|
Activity Number:
|
96
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #308596 |
|
Title:
|
A Decision-Theoretic Approach to the Variable Selection Problem
|
|
Author(s):
|
Dhruv Sharma*+ and Sujit Ghosh
|
|
Companies:
|
North Carolina State University and North Carolina State University
|
|
Address:
|
2824 Apt 201 Avent Ferry Road, Raleigh, NC, 27606,
|
|
Keywords:
|
bayesian analysis ; variable selection ; decision theory
|
|
Abstract:
|
Variable selection within a canonical linear regression framework is a fundamental activity in the analysis of datasets. A common strategy for this problem has been to select a model that minimizes a penalized sum of squares criterion by a constraint optimization method. However the optimality of such a procedure has not been formally studied within a formal decision theoretic framework. This article presents a formal solution by obtaining the Bayes estimator corresponding to a loss function suitable for the problem. The solution is shown to be valid for the "large p small n case" case but would require a version of the stochastic search algorithm to compute the optimal estimator. We present simulation studies to compare the performance of our estimator with some of the popular variable selection approaches available. The proposed method is also illustrated using a real dataset.
|
