R help - Dec 2010 - Confidence Intervals for Odds Ratios in multivariate logistic regression

Hi all,

I am trying to fit a logistic regression for a bivariate response using five
independent variables in a stepwise procedure. My outputs look okay but does
any one know (or is there any literature on) how the confidence intervals
are calculated for the reported odds ratios..?

Thanks!

	[[alternative HTML version deleted]]

See McCullagh and Nelder's GLM book for details -- and also probably
V&R's MASS for a concise summary, although I don't have it at hand
and
can't be sure it's there. Really, practically any book on GLM should
have details.

**HOWEVER** You should realize that all these references are "wrong"
in the sense that the intervals are conditioned on the model choice.
AFAIK, the uncertainty due to model choice, which is typically far
larger imo, is not taken into account. In situations like yours, it
**can** be done, e.g. via bootstrapping the whole stepwise procedure.

NOTE:
*** I would appreciate corrections of these statements if I have it wrong ****

-- Bert

On Wed, Dec 8, 2010 at 8:59 AM, S.M. Raghavan <smraghavan at gmail.com>
wrote:
> Hi all,
>
> I am trying to fit a logistic regression for a bivariate response using
five
> independent variables in a stepwise procedure. My outputs look okay but
does
> any one know (or is there any literature on) how the confidence intervals
> are calculated for the reported odds ratios..?
>
> Thanks!
>
> ? ? ? ?[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>


-- 
Bert Gunter
Genentech Nonclinical Biostatistics

On Dec 8, 2010, at 11:59 AM, S.M. Raghavan wrote:
> Hi all,
>
> I am trying to fit a logistic regression for a bivariate response  
> using five
> independent variables in a stepwise procedure. My outputs look okay  
> but does
> any one know (or is there any literature on) how the confidence  
> intervals
> are calculated for the reported odds ratios..?
There is quite a bit of literature that says doing stepwise approaches  
to  LR results in invalid confidence intervals. (There is also quite a  
bit of literature on the mechanics of constructing confidence  
intervals that should be in any of the standard texts on the subject.)

You might want to read some of the material in the archives:
https://stat.ethz.ch/pipermail/r-help/

This thread (identified with RSiteSearch with alternate parameters to  
bring up postings to r-help) seems particularly rich in contributions  
from people with deep knowledge of the problems and potential solutions:

http://finzi.psych.upenn.edu/Rhelp10/2008-October/175481.html

http://finzi.psych.upenn.edu/Rhelp10/2008-October/175541.html

http://finzi.psych.upenn.edu/Rhelp10/2008-October/175518.html

-- 

David Winsemius, MD
West Hartford, CT

On Wed, 8 Dec 2010, S.M. Raghavan wrote:
> Hi all,
>
> I am trying to fit a logistic regression for a bivariate response using
five
> independent variables in a stepwise procedure. My outputs look okay but
does
> any one know (or is there any literature on) how the confidence intervals
> are calculated for the reported odds ratios..?
>
Bert and David have wanred you about the misleading results that 
confidence intervals can give with stepwise procedures.

There are a number of approaches that are not so misleading.

For one such and perhaps some insights into the problems that David and 
Bert were pointing out, try this:

 	install.packages( "BMA" )
 	library( BMA )
 	?bic.glm
 	example( bic.glm )

HTH,

Chuck

> Thanks!
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
Charles C. Berry                            Dept of Family/Preventive Medicine
cberry at tajo.ucsd.edu			    UC San Diego
http://famprevmed.ucsd.edu/faculty/cberry/  La Jolla, San Diego 92093-0901