Jeff Douglas, PhD
Associate Professor Department of
Statistics University of Illinois-Urbana/Champaign
Friday, January 30th, 2009
12:00 pm - 1:00 pm
|Constrained Finite Mixture Models for Multivariate Categorical Data |
Several developments in finite mixture modeling for psychological diagnosis and educational assessment will be discussed. Specific types of constrained finite mixture models, such as models for cognitive diagnosis, have become increasingly popular in educational testing with the aim of diagnosing the particular skills an examinee may possess. Such techniques - developed with educational purposes in mind, can also be of use in psychology (among other areas) where diagnosis is important, once fully developed.
A review of several models will be given, and the assumptions that distinguish them will be considered. Methods for adaptive item selection, parameter estimation, and model selection will be discussed. Finally, the relationship between these finite mixture models and cluster analysis will be investigated, with theoretical results that indicate when clustering should arrive at the same groups as model-based methods. Applications to educational testing and diagnosis of pathological gambling will be given.
Return to seminar list