Some Robust Approaches for K-component GLMM Mixtures
- Kai Wing Kelvin YAU (Principal Investigator / Project Coordinator)Department of Management Sciences
- Yer Van HUI (Co-Investigator)
DescriptionThe recently developed k-component generalized linear mixed models (GLMM) mixture model is found to be an effective statistical tool to model the heterogeneity for correlated data coming from several latent subpopulations. In the literature, various robust approaches have been applied extensively for parameter estimation in the GLMM and the finite mixtures respectively, such as the robust version of the residual maximum quasi-likelihood estimation in the GLMM based on Huber -function and Minimum Hellinger distance estimation in finite Poisson mixtures. This project aims to develop some robust approaches for the correlated data arising from several latent subpopulations in the context of k-component GLMM mixtures with random effects. Generalizations of the proposed approaches in various directions will also be explored to give further flexibility for modelling correlated data coming from a mixture of distributions with probable contaminations. The applicability of the proposed approaches is assessed via investigating two empirical studies.
|Effective start/end date||1/12/06 → 9/02/10|