Latent Gaussian random fields for large discrete areal data

Google map

Abstract: 

Discrete areal (spatially aggregated) data are common in many fields like sociology, ecology, and public health. Examples are county-level infant mortality data and crime data at the level of a census tract. Answering research questions can involve learning about spatial dependencies, relationships among different variables while accounting for spatial dependence, spatial clustering, and making predictions. Latent Gaussian random field models are widely used for areal data because of their flexibility. I will introduce these models and explain some of the computational challenges with fitting them to large data sets. I will describe recently developed methods for solving the computational problems, focusing on basis representation approaches. (Joint work with John Hughes [Lehigh University] and Bokgyeong Kang [Dongguk University].)