Mixture of Experts Latent Trait Analyzers

The Mixture of Latent Trait Analyzers (MLTA) represents a model-based clustering approach specifically tailored to multivariate categorical data. It accommodates clustering of units through a finite mixture specification, while also modeling the residual latent variability of units within each cluster through a set of multidimensional latent variables (traits). The original formulation is extended to account for the effect of concomitant variables (covariates). These are allowed to affect cluster formation, the conditional outcome distribution, both (as in standard mixtures of experts models), or neither. Overall, the proposal improves the flexibility of the original MLTA specification, as well as its capacity to reflect the complexity of the data.