[R-sig-eco] R-sig-ecology Digest, Vol 9, Issue 9

[Philip Dixon]

Finding a good summary of the X matrix and finding good predictors of Y are 
two very different objectives.

PCA on the X matrix followed by regression works well when you believe that 
the relevant information for predicting Y lies in one of the first few 
eigenvectors.  The relevant information for prediction may be in one of the 
last eigenvectors.

I suggest you consider PLS regression (PLS partial least squares) or the LASSO.
There are R packages for both.

Philip Dixon