Contrasts of risk differences with mlogit margins

[Melvin Donaldson]

Hello

The Stata manual has a good primer on implementing mimrgns after mi est: mlogit but even after searching through the Statalist, I'm having trouble calculating contrasts between outcomes.

webuse sysdsn1
mlogit insure age i.male i.nonwhite i.site
margins nonwhite, predict(outcome(Indemnity))

And that's pretty great. And I can get the contrasts of nonwhite for each level of the outcome by omitting the outcome expression and including the contrast operator.

margins r.nonwhite

What I would like is a point estimate of the contrasts of contrasts with a confidence interval. If indemnity (outcome(1)) is a reference, how do I obtain the 2 contrasts (of contrasts), the difference between the nonwhite contrast in outcome(2) vs outcome(1)? over(insure) does not get me what I want. within does not work because insure is the outcome and not a covariate.

I believe what I want is:

margins, dydx(nonwhite) pwcomp

but I'd ultimately like to do this with mimrgns, which doesn't accept pwcomp.

Thanks for your help!

[Melvin Donaldson]

So far, I've been able to get the estimates and CIs by posting the mimrgns results and manually calculating the contrasts with lincom. I'm looking for a more streamlined version but this at least works for now.