Assessing the goodness of fit of a latent variable model for ordinal data