Fitting measures for ordinal data models

A relevant issue for validating models is the assessment of goodness-of-fit and related measures of predictive ability. When data are nominal, and specifically ordinal, the main problem is the absence of a standard paradigm as in the regression framework for residual variability; in fact, several measures have been proposed. In this contribution we explore fitting measures for ordinal data when these are modelled by a mixture distribution. Some new indexes are evaluated and a comparison with previous proposals is performed by means of simulated and real data sets.