Evaluating Latent Class Models When Ignoring and Accounting for Clustering

• In Event: 45.075 - Multilevel Models, Multiple Imputation, and Latent Models: Unrelated but Equally Important Topics

Abstract

Multilevel latent class analysis (MLCA) has been increasingly used to investigate unobserved population heterogeneity while taking into account data dependency. This study examined the impact of ignoring clustering on class enumeration in LCA and the performance of MLCA in terms of class enumeration and assignment accuracy. In these investigations, we distinguished latent class indicators that were meaningful at the within-level only or at both the within- and between-levels (i.e., within indicators and within-between indicators). Results showed that ignoring clustering could still lead to the correct number of within-level classes for within indicators but not for within-between indicators. MLCA performed reasonably well with more accurate class enumeration for within indicators than within-between indicators. Recommendations for applied researchers are discussed.

Authors

• Yan Wang, University of Massachusetts - Lowell

• Seang-Hwane Joo, ETS

• Seok Joon Chun, South Dakota State University

• Abeer A. Alamri, University of South Florida

• Phil Seok Lee, George Mason University

• Eunsook Kim, University of South Florida

• Stephen E. Stark, University of South Florida