White, Robert G., and Alberto Palloni
Working paper no. 2009-11
Sample selection bias is a chronic problem in longitudinal studies that is particularly problematic for studies concerning the relationship between health and socio-economic status. This paper adopts two alternate methods for handling sample selection bias attributable to survey attrition and item non-response. Both methods are applied to examine the magnitude of bias in the effects of childhood cognition and behavior on the adult socio-economic gradient in health. A method for sample selection correction with multiple imputation for item non-response is implemented to account for different sources of sample selection bias over time. Estimates of a life course model of health and socioeconomic attainment demonstrate that sample selection bias inflates estimates of socioeconomic gradients. The proposed correction for sample selection bias also suggests that the effects of early child non-cognitive skills rather than cognitive skills may play an important role in the early life origins of adult socioeconomic gradients.