Data Augmentation Methods for Bayesian Modeling of Spatially Dependent Categorical Data

• Activity Number: 611
• Type: Contributed
• Date/Time: Thursday, August 6, 2009 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Candace Berrett*+ and Catherine Calder
• Companies: The Ohio State University and The Ohio State University
• Keywords: spatial statistics ; Bayesian modeling ; MCMC ; discrete data

Abstract

Spatially-dependent categorical data are arising more frequently with the increased availability of satellite-derived remote sensing imagery. Using a probit link function within a Bayesian framework, data augmentation techniques enable traditional spatial covariance models for continuous data to be used in the analysis of discrete spatial data. Despite the inherent benefits in terms of parameter interpretability provided by the addition of auxiliary variables, parameters capturing the strength of spatial dependence in these models may be only weakly identifiable. In addition, Markov chain Monte Carlo (MCMC) algorithms which make use of these data augmentation schemes can be highly inefficient and can be difficult to tune. We address these computational challenges and propose strategies to make model-fitting more computationally feasible in analyses of categorical spatial data.