Effect of violation of the parallel slopes assumption on estimated probabilities in a xtologit

[Sam McCaw]

Hello,

I am estimating a ordered response panel model as a predictive analysis. Since a critical assumption of the ordered logit is the parallel slopes assumption, I was wondering how a violation of this assumption will affect the predicted probabilities.

In other words, what is the nature of the impact of a violation in the assumption on the probabilities?

I am thinking of including an mlogit to relax this assumption. Is that an appropriate choice?

Many thanks.

SAM

[Joseph Coveney]

Maybe

help simulate

I'm not trying to be snide: I assume that you've looked at the literature and that it says something like: "it's bad" and "the greater the violation, the worse it is"; one way to find how much of a difference it makes to you in your circumstances is to simulate under your conditions and see how bad things are for those characteristics of the fitted model's predictions that matter to your research question.

You could use the same strategy, a simulation approach, to see whether mlogit provides solution to your problem.

[Sam McCaw]

Thanks, Joseph. Yes, I have looked at the literature and it certainly suggests that when violated, the xtologit is not the appropriate model to use which is why I am also using an mlogit. My question was more on the nature of how I can expect it to affect the probabilities specifically so I could understand it better. I will try to simulate the violation. Grateful if you can suggest any literature that specifically talks about the impact on the violation of the probability estimates. Many thanks.

SAM

[Joseph Coveney]

I'm not familiar with it specifically, sorry.