|
Activity Number:
|
108
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
IMS
|
| Abstract - #309344 |
|
Title:
|
Labeling Issue in Finite Mixture Model: A Frequentist View
|
|
Author(s):
|
Daeyoung Kim*+ and Bruce G. Lindsay
|
|
Companies:
|
The Pennsylvania State University and The Pennsylvania State University
|
|
Address:
|
326 Thomas Building, University Park, PA, 16802,
|
|
Keywords:
|
Finite mixture model ; Labelling non-identifiability ; Asymptotic identifiability ; Parametric bootstrap
|
|
Abstract:
|
Our research addresses the problems caused by labeling nonidentifiability in a parametric finite mixture model from the frequentist view. We propose a nonparametric method to see if the labels on the parameters are well defined so as to be useful in deciding Monte Carlo (bootstrap) standard errors for the parameters. Although there is a form of asymptotic identifiability for consistent labeling on the parameters, this will not work well when the components densities are not well separated or the sample size is not large. The nonparametric approach we investigate provides information on the labeling identifiability for parameters by constructing a data set that contains each bootstrapped parameter estimate multiple times, one for each possible labeling. If this data set breaks into well-separated clusters, then we consider asymptotic identifiability to be reasonable.
|
