Calculating AMEs after mlogit using mimrgns

[Sarah Thea Smith]

I am trying to calculate AMEs that have resulted from a multinomial logistic regression using multiple imputation. My dependent variable (y) has 3 levels. At present, I have tried the following syntax to do so:

quietly mi estimate, dots post: mlogit y x1 x2 x3

mimrgns, dydx (*) predict(pr)

When I do so, Stata gives me the output for the base category only as far as I understand: "Expression: Pr(y==0), predict(pr)", but I am interested in calculating AMEs for y=1 and y=2. How can I go about doing so?

Thank you!

[Andrew Musau]

mimrgns is from SSC (FAQ Advice #12). Even with the official margins command, you need to include nothing within -predict()- to get the full set of marginal effects.

mimrgns, dydx (*) predict()

[Sarah Thea Smith]

Thank you for your quick response. It didn't fully work, but did lead me to the correct answer I believe.

It seems that while predict(pr) is the margin command's default, predict(xb) is the mimrgns command default, so when I left the predict() option blank as you suggest, the output resulted in: "Expression: Linear prediction, y==0, predict(xb)" as well as showing "0 (omitted)" for all of the variables.

However, having looked into the pr options more, it seems that the following solves my issue:

 
 mimrgns, dydx (*) predict(default)