Parametric or Nonparametric Multilevel Factor Mixture Modeling? A Comparison Through a Simulation Study

• In Event: Fresh Perspectives on Estimating Complex Multilevel Models: Counts, Mixtures, Interactions, and Machine Learning

Abstract

Multilevel factor mixture modeling (MLFMM) is used to investigate unknown heterogeneity and takes into account data nesting common in educational data. However, a dearth of studies investigated the performance of parametric vs. nonparametric approaches in MLFMM. We investigate the performances of parametric vs. nonparametric MLFMM under four scenarios that would occur in real research settings through a series of Monte Carlo simulation studies. The preliminary results of Case 1 showed that using the parametric approach, when the nonparametric model was the true model, yielded unbiased estimates of factor mean difference between two latent classes but led to severe bias in class proportions and low accuracy in class assignment in some simulation conditions. We provide practical recommendations for applied researchers.

Authors

• Eunsook Kim, University of South Florida

• Chunhua Cao, University of Alabama

• Yan Wang, University of Massachusetts Lowell