[R] Better idea than Poisson?

[Prof Brian Ripley]

On Fri, 25 Jan 2002, Patrick Connolly wrote:

>
> My data has three factors, a discrete response and an offset column.
> >From the summary of a glm object derived using three way interactions,
> I get this deviance information:
>
> (Dispersion parameter for poisson family taken to be 1)
>
>     Null deviance: 39244  on 896  degrees of freedom
> Residual deviance: 11913  on 795  degrees of freedom
> AIC: 13905
>
> Number of Fisher Scoring iterations: 6
>
>
> The rather high residual deviance could indicate that the data does
> not fit a poisson distrubution very well.  That's believable since in
> quite a number of combinations of the three factors, zero occurs more
> often that one might expect for a Poisson distribution.
>
>
> Are there suggestions as to what might be a better way to analyse this
> data?

I would at least try a negative binomial (glm.nb in MASS), but as that
seems to be a very large degree of over-dispersion it might not fit much
better.  You could fairly easily write code for a zero-augmented Poisson
(a mixture of a Poisson and always zero) by a direct likelihood
maximization.

-- 
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

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._