Please read the Help for predict.glm carefully to make sure you are
not confusing predicted response on the linear scale (log odds) with
that on the probability scale.
The warning is just that: a warning. It means that you have fitted
PROBABILITIES on the boundary, which might compromise the iterative
fitting algorithm and inference thereon. Ergo: examine this carefully
before bithely proceeding.
-- Bert
On Wed, Mar 2, 2011 at 8:10 AM, <patsko at gmx.de>
wrote:> Hi there,
>
> I am encountering a problem with the GLM tool performing logistic
regression. After computing a warning appears, saying ?glm.fit: fitted
probabilities numerically 0 or 1 occurred?. A prediction of new values confirms
the problem as the model does not produce regular probability estimates but
values which are way higher than 1 and lower than 0 in many cases.
> I have tried both methods setting the family=binomial and
family=binomial(?logit?) so this can?t be the reason that causes the error.
>
> As an alternative solution I have considered to resort to the Logistic tool
from the RWeka package. The manual says that it exists for building multinomial
logistic regression models. I can?t image it would be a problem but can anyone
confirm that it indeed is possible to use the algorithm for also computing
binary models?!
>
> Best regards
>
> Patrick
> --
> Schon geh?rt? GMX hat einen genialen Phishing-Filter in die
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
--
Bert Gunter
Genentech Nonclinical Biostatistics