Bayesian Semiparametric Modeling of Determinants of Stunning in Nigeria

The effect of childhood malnutrition varies across geographical location in any given country. Studies on childhood malnutrition without geographic information mask the spatial effect. This study explores the spatial effect and the determinants of stunting in Nigeria.Using the 2013 Nigerian Demographic Health Survey (NDHS) data, we specified a model that simultaneously measures the fixed effect of categorical covariates, nonlinear effect of continuous variable, spatial effect and random effect of the community and households using the diffuse prior, the P-spline with second-order random walk, Markov random field prior and the exchangeable normal priors respectively. The dependent variable was specified as 1 if a child under five years (U5) is stunted and 0 otherwise. The logistic distribution was used to capture the binomial distribution of the dependent variable; the choices of hyperparameters were varied to check for the sensitivity of the priors on the posterior distribution.