Skip to content

stepwise in svyglm???

5 messages · Diana Marcela Martinez Ruiz, Bert Gunter, Frank E Harrell Jr +1 more

#
Stepwise variable selection is invalid in this context.
Frank

Diana Marcela Martinez Ruiz wrote
-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/stepwise-in-svyglm-tp4637679p4637680.html
Sent from the R help mailing list archive at Nabble.com.
#
... which begs the question: In what context is it valid?  ;-)

-- Bert
On Tue, Jul 24, 2012 at 3:00 PM, Frank Harrell <f.harrell at vanderbilt.edu> wrote:

  
    
#
None, except when the bootstrap is used correctly to fully document how well
or poorly the modeling strategy worked and one is not interested in doing
better or hasn't the time to do so.

Cheers,
Frank

Bert Gunter wrote
-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/stepwise-in-svyglm-tp4637679p4637800.html
Sent from the R help mailing list archive at Nabble.com.
#
On Tue, Jul 24, 2012 at 2:48 PM, Diana Marcela Martinez Ruiz
<dianammr06 at hotmail.com> wrote:
If that's actually what you want to do (which as Frank points out, it
probably isn't), you'll have to do it by hand.

The function step() uses AIC. As far as I know, no-one has yet
constructed valid analogues of AIC,BIC,CIC, ... under complex sampling
(Alastair Scott and I are looking into it), so p-values are the only
option, making the process even less useful.

   - thomas