Margins for multiple terms

[Bill Smith]

Suppose I have:

logistic y a b

where a and b are continuous.

I can run

margins, at(a==10 20 30)) atmeans

and

margins, at(b==(50 60 70)) atmeans

to obtain adjusted predictions at selected values of a at the mean of b and selected values of b at the mean of a (I think this is correct).

Is there some way to do this simultaneously? I'm saving post-estimation results from margins and I think I'll have to run my regression again unless perhaps I can use preserve and restore. Any advice appreciated.

[Clyde Schechter]

I don't see any way to combine those into a single -margins- command.

But you definitely do not have to re-run the regression to use -margins- a second time. You can run as many different -margins- commands as you want after a single regression. -margins- for a given estimation command remains available until you run a new estimation command.

[Marcos Almeida]

The examples I ever saw sofar with - margins, at() - used "=" instead of "==". To my surprise, I tested - margins - a few minutes ago with the double equal sign and it worked the same way as the equal sign. That said, in order to avoid potential problems (albeit I cannot envisage at this very moment), I fear we'd better follow the Stata manual (http://www.stata.com/manuals13/rmargins.pdf), the Stata Journal (http://www.stata-journal.com/sjpdf.h...iclenum=st0260) and a series of papers and hand outs, and stick to the first option.

[Richard Williams]

Is this what you have in mind?

webuse nhanes2f, clear
logit diabetes height weight
margins, at(height = (150 175 200) weight) at(weight = (30 100 175) height)
margins, at(height = (150 175 200) _continuous) at(weight = (30 100 175) _continuous)

The last variation might be handy if you have a bunch of other vars. See help margins##atspec.

[Bill Smith]

This is the solution that I need:

margins, at(height = (150 175 200) weight) at(weight = (30 100 175) height)

That gives me margins of height at the mean of weight and margins of weight at the mean of height. Perfect. Thanks for your help.