AW: RE: [R] Mass: lda and collinear variables

[ripley@stats.ox.ac.uk]

On Mon, 24 Feb 2003, Till Baumgaertel wrote:

> On one dataset I have 220 cases of physical data (the variables represent
> points on a curve). The Original data set holds about 1800 variables but
> I reduce this to a power of 2 (32, 64, ...) with an approximation method
> so that the representation of one curve (one observation) with say 32 points
> (variables) does not differ too much from the original representation with
> 1800 variables.
> 
> What I was also wondering about: Why does lda tell me about warnings and
> not *errors*?

Because they are not errors, but they probably indicate problems with 
your dataset.  Just occasionally it is right to use highly collinear 
inputs.  It may even be in your case, but it sounds as if you need to take 
a more orthogonal representation.


BTW, it is MASS not Mass!

-- 
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