|
Activity Number:
|
147
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #309664 |
|
Title:
|
Bayesian Model Averaging Using Adaptive Sampling
|
|
Author(s):
|
Joyee Ghosh*+ and Merlise A. Clyde
|
|
Companies:
|
Duke University and Duke University
|
|
Address:
|
Box 90251, Durham, NC, 27708-0251,
|
|
Keywords:
|
Bayesian Model Averaging ; Sampling Without Replacement ; Two Urn Model
|
|
Abstract:
|
Implementing Bayesian Model Averaging (BMA) when the number of explanatory variables, p, is large is a challenging task. Our focus will be on problems where the marginal likelihoods can be calculated analytically. When p is greater than 25 calculating the marginal likelihoods of all models becomes computationally intractable and typically BMA is based on a subset of models. We use an adaptive sampling algorithm due to Clyde and Littman to choose this subset which samples models without replacement. Our motivation for sampling without replacement is that once we sample a model we can calculate its marginal likelihood analytically and would not like to revisit it. We use some ideas from the Bayesian approach to finite population sampling to estimate the posterior probability of the unsampled models, based on this we decide whether we need to continue sampling.
|
