[R] stepwise with F statistics

Frank E Harrell Jr f.harrell at vanderbilt.edu
Tue Oct 13 14:15:48 CEST 2009


Lucas Sevilla García wrote:
> Hi Community R
> 
> I need to make a stepwise using F statistics as a criteria to choose variables.I have 3 independant variables and one dependant variable, and I need to choose the best model fitting to my data using F statistics. The problem is I haven't found any package to do such operation. I have found packages to choose variables using significance level like "stepback" from "maSigPro" package.  However, I haven't found something similar using with F-statistics. I tried grasp package as well but the documentation of this package is not clear. I read in several messages from the community that it's possible to do a stepwise using F statistics as criteria, but they don't specify how to do it, so, I would appreciate if anyone can help me with some indication about any package that could do what I need. Thanks in advance. 
> 
> Lucas
>  		

Why not just postulate a model and fit it?  Why the need for removing 
any of your variables?

Stepwise variable selection requires bootstrapping for validation / 
stability checking.
-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University




More information about the R-help mailing list