[R] Step wise regression

Frank E Harrell Jr f.harrell at vanderbilt.edu
Tue May 3 17:27:39 CEST 2005

(Ted Harding) wrote:
> On 03-May-05 Frank E Harrell Jr wrote:
>>walmir-rodrigues wrote:
>>>Dear Fellows,
>>>How can I do to proced a step wise regression in R, if it´s possible ?
>>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)
>>:-)   -Frank
> Frank, you are a sneaky subversive! But you have given me a
> technical clue for a project long near to my heart (with
> Andy's important "random parsimony" refinement).

Now I get Andy's comment - missed before that we need to randomly set 
parsimony level then sample without replacement.

> This is to implement software-driven Clinical Trials articles,
> for submission to standard peer-reviewed journals.
> The underlying engine would be the dada-engine. For samples
> of what this can generate, visit
>   http://www.elsewhere.org/cgi-bin/postmodern

What a great site!

> repeatedly (or click on the "To generate another essay, follow
> this link" link you will find just below the article generated;
> repeat ad libitum).
> Also at the foot of the page, you will find links to explanations
> of how it works. See also
>   http://dev.null.org/dadaengine/
> The basis is that articles of certain kinds have a predictable
> structure, and utilise terminology, phrases, sentence-structures
> and substantive elements drawn from typical usage, and are such
> that the real thing is indistiguishable from the results of sampling
> these elements at random, under the control of a rulebased recursive
> transition network automaton.
> Tailor-made for Clinical Trials, methinks. (Mind you, someone
> once did a lot of work setting up the rulebase for the PostModern
> Lit Crit genre which is what you get in the above samples).

Great idea!!


> Best wishes,
> Ted.

Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University

More information about the R-help mailing list