[R] under (and over) dispersion in Poisson regression

Prof Brian Ripley ripley at stats.ox.ac.uk
Tue Jan 3 18:03:46 CET 2006


This most often indicates a problem with dispersion estimate.  See the 
cautionary tale in MASS4 chapter 7.  If you have a reliable dispersion
estimate that low for genuine counts, they are either not independent or 
not Poisson (for example, limited), and one would want to find out what is 
going on.

On Tue, 3 Jan 2006, John Sorkin wrote:

> I am trying to use Poisson regression to model count data. My results 
> are suggestive of under dispersion (0.79). How close to one does one 
> want the measure of dispersion to be before one accepts the results of 
> the analysis?
>
> I know that there is no definitive answer to my question, but I would 
> like to get some sense of general practice.

-- 
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




More information about the R-help mailing list