Hello everyone,
I have what seems to me to be a very basic problem with marginal effects following a mlogit model, but I can't get my head around it. Browsing other Statalist posts as well as the Stata help has not solved the problem either (unless I missed "the" perfectly fitting post or help section). Now to the problem: I am aware of the fact that I can display relative risk ratios (RRR) via the "rrr" option when estimating a "mlogit" model. However, I run into problems when calculating the conditional marginal effects for an interaction via the "margins" command as relative risks ratios. I can easily calculate the conditional AMEs, but I have major problems when I try to get the conditional RRR. I have tried many different combinations (some thoughtful, some more trial and error), but have been unable to determine the conditional RRR for each outcome of the mlogit model. I've seen Statalist posts about logit models where additionally storing the baseline odds and using these in a further step solved the problem, but is there really no convenient way to get the RRR (or OR) directly using margins?
As an example, one could imagine the following simple (and also rather pointless) model for which the conditional AMEs are calculated:
Ultimately, I personally don't care if I would have to estimate the conditional RRR for each result with a separate "margins" command, or if the conditional RRR is displayed for all results at once (as above).
Many thanks in advance!
Marvin
I have what seems to me to be a very basic problem with marginal effects following a mlogit model, but I can't get my head around it. Browsing other Statalist posts as well as the Stata help has not solved the problem either (unless I missed "the" perfectly fitting post or help section). Now to the problem: I am aware of the fact that I can display relative risk ratios (RRR) via the "rrr" option when estimating a "mlogit" model. However, I run into problems when calculating the conditional marginal effects for an interaction via the "margins" command as relative risks ratios. I can easily calculate the conditional AMEs, but I have major problems when I try to get the conditional RRR. I have tried many different combinations (some thoughtful, some more trial and error), but have been unable to determine the conditional RRR for each outcome of the mlogit model. I've seen Statalist posts about logit models where additionally storing the baseline odds and using these in a further step solved the problem, but is there really no convenient way to get the RRR (or OR) directly using margins?
As an example, one could imagine the following simple (and also rather pointless) model for which the conditional AMEs are calculated:
Code:
sysuse auto mlogit rep78 c.weight##i.foreign margins foreign, dydx(weight)
Many thanks in advance!
Marvin
