*urgent finish in a hour*Read a Journal Article Related To The Course Material (Labour Economics) And Come Up 2 Questions Relevant To The Article.

Canadian Journal of Economics / Revue canadienne d’économique, Vol. 48, No. 2 May 2015. Printed in Canada / Mai 2015. Imprimé au Canada

0008–4085 / 15 / 437–463 / © Canadian Economics Association

The effect of universal child benefits on labour supply

Tammy Schirle Department of Economics, Wilfrid Laurier University

Abstract. Using a difference-in-differences estimator, I find the Canadian Universal Child Care Benefit has significant negative income effects on the labour supply of married in- dividuals. The likelihood of lower-educated mothers to participate in the labour force is reduced 3.2 percentage points when receiving the benefit. Median hours worked per week among lower-educated mothers is reduced by 1.9 hours. The effects on higher-educated mothers are substantial, with median hours worked among higher-educated mothers re- duced by nearly one hour per week. For men, the evidence suggests small but significant income effects on labour supply, consistent with the literature on labour supply elasticities.

Résumé. Les effets de la prestation universelle pour la garde d’enfants sur l’offre de travail. À l’aide d’une calibration utilisant les différences dans les différences, on montre que la PUGE a eu un effet de revenu négatif et significatif sur l’offre de travail des gens mariés. La probabilité des mères moins instruites à participer à la main d’œuvre est réduite de 3.2 points de pourcentage quand elles reçoivent cette prestation. Le nombre médian d’heures de travail par semaine chez le même groupe est réduit de 1.9 heures. Pour ce qui est des mères davantage instruites, les effets sont substantiels : le nombre médian d’heures de travail par semaine est réduit de près d’une heure. Pour les hommes, les résultats suggèrent un effet de revenu faible mais significatif sur l’offre de travail qui est en accord avec ce que suggère la littérature spécialisée sur les élasticités de l’offre de travail.

JEL classification: J22, J18

Data for this study was accessed at the South Western Ontario Research Data Centre as part of the Statistics Canada Research Data Centres Program. The opinions expressed herein do not represent the opinions of Statistics Canada. The study benefited from a short-term research grant provided by Wilfrid Laurier University. I would like to thank Kirill Savine for research assistance. I would like to thank Hideki Ariizumi, Ana Ferrer, Nicole Fortin, David Green, Stephen Jones, Shelly Lundberg, René Morissette, Arthur Sweetman, Michael Veall as well as participants of the Laurier Economics Internal Seminar Series, the CRDCN 2012 National Conference, the McMaster Economics Seminar Series, the CLSRN 2013 Annual Conference and the University of Victoria Economics Seminar Series for their comments and suggestions. I also thank two anonymous reviewers for their valuable suggestions. Corresponding author: Tammy Schirle, [email protected]

438 T. Schirle

1. Introduction

The effect of child benefits on the labour supply of parents is an issue of substan- tial interest, as most industrialized countries provide income transfers to parents of young children. Often, child benefits are designed to target low-income families and many studies have examined the effects of means-tested benefits on labour supply. In the United States, a large empirical literature has examined the impacts of the Earned Income Tax Credit (EITC) on labour supply and find significant impacts on single parent households.1 A large US literature finds that welfare programs have important labour market effects and similar effects are found in Europe.2 In Canada, Lemieux and Milligan (2008) have shown that more gen- erous social assistance benefits reduce employment. Milligan and Stabile (2007) examine the integration of Canada’s National Child Benefit and provincial social assistance programs and find strong labour market effects for single mothers in terms of employment. Consistent with this, Milligan and Stabile (2009) found that raising social assistance benefits reduced labour supply and increased social assistance receipt by low-education families.3 There is a large literature estimating labour supply elasticities, which suggests low-income workers have much higher labour supply elasticities than high-income workers (see McClelland and Mok (2012)). A study by Crossley and Jeon (2007) finds that Canadian women married to higher-income men respond significantly to changes in effective marginal tax rates. While the labour supply elasticities of married women have fallen substan- tially over time, they remain higher than the elasticities of men (see Blau and Kahn (2007); Heim (2007)).

There are relatively few studies, however, that have examined the impact of child benefits on the labour market behaviour of parents that are not low-income, in part because such universal programs are not as common and programs are of- ten designed in ways that make it difficult to assess their labour market impacts.4

There is reason to believe that child benefits may have effects quite different from other redistribution programs. Kooreman (2000) examines the expenditures of parents who receive transfers under the Dutch child benefit system (which pro- vides a pure demogrant to families with children). His evidence suggests that

1 Hotz and Scholz (2003) and Meyer (2010) review this literature which generally finds that while the earnings subsidy positively affects the labour force participation of single-parent households, there is a modest negative effect on participation of secondary earners in two-parent families. Chetty et al. (2012) have recently found positive effects on earnings on the “phase-in” region of the EITC that are higher in neighbourhoods more knowledgeable about the tax code.

2 See Moffitt (2002) and Immervoll et al. (2007) for summaries. 3 In contrast, Phipps (1995) argues the introduction of an earned-income supplement (which

would provide a subsidy to some low-earner families) would have negligible effects on labour supply.

4 A review of OECD (2011) and OECD (2013) suggests universal child benefits are more common in Europe and nine of 33 OECD countries offered some level of universal child benefits as cash benefits (excluding “birth grants”) in 2007. Sixteen of the 33 OECD countries offered some benefit (as cash benefits or tax credits) that was not means tested. To note, there are some inaccuracies in OECD (2013) including an omission of Canada’s UCCB in their list of child benefits.

Universal benefits and labour supply 439

parents treat child benefits differently from other income, as though they experi- ence a moral obligation to spend a relatively large part of the benefit on children’s goods. In that context, one might expect little or no effect on labour supply. This result could depend in part on which parent receives the benefit. Woolley (2004) suggests that paying child benefits to mothers should result in higher expenditures on food, household items and children’s clothing, as women are more likely than men to control these expenditures. Dooley et al. (2005) have provided some mod- est evidence that the cognitive and behavioural/emotional outcomes of children are associated with the mother’s share of income. Lundberg et al. (1997) have shown that giving child allowances to women in the UK increases expenditures on women’s and children’s clothing relative to men’s clothing.

More generally, there are few opportunities to examine the effect of demogrants on labour supply and obtain estimates of income effects on both the intensive and extensive margins of labour market activity. González (2013) examined the employment decisions of new mothers in response to a substantial birth grant in Spain. She found that mothers were less likely employed in the first year after childbirth when eligible for the grant. There have also been a few studies assess- ing the effectiveness of tax rebates broadly distributed in the United States as economic stimulus (see Shapiro and Slemrod (2003, 2009); Taylor (2011)). They find that such stimulus packages have small effects on spending, but large effects on savings (i.e., repaying debt). Labour supply impacts are not the focus of these studies.

In Canada, the federal government introduced the Universal Child Care Bene- fit (UCCB) in July 2006, a benefit that provides a family with $100 per month per child under the age of 6. For two-parent families, the benefit is paid as a monthly transfer to the mother. The benefit is not a pure demogrant, however, in that the benefit is taxable: the lower-income spouse in a family is required to claim the UCCB income when filing their federal tax returns. As such, the after-tax benefit amount depends on the tax rate applied to the lower-income spouse, not the income of the family. The benefit is sizeable, representing roughly 12–18% of the annual cost of children.5 The UCCB also represents one of the largest transfer programs administered by the Canadian government, representing 4.5% of federal transfers to individuals.6

In this study, I use Canadian Labour Force Survey (LFS) monthly data from 2003 to 2009 to investigate whether individuals changed their labour market be- haviour in response to receiving the UCCB.7 To identify the effect of the UCCB, a difference-in-differences estimator is used to compare the labour market activity

5 Using estimates from Milligan (2005) and based on Phipps (1998), adjusting for the All-Item CPI (CANSIM Table 326–0021), the cost of a first child in 2006 Canadian dollars is $9,738 per year and the cost of a third child is $6,533.

6 Estimates for 2013–2014 government expenditures are available from the Treasury Board of Canada as part of the Government Expenditure Plan and Main Estimates publication, at tbs-sct.gc.ca/ems-sgd/20132014/me-bpd/me-bpd01-eng.asp.

7 I have used the confidential files of the LFS, available to researchers through the Statistics Canada Research Data Centres Program.

440 T. Schirle

of married Canadians whose youngest child is aged 0 to 5 with married Cana- dians whose youngest child is aged 6 to 17, before and after the introduction of the UCCB. Probit models are used to estimate the effect of the UCCB on labour force participation, and employment and unconditional quantile regressions are used to estimate the effect on hours worked. A complementary analysis of fam- ily expenditures is conducted using information from the Canadian Survey of Household Spending (2004–2008).

The results suggest large and significant negative income effects on labour supply for lower-educated mothers on both intensive and extensive margins. Sig- nificant income effects are found for mothers who have attained education be- yond high school as well. The income effect on hours worked may reflect greater flexibility in hours while mothers enjoy job protection and employment benefits associated with children under 12 months of age. The UCCB’s negative effects on father’s labour supply are significant and align with income effect estimates in the literature.

This paper is organized as follows. In section 2, I describe the UCCB and other important child-related benefits in Canada. In section 3, I discuss important theoretical considerations. In sections 4 and 5, respectively, I discuss the data used in this study and the empirical framework used to estimate the effect of the UCCB on labour market activity. In section 6, I discuss the results and their robustness. I also provide some estimates of the UCCB’s effect on family expenditures. Finally, I offer concluding remarks.

2. Child benefits in Canada

The Universal Child Care Benefit has been distributed to parents of children under the age of 6 since July 2006. The UCCB is a taxable $100 per month per child under the age of 6 payment. The stated goals of the program were to “help cover the cost of children.”8 The discussion surrounding the policy’s introduction, however, suggested the intention was for parents to increase their spending on good and services consumed by their children.9

8 This is the description of the UCCB as of October 2012 at servicecanada.gc.ca/eng/goc/universal child care.shtml. In October 2015, the UCCB was described as providing “a monthly benefit to help all Canadian families with young children choose the child care option that best suits their needs, whether they work in the paid labour force or stay at home with their children,” at esdc.gc.ca/eng/child family/child care/index.shtml.

9 The benefit was first publicly discussed as a part of the Conservative Party of Canada’s election platform for a January 2006 election, while the Liberal Party of Canada had been promoting Federal-Provincial agreements and initiatives to support formal child care programs (discussed in the 2005 Federal Budget documents). A representative of the Liberal Party had publicly stated that parents could spend the UCCB payment on “beer and popcorn” if they so choose. A discussion of that comment is found in the CBC archives at cbc.ca/archives/entry/federal- liberals-deride-beer-and-popcorn-money, last accessed October 5, 2015. The Conservative Party won a minority government in the 2006 election and introduced the UCCB in the 2006 Federal Budget (fin.gc.ca/budtoc/2006/budlist-eng.asp).

Universal benefits and labour supply 441

A recipient of the UCCB must be the primary caregiver of the child under the age of 6 and a resident of Canada.10 For married couples, the benefit is paid to mothers. For tax purposes, the parent with the lowest taxable income claims the UCCB income.11 The UCCB is not clawed back against provincial social assistance payments. Note that in shared custody arrangements, until 2010 only one parent was able to receive the benefit. In 2011, provisions were introduced to allow each parent in a shared custody arrangement to receive half of the benefit.

It would be rare for a family to be eligible for the UCCB and not receive it. When first introduced in 2006, parents were automatically enrolled for the UCCB if they received the Canada Child Tax Benefit (CCTB). If ineligible for the CCTB, parents can apply for the UCCB directly to the Canada Revenue Agency. Upon birth of a child, parents are provided with relevant forms and information by hospital staff, and payments can easily be made retroactively for up to 11 months. While it is possible for a family to not apply and to opt not to receive the benefit, there are very low application costs.12 Payments for a child are automatically terminated the month following the child’s sixth birthday.

Introduced in 1998, the Canada Child Tax Benefit (CCTB) and the National Child Benefit Supplement (NCBS) represent the other major federal programs for child-related benefits. These are income-tested benefits available to families with children under the age of 18. The CCTB eligibility income thresholds are slightly higher and clawback rates much lower than for the NCBS.13 The CCTB and NCBS are not taxable and some provinces have integrated the NCBS payments with their social assistance programs (see Milligan and Stabile (2007)).

Though not typically considered a child benefit, it is important to note the availability of Employment Insurance (EI) benefits for maternity and parental leave in Canada. For children born December 31, 2000, or later, the total ma- ternity and parental paid leave time was increased to one year and the threshold for eligibility was reduced to 600 hours of work in the previous year. In line with this, all jurisdictions in Canada increased job protection provisions to allow for 52 weeks of paid or unpaid maternity or parental leave. Those eligible for EI benefits will receive up to 55% of insured earnings.14 In January 2006, Quebec

10 Canadians with citizenship for fewer than 12 months and new residents of fewer than two years may not be eligible.

11 For single parents, provisions were introduced in 2010 that allow single parents to include the UCCB in the income of an eligible dependent.

12 More recently, an automated benefits application can be completed online, allowing birth registration information to be sent directly to the Canada Revenue Agency for UCCB registration so that application costs are near zero.

13 In 2013, the basic CCTB benefit ($119.41 per month per child under 18) was reduced if “adjusted family net income” was more than $43,561 at a rate of 2% in the case of one child and 4% in the case of more than one child. The NCBS was reduced for income above $25,356 in 2013 at a rate of 12.2% for families with one child, 23% for families with two children and 33.3% for families with three or more children. Some provinces provide an additional supplement, and Alberta varies the basic benefit by age of the child.

14 A maximum benefit is set at 55% of the year’s Maximum Insurable Earnings, amounting to $501 per week in January 2013.

442 T. Schirle

introduced its own Quebec Parental Insurance Plan, which is considered more generous than EI.

A few small child-related benefits were introduced over the time period studied in this paper. A children’s fitness tax credit was introduced (effective January 1, 2007) as a non-refundable tax credit for up to $500 of eligible program fees.15

There is some evidence that these benefits have largely benefited higher-income families and has not changed investments in child fitness (see Spence et al. (2010)). The 2006 budget also introduced the Child Disability Benefit in July 2006. The 2007 budget introduced a new $2,000 non-refundable child tax credit for each child under the age of 18 at the end of the tax year.16

An important consideration is the availability and cost of childcare, which varies substantially across Canada and has changed over time. Quebec introduced full-time kindergarten for all five-year-olds and the provision of child care at a cost at the time of only $5-a-day. The $5-a-day childcare was phased in to cover all children aged 4 and under by 2000. It is well known that the policy had a significant impact on childcare use and maternal labour supply (Baker et al. 2008; Lefebvre and Merrigan 2008). Ontario recently introduced full-time kindergarten for five- year-olds, which was available in some schools in 2010. The initiative was being phased in, aiming to have full-day kindergarten in all schools by 2014. Full-day, half-time kindergarten for four-year-olds was simultaneously being introduced in Ontario schools.

To summarize, there are a variety of child-related benefits in Canada that could affect parents’ decisions regarding labour supply. However, no significant changes were made to child-related benefits over the 2003–2009 period that affect only families with children under the age of 6. The exception is the introduction of the Quebec Parental Insurance Plan; however, robustness checks will demonstrate this program is not driving any results in this paper.

3. Theoretical discussion

A simple standard static model of labour supply treats a demogrant like the UCCB as shifting the budget constraint representing individuals’ potential purchases of leisure (non-market) time and consumption goods. The demogrant has a pure income effect, so that individuals receiving the benefit are expected to purchase more normal goods. As long as leisure time is a normal good, one can expect individuals to reduce their hours of work when receiving the UCCB. In this context, a demogrant also raises an individual’s reservation wage, which could induce a departure from the labour force. To the extent that the tax system is progressive and the lower-earning parent is taxed for the UCCB income, the

15 This provides parents with up to $75 in tax relief. 16 The amount of the credit increases with inflation and provides tax relief up to 15% of $2,000.

See (for 2007) budget.gc.ca/2007/plan/bpa5a-eng.html#child.

Universal benefits and labour supply 443

additional income provided by the UCCB is largest for those individuals with the lowest wage offers and/or lowest hours worked.

In light of the static model’s limitations, however, one might not expect to observe this income effect in the data. In a dynamic model of labour supply, the leisure and consumption goods purchased with an anticipated demogrant— though only received for a short period of time—should be distributed over the individuals’ lifetime. If the marginal value of time related to household produc- tion relative to market production is highest when children are young, however, parents may choose to use that leisure time immediately. Also, parents may use the leisure time immediately if they heavily discount their future.

Further, couples might bargain over the distribution of leisure and consump- tion goods. If they are facing similar marginal tax rates and already sharing consumption equally, couples may choose to evenly split the leisure time and consumption goods purchased with the UCCB. If a secondary earner faces much lower marginal tax rates, the couple may choose to distribute leisure purchases to- wards the primary earner as the marginal utility of leisure for the primary earner might be larger. If, however, the primary earner faces a fixed-hours constraint, then the secondary earner may gain more in leisure time as a result of the UCCB.

A formal model of the UCCB’s effects on labour supply is not pursued here. From the existing frameworks, however, it seems reasonable to expect the total labour supply of married couples to be reduced by the receipt of the UCCB. The reduction in labour supply may not be substantial, as couples might choose to smooth their consumption associated with the anticipated benefit. In the presence of fixed hours-constraints on primary earners, the hours of secondary earners to are expected to change more, particularly among those earners in lower income brackets with a lower opportunity cost of leisure time. Full departures from the labour force in response to UCCB receipt would be more likely in this same group.

4. Data

4.1. The Canadian Labour Force Survey The main data source for this study is the Canadian Labour Force Survey (LFS), 2003–2009. Administered by Statistics Canada, the LFS is a large-scale monthly survey with a complex survey design used to ensure accurate estimates of em- ployment and unemployment in relatively small and sparsely populated regions. Dwellings are sampled and followed for six months of the year. Each month a new incoming rotation group (i.e., set of dwellings) is chosen. The confidential micro- data files are being used, allowing me to observe details of characteristics that are masked in the public-use microdata files. Although basic demographic infor- mation is only collected the first month a person is interviewed, the respondents provide the date of birth for all household members. As such, I am able to see the person’s actual age each month. All respondents are identified in terms of their relationship to the family’s reference person which is “normally an adult with

444 T. Schirle

responsibility for the care or support of the family” (Statistics Canada 2012). Of importance here, I can link children to their parent’s records without mistaking them for grandchildren.

Labour market activity on the intensive and extensive margins will refer to individuals’ activity in the reference week of each month’s survey. For the intensive margin, the actual hours worked at all jobs is used in the analysis. Rather than “usual hours” or hours for the main job, the actual hours at all jobs will capture hours worked in temporary and informal employment.

When creating an indicator for UCCB receipt, I am not able to observe whether an individual or their spouse actually receives the UCCB benefit. Rather, I am indicating an expectation that they are eligible—they have at least one child (by birth, marriage, or adoption) under the age of 6 in the household and are observed after July 2006. I falsely identify eligibility in cases where individuals report their children as members of the household but the parent does not have primary custody.17 I falsely identify ineligibility in cases where individuals have primary custody of a child that is not their own (by birth, marriage, or adoption), such as a grandchild. To check the extent to which an error might occur, I used the Survey of Labour and Income Dynamics, in which I am able to observe the ages of each child in a family and whether the family is collecting the UCCB on an annual basis. An error in assigning the UCCB occurs in only 2.5% of families headed by a married couple, with the bulk of errors representing families deemed eligible for the UCCB, but receiving zero benefits. The error in assigning the UCCB is not more frequent among families headed by divorced or separated individuals, or married individuals who have married more than once—in other words this is not exclusively representing complex child custody arrangements. The error in UCCB assignment in SLID did not appear correlated with indicators of labour market activity (such as labour force status and hours worked).

In the analysis that follows, the distinction is made between “higher” and “lower” educated individuals. In the LFS, the highest level of educational at- tainment is recorded for each individual. I refer to individuals with educational attainment at or below high school graduation as “lower” educated and those with any post-secondary training as “higher” educated. The detailed categories for educational attainment are used as control variables in the analysis.18

The sample used in this study represents married women aged 25 to 49 and men married to women aged 25 to 49. The age range was chosen to represent the group of women most likely to have children under the age of 6, have completed their investments in education and have had the opportunity to start career jobs. In 2006, the average age of mothers at their first childbirth was 28 (Milan 2011).

17 Respondents are asked to list the names of “all persons who usually live here” and add a person if “he/she has no other usual residence elsewhere.” It is not obvious how shared custody arrangements are treated (see Statistics Canada (2012)).

18 Education categories include: (i) grade 8 or lower, (ii) grade 9 and 10, (iii) grade 11 to 13 non-graduate, (iv) grade 11 to 13 graduate, (v) some post-secondary education, (vi) trades certificate or diploma, (vii) community college, CEGEP, etc., (viii) university certificate below bachelor’s, (ix) bachelor’s degree, (x) above bachelor’s degree.

Universal benefits and labour supply 445

The fertility rate of women aged 40 to 44 has increased since the early 1980s and was 7.4 per 1,000 women in 2006; the fertility rate of women aged 25 to 29 has been quite stable and was 100.5 per 1,000 women in 2006 (Milan 2011).

The sample includes common-law marriages when referring to married indi- viduals. Only the reference person or their spouse is sampled to ensure matching of children in a household to their own parents. The sample excludes all indi- viduals in same-sex marriages for practical purposes—although labour market decisions within the household conditional on the presence of children are similar to those for opposite-sex couples, the analysis will be based on subsamples of men and women, making the inclusion of same-sex couples more difficult to interpret. I have chosen not to address single parents within this study as their non-labour income and time constraints differ substantially from married and common-law individuals. The resulting sample captures the vast majority of parents of young children. For example, 84% of parents whose youngest child was aged 0 to 5 were married or living in common-law relationships in the 2003–2009 period; 6% of these parents were single (never married).19

In the analysis that follows, the main sample is restricted to parents whose youngest child (in the household) is aged 0 to 17 unless stated otherwise. The treatment group is defined as parents whose youngest child is aged 0 to 5, as this is the group that benefits from the UCCB. The control group is then comprised of parents whose youngest child is aged 6 to 17. As discussed in later sections, several other control groups are tested to ensure robustness of the results. First, I allow for an expansion of the control group to include parents whose youngest child is aged 6 to 24. Second, I restricted the control group to only include parents whose youngest child is aged 8 to 14. Third, I constructed a control group of “childless” individuals who do not report any of their own children under the age of 24 residing in their household. This essentially captures three groups of individuals—those who have not yet had children but intend to do so, those whose children are grown and moved away and those who are truly “childless” and plan to remain that way. I am not able to distinguish between these three groups. Finally, I construct a broader control group of all men or women who are either “childless” or whose youngest child is aged 6 to 24.

4.2. Descriptive statistics Descriptive statistics for married men and women aged 25 to 49 for the period 2003–2009 are provided in table 1. Labour force participation and employment rates are highest among men and lowest among lower-educated mothers. Note that the participation and employment rates of the broader population (including singles) are quite similar to those presented here.20 Not surprisingly, women are much more likely to be working part-time hours. Reporting 40 hours worked is modal for men, which is a standard work week in Canada (noting that 37.5 hours

19 Based on author’s tabulations from the LFS public use microdata files. 20 In CANSIM Table 282–0004, the participation rate of all women aged 25 to 44 was 82% and all

men aged 25 to 44 was 92%.

446 T. Schirle

TABLE 1 Descriptive statistics

All Low-educated High-educated All mothers mothers mothers fathers

In labour force 0.79 0.71 0.82 0.94 Employed 0.75 0.66 0.78 0.90 Hours

Averagea 21.2 19.6 21.9 36.0 40th percentile 12.0 0.0 15.0 38.0 45th percentile 20.0 12.0 21.0 40.0 50th percentile 24.0 20.0 25.0 40.0 75th percentile 37.5 37.5 37.5 45.5

UCCB 0.22 0.17 0.25 0.22 Under 6 0.44 0.35 0.47 0.44 No. of kids by age

Age 0 0.10 0.06 0.11 0.10 Ages 1–2 0.20 0.15 0.22 0.20 Ages 3–5 0.30 0.26 0.32 0.30 Ages 6–12 0.76 0.80 0.74 0.76 Ages 13–17 0.56 0.70 0.50 0.56

Age (years) 37.90 38.08 37.84 40.56 Age group

Ages 25–29 0.11 0.12 0.10 0.05 Ages 30–34 0.20 0.18 0.21 0.15 Ages 35–39 0.26 0.24 0.26 0.23 Ages 40–44 0.26 0.27 0.26 0.27 Ages 45–49 0.17 0.18 0.16 0.20

Education Grade 8 or less 0.02 0.07 0.02 Grades 9–10 0.04 0.13 0.05 Grades 11–13, non-grad. 0.03 0.10 0.03 Grades 11–13, graduate 0.20 0.71 0.19 Some post-secondary 0.06 0.09 0.06 Trades 0.09 0.12 0.16 CEGEP, community coll. 0.27 0.37 0.20 University below BA 0.03 0.04 0.03 Bachelor’s 0.20 0.28 0.17 Above bachelor’s 0.07 0.10 0.10

Spouse’s age 40.56 41.14 40.33 37.91

N 876,723 257,660 619,063 882,464

NOTE: Sample includes married mothers and fathers, with youngest child aged 0 to 17 and wife aged 25 to 49. a Average hours per week includes observations with zero hours.

is commonly defined in employment contracts for full-time positions, particu- larly in the public sector). While the 75th percentile of hours worked by women reaches 40 hours, median hours for women is much lower. The median hours worked for lower-educated mothers is 20 hours per week. Median hours worked for higher-educated mothers is 25 hours per week. This reflects a higher likeli- hood of mothers to take jobs that allow for part-time and flexible hours. There is, however, a significant gap between the employment rates of highly educated mothers (78%) and lower-educated mothers (66%) that suggests these two groups are quite distinct.

Universal benefits and labour supply 447

Another distinction to be made between higher and lower-educated mothers is the ages of their children. For example, lower-educated mothers have more children aged 13 to 17 than higher-educated mothers; higher-educated mothers have more children aged 0 to 5 (see table 1). The difference likely reflects two things. First, lower-educated mothers may have their first child earlier than higher- educated mothers so that by the time they are in this sample (aged 25 to 49) their children are older. Second, as educational attainment among women has risen over time, earlier birth cohorts of women are more likely to be in the lower- educated group. Being older on average in my sample period, the lower-educated women would have older children even if the timing of fertility were the same. It is not clear from this data whether the total fertility of the two groups is different.

Women are generally more educated than men in Canada. Among women aged 25 to 49, 20% had completed a bachelor’s degree while 17% of their husbands had completed a bachelor’s degree (see table 1). Similarly, fewer women in this sample have not attended some form of post-secondary education. Women, however, are more likely to attend college programs while men are more likely to obtain trades certification.

In figure 1, the labour market activity of women, by subsample and the age of youngest child is summarized, in the months before and after the introduc- tion of the UCCB. As one would expect, in each time period the labour supply among mothers of children aged 0 to 5 is lower on both the extensive and inten- sive margins than other mothers. For women with older children, labour supply generally increases over time. For example, (in the top panel of figure 1) lower- educated mothers whose youngest child was aged 6 to 17 worked 22.2 hours on average before the introduction of the UCCB and 22.8 hours after. Lower- educated mothers with young children aged 0 to 5, however, reduced their average hours worked by 0.7 hours—from 14.6 to 13.9. Similar trends are found for the extensive margin—the employment rates of lower-educated mother with children aged 6 to 17 increased by one percentage point and fell by 3 percentage points for lower-educated mothers with children aged 0 to 5.

The patterns for men (figure 2) are not as distinct as they were for women. The labour supply of fathers is generally lower in the months following the introduc- tion of the UCCB; however, the observed changes over time and the differences between those with younger and older children are very small when compared to the patterns observed for women. The most distinct pattern is for lower-educated men. In the top panel of figure 2, the average hours worked among lower-educated fathers whose child was aged 0 to 5 was 1.1 hours lower after the introduction of the UCCB. For lower-educated fathers with older children, average hours fell by only 0.3 hours. With respect to the extensive margin (in the second panel of figure 2, the participation rate of lower-educated men with children aged 0 to 5 was 1.2 percentage points lower after the introduction of the UCCB. For lower-educated fathers with older children, the participation rate fell by only 0.2 percentage points. Any observed decline in labour supply among higher-educated fathers does not clearly vary with the age of their youngest child.

448 T. Schirle

(a)

(b)

(c)

FIGURE 1 Labour market activity of married women aged 25 to 49, 2003–2009 NOTES: Sample includes married mothers aged 25 to 49 whose youngest child is aged 0 to 17. Standard errors are available in the online technical appendix.

Figure 3 describes the labour market activity of mothers and fathers profiled over the age of their youngest child. In Canada, the availability of EI parental leave clearly affects the labour supply of mothers—when the youngest child is less than one, mothers work just over four hours per week. Mothers with one-year- olds, however, work on average 19 hours per week. Note that mothers employed prior to giving birth are labelled employed (on leave) in the LFS (rather than not

Universal benefits and labour supply 449

(a)

(b)

(c)

FIGURE 2 Labour market activity of married men, 2003–2009 NOTES: Sample includes married fathers whose wife is aged 25 to 49 and whose youngest child is aged 0 to 17. Standard errors are available in the online technical appendix.

in the labour force) as long as their child is under one year and they intend to return to work when their job-protected parental leave is over. As such, there is not a clear increase in employment rates after the youngest child turns 1. More interesting in this figure is the smooth transition towards greater labour market activity for women as their children age. In particular, there is not a clear jump at age 6 when children are typically starting their first grade in elementary school.

450 T. Schirle

(a) Mothers

(b) Fathers

FIGURE 3 Labour market activity of married mothers and fathers, by age of youngest child, 2003–2009

NOTE: Sample includes married mothers aged 25 to 49 and fathers whose wife is aged 25 to 49.

For fathers, there is not a clear pattern for labour market activity across the age of their youngest child.21

Overall, the descriptive statistics suggest that women, particularly mothers, are more likely to engage in employment that is part-time and allows for some flexi- bility in their labour market activity. Men, on the other hand, appear subject to fixed hours constraints and are less influenced by the presence of young children. 21 Note that the slight decline in men’s labour market activity as children age is specific to this

sample, representing entry to retirement by older husbands. The decline is not present in a sample of men aged 25 to 49.

Universal benefits and labour supply 451

Over time, there are clear changes in labour market activity that are different for mothers of young children and those without young children. This is preliminary evidence of an effect of the UCCB and is investigated more rigorously in the next sections.

5. Methods

The main estimating equation takes the form:

Yit = ¯0 + ¯1UCCBit + ¯2Under6it + ¯3PostJuly2006it + Xit° + Zit± + eit . (1) The outcome Yit is either hours, participation in the labour force, or employment, for individual i in month t. The variable Under6it indicates that the individual has a child aged 0 to 5 so that ¯2 capture general labour market effects associated with children being young. PostJuly2006it indicates the individual is observed in a month after the introduction of the UCCB in July 2006. The variable UCCBit is an interaction term indicating the individual has a child aged 5 or younger and is observed after the introduction of the UCCB. The equation then represents the standard difference-in-differences estimator.

The vector Xit represents the baseline set of controls used. In the baseline specification, Xit includes controls for the number of children in each age group (0, 1–2, 3–5, 6–12 and 13–17 years), indicators for the individual’s age (30–34, 35–39, 40–44 or 45–49 years) and a set of indicators for educational attainment. The vector Zit captures time and province effects. The baseline regressions include a trend variable, province fixed effects, province-specific trends and seasonality of labour supply captured by indicators for each month of the year and province- specific month indicators.

The model is estimated using a probit model when Yit represents participa- tion or employment. Unconditional quantile (UQ) regressions (following Firpo et al. 2009) are used to estimate the effect of the UCCB on hours worked at var- ious percentiles including the median.22 Hours are also estimated using a Tobit model, as a fairly large portion of women in this sample have zero hours worked. However, as the notion of latent and potentially negative desired hours worked is an odd construct to some (see for example Angrist and Pischke (2009), p.100), I prefer to estimate the UCCB effect using the UQ regressions. OLS estimates are provided for reference, noting that individuals with zero hours worked remain in the sample when estimating the effect of the UCCB. Given that the use of fixed effects in probit models may bias estimates, I have also estimated participation and employment equations using linear probability models. Note that a variety of specifications of time and province effects (Zit ) were tested when estimating

22 I have made use of the Stata programs generously made available by the authors on Nicole Fortin’s webpage at faculty.arts.ubc.ca/nfortin/datahead.html.

452 T. Schirle

equation (1) for hours, participation and employment, and the estimated effect of the UCCB was robust.23

It is important to acknowledge concerns that the use of conventional standard errors for inference is not appropriate for difference-in-differences estimates, par- ticularly in the case where there is only one group that experiences a treatment. For this reason, I also employ the two-step method developed by Donald and Lang (2007) to estimate the effects of the UCCB on labour supply. Similar to the strategies taken by Rossin-Slater et al. (2013) and Baker and Milligan (2008), in the first step I calculate regression-adjusted differences in outcomes between those with children aged 0 to 5 (the treatment group) and those with children aged 6 to 17 (the control group) in each survey month.24 Specifically, the first-stage estimating equation takes the form:

Yit = Át + Under6i ¼t + Xit° + Zit± + eit . (2) and is estimated with OLS.25 The regression is estimated without a constant so that I can include indicators for each month. The time and province effects (Zit ) include province effects and province-specific trends. The vector ¼̂t contains the estimated regression-adjusted differences between the treatment and control group for each survey month t. For the second stage, I collapse the data into 84 survey month cells and estimate the equation:

¼̂t = ½0 + ½1PostJuly2006t + ut . (3) The coefficient ½1 then represents the effect of the UCCB on labour supply. As the second-step equation is estimated using 84 observations, inference is based on the student’s t-distribution with 82 degrees of freedom.

Presented in the following section, several robustness tests are also conducted. To ensure the results are not driven by parents of young children responding differently to the most recent recession, the equation is estimated using only those observed before September 2008. I also provide results that address con- cerns about employment and wage growth among lower-educated men working in Canada’s resource sector, concerns about the growing importance of EI ma- ternity and parental leave benefits and the introduction of the Quebec Parental Insurance Plan in 2006. I also provide results that address concerns about the purity of the control group in this study, since some individuals were observed with 5-year-olds (in the treatment group) who become 6 year-olds within the 6 months of the LFS.

23 Note the estimated effect of the UCCB on hours is a bit more sensitive—for example the estimated effect on hours falls slightly in all models once provincial fixed effects are accounted for in addition to year effects. Notably, this is true for the OLS and Tobit models as well as the UQ regressions (which will have similar concerns for bias associated with fixed effects in probit models). These estimates are available from the author upon request.

24 The notation used here closely follows Rossin-Slater et al. (2013), who calculated the regression-adjusted differences in outcomes for each survey year of their data.

25 It is not clear whether the Donald and Lang methods can be applied to UQ regressions. Determination of an appropriate method for UQ regressions is beyond the scope of this paper.

Universal benefits and labour supply 453

6. Results

6.1. Main estimates The baseline results for the sample of married mothers aged 25 to 49 are presented in table 2. Here, I report the probit estimates for participation in the labour force and UQ (median) estimates for actual hours worked per week. (A comparable set of results for the sample of married fathers is provided in the technical appendix available from the CJE online archive.) Before turning to the main coefficients of interest in this study, it is interesting to consider the coefficients on other covariates.

As expected, women with the youngest children are the least likely to partic- ipate in the labour force and work the fewest hours. In the first column of table 2, the coefficients indicate that having a child aged 0 is associated with an 11 percentage point lower probability of participating in the labour force. Having a additional child aged 13 to 17, on the other hand, is associated with a 1.8 per- centage point lower probability of participation. A similar child-age-gradient is found for median hours worked though the estimates would suggest having a child aged 0 matters most at the extensive margin for women. The coefficients for fathers (table A1 of online technical appendix) tell a similar story—more older children are associated a higher median hours of work—however the coefficients are quite small relative to those for mothers. The age coefficients suggest an ex- pected concave age-hours profile for men and women. The estimated marginal effect of education on participation or median hours has a steep gradient for women, more so than for men. For example, while having a bachelor’s degree is associated with a 7.7 percentage point higher likelihood of participation than high school graduates among women (table 2), the estimated coefficient for men is only 0.006 (0.6 percentage points; see table A1 of online technical apendix). Moreover, attaining only grade 8 or less is associated with a 22 percentage point lower likelihood of participation than high school graduates among women, but only a 7.6 percentage point lower likelihood of participation among men.

The UCCB has a significant and substantial effect on the labour market activity of mothers on both the extensive and intensive margins (table 2). Receipt of the UCCB reduces married women’s participation by 1.3 percentage points. At the median, actual hours worked is reduced by one hour. This implies a much larger income elasticity than what is typically found in the literature for married women.26

In table 3, I present the estimated effect of the UCCB on labour supply for each of the subsamples (mothers and fathers, by educational attainment) and measure of labour supply (participation, employment or hours worked). For each measure of labour supply, estimates based on the relevant modelling op-

26 Using an hourly wage for women of $20 per hour (CANSIM Table 282–0074, 2006 data), the income effect is (20Å(−1. 005))=24 = −0. 8. McClelland and Mok (2012) suggest the income elasticity is in the range 0 to −0.1.

454 T. Schirle

TABLE 2 Model results, married mothers aged 25 to 49

Participation Hours Participation Hours (probit ME) (UQ median) (probit ME) (UQ median)

UCCB −0.013 −1.005 (0.002) (0.203)

Under 6 −0.034 −4.510 (0.003) (0.244) Education (HS grad. omitted)

Post-July 2006 0.010 1.371 Grade 8 or less −0.220 −12.551 (0.003) (0.222) (0.005) (0.382)

No. of kids Grades 9–10 −0.137 −9.224 Age 0 −0.116 −34.266 (0.004) (0.285)

(0.002) (0.180) Grades 11–13, −0.071 −5.021 non-grad. (0.004) (0.312)

Ages 1–2 −0.095 −6.769 (0.002) (0.165) Some postsecondary 0.012 −1.014

Ages 3–5 −0.072 −4.453 (0.002) (0.227) (0.002) (0.147) Trades 0.056 2.523

Ages 6–12 −0.051 −4.613 (0.002) (0.201) (0.001) (0.069) CEGEP, college 0.090 4.741

Ages 13–17 −0.018 −1.778 (0.001) (0.149) (0.001) (0.084) University below BA 0.075 5.002

(0.003) (0.317) Age (ages 25–29 omitted)

Ages 30–34 0.036 2.406 Bachelor’s degree 0.077 3.629 (0.002) (0.183) (0.002) (0.166)

Ages 35–39 0.040 3.507 Above bachelor’s 0.090 5.625 (0.002) (0.192) (0.002) (0.233)

Ages 40–44 0.025 2.095 Constant 357.4 (0.002) (0.212) (122.7)

Ages 45–49 −0.003 0.090 (0.003) (0.237)

NOTES: Robust standard errors in parentheses. All specifications include controls for trends, province effects, province-specific trends, month effects and province-specific month effects. Probit marginal effects are evaluated at the mean values of covariates. Sample for estimating median hours includes observations with zero hours.

tions are presented (including probit, OLS, Tobit, UQ and the Donald and Lang two-step estimator). The UCCB clearly has large and significant negative effects on the likelihood of participation or employment among mothers, particularly those with lower levels of education, and the estimates are quite robust across modelling choices (see panels A and B in table 3). While the UCCB reduces the likelihood of participation or employment of all mothers by one percentage point, the likelihood of participation among lower-educated mothers is reduced by 3.2 percentage points. Among higher-educated mothers, the effect of the UCCB on participation is smaller, reducing the likelihood of participation by less than one percentage point. Notably, the results (though slightly larger in magnitude) re- main significant when the Donald and Lang methods are used to estimate the UCCB effect in participation or employment. The results also suggest a small, but significant, effect on the participation of fathers—receipt of the UCCB reduces a lower-educated father’s likelihood of participation by more than one percentage point (see panel A in table 3).

Universal benefits and labour supply 455

TABLE 3 Model results – Effect of the UCCB on labour supply

All Low-educ. High-educ. All Low-educ. High-educ. mothers mothers mothers fathers fathers fathers

A. Participation LPM −0.0101 −0.0321 −0.0061 −0.0037 −0.0114 −0.0010

(0.0023) (0.0049) (0.0026) (0.0013) (0.0027) (0.0014) Donald-Lang −0.0101 −0.0325 −0.0060 −0.0036 −0.0111 −0.0011

(0.0025) (0.0054) (0.0025) (0.0013) (0.0027) (0.0014) Probit ME −0.0129 −0.0319 −0.0090 −0.0036 −0.0116 −0.0009

(0.0023) (0.0051) (0.0026) (0.0013) (0.0029) (0.0014)

B. Employment LPM −0.0099 −0.0229 −0.0089 −0.0003 −0.0050 0.0008

(0.0024) (0.0051) (0.0028) (0.0017) (0.0034) (0.0019) Donald-Lang −0.0099 −0.0232 −0.0088 0.00002 −0.0042 0.0009

(0.0028) (0.0055) (0.0029) (0.0016) (0.0035) (0.0018) Probit ME −0.0131 −0.0235 −0.0123 −0.0004 −0.0049 0.0007

(0.0025) (0.0053) (0.0028) (0.0017) (0.0035) (0.0019)

C. Hours/week OLS −0.4096 −0.7483 −0.2882 −0.3601 −0.7534 −0.1955

(0.0998) (0.1964) (0.1165) (0.1068) (0.2069) (0.1253) Donald-Lang −0.3995 −0.7571 −0.2814 −0.3266 −0.6767 −0.1830

(0.2283) (0.2698) (0.2702) (0.1645) (0.2684) (0.1917) Tobit −0.6661 −1.4308 −0.4998 −0.4010 −0.9023 −0.2082

(0.1600) (0.3591) (0.1791) (0.1269) (0.2559) (0.1465) UQ – 45th p. −0.9884 −4.2305 −1.0748 −0.0443 −0.1453 −0.0200

(0.2327) (1.1950) (0.2870) (0.0179) (0.0427) (0.0214) UQ – Median −1.0054 −1.8514 −0.9594 −0.0443 −0.1453 −0.0200

(0.2029) (0.5279) (0.2515) (0.0179) (0.0427) (0.0214) UQ – 75th p. −0.2321 −0.4305 −0.1334 −0.4289 −0.6296 −0.2487

(0.0870) (0.1791) (0.0963) (0.1020) (0.1952) (0.1175)

N 876,723 257,660 619,063 882,464 286,339 596,125

NOTES: Robust standard errors in parentheses. All specifications include covariates listed in table 2 and controls for trends, province effects, province-specific trends, month effects and province-specific month effects. Probit marginal effects are evaluated at the mean values of covariates. Sample for estimating hours includes observations with zero hours.

In panel C of table 3, I present the effect of the UCCB on hours worked. Broadly speaking, the estimates suggest the UCCB has a large and significant negative effect on hours worked. The largest negative effects of the UCCB are found for lower-educated mothers. At the median, hours worked are reduced by 1.9 hours for lower-educated mothers. The effect is not nearly as strong at the 75th percentile of hours, as the UCCB’s effect on the 75th percentile of hours worked is only −0.43. The effects on higher-educated mothers are more modest (the UCCB’s effect on median hours is −0.96). The Donald and Lang estimates suggest that caution regarding inference is warranted; although OLS estimates are significant for higher-educated mothers, the Donald and Lang estimates are not. For men, the UCCB has a small effect on hours worked, which appears to be driven by the response of lower-educated men. For all fathers, the UCCB reduces median hours by only 0.04 (or less than 3 minutes per week). The result

456 T. Schirle

appears to be largely driven by the response of lower-educated fathers: at the median, hours worked among lower-educated fathers is reduced by –0.14. The implied income effect is in line with estimates for men found in the literature.27

Interestingly, there is a slightly larger effect on hours at the 75th percentile for men. This might reflect an opportunity and desire for fathers of young children to pass up overtime hours on occasion.

6.2. Further robustness checks In tables 4 and 5, I explore the robustness of the results for married women on the extensive margin and intensive margin, respectively. Compared to the base- line estimates, the exclusion of all observations after September 2008 does not substantially alter the estimated effect of the UCCB on the participation, employ- ment or hours worked by married women (presented in row 1 of tables 4 and 5). As such, the results do not merely reflect a general reduction in labour supply by mothers in response to the recession.28

Another development in recent years is the large increase job opportunities and men’s hourly wages in Canada’s resource sector.29 To account for this, I pro- vide two sets of estimates (in rows 2 and 3 of tables 4 and 5). First, I include a control variable for the spouse’s wage rate (which is set to zero for spouse’s that are not employed and should thus be thought of as an interaction between employment and wages). This inclusion has no effect not the estimated effect of the UCCB on the intensive or extensive margins. Second, I estimate the equation using a sub-sample of married women whose husbands are not low educated (i.e., have completed more than high school). The estimated effect on participation and employment is slightly smaller in this sub-sample, but a larger negative UCCB effect is found for median hours. This may capture the nature of assortative mat- ing and the role of women as secondary earners in families with highly educated

27 Using an hourly wage for men at $23 per hour (CANSIM 282-0074, 2006), the income effect is (23Å(−0. 0443))=40 = −0.026. McClelland and Mok (2012) suggest the income elasticity is in the range 0 to −0.1.

28 Note that this is robust for the higher-educated mothers sub-sample, the estimated effect on the 75th percentiles of hours among lower-educated mothers and the marginal effects based on OLS or the Tobit model for the sample of lower-educated mothers. However, the estimated effect of the UCCB on the 45th and median hours for lower-educated mothers is insignificant when post-recession observations are excluded. Given this discrepancy is specific to results reflecting those working the fewest hours, it is likely this merely reflects our limited ability to use UQ regressions to estimate effects on the part of the distribution closest to zero hours. I cannot exclude the possibility, however, that during recessions low-educated mothers of young children working few hours are more likely than other mothers to act as added workers in the family by raising their hours worked, but not differentially changing their likelihood of participation in the labour force to the same extent. This result likely relates to the nature of employment held by these mothers and how this changes over the business cycle, which requires further investigation beyond the scope of this study.

29 While nominal wages for all employees aged 25 to 54 increased by 22% from 2003 to 2009, the nominal wages of men aged 25 to 54 in the resource sector increased by 31% (tabulations based on CANSIM Table 282–0072).

Universal benefits and labour supply 457

TABLE 4 Robustness checks – Effect of the UCCB on mothers’ participation and employment

Participation Employment

LPM Probit ME LPM Probit ME N

Baseline −0.0101 −0.0129 −0.0099 −0.0131 876,723 (0.0023) (0.0023) (0.0024) (0.0025)

1. Pre-September 2008 sample −0.0102 −0.0128 −0.0100 −0.0134 713,738 (0.0026) (0.0027) (0.0028) (0.0029)

2. Include spouse’s wage rate −0.0101 −0.0129 −0.0100 −0.0131 876,723 (0.0023) (0.0023) (0.0024) (0.0025)

3. Sample exclude low-educ. −0.0033 −0.0060 −0.0049 −0.0080 593,339 husbands (0.0027) (0.0027) (0.0029) (0.0030)

4. Sample excludes Quebec −0.0067 −0.0086 −0.0085 −0.0109 724,748 (0.0025) (0.0026) (0.0027) (0.0028)

5. Sample excludes new parents −0.0114 −0.0136 −0.0101 −0.0130 800,551 (0.0025) (0.0024) (0.0026) (0.0026)

6. Ensure purity of control group −0.0104 −0.0131 −0.0103 −0.0134 862,070 (0.0023) (0.0024) (0.0025) (0.0025)

7. Control group – Mothers of −0.0114 −0.0142 −0.0114 −0.0147 970,672 kids 6 to 24 (0.0022) (0.0022) (0.0024) (0.0024)

8. Control group – Mothers of −0.0081 −0.0115 −0.0095 −0.0135 677,258 kids 8 to 14 (0.0025) (0.0027) (0.0027) (0.0029)

9. Control group – “Childless” −0.0075 −0.0100 −0.0056 −0.0091 680,904 women (0.0024) (0.0026) (0.0026) (0.0028)

10. Control group – All women −0.0095 −0.0117 −0.0088 −0.0119 1,288,271 “childless” and kids aged 6+ (0.0021) (0.0019) (0.0022) (0.0021)

NOTES: Robust standard errors in parentheses. The baseline specification and sample is that pre- sented in table 2. Probit marginal effects are evaluated at the mean values of covariates. See text for a description of each row.

husbands—they may be more likely to participate in the labour force, but condi- tional on participation have more flexible work schedules. This interesting result warrants further study beyond the scope of this paper.

Tables 4 and 5 also provide estimates that account for parameters of Canada’s maternity and parental leave provisions. First, a subsample that excludes Quebec is used to account for Quebec’s more generous parental leave. The results for the UCCB’s effect on the extensive and intensive margins are slightly smaller (in row 4 of the tables). Second, the effect is estimated using a subsample that excludes the parents of infants (under 12 months old). This appears unimportant for the estimated effect of the UCCB on the extensive margin (row 5 of table 4). However, the estimated effect on hours is smaller than for the broader sample (presented in row 5 of table 5). As such, the effect of the UCCB may in part reflect more flexibility in hours for mothers of infants than for mothers of toddlers and preschoolers.

The sixth row of tables 4 and 5 address concerns about the purity of the control group used in this study. Because the LFS interviews the same household for 6 months, there are some parents that will end up in both the treatment and control group as their youngest child turns 6. To account for this, all children observed

458 T. Schirle

TABLE 5 Robustness checks – Effect of the UCCB on mothers’ hours worked per week

UQ – OLS Tobit UQ – 45th median UQ – 75th

Baseline −0.4096 −0.6661 −0.9884 −1.0054 −0.2321 (0.0998) (0.1600) (0.2327) (0.2029) (0.0870)

1. Pre-September 2008 sample −0.3523 −0.5820 −0.7055 −0.8198 −0.3255 (0.1151) (0.1849) (0.2747) (0.2360) (0.0987)

2. Include spouse’s wage rate −0.4091 −0.6652 −0.9887 −1.0056 −0.2315 (0.0998) (0.1600) (0.2327) (0.2029) (0.0870)

3. Sample exclude low-educ. −0.2934 −0.4513 −0.8895 −1.0416 −0.1052 husbands (0.1192) (0.1889) (0.2828) (0.2468) (0.1031)

4. Sample excludes Quebec −0.2868 −0.4285 −0.9731 −0.5935 −0.1844 (0.1110) (0.1778) (0.3893) (0.2170) (0.1001)

5. Sample excludes new parents −0.2039 −0.3163 −0.5738 −0.6127 −0.1167 (0.1110) (0.1632) (0.2099) (0.2170) (0.0810)

6. Ensure purity of control group −0.4302 −0.6987 −1.0088 −1.0253 −0.2478 (0.1009) (0.1617) (0.2361) (0.2052) (0.0880)

7. Control group – Mothers of −0.4700 −0.7660 −1.1096 −1.1349 −0.2511 kids 6 to 24 (0.0964) (0.1552) (0.2234) (0.2038) (0.0818)

8. Control group – Mothers of −0.4678 −0.7432 −1.6325 −1.5973 −0.3600 kids 8 to 14 (0.1129) (0.1775) (0.3801) (0.3252) (0.1029)

9. Control group – “Childless” −0.1576 −0.3966 −0.5707 −0.4150 −0.0669 women (0.1137) (0.1740) (0.3301) (0.3493) (0.0575)

10. Control group – All women −0.3577 −0.6110 −0.7843 −0.6746 −0.1064 “childless” and kids aged 6+ (0.0898) (0.1453) (0.1869) (0.1614) (0.0361)

NOTES: Robust standard errors in parentheses. The baseline specification and sample is that pre- sented in table 2. Sample for estimating hours includes observations with zero hours. See text for a description of each row.

turning 6 over the 6 month period are removed from the sample. The sample is then reweighed so that distribution for the age of the youngest child in the household is preserved. The results for the estimated effect of the UCCB labour supply are nearly identical to the baseline estimates.

The remaining rows (7–10) of tables 4 and 5 estimate the effect of the UCCB when alternative control groups are used. Broadening the control group to include all women aged 25 to 49, broadening the group to include mothers of older children (up to age 24) or narrowing the sample to those whose youngest child is aged 8 to 14 does not substantively change the results. The estimated effect of the UCCB is slightly smaller and even insignificant when the control group is comprised only of women who do not have their own children in the home; however (as noted earlier), it is not clear how to describe the “childless” group of women.

Broadly speaking, the estimates presented herein suggest: (i) the UCCB has large and significant negative effects on the labour force participation and em- ployment of mothers, particularly lower-educated mothers, (ii) the UCCB has large negative effects on hours worked, particularly lower-educated mothers and (iii) the UCCB has small, but significant effects on the participation and hours worked by lower-educated men.

Universal benefits and labour supply 459

6.3. UCCB effects on expenditures and other decisions The results presented thus far have suggested significant negative income effects on labour supply associated with child-related demogrants. In some subsamples, it appears the UCCB reduces individual labour supply to the extent that there might not be an increase in average families’ expenditures, including purchases of child-related goods and services, in response to UCCB receipt.30 To examine this more closely, I make use of the Canadian Survey of Household Spending (2004–2008). I construct a sample of two-parent families headed by (opposite- sex) married couples where the wife is aged 25 to 49 and the youngest child is aged 17 or under. I then test whether the UCCB has a significant effect on various aspects of families’ spending. I consider several categories of spending, including total consumption, food purchases, clothing and household operations.

I estimate a version of equation (1), where Yit represents annual expenditures (in 2006 Canadian dollars). I include indicators for 2006 or later, whether the youngest child is under the age of 6 and their interaction to capture UCCB receipt. As covariates, Xit includes age of husbands and wives, education, working full time full year, the number of children aged 0 to 4 and 5 and 17, and Zit includes a trend and province fixed effects.

The results are presented in table 6. First, the estimates indicate that the UCCB had no significant effect on total household consumption or expenditures (which includes consumption, payroll taxes, insurance premiums and gifts). Within cat- egories of spending, however, there is some indication that families change their consumption patterns (noting results with single stars are statistically significant at the 15% level).31 For example, in households where the mother is more highly educated, there is an increase in purchases of food from stores associated with the UCCB and a reduction in spending on personal care and household fur- nishings. For households with lower-educated mothers, the UCCB is associated with a slight increase in purchases of children’s clothing (by $53) and a slight reduction in expenditures on children’s camps ($26). In the sample of all fami- lies, there is some evidence of a reduction in expenditures on women’s clothing associated with the UCCB. Overall, the evidence is relatively weak but suggests that families shift their expenditures towards items that could benefit children and the increase in purchases of food from stores could potentially represent a higher level of household production. Unfortunately, there does not exist appro- priate time-use data to properly assess whether the UCCB has led to a change in parents’ household production or general activities with children in a way that clearly benefits children.

30 As a reference point, provincial minimum wages ranged from $6.70 in New Brunswick to $8 in British Columbia in 2006. For lower-educated women, a 1.85 hour per week reduction in labour supply implies a minimum $50 to $60 per month reduction in earnings—more than half of the UCCB benefit amount per month.

31 There are over 200 categories of expenditures in the SHS to consider. Other categories were tested but the results were not statistically significant. Tested categories included tobacco and alcohol, education, recreation, health care and miscellaneous expenditures.

460 T. Schirle

TABLE 6 OLS estimates – Effect of the UCCB on annual family expenditures

All Low-educ. High-educ. families mothers mothers

Total consumption −691.9 −1541.3 −484.1 (1401.0) (2302.4) (1735.1)

Total expendituresa 1226.2 417.9 1359 (2087.7) (3118.3) (2678.4)

Food from stores 126.2 −261.7 306.4* (160.7) (304.5) (189.4)

Food from restaurants 21.6 167.8 −88 (92.1) (161.3) (113.0)

Clothing, women and girls −162.3* −261.8 −114.7 (104.9) (208.1) (116.2)

Clothing, child under 4 26 53.8* 12 (20.7) (33.4) (25.6)

Child care 156.7 251.4 36.7 (132.8) (182.1) (170.4)

Children’s camps 10.4 −26.3* 40.7 (20.0) (17.1) (29.5)

Shelter −42.5 94.5 −261.6 (508.0) (795.2) (638.4)

Household furnishings −191.6** −135.2 −207.7* (107.9) (172.6) (136.2)

Household equipment −91.2 −134.9 −81.1 (71.8) (132.6) (87.0)

Transportation −149.9 −736.7 179.2 (568.0) (886.6) (724.8)

Personal care −82.1* −53.3 −93.9* (50.1) (89.8) (60.9)

N 13,620 4,548 9,072

NOTES: Robust standard errors in parentheses. ** and * denote statistical significance at the 10% and 15% levels, respectively. Sample includes mar- ried couples with children under the age of 17 and the wife is aged 25 to 49, 2004–2008. See text for regression details. a Expenditures includes consumption plus personal taxes, insurance pay- ments and pension contributions.

In this study, it is not possible to ascertain the effect of the UCCB on fertility, which may be of concern if there is selection into becoming a new parent related to labour market attachment. Milligan (2005) shows that Quebec transfer payments to new parents (up to $8,000) had a significant effect on fertility. The research on teen fertility suggests that government transfers will have negligible effects on the choices of teens and that economic opportunities are likely to matter more (Kearney and Levine 2012; Ariizumi and Sen 2013). A recent study of a Spanish income transfer to new parents found significant effects on fertility, in part through a reduction in abortions (González 2013). In Canada, there has been a general upward trend in fertility over the 2000–2009 period (Milan 2011). However, there is no clear break in trend between 2005 and 2007. Furthermore, the increase in fertility appears to be falling after 2007. The effects on fertility warrant further investigation beyond the scope of this study.

Universal benefits and labour supply 461

7. Concluding remarks

Universal child benefits in Canada have had significant and substantial negative effects on the labour supply of women, particularly those with lower education. The main estimates in this study indicate that women with high school gradu- ation or less reduced their likelihood of participating in the labour force by 3.2 percentage points and reduced their likelihood of being employed by 2.4 percent- age points. Median hours worked per week is reduced by 1.85 hours for lower- educated women in response to the introduction of the UCCB. The magnitude of the UCCB’s effect on higher-educated mothers is much smaller: the UCCB reduces their likelihood of participation in the labour force by one percentage point. The evidence suggests that the UCCB’s effect on mothers’ hours worked may in part represent flexibility in the choice of hours while children are under 12 months of age and many mothers enjoy job protection and employment ben- efits. The evidence does not suggest there are large effects on the labour supply of fathers. Estimates for the 75th percentile of hours among fathers does suggest a slight and noteworthy reduction in hours worked.

The estimates represent relatively large negative effects on labour supply for lower-educated mothers in two-parent families, on both the intensive and ex- tensive margins. This is not necessarily representing an effect of demogrants on low-income families, rather an effect on secondary earners with children. Rather than spending the entire value of child-related benefits on goods and services for children, the parents are using that income to purchase more time away from the paid labour market. Unfortunately the available time-use data in Canada does not allow us to measure an effect of the UCCB on the time parents spend in activities that directly benefit their children (such as playing or reading with children) or indirectly benefit their children (for example by making more nu- tritious meals for the family or improving their own mental health). As such, this study is unable to determine whether children are benefitting from the ex- tra time parents spend away from the paid labour market when receiving child benefits.

References

Angrist, J. D., and J.-S. Pischke (2009) Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton: Princeton University Press

Ariizumi, H., and A. Sen (2013) “Teen families, welfare transfers, and the minimum wage: Evidence from Canada,” Canadian Journal of Economics 46(1), 338–60

Baker, M., and K. Milligan (2008) “Maternal employment, breastfeeding, and health: Evidence from maternity leave mandates,” Journal of Health Economics 27(4), 871– 87

Baker, M., J. Gruber, and K. Milligan (2008) “Universal child care, maternal labor supply, and family well-being,” Journal of Political Economy 116(4), 709–45

Blau, F. D., and L. M. Kahn (2007) “Changes in the labor supply behavior of married women: 1980–2000,” Journal of Labor Economics 25, 393–438

462 T. Schirle

Chetty, R., J. N. Friedman, and E. Saez (2012) “Using differences in knowledge across neighborhoods to uncover the impacts of the EITC on earnings,” NBER working paper 18232

Crossley, T. F., and S-H. Jeon (2007) “Joint taxation and the labour supply of married women: Evidence from the Canadian tax reform of 1988,” Fiscal Studies 28(3), 343–65

Donald, S. G., and K. Lang (2007) “Inference with difference-in-differences and other panel data,” The Review of Economics and Statistics 89(2), 221–33

Dooley, M., E. Lipman, and J. Stewart (2005) “Exploring the good mother hypothesis: Do child outcomes vary with the mother’s share of income?,” Canadian Public Policy 31(2), 123–44

Firpo, S., N. M. Fortin, and T. Lemieux (2009) “Unconditional quantile regressions,” Econometrica 77(3), 953–73

González, L. (2013) “The effect of a universal child benefit on conceptions, abortions, and early maternal labor supply,” American Economic Journal: Economic Policy 5(3), 160–88

Heim, B. T. (2007) “The incredible shrinking elasticities: Married female labor supply, 19782002,” Journal of Human Resources 42(4), 881–918

Hotz, J. V., and J. Karl Scholz (2003) “The earned income tax credit.” In Means-Tested Transfer Programs in the United States, pp. 141–97, ed. R. A. Moffitt. Chicago: University of Chicago Press

Immervoll, H., H. J. Kleven, C. T. Kreiner, and E. Saez (2007) “Welfare reform in European countries: A microsimulation analysis,” Economic Journal 117(516), 1–44

Kearney, M. S., and P. B. Levine (2012) “Why is the teen birth rate in the United States so high and why does it matter?,” Journal of Economic Perspectives 26(2), 141–63

Kooreman, P. (2000) “The labeling effect of a child benefit system,” American Economic Review 90(3), 571–83

Lefebvre, P., and P. Merrigan (2008) “Child-care policy and the labor supply of mothers with young children: A natural experiment from Canada,” Journal of Labor Economics 26(3), 519–48

Lemieux, T., and K. Milligan (2008) “Incentive effects of social assistance: A regression discontinuity approach,” Journal of Econometrics 142(2), 807–28

Lundberg, S. J., R. A. Pollak, and T. J. Wales (1997) “Do husbands and wives pool their resources? Evidence from the United Kingdom child benefit,” Journal of Human Resources 32(3), 463–80

McClelland, R., and S. Mok (2012) “A review of recent research on labor supply elasticities,” Congressional Budget Office working paper 2012-12

Meyer, B. D. (2010) “The effects of the earned income tax credit and recent reforms.” In Tax Policy and the Economy, Volume 24, pp. 153–80, ed. J. R. Brown. University of Chicago Press

Milan, A. (2011) “Fertility: Overview, 2008,” technical report no. 91-209-X. Ottawa: Statistics Canada

Milligan, K. (2005) “Subsidizing the stork: New evidence on tax incentives and fertility,” The Review of Economics and Statistics 87(3), 539–55

Milligan, K., and M. Stabile (2007) “The integration of child tax credits and welfare: Evidence from the Canadian National Child Benefit program,” Journal of Public Economics 91(1–2), 305–26

(2009) “Child benefits, maternal employment, and children’s health: Evidence from Canadian child benefit expansions,” American Economic Review 99(2), 128–32

Moffitt, R. A. (2002) “Welfare programs and labor supply.” In Handbook of Public Economics, pp. 2393–430, ed. A. J. Auerbach and M. Feldstein. Elsevier

Universal benefits and labour supply 463

OECD (2011) “The balance of family policy tools – Benefit packages, spending by age and families with young children.” In Doing Better for Families, pp. 55–88. Available at dx.doi.org/10.1787/9789264098732-en. OECD Publishing

(2013) “Benefits and wages: Policies.” Accessed March 4, 2014, at oecd.org/els/benefits-and-wages-policies.htm

Phipps, S. (1995) “Canadian child benefits: Behavioural consequences and income adequacy,” Canadian Public Policy 21(1), 20–30

(1998) “What is the income ‘cost of a child’? Exact equivalence scales for Canadian two-parent families,” The Review of Economics and Statistics 80(1), 157–64

Rossin-Slater, M., C. J. Ruhm, and J. Waldfogel (2013) “The effects of California’s paid family leave program on mothers leave taking and subsequent labor market outcomes,” Journal of Policy Analysis and Management 32(2), 224–45

Shapiro, M. D., and J. Slemrod (2003) “Consumer response to tax rebates,” American Economic Review 93(1), 381–96

(2009) “Did the 2008 tax rebates stimulate spending?,” American Economic Review 99(2), 374–79

Spence, J. C., N. L. Holt, J. K. Dutove, and V. Carson (2010) “Uptake and effectiveness of the children’s Fitness Tax Credit in Canada: The rich get richer,” BMC Public Health 10, 356

Statistics Canada (2012) “Guide to the labour force survey,” technical report no. 71-543-G. Ottawa: Statistics Canada

Taylor, J. B. (2011) “An empirical analysis of the revival of fiscal activism in the 2000s,” Journal of Economic Literature 49(3), 686–702

Woolley, F. (2004) “Why pay child benefits to mothers?,” Canadian Public Policy 30(1), 47–69