|
Activity Number:
|
243
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Stat. Sciences
|
| Abstract - #300522 |
|
Title:
|
Bayesian Multivariate Spatial Models and Their Applications
|
|
Author(s):
|
Joon Jin Song*+ and Bani K. Mallick and Malay Ghosh and Shaw-Pin Miaou
|
|
Companies:
|
Texas A&M University and Texas A&M University and University of Florida and Texas Transportation Institute
|
|
Address:
|
Texas Transportation Institute, College Station, TX, 77843-3135,
|
|
Keywords:
|
multivarate spatial models ; Geographic Information sSystem (GIS) ; Markov Chain Monte Carlo (MCMC) ; highway crash data ; posterior propriety
|
|
Abstract:
|
Bayesian methods have been used for analysis of spatial data in the past decade. Most of them assume that spatial variables are univariate. We consider Bayesian hierarchical model with multivariate spatial model specification. Since correlation among dependent variables is expected, the residual component is assumed to have a multivariate normal distribution with unknown covariance matrix. This is a generalization of the seemingly unrelated regression model. Three types of multivariate spatial model are proposed and a general theorem for each case is provided to ensure posterior propriety under noninformative prior. A Deviance Information Criterion is employed for model choice purpose. We illustrate these methods to Texas Highway Crash data at the county level in 1999 which consists of four types of crashes. Markov chain Monte Carlo is used for computation. ArcGIS maps of posterior means of crash rates for each type crash are provided to visualize spatial pattern of crash rates and allow visual inspection of model performance over Texas.
|
