A flexible approach to finite mixture regression models for multivariate mixed responses

We describe a flexible regression model for multivariate mixed responses, where association between outcomes is modeled through discrete, outcome-specific, latent effects, accounting for heterogeneity and dependence. A relevant question arises when independence between outcome-specific random effects should be tested. By relaxing the uni-dimensionality of the standard finite mixture model, we define a multidimensional latent class approach, with possibly different numbers of locations in each margin, and a full association structure.