I estimated an IRT model on 14 items measured on a 5-point likert scale using MLR. M+ indicates that THE CHI-SQUARE TEST IS NOT COMPUTED BECAUSE THE FREQUENCY TABLE FOR THE LATENT CLASS INDICATOR MODEL PART IS TOO LARGE. However, other programs do compute chi-square for this model. Is there a way to bypass this problem and obtain chi-squares?
Another problem: I estimate several different models. In each of these models, there are 8 items on a 5-point scale. The degree of freedom for each chi-square differ because M+ delete a certain number of "cells in the latent class indicator table" "due to extreme value". Could you please indicate a reference that justifies this deletion?
When the frequency table for chi-square is very large as in your situation, there are too many cells ( 5*5*5*etc.) to make the test meaningful. We don't recommend using it with over 8 binary indicators. I think you can find a discussion of this in the Agresti book.
thank you for your prompt answer. I would just like to ask some precisions. I know that the frequency table for a questionnaire with 14 questions and 5 modalities per question is 5^14. However, is there another goodness of fit statistic, given by Mplus, that could help to estimate if the model fits (i.e. absolute fit and not relative fit like AIC)? Second question: concerning the deletion of extreme value, I wish to publish examining several 8 items scales. The degrees of freedom should be the same. However, since the deletion is not always the same, the df differ. I am sure that the reviewers will ask why and I would like to cite a reference explaining the deletion and thus the difference in df.