11:32 a.m.
Dear Tomke,
it is hard to say what's going wrong at your specific case without
additional information. Have you used the "omit"-command?
Best,
Artur
Am 15.02.2016 um 17:21 schrieb Tomke Lindena:
Hi,
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?
regards and many thanks
Tomke
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
2:35 p.m.
I'd bin the idea of using stepwise regression, if I were you. It's an
entirely pointless and atheoretical way of fitting a statistical model. As
a primer, look up what Frank Harrell has to say about it.
C
On 15 February 2016 at 16:21, Tomke Lindena <TomkeLindena(a)web.de> wrote:
Hi,
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?
regards and many thanks
Tomke
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
--
Clive Nicholas
"My colleagues in the social sciences talk a great deal about methodology.
I prefer to call it style." -- Freeman J. Dyson
3 p.m.
New subject: [OT] Re: help logit, stepwise regression
Am 15.02.2016 um 20:35 schrieb Clive Nicholas:
I'd bin the idea of using stepwise regression, if I were you.
It's an
entirely pointless and atheoretical way of fitting a statistical model.
As a primer, look up what Frank Harrell has to say about it.
Well, the most theoretical way of specifying a model would be to pin
down the functional form and coefficients a priori, without all those
annoying empirical influences.
What I'm saying is there's always a tradeoff, and it depends on the
situation and context. Easy solutions such as "don't do stepwise" are
almost certainly not the solution, IMHO (but also in the opinion of
others, big figures included).
cheers,
sven
3:53 p.m.
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
3084
days inactive
3085
days old
5 comments
5 participants
participants (5)
-
Allin Cottrell -
Artur T. -
Clive Nicholas -
Sven Schreiber -
Tomke Lindena