|
Activity Number:
|
330
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 11, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
SSC
|
| Abstract - #300847 |
|
Title:
|
A Model Selection Criterion for Marginal Zero-inflated Regression for Clustered Data
|
|
Author(s):
|
Daniel B. Hall*+ and Zhengang Zhang
|
|
Companies:
|
University of Georgia and University of Georgia
|
|
Address:
|
Dept. of Statistics, Athens, GA, 30602-1952,
|
|
Keywords:
|
GEE ; zero-inflated Poisson ; information criterion ; longitudinal data ; repeated measures ; mixture model
|
|
Abstract:
|
Applying methodology of Rosen, Jiang, and Tanner, Hall and Zhang have recently described methods for fitting marginal versions of zero-inflated regression models to clustered data via the expectation-solution algorithm. This approach involves replacing the estimating equations solved in the M step of the usual EM algorithm for zero-inflated regression under independence with generalized estimating equations (GEEs). These GEEs involve working correlation matrices to account for within-cluster correlation. We propose a model selection criterion which can aid in the selection of appropriate working correlation matrices as well as in the selection of covariates for the linear predictors of the model. The new criterion is akin to Akaike's information criterion (AIC) and also to Pan's adaptation of AIC to the GEE context. The performance of the new criterion is examined via simulation. In addition, the extension of this approach to the more general finite mixture of marginal models context of Rosen et al.'s (2000) paper is considered.
|
