Search Results

Author: Yang, Chih-Chien
Resulting in 1 citation.
1. Yang, Chih-Chien
Finite Mixture Model Selection with Psychometric Applications
Ph.D. Dissertation, University of California - Los Angeles, 1998. DAI-A 59/09, p. 3421, Mar 1999
Cohort(s): NLSY79
Publisher: UMI - University Microfilms, Bell and Howell Information and Learning
Keyword(s): Addiction; Alcohol Use; Data Analysis; Modeling; Modeling, Growth Curve/Latent Trajectory Analysis; Modeling, Mixed Effects; Monte Carlo; Statistical Analysis

Recent statistical advances, for example, Bandeen-Roche, Miglioretti, Zeger and Rathouz (1997); Jedidi, Jagpal and DeSarbo (1997); and Wang and Puterman (1998) have made it feasible to fit finite mixture models in a wide range of applications. With a collection of plausible models for a given data set, problems of model selection arise. Selection among finite mixture models often involves a choice among models with different number of latent classes. As pointed out by several researchers (Everitt, 1981; Aitkin & Rubin, 1985), this can be problematic for traditional likelihood ratio tests because of the unknown distribution of the likelihood ratio. Therefore, it is desirable to investigate information criteria for alternative model selection procedures. Lin and Dayton (1997) showed that the accuracy of widely used criteria, AIC/BIC/CAIC, can be dissatisfying for complex finite mixture models. To improve the accuracy, a relatively newer criterion (Draper, 1995) as well as an adjustment of standard criteria have been suggested. The aim of this study is to investigate the accuracy of information criteria and standard likelihood ratio testing methods for various finite mixture models. Monte Carlo studies in this dissertation show that improvements of accuracy by using the adjusted information criteria are considerable. Guidelines are also provided for practical use of these model fit indices in estimating the number of latent classes, especially when different sample sizes, parameter structures and model complexities are involved. Following these guidelines, the optimum accuracy rates of these model fit indices can be achieved; moreover, situations that can cause low accuracy can be avoided. Finite mixture models for an alcohol dependence and abuse study using datasets from the National Longitudinal Surveys of Youth (NLSY) are also illustrated. Both model selections and interpretations for the finite mixture models are emphasized using these examples.
Bibliography Citation
Yang, Chih-Chien. Finite Mixture Model Selection with Psychometric Applications. Ph.D. Dissertation, University of California - Los Angeles, 1998. DAI-A 59/09, p. 3421, Mar 1999.