[R] truncated poisson in glm / glmmPQL

[Prof Brian Ripley]

The short answer is the the model(s) you want to fit are not glms, so 
neither glm() nor glmmPQL() can be adapted (easily) to fit them.

I think your first task is understand what a glm is.  Then I suggest
specifying precisely what you do want and using maximum-likelihood
estimation (e.g. via optim).

On Tue, 8 Apr 2003, Joern Fischer wrote:

> I'm a postgrad in ecology, and have recently started to use R. I'm planning 
> to model various sets of animal abundance (i.e. count) data in relation to 
> habitat data using glm's and/or glmmPQL's. However, some of my potential 
> response variables have many zeros. From what I gather the "family = ..." 
> option in the command line does not allow for the direct specification of a 
> truncated / zero-modified poisson distribution. My problem is probably not 
> uncommon, but I could not find anything on truncated poisson distributions 
> in the R web pages or documentation.
> 
> Does anyone know of a bit of software that someone may have written to deal 
> with this problem? I'm interested in modifications to the poisson as well 
> as quasipoisson families.

-- 
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