A latent variable probit model for multivariate ordered categorical responses

This paper presents a fully Bayesian approach via Gibbs sampling
for MIMIC models with ordered categorical outcomes. The method is
of particular interest for moderate or medium sample size data
situations as in the study to be presented. Compared to frequentist
methods that are based on large sample theory, estimates and
standard errors of parameters are more reliable. Experience from
simulations and the application to the particular study on changes
of styles of marital conflict resolution suggest that the approach
provides a useful supplementary tool in combination with
traditional methods.