Investigating Approaches to Estimating Covariate Effects in Growth Mixture Modeling: A Simulation Study

Area of Science:

• Statistics
• Psychometrics
• Quantitative Psychology

Background:

• Accurate estimation of covariate effects in mixture modeling is crucial for understanding latent class membership.
• Growth mixture models (GMMs) present unique challenges for identifying latent variables and estimating covariate effects.
• Limited research compares different estimation approaches for covariate effects within GMMs.

Purpose of the Study:

• To investigate and compare the performance of four estimation approaches for covariate effects in logistic regression class membership models within a GMM framework.
• To evaluate the bias and accuracy of these methods under varying degrees of class separation.

Main Methods:

• A Monte Carlo simulation study was conducted.
• Four estimation approaches were compared: conventional three-step, one-step maximum likelihood (ML), pseudo-class (PC), and three-step ML.
• Performance was assessed based on the recovery of covariate effects.

Main Results:

• When latent class separation was large, the one-step ML and three-step ML approaches yielded less biased covariate effect estimates compared to the conventional three-step and PC approaches.
• Poor class separation significantly impacted the estimation of the covariate-latent class relationship when using the three-step ML approach.

Conclusions:

• The one-step and three-step ML methods are preferable for estimating covariate effects in GMMs, particularly when classes are distinct.
• Researchers should exercise caution when using the three-step ML approach in GMMs with poor class separation due to potential estimation issues.