|
Abstract:
|
Random effects models have been widely used in the analysis of heterogeneous and correlated data that typically arise from nested, multilevel, spatial, temporal, and spatiotemporal designs. This paper explores and tackles two important issues related to random effect specification and misspecification in univariate and multivariate spatial modeling of small area health data: The first issue is the specification of random effects ensemble, and the second issue is the specification of random effects ensemble prior. Using real-life data and via Monte Carlo simulations, this study shows that misspecification of random effects ensemble or misspecification of random effects ensemble prior typically leads to poor posterior inference (i.e., considerable bias and poor coverage) with respect to important model parameters such as fixed effects, random effects, and variance components.
|
