The Role of Prior Distributions in Parameter Recovery and Efficiency of Using Stan for Dynamic Structural Equation Modeling

• In Event: Division D Graduate Student in Progress Research Gala

Abstract

The dynamic structural equation modeling (DSEM) approach incorporates multilevel, Bayesian, and traditional SEM to model the dynamic relationship between observed and latent variables. Hamiltonian Monte Carlo (HMC) algorithm avoids the random walk process used by Metropolis Hastings and Gibbs Sampling algorithms for sampling from the parameter space of the posterior distribution, making it generally more efficient. The goal of this research paper is to investigate how much sample size does a weakly informative prior need to accurately recover the DSEM parameters using the advanced dynamic HMC algorithm of Stan.

Author

• Reginald Ziedzor, Southern Illinois University