|
Activity Number:
|
498
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Thursday, August 2, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #308363 |
|
Title:
|
Bayesian Nonparametric Random Effects Selection Using the Centered Dirichlet Process
|
|
Author(s):
|
Mingan Yang*+ and David B. Dunson
|
|
Companies:
|
National Institute of Environmental Health Sciences and National Institute of Environmental Health Sciences
|
|
Address:
|
NIH, 3950 Jackson St Apt H24, Raleigh, NC, 27607,
|
|
Keywords:
|
nonparametric Bayes ; Dirichlet process ; mixed model ; model uncertainty ; Stochastic Search ; Variable selection
|
|
Abstract:
|
Random effects models are widely used to account for dependence in repeated observations, collected on a subject over time or for different subjects in a center. In conducting inferences and building predictive models, there is typically uncertainty in the subsets of predictors to be included in the fixed and random effect components, as well as in the distribution of the random effects. A challenge in nonparametric modeling of random effects distributions is the need for zero mean constraints to avoid bias. This article proposes an approach for subset selection in nonparametric models. By using a centered Dirichlet process mixture of Gaussians, we choose a prior for the random effects density which has support on the space of absolutely continuous densities with zero mean. An efficient parameter-expanded stochastic search Gibbs sampler is developed, which allows inferences.
|
