On the Requirements of Non-linear Dynamic Latent Class SEM: A Simulation Study with Varying Numbers of Subjects and Time Points

• Published in: Structural equation modeling Vol. 30; no. 5; pp. 789 - 806
• Main Authors: Andriamiarana, Vivato, Kilian, Pascal, Kelava, Augustin, Brandt, Holger
• Format: Journal Article
• Language: English
• Published: Hove Routledge 03.09.2023; Psychology Press
• Subjects: Bayesian estimation; Bayesian Statistics; dynamic structural equation model; hidden Markov model; intensive longitudinal data; Latent class analysis; Simulation; Structural equation modeling; Structural Equation Models
• ISSN: 1070-5511; 1532-8007
• DOI: 10.1080/10705511.2023.2169698

Although small sample sizes represent an important issue, few studies investigated the requirements in dynamic latent variable model frameworks (e.g., dynamic structural equation modeling, DSEM; dynamic latent class analysis, DLCA). We conduct a small sample performance study of Bayesian estimation for the non-linear dynamic latent class structural equation model which generalizes DSEM and DLCA to include time-dependent latent class transitions. We simulate data using a two-level (non-linear) dynamic latent class model with a varying number of subjects ( ) and time points (T = 10, 25, 50) which are in our main focus among other simulation conditions. The results show that at least a sample size of with is required to ensure good estimates. Using diffuse priors on the between level, especially for the (co-)variance parameters and the factor loadings should be avoided.