Semiparametric Analysis of the Socio-Demographic and Spatial Determinants of Undernutrition in Two African Countries

We estimate semiparametric regression models of chronic undernutrition (stunting) using the 1992 Demographic and Health Surveys (DHS) from Tanzania and Zambia. We focus particularly on the influence of the child's age, the mother's body mass index, and spatial influences on chronic undernutrition. Conventional parametric regression models are not flexible enough to cope with possibly nonlinear effects of the continuous covariates and cannot flexibly model spatial influences. We present a Bayesian semiparametric analysis of the effects of these two covariates on chronic undernutrition. Moreover, we investigate spatial determinants of undernutrition in these two countries. Compared to previous work with a simple fixed effects approach for the influence of provinces, we model small scale district specific effects using flexible spatial priors. Inference is fully Bayesian and uses recent Markov chain Monte Carlo techniques.