[R] hierarchical partitioning

[vaniscotte]

Dear all,

I am trying to model variation of distribution of species assemblages 
according to environmental variables. For that I use a log linear 
multinomial regression.
In order to select variables that mostly discriminate the assemblages, I 
tried to apply a hierarchical partitioning protocol to my data set.

For that I have adapted the all.regs() for multinomial model.
The problem is that I couldn't adapt partition() as I don't manage to 
access the C code ("hierpart") which is called in this function. This is 
a black box for me.

When I run partition() with the Log Likelihood values obtained for the 
hierarchy of multinomial model, I obtain negative values for Independenr 
contribution.
This suggest that there is a bug somewhere in my protocole.
I suppose that I missused partition(). Maybe my log lokelihood 
(multinomial one) is not appropriate.

As someone had ever met this problem and could help me?

Many thanks

Amélie