[R] function for "two-part" or "two-condition" models

Richard Chandler rchandler at forwild.umass.edu
Mon Aug 22 21:13:41 CEST 2005


Hello list, Andrew, and Richard, 

I wanted to send you a late thanks your for your suggestions and
mention that I recently stumbled upon what I was looking for in the
VGAM package.  If you're interested, see the help file for zapoisson,
which is based on both a binomial and a poisson distribution. 

Richard

Quoting "Remington, Richard" <rnews at kernstat.com>:

> 
> Two alternatives to the zero inflated Poisson (ZIP) model are
> mentioned 
> in Jung, Jhun, and Lee (Biometrics, vol 61, no 2, June 2005,
> p626):
> 
> "Although the ZIP model is more general than the standard Poisson,
> count 
> data with many zeros are often more dispersed than the ZIP model. 
> In 
> this case, the use of a zero-inflated negative binomial (ZINB) 
> distribution or a zero-inflated generalized Poisson distribution is
> a 
> good alternative."
> 
> best,
> Richard
> 
> -- 
> Richard E. Remington
> Statistician
> KERN Statistical Services, Inc.
> PO Box 1046
> Boise, ID 83701
> Tel: 208.426.0113
> KernStat.com
> 
> Andrew Robinson wrote:
> > Hi Richard,
> > 
> > I'm not sure that I can imagine how data can have too many zeros
> to be
> > fit well with zero-inflated Poisson models. Won't the excess
> zeros be
> > accommodated by increasing the the inflation?
> > 
> > In any case, if you want a model that separates the zeros from
> the
> > occurrences before fitting a Poisson model to account for
> variation in
> > abundance then it might be safest to do that split manually.
> > 
> > Another angle to try is to treat it as a special case of a
> finite
> > mixture regression.  I think that some of Jim Lindsey's code will
> fit
> > such models. Google can help you find his wbsite.
> > 
> > An MS student of mine explored these models for regeneration
> modeling.
> > I'd be happy to send you a pdf of his thesis if it would help.
> > 
> > Cheers,
> > 
> > Andrew
> > 
> > On Mon, Jun 27, 2005 at 03:35:30PM -0400, Richard Chandler
> wrote:
> > 
> >>Hello,
> >>
> >>This is an (hopefully) improved question of one I posted several
> weeks
> >>ago. Does anyone know of a function for fitting "two-part"
> models?
> >>These models are designed to handle count data with so many
> zeroes
> >>that they can't be fit well with zero-inflated Poisson models or
> other
> >>'typical' GLMs. My understanding is that they work by first
> fitting a
> >>binomial model to separate the zeros from the occurrences
> (positive
> >>integers) before fitting a Poisson model to account for variation
> in
> >>abundance. 
> >>
> >>I have tried help.search("two-part") and many other similar
> guesses.
> >>
> >>Thanks,
> >>Richard
> >>
> >>-- 
> >>Richard Chandler, M.S. Candidate
> >>Department of Natural Resources Conservation
> >>UMass Amherst
> >>(413)545-1237
> >>
> >>______________________________________________
> >>R-help at stat.math.ethz.ch mailing list
> >>https://stat.ethz.ch/mailman/listinfo/r-help
> >>PLEASE do read the posting guide!
> http://www.R-project.org/posting-guide.html
> > 
> > 
> 
> 


-- 
Richard Chandler
Department of Natural Resources Conservation
UMass Amherst
(413)545-1237




More information about the R-help mailing list