Family Background, Family Income, Maternal Work and Child Development - October 1998

3.1 Review of Empirical Findings

Most of the American research studies published to date on cognitive outcomes or on social and behavioural development of children in their early and late childhood years (after entrance into school) have used the National Longitudinal Survey of Youth-Child Supplement (NLSY-CS), begun in 1986 and repeated every other year.⁹ The NLSY-CS has been used extensively by psychologists, sociologists, and other behavioural scientists to examine the effects of maternal employment on cognitive skills and on social development of the child. Most of this work (Datcher-Loury, 1988; Baydar and Brooks-Gunn, 1991; Desai et al., 1989; Greenstein's, 1993, 1995; Parcel and Menaghan, 1990, 1994, 1997; Moore and Driscoll, 1997; Brooks-Gunn et al., 1998) has attempted to analyse effects of part-time versus full-time work, influences of the timing of work after the birth of a child as well as whether or not families were poor. It is important to bear in mind that these studies look almost exclusively at families in which mothers voluntarily sought and gained employment, and in some cases used very small sample of children. The studies also differ in the way family income, home time investment and hours worked over the child's lifetime are measured, in the age of the children studied, and in the modelling of the different relationship involved. It can be argued that there are selection problems that cloud the interpretation of these data. In other words, the identification problem of section 3 is not solved. Findings among studies conflict, although they suggest that maternal employment itself is not harmful for young child developmental outcomes.

Two studies can be singled out because first they adopt an economic approach with control variables for family background, parental income and mother's time allocation and they also address the selectivity issue of the mother's participation to the labour market. Second, their work can be replicated with the advantages of using a larger and nationally representative of the population of children. Blau and Grossberg (1992), using a sample of 874 children, from the NLSY-CS, aged 36 to 59 months with employed and non-employed mothers, found that maternal employment during the child's first year of life has negative effects on cognitive skills, but employment in second and later years has positive effects, so that the net effect over the first three to four years is close to zero. They suggest that the indirect effect of the increase in family income when mothers work plays an important part in producing the positive total effect in the second and later years. In the same manner, the impact of time spent in female-headed families is not significant when family income is included in the model of analysis. Hill and O'Neill (1994) also analyse cognitive achievement among young children using a sample of 1,861 children from the NLSY-CS (1986 and 1988), and addressing selectivity issues with respect to the mother's fertility status, paid work status and welfare status. They find a significant negative association between a mother's hours at work and her child's cognitive skills after controlling for family income and the mother's human capital (years of schooling in particular), suggesting mother's work may outweigh the positive effects of higher money income. Finally, their results show that a mother's long-term welfare dependency is detrimental to the acquisition of cognitive skills among young children, and this effect is reinforced if the family lives in an underclass neighbourhood.

In another strand of the research literature the objective is not to measure implications of maternal employment but to investigate the main determinants of young children's outcomes and in particular the independent effect family income might have. Using a large array of cognitive and school assessments for young children from the NLSY-CS, the studies of Korenman and Winship¹⁰ (1995), and Currie and Thomas (1995), present results showing that, after controlling for a variety of family and children observable characteristics, maternal skills (measured by AFQT), education and family income have the most powerful effects on children's outcomes.¹¹ The relative impact of these factors varies across outcomes and age of children assessed.

In the same vein, some recent studies assess the effect of parents' "poverty ratio" on children's outcomes. These reduced-form analyses try to estimate what would happen if families were simply given additional money so their income to needs ratio increased from less than the poverty line to one or more times the poverty lines. These estimates control for some of the main parental characteristics that affect both parental income and children's outcomes, independently of their effect on parental income. Although the controls for family background characteristics such as mother's education, family structure and parent's cognitive skills are not similar and the ways or time horizon used to measure family income and outcomes differ, the results of Korenman et al. (1995), Smith et al. (1997), Chase-Lansdale et al. (1997)¹² and Duncan et al. (1994)¹³ show that income matters, but is clearly small. Pronounced poverty (ratio less than 1) and experience of persistent poverty are detrimental to cognitive development for young children and cognitive achievement at school for older children. Moreover poverty is associated with more behavioural problems.

In a replication study with NLSY-CS data, Mayer (1997) concludes that income per se does not appreciably affect child outcomes, typically the effect of doubling family income from $15,000 to $30,000 raises child's test scores much less than half a standard deviation. Although it takes less money to obtain significant increases for very poor families. Mayer also argues that the "true" effect of income is overstated when unobserved parental characteristics are not controlled. She presents a series of tests that provide some support for the hypothesis that family income may not matter much for child outcomes, once family income covers the basic necessities of life. And, Blau (1997),¹⁴ with a more parsimonious reduced-form model, formulates the same conclusion. In summary, the main findings from these studies are that the effect of "permanent" income is much larger than the effect of current income; income effects are small compared to the effects of some important characteristics of the mother, the child and the family.

Whereas income from work appears to improve children's outcomes, welfare participation appears to reduce young children's test scores on a standardized test of vocabulary (Hill and O'Neil, 1994; Brooks-Gunn et al., 1998). However, since welfare above all serves single-parent families, it will affect mainly children living in such families which poses the difficulty of controlling the differences between types of families. Moreover, the estimates might not correctly disentangle welfare participation and income effects since both are strongly correlated and could depend on the same unmeasured factors that affect children's outcomes.

Previous research on these issues by economists in a Canadian context is to our knowledge non existent or scarce. Dooley and Lipman (1996), using longitudinal health data for children from Ontario, examined the association between family status, as well as maternal work and income, and child psycho-social morbidity. They found that young children of poor lone mothers were at particular risk of psychiatric disorder and poor school performance. Lipman, Offord and Dooley (1996), examining preliminary data from the NLSCY, found that four to 11-year-old children from lone-mother families have one or more behavioural problems. But they note that the majority of children from lone mother families do not have these problems and most children with these problems come from two-parent families.

3.2 Hypotheses

The postulate underlying the hypotheses which are tested is that the mother's time matters to the child's development. First, controlling for several relevant factors, we anticipate a small negative effect on children's outcomes of a mother's increased commitment to the labour market. The main alternative hypothesis is that others factors, positively correlated with work, may cut across this negative effect and produce, on balance, a positive or no effect of mother's work. Because some of these factors are present in the regressions, and others not (unobserved characteristics and self-selected behaviours) the direction of the bias of the estimated work effects is assessed. Second, it is expected that, controlling for several family background factors, the independent effect of family income on children's outcomes will be small. However, welfare receipt (indicating children living in a relatively very low-income family) would have negative effects on their outcomes.

• ⁹The NLSY began as a panel of young men and women aged between 14 and 21 years when first interviewed in 1979 and who have been surveyed yearly. Each year, the women are asked about childbearing. They enter the Child Supplement when they become mothers. Their children have been assessed every other year, from 1986 through 1998. Some analyses include cohorts assessed in 1988 and 1990. We are not aware of published studies using more recently assessed cohorts. It is important to note that the children under study are not, themselves, the results of a probability sampling procedure. They are approximately typical of children who have been born to a nationally representative sample of American women who had only reached ages 21 to 28 in 1986. As a result samples used in American published studies over represent children of relatively younger, less educated and disadvantaged (lower social and occupational status) mothers. Minority children are also over represented. Mothers who have postponed childbearing in order to pursue further schooling or employment instead of parenthood have little chance to be in the samples of 1986, 1988, and 1990. Thus, there is a sample bias. The Canadian counterpart of the NLSY, the NLSCY because it is not tainted by this sample selection bias, is superior on that account. Moreover, the sample size of children in the NLSCY is much larger because of the requirements to produce reliable estimates for all children (0 to 11 years of age) in each of the 10 provinces and at the Canadian level for seven key age cohorts.
• ¹⁰This study also includes an index of behavioural problems and of motor and social development for very young children.
• ¹¹Incidentally, Korenman's study shows that parents' social and economic status (SES) - measured by parental education, occupation and income - is a poor and biased index of family background. The Currie study uses the sub-components scores of maternal AFQT rather than relying only on this summary statistic. In some studies, AFQT is taken as a proxy of IQ because they are highly correlated. Results show that some of the skills measured under AFQT are more highly valued by the labour market and have little relationship with the cognitive achievement of children; some measured skills acquired at school or at work, are not associated with wages but do affect children's outcomes.
• ¹²This study includes neighbourhood characteristics and controls for mother's employment status.
• ¹³This study does not use data from the NLSY-CS. But the study of Smith et al (1997) shows that results are similar to the ones obtained with the NLSY-CS data set.
• ¹⁴This study also uses siblings to control for some of the unobserved parental characteristics.