[R] canonical correspondence analysis

Jari Oksanen jarioksa at cc.oulu.fi
Mon Feb 26 15:37:15 CET 2001

>      [Manually approved, since my filter had this looking like spam...
>  MM] 
It was advertisement: no wonder it looked like a spam...

Seriously: there are numerous commercial ordination packages around; PC-Ord, 
Syn-Tax and Canoco being perhaps the most popular. Then there are some lesser 
known and more free packages, such as ADE-4 (http://pbil.univ-lyon1.fr/ADE-4/AD
E-4F.html) for Windows and Mac, and Pierre Legendre's rdacca for Mac, Windows 
and all platforms with gcc(!) -- I cannot get Legendre's URL now, since 
www.umontreal.ca is slow to respond, but you may look at there yourself.  The 
latter does many things that Canoco does, and ADE-4 is even more general (and 
obviously with a nicer GUI, although it doesn't run on my platform, so I don't 
know personally). The classic goodies Decorana and Twinspan are freeware, 
since they were funded by NFS, and can be get from many sources, including my 
homepages.  None of these is R, and so they are not relevant here.  However, 
most things that people do with ordination software, could be done in R, with 
major holes:

* CCA which is the canonical ordination method in community ecology nowadays: 
filling this gap would be really good (this is not my personal opinion, which 
is more sceptical, but the general opinion). I'd glad to see the functions, if 
they are available (and I think the eigenvector, or svd, version should be 
favoured instead of the original power algorithm look-alike).

* Decorana used to be the canonical method before CCA (obviously I use 
`canonical' in a different meaning than ter Braak). The method is sick, but 
the Fortran source code is freely available and could be easily plugged into R 
-- for those who want it.

* The most popular classificaton program in community ecology may be Twinspan 
(Two Way Indicator Species Analysis). It exists just in one incarnation which 
could be plugged into R, although the result object needs some thinking. For 
other clustering methods, R is superiour to many community analysis packages 

* Venables' & Ripley's isoMDS (and perhaps even more interestingly: sammon) 
are already available, but they'd need some extra tools for community 
ecologists. Most importantly, an additional package to `dist' with 
ecologically more relevant distance measures: those that have a good 
rank-order correlation with ecological distances with unimodal species 
responses along ecological axes (`gradients').  By the way, the default 
practice of using metric scaling as the starting configuration in isoMDS is 
almost guaranteed to fall into local optima: I have tested with some data sets 
using random starting configurations and usually managed to find much lower 
stresses than the default (this is in common with most implementations of MDS 
in statistical packages).

* Then we might need an interface for data formats popular in ecology. This 
should be very easy (e.g., `read.CEP').

cheers, jari oksanen

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

More information about the R-help mailing list