Conversation

Frank Harrell: Here is an easy approach that will yield results only slightly less valid than one actually using the response variable:
x <- data.frame(x1, x2, x3, x4, ..., other potential predictors)
x[ , sample(ncol(x))]
Andy Liaw: Hmm... Shouldn't that be something like:
x[, sample(ncol(x), ceiling(ncol(x) * runif(1)))]
-- Frank Harrell and Andy Liaw (about alternative strategies for stepwise regression and 'random parsimony')
R-help (May 2005) #Rstats