Evaluation of Two Methods for Multistep Latent-Class Analysis

• In Event: Poster Session 10
  In Poster Session: Division D Section 2 Poster Session 2

Abstract

This paper investigates the performance of two multi-step analysis techniques that involve a latent-class analysis in the first step, where the first step analysis would function to classify observations into two or more latent classes. Then, a subsequent analytic model would be fit separately for each of the identified classes in the first step. Two techniques investigated were 3-step approach and sampling weight approach. A simulation study has revealed that the 3-step approach performed better overall. However, it was also revealed that SE was uniformly smaller for the sampling weight approach, while bias was uniformly smaller for the 3-step approach. Therefore, this warrants further investigations, and more conditions will be investigated under a more complex model.

Authors

• Akihito Kamata, Southern Methodist University

• Yusuf Kara, Anadolu University

• Chalie Patarapichayatham, Southern Methodist University

• Patrick Lan, Southern Methodist University