[R-sig-eco] variance partition in vegan

[tyler]

On Tue, May 27, 2008 at 05:30:15PM +0200, Alejandro González wrote:
> I have recently used vegan package and I have shown that variance  
> partition is also implemented. As I prefer to use R rather than CANOCO,  
> I would like to know if results are strictly the same, 

I don't think the results will be strictly the same. The vegan
function varpart provides the unbiased estimators of the variation
partitions, as described by Peres-Neto et al. 2006 (reference below).
I could be wrong, but I think using CANOCO will provide the biased
estimates.

> because in some  
> publications using this technique in CANOCO they only take into account  
> significant variables according to MonteCarlo tests, and to my  
> knowledge), with the vegan  approach R does not do so.

I think the second part of your question is whether or not varpart
incorporates variable selection? It doesn't, but there are several
options available in R. The one closest to what CANOCO does is
probably the packfor package, written by Stephane Dray
(http://biomserv.univ-lyon1.fr/~dray/software.php). This package
incorporates the improvements described by Blanchet et al. 2008
(manuscript at
http://www.bio.umontreal.ca/legendre/reprints/index.html).

So, if the results from vegan differ from Canoco it's because vegan
uses improved techniques. If anyone knows different please let me
know, as I'm spending a lot of time lately working with these
functions!


Cheers,

Tyler

@Article{peres-neto_legendre_ea_2006,

  author = {Pedro R. Peres-Neto and Pierre Legendre and St\'{e}phane
            Dray and Daniel Borcard},

  title = {Variation partitioning of species data matrices: estimation
           and comparison of fractions},

  journal =	 {Ecology},
  year =	 2006,
  volume =	 87,
  number =	 10,
  pages =	 {2614--2625}
}



-- 
Who is your computer working for?

http://www.defectivebydesign.org/faq