted according to -Total mortality rates* 35 23 0 12 (1999) -1993-1995 degree of political influence of the position) -White female mortality rate -Ecological - Women registered to vote in 1992/94 (% of -Black female mortality rate -OLS regression electorate that voted) -Female ischaemic heart disease -Women who voted in 1992/94 (% of electorate -Female cerebrovascular disease that voted) -Female malignant neoplasms -Number of institutional resources available to -Female breast cancer women (# of resources) -Cervical cancer -Political participation composite index (combined -Female suicide the four above aspects of women’s political -Female homicide status) -Infant mortality -Mean days limited activity* Kelleher et al. -N = 26 counties in Ireland -Voting patterns for political parties (% of first -Standard mortality ratio 24 1 2 22 (2002) -1996-1997 preference votes for centre-right, -Fair or poor health -Ecological Labour/Democratic Left, Progressive Democrat, -Poor or very poor quality of life -Correlation Other, and Abstainers) -Dissatisfied with health

(Continued on next page)

71 Table 2.5. Empirical Studies on Leftist Politics and Population Health (n = 14) (Continued) Lena & London -N = 50-84 global countries -Left regime norms (coded as left) - Infant mortality 28 0 0 0 (1993) -1965-1970, 1983 -Right regime norms (coded as right) -Life expectancy -Ecological -Left regime * strength (left * state spending) -Child death rate -OLS regression -Right regime * strength (right * state spending)

Lynch et al. -N = 14-16 OECD countries -Women in government (% of women elected) -Total mortality rate* 92 10 0 82 (2001) -1989-1992 -Belong to a trade union (% of trade union -Cause-specific mortality rate* -Ecological members) -Suicide* -Correlation -Homicide* -Life expectancy* -Low birth weight* -Infant mortality rate* -Self-rated health* Muntaner et al. -N = 12-15 OECD countries -Left votes (% of total votes for left parties) -Total mortality rate* 505 104 53 348 (2002) -1989-1992 -Left seats (% of parliamentary seats need to -Cause-specific mortality rate* -Ecological have a majority) -Suicide* -Correlation -Females in government (% of elected seats in -Homicide* national governments occupied by women) -Life expectancy* -Voter turnout (% of electorate voting) -Low birth weight* -Authoritarian legacy -Infant mortality rate* -Social democratic government (years in power) -Self-rated health* -Christian democratic government (years in power) -Union density (% of labour force) -Social pact (scale of 1 o 4, measures the pact between labour and capital) -Industrial disputes (# of disputes involving 1,000 or more) (Continued on next page)

72 Table 2.5. Empirical Studies on Leftist Politics and Population Health (n = 14) (Continued) Navarro et al. -N = 17 OECD countries -Left time governing cumulative (# of months and -Infant mortality rate 36 10 0 26 (2003) -1950-1998 years in government) -Life expectancy at birth* -Ecological -Left vote (% of electorate who voted for left -Correlation parties) -Voter turnout (% of electorate who voted) Navarro et al. -N = 17 OECD countries -Voter participation (% of electorate who voted) -Infant mortality rate 54 9 0 45 (2006) -1972-1996 -Voter partisanship (% of the vote for four different -Life expectancy at birth* -Ecological political traditions) -Correlation -Time in government by political parties (months and years of political parties in power) Reitan (2003) -N = 70-78 districts in Russia -Voter turnout (% of electorate who voted) -Life expectancy, various time 24 20 0 4 -1991-1999 periods* -Ecological -Changes in life expectancy, various -Correlation time periods* Rodríguez-Sanz -N = 17 communities in Spain -Social Democratic government (years of -Total premature mortality* 20 8 1 11 (2003) -1989-1998 government beginning in the 1980s to 1998) -Premature mortality by -Multi-level -Other government (years of government cerebrovascular disease*, cirrhosis*, -Hierarchical linear regression beginning in the 1980s to 1998) ischemic heart disease*, lung cancer*

Ronzio et al. -N = 141 US cities -Mayoral party (coded as Democratic, Republican, -All-cause death rate 2 0 0 2 (2004) -1980 & 1990 non-partisan) -Preventable death rate -Ecological -Mayoral type (coded as city manager or mayor) -OLS regression Smith & Dorling -N = 561 electoral constituencies -Partisan voting (% of votes for Conservative, -Total standard mortality rate 12 6 6 0 (1996) in England/Wales Labour, Liberal, and Abstention) -Total standard mortality rate* -1983,1987, 1992 -Ecological -Multiple regression Smith & Dorling -N = 561 electoral constituencies -Partisan voting (% of votes for Conservative, -Total standard mortality rate 15 9 6 0 (1997) in England/Wales Labour, Liberal, Referendum and Abstention) -Total standard mortality rate* -1983,1987, 1992, 1997 -Ecological -Multiple regression (Continued on next page)

73 Table 2.5. Empirical Studies on Leftist Politics and Population Health (n = 14) (Continued) Notes: OECD = Organization for Economic Co-operation and Development; OLS = Ordinary least squares; US = United States; (#) = number of independent hypotheses tested; (+) = positive association; (-) = negative association; (≠) = no association; * = health outcome stratified by gender

74 CHAPTER THREE. CONCEPTUAL MODEL

The preceding chapter argued that prevailing approaches in social epidemiology tend to focus on mid-level determinants of population health while overlooking the political determinants of population health. Moreover, it is concluded that more theoretical and empirical work is needed to determine whether the major arguments of welfare generosity, leftist politics, and welfare regimes are also applicable within sub-national contexts. Given these knowledge gaps, this dissertation conceptualizes politics as a macro-level determinant of population health among Canadian provinces, and uses a conceptual model informed by political sociology and social epidemiological theories. In this chapter, I provide a supportive rationale on why Canada serves an attractive case to test the relationship between politics and health, and outline a conceptual model on how key concepts and health outcomes are directly and indirectly related to each other.

Politics of Population Health in Canada

In order to assess the relevance of political sociology theories within a single country, I take advantage of political and population health differences among Canadian provinces over the past three decades (1976–2008). Canadian provinces serve as an attractive set of observations to advance theory and empirics for three reasons (Imbeau et al., 2000). First, Canadian provinces are sub-national jurisdictions that possess sufficient political autonomy to achieve different levels of welfare generosity. Provinces can be viewed as a set of political institutions that have the ability to make binding decisions for people and organizations located within its jurisdiction (Rueschemeyer &

75 Evans, 1985). According to Canada’s constitution, political power is divided between federal and provincial governments. The powers and responsibilities of provincial governments outlined in sections 92-93 include expenditures on health, education, and social services, which are key features of Canada’s welfare state. Critical for explaining variations in population health is that provincial political autonomy leads to different levels of welfare generosity across provinces and over time. For example, with respect to total health expenditure per capita in 2011, Newfoundland and Labrador and Alberta spent more per person on health care than any other province, at $6,884 and $6,570, respectively. In contrast, Quebec and British Columbia spent the least on health expenditure per capita, at $5,261 and $5,450, respectively (Canadian Institute for

Health Information, 2011). In another example, between 1989 and 2007, Quebec and

British Columbia are the only two provinces that increased their proportional spending on social services relative to total program expenditures. Without minimizing the impact of the federal government, these figures suggest that provinces do possess key instruments with which to influence the generosity of welfare expenditures (Cairns,

1977).

Second, leftist political factors, expressed as the strength of labour unions, political parties, voter turnout, and women in government, also vary widely across

Canadian provinces. In terms of labour unions, the average provincial union density rate in 2011 was 31.2%, with Quebec (39.3%) as the most unionized jurisdiction, followed by

Newfoundland (39%), Manitoba (36.2%), Saskatchewan (35.3%), Prince Edward Island

(32.7%), and Nova Scotia (31.5%). Provinces below the average unionization rate included British Columbia (30.9%), New Brunswick (29.6%), Ontario (27.8%), and

76 Alberta (23.2%) (Uppal, 2011). Regarding political parties, the New Democratic Party

(NDP) and Parti Quebecois are two parties with leftist and social democratic ideologies that have been successful in provincial elections. The NDP have been historically strong in Saskatchewan, Manitoba, and British Columbia, while the Parti Quebecois has served as the majority government in Quebec from 1976-1984 and 1994-2002, and most recently, won minority power in 2012. As for the relative strength of the Liberals and Conservatives (centre and right-wing political parties), their influence has been traditionally stronger in Ontario, Alberta, and the Atlantic provinces (Cross, 2011).

From 1965 to 2009, voter turnout rates in provincial elections reveal historical differences (Wesley, 2010). With an overall average turnout of 71.5%, Prince Edward

Island has the highest voter turnout (83.8%), followed by Manitoba (77%), and New

Brunswick (76.9%). At the other spectrum, provinces with the lowest turnout rates for provincial elections include Alberta (56.4%), Ontario (62.5%), and Manitoba (67.9%)

(Wesley, 2010). In terms of women in Canadian politics, women currently represent

21.3% of all elected politicians in provincial and territorial legislatures (156 of 734 seats). However, their representation varies across provinces, from a low of 12.7% in

New Brunswick to a high of 30.9% in Quebec. The remaining provinces ranked from highest to lowest include Prince Edward Island (25.9%), Ontario (25.0%), Manitoba

(23.2%), Newfoundland (22.9%), British Columbia (21.5%), Saskatchewan (17.5%),

Nova Scotia (17.3%), Alberta (15.7%), and New Brunswick (12.7%) (CBC, 2013).

A third reason why Canada serves as an attractive case to empirically test the association between politics and health is that levels of population health vary across provinces and time (Ariizumi & Schirle, 2012; Laporte & Ferguson, 2003). Respectively,

77 figures 3.1, 3.2, and 3.3 show the total, male, and female age-standardized mortality rates (all causes of death per 1,000 population) for ten Canadian provinces in 1975 and

2009 (located at end of chapter). This dissertation examines the same dependent outcomes roughly over the same period. These figures reveal a few trends worth explicating. First, Saskatchewan (8.3 per 1,000) and British Columbia (4.8 per 1,000) had the lowest total mortality rates in 1975 and 2009 while Nova Scotia (9.40 per 1,000) and Newfoundland (6.3 per 1,000) had the highest rates in 1975 and 2009. Second, the mortality rates for each province have declined over time, and this downward trend has been remarkably consistent. For example, the percentage change in the Canadian average for total, male, and female mortality rates from 1975 to 2009 are 76.7%, 85.7%, and 65.3%, respectively. Third, some provinces have experienced comparatively greater declines than others. The highest declines in total mortality rates have occurred in Ontario (82.1%), British Columbia (79.1%), and Quebec (78.9%), and the smallest declines in Saskatchewan (44.6%), Newfoundland (45.8%), and Prince Edward Island

(48.3%). And fourth, gender differences in mortality trends differ slightly from total mortality patterns. Whereas the greatest declines occurred in Quebec for both genders

(males, 99.8%; females, 78.9%), the smallest declines for males and females are observed in Newfoundland (45.1%) and in Prince Edward Island (33.1%), respectively.

Overall, it is clear that provinces possess sufficient political autonomy and differ with respect to welfare generosity, leftist politics, and population health. Despite this, there are several dimensions along which provinces differ that may explain current levels of and changes over time in population health. These dimensions include province-level factors such as demographics (e.g., the senior population tends to be

78 lowest in Alberta), labour market characteristics (e.g., unemployment is highest among

Atlantic provinces and lowest among Prairie provinces), and economic development

(e.g., in 2011, Alberta’s GDP per capita (CAD$), $78,154, is more than two times higher than Prince Edward Island, $36,740). These alternative explanations as well as other confounding factors are controlled for in the subsequent analyses.

Previous studies attempting to explain differences in mortality rates across

Canadian provinces have assessed the effects of income inequality (Laporte &

Ferguson, 2003) and unemployment (Ariizumi & Schirle, 2012). To date, no study has conceptualized provinces as sub-national political jurisdictions to test empirical models of welfare generosity, leftist politics, and population health. Next, I present and summarize the key concepts that inform this model.

Conceptual Model: Welfare Generosity, Leftist Politics, and Population Health

At this point, it is useful to acknowledge that this dissertation’s conceptual model is inspired in large part by the work of Navarro et al. (2003, 2006). In two seminal papers published in the International Journal of International Health Services (2003) and the

Lancet (2006), these authors integrated social epidemiology and political sociology approaches and devised a heuristic framework known as the ‘Barcelona model’ to explain the causal pathways through which political factors shape and influence population health and health inequalities. The Barcelona model argues that political dynamics such as electoral behaviour and trade union characteristics influence the degree to which welfare states implement social, welfare, and labour market policies. By

79 extension, these policies determine living conditions and social inequalities, and population levels of health and health inequalities. Nordic countries are cited as supportive evidence since these nations have historically had stronger unions and left political parties, expansive and generous welfare states, full employment policies, more women participating in the labour market, and therefore, less pronounced social inequalities and comparatively better population health outcomes (Borrell et al., 2007;

Navarro et al., 2006).

The Barcelona model offers a compelling narrative on how different forms of working-class power in the form labour unions and left political parties can mobilize to overcome the power of business to expand the welfare state, narrow health inequalities, and improve population health. Only when the working class is allied with the middle class can their power resources be sufficiently mobilized to advocate for generous welfare expenditures. The conceptual model used in this dissertation is a modified extension of Navarro et al.’s (2003, 2006) model. Consistent with the Barcelona model, this dissertation accepts three neo-Marxist positions: 1) capital is a unique power resource because it is disproportionately concentrated in the hands of business owners;

2) political power is also unequally distributed in capitalist democracies; and 3) opponents of egalitarian outcomes rest in a position of default dominance within the electoral arena (Korpi, 1983; Stephens, 1979). This dissertation aims to assess the extent to which the major arguments of the Barcelona model, which was originally developed to understand cross-national differences in health, are also applicable within

Canadian provinces.

80 The conceptual model presented in figure 3.4 outlines direct and indirect connections on how power resources and partisan politics (e.g., union strength; left, centre, and right political parties) and democratic politics (e.g., voter turnout and women in government) are related to provincial welfare generosity (e.g., expenditures on health, social services, and education), which in turn, affect population health (e.g., total, male, and female age-standardized mortality rates) (located at end of chapter). The solid arrows between boxes represent direct pathways between political determinants and population health, and the long dashed arrows indicate indirect pathways.

My model begins with the hypothesis that provincial welfare generosity operates as a proximate determinant of population health. Current studies generally find a positive association between welfare generosity and population health – as government expenditures increase, health outcomes also tends to improve. My conceptual model advances this work in two important ways. First, I conceptualize Canadian provinces as sub-national welfare states that provide welfare services and social transfers, which in turn, influence population health. In existing work, welfare states have largely been conceptualized as nations or regimes even though sub-national political jurisdictions such as Canadian provinces possess sufficient political autonomy to affect population health. Second, given that prior research generally relies on aggregate indicators of spending, my conceptual model considers the effects of both aggregate (health, social services, and education) and disaggregated (eleven sub-expenditures) types of welfare generosity on population health. For example, the impact of health expenditures, an aggregate measure, will also be tested as four disaggregated measures, including hospital care, medical care, preventive care, and other health services. By implication,

81 this dissertation’s focus on aggregate and disaggregated types of provincial welfare generosity provides a more nuanced and detailed understanding on what kinds of public spending matter most to population health.

Given that provincial welfare generosity is hypothesized to have a positive and significant impact on population health, it is also worthwhile to consider the determinants of welfare states and to assess their distal effects on health. Thus, my conceptual model also emphasizes the impact of power resources and political parties on welfare generosity and population health (see chapter 6). Regarding power resources, if labour unions are organized, mobilized, and strong; population health should be enhanced through higher wages, improved working conditions, and increased support for left political parties (e.g., New Democratic Party and Parti Quebecois). To date, the impact of union strength and left political party power on population health has been examined using cross-national samples, which include Canada; however, it remains unknown whether variations in power resources at the provincial level explain variations in population health.

My conceptual model also examines the health consequences of centre and right political parties, which according to partisan politics theory, should have positive and negative associations with population health, respectively. In figure 3.4, note the dotted lines connecting power resources and political parties to provincial welfare generosity and the solid lines to population health. These lines and arrows reflect the two potential pathways through which labour unions and political parties may affect population health.

The first pathway shows that the effects of labour unions and political parties channel

82 through welfare generosity to affect population health, and the second allows for the possibility that unions and parties combine with welfare generosity to influence health.

A third component of my conceptual model emphasizes the roles played by democratic politics in relation to population health (see chapter 6). First, I test the connections between voter turnout, provincial welfare generosity, and population health.

I argue that political parties are inclined to make pro-welfare appeals (e.g., promises of more health care and income security) to “median” voters in hopes of gaining electoral majorities. Such appeals increase voter turnout and pressure governmental incumbents to increase levels of welfare generosity. The connections between voter turnout, left political parties, and welfare effort have been confirmed in the political sociology literature. However, the connection between voter turnout and population health is not.

Second, my conceptual model also explores whether the historical strength of women in government affects population health. Following the lead of Lynch et al. (2001) and

Muntaner et al. (2002), I conceptualize women in government as an indicator of welfare state strength and democratic governance in capitalist democracies. This argument is based on the idea that female representatives in provincial legislatures are more likely to “do politics” differently from their male counterparts. My model posits that if more women occupy positions of political power, the interests of women and children are advanced, and provincial levels of population health are enhanced. As figure 3.4 depicts, higher levels of voter turnout and women in government are hypothesized to affect population health through indirect (e.g., the effects of voter turnout and female politicians channel through welfare generosity) and direct mechanisms (e.g., the effects of voter turnout and female politicians combine with welfare generosity).

83 After testing and revising this dissertation’s conceptual model, I broaden the analysis to examine whether provinces cluster into distinct groups based on leftist political factors and to test whether these distinct groups are predictive of population health outcomes (see chapter 7). According to the most recent generation of welfare state scholarship, welfare states cluster into different institutional regimes based on various characteristics, including for example, decommodification levels (Esping-

Andersen, 1990, 1999), political traditions (Huber & Stephens, 2001), social policy programs (Korpi, 2000; Korpi & Palme, 1998, 2003), and market economies (Hall &

Soskice, 2001a, 2001b). The important idea that I borrow from welfare regime theory is that sub-national political jurisdictions may also cluster into distinct groups. Recent work in the Canadian politics literature has explored whether provinces display notable divergences, stemming from economic, political, and cultural differences. One important study is conducted by Bernard and Saint-Arnaud (2004), who measured Alberta, British

Columbia, Ontario, and Quebec along three welfare regime indicators: social situations, public policy, and civic participation by citizens, and compared these provinces to established welfare regimes (e.g., liberal, conservative, social democratic). Their findings revealed that Alberta resembled the “ultra-liberal” United States and Quebec leaned toward being a social democratic country. This dissertation makes three unique contributions to this work. First, I expand the number of provinces tested as welfare regimes from four to ten. Second, I use cluster methods to classify provinces into distinct groups based on indicators of leftist politics. And third, I test the association between provincial clusters and population health, and assess whether variations in welfare generosity explain mortality rates across regimes.

84 So how do politics affect population health in Canada? According to my conceptual model, I argue that population health will be enhanced where union densities are high, left and centre political parties are strong, voter turnout is high, more women are represented in government, and provincial expenditures are generous. At the same time, I acknowledge that other provincial-level factors may also affect population health and therefore control for these alternative explanations, including provincial characteristics (e.g., transfer payments, outstanding debt), demographic factors (e.g., size of youth and elderly population), labour market characteristics (e.g., female labour force participation, unemployment), and economic development (e.g., low income, provincial Gross Domestic Product). After testing these hypotheses in chapters

5 and 6, I revise the conceptual model based on bivariate and regression results in chapter 8 (Discussion). In the next chapter, I outline the methods and define the variables used in this dissertation.

85 Figure 3.1. Total Age-Adjusted Mortality Rates (all causes of death per 1,000 population) by Province, 1975 and 2009

QC 1975 2009 NB

0 1 2 3 4 5 6 7 8 9 10

Notes: BC = British Columbia, AB = Alberta, SK = Saskatchewan, MB = Manitoba, ON = Ontario, QC = Quebec, NB = New Brunswick, NS = Nova Scotia, PE = Prince Edward Island, NL = Newfoundland, CA = Canada. Source: Statistics Canada. (2013). Table 102-0552 - Deaths and mortality rate, by selected grouped causes and sex, Canada, provinces and territories, annual, CANSIM (database).

86 Figure 3.2. Male Age-Adjusted Mortality Rates (all causes of death per 1,000 population) by Province, 1975 and 2009

QC 1975 2009 NB

0 2 4 6 8 10 12 14

87 Figure 3.3. Female Age-Adjusted Mortality Rates (all causes of death per 1,000 population) by Province, 1975 and 2009

QC 1975 2009 NB

PEI

CAN

0 1 2 3 4 5 6 7 8

88 Figure 3.4. Conceptual Model: Combining Power Resources, Political Parties, Democratic Politics, Provincial Welfare Generosity, and Population Health

Union Strength

Power Resources and Left, Centre, and Political Parties Right Political Parties

Provincial Population Welfare Health Generosity

Voter Turnout and Democratic Politics Women in Government

89 CHAPTER FOUR. METHODS

This chapter is organized in three parts. First, I describe the data sources used to construct a TSCS dataset. Second, I operationally define the dependent and independent variables used in this dissertation. Third, I describe and justify the estimation and post-estimation techniques used in empirical testing.

Data Sources

This dissertation uses a pooled TSCS analysis of 10 Canadian provinces from 1976 to

2008, including Newfoundland, Prince Edward Island, Nova Scotia, New Brunswick,

Québec, Ontario, Manitoba, Saskatchewan, Alberta, and British Columbia. All data are measured at the provincial level, and the unit of analysis is a province-year. By repeating cross-sectional provincial observations over time, this dissertation overcomes the traditional problem of comparative research and gains important returns in terms of data modeling (Kenworthy, 2004). These problems arise when the number of units such as countries or provinces available for comparative analysis is inherently limited (Ragin,

1994). Specifically, the classic ‘small N’ problem is overcome by using datasets of N ×T observations, obtained by arraying cross-sectional data on N country-units and T time periods.

Guided by the design and structure of the Comparative Welfare States Data Set assembled by Huber, Ragin, and Stephens (1997) and updated by Brady, Beckfield, and Stephens (2004), three data sources are used to construct a TSCS database. First, annual provincial data on population health, welfare generosity, union density,

90 equalization transfers, debt charges, dependency ratio, women in labour force, unemployment rate, low income and real GDP per capita are retrieved from CANSIM II

Tables (Canadian Socio-economic Information Management System) through the

University of Toronto’s CHASS (Computing in the Humanities and Social Sciences).

CANSIM is Statistics Canada’s computerized database and retrieval service that includes time-series data on socio-economic, demographic, health, and education statistics. Second, annual provincial data that are not available on CANSIM are retrieved from published documents by Statistics Canada. Third, annual political data on political party power, voter turnout, and women in government are derived from various editions of the Canadian Parliamentary Guides for the years 1960-2008 and supplemented by information from public sources, including provincial Election Offices.

References to data series and variables appear as a codebook in Appendix A.

Measures of Dependent, Independent, and Control Variables

Dependent Variables

Age-standardized mortality rate. The age-standardized mortality rate is a weighted average of the age-specific mortality rates per 1,000 persons, where the 1991

Canadian Census of Population is used as the standard population for the calculation.

Adjusting for age removes the effects of differences in the age structure of populations among provinces and over time. Three measures of age-standardized mortality rates are tested as health outcomes: total (both sexes), male, and female. These dependent measures are available from 1976 to 2008. Data on mortality rates from 1976 to 1999 are retrieved from Statistics Canada, Births and Deaths (1995, 1996) and Statistics

91 Canada, Mortality - Summary List of Causes (1997, 1998, and 1999). Data on mortality rates from 2000 to 2008 are collected from CANSIM II, Table 102-0552. Other studies conducted in Canada have also used mortality rates as an indicator of population health

(Ariizumi & Schirle, 2012; Laporte & Ferguson, 2003; Ross et al., 2000).

Independent Variables

Measures of Provincial Welfare Generosity

Consistent with other research using provincial expenditure data in Canada (Di

Matteo & Di Matteo,1998; Landon, McMillan, Muralidharan, & Parsons, 2006), all expenditure variables are converted into real per capita terms using data from CANSIM

II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Provincial welfare generosity is measured at two levels: aggregate and disaggregate. At the aggregate level, I examine three general measures of welfare spending: health, social services, and education. These three measures are selected for theoretical and substantive reasons. Theoretically, past researchers have found positive associations between these indicators of welfare generosity and population health across nations and within-country studies (Bradley et al. 2010; Dunn et al. 2005; Laporte & Ferguson, 2003). Substantively, almost two-thirds

(64%) of total provincial spending is attributable to health, social services, and education (Statistics Canada, 2009). Provincial welfare generosity measures are available from 1989 to 2008 for a total of 200 cases. Although earlier expenditure data is available in CANSIM matrices 2782 to 2791 and in the Statistics Canada publication

Public Finance Historical (68-512), Landon et al. (2006) found serious differences

92 between currently available data and older series for the years in which the two data series overlapped. Given the unreliability of earlier data, this dissertation uses the best available data circa 1989.

Provincial Welfare Generosity: Aggregate Expenditures

Health expenditure. This variable captures provincial spending made to ensure that necessary health services are available to all citizens (Statistics Canada, 2009).

Residential care facilities and other health and social services institutions providing medical care and professional nursing supervision are considered as institutions providing health services. Health expenditures also include hospitals’ ancillary enterprises (i.e., entities that exist to furnish goods and services to patients, staff and others). Four sub-functions identify the major components of health expenditures: hospital care, medical care, preventive care, and other services. Among Canadian provinces, increases in real per capita spending on total health care have been shown to be associated with significant reductions in mortality (Laporte & Ferguson, 2003). The health expenditure variable is converted into real per capita terms using data from

CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on health spending is available from

1989 to 2008 and collected from CANSIM II, Table 385-0001.

Social services expenditure. Spending on social services include actions taken by provincial governments to offset or to forestall situations where the well-being of individuals or families is threatened by circumstances beyond their control (Statistics

Canada, 2009). These expenditures capture assistance (transfers) and services to

93 individuals who are so disadvantaged that the universal social security services are inadequate to provide for their well-being or who fail to qualify for support from those services. This aggregate measure of welfare generosity is comprised of five sub- functions: social assistance, workers’ compensation benefits, other social services, employee pension plan benefits, and motor vehicle accident compensations. The latter two sub-functions are omitted from disaggregated analyses due to incomplete data. The impact of social expenditures on population health has been less examined in the extant literature, and none of this work has been conducted within a Canadian context. The social services expenditure variable is converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on social services spending is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001.

Education expenditure. Education spending includes the costs of developing, improving, and operating educational systems and the provision of specific education services (Statistics Canada, 2009). Also included are expenditures of universities’ ancillary enterprises (i.e., entities providing goods and services to students, staff and others). Education expenditures can be further disaggregated into four sub-functions: elementary and secondary education, post-secondary education, special retraining services, and other education. Past research among US states has found that per capita expenditures on education (primary, secondary and higher education) are positively associated with all-age mortality and with male and female working-age mortality rates (Dunn et al., 2005). To date, the population health impact of education expenditures has not been examined among Canadian provinces. The education

94 expenditure variable is converted into real per capita terms using data from CANSIM II,

Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from

CANSIM II, Table 326-0021. Data on education spending is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001.

Provincial Welfare Generosity: Disaggregated Expenditures

Aggregate measures of health, social services, and education expenditures can be disaggregated into eleven sub-functions. Health spending includes hospital care, medical care, preventive care, and other health services. Social services spending include social assistance, workers’ compensation benefits, and other social services.

Education spending includes elementary and secondary education, post-secondary education, special retraining services, and other education expenditures. Also, a provincial welfare generosity index is constructed using the standard scores of medical care, other social services, and post-secondary education given that the significance of these expenditures in lowering mortality rates. To date, no research in Canada has examined the population health effects of disaggregated forms of provincial welfare generosity.

Hospital care expenditure. Spending on hospital care refers to all services provided by general hospitals and public health clinics, as well as by acute disease, chronic disease, convalescent, isolation and mental hospitals (Statistics Canada, 2009).

Expenditures pertaining to nursing schools attached to hospitals, and all private, public, and religious hospitals are included except for national defence and veterans hospitals whose costs are allocated to the “National Defence” and “Veterans Benefits” sub-

95 functions respectively. The hospital care expenditure variable is converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index

(CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on hospital care spending is available from 1989 to 2008 and collected from CANSIM II, Table 385-

0001.

Medical care expenditure. Spending on medical care includes general medical care, drug programs, dental and visiting-nurse services, and out-patient care services

(Statistics Canada, 2009). This variable also includes outlays for medical care provided by hospitals, public residential care facilities, workers’ compensation boards, and other public health, and social services institutions. Transfers to private residential care facilities and other health and social services institutions to help them finance their medical care activities are included here. The medical care expenditure variable is converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-

0021. Data on medical care spending is available from 1989 to 2008 and collected from

CANSIM II, Table 385-0001.

Preventive care expenditure. Preventive care spending consists of a wide variety of outlays which are intended to prevent the occurrence of diseases and to mitigate their effect (Statistics Canada, 2009). It covers public health clinics, communicable disease control services (including immunization, treatment, isolation and quarantine outside hospital premises), food and drug inspection services, hospitals which offer preventive services to patients, government establishments (not located in hospitals, e.g., residential care facilities and other health and social services institutions) providing

96 nursing, hygiene and nutrition advisory services, and government organizations conducting research on the causes and consequences of particular diseases or addictions (i.e., cancer treatment foundations). Also included are transfers to private facilities providing preventive care (i.e., private residential care facilities). This variable is converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-

0021. Data on preventive care spending is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001.

Other health services expenditure. This spending variable includes outlays on clinics for the treatment of mental disabilities or emotionally disturbed persons and on laboratory and diagnostic services, grants to health-oriented organizations, and expenditures on other health-related services such as health department administration, health statistics, staff training and other services of health establishments (e.g., hospitals and other health and social services institutions), ambulance services, medical rehabilitation and indemnities to injured persons and their dependants which cannot be allocated to the other sub-functions (Statistics Canada, 2009). Also included are outlays on protection of health and health inspection, and expenditures of ancillary enterprises of health and social services institutions. Spending on other health services is converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on other health services spending is available from 1989 to 2008 and collected from

CANSIM II, Table 385-0001.

Social assistance expenditure. Social assistance expenditures consist of transfer

97 payments (including refundable tax credits) to help individuals and families maintain a socially acceptable level of earnings (Statistics Canada, 2009). Social assistance comprises the following programs: the general welfare payments to disadvantaged individuals, the refundable tax credits and rebates for low-and-middle income individuals or families (which are used more and more as instruments of social policy to offset taxation of the elderly and disadvantaged i.e., property and sales tax credits), outlays relating to contributory plans such as the Canada Pension Plan and the Québec

Pension Plan, and non-contributory plans, such as old age security (including the guaranteed income supplement), family allowance payments and child tax benefits made under federal and provincial governments programs, the employment insurance benefits, Québec Parental Insurance Plan benefits the rent supplement, the spouse’s allowances and the blind and disabled persons allowances. The social assistance variable is converted into real per capita terms using data from CANSIM II, Table 510-

001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM

II, Table 326-0021. Data on social assistance spending is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001.

Workers’ compensation benefits. This variable includes expenditures on administration and for benefits, other than rehabilitation and medical care, related to workers’ compensation schemes (Statistics Canada, 2009). Workers’ benefits are converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-

0021. Data on workers’ compensation benefits is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001.

98 Other social services expenditure. This variable accounts for expenses related to the provision of services to old age, to persons who are unable to lead a normal life due to a physical or mental impairment, to persons temporarily unable to work due to sickness, to households with dependent children, to persons who are survivors of a deceased person (spouse, children, etc.), and to other needy persons (Statistics

Canada, 2009). It also includes direct expenditures of public institutions such as hospitals, residential care facilities, other health and social services institutions that provide social services and transfers to private organizations (e.g., residential care facilities) that provide similar services. This variable is converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI)

(2002 = 100) for each province from CANSIM II, Table 326-0021. Data on other social services is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001.

Elementary and secondary education expenditure. This expenditure encompasses outlays for educational services from kindergarten to senior matriculation

(Statistics Canada, 2009). It also includes expenditure for technical and vocational training which is provided separately at the secondary school level as well as expenditure for general administration and maintenance of standards, contributions of governments, as employers, to teachers’ pension plans, support to students, the construction of buildings and the operation of education programs. Also included are expenses for pupil transportation, text books, electronics, equipment, and supplies used in the education process. Schools for the handicapped, schools for Aboriginal peoples and Inuit and transfers to private elementary and secondary schools come under this sub-function. Elementary and secondary education expenditures are converted into real

99 per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on elementary and secondary education is available from 1989 to 2008 and collected from

CANSIM II, Table 385-0001.

Post-secondary education expenditure. Education spending at the post- secondary level refers to education generally obtained in universities or in degree and non-degree granting community colleges and specialized educational institutions

(Statistics Canada, 2009). Included in these colleges and institutions are teachers’ colleges, advanced technical institutes and junior colleges, collèges d'enseignement général et professionnel (CEGEPs), music conservatories and schools specializing in the instruction and training of artists, and nursing education provided by universities and colleges. Also included are bursaries, scholarships and other types of financial assistance to students (e.g., loan forgiveness, interest relief, etc.) as well as refundable learning tax credits. Post-secondary education expenditures are converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index

(CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on post- secondary education is available from 1989 to 2008 and collected from CANSIM II,

Table 385-0001.

Special retraining services expenditure. Spending on retraining services comprises outlays made for the purpose of upgrading the skills of individuals (Statistics

Canada, 2009). It includes the cost of courses provided under the Federal Manpower

Training Program and the new Labor Market Development Agreement, the purchases of on-the-job training for unemployed insurance recipients, cash allowances or subsidies

100 to workers and persons available for work undergoing training, tax credits intended to encourage systematic employee training by corporations and other similar services.

Expenditures on special retraining services are converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 =

100) for each province from CANSIM II, Table 326-0021. Spending data on special retraining services is available from 1989 to 2008 and collected from CANSIM II, Table

385-0001.

Other education expenditure. This spending category covers outlays that either overlap or cannot be allocated to the other educational sub-functions (Statistics

Canada, 2009). It includes the general administration expenses of departments of education, the costs of statistical and research activities pertaining to education and the expenses of apprenticeship training. Payments made by one government to another or to the private sector to encourage proficiency in the official languages are also included, as are costs of special instructional arrangements such as evening classes and correspondence courses. Expenditures of ancillary enterprises of colleges and universities, e.g., bookstores and cafeterias, are included here. Expenditures on other education are converted into real per capita terms using data from CANSIM II, Table

510-001, and the consumer price index (CPI) (2002 = 100) for each province from

CANSIM II, Table 326-0021. Spending data on other education is available from 1989 to

2008 and collected from CANSIM II, Table 385-0001.

Provincial welfare generosity index. Empirical results in chapter 5 reveal that three types of disaggregated types of welfare generosity are significant predictors of lower age-standardized mortality rates (total, male, and female): medical care, other

101 social services, and post-secondary education. In order to develop a general measure of provincial welfare generosity, I constructed an index by averaging the standard scores (z-scores) of these three disaggregated expenditures. By doing so, I assess the extent to which the provincial welfare generosity index lowers mortality rates and use this index in subsequent analyses in chapters 6 and 7.

Measures of Leftist Politics

To test the population health hypotheses associated with power resources, partisan politics, and political democracy theories, I assess the health effects of six measures: union density, cumulative left party power, cumulative centre party power, cumulative right party power, voter turnout, and cumulative women in government. To the best of my knowledge, this is the first attempt to empirically test the connection between leftist politics and population health in Canada.

Union density. Union density measures the percentage of employees who belong to a trade union. According to Korpi (1983) and Stephens (1979), labour unions are the immediate manifestation of working-class mobilization, and represent one of two important ‘power resources’ (the second being leftist political parties) that influence distributive outcomes in the market and indirectly through the state. Organization in unions results in a shift of power in the market toward union members and social democratic political parties. A handful of studies have examined the impact of unions on population health, finding positive associations with child health outcomes and injury mortality rates (Lynch et al., 2001; Muntaner et al., 2002). Data on union density from

1976 to 1995 is retrieved from CANSIM II, Table 279-0025. Data from 1997-2008 is

102 obtained by dividing the number of employees covered by union agreements by total employees retrieved from CANSIM II, Table 282-0073. Following the lead of McDonald and Myatt (2004), I calculate the missing year (1996) by taking the mean union density rates of 1995 and 1997.

Cumulative left party power. The long-term, historical strength of left political parties is measured as the percentage of cumulative left seats (e.g., Green Party, New

Democratic Party, Parti Québecois, Québec solidaire). Following the lead of Huber et al.

(1997, 2004), this variable tabulates left seats as a proportion of seats held by all provincial government parties in each individual year and then sums these proportions for all years since 1960. Left political parties in Canada can be generally described as egalitarian, collectivist, and interventionist given their strong support for welfare services and social transfers, including for example, reducing healthcare costs (e.g., ending ambulance fees), increasing access to post-secondary institutions (e.g., freezing tuition fees), and implementing anti-poverty policies (e.g., building affordable housing and providing dental care services for low-income groups) (Albo, 2002; Whitehorn, 2007).

The historical strength of left political parties allied with strong trade union movements has been shown to have a positive effect on welfare state expenditures (Castles, 1989;

Huber & Stephens, 2001). Navarro et al. (2006) tested whether the political ideologies of governing parties affect social policy as well as population health. Their findings showed that political parties with egalitarian ideologies (e.g., left parties) are more likely to implement redistributive policies, which in turn, exert a salutary effect on infant mortality and life expectancy at birth (Navarro et al. 2006). Data on cumulative left party power is available from 1976 to 2008 and collected from publications available through

103 Elections Newfoundland, Elections Prince Edward Island, Elections Nova Scotia,

Elections New Brunswick, Chief Electoral Officer of Québec, Elections Ontario,

Elections Manitoba, Elections Saskatchewan, Elections Alberta, and Elections British

Columbia (see References for website links).

Cumulative centre party power. The long-term historical strength of centre political parties is measured as the percentage of cumulative seats for the Liberals, a centre political party, since 1960. The long-term, historical strength of centre political parties is measured as the percentage of cumulative centre seats for the Liberal Party.

Consistent with Huber et al. (1997, 2004), this variable tabulates centre seats as a proportion of seats held by all provincial government parties in each individual year and then sums these proportions for all years since 1960. Historically, the Liberal party has espoused the principles of liberalism, which emphasizes a mix of interventionist and free market values and goals. In this regard, the Liberal party is firmly positioned to the left of the Conservatives and to the right of the NDP (Bjørnskov & Potrafke, 2012).

Although the impact of centrist politics falls outside of theoretical contours of ‘power resources theory’, its inclusion as an explanatory variable is justified on the grounds that cross-national research finds that centre and centre-left political parties do exert a positive effect on welfare state expenditures, but primarily for transfer payments and with less redistributive impact (Castles, 1982; Esping-Andersen, 1990, 1999; Hicks &

Swank, 1992). Data on cumulative centre party power is available from 1976 to 2008 and collected from publications available through Elections Newfoundland, Elections

Prince Edward Island, Elections Nova Scotia, Elections New Brunswick, Chief Electoral

104 Officer of Québec, Elections Ontario, Elections Manitoba, Elections Saskatchewan,

Elections Alberta, and Elections British Columbia (see References for website links).

Cumulative right party power. The long-term, historical strength of right political parties is measured as the percentage of cumulative right seats (Progressive

Conservatives, Wildrose, Alberta Party, Alberta Social Credit Party, Coalition Avenir

Québec, and Saskatchewan Party). This variable tabulates right seats as a proportion of seats held by all provincial government parties in each individual year and then sums these proportions for all years since 1960 (Huber et al., 1997, 2004). Compared to left and centre political parties in Canada, right political parties espouse a different set of values and goals. Right parties are more likely to favour private enterprise (e.g., privatizing public assets), big business (e.g., opposing labour unions), and free markets

(e.g., accepting or supporting social inequalities due to market failures).

Competitiveness, restructuring, deficit reduction, and privatization of Crown corporations are viewed as laudable goals, as are private property rights, tax exemptions for capital gains, and free trade. These policy preferences are largely consistent with fiscal conservatism, which refers to a political and economic philosophy that is primarily committed to avoiding deficit spending, reducing government expenditures, and achieving balanced budgets. Data on cumulative right party power is available from

1976 to 2008 and collected from publications available through Elections

Newfoundland, Elections Prince Edward Island, Elections Nova Scotia, Elections New

Brunswick, Chief Electoral Officer of Québec, Elections Ontario, Elections Manitoba,

Elections Saskatchewan, Elections Alberta, and Elections British Columbia (see

References for website links).

105 Voter turnout. This variable measures voter turnout in each provincial election, in percentages of electorate that voted (Huber et al., 1997, 2004). High voter turnout is hypothesized to be associated with improved levels of population health by means of increased support for left political parties and welfare generosity (Navarro et al., 2006).

Differential rates of voter turnout between the supporters of left parties versus those of centre or right parties implies that support for left parties co-varies with turnout. It follows that disadvantaged groups (e.g., poor, low educated, unemployed), who are the natural constituency of left parties, vote at lower and more variable rates than their advantaged counterparts, who tend to vote at higher and consistent rates for centre or right parties

(Pampel & Williamson, 1989). If provincial changes in voter turnout reflect changes in the rate of political participation among disadvantaged groups, support for the left should vary directly with levels of voter turnout. If rates of turnout partially determine the electoral fortunes of left parties, then voter turnout itself may have significant consequences for population health (Navarro et al., 2006). Data on voter turnout is available from 1976 to 2008 and collected from the provincial websites, including

Elections Newfoundland, Elections Prince Edward Island, Elections Nova Scotia,

Elections New Brunswick, Chief Electoral Officer of Québec, Elections Ontario,

Elections Manitoba, Elections Saskatchewan, Elections Alberta, and Elections British

Columbia (see References for website links).

106 these proportions for all years since 1960 (Huber et al., 2004). This variable tests the hypothesis that women in government are more likely to promote a women-friendly political agenda than men and are more likely to espouse an egalitarian perspective on social, economic, and political issues (Klein, 1984; Sigel, 1996; Swers, 1998; Tremblay,

1998), which in turn, have a positive impact on population health. Existing research has found that higher political participation by women (measured using four aspects of women’s political status: voter registration, voter turnout, representation in elected office and women’s institutional resources) is correlated with lower female mortality rates and lower activity limitations in US States (Kawachi et al., 1999). Data on cumulative women in provincial government is available from 1976 to 2008 and collected from the respective websites of Elections Newfoundland, Elections Prince Edward Island,

Elections Nova Scotia, Elections New Brunswick, Chief Electoral Officer of Québec,

Elections Ontario, Elections Manitoba, Elections Saskatchewan, Elections Alberta, and

Elections British Columbia (see References for website links).

Provincial political regimes. I created a three-category provincial political regime variable: 1) leftist (Saskatchewan, British Columbia, and Manitoba) (reference category); 2) centre-left (Ontario and Quebec), and 3) conservative (Alberta, Nova

Scotia, New Brunswick, Prince Edward Island, and Newfoundland) using hierarchical cluster methods based on cumulative left political party power and cumulative women in government. As shown in chapter 6, these two leftist political variables are shown to be significant predictors of lower mortality rates (total, male, and female). Thus, I assess whether the leftist, centre-left, conservative typology explains population health, whether any inter-provincial regime differences can be explained by the provincial welfare

107 generosity index, and whether the effect of the provincial welfare generosity index differs across regimes in chapter 7.

Control Variables

In all models, I control for several provincial, demographic, labour market, and economic factors that may affect population health, including equalization transfers, debt charges, dependency ratio, women in labour force, unemployment rate, low income, and real

GDP per capita. Controlling for these alternative explanations provides a robust test of the non-spurious causal relationship between welfare generosity, leftist politics, and population health. Failing to control for these variables would yield models with biased and inflated estimates of the causal effects.

Transfers. This variable refers to the current fiscal year dollar value of total transfers, which includes the sum of general and specific purpose transfers, received from other levels of government. Controlling for this variable is important since the revenues available to provincial governments depend on the size of transfers from other levels of government (Statistics Canada, 2009). Key provincial transfers include the

Canada Health Transfer (CHT), the Canada Social Transfer (CST), and Equalization payments. The CHT and CST are federal transfers which support specific policy areas such as health care, post-secondary education, social assistance and social services, early childhood development and child care. Equalization transfers address fiscal disparities among provinces and enable less prosperous provincial governments to provide their residents with public services that are reasonably comparable to those in other provinces. Equalization payments are unconditional in nature – receiving

108 provinces are free to spend the funds according to their own priorities. Transfer expenditures are converted into real per capita terms using data from CANSIM II, Table

510-001, and the consumer price index (CPI) (2002 = 100) for each province from

CANSIM II, Table 326-0021. Data on transfer payments is available from 1989 to 2008 and collected from CANSIM II, Table 384-0001.

Debt charges. The exogenous spending commitment implied by each provincial government’s outstanding debt may also affect welfare generosity levels (Landon et al.

2006). To account for this possibility, the magnitude of this commitment is represented by real per capita expenditures on debt-servicing. Existing research across and within countries finds an inverse association between public debt and government expenditures. For example, Lora and Olivera (2007) assessed the effects of total public debt on social expenditure using an unbalanced panel of 50 countries between 1985 and 2003, and found that higher debt ratios reduce social expenditures. This finding has also been confirmed among Canadian provinces. Provincial debt charges appear to exert a negative impact on welfare generosity, and in particular, on health expenditures

(Landon et al., 2006). Debt charges are converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on debt charges is available from

1989 to 2008 and collected from CANSIM II, Table 385-0001.

Dependency ratio. I include a variable that represents the percentage of the provincial population that is under 18 and over 65 years of age. This demographic variable is expressed as the number of “dependents” for every 100 “workers”: youth

(ages 0 to 17) + seniors (age 65 or older) per 100 workers (aged 18 to 64). Higher

109 dependency ratios imply greater dependence on the labour force to generate government revenues. Controlling for this variable is important because a higher proportion of non-working dependents such as children and seniors may affect the demand for welfare services and social transfers. Increases in the size of children and senior age-groups are often associated with greater social and health needs and parallel increases in welfare spending (Landon et al., 2006; Pampel, 1992; Pampel &

Williamson, 1988; Wilensky, 1975). Data on the dependency ratio variable is available from 1976 to 2008 and collected from CANSIM II, Table 510-0001.

Women in labour force. The percentage of women in the labour force for each province is included as a control variable because of its mediating role between welfare generosity and population health (Navarro & Shi, 2001; Navarro et al., 2006; Raphael &

Bryant, 2004). Comparative research has found that generous welfare states (e.g., social democratic) tend to institute social service policies as a means of creating employment and facilitating the integration of women in the labour force. Such policies are hypothesized to have a positive effect on population health outcomes such as infant mortality rates (Navarro & Shi, 2001; Navarro et al., 2006). Data on the percentage of women in the labour force is available from 1976 to 2008 and collected from CANSIM II,

Table 282-0002.

Unemployment rate. This variable is the percentage of the total labour force unemployed for each province and included as a measure of the current state of the economy. Among studies using provincial time-series data, the association between unemployment and health has produced contradictory evidence, which may be due to modelling assumptions about age. Whereas Laporte and Ferguson (2003) found that

110 higher unemployment rates are associated with higher age-standardized mortality rates, other research has demonstrated a strong procyclical pattern in the mortality rates of middle-aged Canadians (Ariizumi & Schirle, 2012). The latter study found that a one percentage point increase in the unemployment rate actually lowers the predicted mortality rate of individuals in their 30s by almost 2% (Ariizumi & Schirle, 2012). Given that the current study does not estimate the effect of unemployment on age-specific mortality rates, I expect that unemployment to be positively associated with adverse population health outcomes (e.g., age-standardized mortality rates). Provincial data on unemployment rates is available from 1976 to 2008 and collected from CANSIM II,

Table 282-0002.

Low income. This variable refers to the percentage of all persons in low income after tax. Low income is a relative measure of low socioeconomic status, set at 50% of adjusted median household income, and is categorized according to the number of persons present in the household, reflecting the economies of scale inherent in household size. Adjusting for low income is essential given its causal association with poor living conditions, food insecurity, inadequate housing, and other basic prerequisites of population health (Auger & Alix, 2009). Provincial data on persons in low income is available from 1976 to 2008 and collected from CANSIM II, Table 202-

0802.

Real GDP per capita. This variable measures average real income per person for each province, calculated by dividing GDP by population. Provincial GDP is measured in constant 2002 dollars and deflated using the 2002 consumer price index. The market value of all goods and services produced in a given province is likely to be a principal

111 determinant of government expenditures and population health (Elola, Daponte, &

Navarro, 1995; Tresserras, Canela, Alvarez, Sentis, & Salleras, 1992). This variable reflects the magnitude of a province’s tax base, and controls for the overall level of economic development of a province for a given year. Real GDP per capita is converted into real per capita terms using data from CANSIM II, Table 510-001, and the consumer price index (CPI) (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on real GDP per capita and population estimates are available from 1976 to 2008 and collected from CANSIM II, Tables 384-0002 and 510-0001, respectively.

Several other social, demographic, and economic variables are tested or considered as controls but are not presented in final models. Urban population, immigrant population, net migration, and income inequality are omitted due to null results. Aboriginal and racial/ethnic population are omitted to due inconsistent and incomplete data points.

Urban population. The percentage of the population that resides in an urban area is tested as a control variable since living in an urban area has been associated with both favourable and adverse health outcomes (Vlahov & Galea, 2003). On one hand, urban residents tend to enjoy better health than their rural counterparts, which may be due to greater access to and availability of health and social services (Dye, 2008). Past research in Canada supports the idea that urbanization is a beneficial determinant of health, finding that rural Canadians experience poorer health outcomes than their urban peers (Pong, Desmeules, & Lagacé, 2009). On the other, health inequalities between the rich and poor are more pronounced within urban areas, and features of the social environment in cities, including increased density, poverty, crime and violence, are

112 influential sources of psychological stress and poor health. In this dissertation, however, urban population is omitted because bivariate and regression analyses produced null results. Non-significant findings might be explained by this dissertation’s ecological focus on provinces, and not on disaggregated units such as Metropolitan Influence

Zones (Pong, Desmeules, & Lagacé, 2009). Data on urban population is available from

1976 to 2008 and collected from CANSIM II, Table 153-0037.

Immigrant population. This variable refers to the immigrant population as a percentage of the population, and is calculated by dividing total number of immigrants by the population. Adjusting for the percentage of the immigrant population for each province is meant to control for the ‘healthy immigrant effect’, which refers to the well- established finding that immigrants’ health is generally better than that of the Canadian- born, although it tends to decline as their years in Canada increase (Chen, Ng, &

Wilkins, 1996; McDonald & Kennedy, 2004; Newbold, 2005; Ng, Wilkins, Gendron, &

Berthelot, 2005). This variable also controls for the uneven distribution of immigrants across provinces. Since 2001, the majority of the foreign-born population (86.8%) lives in three provinces: Ontario, Québec and British Columbia (Statistics Canada, 2006).

Immigrant population is excluded from analysis because exploratory and regression results indicated a non-significant association with population health. Possible reasons for null findings include inappropriate level of measurement (e.g., provincial versus

Census Metropolitan Areas) and unavailable immigrant characteristics (e.g., birthplace, period of immigration, and area of residence in Canada). Data on the immigrant population is available from 1976 to 2008 and collected from CANSIM II, Table 510-004.

113 Net migration. This variable refers to the net inter-provincial migration as a percentage of the population, and is calculated by dividing total net migrants by the population. From 2008 to 2009, approximately 277,800 persons changed their province or territory of residence within Canada, down from 301,200 the previous year, and the lowest number of interprovincial migrants since 2003/2004 (Milan, 2011). This variable is considered for its potential effects on social inequalities due to changes in population structure, and power resources due to changes in labour mobility. Bivariate and regression results did not find an empirical association between net migration and population health, and hence, this variable is omitted from final models. Data on net migration is available from 1976 to 2008 and collected from CANSIM II, Table 510-012.

Income inequality. This variable measures the relative degree of inequality in the distribution of income in each province. Even though research on income inequality is one of the most popular streams in population health (Kondo et al., 2009; Wilkinson &

Pickett, 2006, 2009) and Navarro et al. (2006) included this variable in their conceptual model, this dissertation omits income inequality due to non-significant results. I tested two measures of income inequality, Gini coefficients of market and after-tax for all family units, using bivariate and regression methods and found null results. These findings are consistent with previous research conducted at ecological levels (e.g., provinces)

(Laporte & Ferguson, 2003; Ross et al., 2000; Sanmartin et al., 2003). The only studies to find significant associations in the predicted direction between income inequality and have either been conducted: 1) at the community-level (Auger, Zang, & Daniel, 2009) or

2) with stratified samples between immigrants and Canadian-born individuals (Auger et al., 2012). Hence, null findings may be explained in part by this dissertation’s focus on

114 provinces and inability to stratify the sample by immigrants and non-immigrants. Data on income inequality is available from 1976 to 2008 and collected from CANSIM II,

Table 202-0705.

Aboriginal and racial/ethnic populations are not available as demographic controls due to a lack of comparable data over time. Because of changes in the ethnic question of the four censuses during the 1980s and 1990s, there are serious problems with the total counts of Aboriginal people, resulting in substantial differences between actual and estimated population figures (Saku, 1999). Over the past three decades, the census has defined Aboriginal in several incomparable ways: self-identification,

Aboriginal ancestry, and ‘registered Indian’ under the Indian Act (Saku, 2010). Likewise, the historical comparability of ethnic origin data has been affected by differences in the question wording, format, examples, and instructions of the ethnic origin questions in the census (Statistics Canada, 2006). For example, prior to 1996, data on visible minorities are derived from responses to the ethnic origin question, in conjunction with other ethno-cultural information, such as language, place of birth, and religion. Taken together, the comparability data issues for these two socially excluded group serves as a limitation. In Canada, existing studies find that Aboriginal and racial/ethnic groups are significantly more likely to experience a wide range of social and health disadvantages than non-Aboriginal and non-racial/ethnic populations (Mikkonen & Raphael, 2010).

Estimation and Post-Estimation Techniques

Estimation Techniques

115 All statistical analyses begin with a summary of descriptive statistics (mean values and total and within-province standard deviations) and correlation matrices to explore bivariate associations between independent, dependent, and control variables.

Following similar TSCS studies in population health (Conley & Springer, 2001; Fritzell et al., 2013; Laporte & Ferguson, 2003; Kangas, 2010; Lundberg et al., 2008), I use Beck

& Katz’s (1995) technique of Prais-Winsten regression with panel-corrected standard errors (PW-PCSE) and a first-order autocorrelation correction (AR(1)) and fixed-unit effects to estimate the connections between provincial welfare generosity, leftist politics, and population health. This model specification has the following notation:

where in equation (1):

= dependent variable (total, male, and female age-standardized mortality rate)

= independent variable (provincial welfare generosity and leftist politics)

= 1 …. N units (10 provinces)

= 1 …. T time points (calendar years from 1976 through 2008)

= 1 …. number of independent variables

= parameter estimates

= fixed-unit effects

= error term where in equation (1a):

116 = is a coefficient of first-order autoregressiveness

This model specification is selected in favour of other strategies for theoretical and econometric reasons (Beck, 2001; Plümper et al., 2005; Podestà, 2002, 2006).

First, PW-PCSE regression is the most appropriate approach to handle the “temporal dominant” nature of the data set. Given that my data set has more time periods (t = 19 –

32 years) than units (n = 10 provinces), existing work has shown that PW-PCSE are more efficient compared to other methods (e.g., feasible generalized least squares) and produces more accurate standard errors. PW-PCSE is also the recommended strategy to correct for two potential violations of OLS standard assumptions in TSCS data: 1) panel heteroscedasticity (e.g., errors tend to be heteroskedastic – that is, they tend to have different variances across units, for instance units with higher values may have a higher error variance), and 2) contemporaneous correlation of errors (e.g., errors tend to be correlated across units due to common exogenous shocks (Reed & Ye, 2011). Beck and Katz’s (1995) PCSE adopts robust standard errors in an OLS estimate. When computing the standard errors and the variance-covariance estimates, PCSE estimates assume that the disturbances are, by default, heteroskedastic and contemporaneously correlated across panels.

Two, equation (1a) notes that is a coefficient of first-order autoregressiveness

(AR(1) correction). The advantage of adding an AR(1) is that it addresses another potential violation of standard OLS assumptions about the error process: serial correlation of errors (e.g., errors tend to be autocorrelated in TSCS data, that is, they are not independent from one time period to the other) (Beck, 2001). Once serial correlation (dynamics) are accounted for, TSCS model parameters can be estimated

117 with PW-PCSE. Based on Monte Carlo simulations, Reed and Ye (2009) recommend using AR(1) when non-spherical errors are present and the primary concern is accurate inference. A common alternative is to include a lagged dependent variable (LDV) on the right-hand side of the equation; however, past research has shown that LDVs contribute to absorbing most the of the theoretically interesting time-series variance in the data

(Plümper et al., 2005). Modeling the dynamics in TSCS data by the LDV biases the estimate because any trend in the dependent variable will be absorbed by the theoretically uninteresting variables such as the LDV, the period dummies, and the intercept. Thus, AR(1) error models are more consistent, while the inclusion of LDV renders estimates necessarily inconsistent if fixed-effects are included (Wooldridge,

2002: 270).

Three, equation (1) also tells us that fixed-unit effects are added to the right side of the equation. To control for the possibility of non-spherical errors in the time and cross-sectional dimension, Beck and Katz (1996: 4) suggests using a two-way fixed effects model that includes dummies for N-1 units and T-1 periods. As equation (1) shows, my model specification corrects for the former non-spherical error but not the latter. The addition of unit-specific dummy variables acknowledges that there may be inherent features of individual province units that affect population health that are not adequately captured by any of the independent variables (regressors) included in the model. Fixed-unit effects models exclusively use the variance in the unit dynamics of the independent variables to estimate the dynamics of the dependent variable. By doing so, this strategy controls for the potential of unobserved heterogeneity or omitted variable bias. Regarding fixed-time effects, I exclude them for reasons similar to using

118 an AR(1) instead of a LDV. It has been shown that the inclusion of LDV and fixed-time effects absorbs much of the trend in the dependent variable, leaving very little variance for the explanatory variables (Plümper et al. 2005).

Fourth, equation (1) tells us that the dependent variable is measured in levels and changes are determined by long-term changes in the independent variable .

Comparative static models in levels are appropriate when theory suggests a dependent variable responds to long-term changes in independent variables (Garrett & Mitchell,

2001; Huber & Stephens, 2001). TSCS models can either specify dependent variables in terms of levels (i.e., population health levels) or short-term (i.e., population health differences from year to year) changes. Existing work has shown that the former approach meets the assumptions underlying welfare state development, and represents a good solution to understanding the long-term impact of political variables on social outcomes (Huber & Stephens, 2001). Hence, I apply the same logic and comparative static approach to population health that welfare state scholars have used to understand the impact of welfare state development on economic outcomes and social policies.

In order to support a model specification in levels (and not in changes), I explore whether the TSCS data is stationary by determining if a unit root is present (Im, Pesaran

& Shin, 1997). A unit root is an attribute of time-series models whose autoregressive parameter is one, or in other words, more than one trend is present in the time-series dimension (e.g., parameters such as means, variances, and autocovariances change over time). If a time-series has a unit root, it is deemed non-stationary, and as a result, estimating a model with serially correlated errors in the presence of a unit root can lead to spurious regressions and misleading results. I use the Im-Pesaran-Shin test (2003) to

119 detect for the presence of a unit root. This test estimates the t-test for unit roots in heterogeneous panels, and allows for individual effects, time trends, and common time effects. Based on the mean of the individual Dickey-Fuller t-statistics of each unit in the panel, the Im-Pesaran-Shin test assumes that all series are non-stationary under the null hypothesis. Table 4.1 reports the findings for various test specifications for total, male, and female age-standardized mortality rates in levels (located at end of chapter).

Findings show that ten out of the 12 test specifications rejected the null that the time- series contain a unit root (e.g., non-stationarity) at p < 0.05. The general conclusion from these tests is to treat the data as stationary and to examine mortality rates in levels.

Given these explanatory details, my approach is arguably the most appropriate to test this dissertation’s hypotheses (Beck, 2001; Plümper et al., 2005; Podesta, 2002,

2006). To demonstrate the robustness and convergence of my results, I also estimated the final models with a variety of alternative techniques: fixed- and random-effects, feasible generalized least squares, Driscoll Kray standard errors. Substantive results are largely consistent with these alternative estimation strategies.

Hierarchical Cluster Analysis

I use hierarchical cluster methods to identify distinct provincial regimes in chapter 7.

This method of cluster analysis uses a predetermined selection criterion to locate the closest pair of provinces and combines them to form a pair (Bambra, 2007a; Everitt,

Landau, & Leese, 2001; Gordon, 1999a). This process continues until all cases are grouped into a single cluster (e.g., joining cases into pairs or joining two pairs). Once

120 provinces are joined in a cluster, they remain joined throughout the rest of the analysis

(Cramer, 2003; Gough, 2001). In this way, the clusters emerge from the data, facilitating the emergence of provincial regimes. In recent years, similar methods have been used in population health research to identify distinct labour market regimes (Chung,

Muntaner, & Benach, 2010; Muntaner, Chung, Benach, & Ng, 2012).

Two methods are used to determine whether provinces form distinct regimes based on cumulative left party power and cumulative women in government. As shown in chapter 6, these two measures combine with provincial welfare generosity to lower mortality rates. First, I use a proximity matrix to perform a hierarchical cluster analysis, which holds the squared Euclidean distance between all variables. The dissimilarity between leftist variables by provinces is placed in a symmetric matrix. The element in the ith row and jth column displays the dissimilarity between the ith and jth variable.

Second, I produce a dendrogram to visually inspect potential clusters by provinces. A dendrogram is a tree diagram frequently used to illustrate the arrangement of clusters produced by hierarchical clustering.

Standardized and Semi-Standardized Coefficients

For each hypothesis tested in full models, I summarize results using standardized and semi-standardized coefficients for continuous and categorical independent variables, respectively. Standardized coefficients are calculated by multiplying the Prais-Winsten regression coefficient ( ) times the standard deviation of the independent variable and dividing by the standard deviation of the dependent variable. Semi-standardized coefficients are tabulated by dividing the Prais-Winsten regression coefficient ( ) by the

121 standard deviation of the dependent variable. By reporting these coefficients, one can compare the relative size of the effects of provincial welfare generosity and leftist politics on population health and interpret how many standard deviations population health would change if the political variable changed by one standard deviation.

Post-Estimation Tests

To ensure accurate model interpretations, after running each regression, I conduct three post-estimation tests. First, I test the residuals against the assumption of normality (e.g., the residuals are identically and independently distributed) with graphical (e.g., histograms, normality probably plots, and kernel density plots) and numerical methods

(e.g., skewness and kurtosis tests). Second, I check for multicollinearity, which occurs when more than two variables are near perfect linear combinations of one another. The concern is that as the degree of multicollinearity increases, the regression model estimates of the coefficients become unstable and the standard errors for the coefficients can get wildly inflated. I use variance inflation (VIF) scores to examine the assumption of no multicollinearity (e.g., a tolerance of less than 0.20 or 0.10 and/or a

VIF of 5 or 10 and above indicates a multicollinearity problem (O'Brien, 2007). Third, given that all specified models have a certain degree of built-in ‘uncertainty’, I conduct sensitivity analyses to determine whether observed results are robust. The analytical goal is to gain confidence that the observed results are significant in a variety of alternative specifications. To this end, I re-estimated all models using jackknife analysis

(Efron & Efron, 1982; Gould, 1995; Wu, 1986). In the jackknife analysis, each province is, in turn, taken out of the pool, and the remainder of the cases are run using the model

122 being tested to determine whether the results are driven by influential cases. This method involves omitting one province at a time. For example, to conduct a jackknife test for a regression using n provinces, I run n number of repeated regressions using subsets of the dataset with n-1 provinces, and pay special attention to changes in coefficients, especially the change in the direction (from positive to negative or vice- versa) of the associations to verify the stability of observed findings.

Correlation matrices and estimation and post-estimation techniques are conducted using Stata/SE 12 (StataCorp., 2011). Figures on standardized and semi- standardized coefficients are produced using Microsoft Excel (2010).

123 Table 4.1. Im, Pesaran, and Shin (2003) Unit Root Tests for Dependent Variables Total Age-Standardized Mortality Rates in Levels t-bar t-tilde-bar Z-t-tilde-bar cv5% p-value Not demeaned, no trend -1.42 -1.38 0.26 -1.98 0.60 Not demeaned, trend -5.08 -3.74 -9.15 -2.60 0.00 Demeaned, no trend -2.51 -2.23 -3.15 -1.98 0.00 Demeaned, trend -4.62 -3.53 -8.32 -2.60 0.00 Male Age-Standardized Mortality Rates in Levels t-bar t-tilde-bar Z-t-tilde-bar cv5% p-value Not demeaned, no trend -1.06 -1.04 1.60 -1.98 0.95 Not demeaned, trend -5.45 -3.89 -9.76 -2.60 0.00 Demeaned, no trend -2.75 -2.38 -3.74 -1.98 0.00 Demeaned, trend -5.01 -3.72 -9.07 -2.60 0.00 Female Age-Standardized Mortality Rates in Levels t-bar t-tilde-bar Z-t-tilde-bar cv5% p-value Not demeaned, no trend -2.11 -1.95 -2.02 -1.98 0.02 Not demeaned, trend -5.70 -3.97 -10.09 -2.60 0.00 Demeaned, no trend -3.27 -2.72 -5.07 -1.98 0.00 Demeaned, trend -5.26 -3.79 -9.36 -2.60 0.00 Notes: Im, Pesaran, and Shin (2003) (IMP) test with no lags included; t-bar = mean of province-specific Dickey-Fuller tests; t-tilde-bar = different estimator of the Dickey–Fuller regression error variance ; Z-t-tilde-bar = transformed t-bar statistic, asymptotic standard normal distribution, Ho: non-stationarity; cv5% = critical value of IPS test; p-value: significance level of test against H; Ho: All panels contain unit roots; Ha: Some panels are stationary; Number of panels = 10; Number of periods = 33.

124 CHAPTER FIVE. PROVINCIAL WELFARE GENEROSITY AND POPULATION HEALTH

Introduction

An emerging body of research in comparative social epidemiology uses an expenses approach to investigate the relationship between welfare generosity and population health (Bradley et al., 2011; Chung & Muntaner, 2006; Conley & Springer, 2001; Dahl & van der Wel, 2013; Dunn et al., 2005; Kangas, 2010; Lundberg et al., 2008). Following the tradition of Wilensky and Lebaux (1965), the expenses approach uses information on public spending on social purposes to quantitatively gauge the extensiveness of welfare generosity, and its impact on population health (Dahl & van der Wel, 2013).

Studies using an expenses approach find strong evidence that countries with more generous welfare programs and more extensive welfare services have better health outcomes than countries with residual welfare states (Bradley et al. 2011; Brennenstuhl et al., 2012; Conley & Springer, 2001; Lundberg et al., 2008; Kangas, 2010). The general conclusion reached is that welfare state generosity serves as a proximate determinant of national levels of population health. An exception to this general rule is the US. In 2005, the US spent more than 7% of its GDP on health care compared to other wealthy nations (16% vs. 9%); however, the US still ranked 25th in life expectancy and 29th in infant mortality among 30 wealthy countries (OECD, 2009).

How does welfare generosity improve population health according to the expenses approach? According to this view, government expenditures are health- promoting ‘welfare resources’. The concept of welfare resources is rooted in the idea that collectively provided services and transfers: 1) ensure that individuals and families

125 maintain a socially acceptable standard of living, and 2) compensate for market failures or family failures or both, which taken together, improves the health of the population

(Dahl & van der Wel, 2013; Fritzell & Lundberg, 2007; Lundberg et al., 2008). This idea is in part a straightforward extension Sen’s (1993) ‘capability’ approach and Titmuss’s concept of ‘command over resources’ (Titmuss, Abel-Smith, & Timuss, 1974).

According to Sen (1993), capabilities refer to the opportunities that individuals have available to them to lead lives of their own choosing. Hence, poor health arises when one is deprived of basic capabilities and when basic capability failure is caused by an inadequate ‘command over resources’, whether through markets, public provision, or non-market channels (Titmuss et al., 1974). Following the work of Titmuss et al. (1974),

Dunn et al. (2005) introduced the idea of ‘real income’ (or ‘effective income’) to population health. By definition, real income refers to ‘‘all receipts which increase an individual’s command over the use of a society’s scarce resources’’ (as cited in Dunn et al., 2005, p. 768). It follows that real income matters to health because it represents a person’s command over resources, whether or not these welfare resources are purchased with cash incomes or provided in the form of public goods and services by the welfare state (Dunn et al., 2005).

In several important respects, welfare resources are akin to SDOH, in that, welfare generosity embodies a wide array of health-promoting resources in cash or in kind, including for example, money, knowledge, prestige, power, and beneficial social connections (Eikemo & Bambra, 2008; Link & Phelan, 1995; Lundberg et al., 2008).

Such welfare resources are hypothesized to have a salutary effect on population health through multiple mechanisms, including for example, the provision of essential welfare

126 services and social transfers, the expansion of the capability to cope with stressful events, and the building human capital (Dahl & van der Wel, 2013; Eikemo & Bambra,

2008).

One of the most important contributions of the welfare generosity literature is to show what kinds of expenditures are significantly associated with population health.

Scholars interested in the relation between government expenditures and population health have studied total government spending (Dunn et al., 2005; Lena & London,

1993; Xu, 2006; Veenhoven & Ouweneel, 1995), health-related expenditures (Arah &

Westert, 2005; Arah et al., 2005; Bradley et al., 2011; Chung & Muntaner, 2006; Dunn et al., 2005; Elgar, 2010; Fayissa, 2001; Filmer & Pritchett, 1999; Laporte & Ferguson,

2003; Liu et al., 2008; Navarro et al., 2003, 2006; Novignon et al., 2012; Rodríguez-

Sanz et al., 2003; Schell et al., 2007), and expenditures related to social concerns

(Bradley et al., 2011; Chung & Muntaner, 2006; Dahl & van der Wel, 2013; Dunn et al.,

2005; Kangas, 2010; Ferrarini & Norstrom, 2010; Lundberg et al., 2008; Muntaner et al.,

2002; Navarro et al., 2003, 2006; Norstrom & Palme, 2010; Rodríguez-Sanz et al., 2003;

Ronzio, 2003; Ronzio et al., 2004; Stuckler et al., 2009; Veenhoven & Ouweneel, 1995;

Veenhoven, 2000).

Despite these contributions, less work has examined the population health effects of welfare generosity within nations and the impact of disaggregated expenditures. First, researchers have almost exclusively examined the impact of welfare generosity at the national level among OECD countries even though significant variations exist within welfare states. For example, Canadian provinces differ widely in the extent to which they redistribute income, protect the vulnerable against economic

127 insecurity, and spend on essential welfare services. Hence, more work is needed to determine whether the relationship between welfare generosity and population health also applies among sub-national jurisdictions. Second, existing studies have primarily conceptualized welfare generosity using aggregate-level expenditures while overlooking the health effects of disaggregated types of spending. For example, spending on health, an aggregate level expenditure, can be further divided into sub-functions such as hospital care, medical care, preventive care, and other health services. The consequence of relying on aggregate measures is that one overlooks the potential causal links between specific expenditures and population health.

This chapter addresses these two limitations by taking advantage of provincial differences within Canada to test whether variations in aggregate and disaggregated types of provincial welfare generosity are associated with population levels of health, while controlling for alternative explanations. The three main questions addressed in this chapter are:

1. At an aggregate level, does provincial welfare generosity affect population health,

net of province-specific factors? If so, what kinds of expenditures matter - health,

social services, or education?

2. At a disaggregated level, which expenditures are predictive of lower total, male,

and female mortality rates – hospital, medical, preventive, other health, social

assistance, workers’ compensation, other social services, elementary/

secondary, post-secondary, special retraining, or other education, net of

province-specific factors?

128 3. What is the overall health effect of a provincial welfare generosity index, an index

that combines significant predictors of improved health at the disaggregated level

into a single measure?

Data and Methods

This study uses a pooled TSCS analysis of Canadian provinces from 1989 to 2008, including Newfoundland, Prince Edward Island, Nova Scotia, New Brunswick, Quebec,

Ontario, Manitoba, Saskatchewan, Alberta, and British Columbia. Data are retrieved from Statistics Canada and its Canadian Socio-Economic Information Management

System (CANSIM) Tables. This sample includes 200 province-years as cases.

Statistical analysis consists of five steps. First, I present mean values and standard deviations (total and within-province) for the variables used in the analysis.

Second, I display a correlation matrix to display the basic associations between explanatory, dependent, and control variables. Third, following the rationale outlined in chapter 4, I use Beck and Katz’s (1995) technique for calculating panel-corrected standard (PCSE) estimates for linear TSCS models where the parameters are estimated by Prais-Winsten regression with a first-order autocorrelation correction

(AR(1)) and fixed-unit effects (e.g., provincial dummy variables). In striving to keep model specifications as parsimonious as possible, I test one aggregate or disaggregated measure of provincial welfare generosity at a time in each regression model. As a consequence, my analysis only determines which expenditures are important for population health, and not necessarily which of the different expenditures are relatively more important than the others.

129 Fourth, regression estimates are converted into standardized coefficients and reported in bar charts (i.e., each bar represents an aggregate or disaggregated expenditure in separate regression models that adjust for alternative explanations) (see figures 5.1, 5.2, and 5.3 located at end of chapter). Each bar represents how many standard deviations mortality rates would change if the welfare generosity expenditure changed by one standard deviation. Fifth, post-estimation techniques are conducted to check for normality of residuals (e.g., histograms and skewness and kurtosis tests), multicollinearity of predictors (e.g., variance inflation scores), and robustness of results

(e.g., jackknife analyses). All analyses are conducted using Stata/SE 12.1.

Dependent Variables

I focus on total, male, and female age-standardized mortality rates. The outcome variable is the weighted average of the age-specific mortality rates per 1,000 persons, where the 1991 Canadian Census of Population is used as the standard population for the calculation. Data is available from 1989 to 2008 for a total of 200 cases.

Independent Variables

The effect of provincial welfare generosity on population health is measured at two levels: aggregate and disaggregated. At the aggregate level, I examine three general measures of welfare spending: health, social services, and education. Health expenditure includes provincial spending made to ensure that necessary health services are available to all citizens. Social service expenditures cover government actions, either alone or in co-operation with the citizenry, to offset or to forestall situations where

130 the well-being of individuals or families is threatened by circumstances beyond their control. Education expenditure includes the costs of developing, improving and operating educational systems and the provision of specific education services.

At the disaggregated level, I test the health effects of eleven expenditures.

Hospital care covers outlays with respect to all kinds of hospital services. Medical care comprises outlays in respect of general medical care and drug programs as well as outlays incurred for dental and visiting-nurse services and on out-patient care services.

Preventive care expenditures intended to prevent the occurrence of diseases and to mitigate their effect. Other health services includes outlays on clinics for the treatment of mental disabilities or emotionally disturbed persons and on laboratory and diagnostic services, grants to health-oriented organizations, and expenditures on other health- related services.

Social assistance consists of transfer payments to help individuals and families maintain a socially acceptable level of earnings. Workers’ compensation benefits refer to expenditures on administration and for benefits, other than rehabilitation and medical care, related to workers’ compensation schemes. Other social services accounts for expenses related to the provision of services to old age, to persons who are unable to lead a normal life due to a physical or mental impairment, to persons temporarily unable to work due to sickness, to households with dependent children, to persons who are survivors of a deceased person and to other needy persons.

Elementary and secondary education encompasses outlays for educational services from kindergarten to senior matriculation. Post-secondary education refers to the kind of education generally obtained in universities or in degree and non-degree

131 granting community colleges and specialized educational institutions. Special retraining services comprise outlays made for the purpose of upgrading the skills of individuals.

Other education covers outlays that either overlap or cannot be allocated to the other sub-functions.

Provincial welfare generosity index is constructed by averaging the standard scores (z-scores) of three disaggregated expenditures: medical care, other social services, and post-secondary education. As shown below, these three measures are the significant predictors of lower for total, male, and female mortality rates. This index combines them into one variable, and becomes a general measure of welfare generosity.

Control Variables

In all models, I control for several provincial, demographic, labour market, and economic factors. Transfers refers to the current fiscal year dollar value of total transfers, which includes the sum of general and specific purpose transfers, received from other levels of government. Debt charges include each provincial government’s outstanding debt.

Dependency ratio is measured as the number of “dependents” for every 100 “workers”: youth (ages 0 to 17) + seniors (age 65 or older) per 100 workers (aged 18 to 64).

Women in labour force refers to the percentage of women in the labour force.

Unemployment rate is the percentage of the total labour force unemployed. Low income refers to the percentage of all persons in low income after tax, set at 50% of adjusted median household income. Real GDP per capita is the gross domestic product in real per capita terms and deflated using the 2002 consumer price index.

132

Results

Table 5.1 provides the mean values and standard deviations (total and within-province) for the variables used in this study (located at end of chapter). The average age- standardized mortality rate (per 1,000 deaths) is higher among males (mean [M] = 8.26,

Within-group standard deviation [WI-GRP SD] = 0.85] than females (M = 5.08, WI-GRP

SD = 0.34). At the aggregate level, health spending per capita is the leading expenditure (M = 2,226.87, WI-GRP SD = 421.05), followed by education (M =

1,809.95, WI-GRP SD = 328.65), and social services (M = 1,123.09, WI-GRP SD =

154.68). The largest disaggregated expenditures per capita within health, social services, and education are, respectively, medical care (M = 897.91, WI-GRP SD =

192.95), social assistance (M = 451.34, WI-GRP SD = 89.27), and elementary school

(M = 983.68, WI-GRP SD = 279.19).

To make these measures more concrete, in table 5.2, I rank provinces from one to ten across total mortality rates, aggregate-level measures, and the provincial welfare generosity index (located at end of chapter). As this table shows, cross-provincial variations exist in levels of population health and welfare generosity. Whereas the lowest mortality rates are among British Columbia, Alberta, and Ontario, the highest rates are concentrated among Atlantic provinces (e.g., Prince Edward Island, Nova

Scotia, and Newfound). New Brunswick, Newfoundland and British Columbia are the only three provinces to spend more than $2,300 per capita on health expenditures. In contrast, Prince Edward Island is the only province to allocate less than $2,000 per capita. In terms of average social service expenditures, Quebec ranked first

133 ($2,044.01), and actually spent 71% more than the second ranked province, Ontario

($1,194.92). Expenditures on education are highest in Newfoundland ($2172.92) and

Alberta ($2041.15), and lowest in Manitoba ($1559.50) and Ontario ($1476.36). With respect to the provincial welfare generosity index, which combines medical care, other social services, and post-secondary education, Quebec, British Columbia, and Alberta ranked the top three, and Manitoba, New Brunswick, and Prince Edward Island rounded out the bottom.

Bivariate Patterns

Table 5.3 displays a correlation matrix between welfare expenditures, total, male, and female mortality rates, and control variables (located at end of chapter). Regarding aggregate-level expenditures, a strong negative and significant relationship exists between health expenditures and total, male, and female age-standardized mortality rates (total, r(200) = -.60, p < .01; male, r(200) = -.65, p < .01; female r(200) = -.44, p <

.01). Expenditures on social services and mortality are negatively associated for total rates, r(200 = -.17, p < .05, and women, r(200) = -.16, p < .05), but not among men, r(200) = -.12, p = n.s). Education spending is negatively correlated with total, r(200) = -

.19, p < .05, and male mortality rates, r(200) = -.22, p < .05; but not among women, r(200) = -.11, p = n.s).

In terms of disaggregated expenditures, I summarize the correlations only for total mortality rates since results are consistent among males and females. Mortality rates are negative correlated with medical expenditures (total, r(200) = -.56, p < .01), preventive spending (total, r(200) = -.27, p < .01), and other health disbursements (total,

134 r(200) = -.47, p < .01). The link between hospital expenditures and mortality rates is non-significant, r(200) = -.12, p = n.s).

Disaggregated social service expenditures shows a negative relationship between mortality rates and worker’s compensation benefits (total, r(200) = -.38, p <

.01) and other social service expenditures (total, r(200) = -.41, p < .01). Conversely, social assistance is positively associated with mortality rates (total, r(200) = .20, p <

.01).

Disaggregated education expenditures reveals that mortality rates are negatively correlated with expenditures on post-secondary education (total, r(200) = -.55, p < .01) and special retraining services (total, r(200) = -.29, p < .01). Whereas no association exists between elementary/secondary school spending and mortality rates (total, r(200)

= -.03, p = n.s.), other education expenditures is positively correlated with mortality

(total, r(200) = .24, p < .01). Overall, the provincial welfare generosity index, comprised of medical, other social services, and post-secondary expenditures, shows the strongest negative association with lower mortality rates (total, r(200) = -0.66, p < .01).

Correlations between control variables and mortality rates confirm their hypothesized effects. Higher mortality rates are significantly associated with transfers

(total, r(200) = .41, p < .01), debt (total, r(200) = .41, p < .01), unemployment (total, r(200) = .69, p < .01), and low income (total, r(200) = .25, p < .01). In contrast, mortality rates are negatively correlated with female labour force participation r(200) = -.68, p <

.01) and GDP per capita r(200) = -.62, p < .01). Dependency ratio is positively associated with total, r(200 = .25, p < .01, and male mortality rates, r(200) = .27, p <

.01), but not among women, r(200) = .13, p = n.s).

135 Taken together, these bivariate patterns suggest a strong association between provincial welfare generosity and population health. Below, I conduct a more rigorous test to determine if indeed provincial welfare generosity reduces mortality rates, net of other explanations.

Overall Aggregate Models

The unstandardized Prais-Winsten regression coefficients ( ) with panel corrected standard errors (PCSE), and standardized coefficients (β) for aggregate expenditures for total, male, and female mortality rates are reported in Tables 5.4a, 5.4b, and 5.4c, respectively (located at end of chapter). These models tested separately if health

(models 1), social services (models 2), and education (models 3) expenditures significantly predicted age-standardized mortality rates while controlling for transfers, debt, dependency ratio, female labour force participation, unemployment, low income, and GDP per capita. Separate model results are also visually displayed as standardized coefficients in figures 5.1, 5.2, and 5.3, where each bar represents how many standard deviations mortality rates would change if the welfare generosity expenditure changed one standard deviation (located at end of chapter).

Health expenditures significantly predicted lower total (β = -.510, z(17) = -3.90, p < .01) and male (β = -.585, z(17) = -4.82, p < .01) mortality rates; however, no significant association is observed between health spending and female mortality rates.

For a standard deviation increase in health expenditures ($421.05), total and male mortality rates should decline by 0.51 (0.27 deaths per 1,000) and 0.58 (0.50 deaths per

1,000) standard deviations, respectively.

136 There is a significant negative effect between social service expenditures and total and female mortality rates (total: β = -.136, z(17) = -2.72, p < .01; female: β = -.158, z(17) = -2.66, p < .01). A similar impact is not observed among male mortality rates. A hypothetical increase in social service spending of $154.68 (one standard deviation) would lower total mortality rates by 0.07 deaths per 1,000 and female rates by 0.05 deaths per 1,000.

In terms of aggregate levels of education expenditures, this generosity measure is only significant in lowering male mortality rates (β = -.154, z(17) = -2.35, p < .05). If education spending increased one standard deviation ($391.35), male mortality rates should decline by 0.13 (per 1,000 deaths). The z-scores of education expenditures for total mortality rates only reached significance at the .10 level (β = -.119, z(17) = -1.74, p = n.s), and is non-significant for women (β = -.044, z(17) = -0.49, p = n.s).

Overall Disaggregated Health Models

Disaggregated health expenditures for total, male, and female mortality rates are reported in Tables 5.5a, 5.5b, and 5.5c, respectively (located at end of chapter).

Mortality rates are modeled as separate functions of provincial expenditures on hospital care (models 1), medical care (models 2), preventive care (models 3), and other health services (models 4).

Spending on hospital care had no significant effect on total, male, or female mortality rates. Medical care expenditures are significantly negative for all three mortality outcomes (total: β = -.361, z(17) = -3.28, p < .01; male: β = -.308, z(17) = -

2.77, p < .01; female: β = -.354, z(17) = -2.64, p < .01). These results suggest that a

137 standard deviation increase in medical care spending ($230.07) should lower total, male, and female mortality rates by 0.20, 0.26, and 0.12 per 1,000 deaths.

Spending on preventive care had a significant negative effect on total and female mortality rates (total: β = -.100, z(17) = -2.26, p < .05; female: β = -.087, z(17) = -2.09, p < .01); however, preventive care spending is significant only at the .10 level among men (β = -.087, z(17) = -1.90, p < .10). If preventive care spending is increased one standard deviation ($43.38), total and female mortality rates should decline by 0.05 and

0.02 per 1,000 deaths. Admittedly, the health benefits yielded by preventive care are comparatively smaller than other disaggregated health expenditures.

Expenditures on other health services are a significant predictor of lower male mortality rates (β = -.326, z(17) = -2.83, p < .01), but not for total and female mortality outcomes. An increase of $224.74 on other health services would lower male mortality rates by 0.28 per 1,000 deaths.

Overall Disaggregated Social Service Models

Respectively, disaggregated social service expenditures for total, male, and female mortality rates are presented in Tables 5.6a, 5.6b, and 5.6c (located at end of chapter).

Expenditures on social assistance (models 1), worker’s compensation benefits (models

2), and other social services are separately regressed on age-standardized mortality rates.

Results indicate that social assistance and worker’s compensation benefits had no significant effect on mortality rates. In contrast, spending on other social services is significantly negative for total (β = -.217, z(17) = -3.73, p < .01), male (β = -.157, z(17) =

138 -2.58, p < .01), and female (β = -.245, z(17) = -3.71, p < .01) mortality rates. For a standard deviation increase in expenditures on other social services, total, male, and female mortality rates should decline by 0.11, 0.13, and 0.08 per 1,000 deaths, respectively.

Overall Disaggregated Education Models

Disaggregated education expenditures for total, male, and female mortality rates are reported in Tables 5.7a, 5.7b, and 5.7c, respectively (located at end of chapter). These models examine the relative importance of elementary/secondary (models 1), post- secondary education (models 2), special retraining (models 3), and other education

(models 4) expenditures in predicting age-standardized mortality rates.

Provincial spending on elementary/secondary school had no significant impact on all three mortality outcomes. In all models, expenditures on post-secondary are found to be significant predictors of lower mortality rates (total: β = -.339, z(17) = -3.29, p < .01; male: β = -.283, z(17) = -2.70, p < .01; female: β = -.328, z(17) = -2.49, p < .01).

A hypothetical increase of $139.26 in post-secondary education spending should lower total, male, and female mortality rates by 0.18, 0.24, and 0.11 per 1,000 deaths.

No significant effect is observed between spending on special retraining and total, male, or female mortality rates. Unexpectedly, other education spending had a significant positive effect on mortality rates (total: β = .131, z(17) = -3.09, p < .01; male:

β = .140, z(17) = 3.03, p < .01); female: β = .150, z(17) = 3.13, p < .01). Although, this variable is found to be significant in an unpredicted direction, its hypothetical impact is comparatively small. For example, a standard deviation increase in other education

139 spending ($14.35) would increase total, male, and female mortality rates by 0.07, 0.11, and 0.05 per 1,000 deaths, respectively.

Provincial Welfare Generosity Index

As revealed in the above analyses, three types of disaggregated expenditures are predictive of total, male, and female mortality rates: medical care, other social services, and post-secondary education. Given their significant and negative impact, I combine these variables by averaging their z-scores to create a provincial welfare generosity index. Table 5.8 displays the results of the Prais-Winsten regressions for the provincial welfare generosity index (located at end of chapter). Total, male, and female mortality rates are regressed on control variables in models 1-3. Results indicate that the provincial welfare generosity index does have a significant impact in lower mortality rates (total: β = -.507, z(17) = -4.87, p < .01); male: β = -.401, z(17) = -3.56, p < .01); female: β = -.534, z(17) = -4.10, p < .01), net of other control variables, and this effect, measured as standardized coefficients, is actually larger than the disaggregated indicators of the index: medical care, other social services, and post-secondary education. For a standard deviation increase in the provincial welfare generosity index, total, male, and female mortality rates is expected to decline by about per 0.27, 0.34, and 0.18 per 1,000 deaths.

Post-estimation Results

An examination of the skewness values and visual inspection of frequency distributions showed that the distributions of the variables are normally distributed. VIF scores

140 confirmed the absence of multicollinearity and jackknife analyses replicated the substantive results (see Appendices B1 – B5 for jackknife results). Moreover, re- analyzing the data using alternative estimation models did not substantially change significant results.

Discussion

This chapter’s aims are two-fold: first, to assess whether aggregate levels of provincial welfare generosity affected population health, and second, to determine which specific disaggregated indicators of spending predicted lower total, male, and female age- standardized mortality rates. Overall, the analyses support the welfare resources perspective and to my conceptual model’s argument that provincial welfare generosity operates as a proximate determinant of population health.

According to the welfare resources perspectives, collectively provided resources has the effect of improving population health by compensating for market or family failures or both when individuals and families fall below socially acceptable standards of living (Dahl & van der Wel, 2013; Fritzell & Lundberg, 2007; Lundberg et al., 2008).

Welfare resources are hypothesized to affect population through multiple mechanisms, including for example, providing essential welfare services and social transfers, expanding the capacity to cope with stressful events, and building human capital (Dahl

& van der Wel, 2013; Eikemo & Bambra, 2008). This chapter finds empirical support for each of these predictions at aggregate and disaggregated levels of provincial welfare generosity.

At the aggregate level, the analyses reveal that provincial expenditures on

141 health, social services, and education are significant predictors of lower mortality rates.

These connections, however, are dependent upon expenditure type and health outcome considered. Health spending is associated with significant decreases in total and male mortality rates but not among female mortality rates. Of the three aggregate measures, health expenditures yielded the greatest returns to population health. For a standard deviation increase in health spending, for example, total mortality rates should decline by more than 0.5 standard deviations. This finding confirms previous studies that have examined the topic (Arah & Westert, 2005; Arah et al., 2005; Bradley et al., 2011; Elgar,

2010; Fayissa, 2001; Filmer & Pritchett, 1999; Navarro et al., 2003, 2006; Novignon et al., 2012; Rodríguez-Sanz et al., 2003; Schell et al., 2007), and in particular, the work of

Laporte and Ferguson (2003), who found the same association among Canadian provinces (e.g., real per capita spending on total health care and age-standardized total all-cause mortality). Moreover, this supports the idea that generous health expenditures operate as a SDOH by ensuring that necessary and high quality health care services are available to all citizens. Within Canada, this implies that health spending is essential to the country’s universal health care system, which meets the health needs of citizens, spreads health costs across the whole society, and protects citizens with lower incomes who cannot afford private health care insurance (Romanow, 2002).

Social services also had a differential effect on mortality rates by gender.

Spending on social services lowered total and female mortality rates but had no such effect on male outcomes. For a standard increase in social service expenditures, total and female mortality rates should decline by 0.14 and 0.16 standard deviations, respectively. Social services refer to benefits, programs, and supports that are designed

142 to protect individuals and families during important life changes. Because social services are an essential component of Canada’s overall social safety net, it makes intuitive sense that social services have a positive effect on population health. This underscores the health-promoting role played by provincial governments to offset or forestall situations where the well-being of individuals and families are negatively affected by circumstances beyond their control (Bryant, 2009; Hallstrom, 2009).

Interestingly, aggregate-levels of education spending only had a significant negative effect on male mortality rates. For a standard deviation increase in education spending, male mortality rates should decline by 0.15 standard deviations. The established importance of education as a SDOH provides insightful clues on how increased expenditures on education might reduce male mortality rates, including for example, building human capital, increasing social mobility, shaping future occupational opportunities, raising earning potential, promoting health literacy, and providing access to other essential SDOH such as employment security and working conditions

(Galobardes et al., 2006; Mikkonen & Raphael, 2010; Ronson & Rootman, 2009).

Because education is a major determinant of population health, this idea can broadened to suggest that generous investments in education over the life-course, beginning with elementary and secondary schooling, extending to post-secondary education, and continuing with skills retraining, increases the socio-economic position of individuals and families and results in better health outcomes.

Contrary to expectations, some aggregate-level expenditures had no statistically significant impact on population health. Health spending is not associated with female mortality rates, expenditures on social services had no impact on male mortality rates,

143 and education spending is insignificant with respect to total and female mortality rates.

These results, however, should be interpreted with caution. One likely explanation for insignificant findings is that aggregate expenditures may be too broad to capture the precise links between provincial welfare generosity and population health. For example, aggregate findings suggest that health expenditures are not associated female mortality rates; however, once health expenditures are disaggregated into its sub-functions, a more detailed and nuanced association emerged. Disaggregated analyses found that medical care and preventive care are both associated with lower female mortality rates.

This endorses the methodological approach taken here, which augments aggregate indicators of welfare generosity with lower levels of government spending.

Regarding this chapter’s second aim, significant predictors of lower total and female mortality rates included four expenditures: medical care, preventive care, other social services, and post-secondary education. For example, a standard deviation increase in medical care would reduce total and female mortality rates by 0.36 and 0.35 standard deviations, respectively. In terms of male mortality rates, these disaggregated expenditures are significant with the exception of preventive care. Instead of preventive care, other health services are associated with lower male mortality rates. For a standard deviation increase in other health services, male mortality rates should decline by 0.33 standard deviations.

Because expenditures on medical care, other social services, and post- secondary education are found to be significant and negative predictors of lower mortality rates across all three health outcomes, these three spending measures are combined to construct an overall provincial welfare generosity index. Overall, this

144 combined index had a larger negative effect on mortality rates than any of the individual measures used to construct the index. For a standard deviation increase in the provincial welfare generosity index, for example, total, male, and female mortality rates should decline by 0.51, 0.40, and 0.53 standard deviations. Clearly, provincial levels of welfare generosity have a significant impact in improving population health (lowering all- cause age-standardized mortality rates).

Disaggregated analyses confirm that provincial welfare generosity has a salutary effect on population health through multiple expenditures, and in particular, the provision of welfare services (e.g., medical care), the expansion and protection of capabilities (e.g., other social services), and the building of human capital (e.g., post- secondary education). First, welfare services such as medical care clearly matters for population health. When high-quality medical care is available to citizens, mortality rates decline. This finding reinforces the role of provinces in improving population health by delivering superior general medical care, drug programs, dental and visiting-nurse services, and out-patient care services (Bunker, 2001). Second, as provinces invest more in other social services, population health tends to improve. Other social services are primarily directed toward vulnerable groups such as the elderly, disabled, and ailing as well as to households with dependent children and persons who are survivors of a deceased person. Continued investments in these key services appear to be a good strategy to strengthen social safety nets, further protect vulnerable groups, and improve population levels of health (Hallstrom, 2009). Third, the analysis here also shows that increased spending on post-secondary education is associated with lower mortality rates. Education is widely acknowledged as an important SDOH (White, Blane, Morris,

145 & Mourouga, 1999; Morris, Blane, & White, 1996); however, it seems that expenditures on post-secondary education in particular yield the greatest returns to population health.

This finding is consistent with the work of Dunn et al. (2005), who also found a strong inverse association between higher education and mortality among US states. The health effects of higher education may be explained in part by human capital theory

(Becker, 2009). It follows that that post-secondary education in an investment that builds capital in the form of skills and knowledge, which in turn, improves health through future employment, higher income, and increased cultural literacy (Smith et al., 1998;

Kelleher, 2002; Kaufman, 2002; Lynch & Kaplan, 2000; Mirowsky & Ross, 2003; Ross &

Wu, 1995). Another possible explanation is that increased investments in post- secondary training increases the number of educational opportunities available to young adults, which has the unintended effect of minimizing their premature mortality risks due to accidents, suicides, and homicides (Dunn et al., 2005).

There is at least one surprise regarding the hypothesized impact of provincial welfare generosity on population health. Spending on ‘other education’ is positively and significantly associated with total, male, and female mortality rates. Expenditures on

‘other education’ include a range of unrelated education expenses such as administrative aspects of education departments, research activities, apprenticeship training, official language training, evening classes, and correspondence courses. It is difficult to explain this result. A careful search of the extant literature on the possible link between these education expenses and population health failed to produce any explanatory clues. Given the novelty of this finding and its lack of theoretical support, more work is need to determine whether this association is robust (e.g., other education

146 spending increases mortality rates) or selective (e.g., improving health leads to increases on other education expenditures) or spurious (e.g., type II error, or acceptance of a false relation) or artefactual (e.g., the grouping of various unrelated education expenditures amounts to measurement error). With this said, it is worth noting that the counter beneficial effect of other education expenditures on health is comparatively small. For a standard deviation increase on other education spending, total, male, and female mortality rates would increase by 0.13, 0.14, and 0.15 standard deviations.

In sum, this chapter finds that aggregate and disaggregated levels of provincial welfare generosity are significant macro-social determinants of population health among

Canadian provinces. The results allow for an interpretation that provincial expenditures are health-promoting welfare resources which ensure citizens possess basic capabilities and an adequate command over resources. In the next chapter, I build on these results and examine the population health effects of power resources, political parties, and democratic politics given that so much political sociology research has confirmed their causal impact on welfare state development and generosity (Hewitt, 1977; Hicks &

Misra, 1993; Hicks & Swank, 1984; Korpi, 1983; Stephens, 1979).

147

Table 5.1. Summary Statistics, 1989-2008 Variable Mean Total Within-Group Min Max Std. Dev. Std. Dev. Dependent Total age-standardized mortality rateb 6.49 0.67 0.54 4.7 8.2 Male age-standardized mortality rateb 8.26 1.02 0.85 4.7 10.5 Female age-standardized mortality rateb 5.08 0.44 0.34 4.2 6.3

Provincial Welfare Generosity c Aggregate Expenditures Real health expenditure per capita 2226.87 436.67 421.05 1252.59 3624.89 Real social services expenditure per capita 1123.09 375.54 154.68 563.66 2711.82 Real education expenditure per capita 1809.95 391.35 328.65 1169.76 5667.58 Disaggregated Expenditures Real hospital care per capita 871.29 145.71 120.40 546.09 1538.50 Real medical care per capita 897.91 230.07 192.95 484.1 1523.91 Real preventive care per capital 49.94 43.38 26.93 -105.8 192.29 Real other health per capita 407.70 224.74 197.67 52.89 1029.15

Real social assistance per capita 451.34 124.53 89.27 234.96 811.23 Real worker’s compensation per capita 156.31 42.97 21.56 86.47 245.68 Real other social services per capita 424.41 147.80 118.22 40.23 1201.13

Real elementary per capita 983.68 343.44 279.19 540.74 4713.73 Real post-secondary per capita 749.94 139.26 118.88 448.07 1203.25 Real special education per capital 50.46 43.42 26.06 0 171.88 Real other education per capital 25.73 14.35 10.70 4.50 65.75

Provincial Welfare Generosity Indexd 1.32e-07 0.77 0.70 -1.85 2.48 (Continued on next page)

148 Table 5.1. Summary Statistics: 1989-2008a (Continued) Variable Mean Total Within-Group Min Max Std. Dev. Std. Dev. Controls Real cash transfers per capita 1938.92 1000.64 385.83 487.91 7272.82 Real debt per capita 968.79 335.98 197.31 135.92 1981.13 Dependency ratio 57.90 5.81 2.91 48.96 73.86 Female labour force 58.55 4.67 2.34 46.31 67.62 Unemployment 9.55 3.82 1.79 3.4 20.1 Low income 14.15 2.96 1.47 7 22 Real GDP per capita 31974.04 8502.49 5804.34 19297.4 66100.1 Notes: a Number of observations = 200 (10 provinces over 20 years). b per 1,000 deaths. c All expenditure data is converted into real per capita terms and with the consumer price index (CPI) (2002 = 100) for each province. d Provincial welfare generosity index is the average standard scores (z-scores) of medical, other social services, and post-secondary education expenditures.

149 Table 5.2. Total Age-Standardized Mortality Rates, Aggregate Expenditures, and Provincial Welfare Generosity Index by Provinces, 1989-2008a Province TASMRb, c Health Social Services Education Provincial Welfare Expendituresd,c Expendituresd,c Expendituresd,c Generosity Indexe Newfoundland 7.32 (10) 2348.67 (2) 1067.78 (6) 2172.92 (1) .142 (6) Prince Edward Island 6.71 (8) 1996.42 (10) 786.61 (9) 1968.17 (3) -.710 (10) Nova Scotia 6.80 (9) 2257.20 (5) 782.82 (10) 1796.12 (7) .004 (7) New Brunswick 6.58 (7) 2364.31 (1) 872.70 (8) 1854.48 (4) -.537 (9) Quebec 6.44 (5) 2075.38 (9) 2044.01 (1) 1831.18 (6) .394 (1) Ontario 6.19 (3) 2259.32 (4) 1194.92 (2) 1476.36 (10) .166 (5) Manitoba 6.57 (6) 2244.59 (6) 1169.00 (4) 1559.50 (9) -.087 (8) Saskatchewan 6.27 (4) 2243.08 (7) 1010.05 (7) 1561.95 (8) .177 (4) Alberta 6.14 (2) 2138.23 (8) 1169.86 (3) 2041.15 (2) .197 (3) British Columbia 5.83 (1) 2341.53 (3) 1133.17 (5) 1837.63 (5) .255 (2) Notes: a Number of observations = 200 (10 provinces over 20 years). b TASMR = total age-standardized mortality rate (per 1,000 deaths). c Number in parenthesis indicates ranking out of 10. d All expenditure data is converted into real per capita terms and with the consumer price index (CPI) (2002 = 100) for each province. e Provincial welfare generosity index is the average standard scores (z-scores) of medical, other social services, and post-secondary education expenditures.

150 Table 5.3. Pairwise Correlation Matrix for Main Variables in Analyses Panel A 1 2 3 4 5 6 7 8 9 10 11 12 13 1) ASMR 1 2) Male ASMR 0.98** 1 3) Female ASMR 0.94** 0.89** 1 4) Health -0.60** -0.65** -0.44** 1 5) Social Services -0.17* -0.12 -0.16* -0.01 1 6) Education -0.19** -0.22** -0.11 0.43** -0.04 1 7) Hospital -0.12 -0.13 0.00 0.59** 0.02 0.40** 1 8) Medical -0.56** -0.61** -0.41** 0.85** 0.18* 0.29** 0.38** 1 9) Preventive -0.27** -0.23** -0.25** -0.10 0.49** 0.02 -0.03 0.03 1 10) Other Health -0.47** -0.50** -0.39** 0.71** -0.31** 0.27** 0.11 0.37** -0.40** 1 11) Soc. Assist. 0.20** 0.19** 0.22** -0.21** 0.57** -0.19** -0.02 -0.06 0.27** -0.39** 1 12) Work. Comp. -0.38** -0.35** -0.33* 0.16* 0.64** -0.05 0.08 0.39** 0.52** -0.25** 0.35** 1 13) Other SS. -0.41** -0.39** -0.36** 0.26** 0.66** 0.12 0.08 0.43** 0.40** -0.05 0.03 0.40** 1 14) Element/Sec. 0.03 -0.01 0.09 0.20** -0.11 0.92** 0.35** 0.10 0.00 0.04 -0.08 -0.12 0.02 15) Post-Second. -0.55** -0.55** -0.46** 0.64** 0.07 0.46** 0.19** 0.53** 0.08 0.56** -0.34** 0.14 0.24** 16) Special Retr. -0.29** -0.25** -0.29** 0.26** 0.43** 0.18* 0.15* 0.15* 0.03 0.24** 0.00 0.21** 0.34** 17) Other Ed. 0.24** 0.22** 0.22** 0.07 -0.43** 0.17* 0.22** -0.10 -0.39** 0.17* -0.12 -0.37** -0.49** 18) PWG Index -0.66** -0.67** -0.53** 0.75** 0.39** 0.38** 0.28** 0.84** 0.22** 0.38** -0.16* 0.40** 0.72** 19) Transfers 0.41** 0.38** 0.42** 0.18* -0.34** 0.45** 0.37** 0.00 -0.47** 0.21** -0.18* -0.54** -0.20** 20) Debt 0.41** 0.39** 0.38** -0.25** -0.25** -0.29** -0.13 -0.26** -0.43** -0.05 0.00 -0.50** -0.17* 21) Depend. Ratio 0.25** 0.27** 0.13 -0.45** -0.26** -0.46** -0.47** -0.51** -0.32** 0.01 -0.18* -0.36** -0.17* 22) Female LFP -0.68** -0.65** -0.64** 0.31** 0.02 0.08 -0.12 0.23** 0.13 0.42** -0.23** 0.14 0.21** 23) Unempl. 0.69** 0.68** 0.65** -0.32** -0.13 0.15* 0.20** -0.35** -0.16* -0.37** 0.17* -0.25** -0.31** 24) Low Income 0.25** 0.21** 0.26** 0.22** -0.08 0.14 0.26** 0.16* -0.30** 0.17* -0.11 -0.24** -0.02 25) GDP pc -0.62** -0.64** -0.52** 0.52** 0.14* 0.31* 0.02 0.54** 0.16* 0.40** -0.15* 0.29** 0.37** (Continued on next page)

151 Table 5.3. Pairwise Correlation Matrix for Main Variables in Analyses (Continued) Panel B 14 15 16 17 18 19 20 21 22 23 24 25 14) Element/Sec. 1 15) Post-Second. 0.10 1 16) Special Retr. 0.03 0.16* 1 17) Other Ed. 0.20** -0.02 -0.34** 1 18) PWG Index 0.09 0.76** 0.28** -0.26** 1 19) Transfers 0.50** 0.02 -0.14* 0.47** -0.08 1 20) Debt -0.12 -0.44** -0.26** 0.13 -0.37** 0.43** 1 21) Depend. Ratio -0.35** -0.34** -0.29** 0.14* -0.44** -0.01 0.55** 1 22) Female LFP -0.10 0.43** 0.16* -0.01 0.37** -0.47** -0.44** 0.05 1 23) Unempl. 0.33** -0.32** -0.29** 0.30** -0.42** 0.61** 0.24** -0.10 -0.74** 1 24) Low Income 0.14* 0.02 -0.01 0.13 0.07 0.63** 0.51** 0.02 -0.60** 0.43** 1 25) GDP pc 0.10 0.57** 0.31** -0.24** 0.64** -0.40** -0.56** -0.33** 0.67** -0.65** -0.40** 1 Notes: ASMR = age-standardized mortality rate; Soc. Assist = social assistance; Work. Comp. = worker’s compensation benefits; Other SS. = other social services; Element/Sec. = elementary/secondary; Post-Second. = post-secondary; Special Retr. = special retraining; PWG Index = provincial welfare generosity index; Depend. Ratio = dependency ratio; Female LFP = female labour force participation; Unempl. = unemployment; GDP pc = Gross domestic product per capita. N = 200. * p < .05; ** p < .01.

152 Table 5.4a. PW-PCSE Models of Aggregate Expenditures on Total Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Variables SE β SE β SE β Health -0.001*** 0.000 -0.510 (-3.90) Social Services -0.000*** 0.000 -0.136 (-2.72) Education -0.000* 0.000 -0.119 (-1.74) Transfers 0.000 0.000 0.000 0.000 0.029 0.000 0.000 0.086 (0.31) 0.014 (0.65) (1.38) Debt -0.001*** 0.000 -0.000** 0.000 -0.164 -0.000** 0.000 -0.154 (-3.00) -0.225 (-2.28) (-2.03) Dependency Ratio 0.032 0.022 0.060*** 0.021 0.321 0.051** 0.022 0.275 (1.49) 0.174 (2.88) (2.30) Female LFP -0.030 0.027 -0.065* 0.026 -0.284 -0.077*** 0.027 -0.333 (-1.13) -0.132 (-2.50) (-2.89) Unemployment 0.058* 0.033 0.066 0.035 0.219 0.058 0.037 0.192 (1.73) 0.191 (1.91) (1.54) Low income -0.016 0.020 -0.039** 0.018 -0.106 -0.045** 0.020 -0.123 (-0.77) -0.043 (-2.12) (-2.31) GDP per capita 8.00e-07 0.000 -0.000 0.000 -0.125 -8.57e-06 0.000 -0.092 (0.08) 0.009 (-1.10) (-0.80) Constant 7.838*** 2.529 8.475*** 2.581 9.421*** 2.710 (3.10) (3.28) (3.48) R2 0.868 0.856 0.8581 Wald 2 535.11 754.33 765.85 df 17 17 17 Rho 0.328 0.308 0.335 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

153 Table 5.4b. PW-PCSE Models of Aggregate Expenditures on Male Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Variables SE β SE β SE β Health -0.001*** 0.000 -0.585 (-4.82) Social Services -0.000* 0.000 -0.087 (-1.72) Education -0.000** 0.000 -0.154 (-2.35) Transfers 0.000 0.000 0.000 0.000 0.000* 0.000 (0.18) 0.007 (0.46) 0.019 (1.84) 0.109 Debt -0.001*** 0.000 -0.001** 0.000 -0.001** 0.000 (-3.29) -0.231 (-2.35) -0.163 (-2.09) -0.150 Dependency Ratio 0.038 0.034 0.085*** 0.033 0.072** 0.035 (1.12) 0.131 (2.58) 0.292 (2.05) 0.246 Female LFP -0.052 0.040 -0.128*** 0.041 -0.134*** 0.041 (-1.31) -0.144 (-3.13) -0.354 (-3.29) -0.368 Unemployment 0.087* 0.049 0.091* 0.054 0.091 0.056 (1.77) 0.183 (1.69) 0.191 (1.63) 0.192 Low income -0.016 0.029 -0.057** 0.029 -0.067** 0.029 (-0.55) -0.028 (-1.98) -0.098 (-2.30) -0.117 GDP per capita 6.85e-06 0.000 -0.000 0.000 -0.000 0.000 (0.46) 0.047 (-1.00) -0.105 (-0.66) -0.070 Constant 11.426*** 3.875 13.127*** 4.072 14.123*** 4.207 (2.95) (3.22) (3.36) R2 0.868 0.849 0.855 Wald 2 422.80 615.77 625.82 df 17 17 17 Rho 0.395 0.383 0.403 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

154 Table 5.4c. PW-PCSE Models of Aggregate Expenditures on Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Variables SE β SE β SE β Health -0.000 0.000 -0.222 (-1.30) Social Services -0.000*** 0.000 -0.158 (-2.66) Education -0.000 0.000 -0.044 (-0.49) Transfers 0.000 0.000 0.020 0.000 0.000 0.044 0.000 0.000 0.038 (0.31) (0.68) (0.43) Debt -0.000* 0.000 -0.191 -0.000* 0.000 -0.163 -0.000 0.000 -0.158 (-1.82) (-1.70) (-1.60) Dependency Ratio 0.020 0.017 0.169 0.030* 0.0158 0.249 0.025 0.0166 0.213 (1.19) (1.85) (1.51) Female LFP -0.017 0.022 -0.114 -0.018 0.019 -0.124 -0.030 0.019 -0.206 (-0.77) (-0.95) (-1.54) Unemployment 0.050* 0.028 0.264 0.060** 0.028 0.318 0.049* 0.030 0.259 (1.77) (2.16) (1.65) Low income -0.025 0.018 -0.109 -0.032** 0.015 -0.138 -0.034** 0.016 -0.149 (-1.43) (-2.17) (-2.16) GDP per capita -4.85e-06 7.36e-06 -0.083 -8.75e-06 7.48e-06 -0.149 -7.30e-06 7.65e-06 -0.125 (-0.66) (-1.17) (-0.95) Constant 5.612*** 1.953 5.367*** 1.866 6.095*** 1.964 (2.87) (2.88) (3.10) R2 0.737 0.737 0.735 Wald 2 975.61 1262.85 847.95 df 17 17 17 Rho 0.125 0.099 0.128 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

155 Table 5.5a. PW-PCSE Models of Disaggregated Health Expenditures on Total Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Hospital -0.000 0.000 -0.069 (-1.20) Medical -0.001*** 0.000 -0.361 (-3.28) Preventive -0.002** 0.001 -0.100 (-2.26) Other Health -0.001* 0.000 -0.217 (-1.80) Transfers 0.000 0.000 0.015 0.000 0.000 0.034 0.000 0.000 0.011 -0.000 0.000 -0.011 (0.32) (0.74) (0.26) (-0.25) Debt -0.000** 0.000 -0.175 -0.000** 0.000 -0.182 -0.000** 0.000 -0.153 -0.001** 0.000 -0.183 (-2.31) (-2.52) (-2.05) (-2.32) Dependency Ratio 0.054** 0.022 0.290 0.047** 0.021 0.256 0.054** 0.021 0.292 0.044** 0.021 0.236 (2.47) (2.28) (2.54) (2.07) Female LFP -0.077*** 0.027 -0.332 -0.038 0.029 -0.166 -0.088*** 0.027 -0.380 -0.068*** 0.026 -0.296 (-2.89) (-1.33) (-3.24) (-2.60) Unemployment 0.057 0.036 0.190 0.059* 0.034 0.195 0.059 0.037 0.196 0.035 0.037 0.117 (1.58) (1.74) (1.61) (0.96) Low income -0.038* 0.020 -0.103 -0.021 0.019 -0.058 -0.036* 0.019 -0.097 -0.034* 0.020 -0.093 (-1.90) (-1.12) (-1.87) (-1.73) GDP per capita -9.51e-06 0.000 -0.102 -1.66e-06 0.000 -0.018 -7.10e-06 0.000 -0.076 -9.10e-06 0.000 -0.098 (-0.88) (-0.16) (-0.68) (-0.86) Constant 9.199*** 2.683 7.066*** 2.641 9.607*** 2.708 9.320*** 2.604 (3.43) (2.68) (.326163) (3.58) R2 0.857 0.865 0.857 0.863 Wald 2 676.55 576.52 850.20 516.88 df 17 17 17 17 Rho 0.335 0.346 .3261632 0.356 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

156 Table 5.5b. PW-PCSE Models of Disaggregated Health Expenditures on Male Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Hospital -0.001* .0003907 -0.094 (-1.70) Medical -0.001*** 0.000 -0.308 (-2.77) Preventive -0.003* 0.001 -0.087 (-1.90) Other Health -0.001*** 0.000 -0.326 (-2.83) Transfers 0.000 0.000 0.012 0.000 0.000 0.024 0.000 0.000 0.008 -0.000 0.000 -0.024 (0.28) (0.57) (0.21) (-0.66) Debt -0.001** 0.000 -0.178 -0.001** 0.000 -0.177 -0.001** 0.000 -0.152 -0.001*** 0.000 -0.192 (-2.46) (-2.55) (-2.14) (-2.65) Dependency Ratio 0.078** 0.034 0.268 0.069** 0.033 0.236 0.081** 0.033 0.276 0.055* 0.032 0.188 (2.29) (2.08) (2.43) (1.70) Female LFP -0.134*** 0.040 -0.369 -0.089** 0.045 -0.245 -0.152*** 0.041 -0.418 -0.113*** 0.039 -0.311 (-3.32) (-2.00) (-3.67) (-2.92) Unemployment 0.088 0.055 0.186 0.083 0.052 0.174 0.084 0.055 0.177 0.039 0.053 0.082 (1.61) (1.60) (1.52) (0.74) Low income -0.051* 0.030 -0.089 -0.034 0.029 -0.059 -0.051* 0.028 -0.088 -0.042 0.030 -0.073 (-1.73) (-1.17) (-1.78) (-1.43) GDP per capita -0.000 0.000 -0.076 -2.58e-06 0.000 -0.018 -9.71e-06 0.000 -0.066 -9.46e-06 0.000 -0.065 (-0.71) (-0.18) (-0.64) (-0.64) Constant 13.681*** 4.161 11.247*** 4.143 14.240*** 4.174 13.849*** 3.889 (3.29) (2.71) (3.41) (3.56) R2 0.853 0.857 0.852 0.863 Wald 2 573.92 449.00 630.31 476.14 Df 17 17 17 17 Rho 0.409 0.415 0.410 0.441 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

157 Table 5.5c. PW-PCSE Models of Disaggregated Health Expenditures on Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Hospital 0.000 0.000 0.072 (0.98) Medical -0.001*** 0.000 -0.354 (-2.64) Preventive -0.001** 0.001 -0.087 (-2.09) Other Health -0.000 0.000 -0.099 (-0.73) Transfers 3.26e-06 0.000 0.004 0.000 0.000 0.053 0.000 0.000 0.012 1.16e-06 0.000 0.001 (0.06) (0.80) (0.18) (0.02) Debt -0.000 0.000 -0.145 -0.000* 0.000 -0.186 -0.000 0.000 -0.153 -0.000 0.00 -0.170 (-1.44) (-1.89) (-1.53) (-1.61) Dependency Ratio 0.027* 0.016 0.233 0.022 0.016 0.190 0.025 0.017 0.217 0.022 0.016 0.191 (1.66) (1.42) (1.53) (1.36) Female LFP -0.034* 0.020 -0.237 -0.002 0.021 -0.015 -0.035* 0.020 -0.240 -0.027 0.020 -0.189 (-1.73) (-0.10) (-1.71) (-1.39) Unemployment 0.043 0.030 0.224 0.056** 0.028 0.295 0.053* 0.030 0.279 0.043 0.031 0.225 (1.43) (2.04) (1.79) (1.39) Low income -0.035** 0.016 -0.150 -0.019 0.017 -0.083 -0.029* 0.016 -0.126 -0.031** 0.016 -0.136 (-2.14) (-1.16) (-1.87) (-2.01) GDP per capita -8.56e-06 7.66e-06 -0.146 -2.01e-06 7.07e-06 -0.034 -5.29e-06 7.60e-06 -0.090 -7.53e-06 7.69e-06 -0.129 (-1.12) (-0.29) (-0.70) (-0.98) Constant 6.116*** 1.968 4.418*** 1.956 6.208*** 2.005 6.104*** 1.958 (3.11) (2.26) (3.10) (3.12) R2 0.744 0.745 0.738 0.740 Wald 2 716.78 1161.40 1325.74 807.59 df 17 17 17 17 Rho 0.152 0.127 0.127 0.143 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

158 Table 5.6a. PW-PCSE Models of Disaggregated Social Services Expenditure on Total Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Variables SE β SE β SE β Social Assistance -0.000 0.000 -0.006 (-0.11) Workers’ Compensation -0.001 0.001 -0.053 Benefits (-1.19) Other Social Services -0.001*** 0.000 -0.217 (-3.73) Transfers 0.000 0.000 0.010 0.000 0.000 0.016 0.000 0.000 0.021 (0.23) (0.35) (0.47) Debt -0.000** 0.000 -0.161 -0.000** 0.000 -0.174 -0.000** 0.000 -0.165 (-2.20) (-2.26) (-2.42) Dependency Ratio 0.055*** 0.021 0.298 0.058*** 0.022 0.311 0.070*** 0.021 -0.297 (2.58) (2.68) (3.38) Female LFP -0.082*** 0.027 -0.357 -0.076*** 0.027 -0.328 -0.055** 0.025 0.281 (-3.10) (-2.80) (-2.20) Unemployment 0.050 0.036 0.167 0.056 0.036 0.186 0.065* 0.035 -0.126 (1.39) (1.58) (1.83) Low income -0.040** 0.020 -0.109 -0.037* 0.019 -0.100 -0.038** 0.018 0.000 (-2.04) (-1.90) (-2.10) GDP per capita -0.000 0.000 -0.117 -0.000 0.000 -0.115 -2.57e-06 0.000 -0.028 (-0.98) (-1.00) (-0.26) Constant 9.366*** 2.778 8.925*** 2.684 6.687*** 2.577 (3.37) (3.33) (2.59) R2 0.857 0.858 0.857 Wald 2 602.83 776.33 1503.69 df 17 17 17 Rho 0.346 0.346 0.276 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

159 Table 5.6b. PW-PCSE Models of Disaggregated Social Services Expenditure on Male Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Variables SE β SE β SE β Social Assistance -0.000 0.001 -0.014 (-0.24) Workers’ Compensation -0.001 0.002 -0.027 Benefits (-0.62) Other Social Services -0.001*** 0.000 -0.157 (-2.58) Transfers 0.000 0.000 0.008 0.000 0.000 0.009 0.000 0.000 0.015 (0.19) (0.23) (0.36) Debt -0.001** 0.000 -0.159 -0.001** 0.000 -0.167 -0.001** 0.000 -0.161 (-2.31) (000) (-2.35) Dependency Ratio 0.080** 0.034 0.275 0.084** 0.033 0.287 0.097*** 0.034 0.333 (2.38) (2.51) (2.89) Female LFP -0.147*** 0.041 -0.405 -0.140*** 0.041 -0.386 -0.114*** 0.040 -0.314 (-3.60) (-3.38) (-2.82) Unemployment 0.075 0.054 0.158 0.078 0.054 0.164 0.089 0.055 0.188 (1.38) (1.44) (1.62) Low income -0.057* 0.029 -0.098 -0.053* 0.029 -0.092 -0.054* 0.028 -0.094 (-1.92) (-1.83) (-1.91) GDP per capita -0.000 0.000 -0.101 -0.000 0.000 -0.097 -5.69e-06 0.000 -0.039 (-0.93) (-0.92) (-0.38) Constant 14.200*** 4.343 13.594*** 4.143 10.995*** (3.27) (3.28) (2.64) R2 0.850 0.851 0.853 Wald 2 545.97 616.56 826.85 df 17 17 17 Rho 0.415 0.418 0.380 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

160 Table 5.6c. PW-PCSE Models of Disaggregated Social Services Expenditure on Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Variables SE β SE β SE β Social Assistance -0.000 0.000 -0.005 (-0.07) Workers’ Compensation -0.000 0.001 -0.014 Benefits (-0.26) Other Social Services -0.001*** 0.000 -0.245 (-3.71) Transfers 0.000 0.000 0.012 0.000 0.000 0.014 0.000 0.000 0.026 (0.19) (0.21) (0.40) Debt -0.000* 0.000 -0.159 -0.000 0.000 -0.164 -0.000* 0.000 -0.168 (-1.66) (-1.56) (-1.82) Dependency Ratio 0.027 0.016 0.219 0.026 0.017 0.224 0.038** 0.016 0.326 (1.59) (1.56) (2.45) Female LFP -0.031* 0.019 -0.215 -0.030 0.020 -0.206 -0.011 0.018 -0.076 (-1.67) (-1.52) (-0.62) Unemployment 0.048 0.029 0.252 0.049* 0.029 0.257 0.059** 0.028 0.309 (1.63) (1.69) (2.09) Low income -0.033** 0.016 -0.144 -0.033** 0.016 -0.141 -0.030** 0.014 -0.129 (-2.12) (-2.06) (-2.09) GDP per capita -7.95e-06 8.02e-06 -0.136 -7.91e-06 7.67e-06 -0.135 -1.17e-06 6.76e-06 -0.020 (-0.99) (-1.03) (-0.17) Constant 6.099*** 1.951 5.998*** 1.979 3.967*** 1.810 (3.13) (3.03) (2.19) R2 0.738 0.738 0.744 Wald 2 789.03 808.68 3667.97 df 17 17 17 Rho 0.138 0.138 0.068 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

161 Table 5.7a. PW-PCSE Models of Disaggregated Education Expenditures on Total Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE Β Elementary/ -0.000 0.000 -0.058 Secondary (-0.98) Post-secondary -0.002*** 0.000 -0.339 (-3.29) Special Retraining -0.000 0.001 -0.019 Services (-0.30) Other Education 0.007*** 0.002 0.131 (3.09) Transfers 0.000 0.000 0.052 0.000 0.000 0.026 0.000 0.000 0.011 -0.000 0.000 -0.011 (0.83) (0.60) (0.24) (-0.24) Debt -0.000** 0.000 -0.152 -0.001*** 0.000 -0.206 -0.000** 0.000 -0.168 -0.000** 0.000 -0.155 (-1.98) (-2.93) (-2.05) (-2.12) Dependency Ratio 0.055** 0.022 0.298 0.028 0.022 0.151 0.055*** 0.021 0.297 0.048** 0.020 0.257 (2.51) (1.29) (2.59) (2.35) Female LFP -0.081*** 0.027 -0.351 -0.057** 0.026 -0.247 -0.081*** 0.026 -0.351 -0.094*** 0.027 -0.408 (-3.02) (-2.23) (-3.08) (-3.52) Unemployment 0.055 0.038 0.182 0.046 0.034 0.153 0.050 0.037 0.165 0.050 0.036 0.165 (1.43) (1.35) (1.34) (1.37) Low income -0.043** 0.020 -0.117 -0.037* 0.019 -0.100 -0.041** 0.019 -0.111 -0.039** 0.019 -0.106 (-2.16) (-1.93) (-2.13) (-2.06) GDP per capita -9.49e-06 0.000 -0.102 -0.000 9.97e-06 -0.114 -0.000 0.000 -0.119 -8.76e-06 0.000 -0.094 (-0.87) (-1.07) (-1.05) (-0.84) Constant 9.261*** 2.719 10.531*** 2.610 9.344*** 2.647 10.291*** 2.638 (3.41) (4.03) (3.53) (3.90) R2 0.859 0.860 0.853 0.860 Wald 2 671.73 1144.44 660.33 1170.12 df 17 17 17 17 Rho 0.350 0.295 0.323 0.335 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

162 Table 5.7b. PW-PCSE Models of Disaggregated Education Expenditures on Male Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Elementary/ -0.000* 0.000 -0.098 Secondary (-1.75) Post-secondary -0.0029*** 0.001 -0.283 (-2.70) Special Retraining -0.001 0.002 -0.024 Services (-0.40) Other Education 0.011*** 0.004 0.140 (3.03) Transfers 0.000 0.000 0.081 0.000 0.000 0.018 0.000 0.000 0.008 -0.000 0.000 -0.012 (1.37) (0.45) (0.20) (-0.32) Debt -0.001** 0.000 -0.145 -0.001*** 0.000 -0.195 -0.001** 0.000 -0.168 -0.001** 0.000 -0.153 (-1.99) (-2.85) (-2.17) (-2.20) Dependency Ratio 0.080** 0.035 0.274 0.047 0.034 0.160 0.082** 0.033 0.280 0.070** 0.032 0.240 (2.30) (1.36) (2.46) (2.22) Female LFP -0.143*** 0.041 -0.392 -0.111*** 0.040 -0.306 -0.143*** 0.040 -0.395 -0.165*** 0.041 -0.453 (-3.48) (-2.76) (-3.55) (-4.03) Unemployment 0.087 0.060 0.182 0.070 0.053 0.148 0.072 0.056 0.152 0.071 0.054 0.149 (1.52) (1.33) (1.30) (1.31) Low income -0.064** 0.030 -0.111 -0.051* 0.029 -0.088 -0.057** 0.029 -0.099 -0.053* 0.029 -0.092 (-2.16) (-1.75) (-1.97) (-1.86) GDP per capita -0.000 0.000 -0.078 -0.000 0.000 -0.089 -0.000 0.000 -0.099 -0.000 0.000 -0.075 (-0.72) (-0.89) (-0.95) (-0.72) Constant 13.850*** 4.223 15.328*** 4.074 13.912*** 4.113 15.408*** 4.064 (3.28) (3.76) (3.38) (3.79) R2 0.855 0.8555 0.849 0.857 Wald 2 582.09 759.43 594.62 805.01 df 17 17 17 17 Rho 0.423 0.383 0.402 0.427 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

163 Table 5.7c. PW-PCSE Models of Disaggregated Education Expenditures on Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Elementary/ 0.000 0.000 0.014 Secondary (0.18) Post-secondary -0.001** 0.000 -0.328 (-2.49) Special Retraining 0.000 0.001 0.022 Services (0.28) Other Education 0.005*** 0.002 0.150 (3.13) Transfers 2.68e-06 0.000 0.003 0.000 0.000 0.034 0.000 0.000 0.013 -0.000 0.000 -0.023 (0.03) (0.52) (0.20) (-0.36) Debt -0.000 0.000 -0.162 -0.000** 0.000 -0.206 -0.000 0.000 -0.154 -0.000 0.000 -0.153 (-1.63) (-2.12) (-1.39) (-1.60) Dependency Ratio 0.026 0.016 0.222 0.009 0.018 0.081 0.026 0.017 0.219 0.020 0.015 0.172 (1.58) (0.53) (1.53) (1.32) Female LFP -0.031 0.019 -0.213 -0.016 0.018 -0.110 -0.032 0.020 -0.219 -0.040** 0.020 -0.279 (-1.59) (-0.89) (-1.61) (-2.07) Unemployment 0.047 0.030 0.248 0.043 0.029 0.229 0.049 0.030 0.257 0.047 0.029 0.247 (1.55) (1.51) (1.64) (1.63) Low income -0.033** 0.016 -0.141 -0.031** 0.015 -0.133 -0.033** 0.016 -0.143 -0.032** 0.015 -0.138 (-2.03) (-2.02) (-2.14) (-2.14) GDP per capita -8.13e-06 7.87e-06 -0.139 -8.32e-06 7.38e-06 -0.142 -7.97e-06 7.60e-06 -0.136 -6.00e-06 7.12e-06 -0.102 (-1.03) (-1.13) (-1.05) (-0.84) Constant 6.052*** 1.963 6.825*** 2.035 6.095*** 1.969 6.801*** 1.915 (3.08) (3.35) (3.09) (3.55) R2 0.738 0.743 0.737 0.739 Wald 2 776.61 1303.57 818.09 1117.72 df 17 17 17 17 Rho 0.139 0.108 0.136 0.111 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

164 Table 5.8. PW-PCSE Models of Provincial Welfare Generosity Index on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1989- 2008 Total Male Female Model 1 Model 2 Model 3 Variables SE β SE β SE β Provincial Welfare -0.391*** 0.080 -0.507 -0.487*** 0.137 -0.401 -0.259*** 0.063 -0.534 Generosity Index (-4.87) (-3.56) (-4.10) Transfers 0.000 0.000 0.049 0.000 0.000 0.031 0.000 0.000 0.068 (1.11) (0.77) (1.06) Debt -0.001*** 0.000 -0.206 -0.001*** 0.000 -0.190 -0.000** 0.000 -0.214 (-3.26) (-2.89) (-2.37) Dependency Ratio 0.048*** 0.019 0.257 0.070** 0.031 0.241 0.023 0.015 0.199 (2.57) (2.26) (1.59) Female LFP -0.017 0.026 -0.076 -0.065 0.043 -0.179 0.013 0.018 0.088 (-0.68) (-1.51) (0.71) Unemployment 0.067** 0.032 0.223 0.093* 0.052 0.196 0.059** 0.026 0.312 (2.08) (1.77) (2.26) Low income -0.025 0.018 -0.067 -0.039 0.029 -0.067 -0.020 0.015 -0.087 (-1.38) (-1.36) (-1.36) GDP per capita 1.90e-06 9.18e-06 0.020 1.28e-06 0.000 0.009 9.47e-07 6.23e-06 0.016 (0.21) (0.09) (0.15) Constant 4.743*** 2.497 8.378*** 4.161 2.764*** 1.754 (1.90) (2.01) (1.58) R2 0.864 0.858 0.756 Wald 2 1041.43 749.10 1987.89 df 17 17 17 Rho 0.247 0.362 0.058 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product; N = 200. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

165 Figure 5.1. Standardized Effects of Aggregate and Disaggregated Provincial Welfare Generosity Expenditures for Total Age-Standardized Mortality Rates

Aggregate Expenditures Health Social Services Education

Disaggregated Expenditures Hospital Medical Preventive Other Health

Social Assistance Worker's Compensation Other Social Services

Elementary/Secondary Post-secondary Special Retraining Other Education

Provincial Welfare Generosity Index -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 -1E-15 0.1 0.2

Notes: Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See model 1 of Tables 5.4a, 5.5a, 5.6a, 5.7a and Table 5.8 for details.

166 Figure 5.2. Standardized Effects of Aggregate and Disaggregated Provincial Welfare Generosity Expenditures for Male Age-Standardized Mortality Rates

Aggregate Expenditures Health Social Services Education

Disaggregated Expenditures Hospital Medical Preventive Other Health

Social Assistance Worker's Compensation Other Social Services

Elementary/Secondary Post-secondary Special Retraining Other Education

Provincial Welfare Generosity Index

-0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 -1E-15 0.1 0.2

Notes: Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See model 2 of Tables 5.4b, 5.5b, 5.6b, 5.7b and Table 5.8 for details.

167 Figure 5.3. Standardized Effects of Aggregate and Disaggregated Provincial Welfare Generosity Expenditures for Female Age-Standardized Mortality Rates

Aggregate Expenditures Health Social Services Education

Disaggregated Expenditures Hospital Medical Preventive Other Health

Social Assistance Worker's Compensation Other Social Services

Elementary/Secondary Post-secondary Special Retraining Other Education

Provincial Welfare Generosity Index

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 -1E-15 0.1 0.2

Notes: Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See model 3 of Tables 5.4c, 5.5c, 5.6c, 5.7c and Table 5.8 for details.

168 CHAPTER SIX. LEFTIST POLITICS, PROVINCIAL WELFARE GENEROSITY, AND POPULATION HEALTH

Introduction

Chapter 5 demonstrated the significant impact of provincial welfare generosity in improving population health (i.e., lowering mortality rates) among Canadian provinces.

Because provincial welfare generosity has such a significant influence, it is worthwhile to move backwards in the causal process. Moving backwards is akin to moving further upstream and considering whether the political causes of provincial welfare generosity are also associated with population health. To this end, I turn to the welfare state literature to conceptualize and test leftist politics as a macro-social determinant of population health. Recall that leftist politics refers to political forces, processes, and institutions committed to more egalitarian outcomes and more generous welfare states.

For the purposes of this dissertation, I define leftist politics as power resources, partisan politics, and democratic politics, and test the hypotheses outlined in table 2.1 in chapter

Power resources theory applies a class-centric theoretical lens to welfare state development, and stresses the importance of two general types of lower class resources – the market power of labour unions and the political power of left parties

(Esping-Andersen, 1990, 1999; Huber et al., 1993; Korpi, 1978, 1983, Stephens, 1979).

This theory begins with two realist premises about who possesses power and how power is exercised in capitalist democracies. First, business, owners, and managers possess more power than the working-class because they control the means of production and thus control the primary delivery of economic resources to the

169 population. Second, because business possesses greater amounts of wealth, it has more resources to deploy in the political arena to advance its economic interests, which includes for example, a minimalist welfare state, low levels of government regulation, business-friendly policies, anti-union measures, and subsidies for profit-making.

To alter these unequal power relations, the theory goes on to argue that labour unions and left political parties represent the interests of the working class, and mobilize class-based political action in the workplace and elections (Korpi, 1983). Within the workplace, workers and the labour unions representing them can strike and interrupt the ability of business to make profits. Moreover, labour unions often support left political parties given their shared ideological interests. With backing from unions, left parties can push for an expansion of the welfare state to protect workers and the poor and guard against the economic insecurity that is inherent in capitalism. Thus, labour unions and left parties are types of working-class power resources that pressure, advocate, and institutionalize greater levels of egalitarianism. Although power resources theory was originally developed to explain welfare state development, expansion, and generosity, it offers a class-based perspective to better understand the potential connections between politics and population health. Market power resources in the form of labour unions are theorized to affect population health by improving workplace safety measures and supporting left political parties. Aligned with labour unions, left political parties are expected to improve population health given their ideological commitment to providing generous welfare services and social transfers. The political ideology of left parties in Canada such as the New Democratic Party and Parti Quebecois can be broadly described as being egalitarian, collectivist, and interventionist, and further

170 characterized as supporting a broad safety net of social welfare programs, including universal child care, education, and Medicare (Castles & Mair, 1984; Cross, 2011). In addition to these values and goals, the Parti Quebecois is also committed to sovereignty and political independence.

Although power resources theory has been cited as the dominant explanation of welfare state development in the extant comparative literature (Orloff, 1996), its major arguments have been critiqued for overlooking the importance of ideology (Brooks &

Manza, 2007), viewing politics as a game (Korpi, 1989), underappreciating race and gender (Orloff, 1993; Quadagno, 1994), and ignoring the impact of non-left political parties. Regarding the latter limitation, existing studies using a partisan politics approach find that the political power of social democratic (left-wing) parties as well as

Christian democratic (centre-wing) parties are both significant predictors of welfare state development, expansion, and generosity (Allan & Scruggs, 2004; Esping-Andersen,

1985, Garret, 1998; Hicks, 1999; Hicks & Swank, 1992; Huber & Stevens, 2000, 2001;

Iversen & Cusack, 2000). According to partisan political theory (Hibbs Jr., 1992), a major determinant of variation in policy choices and policy outputs in constitutional democracies is the party composition and ideology of government (Alesina, 1995;

Castles, 1982; Garrett & Lange, 1989; Huber & Stephens, 2001). The central arguments of partisan theory are straightforward: first, political parties are major determinants of public policy in democratic politics, and second, the political ideology of political parties influence public policy in predictable ways. Because different political parties are guided by different interests and ideologies, it is expected that they will pursue different policies

(Blais, Blake, & Dion, 1993). Huber and Stephens (2001) synthesized these ideas to

171 propose power constellations theory, which downplays the importance of labour unions to argue that social democratic parties, Christian democratic parties, and social movements are the most important engines of distinct welfare-state trajectories, with research demonstrating that party incumbency directly and indirectly affects a country’s level and type of social inequality (Huber et al., 2006; Moller et al., 2003).

Unsurprisingly, research on partisan politics is highly dependent on the classification of the ideological stance of political parties, usually measured in terms of some more or less explicit left-right ideological scale (Castles & Mair, 1984; Cross,

2011). At the most basic level, left-right political scales range from social democracy on the left and to conservatism on the right with moderatism falling in the centre. Compared to left political ideology described in the preceding paragraph, right political parties in

Canada (e.g., Social Credit, Canadian Alliance, Equality Party, Reform Party,

Saskatchewan Party, and Conservative) espouse a different set of social values and ideological goals. Right-wing parties tend to favour private enterprise, big business, and free markets as well as competitiveness, restructuring, deficit reduction, and privatization of Crown corporations. Right political parties are commonly associated with fiscal conservatism, which includes the explicit goal of reducing government spending.

As for centre political parties such as the Liberal Party, their ideological preferences reflect a moderate or centrist position that falls between the left and right of the political spectrum (Blais, 2005). In doing so, centre parties generally adopt a combination of conservative and liberal ideals geared towards a common goal, whether it is individual freedom, equality, or the well-being of the people, and usually embrace the status quo while not supporting radical changes in government structure, law, or general principles

172 of governance (Cross, 2011).

Given these ideological differences, this leads one to ask whether political parties have a predictable influence on population health. According to Navarro et al. (2006), the answer is a yes, and in their words, “Indeed, if elected political representatives were not able to influence public policy, there would be a severe crisis of democracy” (p.

1033). It follows that political parties in constitutional democracies pursue different public policies that determine varying levels of welfare resources, which then affect population health. Applying partisan political theory to explain population health leads to clear expectations regarding the possible effects of political parties – left and centre political party power should be associated with increased levels of welfare generosity and better population health and right political party power should produce the opposite.

Whereas power resources and partisan political theories stress the importance of labour unions and political parties as precursors of welfare state development, theories of political democracy focus on different types of political participation such as voter turnout and women in government as key factors that affect welfare states. Pampel and

Williams (1988) were among the first welfare state scholars to examine whether voter turnout is associated with higher levels of public spending. These authors suggest that lower rates of voter turnout are a consequence of high abstention rates among disadvantaged groups (Huber et al., 1993; Pampel & Williamson, 1988). Critical to this argument is that these disadvantaged groups are also those most in need of welfare services and social transfers, which includes for example, the poor, single parents, recent immigrants, and the unemployed. Given this, higher voter turnout rates are interpreted as greater levels of political participation of previously excluded voters. If

173 new voters participate in electoral politics, it is hypothesized that party incumbents will increase welfare generosity to appease new voters, and that these new voters will support left-leaning parties, thus skewing the class character of elections toward the left

(Dye, 1979; Hicks & Swank, 1992; Iversen, 2001; Iversen & Soskice, 2006; Lijphart,

1997; Pampel & Williamson, 1988; Piven & Cloward, 1994). When this logic is applied to population health, increases in voter turnout should translate into greater support for left political parties and higher levels of welfare generosity, which taken together, has a positive impact on population health (e.g., lower mortality rates).

The historical strength of women in government has also been conceptualized as an indicator of leftist politics and welfare state strength (Lorber, 2010; Muntaner et al.,

2002; Phillips, 1995; Wängnerud, 2009). According to Phillips’ (1995) ‘political theory of presence’, female politicians are better equipped to represent the interests of women and children than male politicians. Women in government are expected to influence political dynamics and priorities in meaningful ways given the qualitative differences between women and men in their everyday lives, including differences related to child- rearing, education and occupations, divisions of paid and unpaid labour, exposure to violence and sexual harassment (Lovenduski & Norris, 2003; Mansbridge, 1999;

Phillips, 1995; Wängnerud, 2000).

So how do female politicians actually make a difference in politics and possibly population health? The leading hypothesis claims that as more women are elected as politicians, the policy agendas of political parties become more receptive to women- friendly policies. Such policies can be defined as women’s rights bills such as pay equity and violence against women and as women’s traditional areas of interest such as care-

174 giving duties (Dodson & Carroll, 1991). In a recent review of existing research programs, Wängnerud (2009) concluded that gender does affect parliamentary processes, and female politicians make a significant contribution to advocating and strengthening the position of women’s interests. The connections between women in government, promotion of women’s issues, and increased welfare effort are established; however, the connection between female politicians and population health are not.

To date, most public health researchers have largely overlooked the direct and indirect effects of leftist politics on population health with the exception of Navarro,

Muntaner, Chung, and Kawachi (Chung & Muntaner, 2006; Kawachi et al., 1999;

Muntaner et al., 2002; Navarro et al., 2003, 2006). This chapter advances this program of research in two ways. First, I conceptualize leftist politics, expressed as power resources (labour unions and left political party), partisan politics (centre and right political parties), and democratic politics (voter turnout and women in government), as macro-social determinants of population health. Second, I apply the same technique to testing pathways between leftist politics and population health that Brady (2009) used to testing leftist politics and poverty. Specifically, I test two plausible pathways that implicate provincial welfare generosity as a proximate and potential mediating factor.

The first pathway suggests that the impact of leftist politics on population health is indirect and channeled through provincial welfare generosity. Under this scenario, provincial welfare generosity fully mediates the effects of unions, political parties, and democratic politics on population health. The second pathway contends that leftist politics combines with provincial welfare generosity to affect population health. Leftist politics have an additional effect on population health net of provincial welfare

175 generosity (e.g., triggering spending on SDOH other than health, social services, and education or enacting health-promoting policies or laws). If labour unions, political parties, and democratic politics as well as provincial welfare generosity are significant predictors of population health, their effects would not mediate each other. For example, trade unions may improve health through higher wages and improved working conditions before and separately from the welfare state (Muntaner et al., 2002). To determine the health effects and pathways of leftist politics, this chapter seeks to address the following four questions:

1. Do power resources, measured in terms of union strength and left political party

strength, affect mortality rates among Canadian provinces, even after controlling

for other province-specific factors?

2. Does the historical strength of political parties affect population health among

Canadian provinces? If so, which political parties are associated with improved

population health (left-, centre-, or right-wing)?

3. Do democratic politics, conceptualized as voter turnout and women in

government, affect population health outcomes, net of other province-specific

factors?

4. What are the specific pathways through which power resources, political parties,

and democratic politics affect population health?

a. Are the effects of leftist politics channeled through provincial welfare

generosity to affect population health? or

b. Do the effects of leftist politics combine with provincial welfare generosity

to affect population health?

176 Data and Methods

This study uses a pooled TSCS analysis of Canadian provinces from 1976 to 2008, including Newfoundland, Prince Edward Island, Nova Scotia, New Brunswick, Quebec,

Ontario, Manitoba, Saskatchewan, Alberta, and British Columbia. Data are retrieved from Statistics Canada and its Canadian Socio-Economic Information Management

System (CANSIM) Tables. This sample includes 330 province-years as cases.

Statistical analysis consists of five steps. First, I present mean values and standard deviations (total and within-province) for the variables used in the analysis.

Second, I display a correlation matrix to display the basic associations between explanatory, dependent, and control variables. Third, the data properties, estimation techniques, and sensitivity analyses are conducted. Following the rationale outlined in chapter 4, I use Beck and Katz’s (1995) technique for calculating panel-corrected standard (PCSE) estimates for linear TSCS models where the parameters are estimated by Prais-Winsten (PW) regression with a first-order autocorrelation correction

(AR(1)) and fixed-unit effects. As was the case in chapter 5, I keep model specifications as parsimonious as possible, and I test one leftist political variable at a time in each model.

In testing the direct and indirect pathways between leftist politics, provincial welfare generosity, and population health, regression models in this chapter are conducted in two progressive steps. First, I test whether power resources, political parties, and democratic politics have any effect on age-standardized mortality rates. In these direct or baseline models, I regress the dependent variable on a political variable while controlling for dependency ratio, female labour force participation, unemployment,

177 low income, and GDP per capita. Second, the nature of these associations are examined through fully adjusted models by running the same model with the addition of provincial welfare generosity as well as province-budget controls, transfers and debt.

Model results allow for three possible interpretations. First, if the political variable is insignificant in both direct and adjusted models, the political variable does not influence mortality rates. Second, if the political variable is significant in the direct model but not in the adjusted model, the impact of the political variable channels through provincial welfare generosity. Third, if the political variable is significant in both the direct and adjusted models, the impact of the political variable combines with provincial welfare generosity to affect mortality rates.

Fourth, regression estimates are converted into standardized coefficients and reported in bar charts (i.e., each bar represents an aggregate or disaggregated expenditure in separate and fully adjusted regression models) (see figures 6.1, 6.2, and

6.3 located at end of chapter). Fifth, post-estimation techniques are conducted to check for normality of residuals, multicollinearity of predictors, and robustness of results. All analyses are conducted using Stata/SE 12.1.

Dependent Variables

Total, male, and female age-standardized mortality rates are tested as dependent health variables. These outcome variables are the weighted average of the age-specific mortality rates per 1,000 persons, where the 1991 Canadian Census of Population is used as the standard population for the calculation. Data is available from 1976 to 2008 for a total of 330 cases.

178

Independent Variables

To test the population health hypotheses associated with power resources, partisan politics, and political democracy theories, I assess the health effects of six measures: union density, cumulative left party power, cumulative centre party power, cumulative right party power, voter turnout, and cumulative women in government. Union density measures the percentage of employees who belong to a trade union, and is obtained by dividing the number of employees covered by union agreements by total employees.

The union density rate for 1996 is missing and calculated by taking mean rates of 1995 and 1997 (McDonald & Myatt, 2004). Cumulative left party power represents the long- term, historical strength of left political parties (e.g., Green Party, New Democratic

Party, Parti Québecois), and is tabulated as the summed proportion of left seats held by all provincial government parties in each individual year since 1960 (Huber et al., 1997,

2004). Cumulative centre party power represents the long-term historical strength of centre political parties (e.g., Liberals), and is measured as the summed proportion of centre seats held by all provincial government parties in each individual year since 1960

(Huber et al., 1997, 2004). Cumulative right party power is defined as the long-term, historical strength of right political parties (Progressive Conservatives, Social Credit

Party), and is measured as the summed proportion of right seats held by all provincial government parties in each individual year since 1960 (Huber et al., 1997, 2004). Voter turnout is measured in percentages of the electorate that voted in each provincial election. Cumulative women in government represents the long-term, historical strength of women in government, and is measured as the summed proportion of female seats

179 held by all provincial government parties in each individual year since 1960 (Huber et al., 2004).

Provincial welfare generosity index. The analyses in chapter 5 reveal that three types of disaggregated types of welfare generosity are significant predictors of lower age-standardized mortality rates (total, male, and female): medical care, other social services, and post-secondary education. I developed a general measure of provincial welfare generosity by constructing an index by averaging the standard scores (z-scores) of these three disaggregated expenditures. I add this index to fully adjusted models in order to test direct and indirect pathways through leftist politics affect population health.

Control Variables

In baseline (direct effect) models, I control for the same variables as in the previous chapter with the exception of provincial factors: dependency ratio, women in labour force, unemployment rate, low income, and real GDP per capita. Dependency ratio is measured as the number of “dependents” for every 100 “workers”: youth (ages 0 to 17)

+ seniors (age 65 or older) per 100 workers (aged 18 to 64). Women in labour force refer to the percentage of women in the labour force. Unemployment rate is the percentage of the total labour force unemployed. Low income refers to the percentage of all persons with low income after tax, set at 50% of adjusted median household income. Real GDP per capita is the gross domestic product in real per capita terms and deflated using the 2002 consumer price index. In adjusted models that include provincial welfare generosity as a mediator, I add transfers and debt charges as province-specific controls. Transfers refers to the current fiscal year dollar value of total

180 transfers, which includes the sum of general and specific purpose transfers, received from other levels of government. Debt charges are measured as real per capita expenditures on debt-servicing.

Results

Table 6.1 provides the mean values and standard deviations (total and within province) for the variables used in this study (located at end of chapter). The average age- standardized mortality rate (per 1,000 deaths) is higher among males (mean [M] = 9.11,

Within-group standard deviation [WI-GRP SD] = 1.31] than females (M = 5.45, WI-GRP

SD = 0.58). Union density rates ranged from a low of 22.1% to a high of 57.6%, with approximately a third of employees belonging to a labour union (M = 34.29, WI-GRP SD

= 3.29). Historical political party strength is highest among right parties (M = 47.50, WI-

GRP SD = 4.92), followed by centre (M = 33.44, WI-GRP SD = 3.79) and left parties (M

= 18.40, WI-GRP SD = 2.70). The percentage of voter turnout averaged 71.5% (WI-

GRP SD = 5.64) while the cumulative proportion of women in government is 6.7% (WI-

GRP SD = 2.53).

As table 6.2 reveals, cross-provincial differences in leftist politics appear to be more pronounced than provincial differences in welfare generosity (located at end of chapter). Between 1976 and 2008, union density rates are highest in Newfoundland

(46.2%) and Quebec (39.5%) and lowest in Prince Edward Island (29.9%) and Alberta

(25.3%). Since 1960, the historical strength of left political parties held a majority of power only in Saskatchewan (55.9%), with moderate levels of support in Manitoba

(38.4%) and British Columbia (38.1%). Four provinces had majority levels of long-term

181 power by centre political parties included Prince Edward Island (56.4%), Quebec

(55.7%), Newfoundland (54.8%), and New Brunswick (54.5%). The cumulative power of right parties is strong in New Brunswick (60%), and in particular, Alberta (91.4%). Rates in voter turnout exceeded 75% in Prince Edward Island, Quebec, Saskatchewan, and

New Brunswick. And in terms of women in government, only two provinces had a cumulative proportion that averaged double-digits, Alberta (16.5%) and British Columbia

(10.3%).

Bivariate Patterns

Table 6.3 displays pairwise correlations between power resources, political parties, democratic politics, and total, male, and female, age-standardized mortality rates

(located at end of chapter). Relatively weak positive correlations are observed between union density and age-standardized mortality rates (total: r(330) = .18, p < .01; male: r(330) = .16, p < .01; female: r(330) = .23, p < .01).

Mortality rates are negatively correlated with cumulative left political party power

(total: r(330) = -.30, p < .01; male: r(330) = -.29, p < .01; female: r(330) = -.31, p < .01), positively associated with cumulative centre political party power (total, r(330) = .24, p

< .01; men, r(330) = .26, p < .01; women r(330) = .21, p < .01), and unrelated with right political party power (total: r(330) = .04, p = n.s.; male: r(330) = .01, p = n.s.; female: r(330) = .08, p = n.s.). In terms of democratic politics, voter turnout and mortality rates are positively associated, (total: r(330) = .040, p = n.s.; male: r(330) = .44, p = n.s.; female: r(330) = .31, p = n.s.) while women in government is strongly and negatively

182 correlated with mortality rates, (total: r(330) = -.65, p < .01; male: r(330) = -.65, p < .01; female: r(330) = -.62, p < .01).

The basic associations between leftist politics and the provincial welfare generosity index are worth noting since provincial expenditures may mediate the link between leftist politics and population health. Table 6.3 shows that the long-term strength of left political parties and women in government are both positively and significantly associated with provincial welfare generosity (left political parties: r(330)

= .24, p < .01; women in government: r(330) = .37, p < .01). This provides some initial support for the idea that these two leftist political variables affect population health by augmenting welfare effort. In contrast, increases in centre political power and voter turnout are associated with lower levels of provincial welfare generosity (centre political parties: r(330) = -.17, p < .05; voter turnout: r(330) = -.53, p < .01). No significant associations are observed between provincial welfare generosity with union density

(r(330) = -.13, p = n.s.) or right political party power (r(330) = -0.08, p = n.s.)

Overall Power Resources Models

Tables 6.4a, 6.4b, and 6.4c displays the results of the Prais-Winsten regressions for total, male, and female mortality rates, respectively. Unstandardized coefficients ( ) with panel corrected standard errors (SE ) and standardized coefficients (β) are reported for power resource variables and provincial welfare generosity (located at end of chapter). Partial support is found for power resources theory. On one hand, the results substantiated the claim that left political party power matters to population health, showing that left parties significantly predicted lower mortality rates across all

183 three outcomes (total: β = -.221, z(15) = -7.45, p < .01); male: β = -.188, z(15) = -6.43, p < .01); female: β = -.253, z(15) = -8.12, p < .01) in direct models (see models 1 of

Tables 6.4a, 6.4b, and 6.4c), despite controlling for other demographic and economic development factors. For a standard deviation increase in left political party power, total, male, and female mortality rates should decline by about 0.37, 0.33, and 0.43 standard deviations, respectively. In fully adjusted models, cumulative left political party power and provincial welfare generosity both remained significant, thus allowing for the interpretation that left political parties combines with provincial expenditures to positively influence health outcomes (see models 2 of Tables 6.4a, 6.4b, and 6.4c). On the other hand, union density had no significant effect on mortality rates in direct and fully adjusted regression models. The insignificance of labour unions on population health is somewhat surprising given its hypothesized impact and previously noted effects (Lynch et al., 2001; Muntaner et al., 2002).

Overall Political Party Models

Tables 6.5a, 6.5b, and 6.5c displays the results of the Prais-Winsten regressions for total, male, and female mortality rates, respectively (located at end of chapter).

Unstandardized coefficients ( ) with panel corrected standard errors (SE ) and standardized coefficients (β) are reported for cumulative centre and right political party power and provincial welfare generosity. The hypotheses underlying partisan politics theory are substantiated. Cumulative centre political party power has a negative influence on mortality rates (total: β = -.220, z(15) = -7.45, p < .01); male: β = -.188, z(15)

= -7.45, p < .01; female: β = -.172, z(15) = -7.45, p < .01), net of other alternative

184 explanations (see models 1 of Tables 6.5a, 6.5b, and 6.5c). A hypothesized increase in centre political party power of one standard deviation corresponds to a decline by 0.22,

0.19, and 0.17 standard deviations in total, male, and female mortality rates. When provincial welfare generosity is added to the fully adjusted model, it has a large and significant effect on mortality rates but centre political party power is less than half as large and becomes statistically insignificant (see models 2 of Tables 6.5a, 6.5b, and

6.5c located at end of chapter). Given that centre political party power is significant in the direct model and insignificant in the fully adjusted regression model, this supports the interpretation that the health effect of centre parties is channeled through provincial welfare generosity.

Models 3 of Tables 6.5a, 6.5b, and 6.5c displays the regression results for cumulative right political party power and age-standardized mortality rates for total, males, and females. Consistent with partisan politics theory, cumulative right political party power significantly predicted higher mortality rates in direct models (total: β = .247, z(15) = -8.76, p < .01); male: β = .260, z(15) = -8.44, p < .01); female: β = .259, z(15) = -

8.16, p < .01). For a standard deviation increase in right political party power, total, male, and female mortality rates should increase by 0.25, 0.26, and 0.26 standard deviations, respectively. When provincial welfare generosity and provincial-level controls are added to the (fully adjusted) model, right political party power remains positive and significant while provincial welfare generosity is associated with lower mortality rates (see models

4 in Tables 6.5a, 6.5b, and 6.5c).

Overall Political Democracy Models

185 Tables 6.6a, 6.6b, and 6.6c displays the results of the Prais-Winsten regressions for total, male, and female mortality rates, respectively (located at end of chapter).

Unstandardized coefficients ( ) with panel corrected standard errors (SE ) and standardized coefficients (β) are reported for voter turnout, women in government, and provincial welfare generosity.

Contrary to expectations, voter turnout is associated with higher mortality rates

(total: β = .092, z(15) = 2.84, p = 0.01; male: β = .099, z(15) = 3.00, p = 0.01; female: β

= .093, z(15) = 2.23, p = 0.01). After adding the provincial welfare generosity index to the direct model, however, the effect of voter turnout becomes insignificant and provincial welfare generosity has a large negative effect in lowering mortality rates (see models 2 in Tables 6.6a, 6.6b, and 6.6c).

As revealed in models 3 in Tables 6.6a, 6.6b, and 6.6c, cumulative women in government had the hypothesized effect of lowing mortality rates across all three outcomes (total: β = -.382, z(15) = -5.98, p < .01); male: β = -.433, z(15) = -6.74, p < .01); female: β = -.289, z(15) = -3.66, p < .01). Specifically, for a standard deviation increase in women in government, for example, total, male, and female mortality rates should decline by 0.38, 0.43, and 0.29 standard deviations, respectively. In models 4 in

Tables 6.6a, 6.6b, and 6.6c, the effects of women in government and provincial welfare generosity are tested in the same model. The analyses reveal that the effect of cumulative women in government remains significant, indicating that this factor combines with welfare generosity to reduce mortality rates.

186 Post-estimation Results

An examination of the skewness values and visual inspection of frequency distributions showed that the distributions of the variables are normally distributed. VIF scores confirmed the absence of multicollinearity and jackknife analyses replicated the substantive results (see Appendices B6 – B8 for jackknife results). Moreover, re- analyzing the data using alternative estimation models did not substantially change significant results.

Discussion

This chapter examined first whether leftist politics are predictive of population health, and second whether the effects of leftist politics on population health channel through or combine with provincial welfare generosity. Overall, the results indicate that the political power of left and centre political parties and women in government have significant negative effects on mortality rates, net of provincial, demographic, and economic factors. Whereas left political parties and women in government combine with provincial welfare generosity to improve population health, the effect of centre political parties is channeled through provincial expenditures. These findings suggest two conclusions on the connections between leftist politics and population health.

First, the most important message of this chapter is that theories of power constellation and politics of presence can be effectively applied to population health in the context of Canadian provinces (Huber & Stephens, 2001; Phillips, 1995). Consistent with the central predictions of power constellations theory, the findings show that political parties matter for population health. As the long-term strength of left and centre

187 political parties increase, mortality rates tend to decline. This finding corroborates the work of Navarro et al. (2003, 2006), who found that political parties with egalitarian ideologies are associated higher life expectancies at birth and lower infant mortality rates among OECD countries. In contrast, the long-term strength of right political parties has the opposite effect. Clearly, the balance of political power among left, centre, and right political parties matter for population health, which supports the argument that different parties, guided by ideologies ranging from social democracy and moderatism to conservatism, pursue different public policies that produce variations in population health.

This chapter also finds support for Phillips’ (1995) politics of presence theory, which argues that female politicians are better positioned to represent the interests of women and children. In applying this idea, I hypothesized that the long-term strength of women in government would have a significant and positive impact on population health. As more women are elected into provincial governments, population health outcomes improve. If the cumulative proportion of women in government increased one standard deviation – about 2.53% - total, male, and female mortality rates should decline by 0.38, 0.43, and 0.29 standard deviations, respectively. The effects of female politicians on population health are substantial, and are even larger than the health impacts of left and centre political parties. This finding supports previous research which has linked women’s representation in elected office and mortality rates among US states (Kawachi et al., 1999). Furthermore, it demonstrates the potential health benefits to the entire population, including men and women, of addressing gender inequalities in political representation. Increasing the presence of women in politics appears to be an

188 effective strategy to not only redress long standing issues of gender inequality but to also improve population levels of health.

A second conclusion of this chapter is that the influence of leftist politics on population health either channels through provincial welfare generosity or combines with provincial welfare generosity. Regarding the former pathway, I find evidence for the channeled explanation with respect to centre political party power. Specifically, centre political party power is significant initially in baseline models but becomes insignificant once provincial welfare generosity is added to the model. This allows for the interpretation that centre parties are associated with population health; however, their impact is indirect and channeled through provincial expenditures. This pathway is consistent with the comparative welfare state literature that finds that leftist politics has the effect of triggering welfare state expansion and generosity (Huber & Stephens,

2001).

In terms of the combined pathway, I find that left political parties and women in government combine with provincial welfare generosity to improve population levels of health. Because these two indicators of leftist politics are significant in baseline models and remain significant after controlling for provincial welfare generosity, this suggests that left political parties and women in government affect population health through direct and indirect mechanisms.

There are at least three possible explanations for the combined pathways between leftist politics, provincial welfare generosity, and population health. The first explanation is analogous to how centre parties are indirectly linked to population health, left parties and women in government trigger greater levels of provincial welfare

189 generosity, which in turn, lower mortality rates. A second possibility is that left political parties and women in government may trigger other forms of provincial spending besides health, social services, and education that yield significant and positive effects on population health. Recall that this dissertation conceptualized provincial welfare generosity using three aggregate-level expenditures: health, social services, and education. Although these three expenditures account for almost two-thirds (64%) of total provincial spending (Statistics Canada, 2007), other expenditures which may also operate as health-promoting welfare resources went unmeasured. These expenditures for example include housing, environment, general government services, and transportation and communication. Thus, a possible reason for the combined pathway is that left parties and women in government trigger other forms of provincial welfare generosity such as housing that lower mortality rates which have not been measured in this dissertation.

A third likely explanation for the combined explanation is that the impact of left parties and women in government on population health extends beyond the provision of welfare services and transfers. In this respect, leftist politics may affect population health through the enactment of health-promoting policies and legislation that are not captured by provincial expenditure data. Under these circumstances, provinces with strong histories of leftist politics enact policies and laws that protect against economic risks, redistribute welfare resources, or institutionalize equality that have the effect of improving population health, net of provincial welfare generosity. Although comparative research on this topic is scarce, there is some evidence to suggest that provincial differences with respect to paid health and family leave policies (Heymann, Gerecke, &

190 Chaussard, 2010), employment protection and labour relation laws (Daku & Heymann,

2010), and child care polices (Baker, Gruber, & Milligan, 2005) are associated with population health. For example, in 1997, Quebec implemented a low-cost universal child care policy along the lines of the European model, which has the potential to improve population health by allowing more mothers of young children to work outside the home and thus increasing the socioeconomic position of the household (Baker,

Gruber, & Milligan, 2005).

Two unexpected results emerged regarding the health effects of leftist politics.

According to proponents of power resources theory, labour unions are a key market resource that facilitates the organization of lower classes and advocates for egalitarian outcomes in capitalist democracies (Korpi, 1978, 1983; Stephens, 1983). This theory goes on to argue that strong labour unions mobilize class-based actions, which in turn, affect workplace democracy (e.g., increasing wages and job security and improving working conditions) and electoral politics (e.g., increasing the political power of left parties). In this dissertation, I hypothesized that labour unions would have a significant and positive effect on population health. Despite the established importance of labour unions for welfare state development and generosity, I find no relationship between provincial levels of unionization and population health. There are several possible explanations for this insignificant finding. First, it is possible that labour unions might be more consequential to other population health outcomes than total, male, and female all-cause age-standardized mortality rates. Existing studies on labour unions and population health have found confirmative evidence with respect to child health profiles and injury mortality rates (Lynch et al., 2001; Muntaner et al., 2002). Second, the

191 applicability of power resources theory and its emphasis on labour unions may have limited value within a liberal nation such as Canada (Esping-Andersen, 1990, 1999). For example, one of the theory’s central tenets is that unions mobilize support for left political parties; however, this dissertation found the contrary: labour unions are actually more strongly associated with centre political party power (r = 0.28) than left parties (r =

0.14). Third, the insignificance of labour unions might be due to measurement issues. It has been suggested that union density is a weak indicator of trade union strength

(Navarro, 2003). A far more telling indicator is the percentage of the labour force covered by collective bargaining agreements; however, this variable is not available over this dissertation’s time frame. Given these theoretical and methodological issues, it would be premature to discount the impact of labour unions on population health.

Clearly, more work is needed to refute or confirm this dissertation’s finding given the importance of working-class power in capitalist democracies to advance egalitarian outcomes.

Another unexpected finding regards the significance of voter turnout with higher mortality rates in baseline models. Guided by theories of political democracy, I hypothesized that population health, provincial welfare generosity, and left political parties should partly be determined by voter turnout as it is a measure of the degree to which citizens are involved in the political process and their belief in their ability to affect it (Huber et al., 1993; Jackman, 1987). To the contrary, I find that none of these predictions hold true. Instead, correlation and regression analyses reveal that increases in voter turnout are associated with higher mortality rates and lower levels of provincial welfare generosity. Moreover, high rates of voter turnout are strongly associated with

192 centre political party power, and unrelated with left parties. These results raise serious concerns regarding the applicability of Pampel and Williamson’s (1988) argument that voter turnout augments welfare generosity at the provincial level in Canada. Given that these associations failed to materialize, it is not completely surprising that the results did not confirm the hypothesized impact of voter turnout on population health. Likewise,

Chung and Muntaner (2006) also found positive associations between voter turnout and mortality rates (infant and under-5). This finding might be due to the fact that voter turnout only measures the degree of political participation, and not the ideological preferences of the electorate. A potentially better indicator of political participation is the

‘percentage of left vote’, which has been shown to be a significant predictor of improved population health (Chung & Muntaner, 2006; Navarro et al., 2003, 2006). With this said, it is worth noting that the impact of voter turnout on mortality rates is relatively small

(e.g., if voter turnout increases a standard deviation, total mortality rates should increase by only 0.09 standard deviations). Moreover, the effect voter turnout becomes insignificant once provincial welfare generosity is added to the model.

So how do political forces, processes, and institutions affect population health among Canadian provinces? Chapters 5 and 6 show when that changes in provincial welfare generosity, political party power, and gender equality in politics, mortality rates respond in consistent and predictable ways. Shifts to the political left and centre, and increases in provincial welfare generosity and of women in government lead to better population health outcomes. In contrast, shifts to the right have the opposite effect.

These answers have been derived by treating provinces as the primary unit of the analysis, meaning that regression models have been estimated using variations within-

193 province. Inspired by the work of Esping-Andersen (1990), this raises other research questions that rely on variations across provinces. Do provinces cluster together based on the long-term strength of leftist politics, thus generating distinct provincial regimes? If so, are provincial regimes predictive of mortality rates? To what extent does provincial welfare generosity explain population health differences between provincial regimes? I explore these issues in the next chapter.

194 Table 6.1. Summary Statistics: 1976-2008 Variable Mean Total Within-Group Min. Max N Std. Dev. Std. Dev. Dependent Total age-standardized mortality rate 7.07 0.97 0.90 4.7 9.4 330 Male age-standardized mortality rate 9.11 1.41 1.31 4.7 12.4 330 Female age-standardized mortality rate 5.45 0.64 0.58 4.2 7.3 330

Power Resources and Democratic Politics Union Density 34.29 6.53 3.29 22.1 57.6 330 Cumulative left political party power 18.40 19.31 2.70 0 61.10 330 Cumulative centre political party power 33.44 20.14 3.79 1.38 66.18 330 Cumulative right political party power 47.50 20.32 4.92 1.65 96.70 330 Voter turnout 71.46 9.73 5.64 40.6 87.6 330 Cumulative women in government 6.67 4.62 2.53 0.11 17.18 330

Provincial Welfare Welfare generosity Index 1.32e-07 0.77 0.70 -1.85 2.48 200 Real cash transfers per capita 1938.92 1000.64 385.83 487.91 7272.82 200 Real debt per capita 968.79 335.98 197.31 135.92 1981.13 200

Controls Dependency ratio 61.08 7.99 6.18 48.96 88.78 330 Female labour force participation 54.99 7.04 5.45 32.00 67.62 330 Unemployment 9.78 3.77 1.98 3.4 20.2 330 Low income 14.63 3.65 2.03 7.00 28.00 330 Real GDP per capita 29,292.88 8,968.51 6,099.73 6,056.90 66,100.10 330 Notes: N = 330 (10 provinces over 33 years, 1976-2008); N = 200 (10 provinces over 20 years, 1989-2008); Expenditure data are in real per capita terms for each province.

195 Table 6.2. Power Resources, Political Parties, and Democratic Politics by Province, 1976-2008 Province Union Cumulative Left Cumulative Centre Cumulative Right Voter Cumulative Densityb Party Powerb Party Powerb Party Powerb Turnoutb Women in Gov’tb Newfoundland 46.42 (1) 0.40 (8) 54.84 (3) 43.01 (9) 74.01 (5) 2.51 (10) Prince Edward Island 29.90 (9) 0.12 (10) 56.36 (1) 43.52 (8) 83.08 (1) 6.08 (4) Nova Scotia 30.96 (7) 4.67 (6) 34.91 (5) 60.00 (2) 71.43 (7) 2.81 (9) New Brunswick 33.28 (6) 0.39 (9) 54.46 (4) 44.47 (7) 76.51 (4) 5.12 (8) Quebec 39.50 (2) 24.64 (4) 55.65 (2) 17.47 (11) 77.64 (2) 5.80 (5) Ontario 30.26 (8) 18.38 (5) 27.74 (6) 53.86 (3) 60.97 (9) 5.55 (6) Manitoba 35.71 (4) 38.35 (2) 11.57 (8) 49.84 (6) 68.07 (8) 6.69 (3) Saskatchewan 33.74 (5) 55.94 (1) 24.45 (7) 19.61 (10) 76.25 (3) 5.38 (7) Alberta 25.34 (10) 2.95 (7) 4.54 (10) 91.42 (1) 54.82 (10) 16.51 (1) British Columbia 37.85 (3) 38.14 (3) 9.92 (9) 51.78 (4) 71.86 (6) 10.29 (2) Notes: a Number of observations = 330 (10 provinces over 33 years). b Number in parenthesis indicates ranking out of 10.

196 Table 6.3. Pairwise Correlation Matrix for Main Variables in Analyses Panel A 1 2 3 4 5 6 7 8 1) ASMR 1 2) Male ASMR 0.99** 1 3) Female ASMR 0.97** 0.93** 1 4) Union Density 0.18** 0.16** 0.23** 1 5) Cum. Left PPP -0.30** -0.29** -0.31** 0.14** 1 6) Cum. Centre PPP 0.24** 0.26** 0.21** 0.28** -0.48** 1 7) Cum. Right PPP 0.04 0.01 0.08 -0.42** -0.47** -0.56** 1 8) Voter Turnout 0.40** 0.44** 0.31** 0.35** 0.00 0.57** -0.57** 1 9) Cum. Women Gov. -0.65** -0.65** -0.62** -0.38** 0.08 -0.55** 0.47** -0.59** 10) Wel. Gen. Index -0.66** -0.67** -0.53** -0.13 0.24** -0.17* -0.08 -0.53** 11) Transfers 0.41** 0.38** 0.42** 0.26** -0.46** 0.58** -0.15* 0.42** 12) Debt 0.41** 0.39** 0.38** 0.28** 0.13 0.13 -0.27** 0.37** 13) Dependency Ratio 0.54** 0.51** 0.47** 0.03 0.14** 0.07 -0.19** 0.37** 14) Female LFP -0.82** -0.79** -0.80** -0.40** 0.23** -0.47** 0.26** -0.52** 15) Unemployment 0.37** 0.38** 0.36** 0.56** -0.55** 0.64** -0.12* 0.51** 16) Low Income 0.32** 0.28** 0.31** 0.41** -0.15** 0.53** -0.40** 0.44** 17) GDP per capita -0.62** -0.63** -0.56** -0.39** 0.19** -0.53** 0.35** -0.72** (Continued on next page)

197 6.3. Pairwise Correlation Matrix for Main Variables in Analyses (Continued) Panel B 9 10 11 12 13 14 15 16 17 9) Cum. Women Gov. 1 10) Wel. Gen. Index 0.37** 1 11) Transfers -0.53** -0.08 1 12) Debt -0.57** -0.37** 0.43** 1 13) Dependency Ratio -0.45** -0.44** -0.01 0.55** 1 14) Female LFP 0.77** 0.37** -0.47** -0.44** -0.56** 1 15) Unemployment -0.47** -0.42** 0.61** 0.24** 0.02 -0.50** 1 16) Low Income -0.52** 0.07 0.63** 0.51** 0.47** -0.64** 0.48** 1 17) GDP per capita 0.74** 0.64** -0.40** -0.56** -0.53** 0.73** -0.59** -0.56** 1 Notes: ASMR = age-standardized mortality rate; Cum. Left PPP = cumulative left political party power; Cum. Centre PPP = cumulative centre political party power; Cum. Right PPP = cumulative right political party power; Wel. Gen. Index = welfare generosity index; Female LFP = female labour force participation; GDP per capita = Gross domestic product per capita. N = 330. * p < .05; ** p < .01.

198 Table 6.4a. PW-PCSE Models of Power Resources and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Welfare -0.392*** 0.079 -0.508 -0.346*** 0.076 -0.448 Generosity Index (-4.94) (-4.55) Union Density 0.001 0.010 0.002 -0.013 0.013 -0.080 (0.06) (-1.00) Cumulative Left -0.074*** 0.010 -0.221 -0.061*** 0.015 -0.307 Party Power (-7.45) (-3.99) Transfers 0.000 0.000 0.047 0.000 0.000 0.038 (1.05) (0.91) Debt -0.001*** 0.000 -0.240 -0.001*** 0.000 -0.192 (-3.15) (-3.34) Dependency -0.009 0.012 -0.048 0.053*** 0.019 0.287 -0.002 0.010 -0.011 0.054*** 0.017 0.291 Ratio (-0.77) (2.77) (-0.20) (3.13) Female LFP -0.145*** 0.014 -0.628 -0.017 0.026 -0.073 -0.120*** 0.012 -0.521 -0.001 0.024 -0.003 (-10.30) (-0.66) (-10.23) (-0.03) Unemployment 0.023 0.022 0.075 0.070** 0.032 0.231 0.019 0.018 0.064 0.062** 0.030 0.205 (1.05) (2.18) (1.08) (2.08) Low income -0.035*** 0.013 -0.096 -0.030 0.018 -0.082 -0.032*** 0.012 -0.087 -0.026 0.016 -0.071 (-2.61) (-1.63) (-2.69) (-1.59) GDP per capita -0.000* 6.64e-06 -0.159 -1.14e-06 9.41e-06 -0.012 -0.000*** 6.07e-06 -0.177 -7.54e-08 8.57e-06 -0.001 (-2.22) (-0.12) (-2.71) (-0.01) Constant 17.244*** 1.448 4.960** 2.427 15.593*** 1.235 3.677 2.308 (11.91) (2.04) (12.63) (1.59) R2 0.897 0.862 0.910 0.867 Wald 2 528.50 1139.59 666.12 1480.87 df 15 18 15 18 Rho 0.533 0.227 0.480 0.203 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

199 Table 6.4b. PW-PCSE Models of Power Resources and Provincial Welfare Generosity on Male Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Welfare -0.489*** 0.136 -0.403 -0.401*** 0.129 -0.330 Generosity Index (-3.59) (-3.10) Union Density 0.008 0.015 0.020 -0.015 0.022 -0.059 (0.52) (-0.70) Cumulative Left -0.103*** 0.016 -0.188 -.0122*** 0.027 -0.386 Party Power (-6.43) (-4.54) Transfers 0.000 0.000 0.031 0.000 0.000 0.021 (0.77) (0.55) Debt -0.001*** 0.000 -0.214 -0.001*** 0.000 -0.171 (-2.78) (-2.90) Dependency -0.018 0.018 -0.062 0.077** 0.032 0.263 -0.009 0.015 -0.030 0.083*** 0.029 0.283 Ratio (-1.02) (2.37) (-0.57) (2.87) Female LFP -0.215*** 0.022 -0.592 -0.063 0.043 -0.175 -0.180*** 0.018 -0.496 -0.033 0.040 -0.090 (-9.79) (-1.48) (-9.87) (-0.81) Unemployment 0.063** 0.032 0.133 0.098* 0.052 0.206 0.065** 0.027 0.137 0.084* 0.048 0.178 (1.98) (1.87) (2.41) (1.77) Low income -0.049** 0.019 -0.085 -0.045 0.030 -0.078 -0.049*** 0.018 -0.085 -0.041 0.027 -0.071 (-2.52) (-1.52) (-2.76) (-1.53) GDP per capita -0.000 9.35e-06 -0.081 -1.91e-06 0.000 -0.013 -0.000* 8.50e-06 -0.111 -1.77e-06 0.000 -0.012 (-1.26) (-0.13) (-1.90) (-0.13) Constant 23.550*** 2.286 8.554** 4.097 21.500*** 1.951 6.247 3.825 (10.30) (2.09) (11.02) (1.63) R2 0.871 0.856 0.885 0.867 Wald 2 553.65 738.11 784.94 893.83 df 15 18 15 18 Rho 0.564 0.344 0.500 0.327 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

200 Table 6.4c. PW-PCSE Models of Power Resources and Provincial Welfare Generosity on Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Welfare -0.260*** 0.063 -0.534 -0.234*** 0.061 -0.483 Generosity Index (-4.10) (-3.85) Union Density 0.000 0.009 0.003 0.002 0.010 0.018 (0.01) (0.18) Cumulative Left -0.054*** 0.007 -0.253 -0.032*** 0.012 -0.255 Party Power (-8.12) (-2.67) Transfers 0.000 0.000 0.070 0.000 0.000 0.052 (1.03) (0.86) Debt -0.000* 0.000 -0.206 -0.000** 0.000 -0.203 (-1.90) (-2.44) Dependency -0.008 0.009 -0.067 0.022 0.016 0.191 -0.001 0.007 -0.008 0.027* 0.014 0.228 Ratio (-0.90) (1.43) (-0.13) (1.89) Female LFP -0.093*** 0.010 -0.637 0.013 0.018 0.088 -0.073*** 0.009 -0.499 0.022 0.018 0.150 (-9.06) (0.70) (-8.48) (1.19) Unemployment 0.005 0.018 0.024 0.059** 0.026 0.312 -1.73e-06 0.015 0.000 0.055** 0.025 0.291 (0.26) (2.26) (-0.00) (2.23) Low income -0.033*** 0.012 -0.142 -0.019 0.014 -0.083 -0.029*** .010525 -0.126 -0.021 0.014 -0.091 (-2.68) (-1.33) (-2.77) (-1.51) GDP per capita -0.000** 5.87e-06 -0.227 1.41e-06 6.84e-06 0.024 -0.000*** 5.36e-06 -0.237 -3.74e-07 6.04e-06 -0.006 (-2.27) (0.21) (-2.60) (-0.06) Constant 12.435*** 0.999 2.730 1.752 10.962*** 0.852 2.216 1.704 (12.44) (1.56) (12.86) (1.30) R2 0.854 0.756 0.870 0.763 Wald 2 575.62 1878.31 682.45 3683.54 df 15 18 15 18 Rho 0.386 0.058 0.321 0.016 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

201 Table 6.5a. PW-PCSE of Models Centre and Right Political Parties and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Provincial Welfare -0.368*** 0.087 -0.477 -0.286*** 0.090 -0.371 Generosity Index (-4.25) (-3.18) Cumulative Centre -0.040*** 0.007 -0.220 -0.009 0.012 -0.065 Party Power (-5.63) (-0.80) Cumulative Right 0.0451*** 0.005 0.247 0.031*** 0.012 0.278 Party Power (8.76) (2.60) Transfers 0.000 0.000 0.047 0.000 0.000 0.036 (1.06) (0.83) Debt -0.001*** 0.000 -0.187 -0.000** 0.000 -0.133 (-2.75) (-2.19) Dependency Ratio -0.013 0.010 -0.072 0.044** 0.019 0.239 -0.009 0.009 -0.048 0.038** 0.018 0.205 (-1.30) (2.31) (-1.00) (2.14) Female LFP -0.155*** 0.013 -0.671 -0.028 0.030 -0.121 -0.140*** 0.011 -0.606 -0.047* 0.028 -0.202 (-12.28) (-0.94) (-13.19) (-1.69) Unemployment 0.009 0.019 0.031 0.066* 0.032 0.218 0.004 0.017 0.012 0.060** 0.030 0.198 (0.49) (2.05) (0.21) (1.96) Low income -0.035*** 0.013 -0.094 -0.022 0.018 -0.059 -0.031*** 0.012 -0.085 -0.014 0.017 -0.039 (-2.70) (-1.19) (-2.58) (-0.86) GDP per capita -0.000*** 6.35e-06 -0.158 2.83e-06 9.41e-06 0.030 -0.000*** 6.03e-06 -0.169 3.74e-06 8.80e-06 0.040 (-2.31) (0.30) (-2.60) (0.43) Constant 18.380*** 1.370 5.589** 2.759 12.941*** 1.174 4.224* 2.366 (13.41) (2.03) (11.02) (1.78) R2 0.905 0.864 0.915 0.865 Wald 2 799.12 1061.64 1017.48 1218.44 df 15 18 15 18 Rho 0.461 0.245 0.418 0.214 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

202 Table 6.5b. PW-PCSE Models of Centre and Right Political Parties and Provincial Welfare Generosity Male Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Provincial Welfare -0.446*** 0.139 -0.367 -0.304** 0.142 -0.250 Generosity Index (-3.21) (-2.14) Cumulative -0.065*** 0.011 -0.018 0.021 -0.082 Centre Party (-5.70) -0.188 (-0.87) Power Cumulative Right 0.069*** 0.008 0.260 0.059*** 0.021 0.342 Party Power (8.44) (2.83) Transfers 0.000 0.000 0.029 0.000 0.000 0.020 (0.72) (0.50) Debt -0.001** 0.000 -0.168 -0.000* 0.000 -0.106 (-2.47) (-1.79) Dependency -0.023 0.016 -0.080 0.064** 0.032 0.218 -0.017 0.014 -0.057 0.052* 0.030 0.177 Ratio (-1.45) (1.99) (-1.18) (1.74) Female LFP -0.228*** 0.020 -0.627 -0.083* 0.046 -0.229 -0.204*** 0.017 -0.563 -0.112*** 0.043 -0.309 (-11.48) (-1.81) (-12.20) (-2.59) Unemployment 0.047* 0.028 0.099 0.091* 0.052 0.192 0.041 0.026 0.086 0.083* 0.048 0.175 (1.68) (1.76) (1.57) (1.72) Low income -0.049*** 0.019 -0.085 -0.034 0.029 -0.058 -0.044** 0.018 -0.076 -0.022 0.027 -0.038 (-2.64) (-1.17) (-2.50) (-0.83) GDP per capita -0.000 8.85e-06 -0.089 2.72e-06 0.000 0.019 -0.000* 8.37e-06 -0.103 3.89e-06 0.000 0.027 (-1.48) (0.19) (-1.80) (0.28) Constant 25.295*** 2.170 9.901** 4.392 16.932*** 1.885 6.839* 3.999 (11.66) (2.25) (8.98) (1.71) R2 0.884 0.859 0.895 0.863 Wald 2 776.24 779.58 986.48 965.67 df 15 18 15 18 Rho 0.498 0.361 0.458 0.333 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

203 Table 6.5c. PW-PCSE Models of Centre and Right Political Parties and Provincial Welfare Generosity Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Provincial Welfare -0.222*** 0.071 -0.457 -0.165** 0.074 -0.340 Generosity Index (-3.10) (-2.22) Cumulative -0.026*** 0.005 -0.172 -0.013* 0.008 -0.144 Centre Party (-5.35) (-1.71) Power Cumulative Right 0.030*** 0.004 0.259 0.024*** 0.009 0.353 Party Power (8.16) (2.87) Transfers 0.000 0.000 0.059 0.000 0.000 0.039 (0.95) (0.64) Debt -0.000* 0.000 -0.169 -0.000 0.000 -0.115 (-1.72) (-1.34) Dependency -0.010 0.007 -0.082 0.019 0.015 0.158 -0.006 0.006 -0.049 0.016 0.013 0.137 Ratio (-1.32) (1.24) (-0.90) (1.20) Female LFP -0.098*** 0.009 -0.673 -0.004 0.025 -0.029 -0.087*** 0.007 -0.597 -0.014 0.022 -0.099 (-11.18) (-0.17) (-11.65) (-0.66) Unemployment -0.006 0.016 -0.031 0.057** 0.026 0.298 -0.011 0.015 -0.060 0.051** 0.025 0.267 (-0.37) (2.16) (-0.78) (2.01) Low income -0.032*** 0.011 -0.138 -0.015 0.014 -0.063 -0.028*** 0.011 -0.123 -0.010 0.013 -0.043 (-2.82) (-1.01) (-2.69) (-0.74) GDP per capita -0.000** 5.53e-06 -0.212 2.67e-06 6.36e-06 0.046 -0.000* 5.27e-06 -0.217 3.01e-06 5.83e-06 0.051 (-2.25) (0.42) (-2.41) (0.52) Constant 13.000*** 0.924 4.058* 2.189 9.195*** 0.787 2.555 1.643 (14.07) (1.85) (11.69) (1.55) R2 0.858 0.759 0.871 0.766 Wald 2 801.65 1986.90 1044.57 2347.71 df 15 18 15 18 Rho 0.292 0.055 0.244 0.025 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

204 Table 6.6a. PW-PCSE Models of Voter Turnout, Women in Government and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Provincial Welfare -0.353*** 0.083 -0.458 -0.292*** 0.082 -0.379 Generosity Index (-4.26) (-3.55) Voter Turnout 0.015*** 0.005 0.092 0.008 0.005 0.087 (2.84) (1.60) Cumulative Women -0.136*** 0.023 -0.382 -0.095*** 0.025 -0.447 in Government (-5.98) (-3.81) Transfers 0.000 0.000 0.045 0.000 0.000 0.037 (1.00) (0.91) Debt -0.001*** 0.000 -0.215 -0.000*** 0.000 -0.172 (-3.49) (-2.97) Dependency Ratio -0.005 0.011 -0.029 0.048*** 0.018 0.258 0.001 0.010 0.004 0.033** 0.016 0.175 (-0.49) (2.62) (0.07) (2.04) Female LFP -0.136*** 0.014 -0.590 -0.020 0.026 -0.085 -0.087*** 0.016 -0.378 -0.003 0.025 -0.015 (-9.81) (-0.77) (-5.37) (-0.14) Unemployment 0.014 0.020 0.045 0.061* 0.033 0.202 0.005 0.018 0.016 0.053* 0.030 0.175 (0.67) (1.87) (0.27) (1.76) Low income -0.031** 0.013 -0.085 -0.024 0.018 -0.066 -0.022* -0.060 -0.012 0.017 -0.033 (-2.46) (-1.38) (-1.74) (-0.69) GDP per capita -0.000 6.44e-06 -0.133 2.44e-06 9.14e-06 0.026 -0.000* 0.013 -0.125 -1.63e-06 0.000 -0.018 (-1.92) (0.27) (-1.87) (-0.18) Constant 15.603*** 1.522 4.459* 2.487 15.151*** 6.300*** 2.329 (10.25) (1.79) (11.34) (2.70) R2 0.890 0.863 0.909 0.00 Wald 2 519.48 1280.97 814.82 0.00 df 15 18 15 18 Rho 0.542 0.232 .542 0.259 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

205 Table 6.6b. PW-PCSE Models of Voter Turnout, Women in Government and Provincial Welfare Generosity on Male Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Provincial Welfare -0.421*** 0.139 -0.347 -0.296** 0.136 -0.244 Generosity Index (-3.03) (-2.18) Voter Turnout 0.023*** 0.008 0.099 0.015* 0.008 0.101 (3.00) (1.92) Cumulative Women -0.224*** 0.033 -0.433 -0.189*** 0.042 -0.564 in Government (-6.74) (-4.56) Transfers 0.000 0.000 0.027 0.000 0.000 0.021 (0.66) (0.56) Debt -0.001*** 0.000 -0.202 -0.003 0.015 -0.010 -0.001** 0.000 -0.146 (-3.20) (-0.19) (-2.52) Dependency Ratio -0.012 0.017 -0.041 0.071** 0.030 0.242 -0.120*** 0.023 -0.330 0.039 0.026 0.135 (-0.71) (2.32) (-5.26) (1.49) Female LFP -0.200*** 0.021 -0.550 -0.068 0.043 -0.186 0.040 0.026 0.084 -0.037 0.041 -0.101 (-9.40) (-1.59) (1.56) (-0.90) Unemployment 0.054* 0.030 0.113 0.084 0.053 0.177 -0.032* 0.018 -0.056 0.068 0.047 0.143 (1.80) (1.60) (-1.77) (1.44) Low income -0.046** 0.019 -0.079 -0.039 0.028 -0.067 -7.85e-06 8.66e-06 -0.054 -0.017 0.028 -0.029 (-2.47) (-1.36) -0.91) (-0.60) GDP per capita -9.05e-06 9.02e-06 -0.062 2.34e-06 0.000 0.016 20.410*** 1.977 -4.94e-06 0.000 -0.034 (-1.00) (0.16) (10.32) (-0.35) Constant 21.082*** 2.35 7.760* 4.123 0.888 11.448*** 3.823 (8.97) (1.88) (2.99) R2 0.875 0.858 924.41 0.872 Wald 2 620.95 1008.09 15 1529.33 df 15 18 0.558 18 Rho 0.567 0.341 330 0.372 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

206 Table 6.6c. PW-PCSE Models of Voter Turnout, Women in Government and Provincial Welfare Generosity on Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Model 1 Model 2 Model 3 Model 4 Variables SE β SE β SE β SE β Provincial Welfare -0.221*** 0.065 -0.456 -0.212*** 0.065 -0.437 Generosity Index (-3.42) (-3.25) Voter Turnout 0.010** 0.004 0.093 0.008* 0.004 0.133 (2.23) (1.96) Cumulative Women -0.066*** 0.018 -0.289 -0.043** 0.021 -0.321 in Government (-3.66) (-2.09) Transfers 0.000 0.000 0.059 0.000 0.000 0.057 (0.91) (0.92) Debt -0.000** 0.000 -0.226 -0.000** 0.000 -0.191 (-2.53) (-2.34) Dependency Ratio -0.006 0.008 -0.048 0.023 0.014 0.201 -0.003 0.008 -0.022 0.016 0.012 0.140 (-0.68) (1.64) (-0.33) (1.33) Female LFP -0.087*** 0.010 -0.600 0.010 0.018 0.070 -0.063*** 0.013 -0.434 0.019 0.018 0.129 (-8.58) (0.58) (-4.78) (1.02) Unemployment -0.002 0.018 -0.010 0.052** 0.026 0.275 -0.006 0.017 -0.030 0.052** 0.026 0.274 (-0.11) (2.00) (-0.34) (2.04) Low income -0.030*** 0.012 -0.130 -0.020 0.014 -0.086 -0.025** 0.012 -0.107 -0.014 0.015 -0.059 (-2.60) (-1.38) (-2.08) (-0.93) GDP per capita -0.000** 5.68e-06 -0.195 1.42e-06 6.27e-06 0.024 -0.000** 5.67e-06 -0.200 -9.66e-07 6.58e-06 -0.016 (-2.00) (0.23) (-2.07) (-0.15) Constant 11.375*** 1.102 2.528 1.745 11.290*** 1.018 3.511** 1.592 (10.32) (1.45) (11.09) (2.21) R2 0.857 0.759 0.862 0.762 Wald 2 629.52 3500.27 528.50 2599.06 df 15 18 15 18 Rho 0.392 0.045 647.00 0.054 N 330 200 330 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction and fixed unit effects (not shown); b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); N = number of observations; LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

207 Figure 6.1. Standardized Effects of Political Variables and Provincial Welfare Generosity Index for Total Age-Standardized Mortality Rates

Union Density Union Density (Adjusted) Provincial Welfare Generosity Index

Cumulative Left Political Party Power Cumulative Left Political Party Power (Adjusted) Provincial Welfare Generosity Index

Cumulative Centre Political Party Power Cumulative Centre Political Party Power (Adjusted) Provincial Welfare Generosity Index

Cumulative Right Political Party Power Cumulative Right Political Party Power (Adjusted) Provincial Welfare Generosity Index

Voter Turnout Voter Turnout (Adjusted) Provincial Welfare Generosity Index

Cumulative Women in Government Cumulative Women in Government (Adjusted) Provincial Welfare Generosity Index -0.8 -0.6 -0.4 -0.2 0 0.2 0.4

Notes: Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See Tables 6.4a, 6.5a, and 6.6a for details.

208 Figure 6.2. Standardized Effects of Political Variables and Provincial Welfare Generosity Index for Male Age-Standardized Mortality Rates

Union Density Union Density (Adjusted) Provincial Welfare Generosity Index

Cumulative Left Political Party Power Cumulative Left Political Party Power (Adjusted) Provincial Welfare Generosity Index

Cumulative Centre Political Party Power Cumulative Centre Political Party Power (Adjusted) Provincial Welfare Generosity Index

Cumulative Right Political Party Power Cumulative Right Political Party Power (Adjusted) Provincial Welfare Generosity Index

Voter Turnout Voter Turnout (Adjusted) Provincial Welfare Generosity Index

Cumulative Women in Government Cumulative Women in Government (Adjusted) Provincial Welfare Generosity Index

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4

Notes: Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See Tables 6.4b, 6.5b, and 6.6b for details.

209 Figure 6.3. Standardized Effects of Political Variables and Provincial Welfare Generosity Index for Female Age-Standardized Mortality Rates

Union Density Union Density (Adjusted) Provincial Welfare Generosity Index

Cumulative Left Political Party Power Cumulative Left Political Party Power (Adjusted) Provincial Welfare Generosity Index

Cumulative Centre Political Party Power Cumulative Centre Political Party Power (Adjusted) Provincial Welfare Generosity Index

Cumulative Right Political Party Power Cumulative Right Political Party Power (Adjusted) Provincial Welfare Generosity Index

Voter Turnout Voter Turnout (Adjusted) Provincial Welfare Generosity Index

Cumulative Women in Government Cumulative Women in Government (Adjusted) Provincial Welfare Generosity Index -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 -1E-15 0.1 0.2 0.3 0.4 0.5

Notes: Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See Tables 6.4c, 6.5c, and 6.6c for details.

210 CHAPTER SEVEN. HIERARCHICAL CLUSTER OF LEFTIST POLITICS AND POPULATION HEALTH: A TAXONOMY OF PROVINCIAL REGIMES

Introduction

The previous two chapters tested the central predictions of welfare generosity and leftist politics for population health outcomes in Canada. Chapter 5 demonstrated that medical care, other social services, and post-secondary education are significant and negative predictors of lower total, male, and female mortality rates. These three disaggregated expenditures are combined to create a provincial welfare generosity index that had the largest negative effect on mortality rates. Chapter 6 focused on the health impact of power resources, political parties, and democratic politics, and assessed the pathways through which leftist politics affects population health. The findings show that left political party power and women in government combine with provincial welfare generosity to improve population health, while centre political parties channel their influence through provincial welfare generosity to produce better health outcomes.

These findings provide evidence in support of this dissertation’s conceptual model. In this chapter, I integrate these findings and undertake exploratory analyses to assess the degree to which welfare regime theory can be applied among Canadian provinces to explain population health.

Applying Welfare Regime Theory to Canadian Provinces

Current scholarship on welfare states is heavily influenced by the work of Esping-

Andersen (1990, 1999), which contends that welfare states cluster into different types

211 that are not directly comparable. These institutional clusters, often referred to as

‘regimes’, reflect the genetic historical legacies of social policy development and the state’s particular institutionalized tradition of intervention into the market. In addition to

Esping-Andersen’s influence tripartite typology (social democratic, liberal, and conservative welfare regimes), the extant comparative literature has developed several other institutional typologies that range from political traditions (Huber & Stephens,

2001) and social welfare programs (Korpi & Palme, 1998, 2003) to labour markets

(Chung et al., 2010; Muntaner et al., 2012) and market economies (Hall & Soskice,

2001a, 2001b).

Research on comparative typologies has been so productive that scholars have applied such thinking to identifying potential regimes within nations, including Canada and its provinces. In order to better understand the current state of welfare regime research in Canada, I summarize the relevant works of Boychuk (1998), Bernard and

Saint-Arnaud (2004), and McGrane (2010). The first comprehensive study to apply welfare regime theory within Canada and its provinces was Boychuk (1998). In his book entitled Patchworks of Purpose, Boychuk (1998) argued that Canada does have one social assistance system but rather ten variants that reflect the particular policy goals of each province. Provincial social assistance systems have followed significantly different paths in their historical development even though they are funded under the same federal cost-sharing arrangements. To examine contemporary patterns of social assistance provision specific to particular provinces, Boychuk (1998) proposed five broad groups of provincial assistance regimes: conservative (Ontario, Manitoba, Nova

Scotia, and Newfoundland), market-performance (New Brunswick and British

212 Columbia), market/family enforcement (Saskatchewan and Alberta), redistribute (Prince

Edward Island), and a combined market-performance and market/family enforcement

(Quebec). Conservative assistance regimes make a strong distinction between deserving and undeserving recipients, and assistance to the latter group is particular harsh. Market-performance provinces are distinct in their active promotion of labour market participation through positive employment and incentive programs. The market/family enforcement regime is notable for enforcing the market and family by actively stigmatizing recipients, which has the consequence of making assistance relatively unattractive for all recipients. The social assistance system in Quebec includes positive inducements to labour market participation as well as certain punitive measures. Prince Edward Island’s system resembles a redistributive model, whereby assistance is relatively non-stigmatizing and non-stratifying.

Bernard and Saint-Arnaud (2004) conceptualized Alberta, British Columbia,

Ontario, and Quebec as separate welfare regimes to examine whether these provinces display notable divergences, which translate into different social policies. Using hierarchical cluster methods, provinces are compared and contrasted along three sets of indicators: social situations (e.g., unemployment rate, GDP, inflation, female labour participation rate, infant mortality, life expectancy), public policy (e.g., government expenditures, payroll taxes, debt interest payments, number of physicians per 1,000 people), and civic participation (e.g., voter turnout, daily newspapers read per 1,000, union membership). Initial hierarchical analysis revealed that all four provinces qualify as liberal welfare regimes compared to social democratic and conservative regimes.

Subsequent analyses that created additional clusters show that Quebec shared strong

213 similarities with Nordic countries while Alberta resembled the “ultra-liberal” regime of the

United States. In terms of British Columbia and Ontario, these two provinces had regime profiles that fell in the middle of Quebec and Alberta, and remained closely aligned with liberal welfare regimes.

Most recently, McGrane (2010) used welfare regime theory to compare childcare policy and programs of Canada’s ten provinces and to classify provinces into distinct childcare regimes. Using multidimensional scaling methods with existing childcare

Canadian data, McGrane used four variables to perform cross-provincial classifications and comparisons: availability, affordability, provincial government spending, and quality from 1998 and 2008. Analyses suggested three distinct childcare regimes. The first regime, referred to as the neo-liberal group, consisted Prince Edward Island, Alberta,

Nova Scotia, and Saskatchewan, Newfoundland and New Brunswick. These provincial governments spent relatively moderate amounts on childcare and did not display high levels of quality, affordability, or availability. The second regime represented an inclusive liberal group of Ontario and British Columbia. These two provinces are characterized as high quality, high spending, and high availability but also having fees that are much higher than all other provinces as well as provincial state intervened to create high quality standards and regulated spaces but little had been done to enhance affordability. The third and last regime consists of social democratic provinces such as

Quebec and Manitoba. Relative to the other provinces, these two provinces exhibit good quality, large per-child provincial government spending, high availability of regulated spaces as well as moderately priced fees.

214 Taken together, these studies underscore the potential value of exploring how provinces may operate as sub-national welfare regimes that are distinct from the interventions of the federal government (Bernard & Saint-Arnaud, 2004; Boychuk, 1998,

McGrane, 2010). However, no work to date has applied welfare regime theory in

Canada to understand possible connections between provincial regimes and population levels of health. To this end, I borrow three central ideas from the comparative literature to explore possible connections between provincial regimes, provincial welfare generosity, and population health. First, I apply the idea that welfare states cluster into different types that are not necessarily directly comparable to explore whether provinces cluster into distinct regimes. Given the importance of left political party power and women in government in lowering mortality rates, net of welfare generosity, I use these two variables to examine whether individual provinces cluster together to form distinct

‘political regimes’ and whether these regimes are associated with population health.

According to welfare regime theory, population health outcomes should follow a gradient, whereby the most leftist regime (provinces with the strongest histories of left political party and women in government) should experience better health than provincial regimes that are less left and lean more toward the right political spectrum

(Bambra, 2005; Eikemo & Bambra, 2008).

Second, I test the idea that welfare regimes reflect core institutional differences that cannot be captured by simply analyzing the quantitative levels of welfare generosity. For my purposes, I examine whether provincial regimes can be explained by differences in the provincial welfare generosity index that was developed and tested in chapters five and six. According to welfare regime theory, measures of welfare

215 generosity are inadequate for testing fundamental differences between welfare regimes.

For example, Esping-Andersen (1990) argued that “Expenditures are epiphenomenal to the theoretical substance of welfare states” (p. 19), and “Welfare states may be equally large or comprehensive, but with entirely different effects on social structure” (p. 58).

This implies that using welfare generosity to understand political jurisdictions neglects and obscures deeper institutional sources of variations in population health. If this idea holds true, analyzing provincial regimes instead of provincial welfare generosity should yield a greater understanding of population health.

Third, provincial welfare generosity are expected to have different effects across provincial regimes. Because of different historically institutionalized traditions of left political party power and women in government, the consequences of welfare generosity for population health are expected to vary across provincial regimes. In his

1999 book, Social Foundations of Postindustrial Economies, Esping-Andersen contrasts the effects of Scandinavian social democracy, focused on equalizing resources, against the liberal Anglo-Saxon welfare state, which selectively sponsors disadvantaged groups for mobility. Among liberal regimes, for example, “Market inequalities are unlikely to be affected much by social redistribution,” and in social democratic regimes, “The distribution of resources and life chances should be additionally egalitarian, creating homogeneity not only within the working class, but also between the social classes.”

Based on this line of reasoning, one might expect that provincial welfare generosity has different effects on population health across provincial regimes, meaning that an interactive effect exists between government spending and provincial regimes.

Within this theoretical context, this chapter’s research questions are three-fold:

216 1. Do Canadian provinces cluster together across political factors, measured with

indicators of left political party power and women in government?

2. Are these provincial regimes predictive of population health outcomes? If so, can

inter-regime differences be explained by welfare generosity?

3. What are the interactive effects of provincial regimes and welfare generosity on

mortality rates? In other words, does welfare generosity moderate the effect of

provincial regimes on mortality rates?

Data and Methods

This exploratory study uses a pooled TSCS analysis of Canadian provinces from 1976 to 2008, including Newfoundland, Prince Edward Island, Nova Scotia, New Brunswick,

Quebec, Ontario, Manitoba, Saskatchewan, Alberta, and British Columbia. Data are retrieved from Statistics Canada and its Canadian Socio-Economic Information

Management System (CANSIM) Tables.

Statistical analyses involve six steps. First, I present the mean values left political party power, women in government, and provincial welfare generosity by province, and display correlations between explanatory and dependent variables. Second, I use hierarchical cluster methods to classify provinces left political party power and women in government so that each province in a cluster is similar to the others in that cluster and different from provinces in the other clusters. In this way, the clusters represent different provincial welfare regimes. Hierarchical cluster analysis locates the closest pair of provinces and combines them to form a pair; this (joining cases into pairs or joining two pairs) continues until all cases are in one cluster. Once provinces are joined in a cluster, they remain joined throughout the rest of the analysis (Cramer, 2003; Gough, 2001). In

217 this way, the clusters emerge from the data, facilitating the emergence of provincial welfare state taxonomies. I use two methods to determine whether provinces form distinct regimes along two measures of Leftist politics: cumulative left political party power and cumulative women in government. I first use a proximity matrix to perform a hierarchical cluster analysis, which holds the squared Euclidean distance between all variables. The dissimilarity between leftist variables by provinces is placed in a symmetric matrix. The element in the ith row and jth column displays the dissimilarity between the ith and jth variable. I also produce a dendrogram (e.g., cluster tree) to visually illustrate the arrangement of provincial clusters produced by the hierarchical analyses.

Third, I conduct one-way analysis of variance to test effect of provincial regimes on mortality rates. Fourth, I use Beck and Katz’s (1995) technique for calculating panel- corrected standard (PCSE) estimates for linear TSCS models where the parameters are estimated by Prais-Winsten regression with a first-order autocorrelation correction

(AR(1)). Regression models are progressively adjusted in three steps. In models 1, the effects of provincial regimes are tested; models 2 add provincial welfare generosity; and models 3 tests for interactive effects between regimes and welfare generosity. Because these models include provincial regimes which are coded as binary variables (0,1), fixed-unit effects are not used. Provincial regimes and fixed-unit dummies are both time invariant variables and adding them in the same model eliminates ‘too much’ cross- sectional variance (Huber & Stephens, 2001) and makes it impossible to estimate the effect of provincial regimes (Beck, 2001; Wooldridge, 2000, 2002).

218 Fifth, regression estimates are converted into standardized and semi- standardized coefficients and are reported in bar charts (i.e., each bar represents a provincial regime in a fully adjusted regression model) (see figures 7.3, 7.4, and 7.5 located at end of chapter). Sixth, post-estimation techniques are conducted to check for normality of residuals, multicollinearity of predictors, and robustness of results. All analyses are conducted using Stata/SE 12.1.

Dependent Variables

Total, male, and female age-standardized mortality rates are tested as outcome variables. These measures are the weighted average of age-specific mortality rates per

1,000 persons, where the 1991 Canadian Census of Population is used as the standard population for the calculation. Data is available from 1976 to 2008 for a total of 330 cases.

Independent Variables

Cumulative left party power represents the long-term, historical strength of left political parties (e.g., Green Party, New Democratic Party, Parti Québecois), and is tabulated as the summed proportion of left seats held by all provincial government parties in each individual year since 1960 (Huber et al., 1997, 2004). Cumulative women in government represents the long-term, historical strength of women in government, and is measured as the summed proportion of female seats held by all provincial government parties in each individual year since 1960 (Huber et al., 2004). Provincial welfare generosity index combines three types of disaggregated types of welfare generosity: medical care, other

219 social services, and post-secondary education.

Control Variables

I control for the same variables as in the previous chapters: dependency ratio, women in labour force, unemployment rate, low income, and real GDP per capita. Dependency ratio is measured as the number of “dependents” for every 100 “workers”: youth (ages 0 to 17) + seniors (age 65 or older) per 100 workers (aged 18 to 64). Women in labour force refer to the percentage of women in the labour force. Low income refers to the percentage of all persons in low income after tax, set at 50% of adjusted median household income. Unemployment rate is the percentage of the total labour force unemployed. Real GDP per capita is the gross domestic product in real per capita terms and deflated using the 2002 consumer price index. The provincial welfare generosity index is added as a mediator in models two, and I add transfers and debt charges as province-specific controls. Transfers refers to the current fiscal year dollar value of total transfers, which includes the sum of general and specific purpose transfers, received from other levels of government. Debt charges are measured as real per capita expenditures on debt-servicing.

Results

Table 7.1 outlines the summary statistics for left political party power, women in government, and welfare generosity across provinces (located at end of chapter). The last three rows present pairwise correlation results for each political variable and total, male, and female age-standardized mortality rates. As table 7.1 reveals, there are

220 substantial cross-provincial differences among the political variables under study.

Whereas, the cumulative average of left political party power exceeds 38% in British

Columbia, Manitoba, and Saskatchewan, this average below 5% in Alberta, New

Brunswick, Nova Scotia, Newfoundland, and Prince Edward Island. The historical strength of left parties in Quebec and Ontario are 18% and 24%, respectively.

Regarding cumulative women in government, Alberta and British Columbia stand out with averages above 10%. The majority of provinces have about 5% to 6% of their legislatives represented by women except for Newfoundland and New Brunswick who average less than 3%. Quebec ranked the highest among provinces with positive welfare generosity indexes (e.g., z-scores above the provincial mean), followed in generosity order by British Columbia, Alberta, Saskatchewan, Ontario, and

Newfoundland. Nova Scotia approximated the average in provincial welfare generosity and Manitoba, New Brunswick, and Prince Edward Island are below the mean with negative scores.

In terms of correlation results, a moderate negative relationship was observed between left political party power and mortality rates (total, r(200) = -.30, p < .01; men, r(200) = -.29, p < .01; women r(200) = -.31, p < .01). Much stronger negative associations exist between all three dependent variables and women in government

(total, r(200) = -.65, p < .01; men, r(200) = -.65, p < .01; women r(200) = -.62, p < .01) and provincial welfare generosity (total, r(200) = -.66, p < .01; men, r(200) = -.67, p

< .01; women r(200) = -.53, p < .01).

Hierarchical Clusters

221 The results of the hierarchical cluster analysis are shown in the proximity matrix (Table

7.2) and the dendrogram (Figure 7.1) (both table and figure are located at end of chapter). The proximity matrix reveals the distances between the provinces when they are clustered using the variables, left political party power and women in government.

The closest pairing with a value of 1.0 is between New Brunswick and Prince Edward

Island. Conversely, the greatest distance exists between Saskatchewan and Prince

Edward Island with a value of 55.8. Atlantic provinces are located close enough to form a distinct cluster. Newfoundland, Prince Edward Island, Nova Scotia, and New

Brunswick range in distance from 1.0 to 5.6, for example. British Columbia and

Manitoba are only 3.6 apart and Ontario and Quebec are relatively near with a distance value of 6.3, suggesting the possibility of two additional clusters. As for Alberta and

Saskatchewan, the former is most closely aligned with Prince Edward Island with a distance value of 10.8, and the latter shares a proximity score of 17.6 with Manitoba.

Figure 7.1’s dendrogram graphically presents the information concerning which variables are grouped together by province at various levels of (dis)similarity. At the left of the dendrogram, each province is considered its own cluster. Horizontal lines extend across for each province, and at various (dis)similarity values, these lines are connected to the lines from other provinces with a vertical line. The provinces continue to combine until all the observations are grouped together. Based on the length of horizontal lines and the range of the (dis)similarity axis, four visual clues stand out. First, the length and arrangement of the horizontal lines indicate a large degree of difference between

Saskatchewan, British Columbia, and Manitoba on one hand, and Ontario, Quebec,

Alberta, Nova Scotia, New Brunswick, Prince Edward Island, and Newfoundland on the

222 other. Second, the dendrogram confirms the pairing of Ontario and Quebec. Third, as expected, the largest grouping consists of Nova Scotia, New Brunswick, Prince Edward

Island, and Newfoundland. Fourth, Alberta and Saskatchewan can either be conceptualized as separate clusters or grouped together with their connecting branches.

These two provinces are interesting cases given that Saskatchewan and Alberta display the highest averages of left political party power (55.9%) and cumulative women in government (16.5%), respectively.

Based on proximity matrix and dendrogram results, the hierarchical analysis of left political party power and women in government suggests a typology of three provincial regimes. Saskatchewan, British Columbia, and Manitoba group together to form the first cluster, Ontario and Quebec for the second, and Alberta, Nova Scotia,

New Brunswick, Prince Edward Island, and Newfoundland emerge as a third cluster. In turn, I label these three clusters as leftist, centre-left, and conservative. These three regimes are conceptualized as provincial regimes and coded as new binary (0,1) independent variables. Next, I assess whether these provincial regimes are predictive of total, male, and female mortality rates.

Provincial Regimes and Population Health

Figure 7.2 displays the box plots for total, male, and female age-standardized mortality rates across the three provincial regimes. The distribution of mortality rates follow a subtle gradient with the lowest rates observed among the leftist regime (total: M = 6.75,

SD = 0.84; male: M = 8.61, SD = 1.24; female: M = 5.23, SD = 0.53), followed by the centre-left regime (total: M = 7.06, SD = 1.12; male: M = 9.19, SD = 1.67; female: M =

223 5.48, SD = 0.74). The conservative regime had the highest mortality rates (total: M =

7.27, SD = 0.92; male: M = 9.37, SD = 1.32; female: M = 5.57, SD = 0.62).

One-way analysis of variance tests show that the effect of provincial regimes was significant for total, male, and female age-standardized mortality (see Tables 7.3a, 7.3b,

7.3c located at end of chapter). Post hoc analyses using Bonferroni post hoc criterion for significance indicated that total mortality rates are significantly lower in the leftist regime than the conservative regime (mean mortality rate difference = 0.52, p < 0.01); however, no significant difference was observed between leftist and leftist-centre provinces (mean mortality rate difference = 0.31, p = n.s.). Bonferroni post hoc analyses by gender showed that males in leftist provinces had significantly lower mortality rates compared males in centre-left and conservative regimes (e.g., mean mortality rate difference males in leftist provinces vs. males in centre-left provinces = 0.52, p < 0.02; mean mortality rate difference males in leftist provinces vs. males in conservative provinces = 0.76, p < 0.01). The same advantages are observed among females in leftist provinces; however, the mean differences are more narrow (e.g., mean mortality rate difference females in leftist provinces vs. females in centre-left provinces = 0.25, p

< 0.04; mean mortality rate difference females in leftist provinces vs. females in conservative provinces = 0.34, p < 0.01). Although these provincial regime variations may appear small, they do suggest that patterns in mortality rates correspond to leftist political factors – the historical strength of left parties and women in government are associated with gains in population health. However, a more rigorous test is needed before drawing any conclusions. To this end, I test the predictive value of provincial

224 regimes in the next section by estimating PW-PCSE regression models while controlling for other influences on mortality rates.

Models of Provincial Regimes and Population Health

The unstandardized Prais-Winsten regression coefficients (b) with panel corrected standard errors (SE ), and semi-standardized and standardized coefficients (β) for both sexes, male, and female mortality rates are reported in Tables 7.4a, 7.4b, and 7.4c, respectively (located at end of chapter). Semi-standardized coefficients are calculated for provincial regimes for centre-left and conservative regimes, with leftist regimes as the reference category.

In models 1, both centre-left and conservative regimes had significantly higher total, male, and female mortality rates compared to leftist regimes, which confirm the box plots patterns displayed in figure 7.2. Respectively, centre-left and conservative regimes have 0.47 and 0.38 standard deviations higher rates of total mortality than do leftist regimes, even after adjusting for dependency ratio, female labour force participation, unemployment, low income, and GDP per capita.

In models 2, the provincial welfare generosity index is added and tested as a mediator that links provincial regimes and mortality rates. Transfer payments and debt charges are also added as province-specific controls. For total and female mortality rates, the effects of centre-left and conservative regimes become insignificant.

Differences in provincial welfare generosity account for differences in total and female mortality rates between leftist and centre-left regimes and between leftist and conservative regimes.

225 With respect to male mortality rates, the effect of the centre-left regime becomes insignificant when provincial welfare generosity is added; however, the impact of the conservative provinces remains significant. Thus, male mortality rates in conservative regimes cannot be fully explained through differences in provincial welfare generosity compared to the leftist regime.

In models 3, I test whether mortality rate differences between provincial regimes depends on provincial welfare generosity. These models test if the effects of provincial welfare generosity on mortality rates in centre-left and conservative regimes are significantly different from the effect of welfare generosity on mortality rates in leftist provinces. No significant interactions are observed across all three dependent variables.

The interactive effect of provincial regimes and welfare generosity only approached significance at the 0.10 level between the conservative regime and provincial welfare generosity for total mortality rates (β = .099, z(12) = 1.69, p < .10) and the centre-left regime and welfare generosity for female mortality rates (β = .072, z(12) = -1.71, p < .10). Therefore, the general conclusion reached here is that total, male, and female mortality rates can be adequately explained by simply examining welfare generosity levels, and it is not particularly essential to incorporate provincial regimes into the analysis.

Post-estimation Results

226 substantive results (see Appendix B9 for jackknife results). Re-analyzing the data using alternative estimation models did not substantially change significant results.

Discussion

Guided by the central tenets of welfare regime theory, this chapter tested three hypotheses on the links between provinces, welfare generosity, and population health.

The first hypothesis contends that provinces can be grouped into different regimes based on the historical strength of leftist politics, which are predictive of population health outcomes. A second expectation is that these provincial regimes reflect core differences that cannot be adequately explained by different levels of welfare generosity. And the third hypothesis maintains that provincial welfare generosity will have a differential impact on population health across provincial regimes. The analyses reveal both confirmatory and contradictory evidence for these predictions.

Based on the cumulative strength of left political parties and women in government, I find that provinces cluster into three distinct regimes: 1) leftist

(Saskatchewan, British Columbia, and Manitoba), 2) centre-leftist (Ontario and

Quebec), and 3) conservative (Alberta, Nova Scotia, New Brunswick, Prince Edward

Island, and Newfoundland). These regimes are also predictive of population health outcomes in expected directions. Specifically, I find that total, male, and female mortality rates are lowest among the leftist regime, followed by the centre-leftist regime, and highest among the conservative regime. These findings support Esping-Andersen’s

(1990, 1999) argument that core differences in leftist politics matter and augment existing classifications developed within Canada (Bernard & Saint-Arnaud, 2004;

227 Boychuk, 1998, McGrane, 2010). Despite the fact that provinces are sub-national entities that operate under a federal government, the results demonstrate the advantage of viewing provinces as distinct leftist political regimes, which in turn, are associated population levels of health. Some provinces are more egalitarian than other provinces, and this has consequences for population health.

Regarding the second hypothesis, I find that provincial regime differences in population health are mostly explained by differences in provincial welfare generosity.

When models are adjusted for welfare generosity, the analyses reveal that compared to the leftist regime (reference category), the effects of centre-left and conservative regimes become insignificant for most health outcomes. This supports a general conclusion that provincial welfare generosity explains variations in population health better than the long-term differences in leftist politics across provincial regimes.

However, there is one exception to this pattern. Compared to the leftist regime, the conservative regime remains significant with respect to male mortality rates even after provincial welfare generosity is added to the model. There are a couple of possible explanations for this finding. First, conservative provinces may not provide adequate levels of welfare generosity that extend beyond health, social services, and education,

For example, less than generous provisions in housing or on the environment might explain the poorer health among males in conservative provinces. Second, the historical differences in political ideology between leftist and conservative regimes might explain the regime differences in male mortality rates. According to welfare regime theory, provincial welfare generosity is unable to adequately explain male mortality differences because leftist and conservative provinces diverge in fundamental and qualitative ways.

228 Compared to the social democratic preferences of leftist provinces, the adherence to fiscal conservative policies among Alberta, Nova Scotia, New Brunswick, Prince Edward

Island, and Newfoundland might lead to unintended population health consequences among males. Clearly, more research is needed to clarify this finding. For the most part, however, provincial welfare generosity is able to explain population health differences better than the leftist political histories of provincial regimes.

Lastly, I find no evidence for the third hypothesis, or the claim that provincial welfare generosity has difference effects on population health across provincial regimes. Collectively provided welfare resources appear to yield the same health- promoting effects in leftist provinces as they do in centre-leftist and conservative regimes. Provincial welfare generosity improves population health, regardless of provincial regime.

229 Table 7.1. Summary Statistics of and Correlations between Leftist Politics and Total, Male, and Female Mortality Rates

Province Left Political Women in Welfare Party Powera,d Governmentb,d Generosityc,e Newfoundland 0.40 2.51 0.14 Prince Edward Island 0.12 6.08 -0.71 Nova Scotia 4.67 2.81 0.00 New Brunswick 0.39 5.12 -0.54 Quebec 24.64 5.80 0.39 Ontario 18.38 5.55 0.17 Manitoba 38.35 6.69 -0.09 Saskatchewan 55.94 5.38 0.18 Alberta 2.95 16.51 0.20 British Columbia 38.14 10.29 0.26

Correlation with TASMRe -0.30*** -0.65*** -0.66*** Correlation with MASMRe -0.29*** -0.65*** -0.67*** Correlation with FASMRe -0.31*** -0.62*** -0.53*** Notes: TASMR = total age-standardized mortality rate; MASMR = male age-standardized mortality rate; FASMR = female age-standardized mortality rate. a Measured as cumulative left seats (Green, NDP, PQ) as a percentage of seats held by all government parties since 1960. b Measured as cumulative average of seats held by women as a percentage of total seats in provincial parliament since 1960. c Measured as the average of z-scores of medical, other social services, and post-secondary education disaggregated expenditures. d Based on 330 observations. e Based on 200 observations. *** p < 0.000.

230 Table 7.2. Hierarchical Cluster Analysis Proximity Matrix across Canadian Provinces (squared Euclidian distance)

1 2 3 4 5 6 7 8 9 10 1) Newfoundland 0 2) Prince Edward Island 3.6 0 3) Nova Scotia 4.3 5.6 0 4) New Brunswick 2.6 1.0 4.9 0 5) Quebec 24.5 24.5 20.2 24.3 0 6) Ontario 18.2 18.3 14.0 18.0 6.3 0 7) Manitoba 38.2 38.2 33.9 38.0 13.7 20.0 0 8) Saskatchewan 55.6 55.8 51.3 55.6 31.3 37.6 17.6 0 9) Alberta 14.2 10.8 13.8 11.7 24.2 18.9 36.7 54.2 0 10) British Columbia 38.5 38.3 34.3 38.1 14.2 20.3 3.6 18.5 35.7 0

231 Table 7.3a. Summary of One-way Analysis of Variance: Provincial Regimes and Total Age-Standardized Mortality

Sum of Squares Df Mean Square F Between Groups 16.40 2 8.20 9.21*** Within Groups 290.91 327 .89 Total 307.31 329 .93 *** p < 0.0001 Rates

Table 7.3b. Summary of One-way Analysis of Variance: Provincial Regimes and Male Age-Standardized Mortality Rates

Sum of Squares Df Mean Square F Between Groups 36.07 2 18.04 9.58*** Within Groups 615.85 327 1.88 Total 651.92 329 1.98 *** p < 0.0001

Table 7.3c. Summary of One-way Analysis of Variance: Provincial Regimes and Female Age-Standardized Mortality Rates

Sum of Squares df Mean Square F Between Groups 7.19 2 3.60 9.30*** Within Groups 126.49 327 .39 Total 133.68 329 .41 *** p < 0.0001

232 Table 7.4a. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976- 2008. Model 1 Model 2 Model 3 Variables b SE β b SE β b SE β Leftist Regime (reference) Centre-Left 0.422*** 1.392 0.469 0.235 0.151 0.434 0.321** 0.141 0.595 Regime (3.03) (1.56) (2.28) Conservative 0.340*** 1.449 0.379 0.217* 0.113 0.401 0.263** 0.111 0.487 Regime (3.27) (1.91) (2.37) Welfare Generosity -0.314*** 0.087 -0.448 -0.325*** 0.097 -0.463 (-3.62) (-3.35) Centre-Left * -0.170 0.118 -0.103 Welfare Generosity (-1.44) Conservative * 0.167* 0.099 0.148 Welfare Generosity (1.69) Transfers 0.000 0.000 0.079 0.000 0.000 0.081 (0.82) (0.85) Debt -0.000 0.000 -0.006 -0.000 0.000 -0.070 (-0.05) (-0.51) Dependency Ratio 0.032*** 0.010 0.343 0.031*** 0.011 0.334 0.038*** 0.011 0.405 (3.32) (2.83) (3.52) Female LFP -0.095*** 0.013 -0.824 -0.062*** 0.018 -0.535 -0.063*** 0.017 -0.542 (-7.27) (-3.51) (-3.71) Unemployment -0.006 0.020 -0.039 0.041* 0.024 0.293 0.038* 0.022 0.272 (-0.28) (1.76) (1.74) Low income -0.035*** 0.013 -0.193 -0.10 0.019 -0.054 -0.009 0.019 -0.050 (-2.82) (-0.51) (-0.49) GDP per capita -6.37e-06 5.73e-06 -0.100 0.000* 7.24e-06 0.208 6.79e-06 7.45e-06 0.107 (-1.11) (1.83) (0.91) Constant 10.851*** 1.232 7.418*** 1.276 7.381*** 1.225 (8.80) (5.81) (6.03) R2 0.879 0.874 0.875 Wald 2 186.84 300.00 300.85 df 7 10 12 (Continued on next page)

233 Table 7.4a. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Total Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008. (Continued) Rho 0.698 0.533 0.492 N 330 200 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction; b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

234 Table 7.4b. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Male Age-Standardized Mortality Rates in Canadian Provinces, 1976- 2008. Model 1 Model 2 Model 3 Variables b SE β b SE β b SE β Leftist Regime (reference) Centre-Left 0.703*** 0.226 0.537 0.547* 0.299 0.643 0.694** 0.291 0.816 Regime (3.11) (1.83) (2.38) Conservative 0.444*** 0.166 0.339 0.490** 0.233 0.577 0.559** 0.229 0.658 Regime (2.67) (2.81) (2.45) Welfare Generosity -0.409*** 0.131 -0.371 -0.431** 0.174 -0.391 (-3.13) (-2.48) Centre-Left * -0.280 0.208 -0.108 Welfare Generosity (-1.35) Conservative * 0.262 0.175 0.148 Welfare Generosity (1.50) Transfers 0.000 0.000 0.045 0.000 0.000 0.048 (0.54) (0.59) Debt -0.000 0.000 -0.083 -0.000 0.000 -0.143 (-0.70) (-1.08) Dependency Ratio 0.048*** 0.015 0.328 0.063*** 0.019 0.432 0.073*** 0.019 0.499 (3.22) (3.26) (3.85) Female LFP -0.131*** 0.020 -0.722 -0.100*** 0.026 -0.550 -0.100*** 0.026 -0.547 (-6.68) (-3.78) (-3.90) Unemployment 0.014 0.028 0.065 0.044 0.037 0.196 0.040 0.034 0.178 (0.51) (1.19) (1.15) Low income -0.058*** 0.018 -0.203 -0.017 0.028 -0.059 -0.015 0.027 -0.052 (-3.16) (-0.60) (-0.55) GDP per capita -5.37e-06 8.34e-06 -0.054 9.24e-06 0.000 0.092 -6.23e-07 0.000 -0.006 (-0.64) (0.85) (-0.06) Constant 13.874*** 1.941 9.794*** 2.137 9.633*** 2.070 (7.15) (4.58) (4.65) R2 0.846 0.859 0.863 Wald 2 200.34 195.93 218.66 Df 7 10 12 (Continued on next page)

235

Table 7.4b. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Male Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008. (Continued) Rho 0.702 0.587 0.558 N 330 200 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction; b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

236 Table 7.4c. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008. Model 1 Model 2 Model 3 Variables b SE β b SE β b SE β Leftist Regime (reference) Centre-Left Regime 0.247*** 0.082 0.426 0.067 0.056 0.196 0.127** 0.051 0.373 (3.00) (1.12) (2.51) Conservative 0.218*** 0.077 0.376 0.042 0.056 0.124 0.082 0.059 0.240 Regime (2.84) (0.70) (1.38) Welfare Generosity -0.183** 0.080 -0.414 -0.182** 0.089 -0.412 (-2.30) (-2.04) Centre-Left * -0.123* 0.072 -0.119 Welfare Generosity (-1.71) Conservative * 0.111 0.078 0.156 Welfare Generosity (1.42) Transfers 0.000 0.000 0.112 0.000 0.000 0.109 (0.77) (0.75) Debt 0.000 0.000 0.188 0.000 0.000 0.117 (1.08) (0.60) Dependency Ratio 0.013* 0.007 0.214 0.006 0.008 0.097 0.010 0.009 0.175 (1.75) (0.66) (1.18) Female LFP -0.071*** 0.010 -0.980 -0.036** 0.014 -0.493 -0.036** 0.015 -0.495 (-6.99) (-2.42) (-2.47) Unemployment -0.010 0.018 -0.114 0.039** 0.018 0.441 0.037* 0.019 0.411 (-0.58) (2.02) (1.92) Low income -0.029** 0.012 -0.255 -0.014 0.016 -0.126 -0.014 0.018 -0.121 (-2.53) (-0.80) (-0.78) GDP per capita -4.55e-06 5.09e-06 -0.114 0.000** 4.52e-06 0.350 8.88e-06 5.41e-06 0.222 (-0.89) (2.87) (1.64) Constant 9.111*** 0.895 5.953*** 0.899 5.934*** 0.997 (10.18) (5.90) (5.95) R2 0.837 0.791 0.786 Wald 2 137.30 205.74 242.36 df 7 10 12 (Continued on next page)

237 Table 7.4c. PW-PCSE Models of Provincial Regimes and Provincial Welfare Generosity on Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008. (Continued) Rho 0.608 0.418 0.381 N 330 200 200 Notes: PW-PCSE = Prais-Winsten regression with correlated panels corrected standard errors; Models include first-order serial autocorrelation correction; b = unstandardized coefficient; SE b = unstandardized coefficient standard error; β = standardized coefficient; z-scores are in parentheses; R2 = coefficient of determination; df = degrees of freedom; Rho = common autoregressive term (1); LFP = labour force participation; GDP = gross domestic product. * p < .10; ** p < .05; *** p < .01 (two-tailed tests).

238 Figure 7.1. Dendrogram - Hierarchical Cluster Analysis of Left Political Party Power and Women in Government among Canadian Provinces

NL 0 2000 4000 6000 L2squared dissimilarity measure

Notes: SK = Saskatchewan; BC = British Columbia; MB = Manitoba; ON = Ontario; QC = Quebec; AB = Alberta; NS = Nova Scotia; NB = New Brunswick; PE = Prince Edward Island; NL = Newfoundland and Labrador.

239

Figure 7.2. Box Plots of Total, Male, and Female Age-Standardized Mortality Rates across Provincial Regimes

6 4

Leftist Centre-Left Conservative TASMR MASMR FASMR

Notes: Leftist = Saskatchewan, British Columbia, and Manitoba; Centre-Left = Ontario and Quebec; Conservative = Alberta, Nova Scotia, New Brunswick, Prince Edward Island, and Newfoundland; TASMR = Total age-standardized mortality rates; MASMR = Male age-standardized mortality rates; FASMR = Female age- standardized mortality rates.

240 Figure 7.3. Semi-Standardized and Standardized Effects of Provincial Regimes and Provincial Welfare Generosity Expenditures for Total Age-Standardized Mortality Rates

Model 1

Centre-Left

Conservative

Model 2

Centre-Left

Conservative

Provincial Welfare Generosity Index

Model 3

Centre-Left

Conservative

Welfare Generosity

Centre-Left * Welfare Generosity

Conservative * Welfare Generosity

-0.6 -0.4 -0.2 -1E-15 0.2 0.4 0.6 0.8

Notes. Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See Tables 7.3a, 7.4a, and 7.5a for details.

241 Figure 7.4. Semi-Standardized and Standardized Effects of Provincial Regimes and Provincial Welfare Generosity Expenditures for Male Age-Standardized Mortality Rates

Model 1

Centre-Left

Conservative

Model 2

Centre-Left

Conservative

Provincial Welfare Generosity Index

Model 3

Centre-Left

Conservative

Welfare Generosity

Centre-Left * Welfare Generosity

Conservative * Welfare Generosity

-0.5 -0.3 -0.1 0.1 0.3 0.5 0.7 0.9

Notes. Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See Tables 7.3b, 7.4b, and 7.5b for details.

242 Figure 7.5. Semi-Standardized and Standardized Effects of Provincial Regimes and Provincial Welfare Generosity Expenditures for Female Age-Standardized Mortality Rates

Model 1

Centre-Left

Conservative

Model 2

Centre-Left

Conservative

Provincial Welfare Generosity Index

Model 3

Centre-Left

Conservative

Welfare Generosity

Centre-Left * Welfare Generosity

Conservative * Welfare Generosity

-0.6 -0.4 -0.2 -1E-15 0.2 0.4 0.6

Notes: Black bars are significantly different from zero (p < 0.05). Gray bars are not significantly different from zero. See Tables 7.3c, 7.4c, and 7.5c for details.

243 CHAPTER EIGHT. DISCUSSION

This chapter consists of three sections. First, I summarize major findings by revising this dissertation’s original conceptual model based on the results from chapters 5 and 6.

Second, I outline dissertation limitations that constrain the generalizability or utility of major findings. And third, I offer directions for future research that augment this dissertation’s findings.

Revision of Conceptual Model and Summary of Findings

Based on the findings from chapters 5 and 6, this dissertation’s original conceptual model can now be revised to more accurately reflect the connections between leftist politics, provincial welfare generosity, and population health among Canadian provinces

(see figure 8.1 located at end of chapter). The solid arrows between boxes show relationships that are confirmed using regression analyses, net of provincial, demographic, and economic factors, while the dotted arrows indicate basic associations confirmed with correlations.

Provincial Welfare Generosity Improves Population Health

According to welfare generosity theory, increases in provincial expenditures are associated with better population health outcomes (Dahl & van der Wel, 2013; Fritzell &

Lundberg, 2007; Lundberg et al., 2008). I find support that provincial welfare generosity reduces mortality rates using multiple aggregate and disaggregated measures of provincial welfare expenditures. This is consistent with the idea that provincial welfare

244 generosity operates as a proximate determinant of population health among Canadian provinces. The analyses also reveal that not all provincial welfare expenditures matter.

Kanga’s (2010) assertion that “bigger is better” comes with an important caveat - general increases in welfare generosity are insufficient to improve population health.

More accurately, I find that targeted spending on medical care, other social services, and post-secondary education yields the largest returns to the public’s health. In this respect, improving population health through the provision of health-promoting welfare resources is squarely within the jurisdictional powers of provincial governments. As figure 8.1 shows, the connection between provincial welfare generosity and population health remains unchanged; however, our understanding of which welfare expenditures reduce mortality rates is now more nuanced and precise.

As Left and Centre Political Party Power Increases, Mortality Rates Decline

To understand how leftist politics affect population health among Canadian provinces, this dissertation finds evidence in favour of power constellations theory

(Huber & Stephens, 2001). Despite the fact that power constellations theory was originally developed to explain welfare state development among OECD countries, I find that its central predictions are also applicable to understand provincial differences in population health. The political power of left and centre parties have expected impacts on mortality rates. Provinces with stronger left and centre political parties have significantly lower mortality rates. These findings substantiate the general idea that different political parties pursue different policies, and in this case, shifts to the political

245 left and centre are associated with generous welfare expenditures, which in turn, improve population health (Blais, Blake, & Dion, 1993).

Furthermore, the pathways through which these two parties affect population health are different. Whereas left parties combine with provincial welfare generosity to improve population health, the impact of centre parties channels through provincial expenditures. I also find that right political party power is positively associated with mortality rates, suggesting that right parties are more likely to pursue fiscal conservative policies, which in turn, have a negative impact on population health. Figure 8.1 reflects these new findings and pathways: left parties are connected directly to provincial welfare generosity and population health, centre parties are connected indirectly to population health through welfare generosity, and right parties have been omitted given their negative health effects.

As More Women are Elected into Government, Mortality Rates Decline

Political parties are not the only leftist political factor that improves population health. This dissertation also finds that the gender equality in politics has a positive impact. According to the politics of presence theory, gender has an effect on parliamentary processes, and specifically, female politicians strengthen the position of women’s interests in governments (Phillips, 1995). My conceptual model broadens this idea through an examination of the population health effects of the long-term strength of women elected into political office. The results support the conclusion that women in government have a significant impact in reducing mortality rates, and in fact, this political factors turns out to have the largest effect of any of the leftist political indicators.

246 Similar to left political party power, women in government remains significant after controlling for provincial welfare generosity, demonstrating that women in government combines with welfare spending to reduce mortality rates. Figure 8.1 shows a combined pathway which connects women in government to provincial welfare generosity as well as to population health.

Unexpectedly, Union Density and Voter Turnout Do Not Reduce Mortality Rates

Contrary to the expectations of theories power resources and political democracy, I find that union density and voter turnout do not improve population health

(reduce mortality rates). However, the results from bivariate associations provide some tentative hints on how labour unions and voter turnout might be related to other leftist political factors that are predictive of lower mortality rates. I find significant and positive associations between union density and left political parties and union density and centre political parties. Interestingly, the latter association is even stronger than the former, which contradicts power resources’ idea that working-class interests are organized and mobilized through labour unions and left parties (Korpi, 1979; Stephens,

1983). Labour unions align with and support the election of leftist parties, and leftist political parties are supposed to promote conditions which allow labor unions to thrive

(Neumann, Pedersen, & Westergard-Nielson, 1991). With this said, it is possible that labour unions affect health through their continued support of left and centre political parties. Similarly, increases in voter turnout are associated with the long-term strength of centre political parties, not left political parties, which is inconsistent with political democracy theories (Hicks & Swank, 1992; Piven & Cloward, 1994; Iversen, 2001).

247 Therefore, if unions and voting matter for population health, their respective impacts might be limited to increasing the long-term political power of left and centre parties, which in turn, improve population health through channeled and combined pathways. At best, these suggested associations are conjectures. Clearly, more research is needed to confirm or refute these possible interrelationships. Figure 8.1 depicts these bivariate associations with dotted arrows.

Provinces Cluster into Distinct Regimes which are Predictive of Population Health

In addition to testing and revising the above conceptual model, this dissertation assessed the extent to which Esping-Andersen’s (1990, 1999) welfare regime theory is applicable within the context of Canadian provinces and population health. My strategy was to examine whether provinces cluster together in meaningful ways along the two leftist political indicators that combine with provincial welfare generosity to improve population health: left political party power and women in government. Following this strategy achieved two objectives. First, it allows an examination of how provinces may be viewed as regimes which reflect historical differences in their commitment to egalitarian outcomes. Historical differences in leftist politics produced a three-regime model with Saskatchewan, British Columbia, and Manitoba forming the most egalitarian regime (leftist); Alberta, Nova Scotia, New Brunswick, Prince Edward Island, and

Newfoundland representing the most conservative regime; and Ontario and Quebec in the middle of these two positions, representing a centre-left regime. Although provinces belong to the same liberal welfare regime at a national level, the results support the idea

248 that provinces are sufficiently dissimilar in terms of political differences that justify a welfare regime approach (Bernard & Saint-Arnaud, 2004; Morel, 2002).

Second, it allows for the first time an analysis of how provincial regimes, provincial welfare generosity, and population health might are connected. Consistent with welfare regime theory, the effect of provincial regimes was significant, such that average mortality rates are the lowest and highest for leftist and conservative regimes, respectively. However, less support for welfare regime theory is found when provincial welfare generosity is tested as a mediating factor between provincial regimes and mortality rates. For the most part, I find provincial welfare generosity is able to explain population health differences better than the leftist political histories of provincial regimes. The lone exception is male mortality rates in the conservative regime compared to the leftist regime. I also find that provincial welfare generosity reduces mortality rates, regardless of provincial regime. The results provide partial support to

Esping-Andersen’s claims. On one hand, provincial regimes are associated with predictable population health differences; however, most of these differences can be adequately explained by provincial levels of welfare generosity.

Dissertation Limitations

A number of caveats need to be noted regarding this dissertation’s design and methods that might constrain the generalizability or utility of major findings. The first limitation relates to the unavailability of Aboriginal and racial/ethnic population as demographic controls. As noted in the Methods, these two variables are not available due to a lack of comparable data over time. Due to serious problems with the total counts of Aboriginal

249 people in the four censuses during the 1980s and 1990s, there are substantial differences between actual and estimated population figures (Saku, 1999). For example, the census has defined Aboriginal in several inconsistent ways that range from self-identification and Aboriginal ancestry to ‘registered Indian’ under the Indian Act

(Saku, 2010). Similarly, the historical comparability of racial or ethnic origin data has been affected by significant differences in the wording, format, examples, and instructions of census questions (Statistics Canada, 2006). For example, prior to 1996, data on visible minorities are derived from responses to the ethnic origin question, in conjunction with other ethno-cultural information, such as language, place of birth, and religion. Clearly, this is an important limitation given that Aboriginal and racial/ethnic groups are significantly more likely to experience a wide range of social and health disadvantages than non-Aboriginal and non-racial/ethnic populations (Mikkonen &

Raphael, 2010). As data on Aboriginal and racial/ethnic populations gain more reliability over time, future work using TSCS methods should conduct stratified analyses among these two groups or include them as potential confounding variables.

Second, the available time frame to estimate the effects of provincial welfare generosity on mortality rates was considerably shorter than desired. Unfortunately, 1989 is the earliest year for which valid and reliable data on aggregate and disaggregated provincial expenditures are available from Statistics Canada (Landon et al., 2006). As a matter of fact, Statistics Canada recommends against using earlier expenditure data that is available in CANSIM matrices 2782 to 2791 and in the Statistics Canada publication Public Finance Historical Data (68-512). According to Landon et al. (2006), there are serious discrepancies between the available data used in this dissertation and

250 the older series for the years in which the two data series overlapped. As a consequence, provincial welfare generosity analyses are limited to a total of 200 cases

(1989-2008), and leftist political analyses had a total of 330 cases (1976-2008). The only expenditure variable that is available for a longer period of time is health (Canadian

Institute for Health Information, 2004, 2011). Despite the shorter than desired time frame, the analyses still found significant spending effects on mortality rates.

Third, this dissertation focused on all-cause age-standardized mortality rates, which is not an inherent limitation in and of itself, but still has important implications. For example, the effect of politics on specific-causes of mortality rates remains unknown. In fact, the focus on all-cause deaths might partially explain the insignificant finding between labour unions and mortality rates. For example, Muntaner et al. (2002) found strong associations between labour unions and injury mortality. In terms of age- standardized mortality rates, the clear advantage of this outcome is that provides a single numerical indicator that allows different provinces to be compared over time.

However, at the same time, my focus on age-standardized rates overlooks how politics may exert a differential effect on age groups. For example, Ariizumi and Schirle (2012) estimated the effect of unemployment rates on Canadian age-specific mortality rates and found that a one percentage point increase in the unemployment rate lowers the predicted mortality rate of individuals in their 30s by nearly 2%. To their credit, these authors moved beyond focusing on age-standardized rates or working-age population

(18-64) to find that improved health is driven by those aged 30-39. Given these limitations, future politically-oriented studies among Canadian provinces should expand

251 mortality outcomes to include cause-specific rates and stratify mortality rates by age- groups.

Fourth, a potential limitation regards the implicit assumption of uniform lags with respect to time invariant variables. If changes in the political variables are rare and of little persistence, a good estimate critically depends on the selection of a correct lag structure (Plümper et al., 2005). When estimating the effect of institutional changes on a dependent variable, lags may seriously differ from unit (province) to unit (province). In other words, the lag it takes for an independent variable to affect a health outcome may not be identical across provinces. Because the estimation of unit dependent lags is difficult and time-consuming, most researchers either do not lag the independent variables or arbitrarily choose uniform lags – most often one-year lags. In this dissertation, I chose the former strategy to keep models as simple as possible and to maintain all possible cases. I am confident this decision did not significantly affect observed coefficients and levels of significance since the institutional settings and environments between provinces are more alike than different (Döring, 1995; Tsebelis,

2002). Since provinces operate under the jurisdiction of a federal government, legislative institutions, electoral systems, and levels of government autonomy are homogenous across the provinces. Future population health research that compares countries with different institutional settings should pay more attention to determining the correct lag length using various recommended methods (e.g., the t-statistic, the R2, the AIC (Akaike Information Criterion), or the BIC (Bayesian Information Criterion))

(Plümper et al., 2005).

252 Fifth, since this dissertation uses a static approach to examine the links between politics and health, it implicitly assumes that parameter values remain constant over time (Kennedy, 1998; Plümper et al., 2005). Thus, my estimation strategy assumes that the population health effects of political variables are stable over time. As a consequence, possible ‘structural changes or breaks’ such as changes in slopes and changes in error variances are overlooked (Maddala, 1998). In more concrete terms, it has been argued that leftist policies have changed over time and governments dominated by left parties no longer implement a Keynesian macro-economic policy

(Cusack, 1997; Piazza, 2001; Plümper et al., 2005). To adequately control for this hypothesis, future work should allow for parameter heterogeneity. One suggested option involves identifying time periods, adding time period dummies, and creating interaction effects between all time periods dummies and the suspected variable that has unstable effect over time (e.g., cumulative left political party power x 1976-1980, cumulative left political party power x 1981-1985 …. cumulative left political party power x 2000-2004) (Plümper et al., 2005). A second option is to a build two-level conditional hierarchical model to explain variability over time. At the lower level, time periods are introduced as random effects, which are then nested within units of analysis at a higher level (e.g., countries, provinces, states), which are entered as a fixed effects indicator.

Chung and Muntaner (2007) took advantage of this latter approach to show that infant mortality rates among welfare regimes converged in the 1980s; however, the trend for low birth weight actually increased in the 1990s between the social democratic regime and other regimes.

253 Suggestions for Future Research

In addition to the suggested research directions noted above (e.g., incorporating data on Aboriginal and racial/ethnic groups; testing specific causes of mortality among different age-groups; selecting correct lag structures; and testing the conditional effects of time), I suggest below a program of research within Canada that augments this dissertation’s major findings.

Expanding the Scope of Provincial Welfare Generosity Measures

This dissertation selectively focused on health, social services, and education expenditures as indicators of provincial welfare generosity. Despite that fact that these three types of spending account for almost two-thirds of all provincial spending, the remaining (unmeasured) expenditures may also exert a significant and positive impact on population health. For example, unexamined provincial expenditures that may also operate as health-promoting welfare resources include spending on environmental protection (e.g., pollution control, water purification), labour, employment, and immigration (e.g., promotion of improved working conditions, assistance to immigrants), housing (e.g., housing operations and assistance), and transportation (e.g., road transport, public transit) (Statistics Canada, 2009). Thus, future work should expand upon the breadth of provincial welfare expenditures available and test them as possible predictors of population health. In question form, “Do other kinds of aggregate and disaggregated types of provincial welfare generosity, besides health, social services, and education, lower total, male, and female age-standardized mortality rates, net of province-specific factors?”

254

Expanding the Scope of Leftist Political Indicators

Second, this dissertation has only scratched the surface with respect to testing leftist politics as macro-level determinants of population health. Our understanding of the causal links between leftist politics and population health has much room to grow by testing new indicators as well as increasing the sophistication of existing measures.

Future research on the health effects of politics would advance in important ways if the following distal determinants are tested as independent variables:

Union Density by Gender, Sector, and Industry. This dissertation uses union density as a single indicator of market power resources. The average rate of union density among provinces from 1976 to 2008 is 34.3%. However, this single indicator obscures potentially important differences in unionization rates across gender, sectors, and industries that might affect population health. Historically, membership rates in unions by male employees are higher than membership rates by female employees. For example, in 1997, union coverage stood at 35.2% for male employees compared to

32.1% for female employees, converged at 32% in 2005, and has since diverged, with the female and male rates at 32.8% and 30.3% in 2012, respectively (Statistics Canada,

2013). Over the past 15 years, male unionization rates have declined and female rates have increased slightly. Guided by power resources theory, this leads one to speculate whether changing trends in unionization rates by gender have a systematic and predictable influence on male and female mortality rates among Canadian provinces.

Recent trends in unionization rates between the public and private sector warrant closer attention. There has been a notable decline in private sector union coverage,

255 dropping from 19.9% in 2001 to 17.5% in 2010 (Uppal, 2011). Declines in union coverage in the private sector has been attributed to various factors, including legislative changes that are less favourable to certification success, greater management resistance to unionization, industrial regulation, and relative decline in the traditionally more unionized occupations and industries (Human Resources and Skills

Development Canada, 2012). In contrast, union coverage in the public sector has remained relatively unchanged over the same period of time, decreasing slightly to

74.9%. Given these stark differences, future work could consider, “What is the extent to which historical differences in unionization rates between the stronger public sector and weaker private sector explain population health outcomes among Canadian provinces?”

In addition to gender and sector, union coverage also varies significantly by industry. In 2010, union coverage varied across health care and social assistance (21%), education services (18%), public administration (16%), manufacturing (10%), trade (8%), transportation and warehousing (6%), construction (6%), information, culture, and recreation (4%), and other industries (11%) (HRSDC, 2012). Recent trends suggest that unionization rates in health care and social assistance and education services have decreased while coverage rates have increased utilities, transportation and warehousing, and public administration (HRSDC, 2012). Such industry differences and trends reveal a more nuanced understanding of market power resources, leading one to ask whether differences in union coverage across industries have a significant impact on population health among Canadian provinces.

Collective bargaining coverage. This variable refers to employees who are card- carrying members of a union (e.g., union density rates) as well as employees who are

256 not union members but whose jobs are covered by a union contract. A non-union member can be covered by an agreement through the exercise of rights under the Rand

Formula, a declaration of coverage status by the employer, coverage of newly hired employees serving their probationary periods, or the extension or matching practices used by some employers for certain out-of-scope employees (Akyeampong, 2000). It is possible that collective bargaining coverage is a more accurate measure of market power resources. First, collective bargaining rates are higher than union density rates since non-union workers, such as lower level supervisors are covered even though they are not union members (Jackson, 2006). Higher collective bargaining rates suggest greater market power resources to advance the interests of working class. Second, this measure partly reflects important variations in labour power due to organizational differences (Navarro, 2003). Collective bargaining has been theorized as a powerful means to facilitate bargaining coordination; that is, the extent of coordination between unions and employers’ organizations in wage setting and other aspects of industrial relations (e.g., wage scales, working hours, training, health and safety, overtime, grievance mechanisms, and rights to participate in workplace or company affairs) (Aidt

& Tzannatos, 2002). Given this rationale, future work could explore, “Does collective bargaining coverage operate as an effective market power resource that improves population health through the protection of working conditions in the market and through the support of left and centre parties in elections?”

Work stoppages. This dissertation did not test the possible health effect of work stoppages, which refers to all major strikes and lockouts involving a pre-determined number of employees (e.g., 500 or more employees). Strikes and lockouts stoppages

257 can be viewed as a leftist political process since the immediate power resource available to wage earners in capitalist democracies is the ability to quit working and deprive the employer of labor. When workers can bargain collectively and interrupt the ability of business to make profits, their power resources can be more fully actualized and their interests advanced. Significant variations exist across provinces in work stoppages and lockouts. For example, close to half of strikes and lockouts from 2003 to

2005 occurred in Quebec (45.2%), followed by Ontario (31%), and British Columbia

(5.1%) (Akyeampong, 2006). Following the lead of Muntaner et al. (2002), future work should explore, “Do work stoppages have the effect of improving working conditions, increasing wages, strengthening the political power left parties, which in turn, improve population levels of health?”

Electoral competition. This dissertation examines only one facet of political democracy theory, voter turnout. Another leftist political indicator that might reflect the strength of democracies and the interests of the working-class is electoral competition.

According to the democratic politics literature, electoral competition is a nonpartisan phenomenon that augments welfare effort because pro-welfare appeals are made to

“median” voters by competing political parties (Dye, 1979; Pampel & Williamson, 1988).

Future research can broaden this work by asking, “Does electoral party competition have an effect of increasing provincial welfare generosity, which in turn, improves population health among Canadian provinces?” Electoral competition has been measured in the extant literature with the following formula (Hicks & Swank, 1992;

Pampel & Williamson, 1988):

258 where in equation (2):

= 1 …. N political parties (3 to represent left, centre, and right parties)

= 1 …. proportional vote for political party (popular vote for left, centre, and right parties)

Power of women in government. This dissertation tests the politics of presence theory by conceptualizing the long-term, historical strength of women in government as a political determinant of population health. Because this variable is measured as the summed proportion of female seats held by all political parties, it only tells us about presence, not influence. Kawachi et al. (1999) examined the health effects of both presence and influence by assigning a weight to women office-holders in each US according to the degree of political influence of the position (e.g., state representatives are given a weight of 1.0, compared to 1.75 for US senators). A similar weighting scheme can also be applied within the context of Canadian provinces. For example, women in provincial legislatures could be assigned a weight of 1.0, women holding a cabinet position with 1.5, and female premiers with a weight of 2.0. The supporting rationale is that cabinet and premier positions are stronger positions from which to wield power in an executive-centred and party-disciplined parliamentary system (Studlar &

Moncrief, 1997, 1999). Future research on the connections between gender equality in politics and population health could address such questions as, “Are increases in female cabinet ministers significantly and positively associated with greater levels of provincial welfare generosity and improved population health outcomes?”, and, “Which political parties are most likely to appoint women cabinet ministers, and do political

259 partisanship and female politicians interact with each other to affect population health in predictable ways?”

Racial politics of presence. This dissertation examines only the connection between politics of presence and population with respect to women politicians. The politics of presence can also be applied to understanding the health consequences of electing more visible minorities to political office. The basic argument is analogous female politicians - racial and ethnic representatives in political office are assumed to be better positioned to represent the needs and interests of visible minorities, immigrants, and other racialized groups (Bird, 2003, 2004, 2005; Phillips, 1995). This implies that as the number of racial and ethnic politicians increase, these politicians achieve not only symbolic goals but also affect political thinking and actions in substantive ways. For example, racial and ethnic representatives may be more likely than non-racial representatives to promote policies and legislation that address preventable and unfair racial inequalities in social, environmental, and economic conditions. The presence of racial and ethnic politicians in government might bring an under-represented voice to legislative processes that increase the chances that salient race issues are raised and addressed. Despite the plausible links between visible minorities in political power and population health, no research has been undertaken on the topic. Given this void in the extant literature, future research questions include, “Does the presence of racial and ethnic politicians in provincial legislatures operate as a leftist political force that improves population health?”, or, “Which political parties are most likely to have visible minorities elected to political office, and does political partisanship and racial and ethnic politicians interact with each other to affect population health in consistent ways?”, and,

260 “Are increases in visible minority cabinet ministers significantly and positively associated improved population health outcomes?”.

Augmenting Aggregate Measures of Population Health with Health Inequality Measures

This dissertation demonstrates the significant impact of political determinants in lowering total, male, and female age-standardized mortality rates, which are aggregate measures of population health. However, average rates of mortality are an insufficient indicator of health inequalities. By implication, a focus on population health also necessitates a critical understanding of how politics may affect inequalities in health status between population groups (e.g., health inequalities in income, education, occupation, social class, working conditions). The key idea here is that measuring average levels of health is qualitatively different from measuring health inequalities, and both of these approaches imply important trade-offs between policies that improve overall health and that reduce health inequalities (Gakidou, Murray, & Frenk, 2000). In order to advance our current understanding on the empirical connections between political factors and the distribution of population health, future politically-oriented studies should conceptualize health inequalities as dependent variables, or variations in health status across individuals in a population (Murray, Gakidou, & Frenk, 1999). For example, health inequality measures are available through Statistics Canada, including health-adjusted life expectancy, at birth and at age 65, by sex and income (CANSIM

Table 102-0122), and mortality, by selected causes of death, by income adequacy quintile (CANSIM Table 102-0601). Because these measures are not collected over time on an annual basis, methods other than TSCS will have to be employed. Possible

261 questions include, “What is the association between provincial welfare generosity, leftist politics, and mortality gradients by income?”, “Do political factors have a differential health impact on income quintiles?”, and, “Which income quintiles are most likely to benefit from egalitarian politics and welfare generosity?”

Explicating the Social Mechanisms that Connect Politics and Population Health

This dissertation uses quantitative comparative research methods (i.e., TSCS) to address its research questions. Essentially, I test several hypotheses using statistical estimation techniques on data aggregated at the level of provinces and provincial regimes. The epistemological and ontological backgrounds of my scientific approach are straightforward. I have applied a deductive-nomological model to theorize the empirical association between politics and health, and have assumed that a law-like relationship exists between the two variables. As a consequence, this dissertation finds robust statistical associations between leftist politics, welfare generosity, and population health. However, it is important to note that the regression coefficients that link independent and dependent variables only tell us about the direction and strength of the observed associations. Observed associations are discussed in terms of input and output, or between explanans and explanadum, with reference to existing political sociology and social epidemiology theories. The next intellectual challenge is to adequately address the question of, “How do macro-level politics affect population health, and through what social processes and mechanisms does politics influence health?”

Future research should augment this dissertation’s research methods with social

262 theory and methods that explicate the social mechanisms that generate and explain the association between politics and health (Hedström & Swedberg, 1998; Hedström &

Ylikoski, 2010). The idea of mechanism-based explanation has become increasingly influential in the social sciences (Elster, 1989; Tilly, 2001) and political sociology

(Bennet, 2005; Gerring, 2008; Mahoney, 2001). In oversimplified terms, mechanism- based explanations seek to: 1) identify the situational mechanisms by which social structures shape and constrain individual action, desires, and beliefs; 2) describe the action-formation mechanisms that link individual action to desires and beliefs; 3) specify the transformational mechanisms by which individuals, through their actions and interactions, generate various intended and unintended social outcomes; and 4) relate macro-level properties to each other (Coleman, 1990; Hedström & Swedberg, 1998). In several respects, this dissertation has sketched out macro-level relationships between political dynamics and population health (see figure 8.1 located end of chapter). In order to make comprehensive sense of these macro-level associations, the challenge is to understand the whole chain of situational, action-formation, and transformation mechanisms. To this end, future work should use qualitative methods such as explanatory case studies, realist reviews, or concept mapping to pose such questions as (Pawson & Tiller, 1997):

 “What kinds of specific social situations in each province structure the political

views and actions of leftist political actors?”;

 “What combinations of public opinion and actions among leftist political actors

generate a specific action that influences political institutions and population

health?”;

263  “How do leftist political actors interact with each another and forge alliances to

trigger social mechanisms that result in some kind of population health outcome,

be it intended or unintended?”;

 “What are the political contexts and social mechanisms underlying provincial

welfare generosity that bring about lower mortality rates? How do medical care,

other social services, and post-secondary education expenditures meet the

health, social, and educational needs of citizens to reduce mortality rates?”;

 “How do female politicians strengthen women’s interests in government?

How are these potential processes brought about and how do these processes

combine with provincial welfare generosity to affect population health?”;

 “How do left political parties advance their policy platforms in government? How

do these policy platforms combine with provincial welfare generosity to affect

population health?;

 “How do right political parties affect provincial governments in ways that increase

mortality rates? How do right political parties combine with provincial welfare

generosity to affect population health?”;

 “How and why are some provinces more egalitarian than others? What are the

contexts and social mechanisms that explain higher male mortality rates in

conservative provinces compared to leftist provinces?; and

 “Under what political contexts is provincial welfare generosity most likely to

improve population health?”.

264

Figure 8.1. Revised Conceptual Model: Power Constellations, Politics of Presence, Provincial Welfare Generosity, and Age-Standardized Mortality Rates

Union Density Left Political Party Power

Provincial Welfare Generosity Centre Political Age-Standardized Voter Turnout (medical care, other social services, Party Power Mortality Rates & post-secondary education)

Women in Government

265 CHAPTER NINE. CONCLUSION

This dissertation makes two new contributions to the extant literature. First, I show that political sociology theories can be effectively applied within a Canadian context to understand provincial variations in health. Guided by Rose’s ([1985] 2001) “population perspective”, this dissertation demonstrates the connections between leftist politics, provincial welfare generosity, and population health among Canadian provinces.

Second, I use research methods to increase the overall rigor of previous studies. By taking advantage of the strengths of TSCS data, this dissertation increases our confidence on the causal links between politics and health. In this concluding chapter, I briefly outline a set of political strategies and policies informed by this dissertation’s major findings. Before doing so, I review a current debate in the extant literature on the relative importance of political power and public policies in improving population health and reducing health inequalities (Espelt et al., 2008, 2010; Lundberg, 2008, 2010;

Muntaner et al., 2010). This debate’s major arguments provide a useful context to outline this dissertation’s political and policy implications.

Welfare State Policies versus Macro-Level Politics

In 2008, Espelt et al. published a study in the International Journal of Epidemiology that compared inequalities in self-rated health among individuals aged 50 and over, using

Wright’s social class scheme, and among nine European countries coded into three political traditions: Social democracies (Sweden, Denmark and Austria), Christian democracies (The Netherlands, Germany, France and Italy), and Late democracies

266 (Portugal and Spain). Using Poisson regression models with robust variance estimators, these authors find that absolute and relative health inequalities by social class dimensions exist across all three political traditions. However, these differences are more pronounced in Late democracies and among women. For example, the prevalence ratio of poor self-perceived health comparing poorly educated women with highly educated women, was 1.75 (95% CI: 1.39–2.21) in Late democracies and 1.36

(95% CI: 1.21–1.52) in Social democracies. This study sparked an interesting debate on whether efforts to improve health and reduce health inequalities should be devoted to welfare state policies and programs or macro-level politics or both.

In response to Espelt et al. (2008), Lundberg (2008, 2010) took issue with conceptualizing welfare state regimes and political parties as determinants of population health. Although Lundberg acknowledges that politics matter and that politics affect health through policies, he finds little value with studying politics per se. His critique is two-fold. First, Lundberg (2008, 2010) argues that viewing political parties as independent variables in population health research introduces serious misclassification problems. For example, political parties with a certain label (e.g., left-, centre-, or right- wing) cannot be assumed to pursue similar public policies that provide similar health- promoting resources for citizens across place and time. In fact, differences in political and historical contexts given their large scope and complexity are probably more important to population health than similarities among social democratic political parties, for example. Second, Lundberg (2008, 2010) contends that Esping-Anderson’s (1990) welfare regime typology and similar kinds of institutional clusters produce more descriptive information than explanatory mechanisms. As a result, country clusters

267 provide ‘black-box’ descriptions that tell us little about what specific aspects about welfare states are important to population health. More concretely, Lundberg (2008) finds Esping-Anderson’s (1990) typology one-dimensional given that it only focuses on general aspects of cash transfer systems and imprecise given that important cross- national differences may be averaged out within each cluster of countries. The conclusion is that clustering countries on the basis of their political traditions (e.g., political power of social democratic parties) runs the real risk of obscuring more than clarifying what improves health.

Rather than emphasizing the health effects of political power and political parties,

Lundberg (2008, 2010) favours an exclusive focus on how public policies and programs are associated with public health outcomes. Welfare states matter to population health and health inequalities only to the extent that they provide health-promoting resources either directly through services and transfers or indirectly through policies that affect people’s opportunities to generate resources in the market. Thus, research efforts focus on the coverage and generosity of cash transfer programs like unemployment insurance, sickness insurance, family support or pensions (Lundberg et al., 2008). In question form, “What kind of policies and programs work in improving population health and reducing health inequalities?”, “What kinds of welfare state institutions operate as mechanisms at a societal level to affect population health?”

Instead of focusing on the politics of population health, Lundberg (2008, 2010) argues that it is far more important and useful to evaluate the effectiveness and characteristics of welfare state policies and programs. In his words, “If there is no understanding of what works and what does not work, politicians who want to reform

268 society are in a way blindfolded” (p. 2) and “if researchers want to contribute to improved public health policies we should be able to tell what kind of policies that work or not, not what kind of parties’ people should vote into the government” (p. 2).

In response, Espelt et al. (2009) and Muntaner et al. (2010) counter that there is no a priori reason why the effects of politics on health should be confined to policies and programs. Existing studies have already confirmed the idea that political traditions and political parties, expressed in representative institutions, affect population health through direct and indirect mechanisms. For example, political parties committed to redistributional policies are more successful in improving the health and welfare indicators of the majority of the population than are political traditions not sensitive to such redistributional policies (Navarro et al., 2003, 2006). Political traditions are not only associated with cash transfer systems, as Lundberg had suggested, but also with health and social services (Espelt et al., 2009; Muntaner, Lynch, & Oates, 1999). Furthermore, associations between political variables such as number of ‘left cabinet members’ and

’type of political party in government’ remain significant even after adjusting for welfare state policies and programs (Borrell et al., 2009; Chung & Muntaner, 2006).

Given these contributions, Espelt et al. (2008) and Muntaner et al. (2010) argue that ignoring political forces, processes, and institutions with respect to population health is akin to treating the symptoms of a disease than the causes of the ailment.

Treating politics as a peripheral matter has the effect of transforming democracy into a technocracy (e.g., democratic governments being led primarily by technical experts).

Without politically informed frameworks, for example, researchers would lack the analytical tools to adequately explain the historical declines in life expectancy in Russia

269 during the early 1990s (Stuckler, King, & McKee, 2009), the unexpected improvements in social equality and population health in Kerala, India during the 1970s (Ratcliffe,

1978), or the unprecedented public health gains in Venezuela over the past two decades (Muntaner, Salazar, Benach, & Armada, 2006).

The key message is that welfare state policies and programs are not randomly developed or implemented within an apolitical vacuum in democratic societies

(Muntaner & Chung, 2008). To the contrary, public policies and programs reflect the consequences of political patterns and struggles. If political and economic arrangements are viewed as root (distal) causes and public policies and programs as immediate (proximate) causes, we gain a deeper and more complex understanding of how leftist politics might influence social inequalities (income inequality, poverty), welfare services (access to health care and education) and social transfers

(unemployment and family benefits), which in turn, affect population health and health inequalities. Comparing countries in clusters or regimes based on their respective political traditions, welfare state types, or market economies is more than a descriptive exercise. According to Muntaner et al. (2010), cluster or regime approaches augment the study of specific national policies by producing novel insights on why some countries are more effective than others in reducing inequalities and why certain egalitarian policies tend to cluster together in certain societies and not in others.

Another key advantage of comparative approaches that consider politics and policies in unison is that researchers can identify the origins of intersectoral policies that improve health and reduce health inequalities across multiple sectors of national governments. Because egalitarian policies probably require ‘synergic’ effects across

270 multiple sectors of government (e.g., health, labour market, education, environment, social services), such effects are more likely to be identified in certain political traditions opposed to certain policy environments.

Political parties and traditions are only one aspect on how political forces, processes, and institutions may affect population health. The wide-ranging health effects of politics includes other social processes such as grassroots organizing, social movements, wars, strikes, protests and non-government organizations (Muntaner,

Lunch, & Oates, 1999). With this said, it is important to note that Espelt et al. (2008) and

Muntaner et al. (2010) do not claim that politics is a panacea to all matters related to population health and health inequalities. Politics only serves as a necessary starting point for public health, should inform parts of the solution, however, it does not serve as the only solution.

In this dissertation, I show that welfare state policies and programs and macro- level politics are more complementary than different, and that both approaches to population health can be considered within the same program of research. Next, I outline political strategies that augment Espelt et al.’s (2008) and Muntaner et al.’s

(2010) emphasis on politics and I propose policy implications that that support

Lundberg’s (2008, 2010) substantive foci. In this respect, my dissertation provides an important bridge between these two opposing arguments to produce a fuller understanding on how to improve population health. In other words, this dissertation speaks to what kinds of leftist politics should receive greater support and which public policies to sustain and strengthen.

271 Political Strategies

Mobilize support for women in government. Increasing the number of women in provincial governments enhances gender equality in politics, advances the interests of women, and improves population levels of health. In conjunction with organizations such as Equal Voice, a national, multi-partisan group committed to electing more women to all levels of political office in Canada, efforts toward gender equality in politics include promoting electoral changes that increase the numbers of women in politics

(e.g., quotas), conducting outreach with young women to engage them in politics as a future career choice, and raising awareness about the consequences of women's under-representation.

Mobilize support for left and centre political parties. Increasing the political power of left and centre parties shifts the ideological spectrum toward centre-left, promotes a more egalitarian distribution of resources, and improves population health. Although this strategy may appear to be value-laden and ‘too political’, the idea that leftist collective political actors are more committed and more successful to achieving egalitarian outcomes is accepted as conventional wisdom in political sociology. Supporting leftist political forces embodies political relations that determine provincial welfare generosity and affect population health through indirect and direct pathways.

272 Policy Proposals

Augmenting provincial welfare generosity on medical care expenditures.

Increasing provincial expenditures on welfare services in the form of medical care has a positive impact on population health. When high-quality medical care is available to citizens, mortality rates tend to decline. Medical care is expansive and includes expenses on general medical care, drug programs, dental and visiting-nurse services, and on out-patient care services. It also includes expenditures for medical care provided by hospitals, public residential care facilities, Workers’ Compensation Boards, and other public health and social service institutions. This stresses the importance of strengthening current medicare systems and limiting the involvement of for-profit companies in the organization and delivery of medical care services.

Augmenting provincial welfare generosity on other social services expenditures.

Increasing expenditures on social services that assist those in need are associated with significant reductions in mortality rates. The target groups of such spending includes the elderly, persons with physical or mental disabilities, persons temporarily unable to work due to sickness, households with dependent children, and survivors of a deceased person. Since provinces play an essential role in protecting people who find themselves in vulnerable positions, social services such as home care, transportation, counseling, nursery and daycare, essential goods, and rehabilitation for alcohol and drug problems should remain a high priority on provincial budgets. Continued investments in these key services are a solid strategy to strengthen social safety nets and to improve population levels of health.

Augmenting provincial welfare generosity on post-secondary education

273 expenditures. Increasing expenditures on the types of educational training obtained in universities, community colleges, and specialized educational institutions are associated with improving levels of population health. Also included under this expenditure are bursaries, scholarships, and other types of financial assistance (e.g., loan forgiveness, interest relief) that encourage students to pursue post-secondary credentials. A related policy proposal that has the potential to also improve population health requires provincial governments to ensure that the costs associated with attending university and college are not prohibitive. It is within provincial powers to ensure tuition fees are affordable so that students from socioeconomically disadvantaged families are not excluded from higher education.

These political strategies and policy proposals are far from radical in nature.

These initiatives and interventions are stronger in some provinces than others and have a proven track record in other capitalist democracies. For example, Sweden boasts one of the healthiest populations in the world and has the highest number of female politicians in national government. After Sweden’s 2010 national election, female politicians comprised 45% of elected representatives in the Swedish parliament (The

Swedish Institute, 2013). Countries with strong left and centre political parties include social democratic (e.g., Scandinavian nations) and conservative regimes (e.g.,

Germany, Switzerland, Austria, and France), respectively. These regimes are characterized as being more egalitarian and healthier than liberal nations, which tend to have strong right political parties (e.g., US, UK, Canada) (Bambra, 2005). Moreover, the health advantages among Nordic countries might reflect in part their commitment to providing universal post-secondary education. Public expenditures in education are

274 significantly higher in Denmark, Sweden, and Norway in comparison to the OECD average (OECD, 2008).

The overall message of this dissertation is clear. Political forces, processes, and institutions are significant predictors of population health. The significance of this message is encouraging given that politics are actionable within the existing political system. Strengthening leftist politics in Canada is analogous to supporting social movements, organizations, and institutions that are committed to achieving social, economic, and gender equality. From this perspective, the challenge of improving population health and reducing health inequalities reflects ongoing political struggles between men and women, owners and workers, the poor and affluent, Aboriginals and non-Aboriginals, and racialized groups and non-racialized groups (Hofrichter, 2003).

Admittedly, these distinctions between more powerful and less powerful groups are overly simplistic, however, reformative social movements that increase the political power of women (e.g., quotas to elect female politicians), workers (e.g., collective bargaining rights), the poor (e.g., generous social transfers), Aboriginals (e.g., greater self-determination and control over lands and resources), and racialized groups (e.g., employment equity and affirmative action policies) have the real potential to achieve a more egalitarian and healthier society.

275 REFERENCES1

Aidt, T., & Tzannatos, Z. (2002). Unions and collective bargaining. Washington, DC:

World Bank.

Akyeampong, E. B. (2000). Non-unionized but covered by collective agreement.

Perspectives on Labour and Income, 12(3), 33-59.

Akyeampong, E. B. (2006). Increased work stoppages. Ottawa, ON: Statistics Canada.

Albo, G. (2002). Neoliberalism, the state, and the left: A Canadian perspective. Monthly

Review - New York, 54(1), 46-55.

Alesina, A. (1995). Partisan politics, divided government, and the economy. Cambridge,

UK: Cambridge University Press.

Allan, J. P., & Scruggs, L. (2004). Political partisanship and welfare state reform in

advanced industrial societies. American Journal of Political Science, 48(3), 496-

512.

Amenta, E. (2000). Bold relief: Institutional politics and the origins of modern American

social policy. Princeton, NJ: Princeton University Press.

Auger, N., & Alix, C. (2009). Income, income distribution, and health in Canada. In D.

Raphael (Ed.), Social determinants of health: Canadian perspectives (pp. 61-74)

(2nd ed.). Toronto, ON: Canadian Scholars’ Press.

Auger, N., Hamel, D., Martinez, J., & Ross, N. A. (2012). Mitigating effect of immigration

on the relation between income inequality and mortality: A prospective study of 2

million Canadians. Journal of Epidemiology and Community Health, 66(6), e5-e5.

1 References marked with an asterisk indicate studies included in the systematic review conducted in chapter 2.

276 Auger, N., Zang, G., & Daniel, M. (2009). Community-level income inequality and

mortality in Quebec, Canada. Public Health, 123(6), 438-443.

*Arah, O. A., & Westert, G. P. (2005). Correlates of health and healthcare performance:

Applying the Canadian health indicators framework at the provincial-territorial

level. BMC Health Services Research, 5(1), 76.

*Arah, O. A., Westert, G. P., Delnoij, D. M., & Klazinga, N. S. (2005). Health system

outcomes and determinants amenable to public health in industrialized countries:

A pooled, cross-sectional time series analysis. BMC Public Health, 5(1), 81.

Ariizumi, H., & Schirle, T. (2012). Are recessions really good for your health? Evidence

from Canada. Social Science & Medicine, 74(8), 1224-1231.

Avendano, M., Juerges, H., & Mackenbach, J.P. (2009). Educational level and

changes in health across Europe: Longitudinal results from SHARE. Journal of

European Social Policy, 19(4), 301-316.

Bach, S. (2003). International migration of health workers: Labour and social

issues. Geneva, CH: International Labour Office.

Baker, M., Gruber, J., & Milligan, K. (2005). Universal childcare, maternal labor supply,

and family well-being (No. w11832). Cambridge, MA: National Bureau of

Economic Research.

*Bambra, C. (2005). Cash versus services: 'worlds of welfare' and the

decommodification of cash benefits and health care services. Journal of Social

Policy, 34(2), 195−213.

Bambra, C. (2006). Health status and the worlds of welfare. Social Policy and Society,

5(1), 53−62.

277 Bambra, C. (2007a). Defamilisation and welfare state regimes: A cluster analysis.

International Journal of Social Welfare, 16(4), 326-338.

Bambra, C. (2007b). Going beyond the three worlds of welfare capitalism: Regime

theory and public health research. Journal of Epidemiology and Community

Health, 61(12), 1098-1102.

Bambra, C. (2011). Health inequalities and welfare state regimes: Theoretical insights

on a public health ‘puzzle’. Journal of Epidemiology and Community

Health, 65(9), 740-745.

*Bambra, C., & Eikemo, T. A. (2009). Welfare state regimes, unemployment and health:

A comparative study of the relationship between unemployment and self-reported

health in 23 European countries. Journal of Epidemiology and Community

Health, 63(2), 92-98.

Bambra, C., Fox, D., & Scott-Samuel, A. (2005). Towards a politics of health. Health

Promotion International, 20(2), 187-193.

Bambra, C., Fox, D., & Scott-Samuel, A. (2007). A politics of health glossary. Journal of

Epidemiology and Community Health, 61(7), 571-574.

Bambra, C., Netuveli, G., & Eikemo, T. A. (2010). Welfare state regime life courses:

The development of western European welfare state regimes and age-related

patterns of educational inequalities in self-reported health. International Journal

of Health Services, 40(3), 399-420.

Bambra, C., Pope, D., Swami, V., Stanistreet, D., Roskam, A., Kunst, A., & Scott-

278 Samuel, A. (2009). Gender, health inequalities and welfare state regimes: A

cross-national study of 13 European countries. Journal of Epidemiology and

Community Health, 63(1), 38-44.

Beck, N. (2001). Time-series-cross-section data: What have we learned in the past few

years?. Annual Review of Political Science, 4(1), 271-293.

Beck, N., & Katz, J. (1995). What to do (and not to do) with time series cross-section

data. American Political Science Review, 89(3), 634–647.

Beck, N., & Katz, J. (1996). Nuisance versus substance: Specifying and estimating

time-series cross-section models. Political Analysis, 6(1), 1–36.

Beck, N., & Katz, J. (2001). Random coefficients models for time-series cross-section

data. Unpublished manuscript. San Diego, CA: University of California.

Becker, G. S. (2009). Human capital: A theoretical and empirical analysis, with special

reference to education. Chicago, IL: University of Chicago Press.

Beckfield, J., & Krieger, N. (2009). Epi+ demos+ cracy: Linking political systems and

priorities to the magnitude of health inequities—evidence, gaps, and a research

agenda. Epidemiologic Reviews, 31(1), 152-177.

Bennet, A. A. (2005). Case studies and theory development in the social sciences.

Cambridge, MA: MIT Press.

Berkman, L.F., & Kawachi, I. (2000). Social epidemiology. New York, NY: Oxford

University Press.

Berkman, L. F., & Syme, S. L. (1979). Social networks, host resistance, and mortality: A

nine-year follow-up study of Alameda County residents. American Journal of

Epidemiology, 109(2), 186-204.

279 Bernard, P., & Saint-Arnaud, S. (2004). More of the same? The position of the four

largest Canadian provinces in the world of welfare regimes. Research report

FI49 family network. Ottawa, ON: Canadian Policy Research Networks Inc.

Blais, A. (2005). Accounting for the electoral success of the Liberal Party in Canada:

Presidential address to the Canadian Political Science Association. Canadian

Journal of Political Science, 38(4), 821-840.

Blais, A., Blake, D., & Dion, S. (1993). Do parties make a difference? Parties and the

size of government in liberal democracies. American Journal of Political Science,

37(1), 40-62.

*Blakely, T. A., Kennedy, B. P., & Kawachi, I. (2001). Socioeconomic inequality in voting

participation and self-rated health. American Journal of Public Health, 91(1), 99-

104.

Bird, K. (2003). The political representation of women and ethnic minorities in

established democracies: A framework for comparative research. Academy for

Migration Studies in Denmark, Aalborg, DK: Aalborg University.

Bird, K. (2004). Comparing the political representation of ethnic minorities in advanced

democracies. Winnipeg, MB: Annual Meeting of the Canadian Political Science

Association.

Bird, K. (2005). The political representation of visible minorities in electoral

democracies: a comparison of France, Denmark, and Canada. Nationalism and

Ethnic Politics, 11(4), 425-465.

Bird, R.M., & Slack, N.E. (1993). Urban public finance in Canada (2nd ed.). Toronto,

ON: Wiley.

280 Bjørnskov, C., & Potrafke, N. (2011). Political ideology and economic freedom across

Canadian provinces. Eastern Economic Journal, 38(2), 143-166.

Bobbio, N. (1996). Left and right: The significance of a political distinction. Chicago, IL:

University of Chicago Press.

Borrell, C., Espelt, A., Rodríguez-Sanz, M., & Navarro, V. (2007). Politics and health.

Journal of Epidemiology and Community Health, 61(8), 658-659.

Borrell, C., Espelt, A., Rodríguez-Sanz, M., Burström, B., Muntaner, C., Pasarín, M. I.,

... & Navarro, V. (2009). Analyzing differences in the magnitude of

socioeconomic inequalities in self-perceived health by countries of different

political tradition in Europe. International Journal of Health Services, 39(2), 321-

341.

Boychuk, G. W. (1998). Patchworks of purpose: The development of provincial social

assistance regimes in Canada. Kingston, ON and Montreal, QC: McGill-Queen's

University Press.

*Bradley, E. H., Elkins, B. R., Herrin, J., & Elbel, B. (2011). Health and social services

expenditures: Associations with health outcomes. BMJ Quality & Safety, 20(10),

826-831.

Bradley, D., Huber, E., Moller, S., Nielsen, F., & Stephens, J. D. (2003). Distribution and

redistribution in postindustrial democracies. World Politics, 55(2), 193-228.

Brady, D. (2003). The politics of poverty: Left political institutions, the welfare state, and

poverty. Social Forces, 82(2), 557-588.

Brady, D. (2005). The welfare state and relative poverty in rich western democracies,

1967–1997. Social Forces, 83(4), 1329-1364.

281 Brady, D. (2009). Rich democracies, poor people: How politics explain poverty. Oxford,

UK: Oxford University Press.

Brennenstuhl, S., Quesnel-Vallée, A., & McDonough, P. (2012). Welfare regimes,

population health and health inequalities: A research synthesis. Journal of

Epidemiology and Community Health, 66(5), 397-409.

Brooks, C., & Manza, J. (2008). Why welfare states persist: The importance of public

opinion in democracies. Chicago, IL: University of Chicago Press.

Brunner, E., & Marmot, M. G. (1999). Social organization, stress, and health. In M.G.

Marmot & R.G. Wilkinson (Eds.), Social determinants of health (17-43).

Oxford, UK: Oxford University Press.

Bryant, T. (2006). Politics, public policy and population health. In D. Raphael, T. Bryant,

& M. Rioux (Eds.), Staying alive: Critical perspectives on health, illness, and

health care (239-266). Toronto, ON: Canadian Scholars’ Press.

Bryant, T. (2009). An introduction to health policy. Toronto, ON: Canadian Scholars’

Press.

Brym, R. J., Gillespie, M. W., & Lenton, R. L. (1989). Class power, class mobilization,

and class voting: The Canadian case. Canadian Journal of Sociology, 14(1), 25-

44.

Bunker, J. P. (2001). The role of medical care in contributing to health improvements

within societies. International Journal of Epidemiology, 30(6), 1260-1263.

Burris, V. (2001). The two faces of capital: Corporations and individual capitalists as

political actors. American Sociological Review, 66(3), 361-381.

Burstrom, B., Whitehead, M., Clayton, S., Fritzell, S., Vannoni, F., & Costa, G. (2010).

282 Health inequalities between lone and couple mothers and policy under different

welfare regimes − the example of Italy, Sweden and Britain. Social Science &

Medicine, 70(6), 912−920.

Cairns, A.C. (1977). The governments and societies of Canadian federalism. Canadian

Journal of Political Science, 10(4), 695-734.

Canadian Institute for Health Information. (2004). Improving the health of Canadians:

Patterns of health and disease are largely a consequence of how we learn, live

and work: summary report. Ottawa, ON: Canadians Institute for Health

Information.

Canadian Institute for Health Information. (2011). National health expenditure trends,

1975 to 2011. Ottawa, ON: Canadians Institute for Health Information.

Castles, F. G. (1982). The impact of parties: Politics and policies in democratic capitalist

states. London, UK; Beverly Hills, CA: Sage Publications.

Castles, F. G. (1989). Explaining public education expenditure in OECD nations.

European Journal of Political Research, 17(4), 431-448.

Castles, F. G., & Mair, P. (1984). Left–right political scales: Some ‘expert’ judgments.

European Journal of Political Research, 12(1), 73-88.

Castles, F. G., & Mitchell, D. (1992). Identifying welfare state regimes: The links

between politics, instruments and outcomes. Governance, 5(1), 1-26.

CBC. (2013). Women in Canadian politics. Canadian Broadcasting Corporation.

Retrieved from http://www.cbc.ca/news/interactives/map-cda-womenpolitics/

Chen, J., Ng, E., & Wilkins, R. (1996). The health of Canada’s immigrants in 1994

283 -95. Health Reports, 7(4), 33-45.

*Chung, H., & Muntaner, C. (2006). Political and welfare state determinants of infant

and child health indicators: An analysis of wealthy countries. Social Science &

Medicine, 63(3), 829-842.

*Chung, H., & Muntaner, C. (2007). Welfare state matters: A typological multilevel

analysis of wealthy countries. Health Policy, 80(2), 328-339.

Chung, H., & Muntaner, C. (2008). Welfare regime types and global health: an emerging

challenge. Journal of Epidemiology and Community Health, 62(4), 282-283.

Chung, H., Muntaner, C., & Benach, J. (2010). Employment relations and global health:

A typological study of world labor markets. International Journal of Health

Services, 40(2), 229-253.

*Chung, H., Ng, E., Ibrahim, S., Karlsson, B., Benach, J., Espelt, A., & Muntaner, C.

(2013). Welfare state regimes, gender, and depression: A multilevel analysis of

middle and high income countries. International Journal of Environmental

Research and Public Health, 10(4), 1324-1341.

Coburn, D. (2004). Beyond the income inequality hypothesis: Class, neo-liberalism, and

health inequalities. Social Science & Medicine, 58(1), 41-56.

Coburn, D. (2006). Health and health care: A political economy perspective. In D.

Raphael, T. Bryant, & M. Rioux (Eds.), Staying alive: Critical perspectives on

health, illness, and health care (65-92). Toronto, ON: Canadian Scholars’

Press.

Coleman, J.S. 1990. Foundations of social theory. Cambridge, MA: Belknap.

284 CSDH. (2008). Closing the gap in a generation: Health equity through action on the

social determinants of health. Final Report of the Commission on Social

Determinants of Health. Geneva, CH: World Health Organization.

*Conley, D., & Springer, K. W. (2001). Welfare state and infant mortality. American

Journal of Sociology, 107(3), 768-807.

Cusack, T. R. (1997). Partisan politics and public finance: Changes in public spending

in the industrialized democracies, 1955–1989. Public Choice, 91(3-4), 375-395.

Cramer, D. (2003). Advanced quantitative data analysis. Maidenhead, UK: McGraw-Hill.

Cross, W. (2011). Political parties. Vancouver, BC: University of British Columbia Press.

Dahl, E., Fritzell, J., Lahelma, E., Martikainen, P., Kunst, A. & Mackenbach, J.P. (2006).

Welfare state regimes and health inequalities. In J. Siegrist & M. Marmot (Eds.),

Social inequalities in health: New evidence and policy implications (193-222).

Oxford, UK: Oxford University Press.

*Dahl, E., & van der Wel, K. A. (2013). Educational inequalities in health in European

welfare states: A social expenditure approach. Social Science & Medicine, 81,

60-69. doi: 10.1016/j.socscimed.2012.12.010

Daku, M., & Heymann, J. (2010). Canadian labour policies: How do we measure up?.

Montreal, QC. Annual Meeting of the Canadian Political Science Association.

Di Matteo, L., & Di Matteo, R. (1998). Evidence on the determinants of Canadian

provincial government health expenditures: 1965–1991. Journal of Health

Economics, 17(2), 211-228.

Dodson, D.L., & Carroll, S.J. (1991). Reshaping the agenda: Women in state

legislatures. New Brunswick, NJ: Center for the American Woman and Politics.

285 Domhoff, G. W. (1998). Who rules America?: Power and politics in the year 2000.

Mountain View, CA: Mayfield Publishing Company.

Döring, H. (1995). Time as a scarce resource: Government control of the agenda. In H.

Döring (Ed.), Parliaments and majority rule in Western Europe (223-246). New

York, NY: St. Martin’s Press.

Dorling, D., Dunn, J. R., Henriksson, G., Glover, J., Lynch, J., & Weitoft, G. R. (2005).

Metropolitan income inequality and working-age mortality: A cross-sectional

analysis using comparable data from five countries. Journal of Urban

Health, 82(1), 101-110.

Doyal, L. (1979). The political economy of health. London, UK: Pluto Press.

Doyle, Y. G., Furey, A., & Flowers, J. (2006). Sick individuals and sick populations: 20

years later. Journal of Epidemiology and Community Health, 60(5), 396-398.

Dragano, N., Siegrist, J., & Wahrendorf, M. (2011). Welfare regimes, labour policies

and unhealthy psychosocial working conditions: A comparative study with 9917

older employees from 12 European countries. Journal of Epidemiology and

Community Health, 65(9), 793-799.

*Dunn, J. R., Burgess, B., & Ross, N. A. (2005). Income distribution, public services

expenditures, and all-cause mortality in US States. Journal of Epidemiology and

Community Health, 59(9), 768-774.

Dunn, J.R., Frohlich, K., Ross, N.A., Curtis, L., & Sanmartin, C. (2005). Role of

geography in inequalities in health and human development. In J. Heymann, C.

Hertzman, M. Barer, & R.G. Evans (Eds.), Healthier societies: From analysis to

action (237-266). New York, NY: Oxford University Press.

286 Dyck, R. (2000). Canadian politics: Critical approaches (3rd ed.). Scarborough,

ON: Nelson Thompson Learning.

Dye, C. (2008). Health and urban living. Science, 319(5864), 766-769.

Dye, T. 1979. Politics versus economics: The development of the literature on policy

determination. Policy Studies Journal, 7(4), 652-662.

Efron, B., & Efron, B. (1982). The jackknife, the bootstrap and other resampling plans

(Vol. 38). Philadelphia, PA: Society for Industrial and Applied Mathematics.

Eikemo, T. A., & Bambra, C. (2008). The welfare state: A glossary for public health.

Journal of Epidemiology and Community Health, 62(1), 3-6.

*Eikemo, T.A., Bambra, C., Judge, K., & Ringdal, K. (2008a). Welfare state regimes

and differences in self-perceived health in Europe: A multilevel analysis. Social

Science & Medicine, 66(11), 2281-2295.

*Eikemo, T.A., Huisman, M., Bambra, C., & Kunst, A.E. (2008b). Health inequalities

according to educational level in different welfare regimes: A comparison of 23

European countries. Sociology of Health and Illness, 30(4), 565-582.

Elections Alberta. (n.d.). Retrieved from http://www.elections.ab.ca

Elections British Columbia. (n.d.). Retrieved from http://www.elections.bc.ca/

Elections Manitoba. (n.d.). Retrieved from http://www.electionsmanitoba.ca

Elections New Brunswick. (n.d.). Retrieved from http://www.electionsnb.ca

Elections Newfoundland and Labrador. (n.d.). Retrieved from

http://www.elections.gov.nl.ca

Elections Prince Edward Island. (n.d.). Retrieved from http://www.electionspei.ca/

287 Elections Nova Scotia. (n.d.). Retrieved from http://electionsnovascotia.ca/

Elections Ontario. (n.d.). Retrieved from http://www.elections.on.ca

Elections Prince Edward Island. (n.d.). Retrieved from http://www.electionspei.ca

Elections Quebec. (n.d.). Retrieved from http://www.electionsquebec.qc.ca/english/

Elections Ontario. (n.d.). Retrieved from http://www.elections.on.ca/en-ca

Elections Manitoba. (n.d.). Retrieved from http://www.electionsmanitoba.ca/en

Elections Saskatchewan. (n.d.). Retrieved from http://www.elections.sk.ca/

Elections Alberta. (n.d.). Retrieved from http://www.elections.ab.ca/Public%20Website/index.htm

*Elgar, F. J. (2010). Income inequality, trust, and population health in 33 countries.

American Journal of Public Health, 100(11), 2311-2315.

Elola, J., Daponte, A., & Navarro, V. (1995). Health indicators and the organization of

health care systems in Western Europe. American Journal of Public Health,

85(10), 1397-1401.

Elster, J. (1989). Nuts and bolts for the social sciences. Cambridge, UK: Cambridge

University Press.

288 Espelt, A., Borrell, C., Rodríguez-Sanz, M., Muntaner, C., Pasarín, M. I., Benach, J., ...

& Navarro, V. (2008). Inequalities in health by social class dimensions in

European countries of different political traditions. International Journal of

Epidemiology, 37(5), 1095-1105.

Espelt, A., Borrell, C., Rodríguez-Sanz, M., Muntaner, C., Pasarín, M. I., Benach, J., ...

& Navarro, V. (2010). Answer to the commentary: Politics and public health

—some conceptual considerations concerning welfare state characteristics and

public health outcomes. International Journal of Epidemiology, 39(2), 630-632.

Esping-Andersen, G. (1990). The three worlds of welfare capitalism (Vol. 6).

Cambridge, UK: Polity press.

Esping-Andersen, G. (1999). Social foundations of postindustrial economies. Oxford,

UK: Oxford University Press.

Esping-Andersen, G. (1985). Politics against markets: The social democratic road to

power. Princeton, NJ: Princeton University Press.

Evans, R. G., Hertzman, C., & Morgan, S. (2007). Improving health outcomes in

Canada. In J. Leonard, C. Ragain, & F. St-Hilaire (Eds.). A Canadian priorities

agenda: policy choices to improve economic and social well-being (291-325).

Kingston, ON and Montreal, QC: McGill-Queen's University Press.

Evans, R.G. (2002). Interpreting and addressing inequalities in health: From Black to

Acheson to Blair to. London, UK: Office of Health Economics.

Evans, R. G., Barer, M. L., & Marmor, T. R. (Eds.). (1994). Why are some people

healthy and others not?. The determinants of the health of populations.

Piscataway, NJ: Transaction Publishers.

289 Evans, R., G., Hertzman, C., & Morgan, S. (2007). Improving health outcomes in

Canada. In J. Leonard, C. Ragen, & F. St-Hilaire (Eds.), A Canadian priorities

agenda: Policy choices to improve economic and social wellbeing (291-325).

Ottawa, ON: Institute for Research on Public Policy.

Everitt, B. S., Landau, S., & Leese, M. (2001). Clustering analysis. London, UK: Arnold.

*Farfan-Portet, M. I., Popham, F., Mitchell, R., Swine, C., & Lorant, V. (2010). Caring,

employment and health among adults of working age: Evidence from Britain and

Belgium. The European Journal of Public Health, 20(1), 52-57.

*Fayissa, B. (2001). The determinants of infant and child mortality in developing

countries: The case of Sub-Sahara Africa. The Review of Black Political

Economy, 29(2), 83-98.

*Ferrarini, T., & Norström, T. (2010). Family policy, economic development and infant

mortality: A longitudinal comparative analysis. International Journal of Social

Welfare, 19(s1), S89-S102.

Ferrera, M. (1996). The 'Southern model' of welfare in social Europe. Journal of

European Social Policy, 6(1), 17-37.

*Filmer, D., & Pritchett, L. (1999). The impact of public spending on health: Does money

matter?. Social Science & Medicine, 49(10), 1309-1323.

Filmer, D., Hammer, J. S., & Pritchett, L. (1998). Health policy in poor countries: Weak

links in chain (No. 1874). Washington, DC: World Bank Publications.

Franco, Á., Álvarez-Dardet, C., & Ruiz, M. T. (2004). Effect of democracy on health:

Ecological study. BMJ: British Medical Journal, 329(7480), 1421-1423.

290 Frie, K. G., Eikemo, T. A., & von dem Knesebeck, O. (2010). Education and self-

reported health care seeking behaviour in European welfare regimes: Results

from the European Social Survey. International Journal of Public Health, 55(3),

217-220.

Fritzell, J., Kangas, O., Bacchus Hertzman, J., Blomgren, J., & Hiilamo, H. (2013).

Cross-temporal and cross-national poverty and mortality rates among developed

countries. Journal of Environmental and Public Health. doi: 10.1155/2013/915490

Fritzell, J., & Lundberg, O. (eds.) (2007). Health inequalities and welfare resources:

Continuity and change in Sweden. Bristol, UK: Policy Press.

Gakidou, E. E., Murray, C. J., & Frenk, J. (2000). Defining and measuring health

inequality: An approach based on the distribution of health expectancy. Bulletin

of the World Health Organization, 78(1), 42-54.

Galobardes, B., Shaw, M., Lawlor, D. A., Lynch, J. W., & Smith, G. D. (2006). Indicators

of socioeconomic position (part 1). Journal of Epidemiology and Community

Health, 60(1), 7-12.

Gangl, M. (2006). Scar effects of unemployment: An assessment of institutional

complementarities. American Sociological Review, 71(6), 986-1013.

Garrett, G. (1998). Partisan politics in the global economy. Cambridge, UK: Cambridge

University Press.

Garrett, G., & Lange, P. (1989). Government partisanship and economic performance–

when and how does who governs matter. Journal of Politics, 51(3), 676-693.

Garrett, G., & Mitchell, D. (2001). Globalization, government spending and taxation in

the OECD. European Journal of Political Research, 39(2), 145-177.

291 Gattinger, M., Saint-Pierre, D., & Gagnon, A. C. (2008). Toward subnational

comparative cultural policy analysis: The case of provincial cultural policy and

administration in Canada. The Journal of Arts Management, Law, and Society,

38(3), 167-186.

Gehlert, S., Sohmer, D., Sacks, T., Mininger, C., McClintock, M., & Olopade, O. (2008).

Targeting health disparities: A model linking upstream determinants to

downstream interventions. Health Affairs, 27(2), 339-349.

Gerring, J. (2008). The mechanismic worldview: Thinking inside the box. British Journal

of Political Science, 38(01), 161-179.

Giddens, A., Ociepka, F., & Zujewicz, W. (1973). The class structure of the advanced

societies. London, UK: Hutchinson.

Gil-González, D., Ruiz-Cantero, M. T., & Álvarez-Dardet, C. (2009). How political

epidemiology research can address why the millennium development goals have

not been achieved: Developing a research agenda. Journal of Epidemiology and

Community Health, 63(4), 278-280.

Goldthorpe, J. H. (1997). Current issues in comparative macrosociology: A debate on

methodological issues. Comparative Social Research, 16, 1-26.

Goodin, R. E. (Ed.). (1999). The real worlds of welfare capitalism. Cambridge, UK:

Cambridge University Press.

Gordon, A. D. (1999a). Classification (2nd ed.). Boca Raton, FL: Chapman & Hall/CRC.

Gordon, D. (Ed.). (1999b). Inequalities in health: The evidence; the evidence presented

to the independent inquiry into Inequalities in health, chaired by Sir Donald

Acheson. Bristol, UK: The Policy Press.

292 Gordon, D., & Townsend, P. (Eds.). (2000). Breadline Europe: The measurement of

poverty. Bristol, UK: The Policy Press.

Gough, I. (2001). Social assistance regimes: A cluster analysis. Journal of European

Social Policy, 11(2), 165–170.

Gould, W. W. (1995). Jackknife estimation. Stata Technical Bulletin, 24, 25–29.

Granados, J. A. T. (2010). Politics and health in eight European countries: A

comparative study of mortality decline under social democracies and right-wing

governments. Social Science & Medicine, 71(5), 841-850.

Haggard, S., & Kaufman, R. R. (2008). Development, democracy and welfare states:

Latin America, East Asia, and Eastern Europe. Princeton, NJ: Princeton

University Press.

Hague, R., & Harrop, M. (2010). Comparative government and politics. An introduction.

(8th ed.). Basingstoke, UK: Palgrave Macmillan.

Hall, P. A., & Soskice, D. W. (Eds.). (2001a). Varieties of capitalism: The institutional

foundations of comparative advantage. Oxford, UK: Oxford University Press.

Hall, P.A, & Soskice, D. (2001b). An introduction to varieties of capitalism. In P.A. Hall,

& D, Soskice (Eds.), Varieties of capitalism: The institutional foundations of

comparative advantage (1-68). Oxford, UK: Oxford University Press.

Hall, P, A., & Taylor, R.C.R. (2009). Health, social relations and public policy. In P.A.

Hall, & M. Lamont (Eds.), Successful societies: How institutions and culture affect

health (82-103). Cambridge, UK: Cambridge University Press.

293 Hallstrom, L. (2009). Public policy and the welfare state. In D. Raphael (Ed.), Social

determinants of health: Canadian perspectives (2nd ed.) (336-349). Toronto, ON:

Canadian Scholars’ Press.

Hawley, A. H. (1992). The logic of macrosociology. Annual Review of Sociology, 18(1),

1-15.

Headey, B., Goodin, R. E., Muffels, R., & Dirven, H. J. (1997). Welfare over time: Three

worlds of welfare capitalism in panel perspective. Journal of Public Policy,

17(3), 329-359.

Heclo, H. (1974). Modern social politics in Britain and Sweden: From relief to income

maintenance . New Haven, CT: Yale University Press.

Hedström, P., & Swedberg, R. (Eds.). (1998). Social mechanisms: An analytical

approach to social theory. Cambridge, UK: Cambridge University Press.

Hedström, P., & Ylikoski, P. (2010). Causal mechanisms in the social sciences. Annual

Review of Sociology, 36, 49-67.

Hewitt, C. (1977). The effect of political democracy and social democracy on equality in

industrial societies: A cross-national comparison. American Sociological Review,

42(3), 450-464.

Heymann, S. J., Gerecke, M., & Chaussard, M. (2010). Paid health and family leave:

The Canadian experience in the global context. Canadian Journal of Public

Health, 101(7), S9-S15.

Hibbs Jr, D. A. (1992). Partisan theory after fifteen years. European Journal of Political

Economy, 8(3), 361-373.

Hicks, A. M. (1999). Social democracy and welfare capitalism: A century of income

294 security politics. Ithaca, NY: Cornell University Press.

Hicks, A., & Misra, J. (1993). Political resources and the growth of welfare in affluent

capitalist democracies, 1960-1982. American Journal of Sociology, 99(3), 668-

710.

Hicks, A., & Swank, D. H. (1984). Governmental redistribution in rich capitalist

democracies. Policy Studies Journal, 13(2), 265-286.

Hicks, A. M., & Swank, D. H. (1992). Politics, institutions, and welfare spending in

industrialized democracies, 1960-82. The American Political Science Review,

86(3), 658-674.

Hofrichter, R. (2003). Health and social justice: Politics, ideology, and inequity in the

distribution of disease (Vol. 11). Hoboken, NJ: Jossey-Bass Inc Pub.

Huber, E., Mustillo, T., & Stephens, J. D. (2008). Politics and social spending in Latin

America. The Journal of Politics, 70(2), 420-436.

Huber, E., Nielsen, F., Pribble, J., & Stephens, J. D. (2006). Politics and inequality in

Latin America and the Caribbean. American Sociological Review, 71(6), 943-963.

295 Huber, E., Ragin, C., & Stephens, J. D. (1993). Social democracy, Christian democracy,

constitutional structure, and the welfare state. American Journal of Sociology,

99(3), 711-749.

Huber, E., Ragin, C., & Stephens, J. D. (1997). Comparative welfare states data set.

Chapel Hill, NC: University of North Carolina.

Huber, E., Ragin, C., Stephens, J. D., Brady, D., & Beckfield, J. (2004). Comparative

welfare states data set. Chicago, IL: Northwestern University; Chapel Hill, NC:

University of North Carolina; Durham, NC: Duke University; and Bloomington, IN:

Indiana University.

Huber, E., & Stephens, J. D. (1993). Political parties and public pensions: A quantitative

analysis. Acta Sociologica, 36(4), 309-325.

Huber, E., & Stephens, J. D. (2000). Partisan governance, women's employment, and

the social democratic service state. American Sociological Review, 65(3), 323-

342.

Huber, E., & Stephens, J. D. (2001). Development and crisis of the welfare state:

Parties and policies in global markets. Chicago, IL: University of Chicago Press.

HRSDC. (2012). Collective bargaining in Canada 2011. Cat. No. LT-247-06-12E.

Ottawa, ON: Human Resources and Skills Development Canada. Retrieved from

http://publications.gc.ca/collections/collection_2012/rhdcc-hrsdc/HS21-5-2011-1-

eng.pdf

Im, K. S., Pesaran, M. H., & Shin, Y. (1997). Testing for unit roots in heterogeneous

panels. Unpublished manuscript. Cambridge, UK: University of Cambridge.

296 Im, K. S., Pesaran, M. H., & Shin, Y. (2003). Testing for unit roots in heterogeneous

panels. Journal of Econometrics, 115(1), 53-74.

Imbeau, L.M., Landry, R., Milner, H., Pétry, F., Crête, J., Forest, P., & Lemieux, V.

(2000). Comparative provincial policy analysis: A research agenda. Canadian

Journal of Political Science, 33(4), 779-804.

International Labour Organization. (2007). Equality at work: tackling the challenges:

global report under the follow-up to the ILO declaration on fundamental principles

and rights at work: report of the Director-General: International Labour

Conference, 96th Session, 2007 (Vol. 1). Geneva, CH: International Labour

Organization.

Iversen, T. (2001). The dynamics of welfare state expansion: Trade openness,

deindustrialization and partisan politics. In P. Pierson (Ed.), The New Politics of

the Welfare State (45-79). Oxford, UK: Oxford University Press.

Iversen, T., & Cusack, T. R. (2000). The causes of welfare state expansion:

deindustrialization or globalization?. World Politics, 52(3), 313-349.

Iversen, T., & Soskice, D. (2006). Electoral institutions and the politics of coalitions: Why

some democracies redistribute more than others. American Political Science

Review, 100(2), 165-181.

Jackman, R. W. (1987). Political institutions and voter turnout in the industrial

democracies. The American Political Science Review, 81(2), 405-423.

Jackson, A. (2006). Rowing against the tide: The struggle to raise union density in a

hostile environment. In P. Kumar, & C.R. Schenk (Eds.), Paths to union renewal:

Canadian experiences (61-78). Toronto, ON: University of Toronto Press.

297 Janoski, T. (Ed.). (2005). The handbook of political sociology: States, civil societies, and

globalization. Cambridge, UK: Cambridge University Press.

*Kangas, O. (2010). One hundred years of money, welfare and death: Mortality,

economic growth and the development of the welfare state in 17 OECD countries

1900–2000. International Journal of Social Welfare, 19(s1), S42-S59.

Kaplan, G. A. (2007). Health inequalities and the welfare state: Perspectives from social

epidemiology. Norwegian Journal of Epidemiology, 17(1), 9-20.

Kaplan, G. A., Pamuk, E. R., Lynch, J. W., Cohen, R. D., & Balfour, J. L. (1996).

Inequality in income and mortality in the United States: Analysis of mortality and

potential pathways. BMJ: British Medical Journal, 312(7037), 999-1003.

*Karim, S.A., Eikemo, T. A., & Bambra, C. (2010). Welfare state regimes and population

health: Integrating the East Asian welfare states. Health Policy, 94(1), 45-53.

Kaufman, J. S. (2002). Whad'ya know? Another view on cultural literacy. Epidemiology,

13(5), 500-503.

Kawachi, I., Kennedy, B. P., & Glass, R. (1999). Social capital and self-rated health: A

contextual analysis. American Journal of Public Health, 89(8), 1187-1193.

*Kawachi, I., Kennedy, B. P., Gupta, V., & Prothrow-Stith, D. (1999). Women's status

and the health of women and men: A view from the States. Social Science &

Medicine, 48(1), 21-32.

Kawachi, I., Kennedy, B. P., Lochner, K., & Prothrow-Stith, D. (1997). Social capital,

income inequality, and mortality. American Journal of Public Health,87(9), 1491-

1498.

298 Kawachi, I., Kennedy, B. P., & Wilkinson, R. G. (Eds.). (1999). Income inequality and

health (Vol. 1). New York, NY: New Press.

Kawakami, T., Kogi, K., Toyama, N., & Yoshikawa, T. (2004). Participatory approaches

to improving safety and health under trade union initiative. Industrial

Health, 42(2), 196-206.

Keating, D. P., & Hertzman, C. (Eds.). (1999). Developmental health and the wealth of

nations: Social, biological, and educational dynamics. New York, NY: Guilford

Press.

Kerr, C., Dunlop, J. T., & Harbison, F. H. (1960). Industrialism and industrial man: The

problems of labor and management in economic growth. Cambridge, MA:

Harvard University Press.

Kelber, M. (Ed.). (1994). Women and government: New ways to political power.

Westport, CT: Greenwood Publishing Group.

*Kelleher, C., Timoney, A., Friel, S., & McKeown, D. (2002). Indicators of deprivation,

voting patterns, and health status at area level in the Republic of Ireland. Journal

of Epidemiology and Community Health, 56(1), 36-44.

Kelleher, J. (2002). Cultural literacy and health. Epidemiology, 13(5), 497-500.

Kellermann, M. (2005). Power resources theory and inequality in the Canadian

provinces. Unpublished manuscript. Cambridge, MA: Harvard University.

Kelly, N. J. (2011). The politics of income inequality in the United States. Cambridge,

UK: Cambridge University Press.

Kennedy, P. (1998). A guide to econometrics (4th ed.). Cambridge, MA: MIT Press.

299 Kenworthy, L. (1999). Do social-welfare policies reduce poverty? A cross-national

assessment. Social Forces, 77(3), 1119-1139.

Kenworthy, L. (2004). Welfare states, real incomes and poverty. Syracuse, NY: Maxwell

School of Citizenship and Public Affairs, Syracuse University.

Kersbergen, K. V. (1995). Social capitalism. A study of Christian democracy and the

welfare state. London, UK: Routledge.

Kittel, B., & Obinger, H. (2003). Political parties, institutions, and the dynamics of social

expenditure in times of austerity. Journal of European Public Policy, 10(1), 20-45.

Klein, E. (1984). Gender politics: From consciousness to mass politics. Cambridge, MA:

Harvard University Press.

Kondo, N., Sembajwe, G., Kawachi, I., van Dam, R. M., Subramanian, S. V., &

Yamagata, Z. (2009). Income inequality, mortality, and self rated health: Meta-

analysis of multilevel studies. BMJ: British Medical Journal, 339, b4471.

Korpi, W. (1978). The working class in welfare capitalism: Work, unions and politics in

Sweden. London, UK: Routledge & Kegan Paul.

Korpi, W. (1983). The democratic class struggle. London, UK: Routledge & Kegan Paul.

Korpi, W. (1989). Power, politics, and state autonomy in the development of social

citizenship: Social rights during sickness in eighteen OECD countries since

1930. American Sociological Review, 54(3), 309-328.

Korpi, W. (2000). Faces of inequality: Gender, class, and patterns of inequalities in

different types of welfare states. Social Politics: International Studies in Gender,

State & Society, 7(2), 127-191.

300 Korpi, W. (2006). Power resources and employer-centered approaches in explanations

of welfare states and varieties of capitalism: Protagonists, consenters, and

antagonists. World Politics, 58(2), 167-206.

Korpi, W., & Palme, J. (1998). The paradox of redistribution and strategies of equality:

Welfare state institutions, inequality, and poverty in the Western

countries. American Sociological Review, 63(5), 661-687.

Korpi, W., & Palme, J. (2003). New politics and class politics in the context of austerity

and globalization: Welfare state regress in 18 countries, 1975-95. American

Political Science Review, 97(3), 425-446.

Lahelma, E., & Arber, S. (1994). Health inequalities among men and women in

contrasting welfare states Britain and three Nordic countries compared. The

European Journal of Public Health, 4(3), 213-226.

Lahelma, E., Arber, S., Rahkonen, O., & Silventoinen, K. (2000). Widening or

narrowing inequalities in health? Comparing Britain and Finland from the 1980s

to the 1990s. Sociology of Health & Illness, 22(1), 110-136.

Lahelma, E., Kivelä, K., Roos, E., Tuominen, T., Dahl, E., Diderichsen, F., ... & Yngwe,

M. Å. (2002). Analysing changes of health inequalities in the Nordic welfare

states. Social Science & Medicine, 55(4), 609-625.

Landon, S., McMillan, M. L., Muralidharan, V., & Parsons, M. (2006). Does health-care

spending crowd out other provincial government expenditures?. Canadian Public

Policy, 32(2), 121-141.

*Laporte, A., & Ferguson, B. S. (2003). Income inequality and mortality: Time series

evidence from Canada. Health Policy, 66(1), 107-117.

301 *Lena, H. F., & London, B. (1993). The political and economic determinants of health

outcomes: A cross-national analysis. International Journal of Health

Services, 23(3), 585-602.

Lijphart, A. (1997). Unequal participation: democracy's unresolved dilemma. American

Political Science Review, 91(1) 1-14.

Link, B. G., & Phelan, J. (1995). Social conditions as fundamental causes of

disease. Journal of Health and Social Behavior, 35, 80-94.

*Liu, L., Cline, R. R., Schondelmeyer, S. W., & Schommer, J. C. (2008). Pharmaceutical

expenditures as a correlate of population health in industrialized nations. The

Annals of Pharmacotherapy, 42(3), 368-374.

Lenski, G. (1966). Power and privilege: A theory of social stratification. New York, NY:

McGraw Hill.

Lora, E., & Olivera, M. (2007). Public debt and social expenditure: Friends or

foes?. Emerging Markets Review, 8(4), 299-310.

Lorber, J. (2005). Gender inequality. Los Angeles, CA: Roxbury Publishing Company.

302 Lovenduski, J., & Norris, P. (2003). Westminster women: The politics of presence.

Political Studies, 51(1), 84-102.

Lundberg, O. (2008). Commentary: politics and public health—some conceptual

considerations concerning welfare state characteristics and public health

outcomes. International Journal of Epidemiology, 37(5), 1105-1108.

Lundberg, O. (2010). Politics and public health—some conceptual considerations

concerning welfare state characteristics and public health outcomes.

International Journal of Epidemiology, 39(2), 632-634.

Lundberg, O., Fritzell, J., Yngwe, M. Å., & Kölegård, M. L. (2010). The potential

power of social policy programmes: Income redistribution, economic resources

and health. International Journal of Social Welfare, 19(s1), S2-S13.

*Lundberg, O., Yngwe, M. Å., Stjärne, M. K., Elstad, J. I., Ferrarini, T., Kangas, O., ... &

Fritzell, J. (2008). The role of welfare state principles and generosity in social

policy programmes for public health: an international comparative study. The

Lancet, 372(9650), 1633-1640.

Lynch, J., Due, P., Muntaner, C., & Smith, G. D. (2000). Social capital - is it a good

investment strategy for public health?. Journal of Epidemiology and Community

Health, 54(6), 404-408.

Lynch, J., & Kaplan, G. (2000). Socioeconomic position. In L.F. Berkman, & I. Kawachi

(Eds.), Social epidemiology (13-25). Oxford, UK: Oxford University Press.

Lynch, J., Smith, G. D., Harper, S. A., Hillemeier, M., Ross, N., Kaplan, G. A., &

Wolfson, M. (2004). Is income inequality a determinant of population health? Part

1. A systematic review. Milbank Quarterly, 82(1), 5-99.

303 *Lynch, J., Smith, G. D., Hillemeier, M., Shaw, M., Raghunathan, T., & Kaplan, G.

(2001). Income inequality, the psychosocial environment, and health:

Comparisons of wealthy nations. The Lancet, 358(9277), 194-200.

Lynch, J. W., Smith, G. D., Kaplan, G. A., & House, J. S. (2000). Income inequality and

mortality: importance to health of individual income, psychosocial environment, or

material conditions. BMJ: British Medical Journal, 320(7243), 1200-1204.

Macintyre, S. (1997). The Black Report and beyond what are the issues?. Social

Science & Medicine, 44(6), 723-745.

Mackenbach, J. P. (2009). Politics is nothing but medicine at a larger scale: Reflections

on public health’s biggest idea. Journal of Epidemiology and Community Health,

63(3), 181-184.

Mackenbach, J. P. (2013). Political conditions and life expectancy in Europe, 1900-

2008. Social Science & Medicine, 82, 134-146.

doi: 10.1016/j.socscimed.2012.12.022

Maddala, G.S. (1998). Recent developments in dynamic econometric modeling: A

personal viewpoint. Political Analysis, 7(1), 59–87.

Mahoney, J. (2001). Review essay: Beyond correlational analysis: Recent innovations

in theory and method. Sociological Forum, 16(3), 575-593.

Mansbridge, J. (1999). Should blacks represent blacks and women represent women?

A contingent "yes". Journal of Politics, 61(3), 628-657.

Marmot, M. G. (1998). Improvement of social environment to improve health. The

Lancet, 351(9095), 57-60.

304 Marmot, M. G. (2004). Status syndrome: How social standing directly affects your health

and life expectancy. London, UK: Bloomsbury.

Marmot, M. (2005). Social determinants of health inequalities. The Lancet, 365(9464),

1099-1104.

Marmot, M., & Wilkinson, R. (Eds.). (2009). Social determinants of health. Oxford, UK:

Oxford University Press.

Matland, R. E., & Studlar, D. T. (1998). Gender and the electoral opportunity structure in

the Canadian provinces. Political Research Quarterly, 51(1), 117-140.

McAllister, I., & Studlar, D. T. (2002). Electoral systems and women's representation: A

long‐term perspective. Representation, 39(1), 3-14.

McCrate, E. (1997). Welfare and women's earnings. Politics & Society, 25(4), 417-442.

McDonald, J. T., & Kennedy, S. (2004). Insights into the ‘healthy immigrant effect’:

health status and health service use of immigrants to Canada. Social Science &

Medicine, 59(8), 1613-1627.

McGrane, D. (2010). Diverging paths? A classification of the childcare regimes of

Canadian provinces. Winnipeg, MB: Annual Meeting of the Prairie Political

Science Association.

McKinlay, John B., & Sonja M. McKinlay. (1977). The questionable contribution of

medical measures to the decline of mortality in the United States in the twentieth

century. The Milbank Memorial Fund Quarterly. Health and Society, 55(3), 405-

428.

305 McLeod, C. B., Hall, P. A., Siddiqi, A., & Hertzman, C. (2012a). How society shapes the

health gradient: Work-related health inequalities in a comparative

perspective. Annual Review of Public Health, 33, 59-73. doi: 10.1146/annurev-

publhealth-031811-124603

McLeod, C. B., Lavis, J. N., MacNab, Y. C., & Hertzman, C. (2012b). Unemployment

and mortality: A comparative study of Germany and the United States. American

Journal of Public Health, 102(8), 1542-1550.

Merton, R. K. (1968). Social theory and social structure. Glencoe, IL: Free Press.

Microsoft. (2010). Microsoft Excel [computer software]. Redmond, WA: Microsoft.

Mikkonen, J., & Raphael, D. (2010). Social determinants of health: The Canadian facts.

Toronto, ON: School of Health Policy and Management, York University.

Milan, A. (2011). Migration: Interprovincial, 2008/2009, Cat. No. 91-209-X. Ottawa, ON:

Statistics Canada.

Mirowsky, J., & Ross, C. E. (2003). Education, social status, and health. Piscataway,

NJ: Transaction Books.

Mitchell, R., Dorling, D., & Shaw, M. (2000). Inequalities in life and death. Bristol, UK:

Policy Press.

Moller, S., Huber, E., Stephens, J. D., Bradley, D., & Nielsen, F. (2003). Determinants of

relative poverty in advanced capitalist democracies. American Sociological

Review, 68(1), 22-51.

Morel, S. (2002). The insertion model or the workfare model?: The transformation of

social assistance within Quebec and Canada. Ottawa, ON: Status of Women

Canada.

306 Morris, J. N., Blane, D. B., & White, I. R. (1996). Levels of mortality, education, and

social conditions in the 107 local education authority areas of England. Journal of

Epidemiology and Community health, 50(1), 15-17.

Mulé, R. (2001). Political parties, games and redistribution. Cambridge, UK: Cambridge

University Press.

Muntaner, C. (2004). Commentary: Social capital, social class, and the slow progress of

psychosocial epidemiology. International Journal of Epidemiology, 33(4), 674-

680.

Muntaner, C., Borrell, C., Espelt, A., RodrÍguez-sanz, M., PasarÍn, M. I., Benach, J., &

Navarro, V. (2010). Politics or policies vs politics and policies: A comment on

Lundberg. International Journal of Epidemiology, 39(5), 1396-1397.

Muntaner, C., Borrell, C., Kunst, A., Chung, H., Benach, J., & Ibrahim, S. (2006). Social

class inequalities in health: Does welfare state regime matter? In D. Raphael, T.

Bryant, & M. Rioux (Eds.), Staying alive: Critical perspectives on health, illness,

and health care (139-158). Toronto, ON: Canadian Scholars’ Press.

Muntaner, C., Borrell, C., Ng, E., Chung, H., Espelt, A., Rodriguez‐Sanz, M., ... &

O’Campo, P. (2011). Politics, welfare regimes, and population health:

Controversies and evidence. Sociology of Health & Illness, 33(6), 946-964.

Muntaner, C., Borrell, C., Ng, E., Chung, H., Espelt, A., Rodriguez-Sanz, M., ... &

O’Campo, P. (2012). Locating politics in social epidemiology. In P. O’Campo, &

J.R. Dunn (Eds), Rethinking Social Epidemiology (175-202). Netherlands:

Springer.

307 Muntaner, C., & Chung, H. (2008). Macrosocial determinants, epidemiology, and

health policy: Should politics and economics be banned from social determinants

of health research?. Journal of Public Health Policy, 29(3), 299-306.

Muntaner, C., Chung, H., Benach, J., & Ng, E. (2012). Hierarchical cluster analysis of

labour market regulations and population health: A taxonomy of low-and middle-

income countries. BMC Public Health, 12(1), 286.

Muntaner, C., Ng, E., & Chung, H. (2012). Better health: An analysis of public policy and

programming focusing on the determinants of health and health outcomes that

are effective in achieving the healthiest populations. Ottawa, ON: Canadian

Health Services Research Foundation.

Muntaner, C., & Lynch, J.W. (1999). Income inequality, social cohesion, and class

relations: A critique of Wilkinson’s neo-Durkheimian research program.

International Journal of Health Services, 29(1), 59-81.

*Muntaner, C., Lynch, J. W., Hillemeier, M., Lee, J. H., David, R., Benach, J., & Borrell,

C. (2002). Economic inequality, working-class power, social capital, and cause-

specific mortality in wealthy countries. International Journal of Health

Services, 32(4), 629-656.

Muntaner, C., Lynch, J.W., & Oates, G.L. (1999). The social class determinants of

income inequality and social cohesion. International Journal of Health Services,

29(4), 699-732.

Muntaner, C., Ng, E., Vanroelen, C., Christ, S., & Eaton, W. W. (2013). Social

stratification, social closure, and social class as determinants of mental health

308 disparities. In C.S. Aneshensel, J.C. Phelan, & A. Bierman, Handbook of the

Sociology of Mental Health (205-227). Netherlands: Springer.

Muntaner, C., Rai, N., Ng, E., & Chung, H. (2012). Social class, politics, and the spirit

level: Why income inequality remains unexplained and unsolved. International

Journal of Health Services, 42(3), 369-381.

Muntaner, C., Salazar, R. M. G., Benach, J., & Armada, F. (2006). Venezuela's Barrio

Adentro: An alternative to neoliberalism in health care. International Journal of

Health Services, 36(4), 803-811.

Murray, C. J. L., Gakidou, E. E., & Frenk, J. (1999). Critical reflection-health

inequalities and social group differences: What should we measure?. Bulletin of

the World Health Organization, 77(7), 537-544.

Myles, J., & Quadagno, J. (2002). Political theories of the welfare state. Social Service

Review, 76(1), 34-57.

Navarro, V. (Ed.). (2000). The political economy of social inequalities: Consequences

for health and quality of life. Amityville, NY: Baywood Publishing Company.

Navarro, V. (2003). Introduction: Objectives and purposes of the study. International

Journal of Health Services, 33(3), 407-417.

Navarro, V. (Ed.). (2007). Neoliberalism, globalization, and inequalities: Consequences

for health and quality of life. Amityville, NY: Baywood Publishing Company.

*Navarro, V., Borrell, C., Benach, J., Muntaner, C., Quiroga, A., Rodriquez-Sanz, M., ...

& Pasarin, M. I. (2003). The importance of the political and the social in

explaining mortality differentials among the countries of the OECD, 1950-1998.

International Journal of Health Services, 33(3), 419-494.

309 Navarro, V., & Muntaner, C. (2004). Political and economic determinants of population

health and well-being: Controversies and developments. Amityville, NY:

Baywood Publishing Company.

*Navarro, V., Muntaner, C., Borrell, C., Benach, J., Quiroga, Á., Rodríguez-Sanz, M., ...

& Pasarín, M. I. (2006). Politics and health outcomes. The Lancet, 368(9540),

1033-1037.

*Navarro, V., & Shi, L.Y. (2001). The political context of social inequalities and health.

International Journal of Health Services, 31(1), 1−21.

Newbold, B.K. (2005). Self-rated health within the Canadian immigrant population:

Risk and the healthy immigrant effect. Social Science & Medicine, 60(6), 1359-

1370.

Neumann, G., Pedersen, P. J., & Westergård-Nielsen, N. (1991). Long-run international

trends in aggregate unionization. European Journal of Political Economy, 7(3),

249-274.

Ng, E., Wilkins, R., Gendron, F., & Berthelot, J.M. (2005). Dynamics of immigrants’

health in Canada: Evidence from the National Population Health Survey. Cat. No.

82-618. Ottawa, ON: Statistics Canada.

Nordenmark, M., Strandh, M., & Layte, R. (2006). The impact of unemployment benefit

system on the mental well-being of the unemployed in Sweden, Ireland and

Great Britain. European Societies, 8(1), 83-110.

*Norström, T., & Palme, J. (2010). Public pension institutions and old‐age mortality in a

comparative perspective. International Journal of Social Welfare, 19(s1), S121-

310 S130.

*Novignon, J., Olakojo, S. A., & Nonvignon, J. (2012). The effects of public and private

health care expenditure on health status in sub-Saharan Africa: New evidence

from panel data analysis. Health Economics Review, 2(1), 1-8.

O’Brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors.

Quality & Quantity, 41(5), 673-690.

O’Campo, P., & Dunn J.R. (eds.) (2011). Rethinking social epidemiology: Towards a

science of change. New York, NY: Springer.

O'Connor, J. S. (1989). Welfare expenditure and policy orientation in Canada in

comparative perspective. Canadian Review of Sociology, 26(1), 127-150.

OECD. (1994). Social expenditure data. Paris, FR: Organisation for Economic Co-

Operation and Development.

OECD. (2008). Growing unequal? Income distribution and poverty in OECD Countries.

Paris, FR: Organisation for Economic Co-Operation and Development.

OECD. (2009). Health data. Paris, FR: Organisation for Economic Co-Operation and

Development.

311 Orloff, A. S. (1993). Gender and the social rights of citizenship: the comparative

analysis of gender relations and welfare states. American Sociological Review,

58(3), 303-328.

Orloff, A. (1996). Gender in the welfare state. Annual Review of Sociology, 22, 51-78.

Osberg, L., & Xu, K. (1999). Poverty intensity: How well do Canadian provinces

compare?. Canadian Public Policy, 25(2), 179-195.

Pampel, F. C. (1992). Age, class, politics, and the welfare state. Cambridge, UK:

Cambridge University Press.

Pampel, F. C., & Williamson, J. B. (1988). Welfare spending in advanced industrial

democracies, 1950-1980. American Journal of Sociology, 93(6), 1424-1456.

Pawson, R., & Tiller, N. (1997). Realistic evaluation. Thousand Oaks, CA: Sage.

Pega, F., Kawachi, I., Rasanathan, K., & Lundberg, O. (2013). Politics, policies and

population health: A Commentary on Mackenbach, Hu and Looman

(2013).Social Science & Medicine.

http://dx.doi.org/10.1016/j.socscimed.2013.06.007

Piazza, J. (2001). De-linking labor: Labor unions and social democratic parties under

globalization. Party Politics, 7(4), 413-435.

Phillips, A. (1995). The politics of presence. Oxford, UK: Oxford University Press.

Piven, F. F., & Cloward, R. (Eds.). (1993). Regulating the poor: The functions of public

welfare (Vol. 3). New York: NY: Random House Digital, Inc..

Plümper, T., Troeger, V. E., & Manow, P. (2005). Panel data analysis in comparative

politics: Linking method to theory. European Journal of Political Research, 44(2),

327-354.

312 Podestà, F. (2002). Recent developments in quantitative comparative methodology: The

case of pooled time series cross-section analysis. DSS Papers Soc, 3-02.

Podestà, F. (2006). Comparing time series cross-section model specifications: The case

of welfare state development. Quality and Quantity, 40(4), 539-559.

Pontusson, J., & Rueda, D. (2010). The politics of inequality: Voter mobilization and left

parties in advanced industrial states. Comparative Political Studies, 43(6), 675-

705.

Popham, F., Dibben, C., & Bambra, C. (2013). Are health inequalities really not the

smallest in the Nordic welfare states? A comparison of mortality inequality in 37

countries. Journal of Epidemiology and Community Health, 67(5), 412-418.

Pong, R. W., DesMeules, M., & Lagacé, C. (2009). Rural–urban disparities in health:

How does Canada fare and how does Canada compare with

Australia?.Australian Journal of Rural Health, 17(1), 58-64.

Pryor, F. L. (1968). Public expenditures in communist and capitalist nations.

London, UK: Allen & Unwin.

Putnam, S., & Galea, S. (2008). Epidemiology and the macro-social determinants of

health. Journal of Public Health Policy, 29(3), 275-289.

Quadagno, J. (1994). The color of welfare: How racism undermined the war on poverty.

Oxford, UK: Oxford University Press.

Ragin, C. C. (1994). Introduction to qualitative comparative analysis. In T. Janoski, &

A.M. Hicks (Eds.), The comparative political economy of the welfare state (299-

313 319). Cambridge, UK: Cambridge University Press.

Raphael, D. (2003). Addressing the social determinants of health in Canada: Bridging

the gap between research findings and public policy. Policy Options, 24(3), 35-

40.

Raphael, D. (Ed.). (2004). Social determinants of health: Canadian perspectives.

Toronto, ON: Canadian Scholars’ Press.

Raphael, D. (2006). Social determinants of health: present status, unanswered

questions, and future directions. International Journal of Health Services, 36(4),

651-677.

Raphael, D., & Bryant, T. (2004). The welfare state as a determinant of women’s health:

Support for women’s quality of life in Canada and four comparison

nations. Health Policy, 68(1), 63-79.

Ratcliffe, J. (1978). Social justice and the demographic transition: Lessons from India's

Kerala State. International Journal of Health Services, 8(1), 123-144.

Reed, W. R., & Ye, H. (2011). Which panel data estimator should I use?. Applied

Economics, 43(8), 985-1000.

*Reitan, T. C. (2003). Too sick to vote? Public health and voter turnout in Russia during

the 1990s. Communist and Post-Communist Studies, 36(1), 49-68.

*Richter, M., Rathman, K., Gabhainn, S. N., Zambon, A., Boyce, W., & Hurrelmann, K.

(2012). Welfare state regimes, health and health inequalities in adolescence: A

multilevel study in 32 countries. Sociology of Health & Illness, 34(6), 858-879.

*Rodríguez-Sanz, M., Borrell, C., Urbanos, R., Pasarín, M. I., Rico, A., Fraile, M., ... &

314 Navarro, V. (2003). Power relations and premature mortality in Spain's

autonomous communities. International Journal of Health Services, 33(4), 687-

722.

Romanow, R. J. (2002). Building on values: The future of health care in Canada. Final

report. Ottawa, ON: Commission on the Future of Health Care in Canada.

Ronson, B., & Rootman, I. (2009). Literacy and health literacy: New understandings

about their impact on health. In D. Raphael (Ed.), Social determinants of health:

Canadian perspectives (2nd ed.) (170-186). Toronto, ON: Canadian Scholars’

Press.

*Ronzio, C. R. (2003). Urban premature mortality in the US between 1980 and 1990:

Changing roles of income inequality and social spending. Journal of Public

Health Policy, 24(3/4), 386-400.

*Ronzio, C. R., Pamuk, E., & Squires, G. D. (2004). The politics of preventable deaths:

Local spending, income inequality, and premature mortality in US cities. Journal

of Epidemiology and Community Health, 58(3), 175-179.

Rose, G. (1992). The strategy of preventive medicine. Oxford, UK: Oxford University

Press.

Rose, G. ([1985] 2001). Sick individuals and sick populations. International Journal of

Epidemiology, 30(3), 427-432.

Rose, G., Khaw, K. T., & Marmot, M. (2008). Rose’s strategy of preventive medicine.

Oxford, UK: Oxford University Press.

Ross, N. A., & Wolfson, M. C. (1999). Income inequality and mortality in Canada and

315 the United States: An analysis of provinces/states. Annals of the New York

Academy of Sciences, 896(1), 338-340.

Ross, N. A., Wolfson, M. C., Berthelot, J. M., & Dunn, J. R. (1999). Why is mortality

higher in unequal societies? Interpreting income inequality and mortality in

Canada and the United States (No. 2000: 4). Vancouver, BC: University of British

Columbia-Centre for Health Services and Policy Research.

Ross, N. A., Wolfson, M. C., Dunn, J. R., Berthelot, J. M., Kaplan, G. A., & Lynch, J. W.

(2000). Relation between income inequality and mortality in Canada and in the

United States: Cross sectional assessment using census data and vital

statistics. BMJ, 320(7239), 898-902.

Ross, C. E., & Wu, C. (1995). The link between education and health. American

Sociological Review, 60(5), 719-745.

*Rostila, M. (2007). Social capital and health in European welfare regimes: A multilevel

approach. Journal of European Social Policy, 17(3), 223-239.

Rueschemeyer, D., & Evans, P.B. (1985). The state and economic transformation:

Toward an analysis of the conditions underlying effective intervention. In P.B.

Evans, D. Rueschemeyer, & T. Skocpol (Eds.), Bringing the state back in (46-

47). New York, NY: Cambridge University Press.

Sacker, A., Worts, D., & McDonough, P. (2011). Social influences on trajectories of

self-rated health: evidence from Britain, Germany, Denmark and the USA.

Journal of Epidemiology and Community Health, 65(2), 130-136.

316 Saint-Arnaud, S., & Bernard, P. (2003). Convergence or resilience? A hierarchical

cluster analysis of the welfare regimes in advanced countries. Current

Sociology, 51(5), 499-527.

Saku, J. C. (1999). Aboriginal census data in Canada: A research note. Canadian

Journal of Native Studies, 19(2), 369.

Saku, J. C. (Ed.). (2010). Aboriginal Canadians: Issues and challenges. Beersheba, IL:

Ben-Gurion University of the Negev, Department of Geography and

Environmental Development.

Sanders, A. E., Slade, G. D., John, M. T., Steele, J. G., Suominen-Taipale, A. L., Lahti,

S., ... & Allen, P. F. (2009). A cross-national comparison of income gradients in

oral health quality of life in four welfare states: Application of the Korpi and Palme

typology. Journal of Epidemiology and Community Health, 63(7), 569-574.

Sanmartin, C., Ross, N. A., Tremblay, S., Wolfson, M., Dunn, J. R., & Lynch, J. (2003).

Labour market income inequality and mortality in North American metropolitan

areas. Journal of Epidemiology and Community Health, 57(10), 792-797.

Sanmartin, C., & Ng, E. (2004). Joint Canada/United States Survey of Health, 2002–03.

Ottawa, ON: Statistics Canada.

*Schell, C. O., Reilly, M., Rosling, H., Peterson, S., & Ekström, A. M. (2007).

Socioeconomic determinants of infant mortality: A worldwide study of 152 low-,

middle-, and high-income countries. Scandinavian Journal of Public Health,

35(3), 288-297.

317 Schwartz, S., & Diez-Roux, R. (2001). Commentary: Causes of incidence and causes of

cases—a Durkheimian perspective on Rose. International Journal of

Epidemiology, 30(3), 435-439.

Scruggs, L. (2007). Welfare state generosity across space and time. In J. Clasen, &

N.A. Siegel (eds.), Investigating welfare state change. The 'dependent variable

problem' in comparative analysis (133-165). Cheltenham/Northampton: Edward

Elgar Publishing.

Scruggs, L. (2008). Social rights, welfare generosity, and inequality. In P. Beramendi, &

C. Anderson (Eds.), Democracy, inequality, and representation: A comparative

perspective (62-90). New York, NY: Russell Sage Foundation.

Scruggs, L., & Allan, J.P. (2006a). Welfare-state decommodification in 18 OECD

countries: A replication and revision. Journal of European Social Policy, 16(1),

55-72.

Scruggs, L., & Allan, J. P. (2006b). The material consequences of welfare states benefit

generosity and absolute poverty in 16 OECD countries. Comparative Political

Studies, 39(7), 880-904.

Sen, A. (1993). Capability and well-being. The Quality of Life, 1(9), 30-54.

Shaw, M. (1999). The widening gap: health inequalities and policy in Britain. Bristol, UK:

The Policy Press.

Shaw, M., Dorling, D., & Mitchell, R. (2002). Health, place, and society. London, UK:

Pearson Education.

Siddiqi, A., & Hertzman, C. (2007). Towards an epidemiological understanding of the

effects of long-term institutional changes on population health: A case study of

318 Canada versus the USA. Social Science & Medicine, 64(3), 589-603.

Sigel, R. S. (1996). Ambition & accommodation: How women view gender relations.

Chicago, IL: University of Chicago Press.

Skocpol, T., & Amenta, E. (1986). States and social policies. Annual Review of

Sociology, 12, 131-157.

Smith, G.D. (Ed.). (2003). Inequalities in health: Life course perspectives. Bristol,

UK: Policy Press.

*Smith, G. D., & Dorling, D. (1996). "I'm all right, John": voting patterns and mortality in

England and Wales, 1981-92. BMJ: British Medical Journal, 313(7072), 1573-

1577.

*Smith, G.D., & Dorling, D. (1997). Associations between voting patterns and mortality

remain. BMJ: British Medical Journal, 315(7105), c430-431.

Smith, G. D., Hart, C., Hole, D., MacKinnon, P., Gillis, C., Watt, G., ... & Hawthorne, V.

(1998). Education and occupational social class: Which is the more important

indicator of mortality risk?. Journal of Epidemiology and Community Health,

52(3), 153-160.

Stansfeld, S. A., & Marmot, M. (Eds.). (2002). Stress and the heart: Psychosocial

pathways to coronary heart disease. London, UK: BMJ Books.

StataCorp. (2011). Stata statistical software: Release 12. College Station, TX:

StataCorp LP.

Statistics Canada. (1995). Births and deaths 1995, Cat. No. 84-210-XIB. Ottawa, ON:

Statistics Canada.

319 Statistics Canada. (1996). Births and deaths 1996, shelf tables. Cat. No. 84F0210XPB.

Ottawa, ON: Statistics Canada.

Statistics Canada. (1996). Vital statistics, death database, demography division

(population estimates), and the 1996 census. Ottawa, ON: Statistics Canada.

Statistics Canada. (1997). Mortality – summary list of causes, 1997, shelf tables. Cat.

No. 84F0209XIB. Ottawa, ON: Statistics Canada.

Statistics Canada. (1998). Mortality – summary list of causes, 1998, shelf tables. Cat.

No. 84F0209XPB. Ottawa, ON: Statistics Canada.

Statistics Canada. (1999). Mortality – summary list of causes, 1999, shelf tables. Cat.

No. 84F0209XPB. Ottawa, ON: Statistics Canada.

Statistics Canada. (2006). Ethnic origin reference guide, 2006 census. Cat. No. 97-562-

GWE2006025. Ottawa, ON: Statistics Canada.

Statistics Canada. (2007, June 22). Government spending on social services. Retrieved

from http://www.statcan.gc.ca/daily-quotidien/070622/dq070622b-eng.htm

Statistics Canada. (2009). Financial management system (FSM). Cat. No. 68-F0023X.

Ottawa, ON: Statistics Canada.

Statistics Canada. (2013). Deaths and mortality rate, by selected grouped causes and

sex, Canada, provinces and territories, annual, CANSIM (database) (CANSIM

Table 102-0552). Ottawa, ON: Statistics Canada, 2010.

Stephens, J.D. (1979). The transition from capitalism to socialism. London, UK:

Macmillan.

*Stuckler, D., Basu, S., Suhrcke, M., Coutts, A., & McKee, M. (2009). The public health

effect of economic crises and alternative policy responses in Europe: An

320 empirical analysis. The Lancet, 374(9686), 315-323.

Stuckler, D., King, L., & McKee, M. (2009). Mass privatisation and the post-communist

mortality crisis: A cross-national analysis. The Lancet, 373(9661), 399-407.

Studlar, D. T., & Matland, R. E. (1996). The dynamics of women's representation in the

Canadian provinces: 1975-1994. Canadian Journal of Political Science, 29(2),

269-293.

Studlar, D. T., & Moncrief, G. F. (1997). The recruitment of women cabinet ministers in

the Canadian provinces. Governance, 10(1), 67-81.

Studlar, D. T., & Moncrief, G. F. (1999). Women's work? The distribution and prestige of

portfolios in the Canadian provinces. Governance, 12(4), 379-395.

Svallfors, S. (Ed.). (2007). The political sociology of the welfare state: Institutions, social

cleavages and orientations. Palo Alto, CA: Stanford University Press.

Swank, D. (2002). Global capital, political institutions, and policy change in developed

welfare states. Cambridge, UK: Cambridge University Press.

Swers, M. L. (1998). Are women more likely to vote for women's issue bills than their

male colleagues?. Legislative Studies Quarterly, 23(3), 435-448.

Tarasuk, V. (2009). Food insecurity and health. In D. Raphael (Ed.), Social

determinants of health: Canadian perspectives (205-220) (2nd ed.). Toronto, ON:

Canadian Scholars’ Press.

Tesh, S. N. (1988). Hidden arguments: Political ideology and disease prevention policy.

New Brunswick, NJ: Rutgers University Press.

Thompson, W. (1996). The Left in history: Revolution and reform in twentieth-century

politic. London, UK: Pluto Press.

321 Tilly, C. (2001). Mechanisms in political processes. Annual Review of Political

Science, 4(1), 21-41.

Titmuss, R. M., Abel-Smith, B., & Titmuss, K. (1974). Social policy. London:

Allen & Unwin.

Tremblay, M. (1998). Do female MPs substantively represent women? A study of

legislative behaviour in Canada's 35th Parliament. Canadian Journal of Political

Science, 31(3), 435-466.

Tresserras, R., Canela, J., Alvarez, J., Sentis, J., & Salleras, L. (1992). Infant mortality,

per capita income, and adult illiteracy: An ecological approach. American Journal

of Public Health, 82(3), 435-438.

Tsebelis, G. (2002). Veto players: How political institutions work. Princeton, NJ:

Princeton University Press.

Uppal, S. (2011). Unionization 2011. Perspectives on labour and income. Ottawa, ON:

Statistics Canada.

*van der Wel, K. A., Dahl, E., & Thielen, K. (2012). Social Inequalities in "sickness":

Does welfare state regime type make a difference? A multilevel analysis of men

and women in 26 European countries. International Journal of Health Services,

42(2), 235-255.

*Veenhoven, R. (2000). Well‐being in the welfare state: Level not higher, distribution not

more equitable. Journal of Comparative Policy Analysis: Research and Practice,

2(1), 91-125.

*Veenhoven, R., & Ouweneel, P. (1995). Livability of the welfare-state. Social Indicators

Research, 36(1), 1-48.

322 Vlahov, D., & Galea, S. (2003). Urban health: a new discipline. The Lancet, 362(9390),

1091-1092.

Walker, A., & Wong, C. K. (Eds.). (2005). East Asian welfare regimes in transition: From

Confucianism to globalisation. Bristol, UK: The Policy Press.

Wängnerud, L. (2009). Women in parliaments: Descriptive and substantive

representation. Annual Review of Political Science, 12, 51-69. doi:

10.1146/annurev.polisci.11.053106.123839

Wesley, J. J. (2010). Slack in the system: Turnout in Canadian provincial elections,

1965–2009. Concordia University, Montreal, QC: Annual Meeting of the

Canadian Political Science Association.

White, I. R., Blane, D., Morris, J. N., & Mourouga, P. (1999). Educational attainment,

deprivation-affluence and self reported health in Britain: A cross sectional study.

Journal of Epidemiology and Community Health, 53(9), 535-541.

Whitehorn, A. (2007). Social Democracy and the New Democratic Party. In A. Gagnon,

& B. Tanguay (Eds.), Canadian parties in transition (3rd ed.) (Chapter 6).

Peterborough, ON: Broadview.

Wilensky, H. L. (1974). The welfare state and equality: Structural and ideological roots

of public expenditure. Berkeley, CA: University of California Press.

Wilensky, H. (1976). The ‘new corporatism’, centralization, and the welfare state.

London, UK: Sage.

Wilensky, H. L. (2002). Rich democracies: Political economy, public policy, and

performance. Berkeley, CA: University of California Press.

323 Wilensky, H. L., & Lebeaux, C. N. (1965). Industrial society and social welfare. New

York, NY: Free Press.

Wilkins, R. (2007). Mortality by neighbourhood income in urban Canada from 1971 to

2001. HAMG Seminar. Ottawa, ON: Statistics Canada.

Wilkinson, R. G. (2002). Unhealthy societies: The afflictions of inequality. New York,

NY: Routledge.

Wilkinson, R. G. (2005). The impact of inequality: How to make sick societies healthier.

New York, NY: New Press.

Wilkinson, R. G., & Pickett, K. E. (2006). Income inequality and population health: A

review and explanation of the evidence. Social Science & Medicine, 62(7), 1768-

1784.

Wilkinson R. G. & Pickett K. (2009). The spirit level - Why more equal societies almost

always do better. London, UK: Allen Lane.

*Witvliet, M. I., Arah, O. A., Stronks, K., & Kunst, A. E. (2011). Taking welfare state

regime research globally: An application of the Wood and Gough typology to

individual health. Social Theory & Health, 9(4), 355-366.

*Witvliet, M. I., Kunst, A. E., Stronks, K., & Arah, O. A. (2012). Assessing where

vulnerable groups fare worst: A global multilevel analysis on the impact of

welfare regimes on disability across different socioeconomic groups. Journal of

Epidemiology and Community Health, 66(9), 775-781.

Wood, G., & Gough, I. (2006). A comparative welfare regime approach to global social

policy. World Development, 34(10), 1696-1712.

324 Wooldridge, J.M. (2000). Introductory econometrics: A modern approach. New York,

NY: South-Western College Publishing.

Wooldridge, J.M. (2002), Econometric analysis of cross section and panel data.

Cambridge, MA: MIT Press.

Wright, E. O., Levine, A., & Sober, E. (1992). Reconstructing Marxism: Essays on

explanation and the theory of history. Brooklyn, NY: Verso.

Wu, C. F. J. (1986). Jackknife, bootstrap and other resampling methods in regression

analysis. The Annals of Statistics, 14(4), 1261-1295.

*Xu, K. T. (2006). State-level variations in income-related inequality in health and health

achievement in the US. Social Science & Medicine, 63(2), 457-464.

*Zambon, A., Boyce, W., Cois, E., Currie, C., Lemma, P., Dalmasso, P., ... & Cavallo, F.

(2006). Do welfare regimes mediate the effect of socioeconomic position on

health in adolescence? A cross-national comparison in Europe, North America,

and Israel. International Journal of Health Services, 36(2), 309-329.

325 Appendix A. Data Sources

Details from chapter four, Methods

This appendix provides references to the data used in this dissertation. Data was retrieved from the Canadian Socio-Economic Information Management System

(CANSIM) at Computing in the Humanities and Social Sciences (CHASS), University of

Toronto. For data from CANSIM Table II, the series numbers are presented in the following order: Newfoundland, Prince Edward Island, Nova Scotia, New Brunswick,

Quebec, Ontario, Manitoba, Saskatchewan, Alberta, and British Columbia.

Total age-standardized mortality rate. The total age-standardized mortality rate is a weighted average of the total age-specific mortality rates per 1,000 persons, where the 1991 Canadian Census of Population is used as the standard population for the calculation. Data on total mortality rates from 1975 to 1999 are retrieved from Statistics

Canada births and deaths 1995, Cat No. 84-210-XIB. Ottawa: Statistics Canada, 1995;

Statistics Canada births and deaths 1996, Shelf Tables. Cat No. 84F0210XPB. Ottawa:

Statistics Canada, 1996; Statistics Canada. Mortality – Summary List of Causes, 1997,

Shelf Tables. Cat. No. 84F0209XIB. Ottawa: Statistics Canada, 1997; Statistics

Canada. Mortality – Summary List of Causes, 1998, Shelf Tables. Cat. No.

84F0209XPB. Ottawa: Statistics Canada, 1998; and Statistics Canada. Mortality –

Summary List of Causes, 1999, Shelf Tables. Cat. No. 84F0209XPB. Ottawa: Statistics

Canada, 1999. Data from 2000 to 2008 is available from CANSIM II Table 102-0552, series: v30131168, v30131564, v30131960, v30132356, v30132752, v30133148, v30133544, v30133940, v30134336, v30134732.

326 Male age-standardized mortality rate. The male age-standardized mortality rate is a weighted average of the male age-specific mortality rates per 1,000 persons, where the 1991 Canadian Census of Population is used as the standard population for the calculation. Data on total mortality rates from 1975 to 1999 are retrieved from Statistics

Canada births and deaths 1995, Cat No. 84-210-XIB. Ottawa: Statistics Canada, 1995;

Statistics Canada births and deaths 1996, Shelf Tables. Cat No. 84F0210XPB. Ottawa:

Statistics Canada, 1996; Statistics Canada. Mortality – Summary List of Causes, 1997,

Shelf Tables. Cat. No. 84F0209XIB. Ottawa: Statistics Canada, 1997; Statistics

Canada. Mortality – Summary List of Causes, 1998, Shelf Tables. Cat. No.

84F0209XPB. Ottawa: Statistics Canada, 1998; and Statistics Canada. Mortality –

Summary List of Causes, 1999, Shelf Tables. Cat. No. 84F0209XPB. Ottawa: Statistics

Canada, 1999. Data from 2000 to 2008 is available from CANSIM II Table 102-0552, series: v30131300, v30131696, v30132092, v30132488, v30132884, v30133280, v30133676, v30134072, v30134468, v30134864.

Female age-standardized mortality rate. The female age-standardized mortality rate is a weighted average of the female age-specific mortality rates per 1,000 persons, where the 1991 Canadian Census of Population is used as the standard population for the calculation. Data on total mortality rates from 1975 to 1999 are retrieved from

Statistics Canada births and deaths 1995, Cat No. 84-210-XIB. Ottawa: Statistics

Canada, 1995; Statistics Canada births and deaths 1996, Shelf Tables. Cat No.

84F0210XPB. Ottawa: Statistics Canada, 1996; Statistics Canada. Mortality – Summary

List of Causes, 1997, Shelf Tables. Cat. No. 84F0209XIB. Ottawa: Statistics Canada,

1997; Statistics Canada. Mortality – Summary List of Causes, 1998, Shelf Tables. Cat.

327 No. 84F0209XPB. Ottawa: Statistics Canada, 1998; and Statistics Canada. Mortality –

Summary List of Causes, 1999, Shelf Tables. Cat. No. 84F0209XPB. Ottawa: Statistics

Canada, 1999. Data from 2000 to 2008 is available from CANSIM II Table 102-0552, series: v30131432, v30131828, v30132224, v30132620, v30133016, v30133412, v30133808, v30134204, v30134600, v30134996.

Population estimates. Population data is available from 1976 to 2008 and retrieved from CANSIM II, Table 510-001, series: v466983, v467298, v467613, v467928, v468243, v468558, v468873, v469188, v469503, v469818.

Consumer price index. The consumer price index (CPI) is an indicator of changes in consumer prices experienced by Canadians in each province, and it is used to escalate a given dollar value, over time, to preserve the purchasing power of that value (2002 = 100). It is obtained by comparing, over time, the cost of a fixed basket of goods and services purchased by consumers. CPI data is available from 1989 to 2008 and collected from CANSIM II, 326-0021, series: v41693542, v41693677, v41693811, v41693946, v41694081, v41694217, v41694353, v41694489, v41694625, v41694760.

Health expenditure. This variable captures provincial spending made to ensure that necessary health services are available to all citizens. The health expenditure variable is converted into real per capita terms using population data from CANSIM II,

Table 510-001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-

0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-

0001, series: v645264, v645330, v645396, v645462, v645528, v645594, v645660, v645726, v645792, v645858.

328 Social services expenditure. Spending on social services include actions taken by provincial governments to offset or to forestall situations where the well-being of individuals or families is threatened by circumstances beyond their control. The social service expenditure variable is converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from

CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from

CANSIM II, Table 385-0001, series: v645269, v645335, v645401, v645467, v645533, v645599, v645665, v645731, v645797, v645863.

Education expenditure. Education spending includes the costs of developing, improving, and operating educational systems and the provision of specific education services. The education expenditure variable is converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: v645276, v645342, v645408, v645474, v645540, v645606, v645672, v645738, v645804, v645870.

CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM

II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II,

Table 385-0001, series: v645265, v645331, v645397, v645463, v645529, v645595, v645661, v645727, v645793, v645859.

329 Medical care expenditure. Spending on medical care includes general medical care, drug programs, dental and visiting-nurse services, and out-patient care services.

The medical care expenditure variable is converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: v645266, v645332, v645398, v645464, v645530, v645596, v645662, v645728, v645794, v645860.

CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM

II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II,

Table 385-0001, series: v645267, v645333, v645399, v645465, v645531, v645597, v645663, v645729, v645795, v645861.

001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: v645268, v645334, v645400, v645466, v645532, v645598, v645664, v645730, v645796, v645862.

330 Social assistance expenditure. Social assistance expenditures consist of transfer payments to help individuals and families maintain a socially acceptable level of earnings. The social assistance variable is converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: v645270, v645336, v645402, v645468, v645534, v645600, v645666, v645732, v645798, v645864.

Workers’ compensation benefits. This variable includes expenditures on administration and for benefits, other than rehabilitation and medical care, related to workers’ compensation schemes. Workers’ benefits are converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: v645271, v645337, v645403, v645469, v645535, v645601, v645667, v645733, v645799, v645865.

Other social services expenditure. This variable accounts for expenses related to the provision of services to old age, to persons who are unable to lead a normal life due to a physical or mental impairment, to persons temporarily unable to work due to sickness, to households with dependent children, to persons who are survivors of a deceased person, and to other needy persons. This variable is converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002

= 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: v645274, v645340, v645406, v645472, v645538, v645604, v645670, v645736, v645802, v645868.

331

Elementary and secondary education expenditure. This expenditure encompasses outlays for educational services from kindergarten to senior matriculation.

Elementary and secondary education expenditures are converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: CANSIM II Table 385-0001, series: v645277, v645343, v645409, v645475, v645541, v645607, v645673, v645739, v645805, v645871.

332 Post-secondary education expenditure. Education spending at the post- secondary level refers to education generally obtained in universities or in degree and non-degree granting community colleges and specialized educational institutions. Post- secondary education expenditures are converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: v645278, v645344, v645410, v645476, v645542, v645608, v645674, v645740, v645806, v645872.

Special retraining services expenditure. Spending on retraining services comprises outlays made for the purpose of upgrading the skills of individuals.

Expenditures on special retraining services are converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0001, series: v645279, v645345, v645411, v645477, v645543, v645609, v645675, v645741, v645807, v645873.

Other education expenditure. This spending category covers outlays that either overlap or cannot be allocated to the other educational sub-functions. Expenditures on other education are converted into real per capita terms using population data from

CANSIM II, Table 510-001, and the CPI (2002 = 100) for each province from CANSIM

II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II,

Table 385-0001, series: v645280, v645346, v645412, v645478, v645544, v645610, v645676, v645742, v645808, v645874.

333 Provincial welfare generosity index. This is variable is constructed by averaging the standard scores (z-scores) of medical care, other social services, and post- secondary education. The sources for these variables are described above.

Union density. Union density measures the percentage of employees who belong to a trade union. Data on union density from 1976 to 1995 is retrieved from CANSIM II,

Table 279-0025, series: v810368, v810371, v810374, v810377, v810380, v810383, v810386, v810389, v810392, v810395. Data from 1997 to 2008 is obtained by dividing the number of employees covered by union agreements by total employees. Data on employees with union coverage is collected from CANSIM II, Table 282-0073, series: v3075076, v3075081, v3075086, v3075091, v3075096, v3075101, v3075106, v3075111, v3075116, v3075121. Data on total employees is collected from CANSIM II,

Table 282-0073, series: v3075021, v3075026, v3075031, v3075036, v3075041, v3075046, v3075051, v3075056, v3075061, v3075066. The missing year (1996) is occupied by taking the mean of 1995 and 1997.

Cumulative left party power. The long-term, historical strength of left political parties is measured as the percentage of cumulative left seats. This variable tabulates left seats as a proportion of seats held by all provincial government parties in each individual year and then sums these proportions for all years since 1960. Data is available from 1976 to 2008 and collected from publications available through Elections

Newfoundland, Elections Prince Edward Island, Elections Nova Scotia, Elections New

Brunswick, Chief Electoral Officer of Québec, Elections Ontario, Elections Manitoba,

Elections Saskatchewan, Elections Alberta, and Elections British Columbia.

334 Cumulative centre party power. The long-term, historical strength of centre political parties is measured as the percentage of cumulative centre seats. This variable tabulates centre seats as a proportion of seats held by all provincial government parties in each individual year and then sums these proportions for all years since 1960. Data is available from 1976 to 2008 and collected from publications available through Elections

Newfoundland, Elections Prince Edward Island, Elections Nova Scotia, Elections New

Brunswick, Chief Electoral Officer of Québec, Elections Ontario, Elections Manitoba,

Elections Saskatchewan, Elections Alberta, and Elections British Columbia.

Cumulative right party power. The long-term, historical strength of right political parties is measured as the percentage of cumulative right seats. This variable tabulates right seats as a proportion of seats held by all provincial government parties in each individual year and then sums these proportions for all years since 1960. Data on cumulative right party power is available from 1976 to 2008 and collected from publications available through Elections Newfoundland, Elections Prince Edward Island,

Elections Nova Scotia, Elections New Brunswick, Chief Electoral Officer of Québec,

Elections Ontario, Elections Manitoba, Elections Saskatchewan, Elections Alberta, and

Elections British Columbia.

Voter turnout. This variable measures voter turnout in each provincial election, in percentages of electorate that voted. Data is available from 1976 to 2008 and collected from the provincial websites, including Elections Newfoundland, Elections Prince

Edward Island, Elections Nova Scotia, Elections New Brunswick, Chief Electoral Officer of Québec, Elections Ontario, Elections Manitoba, Elections Saskatchewan, Elections

Alberta, and Elections British Columbia.

335

Cumulative women in government. The long-term, historical strength of women in government is measured as the percentage of cumulative seats held by women, irrespective of party affiliation. This variable calculates female seats as a proportion of seats held by all provincial government parties in each individual year and then sums these proportions for all years since 1960. Data is available from 1976 to 2008 and collected from the respective websites of Elections Newfoundland, Elections Prince

Edward Island, Elections Nova Scotia, Elections New Brunswick, Chief Electoral Officer of Québec, Elections Ontario, Elections Manitoba, Elections Saskatchewan, Elections

Alberta, and Elections British Columbia.

336 Provincial regimes. I created a three-category provincial political regime variable:

1) leftist (Saskatchewan, British Columbia, and Manitoba) (reference category); 2) centre-left (Ontario and Quebec), and 3) conservative (Alberta, Nova Scotia, New

Brunswick, Prince Edward Island, and Newfoundland) using hierarchical cluster methods based on cumulative left political party power and cumulative women in government.

Transfers. This variable refers to the current fiscal year dollar value of total transfers, which includes the sum of general and specific purpose transfers, received from other levels of government. Transfer expenditures are converted into real per capita terms using population data from CANSIM II, Table 510-001, and the CPI (2002

= 100) for each province from CANSIM II, Table 326-0021. Data on general transfers is available from 1989 to 2008 and collected from CANSIM II, Table 384-0001, series: v206512, v206513, v206577, v206578, v206642, v206707, v206772, v206837, v206902, v206967, v207032, v207097. Data on specific transfers is available from 1989 to 2008 and collected from CANSIM II, Table 384-0001, series: v206513, v206578, v206643, v206708, v206773, v206838, v206903, v206968, v207033, v207098.

Debt charges. This category is sub-divided into "Interest" and "Other debt charges", and refers to the amount of money owed by the province. Debt charges are converted into real per capita terms using population data from CANSIM II, Table 510-

001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data is available from 1989 to 2008 and collected from CANSIM II, Table 385-0002, series: v206522, v206587, v206652, v206717, v206782, v206847, v206912, v206977, v207042, v207107.

337 Dependency ratio. I include a variable that represents the percentage of the provincial population that is under 18 and over 65 years of age. This demographic variable is expressed as the number of “dependents” for every 100 “workers”: youth

(ages 0 to 17) + seniors (age 65 or older) per 100 workers (aged 18 to 64). Data on the population over 65 is available from 1976 to 2008 and collected from CANSIM II, Table

510-0001, series: v467001, v467316, v467631, v467946, v468261, v468576, v468891, v469206, v469521, v469836. Data on the population under 18 is available from 1976 to

2008 and collected from CANSIM II, Table 510-0001, series: v467274, v467589, v467904, v468219, v468534, v468849, v469164, v469479, v469794, v470109.

Women in labour force. This variable measures percentage of women aged 15 to

64 who are in the labor force. Data on the percentage of women in the labour force is available from 1976 to 2008 and collected from CANSIM II, Table 282-0002, series: v2461672, v2462302, v2462932, v2463562, v2464192, v2464822, v2465452, v2466082, v2466712, v2467342, v2467972.

338 Unemployment rate. This variable is the percentage of the total labour force unemployed for each province and included as a measure of the current state of the economy. Provincial data on unemployment rates is available from 1976 to 2008 and collected from CANSIM II, Table 282-0002, series: v2461854, v2462484, v2463114, v2463744, v2464374, v2465004, v2465634, v2466264, v2466894, v2467524.

1976 to 2008 and collected from CANSIM II, Table 202-0802, series: v52543638, v52548126, v52548670, v52549214, v52549758, v52550302, v52532214, v52532758, v52533302, v52533846.

339 Real GDP per capita. This variable measures average real income per person for each province, calculated by dividing GDP by population. Real GDP per capita is converted into real per capita terms using population data from CANSIM II, Table 510-

001, and the CPI (2002 = 100) for each province from CANSIM II, Table 326-0021. Data on GDP are available from 1976 to 2008 and collected from CANSIM II, Tables 384-

0002, series: v3839933, v3839979, v3840025, v3840071, v3840117, v3840163, v3840209, v3840255, v3840301, v3840347.

Urban population. This variable is measured as the percentage of the population that resides in an urban area. Data is available from 1976 to 2008 and collected from

CANSIM II, Table 153-0037, series: v32163771, v32163786, v32163801, v32163816, v32163831, v32163846, v32163861, v32163876, v32163891, v32163906.

Table 510-004, series: v391100, v391106, v391107, v391108, v391109, v391110, v391111, v391112, v391113, v391101.

Net migration. This variable refers to the net inter-provincial migration as a percentage of the population, and is calculated by dividing total net migrants by the population. Data is available from 1976 to 2008 and collected from CANSIM II, Table

510-012, series: v446476, v447421, v448366, v449311, v450256, v451201, v452146, v453091, v454036, v454981.

Income inequality. This variable refers to pre-tax-and-transfer income inequality and post-tax-and-transfer inequality, measured as the Gini coefficient for all family units.

340 Data on pre-tax-and-transfer inequality is available from 1976 to 2008 and collected from CANSIM II, Table 202-0705, market income series: v21151801, v21151909, v21152017, v21152125, v21152233, v21152341, v21152557, v21152665, v21152773, v21152881. Data on post-tax-and-transfer inequality is available from 1976 to 2008 and collected from CANSIM II, Table 202-0705, after-tax series: v21151873, v21151981, v21152089, v21152197, v21152305, v21152413, v21152629, v21152737, v21152845, v21152953.

341 Appendix B. Jackknife Results

Appendix B1. Jackknife Estimates of Aggregate Expenditures on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Total Male Female 1 2 3 4 5 6 7 8 9 Variables Health -0.0007** -0.0012** -0.0002 (-3.33) (-2.96) (-1.43) Social Services -0.0005** -0.0005 -0.0003** (-2.96) (-1.91) (-2.91) Education -0.0001 -0.0004 -0.0000 (-0.41) (-0.53) (-0.19) Transfers 0.0000 0.0000 0.0001 0.0000 0.0000 0.0002 0.0000 0.0000 0.0000 (0.28) (0.68) (0.67) (0.12) (0.48) (0.73) (0.31) (0.74) (0.32) Debt -0.0006*** -0.0004*** -0.0004*** -0.0010*** -0.0007*** -0.0006*** -0.0003*** -0.0003** -0.0003** (-4.99) (-4.12) (-3.77) (-5.08) (-4.37) (-3.82) (-3.34) (-3.12) (-2.97) Dependency Ratio 0.0324 0.0596** 0.0510* 0.0382 0.0852** 0.0720 0.0198 0.0291* 0.0249 (1.57) (2.99) (2.06) (1.18) (2.78) (1.93) (1.28) (2.03) (1.50) Female LFP -0.0304 -0.0655*** -0.0768*** -0.0524 -0.128*** -0.134*** -0.0166 -0.0180 -0.0300 (-1.31) (-3.48) (-3.42) (-1.33) (-4.30) (-3.72) (-0.97) (-1.22) (-1.91) Unemployment 0.0577* 0.0661* 0.0578* 0.0869* 0.0908* 0.0911* 0.0502** 0.0604** 0.0492* (2.31) (2.48) (2.24) (2.24) (2.18) (2.26) (2.63) (3.10) (2.51) Low income -0.0157 -0.0390* -0.0454* -0.0161 -0.0566 -0.0675* -0.0252 -0.0319* -0.0344* (-0.89) (-2.10) (-2.27) (-0.61) (-1.96) (-2.05) (-1.80) (-2.20) (-2.24) GDP per capita 0.0000 -0.0000 -0.0000 0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 (0.08) (-1.30) (-0.95) (0.46) (-1.14) (-0.68) (-0.68) (-1.18) (-1.09) Constant 7.838*** 8.475*** 9.421*** 11.43** 13.13*** 14.12*** 5.612** 5.367** 6.095** (3.34) (3.60) (3.65) (3.20) (3.57) (3.56) (3.21) (3.06) (3.30) Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product; N = 200. * p < 0.05, ** p < 0.01, *** p < 0.001

342 Appendix B2. Jackknife Estimates of Disaggregated Health Expenditures on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Total Male Female 1 2 3 4 5 6 7 8 9 10 11 12 Variables Hospital -0.0003 -0.0007 0.0002 (-0.76) (-0.70) (1.16) Medical -0.0010*** -0.0014** -0.0006** (-3.76) (-2.99) (-3.04) Preventive -0.0020 -0.0027 -0.0011 (-1.95) (-1.84) (-1.48) Other Health -0.0006 -0.0014* -0.0002 (-1.52) (-2.57) (-0.67) Transfers 0.0000 0.0000 0.0000 -0.0000 0.0000 0.0000 0.0000 -0.0000 0.0000 0.0000 0.0000 0.0000 (0.27) (0.70) (0.22) (-0.21) (0.26) (0.50) (0.19) (-0.64) (0.04) (0.84) (0.16) (0.02) Debt -0.0005*** -0.0005*** -0.0004*** -0.0005*** -0.0008*** -0.0008*** -0.0007*** -0.0008*** -0.0003** -0.0003*** -0.0003** -0.0003**

(-3.92) (-4.47) (-3.92) (-4.54) (-3.82) (-4.65) (-4.16) (-5.03) (-2.75) (-3.47) (-2.87) (-3.11) Dependency 0.0539** 0.0475* 0.0541** 0.0438 0.0784* 0.0691* 0.0806** 0.0549 0.0272 0.0222 0.0253 0.0223 Ratio (2.63) (2.56) (2.82) (1.91) (2.39) (2.40) (2.66) (1.56) (1.86) (1.57) (1.76) (1.42) Female LFP -0.0767*** -0.0382 -0.0878*** -0.0683*** -0.134*** -0.0890** -0.152*** -0.113*** -0.0345* -0.0022 -0.0349* -0.0274 (-3.85) (-1.81) (-4.63) (-3.40) (-4.21) (-2.62) (-5.20) (-3.82) (-2.32) (-0.14) (-2.33) (-1.84) Unemployment 0.0574* 0.0587* 0.0590* 0.0352 0.0883* 0.0827* 0.0841* 0.0388 0.0426* 0.0561** 0.0530** 0.0428* (2.11) (2.26) (2.20) (1.34) (1.98) (2.03) (2.01) (1.02) (2.11) (2.99) (2.64) (2.00) Low income -0.0378* -0.0214 -0.0355 -0.0343 -0.0514 -0.0339 -0.0507 -0.0422 -0.0346* -0.0191 -0.0291 -0.0314* (-2.02) (-1.14) (-1.88) (-1.87) (-1.85) (-1.18) (-1.78) (-1.52) (-2.35) (-1.26) (-1.90) (-2.23) GDP per capita -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 (-1.06) (-0.19) (-0.79) (-1.02) (-0.80) (-0.19) (-0.72) (-0.71) Constant 9.199*** 7.066** 9.607*** 9.320*** 13.68*** 11.25** 14.24*** 13.85*** 6.116*** 4.418* 6.208*** 6.104*** (3.78) (3.00) (4.08) (3.93) (3.57) (3.04) (3.84) (3.81) (3.53) (2.52) (3.54) (3.47) Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product; N = 200. * p < 0.05, ** p < 0.01, *** p < 0.001

343 Appendix B3. Jackknife Estimates of Disaggregated Social Service Expenditures on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Total Male Female 1 2 3 4 5 6 7 8 9 Variables Social Assistance -0.0000 -0.0001 -0.0000 (-0.12) (-0.27) (-0.08) Workers’ -0.0013 -0.0011 -0.0002 Compensation Benefits (-1.28) (-0.65) (-0.23) Other Social -0.0010*** -0.0011*** -0.0007*** Services (-4.80) (-3.91) (-4.21) Transfers 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.18) (0.30) (0.56) (0.18) (0.22) (0.40) (0.16) (0.19) (0.49) Debt -0.0004*** -0.0005*** -0.0005*** -0.0007*** -0.0007*** -0.0007*** -0.0003** -0.0003** -0.0003** (-3.97) (-4.20) (-4.44) (-4.14) (-4.22) (-4.71) (-3.04) (-2.95) (-3.17) Dependency Ratio 0.0554** 0.0577** 0.0699*** 0.0805* 0.0840** 0.0972*** 0.0256 0.0262 0.0381** (2.66) (2.78) (3.76) (2.47) (2.64) (3.35) (1.73) (1.77) (2.67) Female LFP -0.0823*** -0.0757*** -0.0551** -0.147*** -0.140*** -0.114*** -0.0313* -0.0299* -0.0111 (-4.29) (-3.93) (-3.10) (-4.82) (-4.63) (-3.87) (-2.16) (-2.01) (-0.79) Unemployment 0.0505 0.0561* 0.0648* 0.0750 0.0777 0.0892* 0.0479* 0.0488* 0.0588** (1.87) (2.08) (2.51) (1.83) (1.89) (2.14) (2.41) (2.49) (3.13) Low income -0.0400* -0.0368 -0.0379* -0.0565 -0.0532 -0.0543 -0.0332* -0.0326* -0.0299* (-2.07) (-1.86) (-2.12) (-1.92) (-1.77) (-1.93) (-2.21) (-2.11) (-2.07) GDP per capita -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 (-1.23) (-1.21) (-0.32) (-1.08) (-1.05) (-0.46) (-1.13) (-1.13) (-0.17) Constant 9.366*** 8.925*** 6.687** 14.20*** 13.59*** 10.99** 6.099*** 5.998*** 3.967* (3.74) (3.66) (2.93) (3.58) (3.58) (3.04) (3.36) (3.34) (2.24) Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product; N = 200. * p < 0.05, ** p < 0.01, *** p < 0.001

344 Appendix B4. Jackknife Estimates of Disaggregated Education Expenditures on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1989-2008 Total Male Female 1 2 3 4 5 6 7 8 9 10 11 12 Variables Hospital -0.0001 -0.0003 0.0000 (-0.27) (-0.40) (0.09) Medical -0.0015*** -0.0020*** -0.0009** (-4.32) (-3.66) (-3.15) Preventive -0.0004 -0.0008 0.0003 (-0.29) (-0.39) (0.28) Other Health 0.0066*** 0.0111*** 0.0048** (3.38) (3.67) (2.80) Transfers 0.0001 0.0000 0.0000 -0.0000 0.0002 0.0000 0.0000 -0.0000 0.0000 0.0000 0.0000 -0.0000 (0.43) (0.66) (0.19) (-0.15) (0.55) (0.50) (0.18) (-0.22) (0.03) (0.57) (0.17) (-0.25) Debt -0.0004*** -0.0006*** -0.0005*** -0.0004*** -0.0006** -0.0008*** -0.0007*** -0.0007*** -0.0003** -0.0004*** -0.0003** -0.0003** (-3.46) (-5.15) (-3.60) (-3.90) (-3.08) (-5.32) (-4.02) (-4.02) (-3.02) (-3.66) (-2.69) (-2.72) Dependency 0.0553** 0.0281 0.0551** 0.0476* 0.0802* 0.0466 0.0817** 0.0700* 0.0260 0.00946 0.0256 0.0201 Ratio (2.62) (1.35) (2.76) (2.42) (2.44) (1.46) (2.65) (2.23) (1.74) (0.56) (1.76) (1.41) Female LFP -0.0810*** -0.0570*** -0.0811*** -0.0941*** -0.143*** -0.111*** -0.143*** -0.165*** -0.0310* -0.0160 -0.0318* -0.0405** (-4.21) (-3.39) (-4.19) (-4.84) (-4.63) (-4.19) (-4.85) (-5.26) (-2.13) (-1.18) (-2.13) (-2.74) Unemployment 0.0548* 0.0461* 0.0498 0.0496 0.0866* 0.0703* 0.0724 0.0710 0.0471* 0.0434* 0.0487* 0.0470* (2.02) (1.99) (1.85) (1.90) (2.03) (1.98) (1.75) (1.76) (2.32) (2.31) (2.43) (2.46) Low income -0.0429* -0.0366* -0.0409* -0.0391* -0.0642 -0.0512 -0.0572 -0.0532 -0.0327* -0.0307* -0.0332* -0.0318* (-2.11) (-2.12) (-2.15) (-2.12) (-1.93) (-1.92) (-1.95) (-1.94) (-2.14) (-2.21) (-2.24) (-2.16) GDP per capita -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000 (-1.03) (-1.26) (-1.27) (-1.03) (-0.74) (-1.02) (-1.09) (-0.83) (-1.15) (-1.21) (-1.15) (-0.93) Constant 9.261*** 10.53*** 9.344*** 10.29*** 13.85*** 15.33*** 13.91*** 15.41*** (3.40) (3.73) (3.44) (4.01) (3.75) (4.53) (3.91) (4.32) (3.56) (4.24) (3.76) (4.02) Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product; N = 200. * p < 0.05, ** p < 0.01, *** p < 0.001

345

Appendix B5. Jackknife Estimates of Provincial Welfare Generosity on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1989- 2008 Total Male Female 1 2 3 Variables Welfare Generosity -0.391*** -0.487*** -0.259*** (-6.89) (-5.91) (-5.20) Transfers 0.0001 0.0001 0.0001 (0.87) (0.59) (1.04) Debt -0.0006*** -0.0008*** -0.0004*** (-5.32) (-5.65) (-3.91) Dependency Ratio 0.0477** 0.0705** 0.0232 (2.95) (2.75) (1.71) Female LFP -0.0175 -0.0651* 0.0128 (-1.03) (-2.27) (0.91) Unemployment 0.0672** 0.0930* 0.0594*** (2.86) (2.42) (3.38) Low income -0.0247 -0.0389 -0.0200 (-1.49) (-1.51) (-1.42) GDP per capita 0.0000 0.0000 0.0000 (0.25) (0.11) (0.14) Constant 4.743* 8.378* 2.764 (2.29) (2.51) (1.64) Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product; N = 200. * p < 0.05, ** p < 0.01, *** p < 0.001

346 Appendix B6. Jackknife Estimates of Power Resources on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Total Male Female 1 2 3 4 5 6 7 8 9 10 11 12 Variables Welfare -0.392*** -0.346*** -0.489*** -0.401*** -0.260*** -0.234*** Generosity (-6.93) (-6.42) (-5.99) (-5.10) (-5.18) (-4.73) Union Density 0.0006 -0.0131 0.0078 -0.0152 0.0001 0.0019 (0.06) (-0.97) (0.57) (-0.73) (0.01) (0.16) Left Political Party -0.0737*** -0.0615*** -0.103*** -0.122*** -0.0544*** -0.0321* Power (-6.58) (-3.69) (-5.06) (-4.52) (-10.45) (-2.59) Transfers 0.0001 0.0001 0.0001 0.0000 0.0001 0.0000 (0.86) (0.86) (0.59) (0.51) (1.05) (0.95) Debt -0.0007*** -0.0005*** -0.0009*** -0.0007*** -0.0004** -0.0003*** (-4.63) (-5.18) (-4.47) (-5.40) (-2.93) (-3.82) Dependency -0.0089 0.0533*** -0.0020 0.0541*** -0.0182 0.0767** -0.0087 0.0826*** -0.00781 0.0224 -0.0009 0.0266* Ratio (-0.82) (3.55) (-0.21) (3.44) (-0.96) (3.17) (-0.59) (3.41) (-1.09) (1.66) (-0.14) (1.99) Female LFP -0.145*** -0.0167 -0.120*** -0.0007 -0.215*** -0.0635* -0.180*** -0.0325 -0.0926*** 0.0128 -0.0725*** 0.0218 (-13.22) (-1.00) (-12.47) (-0.04) (-11.10) (-2.26) (-11.70) (-1.25) (-13.43) (0.90) (-11.00) (1.46) Unemployment 0.0227 0.0698** 0.0194 0.0618** 0.0630** 0.0978* 0.0652*** 0.0844* 0.00464 0.0592*** -0.0000 0.0553** (1.66) (3.01) (1.79) (2.97) (2.94) (2.52) (3.71) (2.54) (0.43) (3.36) (-0.00) (3.31) Low income -0.0353* -0.0300 -0.0320* -0.0262 -0.0492* -0.0449 -0.0489* -0.0408 -0.0328** -0.0192 -0.0292** -0.0210 (-2.41) (-1.61) (-2.36) (-1.67) (-2.18) (-1.56) (-2.36) (-1.62) (-2.67) (-1.19) (-2.60) (-1.53) GDP per capita -0.0000* -0.0000 -0.0000** -7.54e-08 -0.0000 -0.0000 -0.0000 -0.0000 -0.0000* 0.0000 -0.0000** -0.0000 (-2.49) (-0.12) (-3.29) (-0.01) (-1.10) (-0.14) (-1.66) (-0.16) (-2.33) (0.18) (-3.00) (-0.06) Constant 17.24*** 4.961* 15.59*** 3.677 23.55*** 8.554* 21.50*** 6.247* 12.43*** 2.730 10.96*** 2.216 (12.72) (2.28) (13.18) (1.89) (9.89) (2.53) (11.25) (2.11) (15.10) (1.50) (14.07) (1.32) N 330 200 330 200 330 200 330 200 330 200 330 200 Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product. * p < 0.05, ** p < 0.01, *** p < 0.001

347 Appendix B7. Jackknife Estimates of Left and Right Political Party Power on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Total Male Female 1 2 3 4 5 6 7 8 9 10 11 12 Variables Welfare -0.368*** -0.286*** -0.446*** -0.304** -0.222*** -0.165* Generosity (-5.68) (-4.51) (-4.72) (-3.27) (-3.74) (-2.60) Centre Political -0.0403*** -0.0093 -0.0648*** -0.0184 -0.0263*** -0.0129 Party Power (-7.52) (-0.93) (-7.93) (-1.18) (-6.87) (-1.50) Right Political 0.0451*** 0.0305** 0.0692*** 0.0590*** 0.0305*** 0.0244** Party Power (10.89) (3.26) (10.05) (3.93) (10.75) (3.04) Transfers 0.0001 0.0001 0.0001 0.0000 0.0001 0.0000 (0.88) (0.89) (0.59) (0.54) (1.02) (0.77) Debt -0.0005*** -0.0004** -0.0007*** -0.0005** -0.0003** -0.0002 (-4.19) (-3.17) (-4.23) (-2.88) (-2.72) (-1.92) Dependency -0.0133 0.0443** -0.00898 0.0381* -0.0233 0.0637* -0.0168 0.0517* -0.00961 0.0185 -0.00576 0.0160 Ratio (-1.43) (2.68) (-1.06) (2.36) (-1.50) (2.38) (-1.20) (2.05) (-1.46) (1.35) (-0.93) (1.15) Female LFP -0.155*** -0.0280 -0.140*** -0.0465* -0.228*** -0.0833* -0.204*** -0.112*** -0.0977*** -0.00416 -0.0867*** -0.0144 (-17.56) (-1.26) (-17.70) (-2.40) (-15.55) (-2.29) (-15.43) (-3.60) (-15.69) (-0.21) (-14.83) (-0.81) Unemployment 0.0093 0.0659** 0.0035 0.0598** 0.0472** 0.0910* 0.0407* 0.0830* -0.00587 0.0566** -0.0113 0.0507** (0.85) (2.79) (0.37) (2.70) (2.67) (2.37) (2.55) (2.37) (-0.70) (3.16) (-1.45) (2.83) Low income -0.0345** -0.0215 -0.0312* -0.0144 -0.0490* -0.0336 -0.0442* -0.0222 -0.0320** -0.0146 -0.0285** -0.00992 (-2.62) (-1.20) (-2.50) (-0.86) (-2.49) (-1.22) (-2.34) (-0.88) (-3.10) (-0.93) (-2.93) (-0.67) GDP per capita -0.0000** 0.0000 -0.0000*** 0.0000 -0.0000 0.0000 -0.0000 0.0000 -0.0000** 0.0000 -0.0000** 0.0000 (-3.20) (0.37) (-3.70) (0.53) (-1.43) (0.24) (-1.67) (0.38) (-2.73) (0.39) (-3.09) (0.48) Constant 18.38*** 5.589* 12.94*** 4.224* 25.29*** 9.901* 16.93*** 6.839* 13.00*** 4.058* 9.195*** 2.555 (16.03) (2.42) (10.59) (2.13) (13.01) (2.60) (8.08) (2.26) (17.25) (2.15) (11.66) (1.50) N 330 200 330 200 330 200 330 200 330 200 330 200 Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product. * p < 0.05, ** p < 0.01, *** p < 0.001

348 Appendix B8. Jackknife Estimates of Voter Turnout and Women in Government on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Total Male Female 1 2 3 4 5 6 7 8 9 10 11 12 Variables Welfare -0.353*** -0.292*** -0.421*** -0.296*** -0.221*** -0.212*** Generosity (-5.89) (-4.98) (-4.88) (-3.37) (-4.12) (-3.98) Voter Turnout 0.0147*** 0.0083* 0.0230*** 0.0152* 0.0096** 0.0080* (3.33) (1.99) (3.38) (2.43) (2.96) (2.06) Women in -0.136*** -0.0953*** -0.224*** -0.189*** -0.0662*** -0.0431* Government (-5.65) (-4.26) (-5.49) (-5.34) (-5.80) (-2.45) Transfers 0.0001 0.0001 0.0001 0.0000 0.0001 0.0001 (0.76) (0.81) (0.48) (0.48) (0.89) (0.98) Debt -0.0006*** -0.0005*** -0.0009*** -0.0006*** -0.0004*** -0.0003*** (-5.47) (-4.40) (-5.95) (-4.04) (-4.07) (-3.53) Dependency -0.0054 0.0480** 0.0007 0.0325* -0.0120 0.0707** -0.0028 0.0393 -0.0056 0.0235 -0.0026 0.0164 Ratio (-0.50) (2.92) (0.06) (2.06) (-0.67) (2.76) (-0.16) (1.61) (-0.76) (1.70) (-0.36) (1.19) Female LFP -0.136*** -0.0197 -0.0871*** -0.0034 -0.200*** -0.0676* -0.120*** -0.0366 -0.0872*** 0.0102 -0.0631*** 0.0187 (-12.45) (-1.17) (-5.33) (-0.20) (-10.87) (-2.41) (-4.52) (-1.40) (-12.48) (0.74) (-6.93) (1.31) Unemployment 0.0135 0.0611** 0.0049 0.0528* 0.0536** 0.0842* 0.0400* 0.0677 -0.00190 0.0522** -0.0057 0.0521** (1.08) (2.60) (0.39) (2.38) (2.66) (2.21) (2.07) (1.90) (-0.19) (2.98) (-0.59) (2.96) Low income -0.0312* -0.0243 -0.0220 -0.0121 -0.0457* -0.0385 -0.0323 -0.0168 -0.0301* -0.0199 -0.0247* -0.0137 (-2.24) (-1.47) (-1.58) (-0.77) (-2.20) (-1.49) (-1.56) (-0.71) (-2.50) (-1.40) (-2.09) (-1.01) GDP per capita -0.0000* 0.0000 -0.0000 -0.0000 -0.0000 0.0000 -0.0000 -0.0000 -0.0000* 0.0000 -0.0000* -0.0000 (-2.28) (0.32) (-1.88) (-0.20) (-0.92) (0.20) (-0.69) (-0.38) (-2.07) (0.20) (-2.09) (-0.13) Constant 15.60*** 4.459* 15.15*** 6.300** 21.08*** 7.760* 20.41*** 11.45*** 11.38*** 2.528 11.29*** 3.511* (11.04) (2.13) (10.49) (3.11) (9.08) (2.33) (8.72) (3.58) (12.23) (1.46) (12.98) (2.03) N 330 200 330 200 330 200 330 200 330 200 330 200 Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product. * p < 0.05, ** p < 0.01, *** p < 0.001

349 Appendix B9. Jackknife Estimates of Provincial Regimes and Provincial Welfare Generosity on Total, Male, and Female Age-Standardized Mortality Rates in Canadian Provinces, 1976-2008 Total Males Female 1 2 3 4 5 6 7 8 9 Variables Leftist Regime (ref) Centre-Left Regime 0.422 0.235 0.321** 0.703 0.547* 0.694** 0.247 0.0666 0.127 (1.51) (1.81) (2.65) (1.54) (2.06) (2.83) (1.93) (0.82) (1.47) Conservative Regime 0.341 0.217 0.263* 0.444 0.490* 0.559** 0.218* 0.0422 0.0815 (1.87) (1.86) (2.59) (1.56) (2.14) (2.74) (2.35) (0.55) (1.11) Welfare Generosity -0.314*** -0.325*** -0.409*** -0.431*** -0.183** -0.182** (-4.77) (-4.21) (-4.68) (-3.49) (-3.22) (-2.85) Centre-Left * -0.170 -0.280 -0.123 Welfare Generosity (-1.85) (-1.89) (-1.80) Conservative * 0.167* 0.262* 0.111* Welfare Generosity (2.38) (2.09) (1.98) Transfers 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.75) (0.86) (0.43) (0.51) (0.96) (1.01) Debt -0.0000 -0.0001 -0.0002 -0.0004 0.0002 0.0001 (-0.08) (-0.86) (-0.98) (-1.67) Dependency Ratio 0.0319* 0.0311*** 0.0377*** 0.0480* 0.0632*** 0.0731*** 0.0125 0.00565 0.0102 (2.58) (3.56) (4.57) (2.34) (3.89) (4.81) (1.91) (0.89) (1.54) Female LFP -0.0952*** -0.0619*** -0.0626*** -0.131*** -0.100*** -0.0996*** -0.0713*** -0.0359** -0.0360*** (-6.77) (-4.39) (-4.74) (-6.04) (-4.62) (-5.04) (-8.31) (-3.33) (-3.42) Unemployment -0.0055 0.0414* 0.0384* 0.0144 0.0436 0.0396 -0.0101 0.0393** 0.0366** (-0.35) (2.25) (2.23) (0.57) (1.57) (1.51) (-0.83) (3.06) (2.87) Low income -0.0353* -0.0098 -0.0092 -0.0583* -0.0169 -0.0150 -0.0292* -0.0145 -0.0139 (-1.98) (-0.51) (-0.50) (-2.09) (-0.63) (-0.60) (-2.11) (-0.86) (-0.86) GDP per capita -0.0000 0.0000** 0.0000 -0.0000 0.0000 -0.0000 -0.0000 0.0000*** 0.0000* (-1.11) (2.62) (1.33) (-0.52) (1.03) (-0.06) (-0.76) (4.35) (2.55) Constant 10.85*** 7.418*** 7.381*** 13.87*** 9.794*** 9.633*** 9.111*** 5.953*** 5.934*** (6.83) (6.42) (7.06) (5.24) (4.94) (5.41) (11.25) (7.20) (7.50) N 330 200 200 330 200 200 330 200 200 Notes: b = unstandardized coefficient; z-scores are in parentheses; LFP = labour force participation; GDP = gross domestic product. * p < 0.05, ** p < 0.01, *** p < 0.001

350