Is universal child care leveling the playing field?

Abstract

We assess the case for universal child care programs in the context of a Norwegian reform which led to a large-scale expansion of subsidized child care. We use non-linear difference-in-differences methods to estimate the quantile treatment effects of the reform. We find that the effects of the child care expansion were positive in the lower and middle parts of the earnings distribution of exposed children as adults, and negative in the uppermost part. We complement this analysis with local linear regressions of the child care effects by family income. We find that most of the gains in earnings associated with the universal child care program relate to children of low income parents, whereas upper-class children actually experience a loss in earnings. In line with the differential effects by family income, we estimate that the universal child care program substantially increased intergenerational income mobility. To interpret the estimated heterogeneity in child care effects, we examine the mediating role of educational attainment and cognitive test scores, and show that our estimates are consistent with a simple model where parents make a tradeoff between current family consumption and investment in children. Taken together, our findings could have important implications for the policy debate over universal child care programs, suggesting that the benefits of providing subsidized child care to middle and upper-class children are unlikely to exceed the costs. Our study also points to the importance of universal child care programs in explaining differences in earnings inequality and income mobility across countries and over time.

Introduction

What is the case for universal child care programs? This question is important for both policy and scientific research. In the United States, most child care is provided by the private market, and publicly subsidized child care (or preschool) is predominantly targeted at low income families or single parents. An alternative model is supplied by the universal public programs found in Canada and the Scandinavian countries where all children are eligible for subsidized child care, regardless of family background. Many developed countries currently consider a move towards a universal child care program. ¹ Critics of such a move argue that society should focus its investment on children from low income families, where returns are likely to be greatest; extending subsidized child care to middle and upper-class children may require a considerable increase in taxes, at the cost of economic efficiency. Proponents of universal programs counter that even if returns are greater for the poor, publicly subsidized child care may still have benefits for middle or upper-class children that exceed its costs.

The challenge in assessing the case for universal child care programs is that the direct evidence on the impact of these policies is scarce. Perhaps as a consequence, the case for universal programs is often made drawing on evidence from targeted programs, suggesting that early childhood interventions can generate learning gains in the short-run and improve the long-run prospects of children from poor families.² As emphasized by Baker (2011), using the evidence from targeted programs for the promotion of universal ones is problematic for a number of reasons. First, the widely cited targeted programs offer levels of education and care that are typically not found in programs for all children. Therefore, it is not clear that universal child care can deliver similar benefits to disadvantaged children. Second, even if universal programs could offer similar levels of care and education, it may be that the benefits for middle or upper-class children do not exceed the costs.

In this paper, we examine the case for universal child care programs. Our context is a reform from late 1975 in Norway, which led to a large-scale expansion of subsidized child care. All children 3–6 years old were eligible regardless of their parents' employment and marital status, and available child care slots were in general allocated according to length of time on the waiting list. The reform we study led to a staged expansion of subsidized child care, across Norway's more than 400 municipalities. To control for unobserved differences between children born in different years as well as between children from treatment and comparison municipalities, we use a difference-in-differences (DiD) identification that exploits the temporal and spatial variation in child care coverage induced by this staged expansion. Roughly speaking, we compare the adult outcomes for 3 to 6 year olds before and after the reform, from municipalities where child care expanded a lot and municipalities with little or no increase in child care coverage. As described in detail below, subsidized child care both before the reform and during the expansion was severely rationed, with informal care arrangements (such as friends, relatives, and unlicensed care givers) servicing the excess demand. In our analysis, we will focus on the years immediately after the reform, when child care coverage increased from 10% to 28%, which likely reflects the abrupt slackening of constraints on the supply side caused by the reform, rather than a spike in local demand. We have found no other reforms or changes taking place in this period which could confound our estimated child care effects. Nevertheless, to increase our confidence in the empirical strategy, we perform a number of specification checks.

We begin by using non-linear DiD methods to estimate how the child care reform affected the earnings distribution of exposed children as adults. The estimated quantile treatment effects (QTEs) allow us to assess the impact of the universal child care program on the lower, middle and upper parts of the earnings distribution. We complement the QTE estimates with local linear regressions of the reform effects by family income. The local linear regression estimates quantify the effects of the universal child care program on the outcomes of middle and upper-class children as compared to children of lower-class parents. Lastly, we examine how the child care reform affected intergenerational income mobility. These results inform us about whether the universal child care program improved social mobility by attenuating the link between family background and children's outcomes.

We find that mean impacts in the population of children as a whole miss a lot, concealing major heterogeneity in the effects of the child care reform. This is an important observation since previous empirical studies of universal child care have focused on mean impacts. While the reform had a small and insignificant mean impact on earnings, QTE-estimates are positive in the lower and middle parts of the earnings distribution and negative in the uppermost part. The local linear regression estimates show that most of the gains in earnings associated with the universal child care program relate to children of low income parents, whereas upper-class children actually experience a loss in earnings because of the program. In line with the differential effects by family income, we find that the child care reform substantially reduced the intergenerational income elasticity. We show that the estimated heterogeneity in child care effects is consistent with a simple model where parents make a tradeoff between family consumption today and investment in children.

To clarify the nature of the relationship between the child care reform and the earnings of exposed children as adults, we examine the mediating role of educational attainment and cognitive test scores. We find a positive mean impact of the universal child care program on educational attainment. Most of the growth in educational attainment stems from a substantial increase in the likelihood of completing high school. The increase in educational attainment is driven largely by children of low income parents, while there is no change in schooling levels of upper-class children. We find no reform effect on children's scores on cognitive tests. All our QTE estimates on test scores are close to zero, and they are sufficiently precise to rule out any economically relevant effect. The absence of a change in cognitive test scores points to the importance of non-cognitive skill development to understand the long-run effects of universal child care.³

Our study is related to a small but growing literature on universal or large-scale child care programs.⁴ Researchers have examined the effect of these programs on several measures of child outcomes. While these papers make important contributions to the evidence base on universal child care policy, there is no consensus across studies (for a review, see Baker, 2011). Consider the research most closely related to our study. Baker et al. (2008) and Lefebvre et al. (2008) analyze the introduction of subsidized, universally accessible child care in Quebec. The fact that pre-reform subsidies were targeted at low-income families implies that the policy change primarily affected middle and high income families. Baker et al. (2008) find negative mean impacts on indices of behavior, while Lefebvre et al. (2008) report negative mean impacts on cognitive test scores. Datta Gupta and Simonsen (2010) exploit variation across municipalities in Denmark in access to center based care. They find that, compared to home care, being enrolled in center-based care at age three does not lead to significant differences in child outcomes at age seven. Havnes and Mogstad (2011b) examine the child care reform from late 1975 in Norway, reporting positive mean impacts on educational attainment and labor force participation.⁵

Our paper expands and clarifies this prior literature in several important ways. First, our findings suggest that the effects of subsidized child care vary systematically across the outcome distribution, and that children of low income parents seem to be the primary beneficiaries. These findings could have important implications for the case for universal child care programs, since the benefits of providing subsidized child care to middle and upper-class children are unlikely to exceed the costs. Second, we are able to look at a broad range of child outcomes over time. This allows us to examine how the child care reform may have affected the labor outcomes of exposed children as adults, by considering impacts on educational attainment and cognitive test scores. An advantage of our measures of long-run outcomes is that we get round the issues of whether short-run impacts of child care persist, and perhaps are amplified, over time.⁶ Third, we are able to hone in on the counterfactual to subsidized child care because there is no evidence of crowd out of maternal care and there is little change in family income. This helps us interpret the expansions to subsidized child care as increases in the amount of time spent in the child care center versus informal care (including relatives and nannies).

Our study also points to potential explanations for the mixed evidence in previous studies of universal child care programs. Our theoretical framework predicts that the introduction of subsidized child care equalizes the outcome distribution of children, while there are no unambiguous predictions about the mean impact. If child care quality is a normal good, we expect a negative correlation between the effects of universal child care policy and family income. This could, for example, help in explaining the negative mean impacts of the child care reform in Quebec, which primarily affected middle and high income families. Differential effects both across different studies and across the outcome distribution in a given study may operate through differences in child care take-up and differences in the impacts of uptake (depending on the quality of the child care center and the counterfactual form of care). Distinguishing empirically between these two sources to heterogeneity would be interesting, but requires data we do not have and an assumption about the policy not shifting the ranks of children in their potential outcome distribution (see Heckman et al., 1997).

Another contribution of our paper is to highlight the importance of universal child care programs in explaining differences in earnings inequality and income mobility across countries and over time. It is widely argued that the early years are a particularly important time for efforts to equalize opportunities, because a good deal of inequality is already apparent by the time children start school, and because children's development may be less amenable to change after they enter school (see e.g. Almond and Currie, 2010, Heckman and Moses, 2014). But it is less clear how much policies can reduce inequality or improve social mobility, or what policies might be effective (see e.g. Black and Devereux, 2011). We show that subsidized child care could level the playing field by reducing inequality in the earnings distribution of exposed children as adults and by attenuating the persistence of income across generations.

The size and detailed nature of our data also allow us to bring new evidence on the usefulness of non-linear DiD methods in assessing policy changes, when only observational data is available and theory predicts heterogeneous treatment effects. To date, non-linear DiD methods have been rarely used in policy evaluation.⁷ To evaluate the sensitivity of the QTE estimates to the assumptions about the counterfactual outcome distribution, we implement three non-linear DiD methods that have been recently proposed but rarely used.⁸ Although the identifying assumptions differ, they all show that the consequences of a policy change could be much more extensive than what mean impacts suggest.

The paper proceeds as follows. Section 2 reviews key facts regarding the child care system in Norway, describes the child care reform and discusses its expected impact on child development. Section 3 describes the empirical strategy and Section 4 presents the data. Section 5 presents our empirical findings and provides sensitivity analyses. Section 6 concludes.