[R] Where is gam?

[Jari Oksanen]

aprasad at fs.fed.us said:
> Could you please clarify.... there has been a lot of interest in gam
> recently in predicting species distribution and environmental
> management (there is in fact a workshop scheduled in Switzerland next
> year on GLM/GAM modelling).

I think that if there is one field where GAMs are "oversold" that must be 
(vegetation) ecology and modelling species responses along environmental 
variables (that we call -- believe or not -- "gradients"). GAMs are used since 
some people are worried about the "correct shape" of species response 
functions, and in particular, against the symmetric Gaussian response 
functions.  In particular they are worried about things like skewness of the 
species responses.  GAMs are used since they seem to impose no constraint in 
the shape and so you can see if your responses *really* are symmetric of if 
they are skewed, flat or in other interesting shapes.  However, these 
questions about shape are parametric questions and they cannot be answered 
with non-parametric models, but these must be subjectively interpreted in 
parametric terms.  It seems that vegetation ecologists are more GAM-fixed that 
they do not even follow the example of Hastie & Tibshirani book where GAMs 
were regularly compared against other models, and quite often, used to select 
a credible parametric model.  The major problem in vegetation ecology is that 
first we use GAM, get a response curve, and then we interpret these shapes in 
parametric terms. So we say, e.g., that we found a skewed response without 
comparing this shape to a symmetric model.  Question about skewness is a 
parametric question (`skewness' *is* a parameter), and so we need nested 
models which differ with this models. We cannot use non-parametric models and 
give them parametric interpretation (`non-parametric' in the sense that the 
parameters there are nothing we could use for interpretation).

In particular in exploratory vegetation ecology where we go out and observe 
and measure things, we can hardly ignore interactions either.

At the moment we are somewhat short of models which could be used for the 
parametric questions of veg.ecologists. I have made some trials with these:

@article{Huisman93,
   author = {J. Huisman and H. Olff and L. F. M. Fresco},
   title = {A hierarchical set of models for species response analysis},
   journal = {Journal of Vegetation Science},
   volume = {4},
   pages = {37-46},
   year = {1993}
} 

It should be easy to fit their models in R, even with ML, although I have done 
this so far only with a program written specifically for these models (because 
I did not know about R when I started).

cheers, Jari Oksanen
-- 
"Infinite variety of Nature, that is pure myth. It is not found in
Nature herself. It resides in the imagination, or fancy, or cultivated
blindness of the man who looks at her." Oscar Wilde


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._