I am using xtlogit with fe options. I notice that the results drop some observations due to all positive or all negative results.

I am wondering how does predict work to those dropped observations? It looks like predict will give values for all observations, even though they are dropped during the regression.