I really don't agree with Mr. Chessel. My experience is very little and
his one is very great, of course, however...
I tried the variance regression method explained by
Beals (1984).
It is very very efficient in discover gradient in the data.
The utility of such a method could be this:
Do you want to uncover any possible (ecological) gradient in your data ?
Yes.
Then you need a method that will show it to you. COA, PCA etc. often
succed in doing that. Bray-Curtis ordination does it more often (I
believe): if there is a simple linear gradient all the axes would be the
same...
That's all.
There is however a strong motivation to use COA: Reciprocal Scaling. I
don't think that such a capability[1] could be possible in Bray-Curtis
Ordination.
You also loose information on species, well, on how them contribute to
ordination. The computation (proceeding via a distance matrix) is slower
than PCA or COA. However it would be a 32 bit executable, faster than any
code I have seen running on a PC with Win95.
The lacks for software couldn't justify the use and the eventual
developement of a good tecnique (Reciprocal Scaling docet).
However, I'm ready to change my point of view, as always (yes, I'm very
modest)
Thank you all for the feedback.
[1]: to see habitat amplitude and site diversity on the factorial plane
----------------------------------------
Federico Spinazzi
email: federico@syspr03.disat.unimi.it
----------------------------------------
This archive was generated by hypermail 2b30 : Sat Feb 10 2001 - 10:21:49 MET