On Sat, 10 Apr 2004, Prof Brian Ripley wrote:
> Look at package leaps. Note that it applies to columns of the
> design matrix and not terms, so can be nonsensical with categorical
> predictors (let alone interactions of such).
>
A reasonable compromise is to run regsubsets() (in leaps) returning the
best hundred or so models of each size, and then refit each model ensuring
that the terms are treated sensibly. I use this sometimes for variable
screening.
-thomas
______________________________________________
[EMAIL PROTECTED] mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
