> Date: Mon, 31 Jan 2000 15:14:22 +0300
> From: Yimin Ye <yye@kisr.edu.kw>
>
> I just got S-Plus 2000, and us it to fit a GLM model. With only main
> effects, the model produces sensible results, but an interactive effect
> is added, the model produces a extraordinarily large negative deviance
> for the interactive effect, making the model residual deviance much
> larger than the one without the interactive effect. Does this mean
> adding the interactive effect is not sensible or the program has some
> problem, because in most cases, adding another fact will help explain
> more of the variation?
That should always happen, if the fits converged and the models really
are nested. (Could you look to see if they had converged, as hitting
the iteration limit is easy to miss?)
> Does anybody have any experience about this? I thank you in advance and
> look forward to any advises.
This used to happen when a fit produced an essentially perfect fit to
some points (e.g. mean 0 for a 0 Poisson observation), but that has been
fixed for some time. There is a serious error in glm.weights in
S-PLUS 2000 release 1, fixed in release 2.
We would certainly need more details to be able to help, preferably
a reproducible example.
--
Brian D. Ripley, ripley@stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|