|
Activity Number:
|
198
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, August 3, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
SSC
|
| Abstract - #302818 |
|
Title:
|
Composite Likelihood Information Criterion for Model Selection in High-Dimensional Data
|
|
Author(s):
|
Xin Gao*+ and Peter X.K. Song
|
|
Companies:
|
York University and University of Michigan
|
|
Address:
|
Dept of Math and Stat, N520 Ross Building, Toronto, ON, M3J 1P3, Canada
|
|
Keywords:
|
Information criterion ; Model selection ; Composite likelihood ; Consistency ; Pseudo-likelihood ; Mis-specified model
|
|
Abstract:
|
For high-dimensional data set with complicated dependency structure, the full likelihood approach often renders to intractable computational complexity. This imposes difficulty on model selection problem as most of the traditionally used information criteria require the evaluation of the full likelihood. We propose a composite likelihood version of the extended Bayesian information criterion and establish its consistency property to select the true unknown model. Under mild regularity conditions, such information criterion based on pseudo-likelihood is consistent allowing the number of potential parameters in the model increases to infinity with the sample size. Simulation studies demonstrate the empirical performance of the proposed method.
|
