[R] discriminant analysis

[Prof Brian Ripley]

I don't think this is `discriminant analysis' as used in English (the 
French translation has a wider meaning).  I suggest you start with 
multinom (in package nnet, part of the standard R distribution) which does 
fit a log-linear model, and rpart (in package rpart) as a classification 
tree is often useful to select variables in this sort of problem (and with 
49 explanatory variables you are presumably hoping that only a few are 
needed).

On Fri, 23 Jul 2004, Pierre.Olivier.Mazagol wrote:

> Hello.
> I have a data base with 50 qualitative variables and a lot of 
> individuals. I try to estimate the links between one of these variables 
> (landcover) and the 49 others (geomorphology, hydrography...). I want to 
> use a "discriminant analysis on qualitative variables" (as DISQUAL in 
> SPAD) or a " log-linear model ". Which R-Package(s) or other methods can 
> you advise me.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595