Computationally Efficient Bayesian Inference for Spatial Generalized Linear Mixed Models

• Activity Number: 91
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 9:30 PM to 10:15 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Saksham Chandra* and Murali Haran
• Companies: Penn State and Penn State
• Keywords: Spatial Generalized Linear Mixed Models ; Composite Likelihood ; Bayesian Inference

Abstract

Spatial Generalized Linear Mixed Models (SGLMMs) are often used to model non-Gaussian spatial data. If such data sets are large, both maximum likelihood and Bayesian approaches for statistical inference and prediction may be computationally slow, if not prohibitive. In this work, we focus on decreasing the computational cost of Bayesian inference for such data. To this end, we investigate an approach based on composite likelihood. We propose to evaluate these modifications for SGLMMs for count data to compare the trade-off between computational cost and estimation/prediction uncertainty.