|
Activity Number:
|
312
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Social Statistics Section
|
| Abstract - #301633 |
|
Title:
|
Determining the Number of Classes in Growth Mixture Modeling: Covariate, Data Generation Scheme, and Class Specificity
|
|
Author(s):
|
Libo Li*+ and Yih-ing Hser
|
|
Companies:
|
University of California, Los Angeles and University of California, Los Angeles
|
|
Address:
|
Integrated Substance Abuse Programs, Los Angeles, CA, 90025,
|
|
Keywords:
|
Growth mixture model ; class enumeration ; data generation scheme ; class specificity ; covariate ; adjusted BIC
|
|
Abstract:
|
Our studies on growth mixture models (GMM) show that the inclusion of covariates to the class-invariant GMM fitting models is generally helpful for their correct class enumeration given model consistency. However, the class-specific GMM models that are less restrictive and have been demonstrated to be capable of parameter and membership recovery perform very poorly for class enumeration except in large samples or when some informative covariates are included. Our extensive discussion of data generation schemes also calls into question the current common practice in GMM applications, which segregates class enumeration and final interpretive models by using different fitting models. In addition, across two studies, we find that the adjusted Bayesian Information Criterion is the best index for class enumeration in the GMM context.
|
