The dynamics of immigrant participation in ... - Wiley Online Library

The dynamics of immigrant participation in entitlement programs: evidence from Canada, 1993–2007 Yuri Ostrovsky

Statistics Canada

Abstract. Using uniquely rich Canadian administrative data from the 1993–2007 period, I find no evidence of rising Employment Insurance (EI) participation across recentimmigrant cohorts and no evidence of rising EI participation with number of years spent in Canada beyond the short period following entrance into the Canadian labour market. I also find little evidence of the rising immigrant participation in Social Assistance (SA), either across recent arrival cohorts or with years in Canada. The results do not appear to be seriously affected by emigration and attrition within each immigrant cohort. While the immigrant participation in traditional ‘welfare’ programs such as Social Assistance has generally declined from 1993 to 2007, I show that Canada Child Tax Benefit and other federal and provincial programs aimed at providing financial assistance to families with children have become a major source of transfer income to immigrant families. JEL classification: J61, J65, I31 La dynamique de la participation des immigrants aux programmes auxquels ils ont droit : résultats canadiens 1993–2007. A l’aide de données administratives canadiennes particulièrement riches pour la période 1993–2007, on découvre qu’il n’y a pas d’évidence de participation plus e´ levée au programme d’Assurance Emploi (AE) d’une cohorte d’immigrants récents a` l’autre dans le temps, non plus que d’évidence de participation accrue a` l’AE des immigrants récents avec le nombre d’années passées au Canada, si ce n’est pour la courte période qui suit l’entrée dans le marché du travail canadien. On ne trouve que peu de support pour l’hypothèse de participation accrue a` l’Assistance Sociale (AS) tant d’une cohorte a` la suivante qu’avec le nombre d’années passées au Canada. Ces résultats ne semblent pas affectés par l’émigration ou l’attrition a` l’intérieur de chaque cohorte. Alors que la participation des immigrants aux programmes de bien-être traditionnels comme l’Assistance Sociale ont généralement décliné entre 1993 et 2007, on montre que la prestation fiscale canadienne pour enfants et autres programmes fédéraux I am grateful to Jorgen Hansen, René Morissette, and two anonymous referees for many helpful comments and suggestions. None of the opinions expressed in this paper should be attributed to Statistics Canada. Email: [email protected] Canadian Journal of Economics / Revue canadienne d’Economique, Vol. 45, No. 1 February / février 2012. Printed in Canada / Imprimé au Canada

0008-4085 / 12 / 107–136 / C Canadian Economics Association

108

Y. Ostrovsky

et provinciaux visant a` fournir de l’assistance financière aux familles avec enfants sont devenus une source majeure de revenus de transfert pour les familles d’immigrants.

1. Introduction The mostly optimistic view about the benefits of immigration that dominated economic research and policy discussions in the 1970s and 1980s has been challenged by numerous recent studies arguing that the labour market outcomes of immigrants in many western countries are deteriorating. It has also been suggested that immigrants increasingly rely on unemployment insurance, welfare, and other forms of government transfer payments. Borjas and Trejo (1991, 1993) were among the first to examine trends in welfare participation rates among immigrants in the United States and warned that such rates were on the rise. Since then, the issue of immigrant participation in various entitlement programs has received considerable attention in the economics literature.1 Such interest can easily be understood: whether or not immigrants represent a welfare burden is an important aspect of the debate about the costs and benefits of immigration. Additionally, immigrants today constitute a substantial portion of labour market participants in most western countries, so the issue of immigrants’ participation in entitlement programs is increasingly important in understanding more general employment-welfare patterns. Baker and Benjamin (1995) were the first to examine immigrants’ participation in major entitlement programs – Employment Insurance 2 (EI) and Social Assistance (SA) – in Canada. Although they concluded that immigrants did not ‘pose an excess burden on Canada’s transfer programs’ (671), they also argued that the participation rates in entitlement programs were higher for those who came to Canada in the 1970s and 1980s compared with previous cohorts, and that there existed a positive correlation between number of years spent in Canada and the incidence of EI and SA. In a subsequent study, however, Crossley, McDonald, and Worswick (2001) showed that the results in Baker and Benjamin lacked robustness and were specific to the survey years examined in their study. Like Baker and Benjamin, they focused on two important questions: (i) has immigrant participation in major entitlement programs been rising across immigrant cohorts? and (ii) does within-cohort immigrant participation in major entitlement programs rise with years in Canada? Based on the analysis of 13 survey years from the Survey of Consumer Finances (SCF) (Baker and Benjamin 1 See Barrett and McCarthy (2008) for a review focusing on the European literature. U.S. studies include Borjas and Hilton (1996), Hao and Kawano (2001), and Haider et al. (2004), among others. 2 Until 1996, the program was known as Unemployment Insurance (UI). In the remainder of the text, I will refer to the program as ‘EI’ regardless of whether I refer to the pre- or post-1996 period.

Immigrant participation in entitlement programs

109

use only two survey years), Crossley et al. argued that the evidence of the rising immigrant participation entitlement programs was, at best, mixed. The possibility of rising immigrant participation in entitlement programs is troubling and, given the lack of conclusiveness in the above-mentioned studies and the evolution of welfare programs in recent years, the matter deserves reexamination. Several specific reasons underscore the importance of revisiting the issue. First, the EI and SA programs underwent major overhauls in the early and mid-1990s. The new EI eligibility rules and changes in the generosity of SA may have created new participation dynamics that are substantially different from the pre-reforms dynamics analyzed in previous studies. It is known that the reforms played an important role in reducing the overall rates of EI and SA participation (Lin 1998; Kneebone and White 2009). However, the post-reforms dynamics of EI and SA participation among immigrants have not yet been documented in the literature. Second, while most studies focus on programs such as EI and SA, a modern western welfare state offers an array of entitlement programs, many of which aim to help families with children. As the generosity of EI and SA in Canada was substantially reduced in the mid-1990s, financial assistance to families with children has expanded through federal and provincial programs aimed specifically at providing financial assistance to low-income families with children, and programs such as the Canada Child Tax Benefit (CCTB) introduced in 1993 have become primary vehicles of government transfers to low-income families with children. Finally, the survey data on participation in entitlement programs used in previous studies are known to be problematic.3 As I show in section 2, administrative data linked to immigration records can be expected to produce more precise estimates of immigrant participation in EI and SA than SCF data used in previous studies. In addition to offering a much larger sample of immigrants, the data from the uniquely rich Longitudinal Administrative Databank (LAD) used in this study have the advantage of being accurate, detailed, and recent. I report several new results. First, I find no evidence of rising EI participation rates across recent arrival cohorts and no evidence that EI participation within cohorts rises with years in Canada beyond the initial short-term increases that most likely reflect the growing eligibility of immigrants for EI benefits. Using an econometric approach similar to the one in Green and Worswick (2004, 2010), I show that once the differences in family market income, age, and composition and other observable differences between immigrants and the reference group are removed, the estimated long-run profiles of EI participation of immigrant men from six post-1990 arrival cohorts are similar to the long-run profiles of men 3 Picot et al. (1998), for instance, report that the SCF captured about 75% to 80% of annual government transfers and that underreporting was particularly severe for Unemployment Insurance (the precursor of EI) and Income Assistance (the precursor of SA).

110

Y. Ostrovsky

from the reference group who entered the labour market around the same time. Second, the share of SA benefits in total transfer payments to immigrant families was markedly lower in the mid-2000s compared with the mid-1990s. It was also lower for the reference group, but the magnitude of the decline was larger for immigrant families, especially couples and unattached individuals. Neither descriptive nor econometric analysis shows any evidence of rising SA participation across recent immigrant cohorts or rising SA participation rates with years in Canada. Finally, I show that the share of child and family benefits in total government transfers to immigrant families has surpassed the share of SA benefits and now represents the single largest source of government transfers to such families.

2. Major entitlement programs in Canada Employment Insurance (EI), known as Unemployment Insurance before 1996, is a self-financing federal program aimed at providing temporary financial relief to laid-off workers while they are looking for a new job. In addition to regular benefits, the EI program offers special benefits to expectant mothers, new parents, and individuals who look after disabled family members. To be eligible for regular benefits, individuals must work a qualifying number of insurable hours, which depends on the unemployment rate in the region where they live. The benefit rate is 55% of the average insured earnings up to a specified maximum. Until 1993, the eligibility for regular benefits was nearly universal (Green and Riddell 1993; Lin 1998). The only exclusions were self-employed individuals (except self-employed fishermen) and those who did not meet weekly earnings requirements. In the 1980s and early 1990s, however, the program ran large deficits and was often criticized for creating incentives to work ‘just enough’ to qualify for benefits (Sargent 1995). In 1993, the program underwent a radical overhaul (Bill C-113). Those who quit, were fired for misconduct or refused suitable employment became ineligible for benefits, and the replacement rate was reduced from 60% to 55% (for low-earnings claimants, it was raised to 60%). Over the next several years, the number of beneficiaries fell sharply. The ratio of regular beneficiaries to all unemployed declined from 83.4% in 1989 to 48.1% in 1996, and the program turned from large deficits to surpluses (Lin 1998). The next round of EI reforms took place in 1996 (Bill C-12), when the program was renamed Employment Insurance to emphasize the objective of promoting employment. Major elements of the 1996 changes to the system included benefit claw-backs and other eligibility restrictions for repeated claimants. Social Assistance (SA) in Canada is under the jurisdiction of provincial governments. The programs are income and asset tested, and the size of benefits and eligibility criteria varies from province to province. Although the eligibility requirements for SA and payment rates are generally similar across provinces, there are differences. The eligibility for SA usually depends on family size, income,


111

assets, and housing costs.4 The latter may include rent or mortgage payments, electricity, heating, and phone costs. The number of individuals receiving SA grew steadily in the 1970s and 1980s and peaked in the early 1990s (Finnie, Irvine, and Sceviour 2004). In the mid1990s, in response to the growing concerns about the rising ‘welfare dependency,’ the provincial governments tightened eligibility rules, reduced benefit levels, and took measures to limit abuse of the system.5 The federal government also took several related steps, essentially transferring all the responsibility for SA expenditures to the provincial governments. Two other developments also had implications for SA participation rates. First, SA payments to families with children were reduced after the introduction of the Canada Child Tax Benefit in 1993. Second, as mentioned above, the EI system underwent major changes during the 1993–1994 period, which may have affected the number of people applying for Social Assistance and other forms of financial help. By the early 2000s, the incidence of SA and the rates of entry into the program declined dramatically for all family types, especially for lone mothers (Finnie, Irvine, and Sceviour 2004; Kneebone and White 2009). All in all, although the magnitude of the effect varied across demographic groups, family types, and provinces, there was a considerable across-the-board decline in SA participation in Canada in the late 1990s. Canada Child Tax Benefit (CCTB) is a federal program that replaced the Child Tax Credit and federal Family Allowance in 1993. Unlike the non-refundable Child Tax Credit, the CCTB is available to parents with incomes below the taxable threshold. Payments under this program are determined according to the number and age of the dependent children and their parents’ income. Eligible parents continue to receive these (non-taxable) benefits until each child reaches her/his 18th birthday. Canadian citizens, permanent residents, refugee claimants, and temporary residents can receive the CCTB. In addition to the basic benefit, low-income families may also qualify for the National Child Benefit Supplement to the CCTB, and in some provinces, SA payments are reduced in proportion to the amount of the supplement. Since 2006, eligible families with children under 6 years of age also receive Universal Child Care Benefit (UCCB) payments for each qualified child. Families with children under 6 years of age who already receive CCTB are also eligible for the UCCB, but other families may also qualify. Provincial family benefits include non-taxable benefits from various provincial programs aimed at helping parents (or guardians) of children under 18. Until 1992, residents of Quebec received Family Allowance payments from both the federal and the provincial governments. Since 1994, provincial family benefits have included various provincial programs such as, for instance, the Child Care Supplement for Working Families in Ontario or the Child Tax Benefit in 4 In addition to financial assistance, the recipients may qualify for other benefits such as municipal tax payments, prescription drug coverage, eye exams and prescription glasses, diabetic supplies, burial expenses, eviction or relocation costs, and employment-related costs. 5 Kneebone and White (2009) provide a detailed review of the SA reforms in Ontario, Alberta, and British Columbia, where the changes were particularly important.

112

Y. Ostrovsky

Saskatchewan. Parents do not need to apply separately for these benefits, and in most provinces (with the exception of Quebec) eligibility is determined based on the information provided in the CCTB application. To get an idea about the relative importance of different entitlement programs, I have computed the shares of transfer income received through each of these programs in total transfers and total income using administrative data from 1994 and 2007. Since CCTB and provincial family benefits are programs that provide financial assistance to families with children, I combined these benefits into a single variable: Child and Family Benefits, or CFB.6 The shares are computed based on a sample of immigrants aged 20 to 59 with less than 15 years in Canada and all other tax filers of the same ages (table 1).7 Two important trends with respect to SA and CFB stand out. First, there has been a sharp decline in the share of SA income in total transfers to all tax filers, including immigrants, particularly to couples and lone parents. For instance, whereas in 1994 the share of SA income in total transfer income received by lone immigrant parents was 65% (one-third of the total income received by these families), the corresponding amount in 2007 was only 25.3% (7.8% of the total income). For other lone parents, the share of SA income in total transfer income declined from 56.7% (19% of the total income) in 1994 to 20.7% (4.5%) in 2007. Second, in contrast to 1994, the share of CFB in all transfers received by immigrant couples and lone parents in 2007 far exceeded the share of their SA income. In 1994, recent immigrant couples received, on average, 30.8% of their total transfer income through SA benefits and 23.5% through CFB; in 2007, the corresponding numbers were 9.1% and 59.2%. Among other couples, the share of SA income in total transfer income dropped from 14% to 7.1% between 1994 and 2007, while the share of CFB rose from 20.4% to 38.7%. The latter increase, however, was markedly smaller than the corresponding increase among immigrant couples (from 23.5% to 59.2%). This first look at the data strongly indicates a major shift in the relative importance of transfer incomes received by immigrant families toward transfers aimed specifically at helping families with children. Whereas in the past much of the financial assistance to immigrants was provided through traditional ‘welfare’ programs such as SA, CFB is now accounting for more than half of total government transfers to immigrant couples. The shift points to the increasing role of demographics and, more specifically, the presence of children in the participation in financial assistance programs. In the remainder of the paper, however, I will focus primarily on the receipts of EI and SA since – except for high-income families – the eligibility for CFB is almost entirely determined by the presence of children in the family. Once the difference in the presence of children between 6 For 2007 this category also includes the UCCB. 7 As immigrant status in the data is available only for tax filers (not the members of their families), the immigrant status of the tax filer determines the immigrant status of a married or common-law couple.

27.9 30.4 6.4 30.9 28.1 30.1 9.6 28.9

4.0 4.0 3.3 4.1 2.2 2.3 2.9 1.4

39.2 44.3 9.5 38.0 34.8 40.9 11.3 27.4

3.2 3.0 3.2 4.1 1.7 1.6 2.5 1.7

36.3 30.8 65.0 46.7 12.1 9.1 25.3 35.3

5.2 4.0 33.3 6.2 1.0 0.7 7.8 1.7

Total income share 22.7 14.0 56.7 35.3 14.1 7.1 20.7 36.6

Transfer income share

Other tax filers

1.8 1.0 19.0 3.8 0.7 0.3 4.5 2.3

Total income share

20.5 23.5 20.1 – 55.8 59.2 60.3 –


Recent immigrants

2.9 3.1 10.3 – 4.4 4.5 18.5 −

Total income share

17.0 20.4 22.0 – 34.5 38.7 59.4 −


Other tax filers

Child and Family Benefits

1.4 1.4 7.4 – 1.6 1.5 13.0 −

Total income share

Lone parents with children under 19 NOTES: ‘Recent immigrants’ are immigrant families with 15 years or less in Canada; ‘other tax filers’ are non-immigrant families and immigrant families with more than 15 years in Canada. SOURCE: LAD

a

1994 (all families) Couples Lone parentsa Unattached 2007 (all families) Couples Lone parentsa Unattached




Total income share

Recent immigrants

Other tax filers

Recent immigrants

Total income share

SA Benefits

EI Benefits

TABLE 1 Shares of EI, SA, and CFB payments in total transfers and total income (%)

114

Y. Ostrovsky

immigrants and non-immigrants is taken into account, statistical analysis is unlikely to provide any additional insights into the differences in the incidence of receiving CFB between the two groups.

3. Data and sample The LAD is a random 20% sample of all Canadian tax filers and their families created from information provided by the Canada Revenue Agency. Individual tax filers in the LAD are linked across years based on their unique identification number, and once selected into the LAD, individuals remain in the LAD regardless of whether they file tax returns in subsequent years. To account for population growth each year, the sample is augmented with new individuals so that it represents 20% of all Canadian tax filers in that year. The filing rate in Canada is very high, largely because low-income individuals have strong incentives to file tax returns to receive various benefits and refundable tax credits.8 Although the LAD is based on individual tax returns, it also allows for familylevel aggregation. The information on spouses and common-law partners is based on the information provided in tax returns or – if necessary – matched according to names, ages, addresses and other relevant information. Based on the reported personal information, individuals can be classified into three broad categories: (i) married or common-law with or without children, (ii) lone parents,9 and (iii) non-family (unattached) individuals and ‘filing children.’10 Those who became permanent residents in Canada before 1980 cannot be identified as immigrants in the LAD. For immigrants who became permanent residents after 1979, administrative data are linked to immigration records from the Longitudinal Immigration Database (IMDB). Although most immigrants begin filing taxes after becoming permanent residents, this is not true for all immigrants. About 20% of immigrants in the LAD had filed taxes before the year in which they received permanent resident status. The share of immigrants who filed taxes before becoming permanent residents is particularly high in certain categories of immigrants such as refugee claimants (about 40%), but very low among, for instance, business-class immigrants (about 3%). The exclusion of immigrants who arrived in Canada before becoming permanent residents would result in a disproportionately larger number of refugee status claimants being dropped out of the sample. Therefore, I kept these immigrants in the 8 According to Statistics Canada estimates, the filing rate for the 15-to-64 age category was almost 90% in 2005. The coverage can be expected to be higher for those 20 and older, as the coverage rate for the 15–19 age group is only about 52% compared with 86% to 98% coverage for other age groups. 9 There is no restriction on the age of children in the LAD. However, I redefined the ‘lone parent’ category to include only those lone parents who have children under 19 (the ages of children are available in the LAD). Other ‘lone parents’ (those with all children over 18) were re-classified as non- family individuals. 10 ‘Filing children’ are non-married persons of any age living with their parents.


115

sample, excluding only the years prior to receiving permanent resident status and defining the year in which they received permanent resident status as their arrival year. Because immigrant-specific variables are available only for the tax filers in the LAD (not their family members), an immigrant family is identified by the immigrant status of the tax filer. Therefore, a mixed (immigrant/nonimmigrant) couple will be identified as an ‘immigrant couple’ only if the tax filer in the LAD is an immigrant spouse or partner. The arrival year of the tax filer also determines the arrival year of the couple.11 The analysis of EI participation is based on the LAD Employment Insurance Benefits (EINS) variable, which in addition to the benefits related to employment loss also includes benefits related to pregnancy or birth or adoption of a child. Since childbearing affects only women, I limit the analysis of EI participation to male filers. An indicator of EI participation in year t takes the value of 1 if a male tax filer had a positive annual amount of EI benefits in that year and zero otherwise. The EI analysis sample consists of (i) male immigrants aged 20 to 59 who became permanent residents in Canada between 1991 and 2002 and (ii) the reference group that includes Canadian-born men aged 20 to 59 and immigrant men of the same ages who became permanent residents before 1980. Given that immigrants constitute only a small proportion of the general population, the presence of pre-1980 immigrants in the reference group should have little effect on the overall levels of welfare participation in that group.12 Because financial assistance and benefits are predicated on the financial situation of all adults in a family, the analysis of SA is conducted at the census family level. ‘Family SA income’ is defined in the following way: for married or common-law couples, the family SA income is the combined SA income of the couple, whereas for lone parents and non-family individuals, the family SA income is the SA income of the lone parent or non-family person.13 Sample selection criteria are the same as the sample selection criteria for the EI sample, and the husband’s age is assumed to be the age of a couple. Although pre-1993 LAD files are also available, in the context of this study, the pre-1993 data may not be directly comparable with the post-1992 data because of the changes in filing incentives related to the introduction of the CCTB in 1993 (Frenette, Green, and Picot 2006, 70–1). The overall filing rates were not much lower before 1993 compared with those of later years, but the differences can be expected to be considerably larger for low-income families and individuals, who

11 According to author’s calculations using the 2006 Census data, about 30% of married and common-law couples with at least one immigrant spouse are ‘mixed.’ Assuming that the proportion of mixed couples is about the same in the LAD as in the Census, and all the non-mixed immigrant couples are identified correctly as immigrant couples, I estimate that about 85% of all immigrant couples (mixed and non-mixed) are identified as such in the study. 12 About 8.5% of all individuals in the 1996 Census were immigrants who arrived in Canada before 1980. 13 In the case of couples, SA income must be reported by a spouse or common-law partner with a higher net income, which further underscores the importance of the family-level analysis.

116

Y. Ostrovsky

are the focus of this study. To ensure consistency of the sample, the analysis is limited to the 1993–2007 period.14 Immigrants who were younger than 18 years old when they arrived in Canada have been excluded from the analysis sample, as the labour market outcomes of those who arrived in Canada as children can be quite different from the labour market outcomes of those who arrived in Canada as adults. For convenience, immigrants (immigrant families in the SA analysis) are grouped into six twoyear arrival cohorts based on their arrival year: those who arrived in 1991/92, 1993/94, 1995/96, 1997/98, 1999/00, and 2001/02. The first cohort was observed from 1993 to 2007, the second one was observed from 1995 to 2007, and so on. The last cohort was observed for five years, from 2003 to 2007. Hence, both study samples are unbalanced panels spanning the period from 1993 to 2007. An exit from the sample may be due to reaching the age of 60, death, emigration, or not filing tax returns. A summary of the sample statistics is given in table 2. One of the advantages of using the LAD is that it is likely to produce more accurate EI and SA participation rates than household surveys like the SCF. Kapsalis (2001) estimates that, compared with that of administrative data, the rate of underreporting of EI in the Survey of Labour and Income Dynamics (SLID) − the successor to the SCF − is about 15%, and the extent of underreporting of EI benefit payments is about 10%. The corresponding numbers for SA are even higher: the rate of underreporting for SA is between 20% and 30% (Kapsalis 2001). A LAD-SCF comparison of the annual EI rates from 1993 to 1996 for all men aged 20 to 59 shows that the EI rates in the SCF are about 13% to 19% lower than in the LAD.15 The underreporting of EI benefits is even greater for immigrants, although the magnitude of underreporting varies widely from year to year. For the 1981/85 arrival cohort, the EI participation rate in 1994 was 56% lower in the SCF than in the LAD, but in 1993 it was only 8% lower. For the 1986/90 arrival cohort, the EI participation rate in 1994 was about 23% lower in the SCF than in the LAD, whereas in 1996 it was only 2% lower. The overall incidence of SA among SCF census families during the 1993–1996 period was about 27% to 35% lower than in the LAD. Like the EI rates, the LAD-SCF differences in SA participation rates computed for immigrant families varied widely from year to year. For the 1981/85 arrival cohort, for instance, the LAD-SCF difference in SA rates was only about 9% in 1996 but more than 35% in 1993. The LAD-SCF comparison above underscores the point made in Crossley, McDonald, and Worswick (2001) about the danger of making inferences about EI and SA participation trends based on weakly identified models relying on few survey years. Not only is there a possibility that survey years in this case 14 Other possible problems with administrative data may include income underreporting related to tax evasion and existing discrepancies between SA receipts in tax data and government files used for issuing SA payments. 15 These results are available on request from the author. The annual incidence of EI (SA) in the LAD is computed on the basis of positive EI (SA) benefits in the corresponding year.


117

TABLE 2 Summary of sample statistics (dollar amounts are given in constant 2007 dollars) Immigrants who arrived in Reference group 1991/92 1993/94 1995/96 1997/98 1999/00 2001/02 EI sample (men, 18,452,300 obs.) EI recipients (%) 15.9 Have children under 19 39.0 Age of the eldest child 10.7 (under 19) Lone parents (%) 1.5 Unattached (%) 33.8 Average age 39.8 Average market income 47,550 Live in large metropolitan 28.7 areas (%) SA sample (families, 47,236,850 obs.) SA recipients (%) 8.9 Have children under 19 43.3 Age of the eldest child 10.7 (under 19) Lone parents (%) 6.0 Unattached (%) 29.9 Average age 39.6 Average market income 62,700 Families with 18.0 self-employment (%) Live in large metropolitan 29.5 areas (%)

14.2 55.3 9.7

11.9 53.7 9.6

10.9 56.1 9.9

10.8 56.3 9.7

11.6 56.9 9.1

12.2 56.1 9.0

1.4 22.9 40.4 30,600 74.9

1.1 21.1 40.3 32,500 74.1

0.9 18.0 40.6 34,100 75.0

0.8 17.8 40.0 34,300 74.0

0.7 17.8 38.7 35,100 71.9

0.7 18.7 38.3 30,950 72.2

13.3 59.6 9.9

9.6 58.1 9.5

8.1 59.6 9.7

7.4 60.4 9.6

7.5 60.6 9.0

7.9 59.6 8.8

6.9 20.2 40.6 43,000 19.8

6.0 19.2 40.5 44,700 19.2

5.0 16.9 40.6 46,050 19.9

4.6 16.2 40.0 46,000 21.1

4.1 16.1 38.8 45,800 20.4

3.5 17.0 38.3 40,900 19.3

73.8

73.9

74.8

74.3

71.9

72.5

SOURCE: LAD

are located at very different points of a business cycle, but the results also will be especially sensitive to differences in response rates and underreporting from one survey year to another. Only when a long series of data is available can participation trends be assessed with reasonable accuracy.

4. The incidence of EI and SA among immigrants Figure 1 shows longitudinal profiles of EI and SA participation for six immigrant cohorts and the reference group described in section 3. An important advantage of the cohort approach is that it allows for distinguishing between the ‘betweencohort’ and ‘within-cohort’ dynamics of immigrant participation in entitlement programs. The former pertains to changes in the average participation rates across immigrant cohorts. The latter is usually understood as a correlation between participation rates and the length of immigrant residency in the host country.

118

Y. Ostrovsky

FIGURE 1 Longitudinal profiles of EI participation SOURCE: LAD

The EI participation rates are computed only for men, and the SA participation rates are computed for all census families. Figure 1 shows that the EI participation rate among the reference group steadily declined from 1993 (0.23) to 2007 (0.13) except for a slight increase in the early 2000s compared with 2000. The shape of each immigrant participation profile visibly reflects two factors: eligibility constraints in the initial years and cyclical fluctuations. Like Siklos and Marr (1996), I find that for most cohorts, the incidence of EI among immigrants initially rises as more immigrants are able to satisfy the eligibility rules, but falls in subsequent years. In 1993, the year in which significant reforms of the EI system were implemented and eligibility for EI benefits was restricted primarily to laid-off workers, the EI rates were still similar for immigrants who had just entered the labour force (1991/92 cohort) and the reference group. For the subsequent cohorts, however, the EI entry rates of immigrants were considerably lower than the EI participation rates of the reference group. Although all EI profiles show sensitivity to cyclical fluctuations − a high albeit declining incidence of EI participation during the years that followed the recession of the early 1990s and a small increase in the incidence of EI during the economic downturn of the early 2000s − the sensitivity of immigrant profiles appears to be greater. This finding is consistent with that of other studies documenting higher sensitivity of immigrants’ earnings and employment relative


119

FIGURE 2 Longitudinal profiles of SA participation SOURCE: LAD

to the ‘natives’ (Solon, Barsky, and Parker 1994). Even at the peak, however, participation rates of all arrival cohorts who entered the labour market starting in 1995 are lower than the participation rates of the reference group. In the long run, immigrant profiles appear to converge to a level below the level of the reference group. Similar to the incidence of EI, the incidence of SA was higher during the recession of the early 1990s but declined during the recovery years, which were also the years that followed the SA reforms described in section 2 (figure 2). Immigrants who came to Canada in the early 1990s had a particularly high incidence of SA during the initial years of their stay. In 1993, more than 24% of immigrant families from the 1991/92 arrival cohort received SA compared with approximately 12% of the families from the reference group. In 2003, only about 11% of immigrant families from the newly arrived 2000/01 cohort received SA income compared with 7% of families from the reference group. Figure 2 also shows that the incidence of SA gradually declined with years in Canada for all immigrant cohorts and eventually fell below the incidence of SA in the reference group. The profiles of more recent immigrant cohorts converge to the profile of the reference group faster than the profiles of their predecessors. Whereas it took the 1991/92 cohort ten years to reach the level of SA participation equal to that of the reference group, it took the 1999/00 and 2001/02 cohorts only three

120

Y. Ostrovsky

years to do the same. In 2007, immigrant families in all six arrival cohorts were less likely to receive SA than the families in the reference group. Although much of this trend is undoubtedly related to the improvements in the labour market environment and SA reforms of the mid-1990s, as noted in the previous section, some of the decline in SA participation and average SA income that occurred in the 1990s is likely explained by the introduction of CCTB and a subsequent shift in the financial assistance to low-income families with children from the SA program to various family-related programs. The cohort profiles of EI and SA participation shown in figures 1 and 2 should be considered with caution. While the availability of long-run cohort profiles allows us to see beyond the impact of short-term cyclical fluctuations, these profiles do not allow for differentiation between time and age effects. The declining incidence of EI and SA with number of years spent in Canada may, at least to some degree, reflect the aging of a cohort. The latter also makes a comparison with the reference group whose age remains relatively constant more ambiguous.

5. Econometric methods Like Crossley, McDonald, and Worswick (2001), I consider a flexible model specification that includes cohort fixed effects, a polynomial function of the number of years since migration, a set of interactions between cohort effects and ‘yearssince-migration’ variables, and a set of control variables. The baseline probit model is intended to capture the changes in the probability of receiving EI and SA both across and within immigrant cohorts. The cross-cohort changes are captured in the model by the cohort fixed effects, whereas the within-cohort trends are captured by the polynomial function of years-since-migration. This is a flexible specification that allows for different ‘assimilation’ paths for different arrival cohorts. Unlike previous studies, I consider a more flexible quartic specification of the years-since-migration polynomial. The right-hand side of the EI model takes the following form:

2001/02

α1 · A + α2 · A2 +

γj · Cji +

j=1991/92

2001/02

+

j=1993/94

2001/02

+

j=1993/94

ϕ2j · θit2 · Cji +

4 k=1

2001/02

2001/02

ϕk · θitk +

ϕ1j · (θit · Cji )

j=1993/94

ϕ3j · θit3 · Cji

j=1993/94 2007 ϕ4j · θit4 · Cji +δ · Rt + dt · Dt +β · X it , t=1994

(1)


121

where A is age, Cj are cohort fixed effects (j = 1991/92, 1993/94, . . . 2001/02), θit is the number of years since migration (θit is set to zero for the reference group), and Xit is a vector of control variables. The latter includes demographic variables such as an indicator of the family type (married or common-law couple without children, lone parent, non-family person), residence in a large metropolitan area (Toronto, Montreal, or Vancouver), province of residence, presence of two children under 18, presence of three or more children under 18, age of the eldest child and annual individual market income. Market income is defined in the LAD as total income before tax net of government transfers.16 In the model of EI participation, the market income variable measures individual market income. To control for macroeconomic conditions and business cycle effects, I also include a variable for the province-specific unemployment rates, Rt . The time effects not captured by Rt are captured by time-specific dummy variables Dt (t = 1994 , . . . , 2007; omitted: 1993). The variance estimates are adjusted for clustering, since the sample includes multiple observations for the same individuals. The specification for the SA model is the same as in (1) with additional controls for families with current students and self-employed individuals, as such families are also included in the analysis. The variables for school attendance and self-employment take a value of 1 if either of the spouses attends school or reports self-employment income. The dependent variable equals 1 if the family SA income is positive and is zero otherwise. The market income variable in the SA model measures family market income. The models defined by equation (1) make it possible to compare the probabilities of collecting EI and SA for immigrants and the reference group with similar characteristics by holding age, family type, income and other common characteristics constant. Yet immigrants to Canada also differ considerably in their own characteristics, and these differences are likely to affect the immigrants’ propensity to participate in entitlement programs. Several previous studies emphasized the effect of immigrants’ origins (Borjas and Trejo 1993), immigrant class (Hansen and Lofstrom 2003), education (Baker and Benjamin 1995; Blume and Verner 2007), and linguistic ability (Siklos and Marr 1996) on the likelihood of their participation in entitlement programs. The advantage of the LAD is that all these variables can be considered simultaneously. Although the information on current education is not available in the LAD, it does have information on the years of education upon arrival, self-reported ability to speak English or French (or both), immigrant class, and country of origin. To examine the effect of immigrant-specific characteristics on the probability of immigrants’ participation in entitlement programs I estimate a model with additional controls for the region of origin, linguistic abilities, immigrant class, and foreign education and present the results along with the results from the baseline models. 16 More specifically, annual market income in the LAD includes earnings from paid employment, net self-employment income, alimony income, dividends, interest, and investment income, limited partnership income, RRSP income, pension and superannuation income, and rental income.

122

Y. Ostrovsky

Viewed in the context of the life-cycle model of human capital accumulation, the model defined by equation (1) has one important limitation. Although it explicitly allows for cross-cohort differences in the paths of human capital accumulation among immigrants (through the cohort effects and interactions between cohort effects and years-since-migration variables), it does not allow for such differences among the reference group. Green and Worswick (2004, 2010) argue that in the context of post-arrival human capital investment the best comparison group for an immigrant cohort should consist of non-immigrants who enter the labour market at about the same time and whose human capital investment path should therefore be similar to the human capital investment path of new immigrant entrants. An important aspect of the Green-Worswick approach is that immigrants are compared with the native-born workers who plan and realize their human capital investment decisions in similar macroeconomic conditions. To implement the Green-Worswick approach I further restrict the sample only to those aged 25 to 59 to focus on individuals and families entering the ‘mature’ labour market. I then defined six entry cohorts that coincide with arrival cohorts for immigrants but also include members of the reference group who were 25 to 26 at the time that immigrants from each cohort arrived in Canada. Immigrants from each cohort are therefore paired with members of the reference group who entered the labour market at approximately the same time and were subject to the same macroeconomic conditions. In the model specification I also allow for differences in the way immigrants and the reference group are affected by years since entry into the labour market by interacting the years-since-entry and cohort variables with a dummy variable for immigrant status (Ii ). There remains an important difference between immigrants and members of the reference group within each entry cohort: unlike the latter, immigrants can enter the labour market at any age between 25 and 59. To account for age differences at the time of entry I also include a quadratic function of ‘potential foreign experience,’ (Pi ), simply defined as the age upon arrival minus 25. The right-hand side of the probit model is then given by

2001/02

γj · Bji +

j=1993/94

⎡

×⎣

4

ϕk ·

ϑitk

+

j=1993/94

γjim · Bji +

2001/02

ϕkj · ϑitk · Bji + δ · Rt + Ii

k=1 j=1993/94

k=1

2001/02

4

4 k=1

ϕkim · ϑitk + ⎤

4

2001/02

ϕkjim · ϑitk · Bji

k=1 j=1993/94

im 2 im + λim · Rt ⎦ + β · Xit , 1 · Pi + λ2 · Pi + δ

(2)

where Bji are entry cohorts, ϑit are years since entry into the labour market, and Xit is the same set of controls as in the baseline model defined by equation (1).


123

Since there is a linear relationship between the time of entry (cohort effects), years since entry, and time effects (year dummies), the latter cannot be accommodated by the model. The macroeconomic effects are captured through the provincespecific unemployment rate, Rt , which is allowed to affect immigrants and the reference group differently through an interaction between Ii and Rt .

6. Estimation results 6.1. Estimation results for EI participation In the first and second columns of table 3, I report derivatives and standard errors for the control variables in the baseline model. The derivatives are computed for a hypothetical married 40-year-old male, with a 5-year-old child and $20,000 in annual market income. The unemployment rate is set at 8%, which is approximately the average unemployment rate for the 1994–2007 period. All other Xit and Dt variables are set to zero. Although the choice of the benchmark individual is arbitrary, partial effects computed at these values seem more intuitive than those computed at, for instance, the means of Xs. All derivative estimates in table 3 are significant at the 95% level. Although the derivative with respect to the linear term in the age function in the EI model is positive, the overall marginal effect of age on the probability of EI participation is negative.17 Among different family types, lone parents are slightly less likely than couples with one child (omitted category) to receive such benefits (−0.007), whereas unattached individuals are less likely to receive such benefits than either couples with one child or lone parents (−0.045). Having two children reduces the probability of receiving EI benefits (−0.014) and having three or more reduces even further the probability of receiving such benefits (−0.018). Table 3 shows that EI participation is procyclical: higher unemployment rates increase the probability of receiving EI benefits (0.004). Place of residence also has an effect on the probability of receiving EI benefits: residents of Toronto, Montreal, and Vancouver are less likely to receive such benefits than other Canadian residents (−0.089). The residents of the Prairies (Manitoba, Saskatchewan, and Alberta) were less likely than Ontarians to receive EI benefits, while the residents of the Atlantic Provinces, Quebec, and British Columbia were more likely to receive such benefits. Not surprisingly, EI participation is strongly and significantly affected by market income (−0.004). The estimates of cohort and years-since-migration effects are not reported in table 3. These effects are summarized in figure 3, which shows the differences between the estimated probabilities of EI participation computed for a benchmark individual from each arrival cohort in each year since migration and a corresponding member of the reference group, holding other variables constant. The 17 The point estimate of the total effect can be easily computed using the derivative estimates in table 3: 0.008 + 2 · (−0.013) · (40/100) = −0.002.

124

Y. Ostrovsky

TABLE 3 Derivative estimates for the EI model, control variables (year effects not shown) Immigrant characteristics included

Baseline model Derivative

St. error

Derivative

Age 0.008∗∗∗ (0.000) 0.008∗∗∗ −0.013∗∗∗ (0.000) −0.013∗∗∗ Age2 = age2 /100 Family composition (omitted: couple with one child) couple with no children −0.033∗∗∗ (0.001) −0.033∗∗∗ lone parent −0.007∗∗∗ (0.002) −0.008∗∗∗ unattached individual −0.045∗∗∗ (0.001) −0.044∗∗∗ 2nd child −0.014∗∗∗ (0.001) −0.014∗∗∗ 3 or more children −0.018∗∗∗ (0.001) 0.019∗∗∗ Age of eldest child −0.003∗∗∗ (0.000) −0.003∗∗∗ Unemployment rate 0.004∗∗∗ (0.000) 0.004∗∗∗ Live in large metropolitan areas −0.089∗∗∗ (0.001) −0.089∗∗∗ Province (omitted: Ontario) Newfoundland and Labrador 0.207∗∗∗ (0.004) 0.205∗∗∗ Prince Edward Island 0.221∗∗∗ (0.005) 0.221∗∗∗ Nova Scotia 0.099∗∗∗ (0.002) 0.100∗∗∗ New Brunswick 0.166∗∗∗ (0.002) 0.166∗∗∗ Quebec 0.122∗∗∗ (0.001) 0.122∗∗∗ Manitoba −0.034∗∗∗ (0.001) −0.033∗∗∗ Saskatchewan −0.011∗∗∗ (0.002) −0.010∗∗∗ Alberta −0.018∗∗∗ (0.001) −0.017∗∗∗ British Columbia 0.052∗∗∗ (0.001) 0.055∗∗∗ Market income (×1000) −0.004∗∗∗ (0.000) −0.004∗∗∗ Region of origin (omitted: Central, Eastern and Southern Europe, and the former USSR) USA, UK, Australia, New Zealand, −0.062∗∗∗ Republic of South Africa, Ireland Western Europe – – 0.057∗∗∗ North Africa and the Middle East – – −0.047∗∗∗ Africa (excl. N. Africa and RSA) – – −0.005 India and the region – – 0.015∗∗∗ China and the region – – −0.077∗∗∗ Latin America and the Caribbean – – 0.006 Japan and South East Asia – – −0.051∗∗∗ Immigrant class (omitted: family class) Entrepreneurs – – −0.198∗∗∗ Skilled workers – − −0.010∗∗∗ Other (incl. refugees) – − −0.010∗∗∗ Years of foreign schooling −0.007∗∗∗ Linguistic ability (omitted: speaks neither English nor French) speaks English – – −0.025∗∗∗ speaks French – – −0.032∗∗∗ speaks both – – −0.043∗∗∗ Pseudo R-sq: 0.271 0.271 Pr (y = 1) predicted at benchmark X 0.093 0.094

St. error (0.000) (0.000) (0.001) (0.002) (0.001) (0.001) (0.001) (0.000) (0.000) (0.001) (0.004) (0.005) (0.002) (0.002) (0.001) (0.001) (0.002) (0.001) (0.001) (0.000) (0.006) (0.007) (0.003) (0.005) (0.004) (0.003) (0.004) (0.004) (0.003) (0.003) (0.003) (0.000) (0.002) (0.004) (0.004)

NOTES: ∗∗∗ p < 0.001, ∗∗ p < 0.01, ∗ p < 0.05. The derivatives for EI are reported for a benchmark male individual who is assumed to be a 40-year-old married resident of Ontario with a 5-year-old child and $20,000 in annual market income. The unemployment rate is set to 8%. In the model with immigrant characteristics included, the ‘years of foreign schooling’ variable is set to 12 for immigrants. All other variables are set to zero.


125

FIGURE 3 Differences in the estimated probabilities of EI participation between immigrants and the reference group NOTES: The profiles are computed for benchmark individuals (see text). The × mark indicates statistical significance at 95% level.

profiles confirm some of the key insights from the descriptive analysis. Controlling for business cycle effects, place of residence, market income, and demographics, the initial rates of EI participation are generally lower for the immigrants compared with the reference group but rise sharply in the next two or three years. With the exception of the 1991/92 cohort, even at the peak, the estimated conditional probability of immigrants receiving EI did not exceed the estimated conditional probability of the members of the reference group receiving EI. In the long run – 5 to 10 years after arriving in Canada – the probability of a benchmark immigrant receiving EI levels off and can be expected to be 5 to 10 percentage points lower than the probability of a member of the reference group with similar observed characteristics receiving EI. After about 10 years, the cross-cohort differences in the probabilities observed at entry essentially disappear. The last two columns in table 3 show the results for a model with additional controls for immigrant-specific characteristics such as dummy variables for the regions of origin; immigrant class; ability to speak English, French, or both; and a continuous variable for the years of foreign schooling (a detailed list of countries included in each region of origin is available upon request). Immigrants from Western Europe have the highest estimated probability of receiving EI benefits, 5.7 percentage points higher than the baseline category – immigrants from the rest

126

Y. Ostrovsky

of Europe and the former USSR. Chinese immigrants are least likely to receive such benefits (−0.077). The reasons behind these differences are probably similar to those usually invoked to explain differences in the earnings paths of immigrants from different ethnic backgrounds, such as differences in the role of immigrant networks in finding and keeping jobs and differences in the concentration of immigrants from particular ethnic backgrounds in occupations (or industries) with higher layoff rates than in other occupations. Among different immigrant categories, family-class immigrants (omitted category) are most likely to receive EI, while the ‘entrepreneurs’ are least likely (−0.198). More years of schooling at arrival and knowledge of at least one official language also reduce the probability of receiving EI.

6.2. Estimation results for SA participation Table 4 shows the results for the models of SA participation. The derivatives are computed for a hypothetical couple with a 40-year-old husband, 5-year-old child, and market income of $20,000, with all other variables set to zero to make the interpretation of marginal effects more intuitive. The choice of $20,000 in market income as opposed to a figure close to the mean family market income in the sample – around $40,000 – is motivated by the fact that families with $40,000 have a very low probability of receiving SA, so the estimated marginal effects may not be particularly informative. The probability of receiving SA benefits by a benchmark family is 0.413. Not surprisingly, lone parents are considerably more likely to receive SA benefits than other families (0.090). The probability of receiving SA also rises with family size: families with two children are more likely to receive SA benefits than families with one child (0.029), and families with three or more children have the highest probability of receiving SA benefits (0.125). The younger are the children, the higher is the probability of receiving SA (−0.012). Only in Quebec are families more likely to receive SA benefits than in Ontario (0.025), whereas Manitoba and Alberta families are among the least likely to receive such benefits (−0.153 and −0.128, respectively). Somewhat surprisingly, higher provincial unemployment rates seem to reduce the probability of receiving SA benefits, albeit only slightly (−0.009). This effect can probably be partially explained by the higher probability of receiving EI during economic downturns. Because the eligibility period for EI depends on local unemployment rates, more people can qualify for EI for a longer period of time when unemployment rates increase, making it less likely that they would apply for SA. The marginal effects of the market income variable for the benchmark family is −0.016: an additional $1,000 of market income will reduce the probability of such a family receiving SA by approximately 1.6 percentage points. Figure 4 shows the years-since-migration profiles of the differences in the estimated probabilities of receiving SA between immigrants and the reference group. Compared with the SA participation profiles in figure 2, these profiles suggest that the high levels of SA participation during the initial years in Canada


127

TABLE 4 Derivative estimates for the SA model, control variables (year effects not shown) Immigrant characteristics included

Baseline model Derivative

St. error

Derivative

Age 0.037∗∗∗ (0.000) 0.037∗∗∗ −0.044∗∗∗ (0.000) −0.044∗∗∗ Age2 = age2 /100 Family composition (omitted: couple with one child) couple with no children −0.122∗∗∗ (0.001) −0.123∗∗∗ lone parent −0.090∗∗∗ (0.001) −0.085∗∗∗ unattached individual −0.188∗∗∗ (0.001) −0.192∗∗∗ 2nd child 0.029∗∗∗ (0.001) 0.026∗∗∗ 3 or more children 0.125∗∗∗ (0.001) 0.114∗∗∗ Age of eldest child −0.012∗∗∗ (0.000) −0.012∗∗∗ Unemployment rate −0.009∗∗∗ (0.000) −0.008∗∗∗ Live in large metropolitan areas −0.057∗∗∗ (0.001) −0.055∗∗∗ Province (omitted: Ontario) Newfoundland and Labrador 0.000 (0.004) −0.014∗∗∗ Prince Edward Island −0.089∗∗∗ (0.004) −0.096∗∗∗ Nova Scotia −0.087∗∗∗ (0.002) −0.088∗∗∗ New Brunswick −0.070∗∗∗ (0.002) −0.074∗∗∗ Quebec 0.025∗∗∗ (0.001) 0.019∗∗∗ Manitoba −0.153∗∗∗ (0.002) −0.150∗∗∗ Saskatchewan −0.054∗∗∗ (0.002) −0.048∗∗∗ Alberta −0.128∗∗∗ (0.001) −0.123∗∗∗ British Columbia −0.014∗∗∗ (0.001) 0.000 Market income (×1000) −0.016∗∗∗ (0.000) −0.016∗∗∗ Region of origin (omitted: Central, Eastern and Southern Europe and the former USSR) USA, UK, Australia, New Zealand, −0.207∗∗∗ Republic of South Africa, Ireland Western Europe – – −0.261∗∗∗ North Africa and the Middle East – – −0.061∗∗∗ Africa (excl. N. Africa and RSA) – – 0.025∗∗∗ India and the region – – −0.145∗∗∗ China and the region – – −0.356∗∗∗ Latin America and the Caribbean – – −0.078∗∗∗ Japan and South East Asia – – −0.236∗∗∗ Immigrant class (omitted: family class) Entrepreneurs – – −0.293∗∗∗ Skilled workers – – 0.025∗∗∗ Other (incl. refugees) – – 0.241∗∗∗ Years of foreign schooling −0.008∗∗∗ Linguistic ability (omitted: speaks neither English nor French) speaks English – – −0.060∗∗∗ speaks French – – 0.071∗∗∗ speaks both – – 0.025∗∗∗ Pseudo R-sq: 0.403 0.410 Pr (y = 1) predicted at benchmark X 0.413 0.412

St. error (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) (0.001) (0.000) (0.000) (0.001) (0.003) (0.004) (0.002) (0.002) (0.001) (0.002) (0.002) (0.001) (0.001) (0.000) (0.008) (0.006) (0.003) (0.005) (0.003) (0.002) (0.004) (0.003) (0.006) (0.004) (0.003) (0.000) (0.003) (0.005) (0.006)

NOTES: ∗∗∗ p