On Sun, 14 Jun 2009, Allin Cottrell wrote: > A further response to Nicolai Striewe's question in > http://lists.wfu.edu/pipermail/gretl-users/2009-May/003346.html : > > There are three circumstances in which gretl deliberately drops > regressors from a specification given by the user: > > 1. (all models) the variable is all zeros > 2. (all models) the variable is exactly collinear with > other included regressors > 3. (panel data, fixed effects estimator) the variable > is time-invariant > > If a regressor is dropped when none of these conditions is > satisfied, that's a bug (unless I'm forgetting any other > legitimate conditions, though I don't think so). 4. Logit/probit models, if a perfect predictor is spotted.