CLASS EXTRACTION AND CLASSIFICATION ACCURACY IN LATENT CLASS MODELS
Open Access
- Author:
- Wu, Qiong
- Graduate Program:
- Educational Psychology
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- September 30, 2009
- Committee Members:
- Pui Wa Lei, Dissertation Advisor/Co-Advisor
Pui Wa Lei, Committee Chair/Co-Chair
Hoi Kin Suen, Committee Member
Jonna Marie Kulikowich, Committee Member
Aleksandra B Slavkovic, Committee Member - Keywords:
- classification accuracy
correct class extraction
latent class models
sample sizes - Abstract:
- Despite the increasing popularity of latent class models (LCM) in educational research, methodological studies have not yet accumulated much information on the appropriate application of this modeling technique, especially with regard to requirement on sample size and number of indicators. This dissertation study represented an initial attempt to inform practitioners of desirable sample sizes and number of indicators under various conditions by using simulated data. The performance of LCM was evaluated in terms of both correct class extraction rates (i.e., correct identification of number of latent classes) and classification accuracy for dichotomous and continuous indicators. Manipulated factors included class separation, true number of latent classes, sample size, number of indicators, and class proportion. For both dichotomous and continuous indicators, class separation and true number of latent classes consistently demonstrated the largest effects on both correct class extraction rates and classification accuracy. The results suggest that LCM may not be recommended for use when class separation is small (operationally defined as .20 difference in indicator means for dichotomous indicators and half a standard deviation difference in indicator means for continuous indicators). When class separation is large (operationally defined as .50 difference in dichotomous indicator means and 2 standard deviation differences in continuous indicator means), models show acceptable correct class extraction rates and classification accuracy when certain conditions on sample size and number of indicators are met. Recommendations on sample size and number of indicators are provided taking into account researchers’ level of knowledge of class separation and true number of latent classes.