I'm running factor models with count data indicators. For multiple group modeling I use the knownclass option for type = mixture. I know how to perform model comparisons of nested models by means of likelihood ratio testing, which allows for testing if a model fits the data better than a more restricted one.
But is there anything I can do for inferring the fit of a model to the data? Especially, if cells of the contingency table for the chisquare-test were removed, because of extreme values?