|
Activity Number:
|
150
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 7, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #307494 |
|
Title:
|
Flexible Bayesian Variable Selection in Multivariate Linear Regression
|
|
Author(s):
|
Nan Lin*+
|
|
Companies:
|
Washington University in St. Louis
|
|
Address:
|
Campus Box 1146, Saint louis, MO, 63130,
|
|
Keywords:
|
Bayesian variable selection ; multivariate regression ; Gibbs sampling ; posterior probability ; mixture model ; latent indicator variable
|
|
Abstract:
|
We extend the stochastic search variable selection method in George and McCulloch (1993) to the linear regression model with multivariate responses. Typically, the predictors are either selected or excluded for all the response variables. Our method allows the selection of predictors different for each response variable by transforming the model representation. In the proposed procedure, we describe multivariate linear regression as a Bayesian hierarchical mixture model, in which latent indicator variables are used to select predictors. Predictors with significant effects can be identified as those with higher posterior probability to be included in the model. A Gibbs sampling scheme is used to generate samples from the posterior distribution. The performance of our method is illustrated using both simulation studies and real data examples.
|
