Keywords: Causality, Doubly robust, Health status disparities, Inverse probability of treatment weights, Kernel density, Longitudinal, Propensity scores, Neighborhood
The literature on neighborhood research abounds with studies that consider associations between neighborhood characteristics and health outcomes with the goal of establishing that some associations are causal. Such analyses are confounded by selection issues; individuals with above average health outcomes may self-select into advantaged neighborhoods. We consider a previously proposed model that is designed to disentangle neighborhood context from confounding due to selection, and we apply it to longitudinal data from the Health and Retirement Study. We improve robustness by applying inverse-probability weighting and doubly robust estimation to the model; multivariate kernel densities are proposed as a tool for estimating weights in longitudinal data. We examine the potential influence of neighborhood socioeconomic status (NSES) on cognitive function. We see a main effect of NSES on cognitive outcomes when an unadjusted model is used. However, this effect is removed when adjustments to improve robustness are incorporated. Thesefindings yield the conclusion that the observed association is explained by selection effects that can be captured using observed covariates.