Skew-Elliptical Spatial Random-Effect Modeling for Areal Data with Application to Mapping Health Utilization Rates

• Activity Number: 29
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Farouk Salim Nathoo*+ and Pulak Ghosh
• Companies: University of Victoria and Indian Institute of Management, Bangalore
• Keywords: areal data ; conditional autoregressive model ; Dirichelet process mixture ; spatial random effects ; skew-elliptical distribution ; disease mapping

Abstract

Mixed regression models incorporating spatially correlated random effects are often used for the analysis of small area health data. In this setting, random effect specifications are typically based on a Gaussian Markov random field; however, in the presence of outliers, or discontinuities, models based on non-Gaussian random effects may be more appropriate. We develop a modeling formulation for generalizing the Gaussian CAR model to the non-Gaussian case, allowing for asymmetric marginal distributions having flexible tail behavior through the parametric skew-t distribution. The skew-t formulation, which can be derived as a parametric scale mixture, is then generalized to a semiparametric model based on Dirichelet process scale mixing. The methodology is illustrated in an analysis of health system utilization, mapping rates of revascularization for cardiac patients in Quebec, Canada.