Race and gender differences in the cognitive effects of childhood overweight

Abstract

The increase in the prevalence of overweight children (ages 6–13 years) in the United States over the past two decades is likely to result in adverse public health consequences. We use data from the children of the National Longitudinal Survey of Youth 1979 cohort to investigate an additional consequence of childhood overweight – its effect on relative cognitive development. To control for unobserved heterogeneity, we estimate individual (child) fixed effect (FE) models and instrumental variable (IV) models. Although recent research suggests that there is a negligible effect of childhood overweight on cognitive ability, our results demonstrate that the effects are uncovered when examining the relationship separately by race. In particular, we find that overweight white boys have math and reading scores approximately an SD lower than the mean. Overweight white girls have lower math scores whereas overweight black boys and girls have lower reading scores. Our results suggest that in addition to well-documented health consequences, overweight children may also be at risk in terms of experiencing adverse education outcomes, which could lead to lower future wages.

Notes

¹We also estimated sibling FE models, the results of which were qualitatively similar to the individual FE models presented here. The motivation for this approach is that differences between siblings remove variance in weight attributable to a shared family environment. However, Cawley (2004) argues that this is not an appropriate way to remove unobserved heterogeneity citing evidence that shared family environments explain a negligible proportion of the variance in weight across siblings. However, others have noted that an obesogenic family environment is an important predictor of children's changes in BMI (Davison et al., 2005). These estimates are available upon request.

²Following the suggestion by Klepinger et al. (1995), we began with a set of potential instruments that included the district density of fast-food outlets, district mean fast-food prices, and mother's historical BMI and sibling BMI. However, Sargan tests repeatedly rejected the overidentification restrictions. As such, we limited the instruments to mother's historical BMI as this is the strongest of the excluded variables as measured by significance in the first-stage equations.