I found this post titled: "Positive coefficient but negative marginal effect in mlogit". EDIT: However I recently had the same "issue" with an ordinal probit model and the marginal effect of a a dummy variable.
In my situation, I am interested in the marginal effect of a dummy on an ordinal variable (post).
Maarten Buis explains in his answer: "This can happen. Remember that the raw coefficients in a [this case] mlogit refer to changes in the log of the probability of one outcome divided by the probability of the baseline outcome (the logarithm of an odds), while the marginal effects refer to changes in the probability of one outcome. A variable can influence the one probability and the baseline probability so that it has a positive effect on the ratio, but a negative effect on the the one probability."
I am sure he is right, but I have some trouble really following what he is saying. I was wondering:
What does this mean for my coefficient of interest, which I expected to be positive? How do I interpret what is going on?
Would someone be willing to give a more "basic" and more elaborate explanation of how this can happen, perhaps with a numerical example?
Related post:
How to understand output from R's polr function (ordered logistic regression)?
