Bayesian variable selection in a class of mixture models for ordinal data: a comparative study