I am estimating a model with several categorical outcomes. It is my understanding that, to obtain odds ratios, I need to use the MLR estimator rather than the WLSMV estimator. (The WLSMV estimator will give me probit regression coefficients rather than logistic regression coefficients, correct?)
If this is accurate, then the only model fit indices available to me are the -2LL, the AIC, and the BIC. One of my coauthors wants to know how well the model fits the data, but these indices cannot tell me that - correct? Is there any way for me to tell how well the model fits? Can anything be done with the -2LL, AIC, or BIC to evaluate the fit of a single model to the data? I'm worried that reviewers will criticize me for not making a statement about overall model fit before proceeding to test the signficance of specific paths.
If your model has positive df for chi-square when running WLSMV (probit) then it has some left-out paths. If so, you can use ML (logit) and let those paths be free to get the H1 model to test your H0 model against.
I am running SEM with categorical variables (some binary; some ordinal). I am using ML method in SAS to fit the model. I was wondering whether I should use polychoric correlation matrix as my input to the model (instead of the default, which I assume is Pearson's correlation). Or is the ML method robust enough even when all variables are categorical? Thank you!