|
Activity Number:
|
20
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #309577 |
|
Title:
|
Path Sampling To Compute Bayes Factors: An Adaptive Approach
|
|
Author(s):
|
Genevieve Lefebvre*+ and Russell Steele and Alain C. Vandal and Sridar Narayanan and Douglas L. Arnold
|
|
Companies:
|
McGill University and McGill University and McGill University and McConnell Brain Imaging Centre and McConnell Brain Imaging Centre
|
|
Address:
|
805 Sherbrooke W, Montreal, QC, H3A 2K6, Canada
|
|
Keywords:
|
Model Selection ; Monte Carlo Integration ; Adaptive Quadrature ; Mixed-Effects Model ; Importance Density
|
|
Abstract:
|
Performing model selection using Bayes factors (BF) is a challenging task, particularly when the models are large and complex. Path sampling (PS) is recognized as one of the most powerful Monte Carlo integration methods for BF estimation. We examine the impact of two tuning parameters of PS, the specification of the importance density and of the grid, which are shown to be potentially very influential. We then propose the use of an algorithm to automate the selection of the grid in PS, the Grid Selection by Adaptive Quadrature (GSAQ) approach. A bound for the bias of the corresponding PS estimator is also provided. We perform a comparison between GSAQ and standard grid implementation of PS using a well-studied dataset; GSAQ is found to yield superior results. GSAQ is then successfully applied to longitudinal regression models in Multiple Sclerosis research.
|
