AIC and BIC for odered probit and ordered logit?

[Gabriele Mauro]

Hi everybody,

I was wondering if it is theoretically sound to compare an ordered probit and a generalised ordered logit with the same set of explanatory variables and of course on the same panel data through AIC and BIC tests. The underlying assumptions are different in terms of error distribution and variability of slope coefficients, I know, but the dominant model in my area of research is ordered probit but I have reasons to think that a generalised ordered logit would be a better choice.

Thanks!

[Maarten Buis]

They are nested, so you can use a conventional likelihood ratio test.