cluster defined by environment followed by mrpp

Hi Gabriel,

for your poll: I am principally against, but practically in favour of 
your suggestion.

Here's why:
Grouping (categorising) things inevitably looses information you may 
want to keep when comparing species and environment. Personally, I try 
to follow Frank Harrell's advice (which was on his earlier homepage, now 
sadly scrapped): "Avoid, at all costs, the categorization of continuous 
variables."

Having said that, your question is somewhat incomplete. What happens 
with the clusters? How will they be represented in the next step? 
Through the first axis of a PCA for each cluster? By one of the cluster 
members?
If there was a clear ecological reason to select one variable from a 
cluster over all others then I would be prepared to find clustering a 
useful step (one that I in fact frequently use).

As always, context may dictate the approach.

Cheers,

Carsten