Centre de recherche sur l’emploi et les fluctuations économiques (CREFÉ) Center for Research on Economic Fluctuations and Employment (CREFE) Université du Québec à Montréal
Cahier de recherche/Working Paper No. 76
Lone Female Headship and Welfare Policy in Canada Martin Dooley McMaster University Stéphane Gascon Applied Research Branch - Human Resources Development Canada * Pierre Lefebvre CREFE - Université du Québec à Montréal Philip Merrigan CREFE - Université du Québec à Montréal
April 1999
Dooley:
Departement of Economics, McMaster University, Hamilton, Ontario, L8S 4M4, email:
[email protected] Gascon: Applied Research Branch, Human Resources Development Canada, 360 Laurier Street West, Narono Building, 7th Floor, Ottawa, Ontario, K1A 0J9, email:
[email protected] Lefebvre: Departement of Economics, UQAM, C.P. 8888, Succ. Centre-Ville, Montréal, Québec, H3C 3P8, : email:
[email protected] Merrigan: Departement of Economics, UQAM, C.P. 8888, Succ. Centre-Ville, Montréal, Québec, H3C 3P8, : email:
[email protected] * The authors benefitted from the comments of Guy Lacroix, Bernard Fortin, Robert Moffitt and workshops at McMaster University and at HEC. The financial support of the Social Sciences and Humanities Research Council, the Canadian International Labour Network, the FCAR Fund from the ministère de l’Éducation du Québec and the Conseil québécois de la recherche sociale is gratefully acknowledged. The data used in this study are from the public use files of the Survey of Consumer Finances. All computations were performed by the authors, and responsibility for the use and interpretation of these data are solely those of the authors. The views expressed herein are those of the authors and do not reflect the opinions of Human Resources Development Canada or of the Federal government.
Résumé: La principale condition d’éligibilité à l’assistance sociale au Canada s’exprime en termes de besoins financiers plutôt que sur la base d’un critère démographique comme aux États-Unis. Nous utilisons une série de coupes transversales répétées sur les années 1981 à 1993 pour estimer un modèle expliquant le statut de famille monoparentale à chef féminin. Nos résultats ne supportent pas l’hypothèse que les niveaux d’assistance sociale pour les familles biparentales et monoparentales sont des déterminants importants de la probabilité qu’une Canadienne soit chef de famille monoparentale. Dans tous les modèles estimés avec des effects fixes provinciaux, les coefficients des variables de niveaux d’assistance sociale sont faibles, statistiquement non significatifs et souvent du mauvais signe. Nous trouvons cependant que la probabilité qu’une femme soit chef de famille monoparentale dépend, comme on peut s’y attendre, de son potentiel à gagner un revenu et de celui de son partenaire potentiel, de son âge et de son niveau d’éducation.
Abstract: The principal qualifying condition for welfare in Canada, unlike the US, is financial need - there are no demographic criteria. We use a time-series of annual, national cross-sections for the period 1981 through 1993 to estimate a model of lone-female headship. Our findings do not support the hypothesis that welfare benefit levels for one-parent and two-parent families are important determinants of the likelihood that a Canadian woman is a lone mother. In all models with provincial fixed effects, the coefficients for welfare benefits are small, statistically insignificant and often of the unexpected sign. We do find that the probability that a woman is a lone mother is generally associated in the expected fashion with her earnings capacity and the earnings capacity of her potential male partner, and with her age and schooling.
Keywords: Lone-female headship, welfare, fixed effects. JEL classification: I3, J1.
I. Introduction
A considerable U.S. literature has evolved concerning the association between the incidence of lone female headship among families with children and welfare policy, specifically the level of benefits available from AFDC, Medicaid and Food Stamps.
A consensus, however, has yet to emerge. A strong, positive association
between the state level of welfare benefits and the incidence of lone motherhood has been found in single crosssections (Moffitt 1991 and Schultz 1994), but not in a time-series of cross-sections (Moffitt 1994) or in panel data (Hoynes 1997).
There are also varying interpretations of the evidence from the Seattle-Denver Experiments
(Hannan and Tuma 1990 and Cain and Wissoker 1993). Canada provides interesting similarities and contrasts with the U.S. The following phenomena have been true of both countries during the recent past. Lone-mother families have been the group most reliant on welfare income save for the disabled. The proportion of all children who live in lone-mother families and the proportion of poor children who live in lone-mother families have increased. Earnings inequality has grown and the earnings of low skill workers have been especially weak. The cost of welfare is shared by the federal and state/provincial levels of government. Real benefit levels vary considerably by state and province and over time. There are also differences between the two countries. One is that there are no demographic criteria for welfare in Canada; the main qualifying condition is financial need. However, the welfare participation rates of employable adults who are not sole-support parents have been very low historically. A second difference is that real welfare benefit levels are considerably higher in Canada and rose by about 20 per cent during our sample period, 1981-1993. Welfare policy has been high on the Canadian policy agenda, but there have been only four studies of the possible impact on lone-female headship and, as in the U.S., a consensus has yet to emerge. In this paper, we use a time-series of annual, national cross-sections for the period 1981 through 1993 to estimate a model of lone
female headship. Our focus is on two sets of explanatory variables: (1) the level of welfare benefits for lone parents and for couples with children; and (2) the level of market wages for women and men. We blend the strengths of two recent U.S. studies. Like Moffitt (1994), we see if the positive association between welfare benefits and the incidence of lone female headship, which has been found in single cross-sections, persists in a time-series of cross-sections with provincial fixed effects. Like Schultz (1994), we estimate the impact of female wages and the wages of potential male partners for all women in our sample. Section II of the paper contains a brief review of the Canadian welfare system and the relevant literature. Our data and descriptive statistics are discussed in Section III.
In Section IV, we present estimates of a probit
model for the conditional likelihood that a woman is lone head of a family with children less than 18. Section V provides a summary and conclusion.
II. Review of the Canadian Welfare System and Literature
Welfare in Canada is a provincial responsibility, but the federal government assumed 50% of program costs in 1967 in return for the following three conditions: financial need was to be the principal qualification for welfare; provincial residency requirements were forbidden; and an appeals process was required.1
Provinces
retained the right, however, to set benefit levels and these have varied considerably between provinces and over time.2 Canada has no program comparable to Food Stamps and SSI in the U.S. Welfare is the principal source of income for families who lack both earned income and access to social insurance programs for unemployment, disability and old age.3 As we show in Section III, welfare participation rates have always been high for lone 1
This arrangement changed in 1989 when the federal government imposed a maximum of 5% on the annual growth rate of federal welfare transfers to the three highest income provinces which are Ontario, Alberta and British Columbia. We do not think that this policy change would have influenced female headship other than through benefit levels for which we control. Subsequent to our data period, the cost-shared arrangement was changed to a block grant. 2
Provinces also have the freedom to set benefit reduction rates, but these were close to 100% in all provinces during our sample period and remain so. As a result, we did not include them in this study. For more information see Dooley (Forthcoming). 3
Most health care costs are covered by universally available and publicly funded insurance. A federal allowance for families with children is targeted at the lower half of the income distribution, but is not conditional on welfare 4
mothers and the disabled. The same was once true of seniors, but is no longer. The welfare participation rates of non-aged couples and unattached individuals have always been very low. Why would one expect variation in Canadian welfare benefits across time or provinces to be associated with the likelihood of lone female headship? It is true that a Canadian wife, unlike her U.S. counterpart, need not leave her husband in order to qualify for welfare but there are other reasons to expect a welfare effect north of the border. For example, Canadian couples on welfare have a weaker incentive to remain together in those provinces and time periods in which welfare benefits for lone parents are high relative to the welfare benefits for two-parent families. However, we do not think that this is the principal explanation. The major reason, in our opinion, for a possible welfare-headship link in Canada is indicated by the very low welfare participation rates among couples with children (2%) and the very high rates (approximately 40%) among lone mothers (see Table 2). The low participation rate for couples reflects both their greater earnings potential (due to the presence of multiple earners and a male earner) and their superior access to more adequate (compared to welfare) income support programs, most importantly, unemployment insurance (Card and Riddell 1993). The high welfare participation rate among lone mothers reflects, we believe, the fact that for many women in troubled marriages, the principal alternative is not economic self-sufficiency but rather single-parenthood on social assistance. The same can be said of many never-married mothers considering the marriage market, i.e., the relevant alternative to a poor marital prospect is lone-motherhood on welfare. Among such women (married women in troubled unions or never-married women with poor marital prospects), one would expect lone motherhood to be more likely in provinces and time periods in which welfare benefits for a lone mother are high relative to the earnings of an actual or potential spouse. There are only four Canadian economic studies which deal with family headship. Using the 1986 Census public use sample, Allen (1993) found a large and significantly positive impact of the level of provincial welfare benefits on the probability that a woman is a lone mother. Shortcomings of this study include the limited independent variables and the lack of sensitivity tests but a more fundamental problem arises from U.S. research. Moffitt (1991) found results similar to Allen’s when he used a recipiency.
5
single U.S. cross section. Using a time-series of U.S. cross-sections, however, Moffitt (1994) found the link between AFDC benefits and lone female headship to be weak in most specifications and non-exis tent in the presence of a fixed effect for each state. One interpretation of the cross-sectional finding is that states which are more tolerant of lone mothers vote for more adequate welfare benefits and have more lone mothers due to a less stigmatizing atmosphere. Hoynes (1996) finds support for this interpretation using panel data for U.S. white women. In this paper, we apply Moffitt’s test to Allen’s finding using Canadian data. Hum and Choudhry (1992) used data from the three-year Canadian Mincome Experiment, the principal focus of which was market work incentives, to assess the link between marital stability and the generosity of a negative income tax plan. Their estimates were quite imprecise and even their qualitative findings are not readily interpretable. Marital splits were least frequent among couples with the least generous plans and couples with the most generous plans. U.S. experimental data have also yielded mixed results (see Hannan and Tuma 1990 and Cain and Wissoker 1993). Lefebvre and Merrigan have used retrospective, family-history data from Statistics Canada's 1990 General Social Survey in two recent papers. Such data offer several advantages over the time-series of crosssections used in this paper. These include the ability to estimate spell durations with proper accounting for censoring and to estimate transition probabilities between specific marital statuses. The General Social Survey, however, has the disadvantages of both small sample size and, most importantly, the absence of any wage or income data. Lefebvre and Merrigan were forced to use aggregate means of female wages and male income by age and year from a different data source. In the first paper (1997), they estimate a hazard function for the dissolution of first (registered or common-law) marriages and obtain results similar to those which we present below.
The estimated effect of the level of welfare benefits for lone parents is not significant. (The level of
welfare benefits for couples was not included.) As expected, the female wage effect is significantly positive (higher wage, higher likelihood of a split), though only for the younger cohort of women, and male income has a significantly negative coefficient for all cohorts. In the second paper (1998), Lefebvre and Merrigan estimate a hazard function for exiting lone motherhood via (first) marriage or remarriage. This data set includes the level of welfare benefits for both lone mothers and couples. Both variables have significant coefficients of the expected sign in the full sample but not 6
when the sample is disaggregated into never-married and previously-married mothers. (Disaggregation does not have a major impact on our findings reported below.) Furthermore, the hazard functions in their 1998 paper do not include the mother’s wage and the earnings of her (potential) spouse. As shown below, our estimates of the welfare benefit coefficients are very sensitive to the inclusion of these two variables. The Canadian Survey of Labour Income and Dynamics will ultimately provide much better data for the study of the dynamics of marriage and divorce in Canada but this longitudinal survey is still in its very early stages.
III. Data and Descriptive Statistics
Our data are from the Survey of Consumer Finances (SCF) which is the Canadian equivalent of the U.S. March Current Population Survey. The estimates reported in this paper were obtained with the SCF Public Use Files for Individuals in the years 1981, 1982 and 1984 through 1993.4 In Canada, approximately 90% of lone mothers are family heads and 10% are sub-family heads (using the terminology of the U.S. Census Bureau). The estimates in the tables below are based on the sample of all lone mothers with a child under 18.5 Finally, we note that the SCF category of “married” includes both registered and common-law unions. The top panel of Table 1 shows that the proportion of all Canadian women who are lone mothers was 6%7% over our sample period. Moffitt (1994, Table A1) reports comparable headship rates of 6.7% and 8.5% for U.S. white women in 1978 and 1988 respectively. Given the differences in demographic criteria for welfare, one would predict lower lone female headship rates in Canada. However, Canada also has higher welfare benefits for lone mothers than does the U.S. as we show below. The marked stability in the Canadian figures contrasts with
4
There is no public use sample for 1983. There is also an SCF Public Use File for families that begins in 1973, but these files contain individual information only for the head and spouse of the head. Hence, individual data are missing for a substantial proportion of all Canadian women and up to 50% of women age 18 to 24. As a check, we did estimate our model with a sample of women age 25 and over from the family files and comment on these results in Section IV. 5 All related persons who live in the same household are referred to as a family by the U.S. Census Bureau and an economic family by Statistics Canada. A lone mother who does not head her own (economic) family is referred to as a sub-family head by the U.S. Census Bureau and a secondary census family head by Statistics Canada. Welfare is very uncommon among sub-family heads in Canada. As a check, we estimated our multivariate model with a sample confined to lone mothers who are (economic) family heads and comment on these results in Section IV. 7
the popular image of steady growth in the prevalence of lone motherhood. This popular image is better reflected in the middle panel of Table 1 which demonstrates an increase in the proportion of all mothers (of children under 18) who are lone mothers, particularly among those under age 25. The difference between the top two panels is due to the decline in the likelihood that a Canadian woman is a married mother (Dooley forthcoming). The bottom panel of Table 1 shows that the proportion of lone mothers who have never married (registered or common-law) increased, especially among those under age 35. This implies that the number of never-married lone mothers is growing more rapidly than the number of previously-married lone mothers which is also true of the U.S. (Moffitt 1992). Table 2 provides information concerning welfare participation, earned income and welfare benefits. The first panel indicates the proportion of individuals or families who report any welfare income during the year.6 The top row shows that the proportion of lone mothers reporting welfare income increased from 38% to 44% over the sample period. Moffitt (1992) reports AFDC participation rates of 42%-44% for U.S. lone mothers in 1985-1987. The second row shows that only 2% of Canadian couples with children report social assistance income. Welfare use is also uncommon among unmarried, childless women but the third row shows that this participation rate did increase as the labour market conditions for younger workers worsened. Our multivariate analysis focuses on the impact of market wages and welfare benefits on the likelihood of lone motherhood. We include measures of earnings capacity for each woman and for her potential (male) partner. These variables indicate the capacity of a woman for economic self-sufficiency and the capacity of a potential mate to generate economic support for a spouse and children. Becker (1991, p. 335-6) provides a theoretical framework in which women with higher earnings capacity are less prone to marriage and more prone to divorce, and the opposite is true for men with higher earnings capacity. Schultz (1994) and others have provided empirical support for these predictions.
6
Welfare income is known to be under-reported on the SCF and, therefore, the participation rates in Table 2 are likely to be biased downwards. The extent of the bias is not easy to determine because detailed national welfare caseload data are not collected and published in Canada on a regular basis. Dooley (forthcoming) estimates that the true participation rates are about 10-15% (not percentage points) higher than those in Table 2. The available evidence also indicates that the degree of under reporting is fairly stable over time and across family types. Given stable measurement errors, the SCF data can measure accurately the differences over time and across family types in the incidence of social assistance income 8
It is particularly important for women that we measure earnings capacity and not actual earnings because a change in marital or child status is often accompanied by a change in hours and weeks of market work. Hence, the currently observed level of annual or weekly earnings for an individual may be a poor indicator of what her earnings potential would be were she to change marital or child status. We use three indicators of earnings capacity for women: hourly wages; weekly earnings in a full-time job; and annual earnings in a full year (48-52 weeks), full-time job. Each measure has advantages and disadvantages, but they all provide similar probit estimates.7 Below, we report the results for full-time weekly earnings and comment on the infrequent instances in which the two other measures yield different conclusions. The second panel of Table 2 indicates that full-time weekly earnings of women grew by 10% from $357 to $394. The full-time weekly earnings of men grew by only 3% from $551 to $566. As a result, the female/male earnings ratio increased from 65% in 1981-85 to 70% in 1990-93. It also true (results not shown here) that the earnings gap between younger and older workers grew over this period especially among men. Our multivariate analysis required imputed earnings measures. Potential earnings must be imputed for the substantial fraction of females who did no paid work during the survey year. Standard techniques (Heckman 1987) were used for this purpose and the selection corrected regressions are available upon request. We also had to impute the earnings capacity of each woman’s potential partner. This requires data for both spouses and we used the SCF public use file for families. We regressed each married man’s earnings on his wife’s age and schooling, the provincial unemployment rate, and a series of dummies for year, province and urban residence. The resulting coefficients (available upon request) were then used to predict the earnings of a potential partner for each woman in the SCF public use file for individuals. Similar probit estimates were obtained using annual earnings of men in full-year, full-time jobs and just annual male earnings. We report on the former in this paper. 7
In principle, the hourly wage is least influenced by hours of market work, but the only hourly measure available in the SCF must be derived using weeks of work and earnings from the pre-survey year and hours of work from the survey week. This poses a particular problem for women who are more likely than men to change work schedules between years. The disadvantage of the annual earnings measure is that it is limited to women who worked 48-52 weeks in the year prior to the survey and, as such, provides the smallest estimating sample and may be subject to the greatest selection bias. For each earnings measure, we excluded the self-employed and unpaid family workers. As a further check for self-employment and data consistency, we also exclude the small number of observations with one of the following: negative earnings; positive earnings and zero weeks worked; or zero earnings and positive weeks worked.
9
Hourly wages are unavailable in the SCF family file. The second set of independent variables of particular interest in our multivariate analysis are the levels of welfare benefits for lone parents and for couples with children. As discussed in Section II, our expectation is that the likelihood of lone motherhood would be positively associated with the former and negatively associated with the latter. The welfare benefit data come from a variety of sources including the provincial gazettes, FederalProvincial Working Party on Income Maintenance (1975), Banting (1982) and the National Council of Welfare (1987, 1989, 1990, 1991, 1992, 1993). Benefits vary over time, among provinces and by family size within provinces. For our probit model, we must select the (potential) benefit levels for families of a given size because many of the women in our sample are childless. Below, we use the benefits for families with two children. Very similar results were obtained with benefit levels for one-child families. The third panel of Table 2 presents the average (weighted by population) weekly values of welfare benefits. How do these compare with U.S. values? Moffitt (1992, Table 3) reports that monthly AFDC benefits for a family of four averaged US$395 in 1982 dollars between 1981 and 1985. He also reports that the sum of 70% of AFDC plus Food Stamps had an average value of US$511 over the same period. We have excluded U.S. Medicaid benefits because Canada has universal public health insurance. We adjusted the Canadian figures for 1981-1985 in Table 2 so as to make them comparable with the U.S. values.8 This yielded a value of $622 for one parent with three children and $625 for a couple with two children. Hence, U.S. cash transfers in the early 1980's were 64% (=395/622) of Canada welfare benefits and the sum of AFDC plus Food Stamps was equal to 82% (=511/622) of the value of Canadian benefits. Blank and Hanratty (1993) also find that Canada has higher welfare benefits. What happened to Canadian welfare benefits over our sample period? Table 2 indicates that these grew by 17%-18% for each type of family. There was no change in the average benefits for a lone parent relative to those for a couple. Moffitt’s benefit data for the U.S. stop at 1987 but the indication from his figures is that the 8
The steps taken were as follows: conversion to a monthly basis (multiplied by 4.33); conversion to 1982 dollars (multiplied by 0.84); conversion to U.S. dollars (multiplied by 0.75); and, in the case of the lone parents, an additional benefit for a third child (multiplied by 1.10). Additional federal and provincial cash transfers for families with children would add about 15% to the Canadian total package. Canadian welfare recipients may also qualify for special drug and other health benefits, subsidized day care and housing.
10
U.S. trend in welfare benefits was either downwards or at best constant at that point. Hence, it is likely that the gap between Canada and the U.S. grew even further over our sample period. The bottom panel of Table 2 shows that Canadian welfare benefits for lone parents grew relative to female full-time weekly earnings and that benefits for couples grew relative to the male full-time weekly earnings. The increase in these welfare/earnings ratios was especially large for younger lone parents and couples.9 Moffitt (1992, Table 3) indicates that there was a slight decline in this same ratio for the U.S. during the period 1981-1986.
IV. Probit Estimates
We estimated our probit model for the incidence of lone parenthood with two different samples. The “restricted sample” contains women who are age 20-44 with thirteen or fewer years of education. These individuals are disproportionately likely to use welfare. The “unrestricted sample” contains women age 16-59 of all educational levels.10 As Moffitt (1994) indicates, a comparison of estimates with these two data sets provides a specification test. If the estimated coefficients for welfare benefits in the restricted sample are true, then they should be greater in magnitude than those in the unrestricted sample. Table 3 contains estimates of four different probit specifications which use data from the restricted sample.11 We discuss the few differences between these estimates and those obtained with the unrestricted sample
9
Dooley (forthcoming) finds that social assistance use by lone mothers under age 35 grew between 1973 and 1991 and that a substantial proportion of this growth can be accounted for by the increase in the value of welfare benefits relative to potential earnings. In contrast, the welfare participation rate of lone mothers age 35 and over dropped which is consistent with the fact that the potential earnings of this age group grew at the same rate as welfare benefits and family size declined considerably.
10
Moffitt used a sample of high school dropouts age 20-44. The SCF data do not permit us to distinguish clearly between high school graduates and dropouts.
The estimates in Table 3 use unweighted data. We estimated all of our models with both weighted and unweighted data and found that it made little difference to either the coefficient or standard error estimates. 11
11
in the text below. In each case, the dependent variable is equal to one if the woman is the lone head of a family or subfamily with one or more children under 18 and is equal to zero otherwise. The constant corresponds to a woman age 20-24 with 10 or fewer years of schooling, residing in Ontario in 1981. The sample means of the conditioning variables are presented in parentheses after each variable name in column (1). Column (6) illustrates the quantitative magnitude of the coefficients in column (5) and will be explained below. Column (2) contains the estimates of the simplest specification. The welfare benefit coefficients have the expected signs and t-ratios which exceed the standard threshold levels for statistical significance.
These
coefficients imply that a $1,000 increase in the annual benefits for lone mothers (an increase of approximately 8% in 1990-1993) would lead to a one percentage point increase in the proportion of women who were lone mothers from 12% to 13%.
A $1,000 increase in the annual benefits for couples (an increase of approximately 7% in
1990-1993) would lead to a 0.5 percentage point decrease in the proportion of women who are lone mothers from 12% to 11.5%. These translate into modest elasticities of approximately 1.0 and -0.5 respectively. The model in column (3) adds a dummy variable for 11-13 years of schooling, the full-time weekly earnings variables and a dummy variable for each sample year. Both the coefficient and the t-ratio for the lone mother’s welfare benefits decline considerably in absolute value. The other estimates are as expected. The dummy variable for women age 25-34 is now significantly negative. The female and male earnings variables both have the expected signs and large t-ratios. A schooling level of 11-13 years, as opposed to 10 years or less, significantly lowers the likelihood of lone motherhood. There are several interpretations for the negative education coefficient (controlling for wages), e.g., schooling may be positively correlated with knowledge about, and access to, more effective methods of birth control. The year dummies generally indicate an upward trend. Column (4) presents the estimates of a model with a fixed effect for each province (the coefficients of which are available upon request). In this case, the welfare benefit variables take on unexpected signs, but they are not statistically significant. This result is quite similar to Moffitt’s in that the presence of a provincial fixed effect eliminates the expected effect of welfare benefits. The addition of the fixed provincial effects does, however, increase (in absolute value) the coefficients for schooling, female earnings and male earnings. 12 12
Our model is similar to Moffitt’s in most respects. The variables which are in our model but not in his include welfare benefits for couples and female and male earnings variables. The variables which are in his model but not in ours include the unemployment rate and the percent employed in manufacturing, trade, services and 12
Many of the lone mothers in our sample have been in that status for a number years prior to the survey. Their marital status decisions may reflect reactions to welfare policy of earlier years and, therefore, one can make a case for lagging the welfare benefits variables. Column (5) contains the estimates of a model with provincial fixed effects and a 5-year lag in welfare benefits. The estimates in Columns (4) and (5) are quite similar to each other and to those (not shown here) obtained with a 3-year lag in welfare benefits.13 We use the final column to illustrate the quantitative implications of the coefficients in column (5). The first entry in column (6) is the predicted probability for the omitted category which is a woman age 20-24 with 10 or fewer years of schooling, residing in Ontario in 1981. This is also the same value (0.10) as the sample proportion (in the restricted sample). The subsequent entries use the coefficients in column (5) to show the impact respectively of a switch in a dummy variable, a $1,000 increase in annual welfare benefits, and a 10% increase in full-time weekly earnings. For example, a woman with the same characteristics as the constant, save that she is age 25-34, has a predicted probability of 0.04 as opposed to 0.10. Similarly, a woman age 35-44 has a predicted probability of 0.03 other things equal. Changes in welfare benefits have quantitatively very small and statistically non-significant effects. The impact of switching from 10 or less years of schooling to 11-13 years to lower the likelihood of being a lone mother from 0.10 to 0.04. The quantitative impacts of the earnings variables are very large. A ten percent increase in female earnings is predicted to increase the proportion of women who are lone mothers from .10 to .20. A ten percent increase in male earnings is predicted to decrease the proportion of women who are lone mothers from .10 to .07. The time trend, conditional on the values of the other variables, was upwards throughout the 1980's, but this was reversed in the recessionary years of the early 1990's.14 We estimated a series of variations on the basic model government. 13
One can also make a case for lagging the earnings variables but we have not done so. Our earnings data come from the SCF unlike the welfare benefits data. To lag earnings, we would have to shorten the length of what is not an overly long (14 years) time-series.
14
The provincial dummy variables, which were included in the model but not reported in Table 3, generally have tratios below conventional threshold levels. One interesting result is that conditional provincial differences in the incidence of lone motherhood have a wider range than do the unconditional provincial differences. The latter range from .08 in Prince Edward Island to .12 in Nova Scotia, Ontario and Manitoba. The conditional provincial differences range from .07 in Quebec to .17 in New Brunswick and Nova Scotia.
13
and found that the estimates contained in Table 3 are generally quite robust. We found, in particular, that the welfare benefit coefficients are never statistically significant and are often of the unexpected sign in the presence of provincial fixed effects regardless of the sample used. We briefly summarize the other findings below and will provide full details upon request. Our first variation was to use the unrestricted sample, i.e., women age 16-59 from all schooling levels. The second variation was to restrict the sample of lone mothers to heads of families, i.e., to exclude the heads of subfamilies, who rarely use welfare, from the sample. Both of these variations yielded estimates quite similar to those in Table 3. We also used two alternative earnings measures for women: hourly wages and annual earnings among full year, full-time workers. The coefficient estimates were all significantly, positive but the elasticities were 15%-25% smaller than those in Table 3. We tried two alternative earnings measures for men: annual earnings among full year, full-time workers and annual earnings among all positive earners (hourly measures for males were not available). These yielded negative but, in some cases, non-significant coefficients. 15 Several readers of early versions of the paper also suggested that the decision to become a previouslymarried lone mother (to separate or divorce) might be more sensitive to welfare policy than the decision to become a never-married lone mother (to have a child and remain unmarried). We estimated our basic model with a sample limited to ever-married women, that is, we excluded all never-married women, both those with and those without children. The results are very similar to those in Table 3. Another suggestion was to exclude the welfare benefit for two-parent families due to its high correlation (.90) with the welfare benefit for lone-parent families. This exclusion produced few differences from Table 3 as did the substitution of welfare benefits for families with one child in place of the benefits for families with two children. A final test was to try a longer time series by using data from the SCF family files which date back to 1973. These data lack individual information for women who are neither the head nor the spouse of the head of a family which is a problem most pronounced among females under age 25. We estimated our probit model with 15
Other variations with the earnings variables included the following: the use of OLS earnings imputations; dropping the male earnings variable; and dropping both earnings variables (and including the unemployment rate and urban residence). In none of the foregoing instances, did we obtain coefficients for the welfare benefits that were of the expected sign and statistically significant. Furthermore, the coefficient and standard error estimates for female earnings, age, education, year and province were generally quite insensitive to these variations.
14
family data for female heads and spouses of heads age 25-59. This added the years 1973, 1975, 1977 and 1979 to our sample (files were released every two years prior to 1981). The estimates too were similar to those in Table 3 in sign, magnitude and statistical significance.
V. Summary and Conclusion
Our purpose in this paper has been to add to the literature on the determinants of lone female family headship in Canada. Our focus was on the link between the likelihood that a women is a lone mother and two sets of independent variables: the level of welfare benefits available to both lone-parent and two-parent families with young children; and the earnings opportunities for both women and their potential (male) partners. Data came from Survey of Consumer Finances for the years 1981 to 1993. A very simple probit specification yielded coefficient estimates for welfare benefits for both lone mothers and couples which were statistically significant and of the expected sign. In any model with provincial fixed effects, however, the welfare benefit coefficients were invariably small, statistically insignificant and often of the unexpected sign. Hence, Allen’s (1993) finding with a single cross-section of a large, positive association between the level of the welfare benefits for a lone parent and the likelihood that a woman is a lone mother did not persist in a time-series of cross-sections with provincial fixed effects. This echoes Moffitt’s (1994) finding with U.S. data. Various measures of female earnings capacity did yield highly significant, positive coefficients. The predicted earnings of a potential (male) partner invariably yielded a negative coefficient but the t-ratios were not always above conventional threshold levels. The likelihood of lone female headship had a very robust (negative) relationship with the level of a woman’s schooling and age. Our central results held up under a wide variety of samples, measures and model specifications. The incidence of lone motherhood does appear to be sensitive to socioeconomic factors such as wages, education and age. However, our findings do not support the hypothesis that the level of available welfare benefits is an important determinant of the likelihood that a Canadian woman is a lone mother.
We hasten to add, however,
that the Canadian literature on this topic is still at an early stage. The eventual availability of multiple waves of 15
data from the new longitudinal Canadian Survey of Labour and Income Dynamics will greatly improve our ability to assess the socioeconomic determinants of transitions into and out of various marital states.
16
Table 1 Incidence of Lone Motherhood and Maritat Proportion of Women Who Are Lone Mothers a 1981-1985
1986-1989
1990-1993
Age 16-24
.04
.03
.04
Age 25-34
.08
.08
.09
Age 35-44
.10
.08
.10
Age 45-59
.04
.03
.03
Total
.06
.06
.07
Proportion of Mothers b Who Are Lone Mothers Age 16-25
.24
.26
.36
Age 25-34
.12
.13
.15
Age 35-44
.13
.11
.14
Age 45-59
.13
.15
.14
Total
.14
.13
.16
Proportion of Lone Mothers Who Are Never Married
a
Age 16-25
.67
.72
.77
Age 25-34
.22
.34
.41
Age 35-44
.06
.10
.12
Age 45-59
.03
.02
.05
Total
.23
.27
.32
Head of family or sub-family with one or more children under 18. b Head or spouse of head of a family or sub-family with one or more children under 18.
17
Table 2 Levels of Welfare Participation, Full-Time Weekly Earnings and Welfare Benefits 1981-1985
1986-1989
1990-1993
Proportion Reporting Any Social Assistance Income Lone Mothers
.38
.36
.44
Couples With Children
.02
.02
.02
Unmarried, Childless Women
.06
.07
.08
Average Full-Time Weekly Earnings (1986$) Women
357
370
394
Men
551
557
566
Average Weekly Social Assistance Benefits (1986$) One Parent with Two Children
207
223
245
Couple with Two Children
229
250
275
One Parent/Couple
.90
.89
.89
Average Weekly Benefits/Full Time Weekly Earnings One Parent/Female
.58
.60
Couple/Male
.42
.45
18
.62 .49
Table 3
(1)
Probit Estimates for Lone Headship Restricted Sample with Full-Time Weekly Wages a (2) (3) (4)
Variable Names (Sample Means in parentheses) Constant b
Simple
(5)
(6)
Time Fixed Welfare Conditional Effect 5 year lag Probability c Trends -1.64 -7.3 -12.0 -12.6 .10 (27.0) (13.8) (4.2) (4.1) Age 25-34 (.18) .02 -.27 -.47 -.48 .04 (.8) (9.0) (4.7) (4.6) Age 35-44 (.39) -.07 -.44 -.62 -.64 .03 (3.6) (12.1) (3.9) (3.8) .02 -.05 .03 .11 Welfare Benefit for Lone Parent with Two .06 (6.3) (1.7) (1.5) (.8) Children ($11,279) -.02 .03 -.01 .10 Welfare Benefit for Two Parent with Two -.02 (2.5) (2.1) (1.5) (.5) Children ($12,624) Education: Grade 11-13 (.57) -.40 -.51 -.52 .04 (17.7) (5.8) (5.6) Female Ln Weekly Full-Time Earnings (5.67) 2.04 4.4 4.4 .20 (10.4) (14.1) (13.5) Potential Male Partner’s Ln Weekly Full-Time -.83 -2.2 -2.1 .07 Earnings (6.12) (4.9) (3.3) (2.9) 1982 .04 .04 .03 .11 (1.2) (.7) (.5) 1984 .09 .11 .09 .12 (2.7) (2.0) (1.6) 1985 .18 .25 .22 .14 (5.2) (4.7) (4.1) 1986 .17 .21 .19 .14 (4.6) (3.0) (2.5) 1987 .19 .29 .26 .15 (5.6) (4.6) (4.0) .43 .39 .19 1988 .29 (8.2) (7.7) (6.8) .63 .58 .24 1989 .42 (10.2) (11.4) (10.2) 1990 .40 .51 .47 .21 (9.8) (6.9) (6.4) 1991 .23 .16 .12 .12 (4.7) (1.3) (1.0) 1992 .34 .35 .30 .16 (6.9) (3.2) (2.8) 1993 .22 .02 -.03 .10 (3.9) (.2) (.2) a Lone headship of a family or subfamily with children under 18. The sample contains 52,709 women age 20-44 with 13 years of schooling or less. The parentheses contain the t-ratios. b The constant corresponds, depending on the specification, to a woman age 20-24 with 10 or fewer years of schooling, residing in Ontario in 1981. A dummy variable was included for each province but the coefficients are not reported here. See footnote 13 for more details. c The first entry in this column is the sample proportion (0.10). The subsequent entries use the coefficients in column (5) to show the predicted probability that a women is a lone mother given a switch in a dummy variable, a $1,000 increase in annual welfare benefits or a 10% increase in full-time weekly earnings.
19
References Allen, Douglas. 1993. "Welfare and the Family: The Canadian Experience." Journal of Labour Economics 11(1): S201-S223. Becker, Gary. 1991. A Treatise on the Family. Cambridge: Harvard University. Cain, G. and Wissoker, D. 1993. Marital Breakups in the Seattle-Denver Income Maintenance Experiment: A Different Conclusion. Mimeo, U. of Wisconsin-Madison. Blank, R. and Hanratty, M. 1993. "Responding To Need: A Comparison of Social Safety Nets in the United States and Canada." In Small Differences That Matter, edited by David Card and Richard Freeman. Cambridge: NBER. Card, D. and Riddell, C. 1993. “A Comparative Analysis of Unemployment in Canada and the US.” In Small Differences That Matter, edited by D. Card and R. Freeman. Cambridge: NBER. Dooley, Martin D. Forthcoming. "The Evolution of Welfare Participation Among Canadian Lone Mothers.” Canadian Journal of Economics. Hannan, M. and Nancy Tuma. 1990. "A Reassessment of the Effect of Income Maintenance on Marital Dissolution in the Seattle-Denver Experiment." American Journal of Sociology, 95: 1270-1298. Heckman, J. J. 1987. “Selection Bias and Self Selection.” The New Palgrave: A Dictionary of Economics. London: Macmillan Press. Hoynes, Hillary. 1997. “Does Welfare Plan Any Role In Female Headship Decisions?” Journal of Public Economics, 65: 89-117. Hum, D. and Choudhry, S. 1992 "Income, Work and Marital Dissolution: Canadian Experimental Evidence." Journal of Comparative Family Studies, XXIII, 2: 249-265. Lefebvre, Pierre and Philip Merrigan. 1997. “Social Assistance and Conjugal Union Dissolution in Canada: A Dynamic Analysis.” Canadian Journal of Economics, XXX, 1: 112-134. _____________________________. 1998. “Welfare, Conjugal Unions and Single Motherhood in Canada: Estimates from a Hazard Model.” The Journal of Human Resources, 33, 3: 742-757. Moffitt, Robert. 1991. “The Effect of the U.S. Welfare System on Marital Status.” Journal of Public Economics, 41, 1: 101-124 _______________. 1992. "Incentive Effects of the U.S. Welfare System: A Review." Journal of Economic Literature, 30(1): 1-61. _______________. 1994. Welfare Effects on Female Headship With Area Effects. The Journal of Human Resources, 29, 2: 621-636. National Council of Welfare. 1987. Welfare in Canada: The Tangled Safety Net. Ottawa: Minister of Supply and Services. _______________________. Various years. Welfare Incomes 1989 (1990, 1991, 1992, 1993). Ottawa: Minister of Supply and Services.
20
Schultz, T. Paul. 1994. “Marital Status and Fertility in the United States.” The Journal of Human Resources, 29, 2: 639-669.
21
