[R] various ordinations

[Jari Oksanen]

camann at babylon.cnrs.humboldt.edu said:
> 1) Are there existing functions for:
>     a) Bray-Curtis (polar) ordination?
>     b) non-metric multidimensional scaling (NMDS)?
>     c) canonical correlation analyses (CCA and DCA)?
>     d) TWINSPAN (doubtful...)? 

There is not (yet:) a vegetation analysis package in R, and from the wording 
above I guess this is what was needed. A few notes:

1a) Would be trivial to write in its basic form (perhaps more complicated with 
enhancements à la Bruce McCune), but is not worth of trying: You should let 
this method rest in peace.

1b) Venables & Ripley (mainly latter, I guess) have  MDS packages in the MASS 
library. Some people claim that isoMDS is a bit too easily trapped into local 
optima, but I am not sure. sammon might be worth trying, since it is more 
local in its configuration whereas isoMDS is the old-fashioned global gradient 
space methods. However, you have to write up your similarity measures, since 
the standard choices in R are not the preferred in vegetation analysis. 
Another problem is that the detection of convergence to global optimum is a 
bit hard work in R, since there are no ready tools for this (although there 
are ready components for these tools).

1c) As you write `CCA', I guess you mean actually canonical *correspondence* 
analysis: Nothing like that is available in R, whereas canonical correlation 
analysis is in `cancor' (package `mva'). This `cancor' is a bit like RDA in 
vegetation ecology jargon, but I am afraid it regards most vegetation data 
sets as ill defined.

1c-d) DCA and TWINSPAN are not available. However, their "BSD style" licence 
would allow incorporation into R. I know of one person in GRASS (GIS) 
development team who is writing a pluggable module of TWINSPAN, and I hope he 
makes it so that it can be incorporated into R with minor modifications. It 
would not be too hard to modify these programs so that you can have them in R, 
but I am now too busy to try this (and the current version of my GNU Fortran 
compiler produces broken shared libraries, and I haven't seen the trouble of 
upgrading yet). I think that basically TWINSPAN and DCA are sick, and 
therefore I haven't given a high priority for these jobs.

So I recommend warmly isoMDS and sammon, and as an alternative to TWINSPAN, 
the numerous cluster analysis methods.

cheers, jari oksanen
-- 
Jari Oksanen - Dept Ecol Env Sci, Univ Helsinki, 15140 Lahti
Ph. +358 3 89220312, Fax -- 89220289, Mobile +358 40 5136529
jari.oksanen at helsinki.fi    http://www.helsinki.fi/~jhoksane



-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
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
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._