A Shared Neighbor Conditional Autoregressive Model for Small Area Spatial Data

• Activity Number: 266
• Type: Invited
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Andrew B. Lawson*+ and Chawarat Rotejanaprasert and Paula Moraga-Serrano and Jungsoon Choi
• Companies: Medical University of South Carolina and Medical University of South Carolina and Lancaster University and Hanyang University
• Keywords: CAR ; spatial ; shared-neighbor ; GCM ; autocorrelation

Abstract

The modeling of spatial correlation within small area data can take various forms. The use of conditional autoregressive prior distributions for effects is commonplace, especially when dealing with aggregated count data in health studies. CAR models are convenient and relatively easy to implement, but suffer from the fact that they have limited flexibility in modeling correlation. We introduce a new CAR model which can accommodate different neighborhood features (including shared neighbors). Further we examine via simulation how this model performs in comparison to Gaussian Component Mixture (GCM) models and standard CAR models. We also consider the application to a small area health data example.