[R] question about regression forest

[Jose Cortinas]

Dear All,

My name is José Cortiñas Abrahantes, I am statistician and work at the
university in Belgium. I started working recently with machine learning
techniques and I finding a fascinating field. The reason of my email is to
ask you a question related to regression forest. I am interested to compare
the fit of linear regression, regression trees, bagging trees and regression
forest for the case in which we have only one predictor variable. In all the
articles that I have found related to regression forest they reported the
advantages of the use of a random subsets of predictors used to grow the
tree with respect to bagging, in my case I have only one, thus it is not 
really contributing. I was
expecting then to see a similar behaviour than bagging, the rsquared values
produced by both methods are very similar indeed, but what I find strange is
that if I take the 2.5 and 97.5 percentile of all rsquared from each tree
grow the interval obtained for regression forest is much narrower than the
one obtained for bagging. Do anyone know why is this? Thanks in advance.

Best regards and best wishes for 2007,
José Cortiñas Abrahantes