Latent Class Mediation: A Comparison of Three Approaches

• In Event: Methods for Multilevel, Categorical, and Multidimensional Data

Abstract

Latent class mediation modeling is designed to estimate the mediation effect when both the mediator and the outcome are latent class variables. We compare the performance of three approaches: adjusted one-step, maximum-probability, and multiple pseudo class draw approaches. Results from the current study indicate the three approaches differ with respect to bias and estimation error, particularly at different levels of measurement quality (i.e., model entropy). The adjusted one-step approach produced accurate mediation estimates regardless of entropy levels. When entropy values are equal to or larger than .90, the maximum probability approach yielded unbiased estimates of the mediation effects, however biases increased at lower entropy levels. The multiple pseudo class draw approach generated substantially biased mediation estimates in all simulation conditions.

Authors

• Yu-Yu Hsiao, University of New Mexico

• Eric Stephen Kruger, The University of New Mexico

• Lee Van Horn, University of New Mexico

• Davood Tofighi, University of New Mexico

• David Mackinnon, Arizona State University

• Katie Witkiewitz, University of New Mexico