Re: [Gretl-users] help logit, stepwise regression
On Mon, 15 Feb 2016, Tomke Lindena wrote:
after calculating an ordered logit regression I wanted to delete
variables
by using "stepwise regression", but it is not working. "data error"
(Datenfehler!) occurs.
I already had a look for the collinearity and deleted some variables
(collinearity around 3) on my own and tried stepwise regression again, but
still "data error".
Is there a solution for my problem?
I guess you're using the "omit" command with the --auto option.
Looking at the implementation of that in the gretl source code I see
the comment, "FIXME this probably still needs work for estimators
other than OLS." So I'm afraid with ordered logit you've hit a case
that is not handled properly (the routine was getting confused by
the cut-point estimates).
Thanks for your report. This case is now fixed in gretl git and
snapshots. But we'll need to look into this more thoroughly -- there
may be some other estimators that are not handled correctly.
Allin Cottrell