UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The growth of social assistance receipt in Canada 2000

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
ubc_2000-566269.pdf [ 7.67MB ]
Metadata
JSON: 1.0089823.json
JSON-LD: 1.0089823+ld.json
RDF/XML (Pretty): 1.0089823.xml
RDF/JSON: 1.0089823+rdf.json
Turtle: 1.0089823+rdf-turtle.txt
N-Triples: 1.0089823+rdf-ntriples.txt
Citation
1.0089823.ris

Full Text

THE GROWTH OF SOCIAL ASSISTANCE RECEIPT IN C A N A D A by Alan A. Stark B.A., The University of Calgary, 1988 M.A., The University of Guelph, 1991 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF GRADUATE STUDIES (Department of Economics) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA August, 2000 © Alan A. Stark, 2000 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department The University of British Columbia Vancouver, Canada DE-6 (2/88) Abstract The research undertaken in this thesis examines social assistance (welfare) receipt in Canada during the 1981-95 period to determine the forces responsible for the dramatic growth in welfare use observed during the 1990s. The influence of changes in welfare benefits, labour market conditions, and the availability of unemployment insurance on welfare use during this period is examined using two distinct, but complementary approaches. The first approach investigates this issue from an aggregate standpoint, using Survey of Consumer Finances micro data to construct welfare usage rates for employable singles without children (male and female) and lone mothers. Separate analyses are performed for each of these sub-groups using aggregate province level data. The second approach attacks the issue from a microeconomic standpoint, employing duration analysis to examine the path leading individuals from employment to welfare receipt. Using the 1988-90 longitudinal file of the Labour Market Activity Survey, semi-parametric duration models are estimated to determine how the job loss, re- employment and welfare take-up processes are affected by incentives in welfare benefits, labour market conditions, availability of unemployment insurance as well as demographic variables. The estimates from the duration analysis are applied to administrative data on inflows of persons into the pool of non-employed to simulate and decompose rates of welfare incidence over the 1984-95 period. Results from these two approaches present a relatively consistent picture of welfare use in Canada during the 1990s. Both approaches find strong evidence of important labour market effects. Thus, the economic downturn of the early 1990s played a significant role in the growth of welfare use during this period, particularly in Ontario and Quebec. The evidence concerning the importance of interactions with the unemployment insurance system and changes in benefit generosity is mixed. Both UI effects and benefit effects are found to be important determinants of welfare use but only among specific types of families. The simulation results indicate these factors can account for only a minor amount of the variation in predicted welfare incidence in the 1990s. ii Table of Contents Page Abstract ii Table of Contents iii List of Tables v List of Figures vi Acknowledgement ix Chapter 1 Introduction and Literature Review 1 Chapter 2 Institutional Framework 7 2.1 The Application Process 8 2.2 Reforms to the Welfare System 8 2.3 Parameterization of the Welfare System 11 Chapter 3 Aggregate Analysis of the Growth of Welfare Receipt in Canada 12 3.1 Introduction 12 3.2 Framework for the Analysis 12 3.3 Data 14 3.3.1 Definitions of Types of Recipients and Comparison with 15 Administrative Data 3.3.2 Selection and Description of Final Sample 17 3.4 Estimation Approach 18 3.5 Results 19 3.5.1.1 Aggregate Approach - Graphical Results 20 3.5.1.2 Aggregate Approach - Regression Results 21 3.5.2.1 Disaggregate Approach - Graphical Results 24 3.5.2.2 Disaggregate Approach - Regression Results 26 3.5.2.3 Decomposition of Regression Results 28 3.6 Conclusions 30 Chapter 4 Duration Analysis of the Transition from Employment to Welfare 34 4.1 Introduction 34 4.2 Theoretical Framework 36 4.3 Data 40 4.3.1 L M A S 40 4.3.2 Identification of States 41 4.3.3 Exclusions from the Sample 44 4.3.4 Selection of Covariates 45 iii Page 4.4 Descriptive Statistics 47 4.4.1 Job Spells 47 4.4.2 Non-employment Spells 48 4.5 Empirical Specification and Empirical Hazard Functions 49 4.5.1 Empirical Specification 49 4.5.2 Empirical Hazard Functions 51 4.6 Duration Models 53 4.6.1 Baseline Hazards 53 4.6.2 Marginal Effects 54 4.7 Conclusions 57 Chapter 5, Simulation of Welfare Incidence 59 5.1 Introduction 59 5.2 Methodology 59 5.3 Data 61 5.4 Results 65 5.4.1 Description of Predicted Incidence Rates 66 5.4.2 Decomposition of Predicted Incidence Rates 68 5.4.2.1 Decomposition of Predicted Incidence Rates - Policy Variables 71 5.4.2.2 Decomposition of Predicted Incidence Rates - Macro Variables 72 5.4.3 Comparison of Predicted Incidence Rates and Actual Caseloads 73 5.5 Conclusions 77 Chapter 6 Conclusions 81 Bibliography 171 Appendix 1: Data 174 Appendix 2: Supplements to Chapter 4 - Duration Analysis 178 Appendix 3: Supplements to Chapter 5 - Simulation of Welfare Incidence 195 iv List of Tables Table Title Page Table 2.1: Exemption Limits for Non-disabled Welfare Recipients 86 - Selected Provinces Table 3.1: Age Distributions 88 Table 3.2: Education Distributions 89 Table 3.3: Weeks Worked Distributions 90 Table 3.4: SA Usage - Aggregate Approach 91 Table 3.5a: SA Usage - Singles without Children - Male 92 Table 3.5b: SA Usage - Singles without Children - Female 93 Table 3.5c: SA Usage - Singles with Children - Female 94 Table 3.6: Percentage Reduction in Variation of Predicted Usage Series 95 due to Restricting Covariates Table 4.1a Unweighted Sample Counts by Outcome and Family Type 96 - Job Spells Table 4. lb Unweighted Sample Counts by Outcome and Family Type 97 - Non-employment Spells Table 4.2a: Spell Distribution by Outcome and Family Type 98 - Job Spells Table 4.2b: Spell Distribution by Outcome and Family Type 99 - Non-employment Spells Table 4.3: Duration Model Estimates 100 Table 5.1 Percentage Reduction in Variation of Predicted Incidence Series 101 due to Restricting Covariates Table A2.1 Unweighted Sample Counts by Family Type - Non-employment 179 Spells ending in Welfare Take-up Table A2.2 Spell Distributions and Average Duration by Family Type - 180 Non-employment Spells ending in Welfare Take-up v List of Figures Figure Title Page Fig. 1.1: S A Cases as a Proportion of Households - March - Canada 102 Fig. 3.1a: SA Cases - Singles Without Children - SCF vs. Admin., All 103 Individuals Under 65 Years of Age Fig. 3.1b: SA Cases - Singles With Children - SCF vs. Admin., All 105 Individuals Under 65 Years of Age Fig. 3.2: SA Cases as a Proportion of Households - Aggregate Approach 107 Fig. 3.3: Unemployment Rate - Men, 25-44 years 110 Fig. 3.4: S A Expend per Case as a Proportion of Min Wage Earnings 113 Fig. 3.5: Min. Weeks of Employment Required to Qualify for UI 116 Fig. 3.6: Year Effects 119 Fig. 3.7a: SA Benefits as a Proportion of Minimum Wage Earnings, Singles 120 without Children - Male (SNM) Fig. 3.7b: SA Benefits as a Proportion of Minimum Wage Earnings, Singles 122 without Children - Female (SNF) Fig. 3.7c: SA Benefits as a Proportion of Minimum Wage Earnings, Singles 124 with Children - Female (SPF) Fig. 3.8a: Average Weeks Worked, Singles without Children - Male (SNM) 126 Fig. 3.8b: Average Weeks Worked, Singles without Children - Female (SNF) 128 Fig. 3.8c: Average Weeks Worked, Singles with Children - Female (SPF) 130 Fig. 3.9a: Minimum Weeks Required to Qualify for UI Singles without 132 Children - Male (SNM) Fig. 3.9b: Minimum Weeks Required to Qualify for UlSingles without 134 Children - Female (SNF) Fig. 3.9c: Minimum Weeks Required to Qualify for UlSingles with 136 Children - Female (SPF) Fig. 3.10a: Year Effects, Singles without Children - Male (SNM) 138 Fig. 3.10b: Year Effects - Weeks Worked for less than 35 Yrs., Singles 139 without Children - Male (SNM) - Low Education Fig. 3.11 a: Year Effects, Singles without Children - Female (SNF) 140 Fig. 3.11 b: Year Effects, Singles without Children - Female (SNF) - Low 141 Education Fig. 3.12a: Year Effects, Singles with Children - Female (SPF) 142 Fig. 3.12b: Year Effects, Singles with Children - Female (SPF) - children 143 less than 7 years Fig. 3.12c: Year Effects, Singles with Children - Female (SPF) - children 144 7-17 years Fig. 4.1a: Empirical Hazard Function - Job Spells 145 Fig. 4.1b: Survival Function - Job Spells 146 Fig. 4.2a: Empirical Hazard Function - Time until Re-employment 147 Fig. 4.2b: Survival Hazard Function - Time until Re-employment 148 VI Figure Title Page Fig. 4.3a: Empirical Hazard Function - Time until Welfare 149 Fig. 4.3b: Survival Hazard Function - Time until Welfare 150 Fig. 4.4: Baseline Hazard Function - Job Spells 151 Fig. 4.5: Baseline Hazard Function - Time until Re-employment 152 Fig. 4.6: Baseline Hazard Function - Time until Welfare 153 Fig. 5.1a: Decomposition of Predicted Incidence, 1984-95, Singles 154 without Children (SN) Fig. 5.1b: Decomposition of Predicted Incidence, 1984-95, Couples 155 without Children (CN) Fig. 5.1c: Decomposition of Predicted Incidence, 1984-95, Couples 156 with Children (CP) Fig. 5.2a: Decomp.of Predicted Incidence - Policy Variables, 1984-95, 157 Singles without Children (SN) Fig. 5.2b: Decomp.of Predicted Incidence - Policy Variables, 1984-95, 158 Couples without Children (CN) Fig. 5.2c: Decomp.of Predicted Incidence - Policy Variables, 1984-95, 159 Couples with Children (CP) Fig. 5.3a: Decomp.of Predicted Incidence - Macro Variables, 1984-95, 160 Singles without Children (SN) Fig. 5.3b: Decomp.of Predicted Incidence - Macro Variables, 1984-95, 161 Couples without Children (CN) Fig. 5.3c: Decomp.of Predicted Incidence - Macro Variables, 1984-95, 162 Couples with Children (CP) Fig. 5.4a: Newly Non-employed, 1984-95, Singles without Children (SN) 163 Fig. 5.4b: Newly Non-employed, 1984-95, Couples without Children (CN) 164 Fig. 5.4c: Newly Non-employed, 1984-95, Couples with Children (CP) 165 Fig. 5.5: Probability of Welfare Take-up - Reference Individual 166 Fig. 5.6a: Predicted Incidence and Actual Cases, 1984-95, Singles 167 without Children (SN) Fig. 5.6b: Predicted Incidence and Actual Cases, 1984-95, All Couples (C) 168 Fig. 5.7a: Derived Average Spell Duration, 1984-95, Singles 169 without Children (SN) Fig. 5.7b: Derived Average Spell Duration, 1984-95, All Couples (C) 170 Fig. A2.1a: Monthly SA Benefits per Adult Equivalent ($1990), Singles 181 without Children (SN) Fig. A2.1b: Monthly SA Benefits per Adult Equivalent ($1990), Couples 183 without Children (CN) Fig. A2.1c: Monthly SA Benefits per Adult Equivalent ($1990), 185 Couples with Children (CP) Fig. A2.2: U E Rate, Men 25-44 187 Fig. A2.3: Min. Weeks of Employment Required to Qualify for UI 189 Fig. A2.4: Weeks of UI Benefits given 20 Weeks Employment 191 vii Figure Title Page Fig. A2.5: Monthly Minimum Wage Earnings 193 Fig. A3.la: SA Benefits per Adult Equivalent, 1981-95, Singles 196 without Children (SN) Fig. A3.lb: SA Benefits per Adult Equivalent, 1981-95, Couples 197 without Children (CN) Fig. A3.1c: SA Benefits per Adult Equivalent, 1981-95, Couples 198 with Children (CP) Fig. A3.2: UI Entitlement Weeks given 20 Weeks Employment, 1981-95 199 Fig. A3.3: EP Rate, Persons 20-34 years, 1981-95 200 viii Acknowledgement During the course of researching and writing this thesis I have received assistance and wise counsel from numerous sources. First and foremost I thank my supervisor, David Green, for his support throughout this endeavour. Without his continued encouragement and insights this work would have suffered greatly. I would also like to thank the members of my committee, Craig Riddell and Jon Kesselman, as well as Paul Beaudry for numerous discussions and suggestions. Additionally, I am very grateful to Gilles Seguin who very generously and patiently shared with me his limitless knowledge regarding welfare programs in Canada. I also wish to thank Anne Tweddle and Stephane Roller for their assistance in acquiring data. The financial support of the Canadian International Labour Network is gratefully acknowledged. Lastly, I wish to thank my family, and in particular my mother, for the patience and support they have exhibited throughout my studies. ix Chapter 1 Introduction and Literature Review The last 15 years in Canada have witnessed a dramatic increase in social assistance (welfare) receipt with particularly large increases in the period beginning with the 1990's. As seen in Figure 1.1, the percentage of households in Canada receiving social assistance grew from 8.5 percent in 1982 to 10.8 percent in 1985, moderately declined to 9.6 percent in 1989 and then increased sharply to a peak of 14.3 percent in 1994.1 As one might expect, this phenomenon has attracted the attention of policy makers, and several provincial governments, notably Alberta and Ontario, have introduced major reforms in their welfare programs. This is not surprising given the increases in program costs that accompanied the increased caseloads. However, these reforms have been implemented without a thorough understanding of the forces responsible for this upsurge. Unlike the U.S. experience [see Moffitt (1992)], these issues have not received much attention in the academic literature. In recent years the lack of research into Canadians' involvement with welfare has begun to be reversed. There is limited work utilizing annual aggregate time series data to understand this upsurge [Brown (1995), Fortin and Cremieux (1998)]. It has been proposed by policy makers and others that the observed increase in welfare usage can be explained by changes in welfare benefit generosity and unemployment conditions. An additional candidate is spillover effects from tightening of the unemployment insurance program. It is these issues which concern Brown (1995) and Fortin and Cremieux (1998).2 Brown's analysis relies primarily on graphical techniques but includes regressions for Alberta, B.C. and Ontario using quarterly data. These regressions are estimated for periods characterized as (i) welfare enrollment peaks and troughs, and (ii) benefit level peaks and troughs. The specific sample periods vary for each province but are all within the 1981-1993 time frame. The models estimated are specified in first differences using a caseload usage rate as the dependent variable which is regressed on an unemployment 1 See Appendix for description of data sources. 2 Fortin and Cremieux consider UI effects, but Brown does not. 3 Regressions are estimated for total cases, and when possible, single employables alone. 1 rate and real benefit levels. The regression coefficients from these models are used to decompose the observed changes in welfare usage and imply that on average across provinces and categories, unemployment explains 22.5% of the change in caseloads for the period examined [Brown (1995), page 83]. With respect to benefit rate effects, the results are less determinate, but in the case of Ontario benefit changes account for 16% of the increase in caseloads [Brown (1995), page 89]. Fortin and Cremieux (1998) extend this work using annual aggregate data for Quebec, Ontario, Alberta and British Columbia for the 1977-1996 time period. They regress the proportion of the population comprised by welfare beneficiaries on independent variables including an unemployment rate, measures of UI availability and generosity, the minimum wage and welfare benefit levels.4 They conclude that social assistance usage has been significantly affected by employment conditions with changes in the unemployment rate reflected one-for-one in the usage rate. Results indicate that usage has also been affected by reforms to the UI program, increasing caseloads as much as 25%. Rising benefit levels also significantly contributed to the growth in caseloads. The increase in benefit levels for Ontario over the 1985-1994 decade is associated with an increase of up to 22% in the caseloads in that province. The authors further find that increases in the minimum wage appear to increase participation in social assistance. However, these studies utilize data that groups together all individuals regardless of family type or employability status.5 Stewart and Dooley (1998a) investigate social assistance usage in Ontario using a panel of region level data for 1983-94. Separate estimations are performed for single males, single females, lone mothers and couples with children. They find that welfare benefits are only positively and significantly related to welfare participation for single men. Furthermore, their results indicate that adequacy of the UI program does not have a significant impact on usage for any group. Labour market conditions are found to play a significant role in use of welfare, but in most cases account for little of the increase in usage. The probability of welfare receipt at a point in time has also been investigated 4 Welfare benefits are constructed as average SA expenditures per beneficiary (including dependents). 2 using microdata. Allen (1993) examines the effect of welfare on family structure including the probability of welfare receipt. Using a sample of low-income women from the 1986 census he finds that the probability of welfare receipt is positively related to benefit generosity and the level of liquid asset exemption levels, while negatively related to the level of education. Charette and Meng (1994) employ a sample of female heads of households from the 1989 Labour Market Activity Survey (LMAS). The authors find that the probability of receipt is positively related to the level of benefits, the level of the earned income exemption and the level of education. There is some evidence that the implicit tax rate negatively affects participation. Also included as a covariate is the provincial unemployment rate but this does not have a significant impact. Christofides et al. (1997) investigate the joint welfare participation-labour supply decisions made by single males, single females, lone fathers and lone mothers using data from the 1988-89 LMAS. They find that that these decisions are not independent and welfare program parameters, such as the basic allowance and program tax rate, generally do not influence these decisions. The probability of being on welfare in a given year is the result of two underlying dynamic processes, namely, the time spent on welfare and movement onto welfare. Substantial effort has been invested in studying time spent on welfare. Previous studies have examined the dynamics of welfare use in B.C. [Barrett (1996), Bruce et al. (1996), M . Cragg (1996), Barrett and M . Cragg (1998)] and Quebec [Duclos et al. (1996), B. Fortin et al (1997), B. Fortin and Lacroix (1998)]. This issue has also been investigated using data for Ontario [Dooley and Stewart (1998b)] and New Brunswick [Barrett (1998)]. None of these studies examine the path leading up to welfare receipt.6 To the author's knowledge, the only existing Canadian study examining involvement with welfare following a job separation is Browning, Jones and Kuhn (1995). Browning et al. use data from the Canadian Out of Employment Panel (COEP) to study the interaction between UI reforms and welfare. As a result of the UI reforms implemented by Bill C- 5 Brown estimates regressions for total cases, and when possible, single employables alone. 6 B. Fortin et al (1997) and Stewart and Dooley (1998a) examine the duration of off-welfare spells for lone mothers in Quebec and Ontario respectively, but do not examine the direct path leading up to SA receipt. 3 113 in April 1993 most individuals had their UI benefits cut from 60 to 57% of insurable earnings. Additionally, individuals who either voluntarily quit without cause or were dismissed were no longer entitled to UI benefits. The first cohorts of the COEP data span this date affording an opportunity to utilize this policy change as a natural experiment. The first part of their study focuses on voluntary quits who may have been disentitled because of Bill C- l 13, comparing UI and welfare take up rates before and after the change in policy. Their results indicate that the availability of SA served to mitigate the incentive and income distribution effects of the policy change implying that UI and SA programs may act as close substitutes. The second part of their study focuses on UI exhaustees to see if their post-separation behaviour is significantly affected by incentives present in SA benefits. They employed a sample of UI claimants who were in the first post-separation spell of insured unemployment at the time of their second interview (40 weeks after job separation) and who were scheduled to exhaust their UI eligibility prior to interview three (60 weeks after job separation). Regression analysis is applied to examine a range of outcomes at the time of the third interview (re-employed, recalled, S A and neither employed or SA) to see how they vary with demographic characteristics and economic variables. They find that welfare benefits do not have a significant role in determining the probability of re-employment while labour market conditions (represented by the unemployment rate) have a pro-cyclical effect. Furthermore, SA benefits do not significantly affect the probability of SA receipt. Labour market conditions are seen to have a pro-cyclical effect on the probability of SA receipt in some specifications, however when controls for education and province are included this effect is essentially zero. The research agenda pursued in this thesis investigates the growth of welfare receipt in Canada using three sets of exercises. Chapter 2 presents a brief description of the institutional framework of the welfare system in Canada and selected reforms that occurred during the 1981-95 period. In chapter 3, the growth of welfare receipt is examined within an aggregate framework. This is closest in spirit to the work of Fortin and Cremieux (1998) and Stewart and Dooley (1998a). I conduct two separate sets of estimations. In the first, I examine annual aggregate data for a panel of 9 provinces 4 (excluding Prince Edward Island) for the 1982-96 period to determine the degree to which observed patterns in social assistance usage can be explained by changes in welfare benefit generosity, labour market conditions and the availability of unemployment insurance. In particular, the analysis revealed evidence of composition effects in the benefit rate measure. That is, measures of benefit rates aggregated over all categories of social assistance recipients are affected by the changing composition of the total social assistance caseload as well as changes in the benefit schedule. These concerns are addressed in the second set of estimations. I examine annual aggregate data for a panel of five provinces from 1981 to 1995 and use data from the Survey of Consumer Finances to construct welfare usage rates and measures of labour market conditions specific to sub-groups of individuals. Benefit rates specific to class of recipient are obtained from published rate schedules. This approach allows for models to be estimated separately for different recipient classes, thus overcoming the difficulty presented by composition effects in the benefit rates. Furthermore, it allows for examination of sub-groups of individuals such as those with low levels of education. Decompositions of the usage series allow for determination of the proportion of the series that is explained by economic and policy variables. I find that that cyclical factors are important when the right measure is used, but there is only qualified support regarding the importance of the availability of UI benefits and the relative generosity of welfare benefits. While this is a step forward, we need to better understand the mechanisms underlying these trends. This motivates the second stage of the analysis. An increase in the total caseload in a month can arise either because duration of welfare spells have been increasing, the rate of movement onto welfare has been increasing or some combination of the two. There has been substantial evidence accumulated on the first, but very little is known about the second. This focus on the duration of welfare spells and the exit rate from welfare is understandable since the main data used so far is administrative data which is usually limited to the information required to administer the program and has little information regarding individuals' behaviour before their involvement with welfare. In chapter 4,1 address this gap in our knowledge examining flows into social assistance using individual level survey data to examine the 5 path leading individuals to social assistance receipt. I characterize this path by three processes. The first process considers the job loss experienced by an individual. The second process considers an individual's transition to re-employment following a job separation. The third process considers an individual's transition to welfare receipt, given that a job separation has occurred. Using the 1988-90 longitudinal file of the Labour Market Activity Survey I estimate semi-parametric duration models to determine how these processes are affected by incentives in welfare benefits, labour market conditions, availability of unemployment insurance as well as demographic variables including level of education. The final exercise in the thesis involves the application of the estimates from the duration analysis to administrative data on the number of persons experiencing job separations to simulate the number of persons who would subsequently take-up welfare. These predicted incidence series are then decomposed in order to determine the degree to which predicted incidence is driven by changes in labour market conditions and policy variables. In particular, it allows for an examination of which factors can account for the dramatic growth in welfare usage experienced during the 1990s. Conclusions and directions for future work are discussed in chapter 6. 6 Chapter 2 Institutional Framework The provision of assistance to people in need has long been an integral part of Canada's social policy. Each province within Canada operates its own welfare system and Manitoba, Ontario and Nova Scotia operate both provincial and municipal programs. Each of these has its own complex system of rules regulating all aspects of the system; however, all operate basically along the same lines. This reflects the means by which welfare expenditures have been financed. During most of the past 35 years social assistance expenditures have been funded under the Canada Assistance Plan (CAP). Under this agreement, in effect from 1966 until 1996 when replaced by the Canada Health and Social Transfer (CHST), provincial expenditures on welfare programs were shared equally with the federal government, provided certain conditions were met. To qualify for shared funding, provinces had to provide assistance to all persons judged to be in need. Furthermore, they had to provide a legal process by which individuals could appeal decisions of welfare officials and could not impose a residency requirement as a condition for eligibility. This funding arrangement underwent an important change in the 1990s. Starting with the 1990-91 fiscal year, the federal government limited the amount it contributed to Ontario, Alberta and British Columbia (the "CAP on CAP"). For these provinces, the maximum federal contribution increased 5% annually relative to 1989-90 levels; welfare program costs in excess of these limits was borne solely by the province. The concept of limiting expenditures was extended to all provinces with the introduction of the Canada Health and Social Transfer (CHST) in 1996. Under this program, existing federal block funding arrangements for medicare and post-secondary education were expanded to cover welfare programs. Additionally, under the CHST, provinces were no longer required to provide assistance to all persons in need, nor were they required by law to have a formal appeal process; only the residency requirement remained. 7 2.1 The Application Process The process facing a person applying for welfare is similar in all provinces. An applicant for welfare is first assigned to a specific category of recipient depending on their employability and the presence and number of dependents. Definitions of these categories vary across provinces. Once the administrative conditions are met, each applicant goes through a "needs test" which compares the budgetary needs of the applicant and any dependents with the assets and income of the household. These vary across provinces and with the category of recipient. The first step of the needs test concerns an applicant's fixed and liquid assets. In most provinces fixed assets such as a home and car are exempt, but before qualifying for welfare applicants are required to liquidate non-exempt fixed assets and use those funds to support themselves. The second part of the needs test concerns a household's income, some of which, such as the federal child tax credit, is considered exempt. If non-exempt income is insufficient to cover needs, the household qualifies for welfare. Once eligibility is established, the benefits an applicant receives is determined. The method used to calculate such payments also varies by province and is intended to cover expenses for food, clothing, shelter, utilities and an allowance for personal and household needs. Additionally, the benefits paid to a household is affected by the rules regarding earnings exemptions. Each province allows welfare recipients to retain a certain amount of earnings without any reduction in their welfare cheques. This amount may be a flat-rate sum and/or a percentage of earnings. 2.2 Reforms to the Welfare System During the 1981-95 period examined in this analysis, there were numerous and varied changes made to welfare programs in Canada. During the last half of the 1980s, reforms generally served to make welfare programs more generous to persons on social assistance. Some changes involved efforts to assist welfare recipients back into the workforce. These included the Federal-Provincial Agreements to Enhance the Employability of Welfare Recipients - the "Four-Corner" Agreements, which began to be 8 implemented in 1985. Under these agreements, funds from both the federal and provincial governments were used to involve social assistance recipients in training programs, including those operating under the Canadian Jobs Strategy (CJS). Some provinces implemented additional initiatives. In particular there were major reforms in provincial welfare systems in Ontario and Quebec. In 1986 the Ontario Ministry of Community and Social Services announced the appointment of the Social Assistance Review Committee (SARC) to undertake a comprehensive review of the welfare system. This resulted in 274 recommendations for welfare reform (see "Transitions", SARC 1988), with the first of these phased in between October 1989 and January 1990. These included the introduction of the Supports to Employment Program (STEP) on October 1, 1989. Among the reforms included in STEP was the standardization of all earnings exemptions at 20% of net earnings (previously gross earnings) beyond the basic levels and the introduction of flat-rate exemptions for training allowances. In 1989 Quebec replaced the Social Aid Act with the Act Respecting Income Security. Under this legislation the Social Aid program was replaced by two new programs: the Financial Support Program for people with severe disabilities and the Work and Employment Incentives program (WEIP) for everyone else. These became effective for new applicants on August 1, 1989 and all applicants as of August 1, 1990. Also created was the Parental Wage Assistance program for low-income families with children to replace the Work Income Supplement Program that covered low-income workers. Under WEIP individuals were placed in specific categories according to their willingness and ability to participate in vocational training, job search assistance, work in community agencies or subsidized employment. Furthermore, benefit rates and income exemption levels were directly linked to this classification. These reforms also ushered in significant increases in the level of potential benefits paid to certain classes of recipients. In Ontario, effective January 1, 1990 there was an increase of 6% in basic welfare rates and the implementation of a revised system of shelter allowances, with a base allowance provided regardless of actual shelter costs and a supplementary allowance to cover actual costs up to designated ceilings. The combination of these two changes had the result of increasing the maximum potential 9 benefits available to individuals on social assistance approximately 15-17%.1 As a result of the reforms in Quebec, some recipients saw their benefits increased while others had them reduced. However, a dramatic change was experienced by childless singles and couples less than 30 years of age. Prior to this date these individuals were paid a much lower benefit rate than similar individuals aged 30 years and older. As a result of the new act, this differential was eliminated and benefits paid these younger individuals more than doubled. 2 Less dramatic increases in nominal welfare benefits were instituted in B.C. in July 1989 and August 1990 (approximately 7% and 6% respectively). Reforms to welfare programs during the 1990s were largely driven by a desire to contain costs in the wake of a rapid increase in welfare cases and reduced federal funding resulting from the "CAP on CAP". Over the 1991-93 period the Social Allowances Program in Alberta was replaced with Programs for Independence. This resulted in not only changes in benefit rates in October 1993, but also a pervasive structural change in administration. Welfare offices were specifically asked to develop initiatives to cut caseloads, including more intensive reviews of new and existing cases, requiring people to attend information sessions, requiring recipients to follow case plans and establishing waiting periods for non-emergency cases. In B.C. efforts began in 1994 to tighten up administrative practices to contain costs such as establishing agreements to share information with other provinces to reduce fraud and, the recovery of benefits paid to individuals waiting for unemployment insurance benefits. Sweeping reforms to the welfare program in Ontario were implemented following the election of the Harris government in June 1995. In October 1995, benefit rates were cut by 22% for all individuals except for seniors and the disabled. These reforms also included plans to implement large-scale workfare programs with harsher penalties for employable persons quitting or losing a job without just cause. Further reforms of this nature continued after 1995 in Ontario and other provinces. 1 The specific magnitude of the increase depends on family type. 2 As a result of the reforms, individuals classified as Available had their benefits increase from $185 to $487 in nominal terms. 10 2.3 Parameterization of the Welfare System Some features of the welfare system are more easily parameterized than others. Program parameters, such as potential benefit rates, asset exemption levels and income exemption levels, may be quantified relatively easily. Data on these are available from published material or may be obtained directly from provincial ministries. Quantifying structural/administrative changes is much more difficult. In the analysis that follows, no attempt is made to control for the impact of specific administrative reforms. Consider the impact of changes in program parameters. Some of these parameters are revised more regularly than others. Nominal benefit rates are revised the most often. In several provinces these are either indexed to another payment, such as in the case of Quebec (linked to the level of Q.P.P. payments, usually revised January each year) or revised at regular intervals. Both N e w Brunswick and Ontario revise their schedules annually. Other provinces revise benefits on a more ad-hoc basis. 3 Changes to exemption levels on earned income and liquid assets occur much less frequently than benefits and often as part of major reforms. Generally, these levels remain constant in nominal terms for several years at a time. Exemption levels for non- disabled individuals in selected provinces are presented in Table 2.1. Therefore, for the purposes of this study I am abstracting from all dimensions of the welfare system other than benefit rates. Inclusion of controls for province wi l l capture the effects of these omitted factors to the extent that these dimensions are unchanged over the sample period. O f course, this is not the case for the Alberta Reforms and STEP reforms in Ontario, but investigation of these effects of these reforms w i l l be explored in future work. 3 For example, Alberta changed in rate schedules once in 1985, 1987, 1988, then not until January 1991. 11 Chapter 3 Aggregate Analysis of Growth of Welfare Receipt in Canada 3.1 Introduction In this chapter, the growth of welfare receipt in Canada is examined within an aggregate framework. I seek to determine if factors often proposed, namely changes in welfare benefits, labour market conditions and the availability of unemployment insurance (UI) can explain the observed increase in social assistance usage, and furthermore, if these results hold up upon scrutiny. I conduct two separate sets of estimations. In the first approach I examine annual aggregate data for a panel of nine provinces (excluding Prince Edward Island) for the 1982-96 period. The initial analysis aggregates all welfare cases. Subsequently, the analysis is refined and separately considers the behaviour of different types of welfare recipients in five provinces over the 1981-95 period. These provinces include New Brunswick, Quebec, Ontario, Alberta and British Columbia. 3.2 Framework of the Analysis In the model, welfare usage is defined as the number of social assistance cases as a proportion of the number of households. It is hypothesized that welfare usage will be inversely related to the level of labour market opportunities. It is further hypothesized that welfare usage will increase with the generosity of welfare benefits relative to wages from alternative employment. Lastly, the restrictiveness of conditions for qualifying for unemployment insurance (UI) benefits is believed to positively affect welfare usage since those receiving UI support are not eligible for full welfare support as well. Social Assistance recipients can be broadly classified one of four types: (i) disabled persons, (ii) aged persons, (iii) single parent families and (iv) unemployed employable persons. Individuals in different categories face different economic circumstances, and thus would be expected to exhibit different responses to changes in generosity of benefits and the availability of employment. 12 The importance of separately examining different classes of welfare recipients cannot be overstated. Individuals who are disabled or above the mandatory retirement age are unlikely to be employed even in favourable economic conditions. Alternatively, those individuals below retirement age who do not suffer from some sort of disability should in principle be willing to accept job offers that pay enough. I will refer to this group as the "potentially employable." It can be argued that lone parents who are the sole support for their children would not be available for work. However, if a sufficiently well paying job were available, child related costs such as daycare could also be covered and still make work a financially attractive alternative to welfare. It is these differences which motivate the two separate analyses performed in this chapter. Use of the second approach not only allows for more accurately assigned potential benefit rates, but also allows for different behaviour among sub-groups. In this light, the first approach may be best viewed as a "first pass" at the data, which provides a benchmark to which the results of the second (disaggregate) approach may be compared. It allows us to examine whether the results from an aggregate approach are robust, or if they may be partially attributed to the aggregation of different classes of welfare recipients and the use of broad measures of labour market conditions and welfare benefit rates. Among these "potentially employable" individuals, the overwhelming majority are accounted for by singles with or without children.1 It is these individuals who are the focus of the second approach. Due to limitations of the data, the disaggregate analysis is confined to individuals residing in New Brunswick, Quebec, Ontario, Alberta and British Columbia. 1 Administrative data shows that singles with and without children account for at least 80% of the caseload at any given time. Source: Inventory of Income Security Programs - various years. It should be noted that this is a point in time measurement and may differ from a measurement which also considers duration. 13 3.3 Data Data on the aggregate welfare caseload by province is available from administrative sources. However, similar data is not available disaggregated by employability and family type; hence, another sources of data is required. Data from the Survey of Consumer Finances are well suited for this purpose. The SCF data set has the advantage that when used with its sampling weights it is representative of the entire Canadian population aged 15 years and older. Furthermore, it offers a wealth of income and demographic information. These features make it an attractive candidate for the task at hand. Respondents reporting they were permanently unable to work allow for the identification of those who would be classified as disabled for purposes of the welfare system. Variables indicating marital status and the presence of children allow for singles with and without children to be separately identified. SCF data are not, however, without limitations. There is only a single variable reporting participation in welfare, namely, the amount of income received from "SA income and provincial income supplements." In addition to the standard income assistance payments, these also include provincial top-up programs for seniors and any other cash payments to persons in need. Hence, it is possible to identify all individuals who reported participating in social assistance at some point within the year, but the specifics of their involvement is unknown. More important limitations stem from the fact that responses to the SCF questionnaire are self-reported and may reflect response bias. Previous work (Dooley 1999) has revealed under-reporting of welfare income in the SCF, which may reflect both the degree of involvement with welfare, and more importantly for the current task, whether or not they were involved with welfare. The current paper investigates time variation in welfare usage rates. To the extent that the degree of under-reporting is consistent across time, the effects of this limitation will be minimized. A second limitation arises from the fact that the SCF is subject to sampling error. 2 The Survey of Consumer Finances is a supplement to the Labour Force Survey. Respondents are interviewed in April each year regarding their experiences in the previous calendar year. 3 It is possible that the degree of their involvement may be inferred from the amount of SA income received. 14 While the weighted data is representative of the entire Canadian population (excluding the territories), the number of respondents reporting involvement with welfare is only a fraction of the total. Accordingly, when examining specific sub-groups of individuals it is important to be aware of the number of data points on which estimates are based. 3.3.1 Definition of Types of Recipients and Comparison with Administrative Data Use of the SCF data requires that I define who is to be included as (a) singles without children and (b) singles with children. Status as a single is defined as currently unmarried and thus may include those who have never married and those who are divorced or widowed. I define status as a parent as reporting the presence of a child less than 18 years of age. Before the SCF data can be used it is critical to investigate how closely it reflects actual caseloads. Administrative data on provincial caseloads broken down by family composition are available from volumes of the Inventory of Income Security Programs. These totals include both disabled persons and seniors, so in order to perform the comparison initial samples (singles without children and single parents) were drawn from the SCF data including all persons regardless of disability status. Individuals age 65 and older were excluded to avoid inclusion of those who did not actually receive welfare income, but rather received some sort of income supplement. (Note that administrative data indicates that only in a very small proportion (1-3%) of cases is the head above 65 years old.) For each of the five provinces of interest, the raw caseload series from the SCF data is plotted against the corresponding administrative series for both types of recipients. It must be remembered that these two series do not reflect the same measurement. While the administrative series reflect the caseload as of March of a given year, the SCF numbers reflect the total number of individuals who reported involvement with welfare at some point in that year. Figure 3. la presents the administrative and SCF caseload series for single This has not as yet been investigated by the author. 15 individuals without children. Overall, the caseload series constructed from the SCF data is in broad agreement with the administrative series with the notable exception of the sharp decline exhibited by the SCF series in B.C. in 1995. The similarity between the two series may be established more exactly by computing the degree of correlation between the two series, presented below in Chart 3.1.5 Chart 3.1: Correlation Coefficients - SCF vs. Administrative Caseloads, 1984-95 Province Singles without Singles with Children New Brunswick 0.52 -0.47 Quebec 0.70 0.51 Ontario 0.98 0.94 Alberta 0.72 0.79 B.C. 0.82 0.88 With the exception of the series for New Brunswick, the administrative and SCF series are highly correlated, especially in the cases of B.C. and Ontario. Therefore, it appears that although involvement with welfare is under-reported, the degree of under-reporting is fairly consistent across time. Figure 3.1b presents caseload series from the SCF and administrative data for singles with children. The SCF series are seen to be more volatile in Quebec and especially New Brunswick. Nevertheless, while broadly consistent with the actual caseload in Quebec (correlation between admin, and SCF series is 0.51), this is certainly not the case in New Brunswick (-0.47). This may reflect the smaller numbers of data points used in deriving estimates for the New Brunswick series. For the remaining provinces, the two series display a much tighter match, with similar degrees of correlation as for singles without children. However, there are exceptions such as the spike exhibited in the SCF series in 1984 in Alberta as well as the leveling off of the caseload in later years, especially in Ontario. 41 have used the March observations for the IISP data in these comparisons. 16 3.3.2 Selection and Description of Final Sample In order to focus on the population of interest, namely, singles with and without children who are potentially employable, the SCF sample was further restricted to include only individuals with reported age 19-64 (19-54 in New Brunswick).6 Due to the very small number of male single parents, only female single parents were retained in the single parent sample. Individuals reporting they were permanently unable to work were excluded. It is instructive to examine how the groups of interest differ between welfare recipients and non-recipients. Table 3.1 presents the age distribution for all five provinces combined. Separate panels are presented for (a) single males with no children (SNM), (b) single females with no children (SNF) and (c) single females with children (SPF). Similar distributions for education level and weeks worked in the reference year are presented in Tables 3.2 and 3.3 respectively. These allow for a basic comparison across recipients and non-recipients in terms of their demographic characteristics and work experience. As seen in Table 3.1, the age distributions exhibit a declining population share of the 18-24 year olds. Although for singles without children the 18-24 year olds are the largest age group in the population, they comprise a smaller proportion of the welfare caseload; this is especially true for females. Particularly striking is that the distributions for the males without children and females without children who receive welfare are essentially inverses of each other. The greatest proportion of males is found in the 18-34 age group, while the largest proportion for females is found in the 55-64 age group. Across both recipients and non-recipients the majority of female single parents are less than 45 years of age. However, unlike non-recipients, welfare recipients are more heavily represented in the 18-24 and 25-34 categories. Recipients are far more likely to be in the 18-24 group than non-recipients. Overall, the education distributions presented in Table 3.2 reveal declining proportions of individuals reporting low levels of education (some high school or less). 5 For years before 1984 either the SCF or administrative data was not available. 6 Age 19 was used as the lower limit in order to ensure that no 17 year olds were retained in the sample. Age 54 was used as the upper limit for N.B. as individuals over age 54 were treated differently than in other 17 This finding is consistent across all 3 groups. Although this is true for recipients and non- recipients alike, it is also true that the bulk of the welfare caseload is comprised of individuals with low education. This is particularly true for females. Within the recipient population we see that for singles with children the caseload shares for the certificate and degree categories are increasing at the expense of those with only elementary education. There is also a distinct shift after 1990 for males and after 1991 for females. Single parent recipients are also becoming more educated and experience a similar shift around 1990. Table 3.3 presents the distributions for weeks worked in the reference year. As expected, a much higher proportion of welfare recipients than non-recipients report working zero weeks. The distribution for males without children is essentially constant over the sample period. However, for both female singles without children and single parents the proportion of cases reporting zero weeks worked declines over time from approximately the low-middle 70's to the middle 60's. There is a corresponding increase in the proportion of females reporting greater than 40 weeks worked, although this does not account for the entire shift. 3.4 Estimation Approach In the model, welfare usage is represented by the number of social assistance cases as a proportion of the number of households. For the first (aggregate) approach this variable is computed using administrative caseload data and data on the number of households. For the second (disaggregate) approach it was computed using SCF data for each of the groups mentioned in section 3.2. In the aggregate approach, the unemployment rate for men aged 24-44 is used as a proxy for the level of labour market opportunities in the economy. In the disaggregate analysis the average number of weeks worked in the reference year by all individuals in the sub-group of interest is used as a proxy for labour market opportunities. For example, in the analysis of welfare usage among single men without children, the value of the "Weeks Worked" variable represents provinces. 18 the average number of weeks worked in the reference year for all single men without children, including both those who did not receive welfare and those who did. For both sets of estimations, the relative generosity of welfare benefits is represented as welfare benefits as a proportion of full-time minimum wage earnings. However, in the aggregate estimations, welfare benefits are computed as average social assistance expenditure per case, whereas in the disaggregate estimations, we use the maximum social assistance benefits payable as computed from published benefit schedules. Lastly, in all estimations the restrictiveness of conditions for qualifying for unemployment insurance (UI) benefits is represented by the inclusion of the minimum number of weeks of insurable employment required to qualify for UI benefits.8 Details regarding data sources and the construction of the variables are provided in the appendix. The availability of panel data in the present analysis allows for the inclusion of separate dummy variables for both provinces and years. Province dummies will capture fixed and persistent province-specific effects. The specification of separate year dummies avoids the restrictions inherent in the specification of a linear trend, affording a fully flexible representation of common year effects i.e. these control for contemporaneous correlations across provinces. Therefore, time invariant differences in welfare usage between provinces are stripped away, as are aggregate trends common to all provinces. Therefore, benefit, labour market and UI effects are identified solely by time series variation within each province. Use of such variation is appropriate when identifying the impact of a provincially set policy variable such as welfare benefit rates. 7 Increased inequality in earnings and hours worked during the sample period has been well documented [e.g. Morissette et. al (1994), Beaudry and Green (1997)]. It is likely that this contributed to the overall upward trend observed in SA usage. The current specification does not control for changes along these dimensions. Investigation of this issue will be the subject of future work. 8 In actuality, such levels are specific to each UI region. The variable used in the analysis is constructed using the appropriate schedules and method but using the provincial unemployment rate. See Appendix for more details. 19 3.5 Results For both the aggregate and disaggregate approaches I will first conduct a graphical analysis, examining trends in the independent variables to ascertain their potential influence on welfare usage. Subsequently a more exact examination will be performed using regression analysis. 3.5.1.lAggregate Approach - Graphical Analysis Figure 3.2 presents social assistance usage for each of the nine provinces in the sample. As can be seen, the pattern exhibited by the corresponding national series in Figure 1.1 is reproduced clearly only in Ontario. Usage series for most provinces display an overall upward trend, with little variability for some (Nova Scotia, Saskatchewan and Manitoba) and a high degree of variability for others (Newfoundland and B.C.). Alberta's series displays an upward trend until 1993 but this is sharply reversed thereafter. The series for both New Brunswick and Quebec remain in the same general range over the period, but Quebec's series fluctuates more. When Figure 3.2 is compared to Figure 3.3, it is seen that the usage patterns are broadly consistent with those exhibited by the unemployment series, but with several exceptions. In the early 1980's Newfoundland experienced increases in unemployment, but welfare usage decreased. Although Alberta experienced a large increase in unemployment following 1983, this had a relatively small impact on usage. Usage then continued to increase until 1987 though the unemployment rate began declining in 1984. The provincial unemployment patterns displayed in the late 1980's are consistent with the usage series with the exception of Nova Scotia where usage continued to steadily increase despite a sharp decline in the unemployment rate following 1987. The most pronounced discrepancies appear in the 1990's. In Newfoundland and Nova Scotia, usage failed to decline despite severe declines in the unemployment rate, and in B.C. usage continued to increase despite a moderate decline in the unemployment rate. The increase in caseloads in Ontario was consistent with increased unemployment in the early 1990's, but the 20 relative increase in caseloads was stronger. Additionally, usage continued to increase until 1994, while unemployment had been declining since 1992. Lastly, it is seen that usage in Alberta declined dramatically after 1993 and continued to fall even after the leveling off of the unemployment rate in 1994. Figure 3.4 presents the relative welfare benefit series for each province. These patterns also appear to be broadly consistent with the usage series. This is especially notable in Alberta where usage declined dramatically following the decrease in benefits after 1992. However, again there are discrepancies. In Newfoundland, usage decreased steadily until the late 1980's although benefits steadily increased, a pattern also evident for New Brunswick beginning in 1984. However, the biggest discrepancy in the 1980's is in Quebec where usage steadily increased and then decreased despite the steady increase in benefits. It is seen that during the 1990's Newfoundland experienced sharp increases in usage despite a leveling off of the benefit rate and B.C. experienced increased usage despite a sharp decline in benefits. Usage also increased in New Brunswick in the early 1990's, a time when benefits were declining. The most striking discrepancy is for Ontario where usage increased drastically after 1990 despite a leveling off and then a decline of the benefit rate. The series depicting the minimum weeks necessary to qualify for UI are presented in Figure 3.5. On the whole, these are broadly consistent with the usage patterns in Ontario, Manitoba and in the 1990's, B.C. However, visual inspection suggests that it is unlikely that changes in this variable can explain much of the observed patterns in welfare usage. In summary, graphical analysis indicates that the observed patterns in welfare usage over the sample period are broadly consistent with changes in relative welfare benefit generosity and labour market conditions. However, this cannot be said of the UI availability measure. 21 3.5.1.2Aggregate Approach - Regression Analysis The relationships apparent in the figures presented thus far may be investigated in a more comprehensive manner using regression analysis. Column (1) of Table 3.4 presents results from the ordinary least squares regression of the welfare usage rate on the relative benefits variable, the unemployment rate, and the minimum UI qualification weeks measure, as well as on the province and year dummy variables. As indicated by the R 2 value of 0.84, the fit of the model is quite good. Furthermore, the estimated coefficients on both the benefit variable and the unemployment rate have the expected sign and are significant.9 Both these measures are proportions, as is the dependent variable; thus these results may be interpreted as implying that a 1% increase in relative benefits results in a increase of 0.12% in the usage rate and a similar increase in the unemployment rate implies a 0.42% increase in the usage rate. Although the minimum UI weeks variable has a positive sign, its effect is insignificant. Looking at the coefficient estimates for the provincial dummies, it is seen that welfare usage is significantly greater than average in Quebec, and lower than average in Saskatchewan and Alberta. The coefficients of the year dummies are insignificant with the exception of 1994, 1995 and 1996. Therefore, after controlling for benefits, unemployment and UI effects, it appears that much of the usage pattern is explained by the model, particularly in the 1980's. Hence, it would appear there is some validity to the claim that these were the factors responsible for the increase in caseloads. It is instructive to decompose the usage rates predicted by the OLS estimation into the component resulting from year effects versus that which is generated by benefit rates, unemployment and UI restrictions. Specifically, an additional specification of the model is estimated in which usage rates are regressed on province and year dummies, but with the three covariates excluded. Such a procedure renders the predicted usage rates to be solely a function of time and essentially replicates the observed national usage series for a given group. Additionally, a restricted predicted series may be computed using the estimates from the full model (including the three covariates) but computed holding a 9 Unless otherwise stated significance is taken to be at the 5 % level. 22 covariate, or group of covariates, at their sample averages. This series reflects the variation in the usage series that is left to be explained by the year dummies after stripping away any variation that can be accounted for by variation in the restricted covariates. Comparison of the restricted series with the unrestricted series indicates the degree to which variation in usage may be accounted for by variation in the restricted covariates. For example, if a specific covariate can explain all the variation in usage then restricting that covariate to equal its sample average will yield a restricted predicted series that is completely flat. On the other hand, if a covariate cannot explain any of the variation in usage then restricting it to equal its sample average will yield a restricted predicted series that exactly tracks the unrestricted predicted series. Unrestricted and restricted series for the regressions are plotted in Figures 3.6. As can be seen, the OLS predicted series displays less variability than the "unrestricted" series, which may be interpreted as indicating that there is less unexplained variation in the usage series captured by the year dummies when benefits, unemployment and UI restrictions are considered. However, the considerable variation remaining in the predicted series indicates that a substantial amount of usage remains unexplained despite the inclusion of these variables. In order explore the robustness of these results another regression was estimated using generalised least squares allowing for an AR1 error structure. These results are presented in column (2) of Table 3.4. Several points immediately stand out. First of all, although the coefficients on both the benefit and unemployment variables remain significant and retain the expected sign, they are approximately 1/2 their previous magnitude. Furthermore, seven additional year coefficients are now significant (five at the 5% level). Inspection of the predicted series (holding benefit, unemployment and UI effects constant) in Figure 3.6 reveals that accounting for the AR1 structure results in less of the usage series explained by changes in benefits, unemployment conditions and UI restrictiveness relative to the OLS case. A possible implication of this result is that usage rates and benefit rates trend together (as do usage and unemployment rates). The quasi-differencing involved in 23 specifying the AR1 error structure effectively removes a stochastic trend and thus it may well be that while the series are broadly correlated, the timing of the turns in the series' are different.10 Another possibility concerns the construction of the benefit variable itself.11 Due to the fact that it represents total welfare expenditures divided by all cases regardless of recipient type, it is possible that this measure suffers from composition effects. A closer look at Figure 3.4 supports this hypothesis. Careful inspection of the relative benefit series indicates that throughout the sample period Alberta offered the most generous welfare benefits. However, in truth, this is not the case. A comparison of real benefit rates by class of recipient reveals that from 1987 onwards, benefit levels in Alberta were among the lowest of the four major provinces. This apparent discrepancy may be resolved by considering the differences in the composition of the welfare caseload across provinces. The distribution of cases across family types differs across both provinces and time. Since benefit rates vary with family type, this implies that measures of benefit rates aggregated over all categories of social assistance recipients are affected by the changing composition of the total social assistance caseload as well as changes in the benefit schedule Results from this analysis of welfare usage using aggregate data suggests that while economic analysis may appear on the surface to allow for statements to be made attributing the observed increase in usage in Canada to changes in relative benefit rates and unemployment conditions, closer attention raises questions about this conclusion. These concerns provide further justification for a more refined analysis, such as the one to which we now turn. 3.5.2.1 Disaggregate Approach - Graphical Analysis Figures 3.7a-3.7c present social assistance usage rates and relative benefit series for each of the five provinces in the sample. Figures 3.7a, 3.7b and 3.7c pertain to male 1 0 The model was also run using a lagged benefit rate with no appreciable differences. 1 1 The model was also run with benefits per case and minimum wage earnings as separate regressors. The only appreciable difference was that in the AR1 estimation, minimum wage earnings became insignificant. 24 singles without children, females without children and single mothers, respectively. As can be seen, only for Ontario is the pattern exhibited by the national usage series in Figure 1.1 reproduced clearly. It must, of course, be noted that Figure 1.1 represents the entire national caseload ~ not just non-disabled singles and single parents. Usage series for males without children display an overall upward trend in most provinces; however, the series for Alberta decreases after 1991 as does the Ontario series after 1993. In Quebec the usage series does exhibit variation over time but doesn't display an overall trend. B.C. displays a sharp decrease in usage in 1995. These patterns are also present in the usage series for females without children with important differences. The growth in usage for females without children in Ontario is not as dramatic as it is for the males and in New Brunswick usage among females without children does not display the upward trend observed among males without children. Furthermore, usage rates are generally slightly higher among women than men. Usage for single mothers displays an overall upward trend only in Ontario and B.C. In both N.B. and Quebec, the pattern is somewhat volatile, but stays in roughly the same range. Alberta's usage series displays sharp growth in the early 1980's but then remains relatively constant until sharply declining after 1992. For males without children the usage patterns are broadly consistent with the relative benefit series, but with several exceptions. In the 1980's relative generosity increased in both Quebec and Ontario; however, usage was stable during this period. In 1986 benefits decreased very sharply in Alberta with little change in usage. Finally, in Ontario and B.C. benefit generosity was relatively constant from the late 1980's onward, yet both experienced dramatic increases in usage in the 1990's. Similar results are found for females without children. Inspection of Figure 3.7c reveals that in the case of single mothers, benefit series are again generally consistent with observed patterns in usage. It is worth noting that the dramatic increase in usage exhibited in Ontario from 1989-1992 is preceded by a similar increase in benefit generosity. Inspection of Figure 3.8a reveals that patterns in weeks worked is very consistent 25 with patterns in usage rates for males without children.12 In almost all periods when weeks worked declined, reflecting depressed economic opportunities, usage was increasing. However, the magnitude of the variation in weeks worked appears unlikely to account for the dramatic increase in caseloads exhibited in Ontario and B.C. in the early 1990's (particularly in B.C.). Similar observations apply to the females without children. The weeks worked series for single mothers is less successful at explaining the observed usage series. Figures 3.9a-3.9c present the series for the number of weeks necessary to qualify for UI benefits. The series for both males and females without children indicate that UI restrictions were consistent with usage patterns observed in Alberta until 1991 and in N.B. and B.C. starting in 1990. With respect to single mothers, UI restrictions appear somewhat consistent with usage patterns exhibited in most provinces beginning with the late 1980's. 3.5.2.2 Disaggregate Approach - Regression Analysis The relationships apparent in the figures presented thus far can be investigated in a more comprehensive manner using regression analysis. Regressions were estimated for each of the three groups separately and are reported in Tables 3.5a, 3.5b and 3.5c. For males without children (Table 3.5a) three regressions are reported. All were estimated using ordinary least squares. Regression (1) uses a measure of employment conditions defined as the average number of weeks worked for all individuals in the group, including both welfare recipients and non-recipients (this is the measure presented in Figure 3.8a). Using this measure it is seen that among the variables proxying for economic factors, (I shall refer to these variables as the 3 main covariates) only weeks worked has a significant effect on welfare usage, with an increase of 1 week of work resulting in a decrease in the usage rate of 0.78%. Usage is higher in Quebec and B.C. than in Ontario, 1 2 Subsequent sections will utilize measures of weeks worked specific to certain age and/or education levels. The series presented in Figure 3.8a-3.8c do not make any such exclusions. 1 3 All models were also estimated using generalized least squares allowing for an AR1 disturbance term. Support for this specification was found only for the models for females without children. As results were essentially unaltered, only the OLS results are discussed in the text. 26 the excluded class. Usage is relatively lower in New Brunswick although this coefficient is significant only at the 10% level. Lastly, it is seen that the year dummies are significant for 1988 and 1992-94. This is important to note as these will play a central role later in the analysis. Regression (2) employs an alternative measure of labour market conditions. As indicated by the age distribution presented in Table 3.1, approximately half of males without children who receive welfare are less than 35 years old. For this reason regression (2) employs as a covariate the average number of weeks worked for those individuals less than 35 years. As seen in column (2) the estimated coefficient for weeks worked remains stable. However, the coefficients for relative benefits and UI qualification weeks both increase, and the coefficient for UI qualification weeks becomes significant. Province dummies exhibit similar significance as in regression (1), except that the year dummy for 1993 is no longer significant. As a final refinement, the model is re-estimated restricting the sample to those individuals having low levels of education (i.e. some high school or less) and using the average weeks worked for persons less than age 35 with low levels of education. Focusing on this particular sub-group it is seen that the coefficient on relative benefits has decreased relative to specification (2). Furthermore, coefficients on both weeks worked and minimum UI qualification weeks increase in magnitude, particularly the latter. It is interesting to note that in (3) additional year coefficients are significant (1986-87) and the coefficient for 1992 is no longer significant. Table 3.5b presents the regression results for females without children. Regressions (1) - (3) include the same covariates as their counterparts in Table 3.5a. Additionally, I present regression (4) which restricts the sample to the less educated and uses the average weeks worked for all individuals with low levels of education. The inclusion of this specification is motivated by the observation that single women without children who receive welfare are more uniformly distributed with respect to age than their male counterparts. Accordingly, use of a more broadly defined measure of weeks worked is indicated in the case of single women. Results are quite different than those for single males. For regression (1) not only are weeks worked significant, but also UI qualification 27 weeks and relative benefits, although benefits is significant only at the 10% level. Moreover, the magnitude of these effects are considerably larger than those for males. Again both Quebec and B.C. have positive significant coefficients. However with the exception of 1988, which is significant only at the 10% level, the regression displays no significant year effects. Use of average weeks worked for those less than 35 years changes these results somewhat, diminishing the magnitude of the coefficients of the weeks worked and minimum UI weeks variables by approximately one-third and rendering the minimum UI weeks coefficient insignificant. The coefficient on relative benefits remains stable and mildly significant. Restricting the sample to less educated single females without children I find that only weeks worked remains significant, whether it be the average for all individuals with low levels of education or those less than age 35. The most noticeable change is the increased magnitude and significance of the year coefficients in regression (3). The regressions for the sample of single mothers are presented in Table 3.5c. All regressions use average weeks worked for all single mothers as the proxy for labour market conditions. Regression (1) includes all single mothers whereas specifications (2) and (3) restrict the sample to those having only young children (less than 7 years old) and those having only older children (7-17 years old) respectively. In all regressions, relative benefits and weeks worked are significant at the 5% level.14 This is somewhat surprising. Single mothers are usually thought to be constrained by responsibilities at home, and thus unable to accept employment. However, this sensitivity to benefits and employment conditions is observed for both single mothers with only young children and those with only older children. The coefficient on weeks worked displays considerable variation across the three specifications, but in all cases is of a greater magnitude than for the specifications for singles without children. Dooley's (1996) analysis of lone mothers in Canada reveals important differences in social assistance 28 3.5.2.3 Decomposition of Regression Analysis In the previous section it was established that welfare usage rates are correlated with economic and policy variables. However, the question remains as to how much of the overall time pattern is explained by the inclusion of these regressors. This can be addressed in the same manner as was employed in the aggregate analysis. Unrestricted and restricted series for the regressions are plotted in various panels of Figures 3.10-3.12. Figure 3.10a presents these "year effects" for single males without children from regressions (1) and (2). As can be seen, predicted series using weeks worked for all individuals displays considerably less variation than the unrestricted series, which may be interpreted as indicating that there is less unexplained variation in the usage series captured by the year dummies when benefits, employment conditions and UI restrictions are considered. The predicted series using average weeks worked for those less than age 35 appears to remove even more of this variation. However, visual inspections can be misleading, so in order to provide a quantifiable measurement, the reduction in variation resulting from restricting a given covariate is computed to be the difference between the coefficients of variation for the unrestricted and restricted series, taken as a proportion of the coefficient of variation of the unrestricted series. Table 3.6 presents these reductions in variation for each of the regressions for both the 1981-95 and 1989-95 periods. As indicated in the table, for the sample of single men without children, restricting the three covariates to equal their sample averages can account for 45% - 51% of the variation in usage over the 1981-95 period. Thus, a substantial amount of the time pattern is explained by inclusion of the three covariates, and even more when the weeks worked reflects those aged less than 35 years of age. Figure 3.10b presents the decomposition for the low educated single males without children. Again, inclusion of the three covariates removes much of the variation in the time pattern (56%). The reduction in variation during the 1989-95 period is greater for the model using all persons and average weeks worked for persons of all ages, but essentially unchanged for the others. usage between different age groups. Such a disaggregation will be pursed in future work. 29 The decompositions for females without children are seen in the various panels of Figure 3.11. Inspection of these panels and Table 3.6 indicate that a lesser proportion of the variation in the unrestricted usage series over the 1981-95 period can be accounted for by inclusion of the three covariates than in the case of single men without children (35%). It is noteworthy that unlike the case for men, use of average weeks worked for those persons less than age 35 does not increase the proportion of variation explained. Restricting attention to the low educated single females without children reveals that a greater proportion of the variation in usage can be explained during the 1981-95 period using the weeks worked measure for all persons (47%). The proportion of variation in usage during the 1989-95 period that can be explained is considerably greater for the model using persons of all education levels (49%-50%). However, for these years, the explainable proportion of variation in usage is lower using the sample of low educated persons and the measure of weeks worked is defined as the average among persons of all ages. Results for single mothers (Figure 3.12 and Table 3.6) indicate that 33% of the variation in usage during the 1981-95 period can be explained by inclusion of the three covariates. Again, as was the case for women without children, use of the measure of average weeks worked for those persons less than age 35 does not increase the proportion of variation explained. Restricting the sample to single mothers with only children 7-17 years of age does not change the amount of usage explained but the three covariates, but for the sample of single mothers with only young (less than 7 years of age) children the proportion of usage explained increases to 41%. During the 1989-95 period, this proportion increased to 53%. Additional decompositions restricting a single covariate at a time allow for the determination of the relative importance of each of the three covariates. The reductions in variation of the usage series from these restrictions are also presented in Table 3.6. These results provide further evidence of the relative importance of labour market conditions in explaining welfare usage, particularly during the 1989-95 period. 30 3.6 Conclusions This chapter has documented the use of social assistance by Canadians in recent years using a two-fold approach. First, an initial investigation was conducted employing an aggregate approach that jointly examines all types of welfare recipients, making no distinctions whether or not persons are disabled or family structure. Results suggests that while economic analysis may appear on the surface to allow for statements to be made attributing the observed increase in usage in Canada to changes in relative benefit rates and labour market conditions , closer attention raises questions about this conclusion. It is reasonable to suspect that households who differ in terms of their physical ability to work and family structure will exhibit different responses to changes in benefit rates, labour market conditions and UI availability. This is particularly true when one considers that households with different characteristics are eligible for different levels of welfare benefits and are likely to face different labour market conditions. These concerns are addressed in the second approach pursued in this chapter, which examines the use of social assistance by unmarried individuals in five Canadian provinces over the period 1981-95. Microdata is employed to construct welfare usage rates and measure of labour market conditions specific to sub-groups of individuals. Welfare benefit rates specific to classes of recipient are obtained from published benefit rate schedules. Separate analysis of these data by type of recipient, namely, single men without children, single women without children and single women with children, reveals important differences in behaviour across these groups. First, consider individuals without children. Although the majority of such individuals, regardless of gender, have low levels of education (high school or less), men are more heavily represented in the younger age categories. Accordingly, use of measures of labour market conditions specific to men less than 35 years of age is appropriate. The corresponding measure for single women without children represents all such women. When all individuals, regardless of education level are included in the sample, results indicate that social assistance usage is mildly significantly affected by the relative 31 generosity of benefits, but only for females. Usage by both genders is significantly affected by employment conditions and the availability of unemployment insurance, however, these effects are strongest for women. When the sample is confined to the less educated, coefficients for relative benefits and UI qualification weeks are no longer significant for females - only employment conditions significantly affect welfare usage. The absence of significant UI effects may reflect a weaker attachment to the labour force for this group. For males, coefficients on both weeks worked and UI effects remain significant and increase in magnitude. This is especially true of the latter. This result is consistent with documented declining labour market opportunities for low skilled young men [Beaudry and Green (1997)]. The decompositions for singles without children indicate that there is more variation in the year effects for males than those for females. Furthermore, for females, there is more variation in the year effects for the less educated. Inclusion of the 3 main covariates explains much of the year effects for the samples with all singles (51% and 35% for males and females respectively) and even more is explained for those with less education (56% and 47% for males and females respectively). Much of this can be attributed to variation in labour market conditions; this is particularly true of the 1989-95 period. The use of social assistance by single mothers is significantly affected by both relative benefits and employment conditions. This result is somewhat counter-intuitive. One might expect that individuals who are the sole source of support for a child would be constrained with respect to their ability to accept employment and thus be less responsive to changes in relative benefits. However, this results holds both for sub-samples of single mothers with only young children (less than age 7) and single mothers with only older children (ages 7-17). In all cases single mothers are more responsive to changes in relative benefits and employment conditions than singles without children. Perhaps these effects reflect the increased urgency of bringing home sufficient funds to support a child. They may also reflect the more complex benefits package available to single mothers relative to that for singles without children. The decompositions for the single mothers indicate that inclusion of the 3 main 32 covariates explains 33% the year effects. For the sample of single mothers with only young children inclusion of the three covariates explains 41% of the variation in usage over the 1981-95 period and 53% over the 1989-95 period. As for the other groups, most of this is attributable to variation in labour market conditions, particularly during the 1989-95 period. These results have important policy implications and underscore the importance of separate examinations for each type of recipient. Fortin and Cremieux (1998) find that at an aggregate level welfare usage is significantly affected by benefit generosity, labour market conditions and adequacy of the UI program, however, I find evidence that these conclusions to not apply equally to all types of recipients. The near complete absence of any significant response to changes in relative benefits for singles without children casts doubt on the belief that changes in benefits have driven changes in the welfare caseload for this group. On the contrary, the evidence suggests that growth of involvement with welfare has been the result of reduced labour market opportunities and tightening of eligibility for unemployment insurance. If it is the case that there is a "hard-core" group of individuals who lack the necessary social and/or work skills to maintain gainful employment, then policies aimed toward "forcing the undeserving poor back onto the right track" will only inflict punishment and result in decreased standards of living. Given that the stated objective of the welfare system is to provide a social safety net for those persons in need, such policies are counter-productive. The results for single mothers have less clear implications for policy. The magnitude and significance of the coefficients on relative benefits and weeks worked suggest that single mothers have a stronger involvement with labour force than might be suspected. However it is worth noting that changes in the availability of unemployment insurance does not significantly influence welfare usage by this group, suggesting this attachment is relatively weak. 33 Chapter 4 Duration Analysis of the Transition from Employment to Welfare 4.1 Introduction In the previous chapter, I employed aggregate level data to investigate the forces responsible for the upsurge in welfare receipt observed in recent years. Results indicate cyclical factors are important when the right measure is used, but there is only qualified support regarding the importance of the availability of UI benefits and the relative generosity of welfare benefits. While this is a step forward, we need to better understand the mechanisms underlying these trends. An increase in the total caseload in a month can arise either because duration of welfare spells have been increasing, the rate of movement onto welfare has been increasing or some combination of the two. There has been substantial evidence accumulated on the first, but very little is known about the second. This focus on the duration of welfare spells and the exit rate from welfare is understandable since the main data used so far is administrative data which is usually limited to the information required to administer the program and has little information regarding individuals' behaviour before their involvement with welfare. In this chapter I address this gap in our knowledge examining flows into social assistance using individual level survey data to examine the path leading individuals to social assistance receipt. This path is characterized by three processes. The first process considers the job loss experienced by an individual. The second process considers an individual's transition to re-employment following a job separation. The third process considers an individual's transition to welfare receipt, given that a job separation has occurred. Using the 1988-90 longitudinal file of the Labour Market Activity Survey I estimate semi-parametric duration models to determine how these processes are affected by incentives in welfare benefits, labour market conditions, availability of unemployment insurance as well as demographic variables including level of education. The current study makes a substantial contribution to our knowledge of the path leading to welfare receipt. It examines the entire path from the start of a job until its 34 completion and onward until an individual is observed to take-up welfare or obtain re- employment. To the author's knowledge it is the only Canadian study that specifically examines the length of time until welfare take-up following a job separation. Furthermore, the time period examined contains substantial welfare reforms that make it well suited to investigate the incentive effects present in welfare benefits rates. In 1989 Quebec replaced the Social Aid Act with the Act Respecting Income Security. Under this legislation the Social Aid program was replaced by two new programs: the Financial Support Program for people with severe disabilities and the Work and Employment Incentives program (WEIP) for everyone else. These became effective for new applicants on August 1, 1989 and all applicants as of August 1, 1990. Also created was the Parental Wage Assistance program for low income families with children to replace the Work Income Supplement Program that covered low income workers. Under WEIP individuals may be categorized as not-participating, participating, not-available and available. Classification depends on their willingness and ability to participate in vocational training, job search assistance, work in community agencies or subsidized employment. Couples may be placed in a mixed category. Some recipients saw their benefits increased while others had them reduced. However, a dramatic change was experienced by childless singles and couples less than 30 years of age. Prior to this date these individuals were paid a much lower benefit rate than similar individuals aged 30 years and older. As a result of the new act, this differential was eliminated and benefits paid these younger individuals more than doubled. 1 This increase in benefits provides an exogenous source of variation which may be analyzed as a natural experiment.2 In 1986 the Ontario Ministry of Community and Social Services announced the appointment of the Social Assistance Review Committee (SARC) to undertake a comprehensive review of the welfare system. This resulted in 274 recommendations for welfare reform (see "Transitions", SARC 1988), with the first of these phased in between October 1989 and January 1990. These included the introduction of the Supports to Employment Program (STEP) on October 1, 1989. Among the reforms included as part 1 As a result of the reforms, individuals classified as Available had their benefits increase from $185 to $487 in nominal terms. 2 B. Fortin and Lacroix (1998) exploit this in their analysis of welfare spell durations in Quebec. 35 of STEP earnings was the standardization of all earnings exemptions at 20% of net earnings beyond the basic levels and the introduction of flat-rate exemptions for training allowances. On January 1, 1990 there was an increase of 6% in basic welfare rates and the implementation of a revised system of shelter allowances, with a base allowance provided regardless of actual shelter costs and a supplementary allowance to cover actual costs up to designated ceilings. The combination of these two changes had the result of increasing the maximum potential benefits available to individuals on social assistance approximately 15-17%.3 Less dramatic increases in nominal welfare benefits were instituted in B .C. in July 1989 and August 1990 (approximately 7% and 6% respectively). This variation in benefit rates, particularly the former two, will be examined to determine the incentive effects of social assistance benefits on the transition out of employment and into welfare receipt. The absence of major reforms in other provinces in the 1988-90 allows individuals in these provinces to serve as a control group. It should be noted that for the purposes of this study I am abstracting from all dimensions of the welfare system other than benefit rates. Inclusion of controls for province will capture the effects of these omitted factors to the extent that these dimensions are unchanged over the sample period. Of course, this is not the case for the STEP reforms, and the effect of these reforms will be explored in future work. 4.2 Theoretical Framework The focus of this paper is to examine the path leading individuals to social assistance receipt. This path is characterized as consisting of three processes. The first process considers an individual's transition out of employment. Individuals experiencing such a transition are deemed to be at risk of taking up welfare. The other two processes concern an individual's behaviour given they are at risk. These processes may be viewed within a stylized search model framework. Specifically, employed individuals receive a wage equal to their marginal product. 3 The specific magnitude of the increase depends on family type. 36 This wage is subject to exogenous productivity shocks arriving at random intervals. Upon receipt of a productivity shock workers re-evaluate the optimality of continuing employment at the new wage rate vs. the alternative option of engaging in unemployed job search.4 Unemployed individuals receive job offers at random intervals. While unemployed, individuals receive non-labour income which may be interpreted as UI benefits (provided they have sufficient insurable weeks to qualify) and/or social assistance benefits. The receipt of social assistance benefits is dependent on the level of UI benefits and the level of assets held by the individual.5 The decision between continuing employment and engaging in unemployed job search is based on a comparison of household utility under each alternative. Within this model there are three processes at work. The first process generates exits from employment. Conditional on having experienced a job separation the second process generates entry into re-employment and the third generates entry into welfare receipt.6 Note that the re-employment process is assumed to be independent of the welfare take-up process. Re-employment entry rates will be conditional on the level of welfare benefits available, but not on having experienced welfare. This stylized model suggests the following reduced form specification for the exit rate from employment into non-employment (N)7 hN(t) = h[UIregs(t), SAregs(t), employment cos ts(t), wage{t), 9{t), X(t)] (4.1) The relative attractiveness of non-employment will increase with changes in UI regulations that increase either the level of benefits or the length of the benefit period, or reduce the number of weeks of employment required for qualification. It will be similarly influenced by changes regarding welfare eligibility and benefit levels. Such changes are expected to increase the exit rate from employment. Alternatively, increases in the overall 4 Alternatively this may be viewed as an institutional rigidity preventing wage decreases, resulting in layoffs. 5 In practice the level of SA benefits also depends on demographic characteristics and the composition of the household. 6 This is not a competing risks model as an individual can become re-employed after welfare take-up. 7 The data does not allow one to distinguish individuals who are actively searching for employment vs. those who do not. Accordingly, the alternative state to employment will be denoted as non-employment. 37 level of wages are expected to decrease the exit rate. Higher recurrent costs of employment (such as day care) decreases the relative attractiveness of employment and thus increase the exit rate. The arrival rate of productivity shocks is denoted by 0 t and the arrival rate of job offers when non-employed is denoted by Xt. A higher 6 t implies that a worker's marginal productivity will be re-evaluated more often, thus increasing the probability of a negative wage adjustment. Accordingly, higher values of 0 t will increase the exit rate. Once non-employed, a higher arrival rate of job offers affords the individual more opportunities to receive a wage offer of sufficient magnitude to make employment attractive. Therefore, increases in A,t will increase the exit rate. Note that the exit rate out of employment is influenced by the arrival rate of job offers, although job offers are not received when employed. The re-employment hazard may be specified as hE(r) - h[UIregs(t),SAregs(t),employmentcosts(t),wage(t),A(t)] (4.2) where all arguments with the exception of A,t have the opposite effect to those outlined for the exit into non-employment. The entry rate into welfare (W) may be expressed as hw it) = h[UIregs(t), SAregs(t), X(t)} (4.3) Higher levels of UI benefits reduce the likelihood that UI income will be "topped up" with welfare income, thus decreasing the entry rate into welfare. Longer durations of UI benefit receipt and fewer weeks of work necessary to qualify for UI benefits will also decrease the welfare entry rate. Higher levels of welfare benefits makes it more likely that a person will qualify for social assistance and thus positively affects the entry rate. Furthermore, higher levels of benefits increase the probability that welfare income will compensate individuals for any stigma costs associated with welfare receipt. 8 Stigma costs of welfare receipt were introduced into the standard static labour supply model in Moffitt (1983). 38 Ideally, the researcher would know the exact UI and welfare regulations and schedules etc. that each individual would face as well as their employment costs. Since this is unknown, the reduced-form hazards are implemented using the following empirical specifications.9 hN(t) = m(t), d(t), s(t), w(t), (9(0, A(r), ep(t), q(t), k(t), x(t) (4.1a) h^t)=m(t),d(t)XOMt),W,ep(OMOM),x(t) (4.2a) hw{t) = m(t), d(t), s(t), A(t), epit), q(t), k(t), x(t) (4.3a) The precise level of potential UI benefits is unknown, however I characterize the UI system using measures of the number of weeks of insurable employment required to qualify for UI benefits (mt) and the potential duration of benefits payable to someone with 20 weeks of insurable employment (dt). The level of potential welfare benefits (from published sources) in terms of equivalent adults in the family is represented by st. The overall level of wages is proxied by the minimum wage in the province, denoted wt. These hazards may be modified to take into consideration cyclical effects. Labour market conditions are likely to affect the job offer arrival rate and can be represented through the inclusion of a measure of the employment-population rate (ept). Similarly, seasonal variation in the offer rate can be captured through the inclusion of quarterly dummies (qt). Demographic characteristics such as the level of educational attainment (kt) may also affect the arrival rate of job offers and also the level of the offered wage. Additional demographic characteristics may also affect the arrival rate, and are denoted xt. 9 Data limitations do not allow for the estimation of the UI take-up process. 39 4.3 Data 4.3.1 LMAS The 1988-90 Labour Market Activity Survey (LMAS) data set is the primary source of data used in this analysis. The LMAS follows an initial cohort for 2 years (1986-87) and a second cohort is followed for 3 years (1988-90). These data include a wealth of job activity information pertaining to all jobs worked (up to five jobs in a given year) as well as demographic information. It also includes information on social assistance receipt and of particular importance to this paper, this information is recorded on a monthly basis for 1989-90.10 However, the LMAS has several limitations regarding its use in the current study. First, there are the limitations which stem from the fact that the L M A S is a survey and as such is subject to response error. As discussed in section 3.3.1 there is substantial under-reporting of welfare receipt in the Survey of Consumer Finances data and this is also true of the LMAS. Furthermore, the LMAS is only a sample. Given that only a fraction of the population report S A receipt, this severely limits the number of observations where individuals report receiving benefits. This has important implications for the current study. Results from chapter 3 and Stewart and Dooley (1998a), highlight the importance of examining different types of households separately when investigating participation in welfare. However, this is not feasible given the size of the available sample. Accordingly, the analysis is conducted pooling three different types of households: singles without children (SNs), couples without children (CNs) and couples with children (CPs).11 The necessity of pooling the data motivates the exclusion of lone parents from the sample as their behaviour is likely to be very different from other types of households. However, differences in behaviour by household type are permitted by including family type as a covariate. Second, the LMAS is completed by individual members of a household but there is no information regarding specific job patterns or earnings of spouses. Furthermore, it is not possible to match up respondents with spouses using the public use file. 1 0 In prior years, such information was only recorded on an annual basis. 1 1 Household type was assigned using the HHCHAR88, HHCHAR89 and HHCHAR90 variables. 40 This presents three obstacles. The first two stem from the fact that unlike unemployment insurance, social assistance is administered on a family basis. Accordingly, inclusion of couples in the analysis implies that we must be concerned with identifying shocks to family earnings, but the data only allows for the identification of shocks to individual earnings. The second obstacle concerns double counting of SA receipt. Who in the household would have responded questions regarding SA receipt? Does only the head of the household answer this question or do all members of the household? Is it a mixture of the two? Without the ability to link respondents and spouses this cannot be known with certainty.12 A third obstacle is presented by the lack of information regarding one's spouse. Until August 1, 1989 there was a differential benefit rate paid to childless singles and couples in Quebec, but the regulations state that the lower rate was paid only to couples where both individuals were less than age 30.1 have no way of knowing the age of the spouse and will therefore assume that the spouse is the same age as the respondent. 4.3.2 Identification of States The primary focus in this study is to follow individuals from the time they experienced the job separation resulting in the loss of earnings that ultimately lead to their receipt of welfare income. Inspection of the data reveals the presence of individuals who simultaneously report receipt of welfare income and low levels of earnings. This is not surprising as each province allows welfare recipients to retain a certain amount of earnings without any reduction in their welfare benefits. Use of a strict definition of states defined as non-employed versus receiving welfare benefits would result in the exclusion of these spells. To avoid this I define a threshold level of earnings that distinguishes high 1 2 It is possible to partially investigate this issue. In addition to the variables regarding individual SA receipt, the LMAS also contains variables indicating SA receipt by the family. Therefore, one can identify respondents who report family receipt of SA benefits in a given month and check if there individuals also report individual SA receipt. Restricting attention to individuals 20-64 years of age I find that among singles without children I find that this is true in approximately 86% of the cases. The corresponding figure for couples without children and couples with children are 66% and 50% respectively. Assuming there are no other adults in the family other than the respondent and spouse this implies that for approximately half the couples in the sample only one person reports individual receipt of SA. This suggests that double counting is not likely to be a severe problem. 41 earnings jobs from low earnings jobs. The L M A S records the specific week that jobs started and ended. For all paid jobs (as opposed to self-employment) there is a derived measure of hourly earnings. Also reported for paid jobs is the average quantity of labour supplied in terms of hours per day, days per week and weeks per month. These variables allow for the construction of a measure of total average monthly earnings received from all paid jobs worked in a given week and this measure is used to define transitions in and out of the low/high earnings states. The threshold level distinguishing these two states was selected as $250 per month (nominal). The sample was constructed such that once an individual was observed to take-up welfare he was no longer followed. In other words, the process into employment (high earnings state) was censored at the time of welfare take-up. This decision was motivated by the focus of this study on entry into social assistance. Similarly, once an individual was observed to re-enter the high earnings state the process into welfare was censored since the process into welfare is conditional on being in the low earnings state. Therefore, this is not strictly a competing risks model, as individuals can become re-employed after welfare take-up. However, given the manner in which the sample is constructed, re- employment and welfare-takeup may be thought of as competing risks. Employment spells are defined as beginning when a transition from the low earnings state to the high earnings state was observed. These could have begun at any week during the 1988-90 period. The end of these spells is defined as the week when the individual's earnings was observed to fall below the earnings threshold or the last week of 1990. Similarly, non-employment spells are defined as beginning with a transition from the high earnings state to the low earnings state. The lack of monthly information on welfare receipt in 1988 presented a complication in the identification of non-employment spells. For those spells beginning in 1988, it is impossible to know if welfare income was received in 1988 after the start of the spell. One approach to deal with the limitation would be to restrict the sample to those spells that began after January 1989. However due to the small number of individuals observed to participate in welfare an alternative 42 approach was followed. Consider an individual experiencing an non-employment spell that began in 1988 and continued into 1989. Suppose they did not report welfare receipt in January 1989, but did so in a subsequent month before the end of the spell. Given that the individual was in the non-employment (low earnings) state throughout the spell, but did not report welfare receipt in January 1989, it is reasonable to assume that they would not have received welfare at any time in 1988 after the start of the non-employment spell. An exception to this is the case of a person receiving welfare benefits while awaiting processing of their UI claim. Unfortunately, UI receipt is not reported on a monthly basis, so such cases cannot be identified. Use of this strategy required that the threshold level of earnings that distinguishes low and high earnings state be sufficiently low that an individual would find it financially attractive to claim welfare if their earnings were below this limit.13 This strategy results in length biased sampling. This may be illustrated by considering two spells that begin with the transition into the low earnings state at some point in 1988. Only if a spell continues into February 1989 is it possible to observe welfare take-up. Accordingly, spells starting in 1988 leading to welfare receipt will tend to be of longer durations that similar spells starting in 1989. Fortunately, there are statistical techniques for dealing with this problem. One approach suggested by Heckman and Singer (1984,1985) involves estimating a separate baseline hazard for spells that start before and after February 1989. 1 4 For the process generating entry into welfare there is an inconsistency due to the fact that job starts and stops are recorded as occurring in a particular week while receipt of social assistance is recorded on a monthly basis. This potentially presents difficulties in identifying spells in which welfare take-up is observed in the same month as a person enters into the high-earnings state. This issue is dealt with by assuming that welfare claims are initiated during the first week of the month. 1 3 Given this level, only singles without children in N.B. and Quebec (less than 30 years prior to Aug, 1989) would have benefit levels below the threshold. 1 4 Only 15% of non-employment spells begin in 1988. Although not applied in the current analysis this will be implemented in future work. 43 4.3.3 Exclusions from the Sample There are several concerns that require candidate spells to be excluded from the analysis. If a candidate spell includes a period of self-employment then it is impossible to determine total earnings with certainty since neither a derived wage or labour supplied is reported for self-employed jobs. Such spells are excluded. Disabled individuals, or those with dependents who are disabled are not subject to the same welfare benefit schedules as individuals designated as employable. Their behaviour is also likely to be sufficiently different to warrant their exclusion. Accordingly, spells are excluded for all individuals who report that they or a family member are disabled and limited at work. Individuals are excluded if they report being a full-time student at any time during the spell. Since welfare benefit rates differ across types of households, only if an individual's household type is known with certainty to be unchanged throughout the spell will their spells be retained in the analysis.15 Until August 1989 there existed a differential benefit rate in Quebec for childless singles and couples less than 30 years of age. With the replacement of the Social Aid Act with the Act Respecting Income Security in August 1989, this differential was eliminated. This event provides an unique opportunity for exploring the incentive effects of welfare benefits on the three processes in the current study. However, in the public use LMAS file, age is recorded in categories, one of which includes individuals between the ages of 25 and 34. In order to retain as many individuals affected by this change as possible, a special age variable was made available to the author which allowed for the retention in the sample of the maximum number of affected individuals.16 Age is recorded when a person rotated into the Labour Force Survey, which would have occurred during July-December 1988. In order to ensure that the appropriate benefit rate was assigned it was only possible to include those individuals whose age group was known with certainty throughout the period. For this reason, individuals reporting their 1 5 Interviews were conducted in the month of January in each of 1989-1991 to collect information about the previous calendar year. If an individual's household type changed between the interview preceding the start of the spell and the interview following completion of the spell, then the spell was excluded. 161 wish to express my gratitude to Stephan Roller for his assistance in this matter. 44 age as 29 or 30 were excluded. Also excluded were those who reported in July 1988 they were 28 years old. This resulted in a sample of 11,797 employment spells (1,678 for singles without children, 2,025 for childless couples and 8,094 for couples with children). The sample of non-employment spells consisted of 1,363 for singles without children, 1,912 for childless couples and 6,512 for couples with children for a total of 9,787 spells. Tables 4.1a and 4.1b present a detailed breakdown of sample counts.17 1 8 4.3.4 Selection of Covariates Although the relatively few number of spells (204) observed to exit into S A preclude separate estimations by family type, differences across family type can be captured through the inclusion of dummy variables. This imposes the restriction that the shape of the baseline hazard function has an identical shape for each type of family, however, the level of the function can vary across type. Exit rates may be influenced by personal characteristics. A dummy variable indicating gender is included to capture differences between men and women. Dummy variables representing different levels of educational attainment are included to allow for differences in exit rates due to differences in the arrival rate of job offers and also the level of the offered wage. Exit rates may also by affected by the age of individual. Age dummies are included to capture such effects. Job offer arrival rates will likely be influenced by labour market conditions. The employment-population rate for persons aged 20-34 years is used as a proxy to capture this effect. Furthermore, it possible that there is seasonal variation in the offer rate. This motivates the inclusion of seasonal dummies. The province specific minimum wage rate acts as a proxy for the overall level of wages in the economy. The relative attractiveness of alternative states will be influenced by the availability, duration and generosity of alternative income sources. Incentive effects in 1 7 Table A2.1 in Appendix 2 presents sample counts for spells observed to exit into welfare under different levels of threshold earnings. Results indicate that altering the threshold has very little impact on the sample. 1 8 In the remainder of the paper I will refer to the high earnings state as "employed" and the low earnings state as "non-employed'. 45 social assistance benefits are proxied by a measure of maximum potential welfare benefits. These are specific to type of household and are expressed in terms of real dollars (December 1990 $'s). In order to adjust for family size, benefits are divided by the number of equivalent adults in the household. In order for allow for non-linear effects in the benefit rate variable, an additional variable is included in the specification which equals 1 if the month to month increase in the benefit rate series is greater than 5% and zero otherwise.19 The availability of UI benefits is proxied by a measure of the minimum number of weeks of insurable employment required to qualify for such benefits. An alternative dimension of unemployment insurance is the duration of the period of benefit entitlement. This effect can be explored by the inclusion of a measure of the number of weeks of UI benefits payable to an individual working 20 weeks.20 These variables were constructed using the relevant regulation schedules and a 3 month average of the seasonally adjusted provincial unemployment rate for both sexes, 15 years and older, for the 3 months preceding a given month. Variables representing labour market conditions, seasonal effects and policy parameters (both SA and UI) are specified as time varying covariates. Plots of covariates are presented as Figures A2.1-A2.5 in Appendix 2. Province dummies are included to capture fixed and persistent province specific effects. These may include differences in welfare programs along dimensions other than benefit rates. As discussed in chapter 3, it is desirable to control for contemporaneous correlations across provinces. In this manner the effects of aggregate trends is stripped away and variables such as benefit rates are identified solely off of time series variation within provinces. In the aggregate analysis this was achieved by including year dummies in the specification. In order to control for similar effects in the current exercise, the series representing welfare benefits, minimum UI qualification weeks, UI entitlement weeks, minimum wages and the employment-population rate were de-trended using a specification that included a cubic in time and provincial dummies. The de-trended series were utilized in the duration analysis. Use of this strategy ensures that the methodologies used in the two exercises as consistent as possible. 1 9 During the sample period there are no decreases in benefit rates in excess of 5%. 2 0 Duration of UI entitlement was based on 20 weeks because this is beyond the range affected by the Variable Entrance Requirement. 46 4.4 Descriptive Statistics 4.4.1 Job Spells The distribution of job spells is presented in Table 4.2a. Of the sample of job spells ending in separations, most (61%) are experienced by couples with children (CPs) while singles without children (SNs) and childless couples (CNs) account for similar proportions (19% and 20% respectively). This pattern is essentially unchanged across spells observed to exit and censored spells. For the entire sample spells are split nearly equally between men and women (51% vs. 49%). However men are represented more heavily among singles without children (57%) whereas men are slightly over-represented (52%) among CNs and slightly under-represented among CPs. Among spells terminating in job separations men account for the majority of spells within each family type. The majority of spells are accounted for by individuals less than 35 years old (63%). This concentration among the younger age groups is strongest among SNs. Within this group, 77% and 65% of spells are accounted for by individuals less than 35 and 31 years old respectively. Spell distribution by age is similar among spells resulting in exits vs. spells which are censored. The distribution of spells by level of educational attainment exhibits a fairly even split between those with and without some post-secondary education (52% have some post-secondary education). Among spells observed to terminate, only 49% have some post-secondary education whereas 56% of censored spells are accounted for by persons with some post-secondary education. Furthermore, a greater proportion of SNs have some post-secondary education; a lower proportion of couples (with and without children) have some post-secondary education. This holds regardless of outcome. Unsurprisingly, the majority of spells are accounted for by individuals residing in Ontario and Quebec (35% and 29% respectively). Censored spells are more heavily represented in Ontario. Among SNs, spells are relatively concentrated in Alberta and B.C., whereas among CPs spells are relatively concentrated in Newfoundland and New Brunswick. 47 4.4.2 Non-Employment Spells Table 4.2b presents non-employment spell distributions. The distribution of spells across the 3 types of households is 18%, 22% and 60% for singles without children, couples without children and couples with children respectively. For this sample there exists considerable variation in distribution across household type for those observed to exit vs. those which are censored. This is particularly true of spells terminating in welfare receipt, where spells are substantially over-represented by SNs. A s was observed for job spells, non-employment spells are concentrated among the younger age groups, however not as severely (55% of spells are accounted for by persons less than 34 years of age vs. 64% of job spells). SNs have the highest concentration of spells by individuals in this age group. Relative to spells that exit into re- employment, spells that exit into welfare are slightly more likely to be associated with persons in lower age groups (46% of spells terminating in re-employment are accounted for by individuals less than 31 years old vs. 50% of spells terminating in welfare). Wi th respect to education level, spells are less likely to be accounted for by individuals with low levels of education for SNs (40%) and more likely for CPs (57%). A s a whole, spells that exit into re-employment follow a similar pattern (34%, 46% and 49% for SNs, C N s and CPs respectively). Spells that exit into welfare are more than proportionately represented by persons without any post-secondary education (65% vs. 53% for all spells). Furthermore, among those spells that exit into welfare, spells accounted for by SNs are the most likely to be associated with individuals with low levels of education (70%, 57% and 62% for SNs, C N s and CPs respectively). The distribution of total spells across provinces is essentially the same as observed for job spells. There is a high concentration of spells terminating in welfare receipt in Quebec and Ontario (46% and 23% respectively). Spells exiting into S A in Quebec are accounted by 63%, 41% and 36% for SNs, C N s and CPs respectively. In Ontario the corresponding proportions are 12%, 23% and 32%. Among CNs , spells terminating in welfare receipt are also concentrated in Alberta (25%). 2 1 Distributions of spells observed to exit into welfare for alternative levels of the low-high earnings threshold are presented in Table A2.2 in Appendix 2. 48 4.5 Empirical Specification and Empirical Hazard Functions 4.5.1 Empirical Specification The empirical analysis is conducted in a hazard function framework. The analysis involves first estimating the empirical hazard rates and the corresponding survival probability functions. The estimate of the hazard rate at time T is calculated as the number of spells terminating at time T as a proportion of spells at risk of terminating at time T. The corresponding survival probability function is defined as the proportion of spells that are at least T months in duration. The empirical hazard rate estimator is limited in that it treats the population as homogenous. The length of spells may vary with the characteristics of the recipient, UI and SA policy parameters and labour market conditions. To control for these effects a proportional hazard duration model is used which specifies the hazard rate as where hj(f) is the hazard rate for person i, ho(t) is the baseline hazard common to all individuals, Zj(t) is a vector of observable characteristics which may vary with t, and p is a parameter vector to be estimated. For different values of Zj(f)p\ the hazard function for individual i is shifted proportionally up or down relative to the baseline. The estimation approach implemented in this paper is an extension of Prentice and Gloecker (1978) and is discussed in Meyer (1990) and Lancaster (1990:172-208). The baseline hazard is estimated nonparametrically as a piece-wise constant function. The time axis is divided into a finite number of intervals and a separate baseline hazard parameter is estimated for each interval. This allows for a very flexible method for estimating the baseline hazard function and avoids the pitfalls associated with the imposition of a parametric functional form on the baseline. Effects of misspecifying the baseline hazard include errors in drawing inferences regarding the presence of duration dependence (Manton, Stallard and Vaupel, 1986; Blank, 1989) and the impacts of (4.4) 49 covariates (Heckman and Singer, 1985; Dolton and van der Klaauw, 1995). This technique also can be easily extended to allow for the effects of unobserved heterogeneity, either parametrically or non-parametrically [Meyer (1990)].22 Derivation of the likelihood function for the piece-wise constant proportional hazard model proceeds as follows. The probability that a spell last at least until time t+1, given that it has lasted to time t, is given by P[Tt <t + l\T, >r] = exp[-exp(r(0 + z , M V )} (4.5) where y(t) = [ ̂ h(u)du ] . The log likelihood for a sample of N spells is then X ^.log(l-exp[-exp(KA:() + ̂ fe,)V)])- Z exp(K0+ *,(')'/?)] (4-6) /•=i ;=i where kj is the observed length of the ith spell, 8j equals one if the spell terminates before being censored, and 8i equals zero if the spell is censored. Implementation of this model requires that spells are censored at a specified duration and that the baseline is divided into intervals. The hazard rate is assumed to be constant within each interval. Job spells with durations in excess of 129 weeks were censored. For the sample of non-employment spells, all those with durations in excess of 77 weeks were censored. The baseline intervals selected are presented below. 2 2 Allowing for unobserved heterogeneity in the model will be pursued in future work, as will joint estimation of the re-employment and welfare take-up processes (See Gilbert et. al, 2000). 50 Definition of Intervals for Baseline Hazards Job Duration Time to Re- employment Time to Welfare Baseline Interval Baseline Interval Baseline Interval Range Length Range Length Range Length (weeks) (weeks) (weeks) (weeks) (weeks) (weeks) 1-40 1 1-40 1 1-6 2 41-60 2 41-52 2 7-10 4 61-90 3 53-55 3 11-15 5 91-125 5 56-70 5 11-30 15 126-129 4 71-77 7 31-50 20 51-77 27 These intervals were selected as to afford the greatest flexibility possible while ensuring sufficient number of observations in each segment to adequately identify the baseline hazard. Inspection of the empirical hazard functions provided additional guidance. 4.5.2 Empirical Hazard Functions Figure 4.1a presents the empirical hazard for the sample of job spells. During the initial weeks of the spell the hazard increases sharply, then maintaining relatively high levels up to week 18. The hazard displays spikes at weeks 10 and 14 which are strongly suggestive of effects resulting from UI qualification requirements.23 After this point the hazard displays a downward trend until week 45, after the hazard declines moderately. The volatility exhibited in the hazard towards the end of the time frame likely reflects increased variability resulting from limited exits in this region. The corresponding empirical survival function is presented in Figure 4.1b. Alternatively, empirical survival probabilities are presented in Chart 4.1. The probability of remaining in a job is seen to initially decline relatively quickly, after which it declines much more moderately. Three months after the start of the job 22% have experienced a job separation. At 6 months this figure increases to 40% and after 1 year 54% have had a job separation. After 2 years only 31% remain employed in their original job. 2 3 During 1989 the minimum number of weeks to qualify for UI ranged from 10-14 depending on the 51 Chart 4.1: Empirical Survival Probabilities Week Job Time to Time to Duration Re-employment Welfare 1 0.996 0.812 0.998 4 0.943 0.672 0.993 13 0.775 0.560 0.984 26 0.600 0.459 0.978 39 0.508 0.385 0.971 52 0.459 0.319 0.963 65 0.416 0.276 0.944 77 0.380 0.250 0.937 91 0.337 104 0.310 117 0.292 129 0.271 The empirical hazard function for the time until re-employment is presented in Figure 4.2a. After experiencing a job separation there is a 19% probability of obtaining immediate re-employment. U p until week 8 the probability of exit decreases sharply, after which the hazard is essentially flat. The empirical survival function presented in Figure 4.2b displays a corresponding sharp decline in the initial weeks following separation, after which it declines at a moderate, albeit increasing rate. Four weeks after separation the survival probability has dropped to 67% and after 4 months the probability of not having left non-employment is 46%. After 1 year 32% remain non-employed and at the censoring point (1.5 years) 25% remain non-employed. The empirical hazard functions for the time until welfare receipt (Figure 4.3a) illustrate the effect of the small number of spells in the data that terminate in welfare take-up and the necessity of pooling the data. 2 4 The hazard is seen to decline up to week 23 after which there is a range in which there is a higher probability of exit. Finally there is a region (week 50-70) where the probability of entry into welfare is substantially higher. After 3 months of non-employment 98.4% of individuals have not received regional unemployment rate. For most of 1990 (until December) the minimum number of weeks was 14. 2 4 In order to more easily see the overall pattern in the hazard function, the series represents a 5-week moving average of the actual series. 52 welfare income. After 1 year only 3.7% of individuals have received welfare benefits and at the censoring point (1.5 years) 6.3% of the those non-employed have received welfare benefits. 4.6 Duration Models Models were estimated for each of the three processes that together characterize the theoretical model. Estimation results are presented in Table 4.2. The corresponding baseline hazard functions are presented in Figures 4.4-4.6 and the estimated survival probabilities are presented below in Chart 4.2. 4.6.1 Baseline Hazards Inspection of the baseline hazard function for job spells (Figure 4.4) reveals that probability of exit from employment declines initially (until approximately week 40) after which is essentially constant. (This provides support for the hypothesis that the increased volatility exhibited by the empirical hazard at high durations is in fact the result of the sparseness of exits observed in this range.) As seen in Figure 4.5, following a job separation the probability of re- employment declines very quickly. The estimated baseline hazard for time to re- employment indicates that during the first week following a job separation the probability of re-employment is 24%. In week 2 this falls to 14% and in week 3, 5%. The U-shaped pattern exhibited by the empirical hazard function for the time to welfare sample is reproduced in the estimated baseline hazard (Figure 4.6). The exit rate into welfare receipt generally declines up to week 10 and then remains constant until week 30 (recall that there is a breakpoint between baseline intervals at week 16). Probability of exit is higher for the week 30-week 50 interval and much higher in the interval beginning with week 50. The higher probability of exit in the week 30-50 range could reflect welfare take- up by individuals exhausting UI benefits. The higher exit rate in subsequent weeks 53 suggests welfare take-up following exhaustion of all personal financial resources and may also reflect a stigma effect. However, as indicated in Figure 4.6, the standard errors of the baseline hazard estimates for the welfare take-up process are sufficiently large that we cannot reject the hypothesis of equality. This undoubtedly reflects the small number of spells observed to end in welfare take-up. Chart 4.2: Duration Model Survival Probabilities Week Job Time to Time to Duration Re- Welfare employment 1 0.998 0.763 0.997 4 0.963 0.598 0.989 13 0.850 0.470 0.976 26 0.723 0.362 0.964 39 0.651 0.283 0.946 52 0.609 0.211 0.919 65 0.566 0.164 0.867 77 0.533 0.137 0.822 91 0.494 104 0.464 117 0.443 129 0.420 4.6.2 Marginal Effects The estimated coefficients for the models are presented in Table 4.2. As can be seen, the estimated coefficients for the "female" variable are negative for all 3 processes. This implies that women tend to take longer to exit any given state. Neither singles without children or childless couples take significantly longer than couples with children to exit employment. Singles without children take the least time to become re-employed and to take-up welfare once non-employed. Once non-employed, couples without children take less time to become re-employed than couples with children, however, they take longer to take-up welfare. Longer times to re-employment among couples with children could reflect the need to consider recurrent employment costs, such as day care. Shorter times until welfare take-up could represent an unwillingness to put-off welfare take-up due to stigma effects because of the needs of the children. Higher levels of educational attainment are generally associated with lower job 54 exit rates and re-employment rates generally increase with education. Hence, more educated individuals have longer job durations and shorter times until re-employment. Shorter times until welfare receipt are usually associated with lower levels of education, with the exception of those who only have a high school degree. It is these persons who take the longest to take-up welfare following a job separation. Higher levels of education wi l l be correlated with higher wages, which in turn would likely be correlated with higher levels of savings. Accordingly, increased time until welfare take-up among the more highly educated could reflect a longer period being required to run down assets before becoming eligible for welfare. Similarly, higher wages w i l l generally imply higher levels of UI benefits, which lengthens time until take-up. Individuals less than 35 years o f age experience shorter job durations, as do individuals over age 54. Time until re-employment is positively related to age, as is time until welfare receipt. Older individuals w i l l likely have higher levels of savings, both because of higher wages and life-cycle effects. A s above, this implies longer time until welfare take-up. With the exception of Saskatchewan, individuals l iving in provinces outside Ontario experience shorter job durations (particularly in Newfoundland, N e w Brunswick and Alberta) and after a job separation take more time to become re-employed (particularly in Quebec, Newfoundland and N e w Brunswick). Once non-employed, only in Alberta is the expected time until welfare take-up significantly longer than in Ontario. There is evidence of significant seasonal effects in job exit rates, increasing toward the end of the calendar year. Re-employment rates also display some seasonal variation with higher entry rates in the second quarter of the year. There is no evidence of seasonal effects in the welfare entry rate. A s expected, I find that job duration is positively affected by the increases in the number of weeks required to qualify for U I benefits. Additionally, I find that weeks of U I entitlement significantly affects time until re-employment, however, the sign is contrary to expectations. The results suggest that higher weeks of U I entitlement are associated with shorter times until re-employment. This apparent contradiction may be reconciled by considering the source of variation being used to identify this effect. Given the 55 employment-population rate is included as a regressor, the estimation is comparing people with the same employment-population rates, but those with higher unemployment rates (since higher unemployment rates are directly linked to higher UI entitlement weeks) are becoming re-employed faster. This suggests that these people may be seasonal workers. Changes in weeks of potential UI benefit entitlement does not significantly affect increase the time until welfare take-up. Labour market conditions do significantly affect job durations and the time until re-employment following a job separation. However, the exit rate from employment is seen to be positively affected by higher employment-population unemployment rates. This result is at odds with evidence found by Jones (1993). In his study of the cyclical and seasonal properties of gross flows of labour in Canada he found that the hazard rate out of employment into non-employment was counter-cyclical, rather than cyclical as indicated here. This also conflicts with Green and Sargent (1998) who found that exit rates from employment are positively related to unemployment rates. Labour market conditions do have the expected significant effect on the re-employment rate; in favourable economic times individuals experience shorter times until re-employment. Perhaps surprisingly, changes in the employment-population rate do not significantly affect time until welfare take-up, although Browning et al. (1995) find a similar finding regarding the probability of observing welfare receipt for UI exhaustees and its relation to unemployment rates. However, it should be remembered that the welfare take-up process is estimated independently of the re-employment process, with exits to re-employment treated as censored. In this framework, time until welfare take-up becomes a question of financial resources other than wage income, in which case this result is not surprising after all. There is no evidence that generosity of welfare benefits affects job durations. However, benefits do have a statistically significant impact on individuals' behaviour after experiencing a job separation. Higher benefits increase the time until re-employment and decrease the time until welfare take-up. Lastly, the overall level of wages in the economy, as represented by the minimum wage level, is seen to increase job durations. 56 4.7 Conclusions In this chapter I employed duration analysis to examine the path leading individuals from employment to welfare receipt; this path is characterized by three processes. The first process considers the job loss experienced by an individual. The second process considers an individual's transition to re-employment following a job separation. The third process considers an individual's transition to welfare receipt, given that a job separation has occurred. The evidence presented in this study must be considered in light of the limitations of the data, primarily the small number of spells observed to exit into welfare and the short sampling window for the data. Nevertheless, the results provide useful insights into these processes. Examination of the baseline hazard functions show that the probability of a job ending in a given period, conditional on lasting until that period is significantly higher during the initial 4 months than in subsequent periods. Furthermore, impacts reflecting the number of weeks to qualify for unemployment insurance are clearly visible. Results indicate that the conditional probability of re-employment following a job separation is initially relative high, but decreases very rapidly during the first few weeks. The conditional probability of welfare take-up following a job separation displays a much different pattern, initially declining but then stabilizing and finally returning to a higher level. This pattern could reflect the time necessary for individuals to run down assets levels sufficiently to qualify for welfare. Although, I do not find evidence that this patterns reflects the time necessary to exhaust UI benefits although such a story would be consistent with the observed pattern in the hazard. It is possible that there is not sufficient variation in the variable used to proxy UI entitlement effects during this period to capture this effect. The estimated coefficients provide data on how these profiles change with demographic, macroeconomic and policy variables. Despite the small number of spells in the data observed to terminate in welfare receipt, results indicate that the models estimated are able to identify the effects represented by the covariates. Most variables are found to have the expected signs, although not all are found to be statistically significant. 57 Notable exceptions are the negative sign on the employment-population rate for the employment process and the positive sign on the UI entitlement weeks variable for the re- employment process. Of particular interest is that the level of welfare benefits does not significantly affect job duration, however, benefits do affect time until re-employment. Thus, the depth of the welfare safety net doesn't affect job duration, but once non- employed, it does affect time until re-employment. These results also provide further evidence of the importance of considering the family composition of households and levels of education when considering welfare take-up. 58 Chapter 5 Simulation of Welfare Incidence 5.1 Introduction In the previous chapter I estimated the baseline hazard for the welfare take-up process for a given reference individual and determined how this profile shifts with variations in personal characteristics, labour market conditions and the policy environment. A similar exercise was performed for the re-employment process faced by individuals who experienced a job separation. In this chapter these estimates are applied to administrative data on the number of newly non-employed in a given month (cohort) in order to simulate how many of that cohort would take-up welfare in each of the months following the job separation. This allows for the predicted incidence series to be decomposed in order to determine the degree to which predicted incidence is driven by changes in labour market conditions and policy variables. In particular, it allows for an examination of which factors can account for the dramatic growth in welfare usage experienced during the 1990s. These predicted incidence series are then compared to administrative data on welfare caseloads. 5.2 Methodology Construction of these incidence series must consider that there are two means by which individuals can exit the pool of non-employed, non-SA individuals; namely, welfare take-up and re-employment. The estimates obtained in the duration analysis were the result of independent estimations in which exits from the pool of non-employed to the alternative state were treated as censored. For example, when considering the length of time until welfare take-up, individuals who became re-employed were treated as censored. These yield the following hazard rates for employment and welfare take-up, respectively. (5.1a) 59 (5.1b) The corresponding survival functions for the employment and welfare take-up processes are The desired unconditional probability of welfare take-up must combine these functions such that both processes are considered. This is achieved in the following manner. In order for an individual to take-up welfare in a given period following job separation, he must not have taken up welfare or become re-employed prior to that period. To implement this rule, I make the assumption that in a given period the "draw" for someone to become re-employed occurs before the "draw" to take-up welfare. Hence, the probability of welfare take-up in the t t h period following a job separation is defined as Thus, the probability of welfare take-up in the first month equals pf (l) = Sf (l) hf (l), and in the second month equals pj (2) = Sf ($)Sf (2)hf (2); probabilities for subsequent months are constructed in a similar manner. A period is defined to be 1 month in duration. This is a logical choice as welfare is administered on a monthly basis, however, it necessitates an adjustment of the hazard rates for the two processes.1 The L M A S data used in the duration analysis records the start and completion dates of jobs in terms of weeks. Accordingly, the estimated baseline hazards for both processes were also specified in weeks. Thus, the estimates of the weekly baseline hazard rates must be aggregated to form monthly hazard rates. Monthly survival rates are computed from these monthly hazard rates. (5.2a) (5.2b) P:(t)=sr{t-i)s?(t)K(t) (5.3) 60 I employ a 36-month horizon following a given month to allow for contributions to the total number of individuals who take-up welfare. That is, for individuals who become non-employed in a given month, I consider only those who would take-up welfare within the subsequent 36 months. Since this horizon is longer than the 77-week horizon used in the duration analysis, I assume that the baseline hazard for the balance of the horizon after week 77 remains constant at the week 77 value. The contributions from a given "source" or "cohort" month to the 36 subsequent "destination" months are computed using the average probability of welfare take-up (averaged over those individuals retained in the province-family type sub-group) for the source month and each of the subsequent 35 months. These average probabilities are multiplied by the total number of newly non-employed in the source month to determine the number who take up welfare in each of the destination months. This procedure is repeated for each cohort of newly non-employed for all months between January 1981 and December 1995. Use of the 36-month horizon implies that only destination months starting with January 1984 w i l l have contributions to welfare take-up from the entire 36 months. Accordingly, the final analysis w i l l be restricted to the 1984-95 period. 5.3 Data To perform this exercise, I require data on the number of newly non-employed for each month in the 1981-95 period. Furthermore, I need the data necessary to compute unconditional probabilities of welfare take-up (i.e. the data series used in the duration analysis) extended over the entire 1981-95 period. The exercise w i l l be performed for each family type (singles without children, couples without children and couples with children) but restricted to the provinces of Quebec, Ontario, Alberta and B .C. Environmental and policy variables are constructed in the same manner as was done for the duration analysis. These include maximum potential welfare benefits per equivalent adult (constructed using published benefit rate schedules and information from provincial ministries), the employment-population rate for individuals aged 20-34 and UI 1 Monthly hazards were creating by taking the sum of the weekly hazards for the weeks in that month. 2 Actually a weighted average using sample weights. 61 entitlement weeks. As was done for the duration analysis, de-trended versions of these series are utilized in the simulations.3 Data on the total number of newly non-employed persons are computed from monthly individual level files of the Labour Force Survey (LFS). I define an individual to be newly non-employed if they report that they have worked in the past, but not currently employed and have been jobless for only 1 month. Individuals meeting these criteria are retained in the LFS sample. To be consistent with the sample used in the duration analysis, the LFS sample includes only individuals between 20-64 years of age who are not full time students. In most respects, this sample corresponds to the sample used in the duration analysis with the following exceptions. First of all, the L M A S sample includes 19 year olds. This is not the case for the LFS sample because in LFS data the age variable is defined such that persons aged 17-19 are grouped together. Thus, the inclusion of persons aged 19 years would necessitate the inclusion of individuals aged 17-18 years. Individuals less than 18 years old are considered to be dependants by welfare authorities; thus their exclusion is warranted. Additionally, the L M A S sample excludes individuals who are permanently unable to work, or who have a family member who is permanently unable to work. It is not possible to identify such individuals in the LFS data. Failure to exclude such individuals is mitigated by the fact that individuals in the LFS sample must have been working the previous month. As was done in the duration analysis, the final sample includes only individuals defined as singles with no children (SN), couples without children (CN) and couples with children (CP). Administrative data on welfare caseloads is available from volumes of the Inventory of Income Security Programs (IISP). These provide data on the number of social assistance cases at the end of March, June, September and December. Published volumes provide this data for the June 1982 to December 1992 period. Additionally, this information was obtained for selected provinces for the period from January 1993 to December 1995 from Human Resources Development Canada. These data have not as 3 Plots of these covariates for the simulation period are presented in Figures A3.1-A3.3 in Appendix 3. 62 yet been released in a volume of IISP and should be considered preliminary. To varying degrees, this information is available disaggregated by family type. For all provinces separate series are presented for singles without children and singles with children. However, only for Quebec and B.C. are separate series presented for couples without children and couples with children; for Ontario and Alberta all cases for couples are grouped together. The duration analysis focuses on employable individuals, excluding those unable to work due to some form of disability. This presents a complication. IISP data does not easily allow for the construction of caseload series for non-disabled individuals disaggregated by family type for all provinces. However, it is possible to estimate these series. For Alberta and British Columbia the construction of these series is relatively straightforward. Volumes of IISP report cases in Alberta as belonging to one of two programs: the Social Allowance program for non-disabled individuals or the Assured Income for Severely Handicapped (AISH) program for disabled individuals.5 AISH series are disaggregated between singles and families. This permits the construction of series presenting the number of welfare cases for non-disabled individuals who are (a) singles without children and (b) couples, (with and without children combined). B.C. spans two distinct regimes. The Guaranteed Available Income for Need (GAIN) program was replaced in April 1990 by Programs for Independence. Under GAIN, non-disabled individuals are assigned to GAFN Income Assistance and disabled individuals are assigned to GAIN Income Assistance and Supplementary Programs. Under Programs for Independence, non-disabled individuals are assigned to the Temporary Assistance Program, while disabled individuals are assigned to the Income Assurance Program. This data allows for the disabled caseload to be disaggregated by my 3 family types throughout the 1984-95 period. For Quebec and Ontario, the matter is more complex. In Quebec, the sample period also spans two separate regimes. The Social Aid Act was in effect up until August 1, 1989 when it was replaced with the Act Respecting Income Security. For the years 4 1 wish to thank Anne Tweddle for her assistance in this matter. 5 These includes Modified AISH 63 covered by the Social A i d Act , disabled and non-disabled individuals were not placed in separate programs (at least the data is not presented in this manner). However, IISP does present a breakdown of the aggregate caseload (not distinguished by family type), as of March 31 of each year, by reason for assistance. This allows for the determination of the aggregate disabled caseload as of March 31, in each of these years. Under the Ac t Respecting Income Security disabled individuals were assigned to the Financial Support Program and IISP presents quarterly caseload data for these individuals. 6 Therefore, it is possible to construct a series representing the aggregate number of cases for disabled individuals as of March 31 (with the exception of March 1990 - data for this year is not reported). During the sample period welfare recipients in Ontario were assigned to either General Welfare Assistance ( G W A ) or Family Benefits ( F B A ) . Long-term unemployable individuals were ultimately assigned to Family Benefits, but were usually assigned initially to G W A . For each program IISP reports the distribution of cases by reason of assistance as of March 31 each year. Thus, as in the case of Quebec, it is possible to compute an estimate of the number of disabled cases (not disaggregated by family type) as of March 31 for each year. It should be noted that I am including cases under G W A for whom the reason for assistance is " i l l health". This is consistent with the procedure followed for Quebec where I include cases reporting either permanent or temporary disability. Although the specific distribution of the disabled caseload across family type is not known for Quebec and Ontario, such a breakdown can be estimated by applying each family type's share of the disabled caseload in B . C . to the aggregate disabled caseload for Quebec and Ontario. Specifically, during the 1984-95 period, B . C . ' s aggregate disabled caseload was comprised of singles without children (80%), couples without children (12%), couples with children (4%) and single with children (4%). Applying these proportions to aggregate disabled caseload data, and then combining with total (including disabled) caseload data allows for the computation of separate non-disabled caseload series for Quebec (singles without children, couples without children and couples with 6 1 also include cases classified as Adults in Institutions. 64 children) and Ontario (singles without children and all couples). It is not possible to estimate separate caseload series for different types of couples in Ontario because the total (including disabled) caseload groups together both types of couples. To summarize, separate data series representing the total number o f social assistance cases for non-disabled individuals are constructed for singles without children and all couples combined for all 4 provinces (Quebec, Ontario, Alberta and B.C . ) . However, separate series for couples with and without children are available only for Quebec and British Columbia. 5.4 Results The section proceeds as follows. First, predicted incidence rates w i l l be presented and discussed. Second, these predicted incidence rates are decomposed to determine the relative importance of the contribution made by different covariates to welfare incidence. Third the predicted incidence rates are compared to administrative data on the number of welfare cases and the existence of a steady state equilibrium in welfare take-up behaviour is considered. Lastly, average spell durations are derived and compared to expected durations estimated from administrative data. In order to facilitate the analysis series are transformed to represent annual data. This is required since non-disabled administrative caseload data reported by quarter is not available for Quebec and Ontario. Furthermore, the predicted incidence series exhibit substantial seasonal variation. Conversion to an annual basis removes this seasonal variation, making evaluation much easier. In order to make the discussion easier, hereafter "singles" should be interpreted as singles without children. Similarly, "couples" should be interpreted to mean all couples including both those with and those without children. 65 5.4.1 Description of Predicted Incidence Rates In this section the predicted rates of welfare incidence generated using the simulation model will be examined. Overall patterns of predicted welfare incidence rates differ substantially across provinces within the 1984-95 period. Profiles of these series are presented in Figures 5.1a-5.1c (labelled "Total Incidence").8 For the most part, the predicted incidence series exhibit very similar patterns among individuals in Ontario and Quebec. In Ontario, predicted welfare incidence was stable until 1989, then increased dramatically until 1993 (incidence increased over 400% between 1989 and 1993). Thereafter, incidence declined (approximately halved relative to 1993) but did not return to 1989 levels. This pattern was observed among individuals of all province-family type sub-groups. Similar patterns were observed among singles and childless couples in Quebec for whom predicted rates of incidence more than doubled between 1988 and 1992. These two groups experienced declining incidence after 1992, but did not return to 1988 levels. Incidence patterns prior to 1988 are distinctly different among childless couples in Quebec. Among this group rates of incidence fell nearly 50% between 1984 and 1988. Predicted rates of welfare incidence among individuals in Alberta and B .C., as well as couples with children in Quebec were substantially different from those seen in Ontario and among childless persons in Quebec. Among all these sub-groups, predicted rates of incidence exhibited a general decline during the 1984 to 1995 period. This decline was greatest during the 1980s. In B .C., between 1984 and 1990 rates of incidence declined approximately 50%, 70%o and 80% among singles, childless couples and couples with children respectively. Among couples with children in Quebec the corresponding decrease in incidence was 70%. The decline among individuals in Alberta was even more pronounced: during these years, rates of incidence decreased in excess of 80% among all three family types. 7 Unless otherwise stated, "welfare incidence" should be taken to mean rates of welfare incidence predicted by the model. 8 These figures also present other series to be discussed in the next section. 66 Between 1990 and 1992 incidence rates were relatively stable among persons in B . C . , but increased among couples with children in Quebec and persons in Alberta. This was particularly true for singles and couples with children in Alberta, for who predicted rates of welfare incidence more than doubled from 1990 to 1992. After 1992, incidence among all family types again declined in Alberta reaching levels in 1995 below those experienced in 1990. Incidence among couples with children in Quebec also declined, returning in 1995 to the same level as in 1990. In B . C . , incidence among singles continued to decline, but was relatively stable for other types of families. Chart 5.1: Average Annual Incidence (1984-95) Quebec Ontario Alberta B.C. Singles without Children 12,768 6,003 4,233 3,500 Couples without Children 8,350 4,419 1,780 1,512 Couples with Children 27,851 14,752 6,147 5,299 Although there are similarities in overall incidence patterns across most provinces and family types, levels of welfare incidence differ substantially. Chart 5.1 presents average rates of incidence over the 1984-95 period for each province-family type sub- group. Across all three family types levels of welfare incidence are considerably larger in Quebec than in other provinces. Levels in Ontario are approximately one half of those in Quebec and levels are relatively similar in Alberta and B . C . , but lower than those observed in Ontario. These results are at odds with relative differences in population across provinces. The population of B . C . is greater than Alberta (on the order of 50%) but average incidence levels are somewhat larger in Alberta. However, the most striking contradiction concerns levels for Ontario and Quebec. The population of Ontario is approximately 50% larger than the population of Quebec, and yet average incidence levels in Quebec are approximately twice those of Ontario. Nevertheless, although these differences are noteworthy, this exercise focuses on trends in predicted incidence rates, rather than levels. Therefore, this issue is not of primary concern. Comparing incidence levels across family type, we observe similar patterns between provinces. In all provinces, incidence levels are lowest for childless couples and 67 highest for couples with children. Relative magnitudes across family type are also similar: relative to levels for childless couples, levels for singles are on the order of 1.5- 2.5 times higher, and levels for couples with children are on the order of 3.5 times higher. 5.4.2 Decomposition of Predicted Incidence Rates When considering which underlying factors are responsible for these incidence series, it is useful to recall the method employed to estimate these rates of incidence. Specifically, for each newly non-employed person in a given cohort (defined by province, family type and month), the unconditional probability of welfare take-up is computed for the month that person became newly non-employed and the subsequent 35 months. For each month, these probabilities were then averaged across all individuals in a given cohort to arrive at a sequence of the probability of welfare take-up in each of the 36 months that would apply to the "average" person in that cohort. Application of these probabilities to the total number of newly non-employed yields the number estimated to take-up welfare in each of the 36 months. Therefore, the number of individuals who take-up welfare w i l l be a function of (a) the number and composition of the pool of newly non-employed individuals, (b) the economic and policy environment and (c) the parameter estimates from the duration analysis. Given that the simulation is performed separately for each province-family type pair, variation across time in the composition of the pool of newly non-employed persons for a given province-family type group w i l l reflect variation in the distribution of persons across age group and level of education. Variation in the economic and policy environment, that is, changes in welfare benefit rates, labour market conditions (employment-population rate for persons aged 20-34 years) and U I generosity (benefit weeks for a person with 20 weeks insurable employment) w i l l be reflected in the hazard rates of both the re-employment process and the welfare take-up process. In turn, these w i l l be reflected in variation in the unconditional probabilities of welfare take-up. Decomposing the predicted incidence series by specific factors is not a simple task. Inferences regarding the source of variation driving variation in the predicted incidence series can be made through comparison of the incidence series with plots of the 68 underlying covariates. Plots of these covariates are presented in Appendix 3. Although useful, this procedure is inexact. Alternatively, the predicted incidence series may be re- simulated holding a covariate, or a group of covariates, at their sample averages. In this manner, comparison of the original (unrestricted) predicted incidence series with the restricted incidence series provides information as to the impact due to inclusion of the restricted variables. That is, differences in the trends of these two series reflect the contribution made by variation in these covariates to the incidence series. Ideally, the unrestricted incidence series could be decomposed into separate mutually exclusive series that would sum to the unrestricted series. For example, the incidence series could be re-simulated holding variable A constant, and then again holding variables A and B constant. The difference between the first and third series would reflect the joint impact of variation in variables A and B . But, it would be incorrect to say that the difference between the second and third series reflects the sole impact of variable B . Due to interaction effects, the relative impact due to variables A and B w i l l depend on the order followed when restricting these variables. This arises not because covariates are interacted in the specification itself, but rather because of the non- linear nature of the hazard function. However, it is possible to make comparisons between predicted incidence series from a restricted model vs. the unrestricted model. For this decomposition exercise it is useful to categorize the covariates into three groups: (a) demographic variables, (b) macroeconomic variables (the inflow rate of newly non-employed and the employment-population rate), and (c) policy variables (welfare benefits and U I entitlement weeks). The initial decomposition w i l l focus on the impacts of the macroeconomic variables vs. impacts due to the policy variables. Accordingly, for each province-family type sub-group two additional simulations are performed, the first holding policy variables equal to their sample averages and the second holding macro variables equal to their sample averages. The predicted incidence rates from these simulations are presented together with the incidence series from the unrestricted simulation in Figures 5.1a-5.1c. In these figures, comparison between the degree of variation in the unrestricted incidence series, denoted as "Total Incidence", and the degree of variation in the series denoted "Policy Vars Constant" indicates how much variation is removed from the 69 unrestricted series when policy variables are held constant. This indicates the importance of changes in policy variables in explaining patterns in predicted incidence derived from the unrestricted model. If a factor can explain all the variation in predicted incidence then restricting that factor to equal its sample average will yield a restricted predicted incidence series that is completely flat. On the other hand, if a factor cannot explain any of the variation in predicted incidence then restricting it to equal its sample average will yield a restricted predicted incidence series that exactly tracks the unrestricted predicted incidence series. For example, in Figure 5.1a, we observe that although the total incidence series among singles in Alberta exhibits substantial variation over the 1984-95 period, the Policy Vars Constant series for this group is relatively stable; only during the 1990s does it display noticeable variation. Hence, most of the observed variation in the predicted incidence series can be accounted for by variation in policy variables. Similarly, for this group we observe that the Macro Vars Constant series closely tracks the total incidence series except for the early 1990s. Thus, changes in macro variables can only partially account for observed variation in the total incidence series and only during these years. It is important to note that this exercise involves comparing the patterns in these series, rather than the levels. For example, although the macro constant series and the total incidence series do not coincide between 1985 and 1987, the decrease in the macro constant series is mirrored in the total incidence series; hence, this decrease in total incidence during these years is not driven by changes in macro variables. Similar inferences apply to couples in Alberta, however, controlling for policy effects is not as successful in removing variation in the predicted series for couples with children, particularly after 1993. Inspection of Figures 5.1a-5.1c reveals that for sub-groups in the other provinces, the case is not as clear cut, that is, neither the policy variables nor the macro variables alone can account for all of the observed variation in the predicted incidence series. Consider the panels presenting series for Ontario. For each family type we observe that while neither the Macro Vars Constant series nor the Policy Vars Constant series are perfectly flat during the 1990s, they both exhibit much less variation than the Total Incidence series during this time period. Therefore, for all family types, holding 70 either the policy variables or the macro variables constant can account for a considerable portion of the growth and subsequent decline exhibited in the total incidence series during the 1990s. However, neither can account for all of it by itself. It is worthwhile noting that changes in policy variables in 1990 can account for the rise in incidence observed during this year for couples, but not singles. Furthermore, it appears that changes in macro variables are particularly important during the 1991-94 period, especially for singles. Impacts of macro variables on predicted incidence during the 1990s are also seen to be very important among all family types in Quebec. For these groups, the Macro Vars Constant series is very stable during this period, while the Policy Vars Constant series closely tracks the unrestricted series. In contrast, the Policy Vars Constant series for singles and childless couples is essentially flat during the 1980s. Hence, changes in policy variables can account for essentially all the variation in predicted incidence during the 1980s, but cannot explain incidence during the 1990s. But, most of the variation in predicted incidence during the 1990s can be explained by changes in macro variables. Discerning the relative importance of macro vs. policy impacts is more difficult in the case of B.C., but, the series in Figures 5.1a-5.1c suggests that controlling for macro effects is more successful in accounting for variation in total incidence during the 1990s than controlling for policy effects. The overall impact due to a covariate or group of covariates over a given period may be quantified as the difference between the coefficients of variation for the unrestricted and restricted series, taken as a proportion of the coefficient of the unrestricted series. These are presented in Table 5.1 for both the 1981-95 and 1989-95 periods. These results support the inferences derived from the figures. Generally, with the exception of Alberta, variation in macro variables can account for more variation in predicted incidence than variation in policy variables. This is particularly true of childless persons in Quebec and Ontario, and among almost all groups outside Alberta in the 1990s. Nevertheless, policy effects are also important, particularly in Alberta and among couples. 71 5.4.2.1 Decomposition of Predicted Incidence Rates - Policy Variables It is instructive to perform additional decompositions to examine the impact of changes in welfare benefit rates vs. UI entitlement weeks. In the same manner as above, additional simulations were performed in which each of these variables in turn was restricted to equal its sample average. The predicted incidence series from these simulations are presented along with the incidence series from the unrestricted simulations in Figures 5.2a-5.2c. Inspection of Figures 5.2a-5.2c clearly indicates that for almost all province- family type sub-groups the UI Entitlement Weeks Constant series almost exactly coincides with the Total Incidence series. Hence, essentially all of the variation in the predicted incidence series that is removed by restricting the policy variables to equal their sample averages is the result of controlling for welfare benefit effects. This is also reflected in Table 5.1. The only notable exceptions to this are observed during the 1992 to 1994 period, among persons in Ontario, and to a lesser extent among singles and childless couples in Quebec. However, even in this period, restricting UI entitlement weeks does not remove all of the variation in predicted incidence. This result is particularly noteworthy. It is often proposed that there are potentially significant interaction effects between reforms in the UI program and welfare usage. These results do not provide evidence of such effects. 5.4.2.2 Decomposition of Predicted Incidence Rates - Macro Variables Inspection of Figures 5.1a-5.1c indicates that macro effects are important in explaining predicted rates of welfare incidence during the 1990s. To better understand the relative importance of changes in the inflow rate of newly non-employed vs. changes in labour market conditions in these macro effects, additional simulations were performed which each of these variables in turn was restricted to equal its sample average. These are presented in Figures 5.3a-5.3c. Inspection of these figures indicates that unlike the decompositions using the policy variables, neither the inflow rate or labour market conditions can by itself account for the macro effects; for all province family-type 72 groups, both factors have noticeable impacts. That is, both the EP Rate Constant series and the Newly Non-employed Constant series remove some of the variation in predicted incidence during the 1990s, but neither series is flat either. However, the series do suggest that during this period, restricting the employment-population rate to its sample average appears to remove more variation that restricting the inflow rate of newly non- employed to equal its sample average. This observation is bome out by Table 5.1, which indicates that generally, variation in employment-population rates accounts for a greater proportion of variation in predicted incidence that variation in the flow of newly non- employed. Hence, although changes in both the inflow rate and labour market conditions can account for some of the variation observed in predicted incidence, it appears that labour market conditions can account for a greater amount. It is somewhat surprising that changes in the inflow rate of newly non-employed does not have a greater impact than that observed in Figures 5.3a-5.3c. Changes in the size of the pool of newly non-employed for a given cohort has a proportionate impact on welfare take-up and inspection of Figures 5.4a-5.4c, plotting annual inflow rates over the period, reveal these series to exhibit cyclical variation which would be expected to show up in the decomposition. However, we must keep in mind that individuals are able to take-up welfare up to three years following a job separation. If it is the case that take-up from a given cohort is widely spread over this 3-year horizon, then the impact of cyclical variation in the inflow rate will be mitigated. Some insight regarding this issue may be gleaned from examination of Figure 5.5, which presents the 36-month profile of welfare take-up probabilities for the reference individual. We see that although in the month following a job separation the probability of take-up is much higher than in subsequent months, by the 3 r d month it has stabilized and remains at this much lower level until the 12th month. Probability sharply increases in the 13 t h month, thereafter smoothly declining but not returning to level seen during the 3rd-12th month until 2 years after job separation. The result of this pattern is that although the probability of take-up by the end of the first few months is very high, the probability of take-up during the first year is actually slightly less than the probability of take-up during the second year. Probability of take-up during the third year is approximately half the probability of take-up during the second year. Of course, this profile does not consider changes in time-varying covariates that 73 alter the shape of this profile among cohorts, but it does suggest that the existence of a "smoothing effect" is possible. 5.4.3 Comparison of Predicted Incidence Rates and Actual Caseloads Figures 5.6a-5.6b once again present predicted welfare incidence rates for singles without children and all couples (with and without children), but also show the actual number of welfare cases accounted for by non-disabled individuals (Note that caseloads refer to the number of cases as of March 31 in each year). Inspection of these series reveal that trends observed in welfare caseloads share some similarities with trends in predicted incidence rates, but there are also distinct differences. Among all province- family type sub-groups the number of welfare cases increased in the early 1990s. However, the magnitude of this increase was more pronounced among some sub-groups, particularly those in Ontario, where between 1990 and 1992 the number of welfare cases among singles more than doubled and more than tripled among couples. Furthermore, in all provinces except Alberta, caseloads continued to grow until 1994, albeit for most sub- groups at a slower rate. In Alberta, the number of cases sharply declined after 1992; for singles cases dropped 44% relative to 1992 levels (53% for couples). These patterns in caseloads are consistent with patterns in predicted incidence rates in the 1990s among both singles and couples in Ontario and Alberta. To a less extent actual caseloads are consistent with predicted incidence rates among individuals in Quebec, except that predicted incidence begins to sharply decline in 1993, while caseloads do not level off until 1995. A similar occurrence happens in Ontario, although predicted incidence in that province doesn't begin to decline until 1994. In B.C. incidence continued to decline in the 1990s when the welfare caseload while caseloads increased. The most notable result from these comparisons is the dramatic difference in the degree of similarly between the predicted incidence and caseload series in Ontario vs. B.C. in the early 1990s. During this period predicted incidence series in Ontario essentially mirrors the rise in the caseload series for all family types, but in B.C. the two series are opposites. Inspection of Figures A3.1 and A3.3 presenting the welfare benefit rates and employment-population rates provides some insight as to why this occurs. 74 During the 1990s benefit rates are trending up in Ontario, but remain essentially constant in B.C. Additionally, the employment-population rate series declines much more sharply in Ontario (and Quebec) than in B.C. during this period. This worsening of labour market conditions and increases in benefit rates would be expected to increase welfare incidence in Ontario, and this is what is observed. Without similar changes in these factors in B.C., there is no variation for which to drive up predicted incidence. Therefore, the observed increase in caseloads in B.C. in the 1990s must be attributed to other factors. One potential candidate is spillover effects from other provinces. It could be that migration from provinces such as Ontario and Alberta contributed to the increase in welfare caseloads in B.C. This issue was considered sufficiently important for the B.C. government to introduce residency requirements for welfare eligibility in 1995. An additional candidate is increases in welfare incidence by individuals awaiting unemployment insurance. This also was considered important by the B.C. government who imposed a policy to recover benefits paid to UI pending cases in 1994. Other factors could be at work, such as an increased "taste for welfare" among the B.C. population. We do not observe the same degree of consistency between predicted incidence and welfare caseloads during the 1980s. Among most sub-groups, welfare caseloads were relatively stable during the 1984-1990 period, most notably among couples in Alberta and all family types in Ontario. Cases among singles in Alberta were also stable from 1987 to 1990, but steadily increased prior to 1987. In Quebec and B.C., welfare cases generally declined until 1990. Thus, we see that the caseload and incidence series exhibit similar patterns among individuals in Ontario, and the two series for singles in Quebec are broadly consistent. However, the decline in predicted incidence among couples in Quebec far exceeds the decline in the caseload as does predicted incidence among couples in Quebec. The two series are very different among both family types in Alberta. If the non-disabled population may be characterized as being in a steady state equilibrium (i.e. the fraction of the non-disabled population in each state and the proportions flowing between states are constant), then the number of welfare cases at a point in time may be decomposed to equal the product of the rate of welfare incidence and average spell duration (Benjamin, Gunderson and Riddell, 1998). To the extent that this assumption holds, it is possible to derive the average welfare spell duration as the 75 ratio of the number of welfare cases to the rate of incidence. These series are presented in Figures 5.7a-5.7b. Only for Ontario and Quebec do the derived average duration series display any sense of stability, but these series exhibit dramatic growth after 1992. Additionally, the Ontario series exhibit a spike in 1989. The series for B . C . exhibit steady growth throughout the entire 1984-95 period while the Alberta series exhibit an overall upward trend with a spike in 1990. These results do not find evidence of a steady state equilibrium in welfare take-up behaviour during the 1984-95 period. However, they do suggest the possible existence of separate regimes, pre and post-1989, across which behaviour regarding welfare take-up differed. Recall that the estimates used in these simulations are estimated from a sample of individuals experiencing job separations in 1989 and 1990. Accordingly, i f welfare take-up behaviour differed after 1989, then these estimates would reflect behaviour during the latter regime, but not the former. Thus, it would not be surprising to find that the simulation results do not suggest a steady state relationship in the pre-1989 period, assuming one did exist. Average derived spell duration for the entire 1984-95 period and the 1990-95 sub- period are presented below in Chart 5.2. Included in the table is the expected welfare spell duration for singles and couples in Quebec and B . C . (Lacroix, 1998). 9 Consider derived average spell duration over the entire period (1984-95). In all provinces, welfare spells experienced by singles are considerably longer than those experienced by couples (on the order o f 4-9 times longer). Among both singles and couples, the longest spells are experienced by persons in B . C . Spells are also relatively long in Ontario albeit shorter than in B . C . Individuals in Quebec and Alberta experience the shortest spells. Among the two provinces for which it is possible to separately compute average spell duration for couples with children vs. those without children, the results are mixed. In Quebec, spell length among these two sub-groups is approximately equal, but in B . C . average spell length among couples without children is twice that for couples with children. 9 Lacroix computes these using estimates from duration analysis of administrative welfare spell data. 76 Chart 5.2: Derived Average Welfare Duration and Expected Duration Derived Average Duration (years) 1984-1995 (1990-1995) Quebec Ontario Alberta B.C. Singles without children 14.7 (13.2) 19.4 (18.9) 11.2 (15.0) 28.5 (40.4) Couples without children 1.4 (1.3) 4.1 (5.9) Couples with children 1.4 (1.8) 3.2 (4.9) All Couples 1.4 (1.6) 2.3 (2.2) 2.2 (2.9) 3.4 (5.1) Expected Duration (years) Quebec Ontario Alberta B.C. Single Men without children 1.8 0.6 Single Women without children 2.0 0.7 Couples without children 1.6 0.5 Couples with children 1.9 0.6 These results are very much at odds with the expected spell duration figures computed from administrative data. Expected spell duration is approximately 3 times longer in Quebec than in B.C., whereas the derived duration results suggest the opposite to true. Additionally, unlike my results, expected spell duration does not significantly differ among family types. However, the most notable difference between my estimates of average spell duration and those computed using administrative data is that my estimates are much larger. My estimated spell length for persons in B.C. is 50 times larger and among all couples is 6 times larger. Among couples in Quebec, derived spell duration is comparable to expected duration, but among singles derived duration is 8 times larger. In summary, predicted rates of welfare incidence are seen to be very consistent with observed welfare caseloads during the 1990s among both singles and couples in Ontario and Alberta, and to a lesser extent Quebec. However, they are diametrically opposed to trends in welfare caseloads in B.C. Therefore, other causes for the growth of welfare cases in B.C. must be sought. Furthermore, the evidence suggests that a steady state equilibrium in welfare take-up behaviour did not exist during the 1984-95 period. 5.5 Conclusions In this chapter I have applied the estimates from the duration analysis to administrative data on the inflow rate of persons into the pool of non-employed to simulate incidence rates of welfare take-up over the 1984-95 period. I find that over this period the predicted incidence rates series exhibit very similar patterns among individuals in Ontario and childless persons in Quebec. Among these sub-groups predicted welfare incidence was relatively stable until 1989 (except for childless couples in Quebec), then increased dramatically until 1993 (1992 for Quebec). Thereafter, incidence declined but did not return to 1989 levels. This pattern was observed among individuals of all province-family type sub-groups. Predicted rates of welfare incidence among individuals in Alberta and B . C . , as well as couples with children in Quebec were substantially different from other sub- groups. Among these individuals, predicted rates of incidence exhibited a general decline during the 1984 to 1995 period, but was greatest during the 1980s. Between 1990 and 1992 incidence rates were relatively stable among persons in B . C . , but increased among couples with children in Quebec and persons in Alberta. Decompositions of these predicted incidence series indicate that among all sub- groups, variation in macroeconomic variables is important in explaining predicted incidence during the 1990s. This is particularly true among individuals in Ontario and Quebec and to a lesser extent, Alberta and B . C . Changes in policy variables can also account for substantial variation in predicted incidence during this period in Ontario and in the pre-1990 period for singles and childless couples in Quebec. However, the strongest policy effects are observed for Alberta, where they can account for essentially all variation in predicted incidence throughout the entire 1984-95 period, although to a less degree for couples with children than other family types. Additional decompositions indicate that essentially all o f these policy effects are driven by changes in welfare benefit rates; changes in U I entitlement weeks had very little impact on predicted incidence. This result is particularly noteworthy. It is often proposed that there are potentially significant interaction effects between reforms in the 78 UI program and welfare usage, but these results do not provide evidence of such effects. However, this may indicate insufficient variation in the U I entitlement weeks measure during the 1988-90 period. Similar decomposition of the macro effects indicate that changes in both the inflow of newly non-employed and labour market conditions had important impacts on predicted incidence, but labour market conditions was the dominant factor. It should be emphasized that these decompositions only provide guidance on which factors are driving the predicted incidence rates; they do not involve use of data on actual incidence rates. The use of such data would be most valuable in this exercise, but is not available. Trends in predicted welfare incidence rates are consistent with observed trends in welfare caseloads in Ontario throughout the entire 1984-95 period. Similar results hold among individuals in Alberta during the 1990s and to a lesser extent Quebec. However, they are diametrically opposed to trends in welfare caseloads in B . C . Therefore, other causes for the growth of welfare cases in B . C . observed during the 1990s must be sought. Furthermore, we do not find evidence o f stable derived welfare spell durations, casting doubt on the existence of a steady state equilibrium in welfare take-up behaviour. A s previously stated we have no data on actual incidence rates, so direct evaluation of the success of this simulation exercise cannot be done. However, for Quebec and B . C . the derived average spell durations can be compared with estimates o f expected spell duration estimated from administrative data. Results indicate that the derived average durations greatly exceed expected durations for singles without children to the point of not being credible, but are relatively similar among couples, particularly among couples in Quebec. These discrepancies are not easily reconciled. The L M A S data on which the probability of welfare take-up is based is known to suffer from under-reporting which would manifest itself as an artificially lower probability of welfare take-up. However, the level of under-reporting is not by itself sufficient to explain these results. It is also unlikely that the degree in under-reporting would vary sufficiently across provinces and family types to account for these differences. This suggests that the parameter estimates used in the construction of the probability of welfare are not sufficiently accurate to simulate welfare take-up behaviour during the 1984-95 period. This could reflect a lack 79 of variation in covariates during the 1989-90 period used for the duration analysis, or an insufficient number of individuals in the sample observed to take-up welfare. The resolution of these issues w i l l be the subject of further work. We should also consider the possible existence of separate regimes during the sample period, say, pre and post-1989, across which behaviour regarding welfare take-up differed. Recall that the estimates used in these simulations are estimated from a sample of individuals experiencing job separations in 1989 and 1990. Accordingly, i f welfare take-up behaviour differed after 1989, then these estimates would reflect behaviour during the latter regime, but not the former. Thus, it would not be surprising to find that the simulation results do not suggest a steady state relationship, assuming one did exist. 80 Chapter 6 Conclusions The research undertaken in this thesis has examined social assistance (welfare) receipt in Canada during the 1981-1995 period. During these years, particularly in the early 1990s, welfare use in Canada grew dramatically. The increase in program expenditures associated with this growth, as well as changes in funding arrangements between the federal and provincial governments, led to extensive reforms of provincial welfare systems. However, these reforms were implemented without a thorough understanding of the forces responsible for this growth. Several candidate explanations have been proposed in discussions of this rise in usage, including changes in welfare programs, changes in labour market conditions and reforms to the unemployment insurance system. The work done in this thesis examines how these factors have influenced the use of welfare during the 1981-95 period. Results suggest that an examination of welfare usage in Canada must consider differences amongst individuals. The need for social assistance will necessarily depend on specific personal circumstances; the financial resources and needs of the household. These differences are reflected in rules governing eligibility for welfare and levels of benefit rates that depend on the employability and family composition of households. One would suspect that an analysis that failed to account for such differences would not yield results that are representative of the population as a whole. In chapter 3,1 conduct such an analysis of welfare usage over the sample period and judge it to be wanting. This can be partially attributed to the aggregation of different classes of welfare recipients and the use of broad measures of labour market conditions and welfare benefit rates. In the second part of the chapter, these concerns are addressed in a subsequent analysis that separately examines welfare usage among (a) male singles without children, (b) female singles without children and (c) single mothers. Results reveal important differences in behaviour between these types of households. Labour market conditions are shown to have a significant impact on welfare usage among all family types, but there is only qualified support regarding the importance of the availability of UI benefits and the relative generosity of welfare benefits. Changes in the generosity of welfare benefits are 81 found to significantly affect welfare usage among single mothers and there is some evidence that changes in benefit generosity affects usage among single women without children. But, this is not the case for single men. Changes in U I availability have a significant impact on welfare use among single men, particularly when the sample is restricted to those individuals with low levels of education. Similar results are found for single women without children, but are no longer statistically significant when only individuals with low levels of education are considered. The importance of labour market conditions and interactions with the UI system motivate the analysis performed in chapter 4. In order to better understand the mechanisms underlying these trends, I employ duration analysis to examine the path leading individuals from employment to welfare receipt. This path is characterized by the job loss experienced by an individual and their subsequent transition to re-employment or welfare receipt. Although sample sizes preclude separate estimations by family type, behavioural differences between these groups are considered through the inclusion of dummy variables. The profile of the baseline hazard function from the estimation of the welfare take-up process indicates that many of those who do take-up welfare following a job loss do not do so immediately. Thus, take-up is delayed until other financial resources, including UI benefits, are exhausted. This suggests the importance of interactions with the UI system when considering welfare usage. However, results from the duration analysis find no evidence that changes in the number of weeks of U I benefit entitlement significantly affects time until welfare take-up. O f particular interest is that the level of welfare benefits does not significantly affect job duration, however, benefits do positively affect time until re-employment and negatively affect time until welfare take-up. Thus, the depth of the welfare safety net doesn't affect job duration, but once non-employed, it does affect time until re- employment and welfare take-up. Labour market conditions are also found to be important. Results indicate that in favourable economic times individuals experience shorter times until re-employment. Although changes in labour market conditions are not found to significantly affect time 82 until welfare take-up, this is not unexpected given the re-employment and welfare take-up processes are estimated independently. Use of such an estimation strategy renders time until welfare take-up to be a question of the availability of financial resources other than wage income, and as such, labour market conditions are not relevant. The duration analysis provides further evidence of the need to consider personal characteristics when examining welfare take-up. Results suggest that time until re- employment and time until welfare take-up are significantly different between family types. In chapter 5,1 apply the estimates from the duration analysis to administrative data on inflows of persons into the pool of non-employed to simulate rates of welfare incidence over the 1984-95 period. This allows for the predicted incidence series to be decomposed in order to determine the degree to which predicted incidence is driven by changes in labour market conditions and policy variables. In particular, it allows for an examination of which factors can account for the dramatic growth in welfare usage experienced during the 1990s. Therefore, this represents a complementary approach to the strategy pursued in chapter 3. Decompositions of these predicted incidence series indicate that among all sub- groups, variation in macroeconomic variables is important in explaining predicted incidence during the 1990s, particularly so in Ontario and Quebec. Changes in policy variables can also account for substantial variation in predicted incidence during this period in Ontario and in the pre-1990 period for singles and childless couples in Quebec. However, the strongest policy effects are observed for Alberta, where they can account for essentially all variation in predicted incidence throughout the entire 1984-95 period, although to a less degree for couples with children than other family types. Additional decompositions indicate that essentially all of these policy effects are driven by changes in welfare benefit rates; changes in UI entitlement weeks have little impact on predicted incidence. This result is particularly noteworthy. It is often proposed that there are potentially significant interaction effects between reforms in the UI program and welfare usage, but these results provide little evidence of such effects. Similar decompositions of the macro effects indicate that changes in both the inflow of newly 83 non-employed and labour market conditions had important impacts on predicted incidence, but labour market conditions was the dominant factor. The focus of this thesis has been to explain the dramatic growth of welfare receipt observed during the early 1990s. Considering the various pieces of evidence together presents a relatively consistent picture and yields important insights. Of course, we must keep in mind that the samples examined in the two analyses are not identical; only singles without children are common to both samples. This is particularly important in light of the different results found among different types of individuals in both the aggregate and duration analyses. It should be emphasized that these decompositions from the simulations only provide guidance on which factors are driving the predicted incidence rates; they do not involve use of data on actual incidence rates. Thus, the degree of faith we put into the simulation results should reflect the "reasonableness" of the results from the duration analysis and the degree to which the simulations are successful in producing "reasonable" predicted incidence series. Estimates from the duration analysis reflect behaviour during the period immediately prior to the upsurge in caseloads, that is, the period of calm before the storm. Furthermore, most of the estimated coefficients have the anticipated signs. As such, they appear to be reasonable. Trends in predicted welfare incidence rates are consistent with observed trends in welfare caseloads in Ontario throughout the entire 1984-95 period, and to a lesser extent, Quebec. This is also true for the 1990s in Alberta. Accordingly the effects of which I am the most confident are those observed in these provinces. The most consistent result coming from both the aggregate and duration is that labour market conditions contributed significantly to the growth of welfare use in Canada. The aggregate analysis finds strong labour market effects among both male and female singles without children as well as lone mothers. The simulation results indicate both the inflow rate of newly non-employed and particularly labour market conditions can account for substantial amount of the variation in predicted welfare incidence observed during the 1990s among singles and couples without children as well as couples with children in both Ontario and Quebec. Thus, the economic downturn of the early 1990s definitely 84 played a significant role in the growth of welfare use during this period. The evidence concerning the importance of interactions with the unemployment insurance system is mixed. The aggregate analysis suggests that UI effects are important determinants of welfare use among single men without children but only qualified support is found for single women without children. No evidence of such effects is found for lone mothers. In contrast, the simulation results find little evidence of UI effects, and this is limited to individuals in Ontario and Quebec during the early 1990s. The absence of a strong UI effect in the simulated incidence series may indicate insufficient variation in the variable used to proxy UI entitlement during the 1988-90 period. Inclusion of additional spells from the Survey of Labour Income Dynamics (SLID) data will provide such variation. This will be pursued in future work. The aggregate analysis finds important benefit effects for lone mothers, but only limited benefit effects for single women without children. There is no evidence of benefit effects for single men without children. Results from the simulations indicate that changes in benefits could account for some of the variation in predicted incidence in the 1990s in Ontario and Alberta as well as in the 1980s in Quebec and Alberta. These discrepancies are not easily reconciled but may reflect differences in the composition of the two data samples. The resolution of these issues will be the subject of further work. In conclusion, the research undertaken in this thesis has provided substantial progress in our understanding of the forces underlying welfare usage in Canada over the 1981-95 period. This research agenda will continue to be pursued. 85 c o — r/5 TS 3 vi • S3 O a S3 mO a E x u a> v s c <u • a c o ft- « • O h * - * * ; • * T t m vo vc OM «A &o &o &<i &o| .2 o ro p» o CN o M r- oo oo io, B w w V) w w w Vi + o Vi cu o r -T f T t T t —I T j " m vo vo e S W 6fl W « M M ro so as ro T t in u-i »o vo vo €̂5" o o o o E as O o o o td IT) O 0 0 0 0 H i - l o O O © O O <-H I I I I I I 0 \ O H N C l t f ) x oi ov ov os ov s- a o E a fa. s? os m CN + o o s? os o o o o T t S3 a . u c c s? ©s th c s? ©s o t/5 »n C S CN + s? ©s + o O CN O in •n + o c <u S3 CL. ob os «n 60 X CN + o os CN + ^ >r. rt Vi + C C N? O S s? O S O «n CN CN + + «n »n r~ r-Vi Vi I S CN + o o s© O O S > O >* u-l O + T t O Vi 2 + Vi "*s O S o CN + l~ Vi CN + <N + O cn O ^ 4i OJO s? C « 9 S (H Jji «n o * CN „ O + 2? S ° ® 3 in V ^ Vi + v> r - + V5 T 3 T 3 s? O S o CN + o VO -o OS O S in CN + O VO o •a o o m Vi + o o m C N Vi o o in Vi + o o o c/3 o o o o t- O <N O < S ^ h H i ir, ov cu CJ vP o E CN © 3 O O O O wo r- o cs CN co Vi Vi Vi O O O O O O -a a Ko M W CS sr wo cs + o WO Vi « wi in f~ 00 00 6? V) W 3 -a a •a c IB E o o s? wo cs + o wo <̂ wo wo £ f- oo " E o o o o O I/O cs" CN" 6<J &<} o o wo cs" OX B o o o o O wo o o ^3 1 « is fi >n •<* Tf O O O cs in ON — rs t~- 00 00 s 3 u J3 so *sZ M .I4 E es OX) c CJ TD ° "S O « O •*? CS + o o CJ M CO 00 e C3 <D ° "3 W0 ™ &o o wo cs + o >/0 s cs la- in JS C3 D O O WO WO T3 — O ° M VO °, 5 s 1 « £ o w O 9i ») i-H O O VO VO CS 00 C\ 5 Table 3.1: Age Distributions Non-Recipients SA Recipients Age Group Age Group Year 18-24 25-34 35-44 45-54 55-64 18-24 25-34 35-44 45-54 55-64 Singles Without Children - Male (SNM) 1981 0.52 0.27 0.09 0.06 0.05 0.31 0.22 0.14 0.15 0.18 1982 0.52 0.28 0.09 0.06 0.05 0.33 0.30 0.14 0.11 0.11 1984 0.50 0.29 0.09 0.06 0.05 0.32 0.28 0.18 0.13 0.09 1985 0.49 0.30 0.10 0.06 0.05 0.32 0.32 0.14 0.10 0.12 1986 0.46 0.32 0.10 0.07 0.06 0.27 0.38 0.21 0.05 0.09 1987 0.42 0.35 0.12 0.06 0.06 0.26 0.30 0.17 0.14 0.12 1988 0.41 0.35 0.12 0.06 0.06 0.16 0.32 0.19 0.16 0.18 1989 0.40 0.34 0.13 0.07 0.06 0.19 0.29 0.25 0.13 0.13 1990 0.38 0.35 0.15 0.06 0.06 0.26 0.27 0.15 0.15 0.16 1991 0.37 0.34 0.16 0.07 0.05 0.25 0.34 0.18 0.10 0.13 1992 0.37 0.34 0.16 0.08 0.06 0.20 0.30 0.25 0.11 0.14 1993 0.37 0.32 0.17 0.09 0.05 0.24 0.35 0.19 0.13 0.09 1994 0.36 0.33 0.17 0.10 0.05 0.22 0.33 0.24 0.12 0.09 1995 0.36 0.34 0.17 0.09 0.05 0.20 0.30 0.23 0.18 0.08 Singles Without Children - Female (SNF) 1981 0.46 0.21 0.08 0.10 0.15 0.17 0.14 0.08 0.17 0.44 1982 0.48 0.21 0.08 0.10 0.14 0.28 0.10 0.12 0.19 0.31 1984 0.43 0.22 0.09 0.10 0.15 0.21 0.15 0.11 0.21 0.32 1985 0.43 0.23 0.09 0.09 0.15 0.19 0.17 0.10 0.20 0.34 1986 0.40 0.23 0.11 0.11 0.15 0.22 0.13 0.14 0.21 0.30 1987 0.39 0.27 0.11 0.10 0.14 0.15 0.18 0.13 0.22 0.33 1988 0.38 0.26 0.11 0.11 0.14 0.15 0.15 0.22 0.19 0.29 1989 0.37 0.26 0.12 0.11 0.14 0.15 0.16 0.17 0.21 0.31 1990 0.37 0.26 0.13 0.11 0.13 0.15 0.18 0.12 0.20 0.35 1991 0.36 0.24 0.13 0.12 0.15 0.16 0.19 0.14 0.22 0.28 1992 0.38 0.24 0.12 0.12 0.14 0.18 0.21 0.18 0.20 0.24 1993 0.36 0.24 0.14 0.13 0.13 0.16 0.18 0.19 0.22 0.25 1994 0.38 0.24 0.13 0.13 0.12 0.20 0.20 0.15 0.20 0.26 1995 0.37 0.24 0.13 0.15 0.12 0.17 0.14 0.21 0.21 0.28 Single With Children - Female (SPF) 1981 0.09 0.33 0.39 0.16 0.03 0.25 0.38 0.22 0.11 0.03 1982 0.06 0.36 0.40 0.13 0.04 0.20 0.43 0.24 0.10 0.03 1984 0.08 0.35 0.42 0.12 0.03 0.20 0.45 0.24 0.10 0.01 1985 0.07 0.35 0.45 0.11 0.01 0.23 0.42 0.26 0.06 0.02 1986 0.08 0.38 0.38 0.14 0.02 0.19 0.45 0.26 0.09 0.01 1987 0.08 0.31 0.45 0.14 0.02 0.17 0.54 0.23 0.04 0.01 1988 0.05 0.33 0.46 0.14 0.01 0.25 0.45 0.23 0.06 0.02 1989 0.08 0.33 0.43 0.15 0.01 0.14 0.50 0.27 0.09 0.01 1990 0.08 0.34 0.45 0.12 0.01 0.18 0.48 0.28 0.06 0.01 1991 0.05 0.35 0.45 0.14 0.00 0.18 0.49 0.27 0.05 0.01 1992 0.04 0.30 0.48 0.16 0.02 0.17 0.52 0.23 0.07 0.01 1993 0.07 0.32 0.47 0.13 0.02 0.15 0.47 0.30 0.07 0.01 1994 0.06 0.31 0.45 0.18 0.01 0.16 0.44 0.31 0.09 0.01 1995 0.07 0.28 0.47 0.18 0.00 0.18 0.39 0.36 0.06 0.01 88 Table 3.2: Education Distributions Non-Recipients SA Recipients Education Group Education Group Year Elcm. Some Some Post Cert. Deg. Elem. Some Some Post Cert. Deg. High Sch. Second. High Sch. Second. Singles Without Children - Male (SNM) 1981 0.10 0.50 0.17 0.12 0.11 0.38 0.41 0.07 0.10 0.03 1982 0.08 0.49 0.18 0.12 0.12 0.31 0.53 0.07 0.06 0.03 1984 0.08 0.50 0.17 0.14 0.11 0.29 0.52 0.08 0.08 0.03 1985 0.08 0.47 0.20 0.14 0.12 0.31 0.49 0.11 0.06 0.04 1986 0.07 0.48 0.18 0.15 0.12 0.26 0.52 0.08 0.08 0.05 1987 0.07 0.44 0.19 0.15 0.14 0.30 0.48 0.09 0.10 0.03 1988 0.07 0.45 0.19 0.16 0.14 0.29 0.45 0.10 0.09 0.07 1989 0.06 0.40 0.17 0.25 0.13 0.30 0.44 0.09 0.13 0.04 1990 0.05 0.41 0.16 0.24 0.14 0.32 0.44 0.05 0.13 0.06 1991 0.05 0.41 0.15 0.25 0.15 0.19 0.51 0.06 0.17 0.06 1992 0.05 0.38 0.16 0.26 0.15 0.17 0.46 0.09 0.22 0.06 1993 0.04 0.39 0.16 0.26 0.15 0.18 0.47 0.12 0.17 0.07 1994 0.05 0.36 0.18 0.27 0.15 0.16 0.44 0.12 0.21 0.06 1995 0.04 0.37 0.16 0.27 0.16 0.18 0.48 0.10 0.16 0.08 Singles Without Children - Female (SNF) 1981 0.10 0.45 0.16 0.18 0.12 0.49 0.39 0.05 0.06 0.02 1982 0.09 0.44 0.18 0.17 0.12 0.44 0.42 0.06 0.05 0.03 1984 0.09 0.43 0.18 0.17 0.14 0.41 0.46 0.06 0.06 0.01 1985 0.09 0.39 0.21 0.17 0.13 0.38 0.48 0.05 0.06 0.04 1986 0.07 0.39 0.19 0.20 0.15 0.37 0.52 0.06 0.03 0.01 1987 0.07 0.37 0.20 0.19 0.16 0.36 0.45 0.10 0.06 0.04 1988 0.08 0.39 0.18 0.19 0.16 0.40 0.40 0.09 0.10 0.02 1989 0.06 0.34 0.19 0.27 0.14 0.41 0.41 0.05 0.10 0.03 1990 0.06 0.33 0.18 0.28 0.15 0.38 0.37 0.07 0.14 0.03 1991 0.05 0.32 0.19 0.28 0.15 0.30 0.46 0.10 0.12 0.03 1992 0.05 0.33 0.18 0.26 0.17 0.18 0.48 0.11 0.17 0.06 1993 0.06 0.29 0.18 0.29 0.18 0.24 0.42 0.09 0.19 0.06 1994 0.05 0.29 0.19 0.30 0.17 0.22 0.43 0.11 0.18 0.06 1995 0.05 0.28 0.18 0.31 0.18 0.24 0.41 0.09 0.20 0.06 Singles With Children - Female (SPF) 1981 0.12 0.57 0.09 0.15 0.07 0.32 0.60 0.06 0.02 0.00 1982 0.10 0.61 0.11 0.10 0.08 0.28 0.61 0.05 0.04 0.01 1984 0.12 0.52 0.14 0.14 0.09 0.23 0.66 0.05 0.05 0.01 1985 0.09 0.53 0.12 0.15 0.12 0.23 0.61 0.08 0.06 0.02 1986 0.06 0.54 0.14 0.11 0.14 0.19 0.65 0.08 0.06 0.02 1987 0.08 0.53 0.10 0.17 0.12 0.15 0.70 0.09 0.06 0.00 1988 0.06 0.50 0.12 0.19 0.13 0.18 0.62 0.10 0.08 0.02 1989 0.07 0.43 0.11 0.26 0.13 0.17 0.61 0.11 0.11 0.01 1990 0.05 0.45 0.12 0.28 0.11 0.13 0.64 0.07 0.14 0.03 1991 0.06 0.43 0.10 0.28 0.13 0.10 0.62 0.14 0.11 0.02 1992 0.05 0.40 0.11 0.33 0.12 0.11 0.58 0.11 0.18 0.02 1993 0.03 0.39 0.13 0.31 0.13 0.08 0.58 0.12 0.20 0.02 1994 0.04 0.37 0.13 0.32 0.13 0.11 0.51 0.16 0.20 0.02 1995 0.04 0.38 0.10 0.34 0.15 0.10 0.50 0.15 0.21 0.04 89 Table 3.3: Weeks Worked Distributions Non-Recipients SA Recipients Weeks Worked Weeks Worked Year 0 1-10 11-20 21-30 31-40 41-52 0 1-10 11-20 21-30 31-40 41-52 Singles Without Children - Male (SNM) 1981 0.06 0.05 0.10 0.10 0.08 0.60 0.53 0.09 0.09 0.10 0.05 0.13 1982 0.08 0.08 0.11 0.09 0.08 0.56 0.55 0.14 0.12 0.04 0.03 0.12 1984 0.07 0.07 0.12 0.09 0.08 0.56 0.57 0.13 0.11 0.06 0.03 0.09 1985 0.06 0.08 0.11 0.10 0.08 0.58 0.54 0.11 0.14 0.09 0.04 0.09 1986 0.06 0.05 0.11 0.09 0.08 0.61 0.41 0.13 0.13 0.09 0.07 0.16 1987 0.06 0.06 0.10 0.07 0.07 0.64 0.50 0.15 0.11 0.10 0.04 0.11 1988 0.05 0.05 0.10 0.07 0.07 0.65 0.56 0.12 0.07 0.06 0.05 0.15 1989 0.05 0.05 0.09 0.08 0.08 0.66 0.60 0.10 0.09 0.07 0.05 0.09 1990 0.05 0.05 0.11 0.08 0.08 0.63 0.50 0.08 0.13 0.08 0.07 0.14 1991 0.07 0.06 0.11 0.08 0.08 0.61 0.52 0.11 0.11 0.09 0.04 0.14 1992 0.07 0.06 0.10 0.08 0.06 0.62 0.45 0.11 0.12 0.11 0.05 0.16 1993 0.09 0.06 0.11 0.07 0.07 0.60 0.55 0.14 0.10 0.05 0.05 0.11 1994 0.08 0.05 0.10 0.08 0.07 0.63 0.56 0.08 0.10 0.12 0.05 0.10 1995 0.09 0.04 0.09 0.08 0.08 0.62 0.53 0.15 0.08 0.07 0.04 0.13 Singles Without Children - Female (SNF) 1981 0.12 0.05 0.08 0.06 0.06 0.62 0.75 0.04 0.02 0.01 0.05 0.12 1982 0.13 0.07 0.09 0.06 0.06 0.60 0.73 0.08 0.05 0.02 0.04 0.09 1984 0.13 0.06 0.08 0.07 0.07 0.59 0.71 0.05 0.06 0.04 0.02 0.10 1985 0.13 0.06 0.09 0.06 0.05 0.61 0.71 0.07 0.07 0.04 0.02 0.09 1986 0.12 0.04 0.08 0.06 0.06 0.64 0.73 0.03 0.06 0.06 0.03 0.09 1987 0.11 0.05 0.09 0.07 0.06 0.62 0.74 0.05 0.07 0.04 0.03 0.07 1988 0.11 0.05 0.08 0.05 0.05 0.65 0.74 0.03 0.07 0.03 0.02 0.11 1989 0.10 0.05 0.08 0.05 0.05 0.67 0.71 0.05 0.03 0.03 0.05 0.13 1990 0.10 0.04 0.09 0.07 0.07 0.63 0.64 0.08 0.08 0.04 0.02 0.14 1991 0.12 0.05 0.08 0.06 0.05 0.65 0.66 0.08 0.07 0.03 0.05 0.11 1992 0.13 0.05 0.07 0.07 0.06 0.62 0.59 0.11 0.08 0.05 0.04 0.13 1993 0.13 0.07 0.08 0.05 0.04 0.64 0.67 0.05 0.04 0.07 0.05 0.13 1994 0.13 0.05 0.09 0.06 0.05 0.63 0.65 0.07 0.09 0.05 0.03 0.11 1995 0.11 0.05 0.07 0.06 0.05 0.66 0.67 0.06 0.06 0.04 0.05 0.13 Singles With Children - Female (SPF) 1981 0.16 0.03 0.05 0.06 0.04 0.66 0.73 0.09 0.06 0.02 0.01 0.08 1982 0.18 0.03 0.06 0.05 0.04 0.65 0.72 0.06 0.04 0.05 0.03 0.10 1984 0.14 0.05 0.05 0.05 0.05 0.66 0.72 0.08 0.07 0.06 0.02 0.05 1985 0.13 0.04 0.06 0.06 0.04 0.66 0.72 0.08 0.06 0.03 0.04 0.07 1986 0.12 0.04 0.05 0.06 0.03 0.69 0.64 0.11 0.08 0.05 0.04 0.08 1987 0.12 0.03 0.06 0.05 0.06 0.68 0.67 0.08 0.05 0.05 0.04 0.11 1988 0.12 0.04 0.06 0.04 0.08 0.66 0.65 0.08 0.06 0.07 0.04 0.10 1989 0.12 0.03 0.05 0.06 0.07 0.68 0.69 0.07 0.06 0.06 0.03 0.10 1990 0.12 0.04 0.05 0.05 0.08 0.67 0.60 0.08 0.07 0.05 0.04 0.16 1991 0.14 0.03 0.05 0.07 0.04 0.67 0.63 0.07 0.06 0.10 0.03 0.10 1992 0.09 0.02 0.06 0.06 0.04 0.72 0.69 0.06 0.05 0.03 0.03 0.14 1993 0.12 0.02 0.05 0.04 0.04 0.73 0.64 0.08 0.07 0.06 0.03 0.11 1994 0.12 0.02 0.04 0.06 0.05 0.70 0.68 0.07 0.07 0.05 0.02 0.11 1995 0.12 0.02 0.06 0.05 0.04 0.71 0.63 0.08 0.08 0.06 0.03 0.12 90 Table 3.4: SA Usage - Aggregate Approach Independent Variables Regression (1) OLS (2) AR1 Coeff. s.e. Coeff. ! i.e. SA Benefits/Min Wage Earnings 0.1249 ** 0.032 0.0709 0 019 Unemployment Rate 0.4176 ** 0.122 0.1870 0 058 Min. UI Qualification Weeks 0.1237 0.157 -0.0636 0 068 Newfoundland -0.0977 1.937 2.2526 1 827 Nova Scotia 1.5196 1.184 1.4952 1 331 New Brunswick 0.8369 1.261 1.6331 1 466 Quebec 4.1739 ** 0.822 4.1577 ** 0 667 Manitoba -0.5788 0.800 -0.9923 1 141 Saskatchewan -2.3897 ** 0.756 -2.0615 * 1 168 Alberta -7.5806 ** 0.959 -6.2734 ** 0 903 B.C. 0.2912 0.505 0.3986 0 713 1983 -0.3569 0.850 0.7251 ** 0 347 1984 0.3805 0.862 1.0633 ** 0 414 1985 0.5613 0.850 1.1617 ** 0 450 1986 0.1668 0.880 0.7706 0 494 1987 0.1312 0.850 0.7253 0 500 1988 -0.1998 0.937 0.2804 0 545 1989 -0.3249 0.895 0.2128 0 541 1990 -0.4598 0.918 0.2215 0 564 1991 -0.9028 0.963 1.0782 * 0 609 1992 0.2230 1.151 2.8413 ** 0 681 1993 1.3953 1.054 3.7196 ** 0 651 1994 2.6063 ** 0.944 4.4107 ** 0 600 1995 3.0218 ** 0.897 4.4821 ** 0 571 1996 2.1711 ** 1.038 3.8413 ** 0 604 Constant -4.1322 2.765 3.0360 ** 1 510 Number of Observations 135 135 Adj. R2=0.81 AR coeff=0. 77 Legend: ** significant at 5% level * significant at 10% level 91 Table 3.5a: SA Usage - Singles Without Children - Male (SNM) Regression (1) (2) (3) All SNM's SNM's Low Educ Wkswkd for Wkswkd for Wkswkd for Independent Variables all SNM's SNM's < 35 yrs. SNM's < 35 yrs. Coeff. s.e. Coeff. s.e. Coeff. s.e. Relative Benefits 0.0301 0.028 0.0425 0.028 0.0318 0.038 Weeks Worked -0.7811 ** 0.131 -0.7545 ** 0.118 -0.8707 ** 0.122 Min. UI Qual. Weeks 0.2256 0.152 0.3712 ** 0.155 0.6374 ** 0.208 New Brunswick -2.7904 * 1.424 -1.7128 1.288 -2.7976 1.779 Quebec^ 4.3946 ** 0.750 5.7853 ** 0.578 9.6810 ** 0.816 Alberta 0.0513 0.430 0.4776 0.428 0.7890 0.595 B.C. 2.6065 ** 0.452 3.3409 ** 0.414 3.7053 ** 0.581 1982 -0.6910 0.751 -0.8550 0.735 -0.7246 1.038 1984 0.2311 0.814 0.7674 0.764 1.3233 1.076 1985 0.4971 0.767 0.6936 0.737 1.4829 1.039 1986 0.8579 0.754 0.8511 0.732 2.4702 ** 1.025 1987 0.9628 0.731 1.0862 0.712 2.3856 ** 0.987 1988 1.3955 * 0.757 1.7610 ** 0.746 2.8279 ** 1.029 1989 0.5126 0.752 0.7552 0.739 1.4855 0.981 1990 0.4842 0.806 0.0572 0.783 0.2696 1.111 1991 1.0948 0.973 0.2383 0.991 0.3053 1.393 1992 2.2175 ** 0.893 1.8608 ** 0.881 1.9485 1.251 1993 1.9858 ** 0.922 0.9348 0.974 0.7630 1.379 1994 3.0539 ** 0.817 2.4822 ** 0.820 2.7906 ** 1.187 1995 1.2384 0.937 0.4348 0.948 0.6122 1.351 Constant 28.8611 * 5.033 24.5330 ** 4.172 27.3482 ** 5.178 Number of Observations 70 70 70 Adjusted R 2 0.93 0.93 0.94 Legend: ** significant at 5% level * significant at 10% level 92 Table 3.5b: SA Usage - Singles Without Children - Female (SNF) Regression (1) (2) (4) (3) All SNF's SNF's - Low Educ. Wkswkd for Wkswkd for Wkswkd for Wkswkd for Independent Variables all SNF's SNF's < 35 yrs. all SNF's SNF's < 35 yrs. Coeff. s.e. Coeff. s.e. Coeff. s.e. Coeff. s.e. Relative Benefits 0.0620 * 0.036 0.0764 * 0.041 0.0544 0.056 0.0368 0.066 Weeks Worked -1.1604 ** 0.203 -0.8252 ** 0.213 -1.1674 ** 0.204 -0.5580 ** 0.189 Min. UI Qual. Weeks 0.4495 ** 0.185 0.2590 0.202 0.4205 0.287 0.2944 0.349 New Brunswick 0.4910 1.543 0.2942 1.826 1.8402 2.360 1.6161 2.923 Quebec 2.7325 ** 1.284 6.3820 ** 0.989 6.6702 ** 1.751 13.9849 ** 1.118 Alberta 0.7558 0.528 0.6476 0.597 1.5590 * 0.893 0.3327 1.012 B.C. 1.6746 ** 0.509 2.4802 ** 0.556 2.6194 ** 0.788 3.3696 ** 0.958 1982 -1.0780 0.827 -1.2305 0.974 -0.9159 1.274 -0.4592 1.518 1984 0.4177 0.928 0.0400 1.093 0.9343 1.442 1.4415 1.713 1985 0.6587 0.910 0.5912 1.038 2.2860 1.426 3.3467 * 1.694 1986 1.1388 0.919 0.4042 1.022 2.1414 1.424 2.7743 1.700 1987 0.7677 0.857 0.6648 0.969 1.4560 1.366 2.7802 ** 1.608 1988 1.4941 * 0.880 1.4411 0.998 2.3238 * 1.380 3.2279 * 1.696 1989 1.0589 0.935 0.2837 1.032 0.6025 1.339 1.2101 1.605 1990 0.5206 0.922 0.0695 1.052 1.2118 1.553 2.5274 1.834 1991 0.3450 1.067 0.6091 1.209 0.8620 1.913 3.7764 * 2.158 1992 -0.8676 1.026 -1.2020 1.238 -1.4387 1.707 0.3054 2.015 1993 0.0323 0.986 -0.4465 1.223 1.4628 1.644 3.5823 * 1.893 1994 0.9016 0.965 0.6091 1.185 1.7534 1.704 3.3537 2.133 1995 0.5271 1.043 0.0216 1.220 2.6112 1.702 2.6136 2.162 Constant 38.7917 ** 7.107 28.8981 ** 7.775 39.7954 ** 7.459 21.8160 ** 7.658 Number of Observations 70 70 70 70 Adjusted R 2 0.92 0.90 0.94 0.91 Legend: ** significant at 5% level * significant at 10% level 93 Table 3.5c: SA Usage - Singles With Children - Female (SPF) Regression (1) (2) (3) All SPF's SPF's with SPF's with only only Independent Variables young child older child Coeff. s.e. Coeff. s.e. Coeff. s.e. Relative Benefits 0.1838 ** 0.070 0.2300 ** 0.114 0.2003 ** 0.059 Weeks Worked -1.6606 ** 0.204 -1.9611 ** 0.276 -1.4895 ** 0.144 Min. UI Qual. Weeks 0.1555 0.477 0.1899 0.842 0.4048 0.415 New Brunswick 4.7124 3.363 4.6747 5.118 6.2474 ** 2.903 Quebec 1.3639 2.524 4.8210 3.797 2.3409 2.217 Alberta 0.1526 1.537 2.0965 2.636 -1.7696 1.320 B.C. 0.8973 1.446 6.0174 ** 2.499 -0.7150 1.250 1982 2.5244 2.225 5.4079 3.924 2.5332 1.941 1984 3.0849 2.531 4.0277 4.488 2.4604 2.196 1985 2.9010 2.509 -1.9063 4.511 4.0793 * 2.182 1986 3.6451 2.610 0.9743 4.545 3.5562 2.251 1987 2.9125 2.647 5.4018 4.541 0.9006 2.316 1988 1.8918 2.639 5.8189 4.564 -1.0121 2.327 1989 -0.1443 2.671 -0.6091 4.558 -1.0336 2.295 1990 4.6950 2.865 3.7339 4.930 3.6811 2.566 1991 3.7769 3.040 3.5374 5.381 3.1773 2.719 1992 8.7144 ** 2.840 10.1093 ** 5.041 4.5282 * 2.603 1993 7.6814 ** 2.682 7.2340 4.713 4.5778 * 2.386 1994 7.3041 ** 2.559 5.7518 4.516 6.1540 ** 2.281 1995 7.5398 ** 2.924 7.6667 5.181 5.5314 ** 2.588 Constant 60.6398 ** 11.56 61.3954 ** 16.51 50.8380 ** 9.903 Number of Observations 70 70 70 Adjusted R 2 0.79 0.66 0.85 Legend: ** significant at 5% level * significant at 10% level 94 — "3 E . « I-- — 1 - = is 2 B • = « — -B — ~£ t« un rn V M - - co oo CO CN — CO TT uo CO VO CO CN CO CO CO CO V O vo "V CN VO CO CN UO Tt r-CN ro Tf UO -H CO T)- CO s cu s_ 2 U Lo e I to .2 £ — 3 CO -O V CJ 8 O © o ca — S •o 01 — >> « IS CO t - 00 CO CO t> oo Tf co uo ov CO Tf uo o CO uo CO ^ CN CO CO CN CO CO uo CO CO CO CN fH uo uo oo I uo o CN ON Tt © 1 cu l-p o © CJ — s •a CJ UO CO V -f* s o ( A "3D W B leg uo CO V vo r-uo uo r-H ro uo uo uo CN Tt UO CO uo UO Tt CN uo CN O CN uo r~ o CN — D. a CS CO rS t- o rS CJ ID uo uo ON ON 0 \ ON uo uo ON ON ON ON UO uo ON ON ON ON uo uo ON ON ON ON f H ON 00 00 ON ON f H ON 00 00 ON ON f H ON 00 00 ON ON f H ON 00 00 ON ON B CS •f* CW B o u I S CJ a CJ ca CJ > £ Pi T3 rS o J2 CJ CJ 2 CJ CJ CJ s 3 a r-N 95 as 13 a £> O >S cu O. >> H E CS s cs cu E o +H s o . 0 s o U n a cs (» -a cu +H J3 .Sf '3 a P 3 es H mm uo cs t~ 00 00 OS OS ro OS OS cs OS o o\ Tf co cs Xr- CO SO CO CS fH a cs t~- OS cs fH U0 OS cs Tf wo OS so 00 00 OS Tf cs os CO UO Tf •H r- cs so os fH so |-~ SO co Tf SO 00 co CO fH r~ cs wo r~ Tf uo 1—I t- os so cs fH 00 so CO r-o so UO fH cs CS fH cs fH fH fH cs cs cs T H H fH cs CS fH fH H ^ i-H fH oo so Tf wo cs Tf OS oo CN OS wo Tf OS ro oo Tf wo wo Tf oo Tf Tf cs o Tf cs Tf so cs OS r H o CS ro oo OS OS ro Os cs cs wo wo OS WO wo oo wo WO <N OS P N o o o wo Tf wo ro ro t» o OS OS OS o so so o so CN Tf ro wo Os oo C D Ti- Tf 00 ' CN 00 ^ 00 ^ ^ ^ 00 ro cs uo Tf ,_ cs CS OS Os _ C-- wo r- Tf ro WO SO o wo OS Tf OS ro SO uo ro OS cs ro ro 00 SO wo OS oo r~. cs cs os o SO cs cs wo wo SO OS oo OS SO CS c* O OS o Tf Tf r H cs ro r H e cs ro wo cs Tf cs o r—1 CN Tf ro r-H (N cs o u cs cs cs CS TI- Tf 00 OS WO ro oo SO oo OS o 00 ro ro oo wo cs c-~ 00 Tf wo oo f~ o CN CN 00 CS wo r- 00 wo 00 cs ro SO Tf r~ r H so r- cs 00 OS Tf HH ro Tf r-~ z © so SO Tf Tf r H cs SO r H CS ro CN Tf cs SO H H ro ro r H ro CN so fH fH ^— SO Tf O 00 OS wo fH r- so cs CS o CS CO Tf so Tf o uo 00 OS Tf Tf CO o Tf fH SO fH CO. fH wo cs OS o CO so uo Os CS Tf SO Tf r~ CS r- CO fH 00 so TW 00 cs O fH SO OS SO 1—1 Tf fH o cs so o Tf o wo o Tf Tf 1̂  Os CO SO Tf o O H fH cs Tf fH Tf Tf Tf SO oo Tf Tf cs o , , WO OS cs Tf o oo Tf o Tf WO OS 00 ro WO Tf o Tf so ro o oo Tf ro I—H os CS o Os © o OO OS ro o Tf ro H H oo e •o P N WO r~ ro WO wo OS cs r- wo r- CS so ro r- ro ro Tf so CN ro CN V U cs cs cs CS o c , , OS o ro CS so SO wo oo ro r- o OS , , ro ro r- o wo Tf OS oo SO OS OS o CJ Ti- o U0 SO OS wo o Tf wo r—< wo ro cs Os oo o o uo Tf oo Tf OS SO © OS uo U U ro Tf t- *~' 1̂  cs r~ r~ Cs r- SO Tf ro c- wo ro O ro SO ro Tf o O ro SO SO WO wo o 00 ro Tf so ro so o CS SO f- so Os wo CS r-r o CS SO cs 00 oo ro o CN Tf o Tf wo ro © o CJ z ro cs SO r H r H so F H SO H H H —1 r H SO uio; C O •*-» - - os 00 f~ o ON Tf 00 se CO r- o r- r- r- 00 o t- 00 r> SO uo Tf t- OS 00 00 t~-3 cs r~ wo co fH t- Tf CO so CO Tf cs CO fH 00 SO wo SO Tf CO 00 r- fH CO CO o CO o •fl cs Tf t~- wo se WO o 00 CS fH t- © 00 r- o Tf wo Tf rH uo fH uo o 00 o H Tf CO r- ^ ^ l-~ fH »—1 1—( r~ rH ̂  fH fH r-~ CS oo e Tf oo CS OS r- r- o ro o OS wo Tf ro WO Tf o ro wo so OS wo o CN o o oo OS ro oo oo o r~ OS f- ro OS ro .—1 WO ro SO oo OS o o Tf ro oo Tf os P N OS Tf Tf o OS oo Tf Tf CO r- Tf cs t-~ OS CN CO ro Os os ro wo wo co -4-» cs cs uo wo ~* wo r~~t uo W cs ro uo _ OS vo SO Tf _ 00 © U0 00 ro SO CS ro ro wo Tf o o OS o r~ oo UO Os 00 t- cs SO 00 o WO WO cs so f- 00 I—1 cs SO r- i—H cs OS r—H 00 SO t-r c SO wo cs cs cs r H cs cs r H cs ro CN cs p H ro f H r-H r - H f—1 cs u — ^ ^ 1/0 r- cs wo cs SO ro WO wo OS r- cs © OS 00 wo OS cs OS o 00 OS r- , - H oo cs , 00 00 r- wo cs CS ro oo Tf cs f- OS OS Os Os ro WO f- SO o r—t oo 00 Tf SO f-z SO ro o ro cs r H H e i-H CN o r H CS r-H CS r—1 © — 3 es es > C S — TfOTfTfTfOSTf — g « CSCOCOTfUOUpsoW r K r ° OS © W0 P H WO W0 W0 © r ° faH - H f s c s c o c o T f u o s o H u a SS ffi OH I S « S td Vi O <A es B a 9 CJ H«J es u C '•S CJ U CJ — CJ o a 'it •v a a s a <- ^ pa fi ft u « i i s _ Z Z C O c o ^ e Q H 3 u w .2 ° « _ .g = -s r •S _ a S -S -S s a "E 3 O U I OH o B CJ Q. 3 O u I z u B CJ xt U 3 O OX) B I Z C/5 B ej OX) CJ -J 96 CU C a a S CU I e o « e 03 cu E o u c s o U ja> "a. £ OS -a O X c p T* S3 H o H 0- U Z u 00 O N C N O N O in C N 00 o 00 O m r-C N m ci Tt e Tt C N in O N C N 1-H C N i—1 r~ r- Tt O N ci NO C N 00 Tt o o r- 00 Tt 00 m o Cl m 00 O N Tt o 90 C N Tt O N 00 rt C N Cl e o O N <n Cl O N r- ci o r-O N C N C N »—i C N O N O N CN o C N CN O N rt Tf O m ro ro NO o r- ro rt CN rt ro m C N ro rt 0 0 CN Tt C S o rt m ON ro C N m 0 0 r- ro in NO 1—< Tt NO NO o ro CN CN rt o oo o ON ro NO m I—1 CN oo CN © CN P- m VO NO NO NO ro in <n NO ro ON CN o Tt NO m NO —i — C N O » H O O N O N CN CN O r H - H C O O ' - i n N 0 0 * M h * O H rOTt — CNcOCNCNON m © oo h •* oo C N V O O m 0 0 O O N » H CN Tt Tt — Tf I-H O N NOCNmTtCNNOĈCN r o r O T t o v o o O T t - H •-iCNTtTf̂ HCNCNON in oo O «n in O N O N NO Cl O N ro OO O N ro rt Cl ro © CN in O N O CN ro N O ro O N V O r- O N ro f~ oo N O CN O N i—< ro N O oo ro NO OO —* oo O N V O z so m CN ro i—I Cl CN CN ro CN Cl CN CN CN Cl V O t-~ NO ^ 00 O N O N V O r~ Cl o O N NO Tt r~ t- Cl Cl NO 00 C N O N Tt Cl CQ m 00 Tt p~ O N O N r- Cl N O Tt 00 m O N Tt © m r- in o r- Tt TM N O 00 m Cl N O Tt 00 m C N C N m Tt O N 00 Cl N O C N <n V O m r- m C N Tt Cl «n Tn  1 0 Cf) Cl Cl Cl in o m Tt CN ro O V O NO ro in ro CN O N 0 0 rt CN m 0 0 m O N CN N O C O O N in r~ 00 vo V O oo O N O N o m ro 00 O O ro f~ O N ro NO 00 V O m m CN Tt —H 00 "O N N O N ro C N CN Tt ro V O ro C N CN v  m t T—* C N m ro Tt ro —H CN CN C N « u U C N C N C N C N O r- NO O m O N O O N O N O 0 0 ro Cl O N vo Tt ro © rt Cl rt m ro r-~ NO CN O N Cl • r o C N NO CN ro r- >o r~ C N Tt O N OO r- in C N © © vo V O O N C N u Z u Tt Tt 00 ' 1 ' 1 00 —* *~1 00 CN 00 Tf m oo <n 0 0 O N Tt oo CN m 0 0 i , ro ro NO Tt m Tt ro Tt O N © O N N O m vo ro V O oo Tt N O rt Tt Cl Tt O N O N Tt O N N O Cl ro Tt O N OO Tt r~ in Cl CN Tt Tt Tt Tt | O N m Tt C N Ov Cl O N C N t~ Tt 1-i 00 O N 00 Tt in Cl m 00 00 Tt Tt C I V O o Tt Tt C N Cl C N o Tt N O Cl C N C N © Tf N O C N C N C N © Q C N C N C N C N H V O rt t- r- >n 0 0 >n V O CN ro r- m CN NO CN in r~ r- © 0 0 Tt vo ro O N r- Tt CN CN i—< CN * H CN ro CN i — i ro CN i-H PH La U O N Tt Cl CN Tt r- o CN ro Tt Cl Tt r- ro m rt ro Cl © CN C- Tt © ro Cl z C N C N C N C N u Tt © Tt CN ro Tt 00 CN vo CN Tt © O N CN r- NO © Tt Tt ro O O CN 0 0 CN Tt Tt CN V O vo N O CN N O C/2 cn t- e m Tt Ov e O N O C N 7-1 e NO Tt C N Cl 00 © C N vo O N © Cl O N o cs V O r~ Tt Tt O N m O N Cl m m Tt C N © NO V O Cl Tt Tt 00 C N 00 00 N O o O N Tt •*-» m o C N o C N O N m t-~ o NO o Tt V O Cl N O © 00 00 O Tt O N vo © 10  ci C N vo vo 1 vo *-< V O ^ O N e O N o CN N O o 0 0 Tt rt o ro © Tt O N r~ © CN r- Tt O N r~ NO m © S CO * H f- o V O CN ro Tt m i—< » H m O N © CN in i—< i—H in 0 0 ro CN M " ro r- T H V O r-- r- ro Tt 1—1 Tt O N © Tt 0 0 ro V O t~~ r~- CN m Tt E g> u CN Tt Tt Tt Tt "5. in rt N O «n r- Tt CN NO NO o \o V O CN f~ rt O N ro Tt N O in >n m ro NO t-~ in N O O N V O CN t- O Tt o CN V O O N C~~ © m CN N O NO CN ro ro O N NO Tt V O s z m Tt o CN CN CN © » r t CN i—1 CN i - H © CN CN i—1 —H o o r- r- Tt r-~ ro CN ro *o 0 0 Tt Tt rt ro ro O N rt r- Tt m Tf © O N r- o O N Tt Tt V O CN vo CN CN CN N O N O rt O N CN © NO NO Tt OO Tt 0 0 © CN NO z m ro 00 CN CN 00 —H CN i—t 00 CN —-< 00 O o JS u Vi % 5 — 2 93 > — e w — i CV O S H T f © T t T t T t O N T f C N ci ci T T m m vo • • • • • • i O N © m » - < m m m © — C N C N C l C l T t v j v o — DuoiioDoJiexiexicXMo cn l i ^ i 1 B ffi * CM ,a (U 1 | E | ? 8f W co O vi U O o H _ B ' fe •a e s s V ^ a « Z Z C O CO ^ 03 2 " 2 13 o U Vi rt '•B « k. o D3 H u a CU "5. 3 O U • 0- U fi CU rt U S o a. 3 O U I z o s cu u S o Ci) c Z C O •o Ci cu OC cu 9 7 tu a. co X l o i cu O. H >> 1 « to •a a a w & o u a o V a x> • f. u +H CO 5 a • as M Tf Ji 3 a H T ot al  51 .0  49 .0  o o fH ro wo 23 .2  21 .3  13 .4  22 .0  10 .3  <N ro Tf o o ro t-' 17 .3  23 .8  14 .7  23 .8  13 .1  © o fH rj Tf ro Tf 28 .5  35 .4  o CO 11 .0  13 .6  o o o © fH VO •* o o _ Tf ON CN <N © ro o © VD _ oo ON VO © CO CN CN ON Tf o vq OH u CO Tf wo o HH CN VD wo OO CN d d o fH od ON w-i CN Tf CN d © w-i w-i CO CN w-i ro CN Os CN © fH vo Tt VO o CN 00 OO ro vq Tf ON CN o r- vq _ © O VO CO Tf OO f H vq oo wo o oo z o CN uo Tf o HH CN CN WO d Tf o fH t-^ wo vd CN <N CN © fH ro d ro vd ro ro © ro © fH CO ro r- o _ H VO Tf wo wo o _ o r- ON vq vq vq vq © CN CN Tf vq OO wo wo © vo SN  r~ WO CN Tt o fH ro ro © ro CN wo ro <N o fH Tf r~ c~ vd CN CN © fH " CN CN ro ro CN w-i r~ © ON T ot al  45 .5  54 .5  o o fH "0 ri 17. 8 23 .2  15 .5  25 .4  11 .6  fH co fH o o fH UO 14 .1  24 .9  11 .6  27 .3  17 .2  © © uo ri CO co 2 5. 4 42 .7  ON ri 10. 2 13 .0  © o HH O O ON _ o wo ON ro CN © ON CN o o HH ON CN VD wo o ro ON ON © ON oo rn o oo >r ed  C P Tf oo wo o fH CN ro CO ON ro ro - d © fH VD w-i r~ CN d Tf CN w-i © ro ro WO CN ro Tf CN ON © d VD C en s(  C N  47 .7  52 .3  o o ro 20 .2  29 .3  ON 13 .2  16 .6  Tf CO © ro © o fH ro 14 .1  26 .3  11 .5  29 .6  14 .8  o o fH ro ON CN 25 .3  45 .7  ro © ON 12 .8  o © fH 19 .7  wo wo e Tf t-- Tf r~ vq wo r- © wo CN f H wo © VO o © ON CN CO t-» 00 p- o wo :o m e SN  Tf WO wo Tf © fH CN CN CN ro —* Tf wo ro CN o fH ro 00 vd Tf ro ro Tf <N © fH Tf CN od ro CN CN 00 © ON O ut t T ot al  54 .5  45 .5  o o fH n 26 .6  20 .1  12 .0  19 .8  wo ON co ro wo fH o o P-00 19. 4 23 .2  16 .7  21 .5  10 .6  © o fH n uo ON Tf 30 .5  30 .7  fH CO 11 .6  14 .0  © o fH O o fH oo CN o 00 vq CN oo CN CN oo wo o CN CO CO wo f H wo © Tf o vo CN o t- CN o f H HH C P CN VO t ~ Tf o fH ON wo CN WO ro w-i CN ON d d o fH ON CN ro CN r~ d CN © r~ vd ON CN ro CO os ro © fH CN VD Exi  vo Tf o f H ON r~ oo Tf wo vo © o uo vq © 00 f H f H © CN CO VO Tf © o © © ro Exi  Z u u-i wo Tf Tf o fH —* CN CN VO CN vd d Tf CN w-i o fH d vd vd CN d WO CN © CN ro Tf ro ON CN Tf CN Tf © fH od _ H © e WO CN VO CN wo WO oo © wo ro WO vo WO vq o ro Tf Tf CN ON CN © vo SN  ON WO Tf o Tf ro ro ON CN - WO CN o vi Tf 00 ON CN CN ON © fH CN ON CN d ro CN vd o fH ON a O T3 VI -H C ™ B C *-fH CU O § tZ H Tf © Tf Tf Tf ON Tf n ro co Tf uo uo vo I I I T I I I ON © U0 - H U0 U0 U0 © f H n n c o c o T f u o v o — < < < < < < < < H e o 15 5 !§> 1 / 3 t» 9 — o 5 « » o •o e S 1. U O U « - .. .- t CU a ~o U k f f | •s e u « s e - _ f , = > Z Z O O t w ^ f - H a s o U I BH u u s o o U I Z u z vi B0 cu -J 98 cu a as a cu E B cu a o Z cu a. > > H B CJ •a a cu E o u 9 o CO a s cu C Xfl Ja CN TT CU 3 e s H o H U z u z cn go N o n \s 2 CN „ m Tt " o o o rt vo 2 W Tf " 1H v> Vi 00 Vi vo O ov ci vo 1-3 CN ci rn cN r-i vi rn o © m _ CN OS oo Tt Tt vd ro rn —* o 00 i - i O O O « SS c» Os vi c>i cN ci 2 H M I H M H ™ — r N T T O v o r - t — o OS Tf m Tt o vo Tt vd — CN 0O t~ Tt OS C~ O oo ro © O vd ro — 2 h O 00 oo o Tt VO Tt Tt 2 o H z Z t/3 r- co © h CN 2 TT in m vo o CN K 2 Tt w-> CN oo o CN K 2 r- ov v> ri © Ov CN H H H fN n H N q o CN vi ov ov 2 CN rt —< — vq Tt CN — OV SO m © ro CN o ov vd Tt vo ro vo — ro O so 00 © OS -H h Tt © K od « (N 2 r-; oo ro © so as oo ro 2 o H OH U z u z cn OS - H © vo ro in © Tf ^ CN 00 © od - 2 t— CN rN vv t- tn — Tf OO © os vq Tt © in « Tf o CN ro Tt vo so r— v> t> vq CN © © od vi rn 2 CN 00 ro © vo © * so - ©' 2 vo in t"~ ro © rt OO 00 os 2 - os f- « o O —' © CN 2 CN —" - « ™ o •S H a E PM u z u z OS CN © VO CO 2 1/1 t ro © o< s Tf " OS -H © Tt in 2 m Tf " so ro ii m Tf " oo vi vi r- ro os rN Tf - H <N Cl Tf f- o o Tt CN f •» H fi H ' H N r t " © m os in m' vd vd so VO Tf Os ro © — CN O O 2 CN OS Tf CN H « d oo CN ro — in © in OS Tt © OS © CN Tt 00 CN 2 in Tf h Tf © ro in cN —• 2 in rt 00 Tf © o © oo rN OS in in © VO od CN <N vd CN © ' CN © vd in OS CN m CN OS © rt OS in oo CN OO rt oo ro o in CN _ oo CO rt o © CN o Tf CN CN ro CN o ro ©' ro od ro CO © ro © CN CN o in OO 00 CN o ro in Tf vO OS CN rt © ro CN in CN CN CN © CN <N CO CN in vd © rt od 10 .9  22 .4  25 .8  r- OS 19 .7  11 .5 o © rt OS Tf Tf Tf 34 .9  32 .2  oo CN rt OS 1 1.7  O O rt © o —H m ro Tt in 00 SO o rt ro © <N rt © in OS Tf CN vd <N OS OS o © rt in in ro © ro CN cd i - 1 © cd in rt o © r- © ro S OS in CN rt OS r~ 00 O 00 in CN CN in CN OS OS OS o —~' ro v i ro vd ro CN cd © rt vd CN ro Tt CN r- Tf o vq oo CN O in vq m © OO OS tn in CN © ©' CN OS © rt CN CO Tf CO ro ro CO © Tt 20 .3  30 .3  14 .0  13 .5  16 .0  OS ir, O O ro vi CN 47 .1  23 .3  1-H 11 .8 00 00 © o o o ro OS SO Tf rt O Tf VO 00 00 Tt ro r- o ro O <N ©' ro © ' ro in OS © Os CN in ro ro CN CJ © rt ©' in in ro in t— CN o © CO CN OS © CO OS © OS OS CN CN in vd CN CO CN © rt © ' CN Tt CN CN O in CN © o SO OO OS in CN O O Tf CN Tf ro CN © VO © OS ©' <N ro OS o CN © © rt ro VO CN ©' © ' © rt od ro CN 16 .9  24 .9  13 .3  23 .5  14 .3  O O rt ro rt Tf 25 .8  38 .4  CN ro 11 .5 13 .4  S © © o rt r- ro Tf m co O CO o rt _ CO oo O in ©' CN vd CN CN CN - O rt in in CN od ro CO © ' © rt o vo in in 00 00 O CO „ O rt t-; CO rt O OS ro Tt CN CO C~ CN CN © rt CO CN Tt CO © Tf © rt OS CN so VO VO CN Os O rt Tt «n Tt 00 CN VO O VO in 00 ©' od CN Tf © rt CN ro SO CN SO © rt OS CN ro rt rt 3 rt i 1 « 5 I O S fa H Tt © Tt TT Tt OS Tf CN ro ro T t VI in so I I I I I I I OS O Vi - H VI VI VI © -HCNCNrorOTtvivo— S oUOJIMOlloODXIOIOXlo <<<<<<< <H •S — T< O & s 33 PH •a « .5 !o BJDTS - i O O W «W W L u .2 ci a a> *r +-1 s #• io .fl • - * ; z C O tJI •< « H c 3 PH U z z on -o e cu on CU 99 Table 4.3: Duration Model Estimates Spells Independent Variable Employment Time to Re-employment Time to Welfare Coeff. s.e. Coeff. s.e. Coeff. s.e. Female -0.1566 ** 0.017 -0.3923 ** 0.018 -0.6461 ** 0.160 Singles w/out Children -0.0176 0.021 0.1526 ** 0.022 1.0332 ** 0.154 Couples w/out Children -0.0441 0.024 0.1239 ** 0.023 -0.5232 ** 0.263 Some High School -0.1467 ** 0.034 -0.0403 0.045 -0.4385 ** 0.203 Grad. High School -0.3690 ** 0.034 0.0394 0.042 -1.6353 ** 0.272 Some Post Secondary -0.1050 ** 0.035 0.2505 ** 0.045 -0.6827 ** 0.260 Certificate/Diploma -0.4485 ** 0.033 0.2746 ** 0.042 -0.9532 ** 0.228 Degree -0.6041 ** 0.038 0.2381 ** 0.044 -1.5517 ** 0.304 Age 19 0.8962 ** 0.039 0.5454 ** 0.050 1.1302 ** 0.384 Age 20-24 0.5081 ** 0.025 0.2458 ** 0.027 0.6078 ** 0.225 Age 25-30 0.1372 ** 0.028 -0.0199 0.025 0.3051 0.241 Age 31-34 0.0171 0.032 0.0602 ** 0.028 0.1131 0.266 Age 45-54 -0.0693 ** 0.035 -0.2753 ** 0.033 -0.4985 * 0.285 Age 55-59 0.1920 ** 0.045 -1.1383 ** 0.053 -0.6526 ** 0.319 Age 60-64 0.2290 ** 0.073 -2.0670 ** 0.082 -1.2405 ** 0.411 Newfoundland 0.4727 ** 0.098 -0.4699 ** 0.149 0.0474 0.611 New Brunswick 0.3319 ** 0.098 -0.3456 ** 0.129 -0.6558 0.976 Quebec 0.2782 ** 0.022 -0.4177 ** 0.032 0.3982 0.245 Saskatchewan 0.1556 0.102 -0.1034 0.120 -0.5761 0.882 Alberta 0.3148 ** 0.045 -0.1058 ** 0.051 0.8283 ** 0.348 British Columbia 0.2921 ** 0.034 -0.1645 ** 0.041 -0.0633 0.354 Q2 0.1919 ** 0.036 0.2557 ** 0.031 -0.0697 0.363 Q3 0.4992 ** 0.037 0.0034 0.035 0.0738 0.352 Q4 0.8441 ** 0.030 -0.0112 0.028 0.3084 0.235 EP Rate - 20-34 yrs. (1% pt). 0.0223 ** 0.006 0.0536 ** 0.007 -0.0223 0.059 Min. UI Qual. Weeks (1 week) -0.0211 ** 0.075 Max. UI Benefit Weeks (1 week) 0.0113 ** 0.004 0.0344 0.033 Monthly Welfare Benefits ($100) -0.0230 0.016 -0.1293 ** 0.019 0.6635 ** 0.120 Welfare Benefit Change > 5% -0.0159 0.048 0.1470 ** 0.043 0.4927 0.335 Min Wage Rate ($) -0.2040 ** 0.070 Number of Observations 11797 9787 9787 LogL fn -34401.11 -23038.21 -1052.27 Legend: ** Significant at 5% * Significant at 10% Note: The baseline category corresponds to 35-44 year old male with elementary education, who married with children and living in Ontario. 100 cS o o r j oo cn cn C N o cn v© cn m Tt Tt un CN T T O NO cn O N Tt i> cn r-» o cn C N C N cn r-H C N C N TJ- cn cn C N cn IT) un un, Tt VO O N N O T t Tt cn un cn un i O O N un cn o un NO Tt N O C N T t < N N O 0 0 O N T-H oo S P Q o "it Z jg ° IS, § u u a cn C N cn un Tt Tt r-H OO cn un ON o r-H cn N O Tt un cn • cn r - un Tt — ' ON C N C N Un rH C N T t Tt O NO C N r-H cn 0 0 NO Tt OS un cn cn C N 0 0 so un C N Tt N O OO so Tt un T t C N N O <=> r-1 ^ NO C N Tt "? r-H VO ^ a U 2 « a o o CU "3D .S U r - Tt NO un C N un C N so un C N un vo 8 ^ un o o cn C N Os 0 0 un cn NO C N un Tt Tt Tt 0 0 r-H r-H Tt O un Un r-H rn cn oo un Tt r - cn C N C N cn O N un Tt Tt rH 0 0 0 0 r~- Tt 0 0 o C N C N 8 C N C N JO ' £ cu PN CU CJ "E H-< i s ^ cu 3 .2 *E a > IT; IT) O N O N O N O S O N O N O N O N tn tn O N ON O N ON tn tn Os O N O N O N tn tn O N O N O N O N Tt O N 00 00 O N O N O 09 CU 3 .2 'E a > o h CJ Tt O N 00 00 O N O N a a CU I fl o Z, >> I T t O N 00 00 O N O N T T O N oo oo O N O N I i T t O N 00 00 O N O N a Da O PH a. a o X i cu 3 a •PH u a i> r>> 3 P M a CM H-< ia cu fl CU P Q tn tn O N O N O N O N T f O N 00 00 O N ON fl a CU fl 101 102  S3S^3 jo jaqxnriNT 104   107 108 109 110 I l l 112 113 114 115 116 117 118 119 120 aiE Î aSesn VS snjsuag VS 3AIJB|3^ 121 122 aura s S E S f l V S 124 S;E>I s8Bsn V S 3«r>j aSBSfi VS 3KH aSesn VS 127 3J1TH 38ESfl vs J I I I I I I I 1 I I I I I I I I I L o o o o m Tf f*l (N P==fJ°M S5I33/VA 129 siirH s8esn VS O 3W£ S S B S f l VS snira s8Esn VS 2 <t> •rH S5[33/Y\ 130 nvyi sSEsn VS 131 132 S)E}J s S E S n VS _l L_l I I I I I I L_ _1 I I I L 133 134 3J13>I 9 § E S n V S 00 \0 (S o _ , _ ( — > » - — < co 0̂ cs O 1 ' ' I I I I I I I I I—I I—I I I I—L_ n — i — i — i — i — i — i — i — i — i — i — i — r O 00 \0 Tt CN O CO <SJ -H r-l -H 135 136 sSesn vs 138 139 9§^:).U90.I3C[ 140 CO M CD 1) +-> O • rH to <D to rH ro V to r-4 CD CD • o * • rH ^ <3 o CO ' J H CD £ H to fl CD o • ON 00 NO 1~ ON ON fl O ON O ON 3 r H PQ r H O .ON hH1 ON £ 00 "ON oo "ON 00 " ON i n a <L) r H <4H w S3 . <D fl CD ~ r r l • • 2 r D r f l r H U r H •rH PH fl o r f l 00 03 ON « r H Ofl fl • rH CO oo ON co CN 9§12XII30 . I9 C [ 141 142 143 144 145 146 147 148 149 150 151 152 cd CD r-H 1) O fi CD o U O fi CD CO CD u a CD CO CD fi u CD £>0 o ' 00 o o o AO o o VO o o o o o o cn o o (N o o o o rt o +-> o CD CD o 1̂- rt rt o m o3 N o CN d 13 o PQ • • f "i • pd 153 154 s,000 s.ooo 155 156 a < CO o „ •8 8 0 -I • + >.2 <L) g < co O « a o U 0 •I s i •8 § •s-S - J £3 u s.OOO s,000 a 0 0 in c 0> fi V B < CO O « c 0 o •8 § org 2w • + s,000 u 0 U « a ffl < CO o fi 0 U •8 S •3.2 - J • + ~i r O 00 s.OOO <4H cd O fi < 1* o o Q • • <N in • fe 157 4 00 i n s,000 s.ooo OX) 158 as 4 s.OOO s.000 • 1—I 159 IT) s.OOO s.ooo bio • r H 160 i n 4 s,000 s.ooo bb 161 Os 4 s.OOO s.000 •1-4 fe 162 o o o o o o o o o o o o S,000 p3A"0|dlU3-UON AjMSNJ S.000 pS^O[dUJ9-UOM A[M3fs[ 163 o wi o *n O vi o o o o o o o s,000 ps^Ojdius-uoM A J M S M S,OOO psfoidura-uoN A " [ M 3 N 164 165 166 167 168 169  Bibliography Allen, Douglas. (1993). "Welfare and the Family: The Canadian Experience," Journal of Labour Economics, 11(1): S201-S223. Barrett, Garry. (1996). "The Duration of Welfare Spells and State Dependence: Evidence from British Columbia", Mimeo, University of New South Wales. Barrett, Garry, (1998). "The Effect of Educational Attainment on Welfare Dependence: Evidence from Canada", Mimeo, University of New South Wales. Barrett, G., M . Cragg, (1998). "An Untold Story: The Characteristics of Welfare Use in British Columbia", Canadian Journal of Economics, 31(1):165-188. Beaudry, P. and D. Green, (1997). "Cohort Patterns in Canadian Earnings: Assessing the Role of Skill Premia in Inequality Trends," National Bureau of Economic Research Working Paper: 6132, August 1997, pages 35. Blank, R.M. (1989). "Analyzing the Length of Welfare Spells", Journal of Public Economics, v.39, 245-273. Brown, David M . , (1995). "Welfare Caseloads in Canada", Helping the Poor: A Qualified Case for Workfare, C D . Howe Institute, Toronto, pp. 37-90. Browning, M . , S, Jones and P. Kuhn. (1995). "Studies of the Interaction of UI and Welfare Using the COEP Dataset", Human Resource and Development Canada. Bruce, R., N. Bailey, W.P. Warburton, J. Cragg and A. Nakamura. (1996). "Those Returning to Income Assistance", Canadian Journal of Economics, 29 (Special Issue, Part I), S33-S38. Charette, M . and Meng. R. (1994). "The Determinants of Welfare Participation of Female Heads of Household in Canada," Canadian Journal of Economics XXVII, 2:290-306. Christofides, L. , Stengos, T. and Swidinsky, R. (1997). "Welfare Participation and Labour Market Behaviour in Canada", Canadian Journal of Economics, X X X , 3:595-621. Cragg, Michael. (1996). "The Dynamics of Welfare Use in Canada," Canadian Journal of Economics, 29 (Special Issue, Part I), S25-S32. Dolton, P. and W. van der Klaauw. (1995). "Leaving Teaching in the UK: A Duration Analysis", Economic Journal, v.l05(l), 431-444. 171 Dooley, Martin (1999). "The Evolution of Welfare Participation Among Canadian Lone Mothers from 1973-1991", Candian Journal of Economics, 32 (1): 589-612. Duclos, J-Y, B. Fortin, G. Lacroix, and H. Roberge. (1996). "La dynamique de la participation a l'aide sociale au Quebec: 1979-1993," Report Prepared for le Ministere de la Securite du Revenu du Quebec March 1996. Fortin, Bernard and G. Lacroix. (1998). "Welfare Benefits, Minimum Wage Rate and the Duration of Welfare Spells: Evidence from a Natural Experiment in Canada", Mimeo, Departement d'economique, Universite Laval. Fortin, Bernard, G. Lacroix and Thibalt, J. (1997). "Welfare Program and Lone Mothers' Welfare Spells in Quebec", Mimeo, CREFA, Universite Laval. Fortin, Pierre and Pierre-Yves Cremieux. (1998). "The Determinants of Social Assistance Rates: Evidence from a Panel of Canadian Provinces, 1976-1996.", Mimeo, Department of Economics, University of Quebec at Montreal. Gilbert, L. , Kamionka, T. and G. Lacroix. (2000). "The Impact of Government-Sponsored Training Programs on the Labor Market Transitions of Disadvantaged Men", Mimeo, CREFA, Universite Laval. Green, A. and W.C. Riddell. (1997). "Qualifying for Unemployment Insurance: An Empirical Analysis", Economic Journal 107, 17-35. Health and Welfare Canada, (various years). "Inventory of Income Security Programs in Canada", Ottawa: Minister of Supply and Services. Heckman, J.J. and B. Singer. (1984). "A Method for Minimizing the Impact of Distributional Assumptions in Econometric Models for Duration Data", Econometrica, v.52(2), 271-320. Heckman, J.J. and B. Singer. (1985). "Social Science Duration Analysis, in J.J. Heckman and B. Singer (eds.), Longitudinal Analysis of Labor Market Data, Cambridge:* Cambridge University Press, 39-110. Lacroix, Guy. (1999). "Reforming the Welfare System: In Search of the Optimal Policy Mix", Mimeo, CREFA, Universite Laval. Lancaster, T. (1990). "The Econometric Analysis of Transition Data", Cambridge: Cambridge University Press. Manton, K .G. , E. Stallard and J.W. Vaupel (1986). "Alternative Models for Heterogeneity of Mortality Risks Among the Aged", Journal of the American Statistical Association, v81(4), 635-644. 172 Meyer, B.D. (1988). "On Unemployment Insurance and Unemployment Spells", Econometrica, v58(4), pp. 757-782. Moffitt, Robert. (1992). "Incentive Effects of the U.S. Welfare System: A Review.", Journal of Economic Literature 30(1): 1-61. Morissette, R., J. Myles and G. Picot (1994). "Earnings Inequality and the Distribution of Working Time in Canada," Canadian Business Economics, 2(3): 3-16. National Council of Welfare, (1987). "Welfare in Canada: The Tangled Safety Net," Ottawa: Minister of Supply and Services. National Council of Welfare, (various years). "Welfare Incomes," Ottawa: Minister of Supply and Services. Prentice, R. and L. Gloeckler. (1978). "Regression Analysis of Grouped Survival Data with an Application to Breast Cancer Data", Biometrics, 34, 57-67. Social Assistance Review Committee (SARC, 1988). "Transitions: Report of the Social Assistance Review Committee." Prepared for the Ontario Ministry of Social Services , Toronto, Ontario, Queen's Printer. Stewart, Jennifer and M.D. Dooley. (1998a). "An Analysis of Changes in Welfare Participation Rates in Ontario From 1983-1994 using Social Assistance Caseload Data," Mimeo, Department of Economics, McMaster University. Stewart, Jennifer and M.D. Dooley. (1998b). "The Duration of Spells on Welfare and Off-Welfare among Lone Mothers in Ontario," Mimeo, Department of Economics, McMaster University. 173 Appendix 1 - Data This appendix describes the sources of the data used in this thesis and the construction of certain variables. A. Chapter 2 Data on asset exemption limits is from the Provincial Gazettes (various years). B. Chapter 3 1. S A Caseload levels and Usage Rates Administrative social assistance caseload data was obtained through Human Resources Development Canada - Social Program Information and Analysis Directorate, both directly and through volumes of the publication "Inventory of Income Security Programs". These data were used to construct usage rates for the aggregate specifications in Chapter 3 (i.e. the specifications that examine al l recipients together). These data were also used to construct Figure 1 and used in comparisons with caseload measures derived from microdata. Survey of Consumer Finances of Individuals microdata files (1981-1995) were used to construct welfare usage rates for the disaggregate analysis. 2. Relative S A Benefit Rates Relative S A benefit rates were constructed as the ratio of a measure of welfare benefits to full time minimum wage earnings (based on 40 hours worked per week). Min imum wage data were obtained from volumes of the Provincial Gazettes (various years). Different measures of welfare benefit rates are used for the aggregate and the disaggregate analyses. (a) Benefit Rates - Aggregate The measure of welfare benefits used in the aggregate analysis is the average total expenditure per case. Expenditure data represents the total social assistance expenditures excluding administrative or "in-kind" benefit costs for the fiscal year ending March 31. Benefits per case is constructed as the total expenditure over the fiscal year, divided by the average number of cases from the start and end of that fiscal year. This measure is then assigned to the March at the end of the fiscal year in question. For example, the expenditures of the 1991-92 fiscal year is divided by 174 the average of caseloads as of March 1991 and March 1992 to arrive at the benefits per case measure for March 1992. Expenditure data was obtained from Human Resources Development Canada - Social Program Information and Analysis Directorate (b) Benefit Rates - Specific to Family type Welfare rates by type of recipient were obtained from volumes of the Provincial Gazettes (various years) and also directly from provincial ministries. Monthly social assistance benefits represent the maximum potential benefits (including basic allowance, shelter and utilities) for individuals classified as employable. Where rate differentials exist for short-term vs. long-term support, the rate for short-term receipt has been used. Until August 1989, childless individuals less in Quebec who were less than 30 years of age were paid a lower benefit rate than similar individuals aged 30 years and older. The benefit rate measure used for single without children represents a weighted average of these two benefit rates, using as weights the proportion of singles without children in these two age groups. Starting April 1984 and lasting until the end of November 1987 a similar benefit differential existed in B.C. with respect to childless individuals less than 26 years or age vs. similar individuals aged 26 years or more. Again, the benefit rate used represents a weighted average of the two rates using the proportions of such individuals in the two age groups. Until June 1985, there existed a benefit rate differential for childless individuals in New Brunswick. Such individuals less than 40 years of age were paid a different rate than individuals aged 40-54 years. A weighted average similar to those described above is used for this group. Note that individuals above 54 years of age are excluded in the disaggregate analysis. Measure of Labour Market Conditions Different measures of labour market conditions were used in the aggregate analysis vs. the disaggregate analysis. The aggregate analysis uses a measure defined as the unemployment rate for men aged 24-44 years. This data series was obtained from CANSIM and its original source is the Labour Force Survey. In the disaggregate analysis, the measure used is the average number of weeks worked during the year among a given family type (note that this includes both welfare recipients and non-recipients). This measure is constructed using the Survey of Consumer Finances data. 175 4. Unemployment Insurance Variables The minimum weeks necessary to qualify for UI is constructed using the relevant regulation schedules and a 3 month moving average of the seasonally adjusted provincial unemployment rate for both sexes, 15 years and older for the December, January and February months preceding March of the year in question. UI regulation data were obtained from volumes of the Canada Gazette (various years). 5. Use of March values Although some variables are available by month (family type-specific benefit rates, minimum wages, UI qualification weeks and unemployment rates), other are only available on an annual basis (SA expenditure, SA usage rates and average weeks worked). The monthly data is made compatible with the annual data by using the values of these series corresponding to the March of the year in question. C. Chapter 4 1. Job Separations, Re-employment and Welfare Take-up The basic data used in the duration analysis regarding job separations, re- employment, welfare take-up and demographic characteristics is from the 1988- 1990 file of the Labour Market Activity Survey. These data were supplemented with the following. 2. SA Benefit Rates As in Chapter 3, welfare rates by type of recipient were obtained from volumes of the Provincial Gazettes, various years and also directly from provincial ministries. Rates specific to family type are used and for couples with children I have employed the rate applicable for two children. For periods after August 1,19891 have selected for individuals in Quebec the rates applicable for individuals in the "Available" category. For individuals in Newfoundland the rates for employable individuals less than 50 years old were used. 3. Minimum Wage Rates As in Chapter 3, minimum wage data were obtained from volumes of the Provincial Gazettes (various years). 176 4. Measure of Labour Market Conditions The employment to population rate for persons aged 20-34 years was obtained from Labour Force Historical Review. 5. Unemployment Insurance Variables The minimum weeks necessary to qualify for UI and the number of weeks of benefits for someone with 20 weeks of insurable employment are constructed using the relevant regulation schedules and a 3 month moving average of the seasonally adjusted provincial unemployment rate for both sexes, 15 years and older for the 3 months preceding the month to which the measure is assigned. UI regulation data were obtained from volumes of the Canada Gazette (various years). D. Chapter 5 Data used in the simulations performed in Chapter 5 employ covariates that are identical, both in terms of source and construction, to those used in the duration analysis in Chapter 4. The constructed probabilities of welfare take-up following a job separation are applied to data from monthly individual level files of the Labour Force Survey. Administrative data used in comparisons with predicted incidence rates are from volumes of the publication "Inventory of Income Security Programs". 177 Appendix 2: Supplements to Chapter 4 - Duration Analysis This appendix presents supplements to the duration analysis of Chapter 4. These include unweighted sample counts and distributions for non-employment spells ending in welfare take-up for various threshold levels of earnings used to distinguish between low- earnings and high-earnings states. Also included are plots of the covariates used in Chapter 4. 178 o l l — SO T t © -M MM 90 T t t- rt rt © o O 90 t- 90 o rs NO rs T f OS NO rt O ro SO o —M T f T f rs ro rs rt o T t NO ro rs rs o T f SO rs rs rs © o MM rs rs rs rs f- SO o so 00 i n 00 so T f rs rs rt NO SO i n ro so rt i n NO 00 rt r~- T f rs r - NO Pri r— T t —M rs r s rs 1-H rs ro rs rt ro rs © <J -M rt rt rt o IT, t ~ i n rs M̂ i n SO © rt rs ro T f rs T f NO ro i n rt ro rs © rs 00 T f © so rs rs z rs rs rs rs u ro OS rs r s rt T f oo rs so rs rs o ON r s so i n o rs T f ro r - SO rs oo rs rs z T t rt so rt rt i—i NO —M rt rt rt so rs rt NO —- 00 1/j ro rt rs ON T f 90 —M rt r- ro © rs 90 90 90 ro rs SO rs m OS 1— rs ro ro SO o HH T t T f rs ro rs rt o T f NO ro rs rs o T f rt SO rs rs rs o o rt rs rs rs rs H SO rt r- OO i n OO SO i n rs rs rt r- SO SO ro SO rt i n r~ OO rt r - T f r~ rs oo PM TT rs rs rs rs ro rs ro CN o u rt rt rt o T f 00 i n ro rt in o rt r s ro T f ro T f l> ro in i—i ro ro O rs OO T f © r - rs ro z —M rs rs rs rs u T f OS ro r s rs T f oo rs so rs ro © OS r s i n o ro T f ro r - r s oo rs ro z T t —H SO rt rt i—i SO rt rt rt rt NO r s <—i NO —- OS in T f *M rt O ro ON rt rs r- T f rt 1—< t- 90 ON 90 T f rt m ro m ON 90 ro T t ro SO © MM T f in rs ro rs rt o T f SO ro rs rs o T f NO rs rs rs o o MM rs rs rs rs H SO rt t - r-~ T t ON in SO r s ro rt t - r - i n rs SO rs i n r - © OO T f ro oo CM T t rs rs rs r s ro r s ro r s © <J MM rt rt rt O ro OS T t ro rs T f r - © rt rs ro T f ro T f r - ro i n rt ro ro © r s T f © r - ro ro Z rt rs rs rs rs u T t o T f rs ro T f 00 rs r - SO rs T f © OS rs r~ SO o T f T f ro oo r - rs oo r s T f z T t rs SO —M rt rt so rt rt rt rt SO r s rt NO — OS IT) T f i—i rs ON ro OS rt rs t - T t MM MM 90 OS 90 T f MM m ro m 90 90 T t T f ro SO o 1-H TT T f rs ro rs o T f SO ro rs rs © T f rt NO rs rs rs O o -M rs rs rs rs H SO rt t - i n 00 i n SO r s ro rt r - in rs so rs i n © OO T f NO ro ON r--PM r - T f rt r s rs rs rs ro rs ro r s rt o U rt rt rt MMm rs Os T f ro rs T t r - o rt rs ro T f ro T f ro i n ro ro o rs t- T f o r~- ro ro z rs rs rs rs u T t o TI- rs ro T f oo rs r - SO rs TT o Os <N r~ SO O T f T f ro oo r - rs oo rs T f z T t rs NS rt rt se rt rt rt rt SO r s rt SO tzi — OS 00 f- MM rs O T t © -M rs r^ i — rs rt r~ 00 O OS r- 1—I m T f NO ON 90 T f r-ro so o rt T f m rs T f rs o T f SO ro rs ro © T f rt SO rs rs rs s s rt rs rs rs rs H SO rs 90 r- i n oo i n rs ro rt 90 r - i n rs SO rs SO 00 r^ © 00 T f ro Os 90 PM T f i—I r s rs rs i—I rs ro rs rt ro rs © U i—I -M rt iH o rs OS T) T f rs T f 00 o —M rs ro T f T f i n r~ ro i n rt ro T f o rs oo T f © r - ro T f z rs rs rs rs u T t rt m rs ro T t OS r s r- so r s m © OS tN t - r~ © m T t ro 00 00 rs oo rs m z T t r s NO rt —M rt NO rt rt —M rt NO r s —M NO — o so 00 —M ro © T t o rt rs 90 rs rt 00 rt OS 90 rt m T t l-~ ON 90 T f 90 T f so © rt T f m rs T f rs -M © T f NO ro rs ro O T t MM NO rs rs rs O o rt rs rs rs rs H r - r s OS SO 00 m rs ro rt ON i n rs SO ro so ON © OO i n ro OS OS PM r - T f rs r s rs rt rs ro rs rt ro rs rt o MM rt rt rt o rt Os i n T f rs T t 00 © rt rs ro T t T t i n ro i n rt ro T t © rs oo T t o r^ ro T f z rs rs rs rs u T t rt IT) r s ro T f Os rs r~ so r s m © ON r s t ~ o m T f ro oo 00 r s oo rs m z T t r s NO rt rt so rt rt rt rt se Csl *—H SO U M — T f © T t T f T t O S T f r t • • S c S rs ro ro TT m m so ™ M M r « o CN o if. M « « m o r ° is H M M r s r s r o r o T f i n s o H "S "o "6 E ® -§ S -2 •S « S .& • 1 | 1 1 E r « 2 — o "r o 1 1 « o H « U % U Q H "i U fj O fj pa £ 'C « o O lm Vt rt V cu Z Z . S ^ 2 « S S c 'u o c o w < n H 179 eg H v DX S -3 s 0) a E >, _o a. E CD I c o Z • a E c s 6M Oil > < a s tu a cs < z a H To ta l 62 .8  37 .2  e o Tf 20 .7  25 .1  11 .9  20 .5  OS ad Tf i/5 ro ci O o 20 .5  30 .4  14 .5  12 .6  15 .9  o ve © o ve ui rl ri 47. 7 22 .7  90 11 .7  ro od e o 22 .1  o C P 56 .1  43 .9  o e cn SO 18 .6  26 .2  17 .4  24 .1  rn vd Tf d VO d o o 19 .9  30 .8  10 .9  13 .9  15 .4  CN Ov o e oo Ov OS CN 35. 6 32 .1  SO CN 10. 2 oo vd o o in CN  76 .0  24 .0  e o Tf m' 23 .8  33 .2  o d 00 od Ov' 20 .1  o o 20 .4  19 .0  16 .2  27 .9  Ov rn 12.7  o e o d OS CN 44. 2 23 .9  O d 23 .6  Tf in e o SN  68 .0  32 .0  o o Tf 22 .4  21 .5  p ob 20. 9 12 .4  11 .2  CN e o rH 21 .5  33 .2  18 .6  00 vd 20 .0  o d o o in CN 65. 0 vo Os m 10.5  11 .0  o o To ta l 63 .1  36 .9  o o Tf 21 .5  24 .9  11 .7  20 .5  00 od rn i/5 rl ci O O 20 .2  30 .5  14 .2  13 .6  15 .6  ui e o if; ui rl ri 46 .9  23 .4  90 11 .8  m od o o 21 .7  o C P 55 .7  44 .4  o o m vd 18. 5 26 .0  17 .2  24 .7  m vd Tf d vo d © O 19 .7  31 .3  10 .9  13 .8  15 .3  OV o o oo Ov OS CN 35. 3 31 .9  VO CN 10. 1 m r-- o o TT C N ' 76 .7  23 .3  o o m rn 23. 2 35 .0  o o vq vq od 00 od 19 .6  O e 19 .8  21 .1  15 .7  27 .1  oo rn 12.4  o o o d OO CN 43. 0 23 .3  O d 25 .7  m «n o o SN  o as SO 3 1. 0 o o Tf 24 .8  20 .8  oo 20 .3  12 .0  10 .8  CN o e 20. 8 32 .2  18 .1  vo Ov 19.4  p d o o Tf CN 63. 0 12 .4  CN 10 .1  10 .7  o e To ta l 62 .9  37 .1  o o n Tf 21 .4  24 .9  11 .5  20 .5  r~ 00 ve ui rl ni © e 20 .3  30 .3  14 .0  13 .5  16 .0  ov i/i o o ro ui ri 47. 1 23 .3  00 11 .8  r- 90 o o 21 .6  O C P 55 .5  44 .5  e o H H >d 18.2  26 .1  17 .0  24 .8  m vd p vo d © e 20 .3  30 .9  10 .6  13 .7  15 .4  Ov o o Tf OS SO CN 35. 8 31 .8  vo CN 10. 3 m r-̂ o o CN  78 .2  21 .8  e o OS »n 21 .6  34 .4  o d vq vn od r~ od 19.3  e o 19 .5  22 .3  15 .5  26 .7  r- ro 12. 2 o © o d OO CN 41. 2 22 .9  O d 25 .3  Ov © o SN  68 .3  31 .7  o o Tf 25 .6  20 .6  r~ i-̂  20 .1  11 .9  10 .7  CN o o 20. 6 31 .8  17 .9  T. Ov 20. 2 o d o o Tf CN 63. 4 12 .3  CN 10 .0  10 .6  o o To ta l 62 .9  37 .1  e o rH rl Tf 21 .5  24 .7  11 .5  20 .5  r~ 90 VO ui rl rn e o H H 20 .3  30 .3  14 .0  13 .5  16 .0  OS ui o o ro ui ri 47. 1 23 .3  f"; 11 .8  00 90 o o 21 .5  o n C P 55 .5  44 .5  o o vd 18. 4 25 .8  17 .0  24 .8  m vd o VO d o o 20 .3  30 .9  10 .6  13 .7  15 .4  OS OOT Tf Ov vq CN 35 .8  31 .8  Tf CN 10. 3 r- r-̂ o o H H C N  78 .2  21 .8  e o ov in 21. 6 34 .4  o d vq <r, od r~ od 19. 3 o o 19 .5  22 .3  15 .5  26 .7  r--rn 12. 2 © © O d 00 CN 41. 2 22 .9  o d 25. 3 OS o o M S 68 .3  31 .7  OOT Tf 25 .6  20 .6  20 .1  11 .9  10 .7  CN o o 20. 6 31 .8  17 .9  vn ON 20. 2 o d e o Tf CN 63. 4 12 .3  CN 10 .0  10 .6  o o To ta l 62 .5  37 .6  o o rl Tf 21 .4  24 .9  11 .7  20 .4  vq od vq ui rl ro o o 20. 5 30 .1  13 .9  13 .4  16 .2  OS ui © o ri i/5 ri 47 .1  23 .4  I - 11 .7  00 od o o 21 .5  e C P 55 .4  44 .6  o o vd 18. 4 25 .8  17 .0  25 .0  m vd o vo d © o HH 20 .2  30 .9  10 .6  13 .7  15 .4  CN OS o o Tf Ov SO CN 35. 8 31 .7  in CN 10. 2 r- o o o rl C N  75 .6  24 .4  e> e in 20. 9 36 .6  o d vq CN od Tf od 18 .6  e> e 22 .2  21 .6  15 .0  25 .8  vq rn 00 OOT o d r~ CN 43 .1  22 .2  o d 24. 4 VO e o SN  67 .8  32 .2  OOT Tf 25 .4  20 .5  Tf od 20 .0  11 .8  10 .6  o CN o o rH 20 .5  31 .6  17 .8  Tf Ov 20 .7  o d © o Tf CN 62. 9 12 .9  CN 10 .0  10 .5  e o fo ta l 63 .5  36 .5  o o Tf 23 .6  24 .2  11 .4  19 .8  Tf 00 Tf ui fi OOT 20 .0  29 .3  13 .5  13 .0  18 .5  00 ui e o ui © ri 45. 8 25 .5  11 .4  i/i 90 e> o 20 .4  D. z u z vi 2 • a. 00 C N O C N 2 »n Tf VO Tf O <n Tf 2. r~ OO Tf o vq Os p SO O HH CN o o O r~ in CN CN Tf CN vd rn CN in d o H H OS Ov CN d rn d CN od r- Ov VO o SOCN Tf vq o CN vq p oo vq oo in d CN vd cn d ~* od od od o CN CN CN in in CN rn Tf Tf m Tf o 00 so o e> »n SO 00 Tf r~ o in CN d CN od d CN d CN o d CN cn Ov d CN d Ov in OO Tf Tf r- m od CN rn m vn cn CN OS o r- — CN o Tf vo d CN rn Tf CN CN d Tf CN Tf CN OsOv CN o vn CN SO CN d o o 2 5 VI " VI a a s ft. ,H — P 2 S u e IS H T f O T f T f T f O S T f — C N f O r O T f l / l i n s o S 1 1 1 1 1 1 1 7 = o s o v i H H U 5 i n i n o 2 H H CN CN CO r o Tf 1/1 so r * n 6d VI O ± i o u vi y T3 = C 3 3 V- <2 6 0 z z U C IS •S 3 3 O O v: E s o u L . i» M 41 « £ H -— k. o < M H 2 o u z 180 T—I—I—I—i—I—I—I—I—I—r J g a : 2 T—I—I—i—r •s .00 C "1.2 Ul Ci .«N 0 o as Os T—I o Q so m o oo 00 ON § ,43 0 ~i i i i r •s .00 C S ~ o 1 "0 c 3 2i ui ~i—i—i r ~! i i i r •5 IT, u - 3 • i—i fe 181 o Os r-H o Q VO m o 00 00 r-H s c o T — i — i — i — i — i — i — r i — i — i — i — i — r "T o o o o •s .» c - 2 ca - £• ~i—i—i—I—I—i—i—r •S erj oo C > - 0 ? 12 o - « -4 <D fl rt 2 rt 3 o 3 o r f l c3 £ C N I <! "5b • S bTj 75 E 182 183 184  •5 _co a - 2 a) 186 187 188 • r H fe 189 CD i • t -H o o C N fe 190 191 192 193 194 Appendix 3: Supplements to Chapter 5 - Simulation of Welfare Incidence This appendix presents supplements to the duration analysis of Chapter 5, namely, plots of the covariates used in Chapter 5. 195 ipuouii3do66I$ ipuouiJ3do66I$ 196 qjuoui lad 0661 $ Hiuoui jsd 0661 $ 197 i|)uoui jsd 0661 $ Wom J 3 d 0661 $ 198 199 in o v~i o i/i o v\ in o "n o <n o <n oo oo r- r- \0 NO «n oo oo r*- \0 NO *n 200

Cite

Citation Scheme:

    

Usage Statistics

Country Views Downloads
France 3 0
United Kingdom 3 0
Japan 2 0
United States 2 0
Russia 1 0
City Views Downloads
Unknown 7 5
Tokyo 2 0
Sunnyvale 1 0
Ashburn 1 0

{[{ mDataHeader[type] }]} {[{ month[type] }]} {[{ tData[type] }]}

Share

Share to:

Comment

Related Items