Dear Statalist:

I am trying to verify something about the -margins- command after xtlogit, fe. When you do xtlogit, fe, it drops a bunch of the cases where there is no variation. Does the -margins- command compute the predicted probability for the estimation sample after the dropped cases, or before the dropped cases?

And just to head this off: I am well aware that there are some major questions about how to interpret the margins after fixed-effects logit.