|
Activity Number:
|
57
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
IMS
|
| Abstract - #309252 |
|
Title:
|
Variable Selection Procedures for Generalized Linear Mixed Models in Longitudinal Data Analysis
|
|
Author(s):
|
Hongmei Yang*+ and Daowen Zhang and Hao Zhang
|
|
Companies:
|
North Carolina State University and North Carolina State University and North Carolina State University
|
|
Address:
|
3002 Kings Ct Apartment C, Raleigh, NC, 27606,
|
|
Keywords:
|
Smoothly clipped absolute deviation ; Penalized quasi-likelihood ; Longitudinal data ; Restricted maximum likelihood ; Generalized linear mixed models ; Variance components
|
|
Abstract:
|
For non-sparse longitudinal data such as count data and binomial data with moderate to large binomial denominators, we propose Penalized Quasi-Likelihood (PQL) procedure for simultaneous model selection and estimation. Due to the low estimation ability of PQL for binary data, we propose three other procedures: Full Likelihood Model Selection (FLMS), Two-stage Penalized Quasi-Likelihood Model Selection (TPQLMS) and approximate Marginal Likelihood Model Selection (AMLMS). Among them, FLMS and PQLMS have the feature of selecting informative variables and estimating regression parameters simultaneously. A robust estimator of standard deviation is derived based on a sandwich formula and tested through simulations for FLMS and PQLMS. A bias correction is proposed to improve the estimation accuracy of PQLMS.
|
