[R] glm.fit warning message

[Prof Brian Ripley]

It is a warning: it may or may not be serious.

Such a fitted value could well occur if you have a cell in the
cross-classification which has only zero observations.  Then the fitted
value may be zero, and the distribution theory for a Poisson glm is
invalid in that case (as the MLE is on the boundary of the parameter
space, for example).

The problem may be overfitting the data, or it may be a structural zero
(that means that zero did not occur by chance, but would always have
happened).

I suggest you need to read more of the literature on Poisson aka
log-linear regressions to better understand what might be happening.

On Thu, 11 Nov 2004, Davies, Jonathan wrote:

> I am feeling my way in the use of GLM's and have come across a warning whilst manually simplifying a model with interaction terms, removing terms one at a time from the maximum model (R1.9.0).
>
>
>
> > model<-glm(midpnts~(AET+tempave+tempvar+MDE+sqrtarea)^2+Lat,family=poisson,weights=weightS)
>
> > model2<-update(model,~.-tempave:tempvar)
>
> Warning message:
>
> fitted rates numerically 0 occurred in: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart,
>
>
>
> I have had limited success in interpreting this message, further when I use the STEP function [step(model)] the two way interaction term is removed from the model (along with several other interaction terms) and no warning message appears. I saw that there had previously been issue with the glm.fit function (in a much earlier version R 1.4.1), though I am more than ready to believe the problem lies with my data.
>
>
>
> I would appreciate comments on whether the problem is with the data, and if so what is the interpretation of the warning message; finally, as the step function removes the interaction term in any case, can I continue the model simplification procedure from where the step function leaves me?
>
>
>
> Thanks,
>
> Jonathan
>
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
>
>

-- 
Brian D. Ripley,                  ripley at 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