226 – Model Selection and Prediction for Complex Data
Spatial Generalized Linear Mixed Models in Small-Area Estimation
Mahmoud Torabi
University of Manitoba
In small area estimation, we need to predict characteristics of the subpopulations based on the coarse scale data. Small area predictors are improved by borrowing information from other areas. These are commonly based on either the linear mixed models (LMMs) or the generalized linear mixed models (GLMMs). However, there are many situations that the characteristics are related to their locations. For example, it is an interest of policy makers (and public) to know the spatial pattern of a rare disease (e.g., chronic disease or cancer) to identify the regions with high risk of disease to implement the prevention. In this talk, we propose small area models in the class of spatial GLMMs (SGLMMs) to be able to predict characteristics and also to obtain corresponding mean squared prediction error (MSPE). We also provide second-order unbiased estimators of MSPE of small area predictors using Taylor expansion and parametric bootstrap approaches. In our simulations, we show that our MSPE estimates perform very well in terms of small area predictors as well as their precisions. The performance of our proposed approach is also evaluated through a real application.