Methods: We employ a fixed-effects instrumental variable design first used by Dahl and Lochner (2005). The approach controls for permanent unobserved heterogeneity via child-level fixed effects and transitory unobserved characteristics with an instrumental variable. Specifically, we use predicted after-tax/transfer family income as an instrument for current family income taking advantage of variation introduced by changes in the Earned Income Tax Credit (EITC).
We apply this strategy to the data on the children of the National Longitudinal Survey of Youth (NLSY) 1979. The sample includes up to 11 interviews with all children born to women in the original NLSY 79 sample. This sample was representative of U.S. children ages 14-22 in 1979. The current analysis includes over 9,000 children with at least one measurement of height and weight over the age of 2. Over 50% of the children in the sample had 4 or more observations.
Results: Overall, the prevalence of obesity in this sample increased from approximately 9% in 1986 to over 20% in 2006. The instrument for pre-tax/transfer income is constructed by predicting income based on a set of exogenous mother characteristics including age, race, education at age 23, AFQT score, and characteristics of the mother's home life at age 14. The taxes and transfers associated with this income are found using the TAXSIM software program available on-line at the National Bureau for Economic Research. The sum of the two values is equal to the predicted after-tax/transfer income. This variable serves as the instrument. The t-statistic on predicted after-tax/transfer income is highly significant in a regression on actual after-tax/transfer income after accounting for fixed effects.
In the child-fixed effects regressions, there is no relationship between changes in pre-tax/transfer income and childhood obesity. With the instrumental variable, however, an increase in current income is shown to reduce the probability of obesity. Specifically, in linear probability models an increase of $1,000 implies a 1% reduction in the prevalence of obesity.
Conclusions: The preliminary evidence from this analysis shows that family income does have a causal effect on childhood obesity. The observed relationship of decreasing prevalence of obesity with increased family income withstands rigorous controls for time-constant and time-varying unobserved characteristics. The preliminary evidence supports the expectation of positive public health effects from policies such as the EITC.