|
Activity Number:
|
193
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #301302 |
|
Title:
|
Bayesian Semiparametric Density Regression with Measurement Error
|
|
Author(s):
|
Ju-Hyun Park*+ and David Dunson
|
|
Companies:
|
The University of North Carolina at Chapel Hill and Duke University
|
|
Address:
|
5523 NC Highway 55 #434, Durham, NC, 27713,
|
|
Keywords:
|
Conditional density model ; Density regression ; Dirichlet process ; Kernel stick-breaking process ; Latent variables ; Nonparametric Bayes
|
|
Abstract:
|
In many applications, it is of interest to study the relationship between a predictor X and a response Y without imposing restrictive parametric assumptions on the conditional response distribution of Y given X. The focus of this article is on developing a semiparametric Bayes approach for flexible conditional response distribution modeling, accommodating predictors that are measured with error without imposing parametric assumptions on the distribution of the missing predictor. Our approach relies on a joint modeling strategy, which uses Dirichlet process mixture models for the distribution of X and kernel stick-breaking process mixtures for the conditional distribution of Y given X. Identifiability issues are considered, and an efficient MCMC algorithm is developed for posterior computation. The methods are illustrated using simulation examples and an epidemiologic application.
|
