Fitting Structural Equation Models via Variational Approximations

Structural equation models are commonly used to capture the relationship between sets of observed and unobservable variables. Traditionally these fitted using frequentist approaches, but recently researchers practitioners have developed increasing interest in Bayesian inference. In settings, inference for is typically performed via Markov chain Monte Carlo methods, which may be computationally intensive with a large number manifest variables or complex structures. Variational approximations can fast alternative; however, they not been adequately explored this class models. We develop mean field variational Bayes approach fitting elemental structural demonstrate how bootstrap considerably improve approximation quality. show that method provide reliable while being significantly faster than methods.