[wenc]

Stepwise regression is not to be recommended because it's very easy to fool oneself.

http://www.sascommunity.org/mwiki/images/e/e2/NYASUG-2007-Ju...

http://www.barryquinn.com/the-statistical-dangerous-of-stepw...

Shrinkage methods like lasso/elasticnet are less susceptible to these problems.

[graycat]

I agree fully. No doubt Breiman took some steps beyond what Tukey did. My only point was that the question and some answers are old.