The short answer is that a Poisson distribution is a discrete
distribution: if that is appropriate to your data the rpart function (in
the package of that name) has a suitable option.
On Mon, 28 Aug 2006, Solomon Dobrowski wrote:
> Hello all. I have heard over and over that CART and its various tree-like
> brethren are "non-parametric" techniques. When I read the
chapter in
> Chambers and Hastie on tree-based models it states that tree-based models
> can be generalized (GTMs) in a manner similar to GLMs by specifying a
> different deviance function to distributions other than the gaussian error
> distribution ( section 9.4.3). I have an application in which the response
> variable is a continuous variable representing tree counts within a unit
> area and thus would be best described by a poisson distribution. The error
> distribution for this data is not gaussian. If this is the case, will the
> gaussian error distribution used in most regression tree packages, be
> appropriate? Are there ways to specify the error distribution in R or
Should
> I log transform the response variable? If the specification of error
> distribution in regression trees is important, than are these techniques
> truly " non-parametric". Thanks for your inputs.
>
>
> Solomon Dobrowski
> Tahoe Environmental Research Center (TERC)
> John Muir Institute of the Environment
> University of California, Davis
> 530 754 9354
>
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> 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
> and provide commented, minimal, self-contained, reproducible code.
>
--
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