|
Activity Number:
|
362
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Teaching Statistics in the Health Sciences
|
| Abstract - #307769 |
|
Title:
|
Overall AIC Selection Strategy for Linear Mixed-Effects Models
|
|
Author(s):
|
Hua Liang*+ and Guohua Zou
|
|
Companies:
|
University of Rochester and Chinese Academy of Sciences
|
|
Address:
|
Medical Center, Dept. of Biostatistics and Computational Biology, Rochester, NY, 14642,
|
|
Keywords:
|
AIC ; Kullback-Leibler information ; model selection ; profile likelihood ; restricted maximum likelihood
|
|
Abstract:
|
The conventional model selection criterion AIC has been parallelly applied to choose candidate models in mixed-effects models by the consideration of marginal likelihood. Its deficiency was recently noticed by Vaida and Blanchard (2005). Correspondingly, the conditional AIC was suggested. We argue that a more overall measure for the difference between the candidate model and true model should be the joint likelihood of data and random effects. In this paper, we develop two joint likelihood-based model selection criteria for linear mixed-effects models. The criteria are the approximately unbiased estimators of the expected Kullback-Leibler information. Simulation studies show that the proposed method outperforms its counterpart: conditional AIC. Our criteria are also applied to choose the variables in a semi-parametric regression model for a real dataset.
|
