[R-sig-eco] Multinomial logist regression

[Michael Denslow]

Hi Steve,

> Hi,
>
> I've been working up a dataset of soils and vegetation and I'd like to run a
> multinomial regression model with this data.  The data includes six possible
> outcomes (vegtation types) and anywhere from 1 to 5 independent predictors
> (nutrient and hydrological properties).  I've been looking at the mlogit
> package, globaltest (bioconductor) package, and glm modeling capabilities.
>
> Using the globaltest package I came up with a solution, but have some
> reservations.
>
> Which package and function is more commonly used for ecological
> investigations of this kind?
>
> How can I develop a plot illustrating the probabilities of each of the six
> response vegetation types for a given concentration of nutrients (the
> independent predictors) ?
>
> Eventually, I will add several predictors to this model. Is there a method
> that I can use that will allow me to use multiple predictors and multiple
> responses ?
>
> If anyone really wants to see the code I'm using, I'll have to get it from
> my office tomorrow.
>
> Thanks
> Steve

I am not sure if I am on the right track here but you might check out
Frank Harrell rms package. It sounds as if the Proportional Odds model
may work for you. There is a paper by Guisan and Harrell that may be
of interest. Also Franks Regression Modeling Strategies book has some
nice examples for plotting such models. I am not sure how often these
methods are used in ecology though.

Guisan, A., and F. E. Harrell. 2000. Ordinal response regression
models in ecology. JOURNAL OF VEGETATION SCIENCE 11:617-626.

Hope this helps,
Michael

-- 
Michael Denslow

I.W. Carpenter Jr. Herbarium [BOON]
Department of Biology
Appalachian State University
Boone, North Carolina U.S.A.
-- AND --
Communications Manager
Southeast Regional Network of Expertise and Collections
sernec.org

36.214177, -81.681480 +/- 3103 meters