From: Bjørn-Helge Mevik <bhs2_at_mevik.net>

Date: Thu 09 Feb 2006 - 01:21:54 EST

Date: Thu 09 Feb 2006 - 01:21:54 EST

Dirk De Becker wrote:

> * Determine the range of the spectrum to be used -> For this, I should

*> be able to calculate the regression coefficients
*

You can get the regression coefficients from a PLSR/PCR with the coef() function. See ?coef.mvr However, using the regression coefficients alone for selecting variables/regions, can be 'dangerous' because the variables are highly correlated.

One alternative is 'variable importance' measures, e.g. VIP (variable importance in projections) as described in Chong, Il-Gyo & Jun, Chi-Hyuck, 2005, Performance of some variable selection methods when multicollinearity is present, Chemometrics and Intelligent Laboratory Systems 78, 103--112. A crude implementation of VIP can be found in http://mevik.net/work/software/pls.html

Another alternative is to use jackknife-estimated uncertainties of the regression coefficients in significance tests. (I don't have any reference or implementation, sorry. :-)

The correlation loadings can also give valuable information about which variables that might be important for the regression. See ?corrplot in the pls package.

-- Bjørn-Helge Mevik ______________________________________________ R-help@stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.htmlReceived on Thu Feb 09 01:40:06 2006

*
This archive was generated by hypermail 2.1.8
: Fri 10 Feb 2006 - 21:35:17 EST
*