|
Activity Number:
|
151
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #303519 |
|
Title:
|
Adaptive Design for Model Selection
|
|
Author(s):
|
Fei Liu*+ and Fan Li and David Dunson
|
|
Companies:
|
University of Missouri-Columbia and Duke University and Duke University
|
|
Address:
|
146 Middlebush Hall, Columbia, MO, 65211,
|
|
Keywords:
|
Adaptive design ; Bayesian statistics ; Decision-theoretic approach ; Linear regression model ; Model Uncertainty
|
|
Abstract:
|
Many scientific problems are concerned with selection of the best model among a list of candidate models. Often, it is the case that one has access to massive data sets of predictors, but collecting data for a new response of interest is expensive. It is thus desirable at each step of the experimentation, to collect a new response whose predictor values are most informative in reducing model uncertainty. We view this as an adaptive design problem. In this paper, we propose a decision-theoretic solution to this problem. We first propose a utility function which appropriately reflects the model uncertainty. Each predictor value are then ranked according to the expected utility and the optimal design point is thus chosen. Additionally, the design will be adaptive as more and more data are included in the analysis. We illustrate our method by simulation studies and real examples.
|
