|
Activity Number:
|
67
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #308245 |
|
Title:
|
Minimum Hellinger Distance Estimation of Finite Mixtures of Generalized Linear Models
|
|
Author(s):
|
Jing Shen*+ and Daniel Hall and Chenhua Zhang
|
|
Companies:
|
IBM T.J. Watson Research Center and University of Georgia and University of Georgia
|
|
Address:
|
71 Charter circle, Ossining, NY, 10562,
|
|
Keywords:
|
Conditional Density ; Robustness ; Mixture Models ; Zero Inflated Poisson ; Zero Inflated Binomial
|
|
Abstract:
|
We study minimum Hellinger distance estimation (MHDE) method for finite mixtures of generalized linear models (FMGLMs). MHDE method as a robust approach is well established for iid case. Recently, it has been extended to mixture models with regression context but leaving asymptotic properties unexamined. In this paper, we define a new MHDE method in the general FMGLMs context based on conditional densities. We prove that our method yields consistent and asymptotic normally estimators. Numerical results suggest that our method is more robust than MLE with the presence of outliers and competitive otherwise. In addition, we demonstrate that our method is more efficient than the previous one in the literature and can be applied more broadly. An example on a cohort study which characterizes ambulatory electrocardiography results of overtly healthy dog is used to illustrate our method.
|
