Bayesian Latent Class Analysis: Sample Size, Model Size, and Classification Precision

• In Event: Bayesian Techniques in Structural Equation Modeling

Abstract

The present study used Markov chain Monte Carlo simulations to compare the classification precision of latent class models using robust maximum likelihood (MLR) versus Bayes estimation. Models with four binary observed indicators and 2, 3, and 4 latent classes were estimated using samples of 1000, 750, 500, 250, 100, and 75 observations. For each condition, entropy and average probabilities for most likely latent class membership were compared. Regardless of sample size, Bayes estimation showed greater classification precision for the 2-class and the 3-class models but was outperformed by MLR for the 4-class model when samples were small (N=75, N=100). As the complexity of the model increased, the number of successful computations decreased for Bayes estimation, particularly for the 4-class model.

Author

• Diana L. Mindrila, University of West Georgia
  Presenting Author