[R] Package random forest: can corr.bias be harmful?

[Christian Kampichler]

There exists an option corr.bias for making regression analyses with the 
R package randomForest. The manual warns “Experimental. Use at your own 
risk.” What does corr.bias precisely do and why can it be dangerous to 
use it? As far as I know the bias correction is based on a linear 
regression between the out-of-bag-prediction and the observed values 
which is used for correction of all predictions made by the model, but 
this does not appear so harmful to me.

-- 
Dr. Christian Kampichler
Sovon Dutch Centre for Field Ornithology
PO Box 6521, 6503 GA Nijmegen, The Netherlands
+31 24 7 410 410
www.sovon.nl