The intersection between child health and education
This report was originally published 21 May 2014.
A presentation and paper by David Figlio, Director, Insitute for Policy Research at Northwestern University.
A large literature documents the effects of neonatal health (commonly proxied by birth weight) on a wide range of adult outcomes such as wages, disability, adult chronic conditions, and human capital accumulation. A series of studies, conducted in a variety of countries, including Canada, Chile, China, Norway, and the United States, have made use of twin comparisons to show that the heavier twin of the pair is more likely to have better adult outcomes measured in various ways.1
While the existing literature makes clear that there appears to be a permanent effect of poor neonatal health on socio-economic and health outcomes, it is important for a variety of policy reasons to know how poor neonatal health affects child development, and whether there are public policies that might act to remediate the negative relationship between early poor health and later-life outcomes. Knowing this relationship can also be useful in helping to understand whether favorable health at birth can shield children against adverse shocks, policy or otherwise. However, we know very little to date about whether the effects of poor neonatal health on cognitive development vary at different ages (say, at kindergarten entry versus third grade versus eight grade), and no existing study identifies whether public policies such as school quality could help to mitigate the effects of poor neonatal health on cognitive development. For that matter, we know very little about whether these effects vary heterogeneously across different demographic or socio-economic groups, and since the existing literature, when it mentions effect heterogeneity at all, rarely presents subgroup-specific findings, it is impossible given the extant literature to know whether early neonatal health and parental inputs are complements or substitutes. As such, while we have strong evidence from twin comparison studies that poor initial health conveys a disadvantage in adulthood, we have little information about the potential roles for policy interventions in ameliorating this disadvantage during childhood.
The reason for these gaps in the literature involves data availability. The datasets that previous researchers have used to study the effects of poor neonatal health on adult outcomes (e.g., Scandinavian registry data, or data matching a mother’s birth certificate to her children’s birth certificates) do not include information on schooling and human capital measures during key developmental years. And even the small number of studies that investigate the effects of birth conditions on test scores rather than adult outcomes (Bharadwaj et al., 2013; Torche and Echevarria, 2011; Rosenzweig and Zhang, 2009) are in developing contexts (e.g., 1990s-era Chile and China) that either lack the diversity necessary to explore heterogeneous effects of poor neonatal health on cognitive development in a manner generalizable to most OECD countries, or a sufficient level of affluence to study these effects in a highly developed context.2,3 And while the Early Childhood Longitudinal Study – Birth Cohort (ECLS-B) of children born in the United States in 2001 oversamples twins, this data set is too recent to investigate outcomes in late elementary school or adolescence, too small to study heterogeneous effects of birth weight, and does not include cognitive outcomes that have high stakes for children.
Another gap in the adult-outcomes literature is that the subjects of that literature are rather old at present; they were necessarily born in the 1970s and earlier. Given the advances in modern neonatology, it is reasonable to believe that poor neonatal health in the 21st century may bear little resemblance to poor neonatal health fifty years ago.4
There have been no studies linking neonatal health to either educational or later outcomes in a highly developed country context using very recent birth cohorts.5
We make use of a major new data source that can fill these gaps in the literature. We match all births in Florida from 1992 through 2002 to subsequent schooling records for those remaining in the state to attend public school. Florida is an excellent place to study these questions because it is large (its population of around 17 million compares to Norway, Denmark, and Sweden combined) and heterogeneous (44.3 percent of mothers are racial or ethnic minorities, and 22.5 percent of mothers were born outside the United States). In addition, Florida is well-known for having some of the strongest education data systems in the United States; Florida, North Carolina, and Texas established the most advanced statewide student longitudinal data systems in the United States during the first half of the 2000s, and Florida has been testing children annually from third through tenth grade for over a decade. For several cohorts, Florida also implemented a universal kindergarten readiness assessment that allows us to explore the effects of birth weight on children’s cognitive outcomes as early as age five. In addition to superb education data quality, Florida offers another major advantage when attempting to match birth and school records: Because children born in Florida are immediately assigned a social security number, and because social security numbers are collected upon school registration, Florida presents the opportunity for particularly effective matches between birth and school records. This allows Florida’s health and education agencies the ability to nearly perfectly match births to school records. As we describe in the next section, our match rate is almost identical to what we would have expected based on American Community Survey data. With these new data, we follow over 1.3 million singletons and over 14,000 pairs of twins from birth through middle school to study the relationship between birth weight and cognitive development.
This paper makes several principal contributions to the literature. First, this paper represents the first analysis of population-level data in a highly developed context to study the effects of neonatal health on cognitive development. Our use of population-level data is important because it permits us the opportunity to estimate heterogeneous effects across a wide variety of demographic and socio-economic dimensions, in order to address both the stability of results across background and the degree to which parental inputs and early health are complements or substitutes.6 Such complementarity could be driven by parents with more resources investing more or less in children with better neonatal health, or could be the result of parents making equal investments but those investments by more educated higher-SES parents being relatively more or less effective at building the human capital of children born with better initial health.
Understanding this complementarity is important because it provides a window into the mechanisms by which neonatal health and parental resources and behavior contribute to human capital development. Whether parental inputs and neonatal health are complements or substitutes also has important implications theoretically (e.g. consider the role complementarities play in the models of human capital accumulation of Cuhna et al. (2006), Cunha and Heckman (2007), Conti and Heckman (2010)), for understanding the distributional effects of investments in infant health, and for guiding the targeting of policies intended to reduce inequalities by improving early life health.
In addition, we use these population-level data to estimate the relationship between birth weight and student outcomes for both twins and singletons. We show that with a richly specified model that holds constant gestation length and restricts attention to the range of birth weights that account for the vast majority of twin births, it is possible to produce what appears to be an unbiased estimate of the effect of birth weight on test scores using the population of singletons. This finding then allows us to show how the results we find for twins, which have strong internal validity, show very similar patterns across subgroups to results estimated from singletons, suggesting the findings have broader external validity than previously known.
Most uniquely, ours is the first study to explore the interaction between schooling factors and the relationship between birth weight and children’s cognitive development. Once children reach school age, they spend considerably more time with adults who are not their parents than they did before school age. Schooling in the most natural place where public policy can play a role in promoting cognitive development amongst children of this age. We seek to understand the degree to which school quality can help to overcome disadvantages associated with poor neonatal health.
We find that the effects of birth weight on cognitive development are roughly constant across a child’s schooling career, and appear to be approximately the same across a wide range of demographic and socio-economic groups. To the extent that there is any systematic relationship we find that parental resources and neonatal health are slightly complementary. In addition, this trajectory is very similar regardless of the quality of the school the children attend. These results suggest that the gaps observed in adulthood associated with poor neonatal health are largely fixed at least by third grade or even kindergarten, indicating that some neonatal health deficits may be very difficult to overcome.
1 Examples of influential previous research include Behrman and Rosenzweig (2004) on schooling and wages, Almond et al. (2005) and Conley, Strully and Bennett (2003) on neonatal outcomes and hospital costs, and Royer (2009) on next generation birth weight, neonatal outcomes and educational attainment, for the United States; Black et al. (2007) on neonatal outcomes, height, IQ, high school completion, employment, earnings and next generation birth weight, for Norway; Oreopoulos et al. (2008) on neonatal outcomes, health outcomes in adolescence, educational attainment and social assistance take up, for Canada; Rosenzweig and Zhang (2012) on educational attainment, wages and weight for height, for China; and Torche and Echevarria (2011) on fourth-grade mathematics test scores, for Chile. In a current working paper, Bharadwaj et al. (2013) study fourth-grade test scores and grades in school, also in Chile.
2 China and Chile have both been rapidly developing over the past two decades, and Chile is today a squarely middle-income country with GDP per capita that is 59 percent of the OECD average when measured in purchasing power parity and 39 percent of the OECD average when denominated by exchange rate. That said, Chile is much wealthier now than in the 1990s when the extant studies’ children were born. This is important context because there have been significant and continuing advances in medical technology over the past several decades that have reduced the lower end of viable birth weights, and possibly changed the life chances of babies of all birth weights. Since access to these technologies is disproportionate to wealthy countries, it is important to examine the effect of poor neonatal health in a setting with widespread access to understand the effects in that context. Indeed, there is reason to believe that fewer Chilean children were viable during their time periods than at a similar time in the United States: Torche and Echevarria (2011) show that the mean birth weight amongst Chilean twins in 1997-1999 was 2500 grams, 107 grams heavier than Florida’s average in 1997-1999.
3 Neither Chilean study can investigate the dynamics of test scores through childhood, as Chile’s staggered
implementation of the SIMCE exam means that they can study fourth grade test scores for some but not all cohorts. Bharadwaj et al. (2013) use teacher-assigned grades, rather than a broadly-comparable outcome measure, as their dependent variable that varies across age.
4 As one example of the temporal differences in neonatology, whereas 50 years ago the threshold for infant viability was around 1500 grams, today the threshold for viability in developed countries is as low as 500 grams or even lower (Lau et al., 2013). As such, it is independently valuable to study the effects of birth weight using a more contemporary set of births than those used in the existing literature.
5 The potential benefits of using more current data from a highly developed country become apparent when we compare the mean birth weight amongst twins in our study of children born after 1992 (2410 grams) to those from previous studies of twins from highly developed countries born in the 1930s through the 1970s (which range from 2517 to 2598 grams, depending on the cohort and country) and those from the late 1990s in Chile (2500 grams).
6 We are certainly not the first paper to conduct heterogeneity analyses. Black, Devereux and Salvanes (2007) mention that they investigated sample splits by income and education and find no significant differences, but do not report their subgroup-specific findings, making it impossible to address the question of whether parental inputs and early health are complements or substitutes. Oreopoulos et al. (2008) report results broken down by birth weight group, gestational length, and APGAR scores, but not by different socio-economic groups. Johnson and Schoeni (2011) report results by parental age and the presence of health insurance, which could reflect a variety of factors other than the key questions that we are interested in studying. Bharadwaj et al.’s (2013) working paper and Torche and Echevarria (2011) split their analyses by maternal education – but the developing Chilean context at the time means that Bharadwaj et al. (2013) only split by high school and over versus middle school or lower education.