Skip to content

poisson with overdispersion

2 messages · Martin Hoyle, Brian Ripley

#
Dear Amanda,
According to "Statistical Computing" by Crawley, p545, you can deal with overdispersion by any one of 3 ways;

1: Carry out significance tests using "F" rather than "Chi" in the anova function for comparison of deviances,
2: Use quasipoisson, and specify a variance function, or,
3: Use negative binomial error (see library(MASS), glm.nb).

To answer the second part of your qu, if you do summary(model), you will get the estimate  and s.e. of the log transform of your  response variable (log is the default link).

Cheers,
Martin.


Martin Hoyle,
School of Life and Environmental Sciences,
University of Nottingham,
University Park,
Nottingham,
NG7 2RD,
UK
Webpage: http://myprofile.cos.com/martinhoyle
I am fitting a poisson model and it appears to have
overdispersion.  I am interested in using the
quasipoisson family (with the glm command).  Will this
account for the overdispersion in the model?  Is there
an additional method for accounting for a dispersion
parameter not equal to 1 (with the glm command)?  

And after fitting the model, how to I obtain the
fitted values and their respective standard errors?

Thank you-
Sarah Watson

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help
#
That's an answer about S-PLUS rather than R.  S-PLUS does not have a
quasipoisson family, and treats its poisson family in an inconsistent way
(sometimes as quasi-Poisson and sometimes not).

For R, use the quasipoisson family or the negative binomial family via
glm.nb.  See section 7.5 of MASS (the book, fourth edition) for fuller
details.
On Thu, 1 May 2003, Martin Hoyle wrote:

            
The model is for the log of the means, not the log of the variable.

You get the fitted values via fitted(fit).  For associated se's you need to
use predict(fit): if you use a quasipoisson model you will automatically
get over-dispersion taken into account.