Class Identification Efficacy in Piecewise GMM with Unknown Turning Points

ABSTRACT

Piecewise GMM with unknown turning points is a new procedure to investigate heterogeneous subpopulations' growth trajectories consisting of distinct developmental phases. Unlike the conventional PGMM, which relies on theory or experiment design to specify turning points a priori, the new procedure allows for an optimal location of turning points based on data. The advantage of the procedure has gained increasing attention in educational and behavioral research, but a major challenging issue, class enumeration performance of the model, has not yet been investigated. The current simulation study compared the performance of PGMMs with unknown turning points in identifying the correct number of latent classes under both Bayesian and ML/EM estimation methods.

KEYWORDS:

• Bayesian estimation
• ML/EM estimation
• model-selection indexes
• piecewise growth mixture model
• turning point

Note

Notes

1. For the 4-class model, the priors for the first three classes were based on estimates from ML/EM, and the priors for the fourth class were specified based on the average of the first three classes.