[R] Poisson GLM warning message

[Prof Brian Ripley]

On Wed, 1 Dec 2010, Anna Berthinussen wrote:

> Hi,
>
> I receive the following warning message when I run a poisson GLM in R:
>
> "glm.fit: fitted rates numerically 0 occurred"
>
> The model summary is shown below. The variable 'Species' consists of
> counts of different species ranging from 0 to 4. I suspect this may
> have something to do with the warning message but I'm not sure. Can
> anybody help?

This is a log-linear model, so that is saying that for at least one 
observation the fitted value is very close to zero (and hence the 
linear prediction is near -Inf: a lp around -30 gives small enough 
fitted rates.).

This normally indicates some sort of separation in the data -- see 
below for a suggestion of where to look.

>
> Thank you!
>
> Anna
>
>
> Call:
> glm(formula = Species ~ Dist + Time + Alt + Elev.road + Noise +
>    Cloud + Temp + Wind.sp + Water + Hab + Crossing, family = poisson,
>    data = bats1)
>
> Deviance Residuals:
>      Min         1Q     Median         3Q        Max
> -2.131083  -0.734566   0.005422   0.548956   2.186938
>
> Coefficients:
>              Estimate Std. Error z value Pr(>|z|)
> (Intercept) -7.496e-01  6.642e-01  -1.128  0.25911
> Dist         1.993e-04  1.233e-04   1.617  0.10593
> Time        -4.753e-04  2.114e-03  -0.225  0.82206
> Alt          2.042e-04  1.265e-03   0.161  0.87180
> Elev.road    2.189e-03  1.960e-03   1.117  0.26418
> Noise       -8.126e+01  4.528e+01  -1.795  0.07268 .

What is 'Noise'?  Is a coefficient of -81 reasonable?  Do all obs with
non-zero noise have observation = 0?  That's the usual cause.

> Cloud       -1.059e-03  2.274e-03  -0.466  0.64150
> Temp         6.660e-02  4.051e-02   1.644  0.10017
> Wind.sp     -7.553e-02  2.420e-02  -3.121  0.00180 **
> WaterY      -6.919e-02  1.744e-01  -0.397  0.69162
> Hab          1.200e-01  5.051e-02   2.375  0.01755 *
> CrossingUnd  9.496e-02  1.214e-01   0.782  0.43420
> ---
> Signif. codes:  0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1
>
> (Dispersion parameter for poisson family taken to be 1)
>
>    Null deviance: 253.49  on 239  degrees of freedom
> Residual deviance: 206.59  on 228  degrees of freedom
> AIC: 681.56
>
> Number of Fisher Scoring iterations: 6
>
> ______________________________________________
> 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.

-- 
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 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595