Goodness of fit measure for a MNL model

J

Jüri

Guest
#1
Hello! I have carried out MNL regression with the same model among different samples and I now need to compare the results against each other. In order to do that I'd need a single indicator that describes the goodness of fit of each model. Does something like this exist for MNL regression?

Would it be acceptable to base conclusions about the differences between samples on pseudo-R? I can live with the fact that it can't be interpreted in any way, but does it have any other disadvantages?
 
#2
I believe that you can. There's also some other information measures (AIC, BIC) that you can use as well to determine the goodness of fit. Regarding the AIC and BIC, the lower the value, the better the fit - but the values do not have an intrinsic meaning.