hi, first let me say that looking into the issue of heteroskedasticity in a binary-choice (probit/logit) setup, I found two existing function packages, namely HIP by Jack and Claudia, and LOGIT_HETERO by Artur. This is showing the great evolution of gretl in connection with user-contributed stuff. (Although Jack probably doesn't really qualify as a "user" in this sense, but anyway.) Ok, now I've also noticed that the 'logit' and 'probit' commands offer a --robust option. However, I wonder whether this is really sensible, because we all know that estimates are inconsistent under heteroskedasticity here. Perhaps this has been discussed before, does it serve a pedagogical purpose or am I missing something? thanks, sven