Applying ordered logit to non-ordered dependent variables

[Tinge Yin]

Dear Statalist,

There are so many convenient codes for ordinal dependent variables (e.g., Extended regression models, eoprobit), almost no advanced code for nominal dependent variables.
My dependent variable has 3 categories and there is no order. The sample is a panel. I need to adjust for sample selection and endogeneity.

I wonder what are the potential problems if I use the models designed for ordered category variables. I found many discussion about the reverse situation: using multinominal if DV is ordinal, but not vice versa. Any suggestion and input will be appreciated.

[Richard Williams]

Unless you can come up with an ordering for the variable, I don't think that makes much sense. But the help for me says

"Although there is no memlogit command, multilevel mixed-effects multinomial logistic models can be fit using gsem; see [SEM] example 41g."

So, I would check out gsem. Also maybe there are some user-written commands that people can suggest. You might also think about whether it would be ok to dichotomize your variable so you could use logit and probit commands.

Also, the user-written -cmp- command estimates an incredible number of models. Maybe it can do what you want.