Efficient Estimation and Prediction for Spatial Generalized Linear Mixed Models

• Activity Number: 176
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Vivekananda Roy*+
• Companies: Iowa State University
• Keywords: Generalized linear mixed models ; geostatistics ; importance sampling ; Markov chain Monte Carlo ; spatial statistics ; spatial prediction

Abstract

Markov chain Monte Carlo (MCMC) methods are often used for parameter estimation and prediction in model-based geostatistics. Due to complexity of the spatial models, often some ad-hoc methods like discretization are used for MCMC sampling from the corresponding posterior distributions. Here we consider a principled approach based on MCMC and efficient importance sampling methods for parameter estimation and model selection in spatial generalized linear mixed models. We introduce a spatial robit model for binomial data and illustrate our methodology in the context of this model. The techniques that we use here can also be applied to other types of geostatistical models. We present results from some simulation studies and a real data application involving prediction of a map of disease severity for a particular plant root disease.