Re: Sample size for logistic regression with multiple covariates

[Daniel_J_Celta@fpl.com]

Inman,
I would try to think of the problem from a slightly different perspective.
I would try to design an experiment under a factorial framework.
You can think of a n^k design where k = the number of independent variables
and n = the number of levels.
Once there, you could think of doing a full factorial or a fractional
factorial design depending on the interactions you think are most
important.
There is a function in Splus "fac.design" that would help you defining the
experiments in preparation for the analysis.
I hope this helps


Daniel J Celta
Project Valuation - Special Valuations
561 691 7653


                                                                                
                                                              
                      "Inman, Brant A.   M.D."                                  
                                                              
                      <Inman.Brant@mayo.edu>           To:       
"'s-news@lists.biostat.wustl.edu'" <s-news@lists.biostat.wustl.edu>          
                      Sent by:                         cc:                      
                                                              
                      s-news-owner@lists.biosta        Subject:  [S] Sample 
size for logistic regression with multiple covariates             
                      t.wustl.edu                                               
                                                              
                                                                                
                                                              
                                                                                
                                                              
                      08/09/2006 05:18 PM                                       
                                                              
                                                                                
                                                              
                                                                                
                                                              







I am designing a mouse experiment for testing a new treatment for bladder
cancer. The idea is to evaluate treatment effectiveness and treatment by
gene interactions. The variables are:


Outcome (Y): binary (0/1); presence/absence of tumor
Covariate1 (X1):  categorical (0/1/2);  genotype of mouse (XX, X0, 00)
Covariate2 (X2):  binary (0/1); presence/absence of treatment


I am having trouble calculating the sample size for this experiment.  Does
anyone know of an S-Plus or R function that has been written to calculate
sample size for a multiple logistic regression situation?


_________________________________________________
Dr. Brant A. Inman,  MD FRCSC
Urologic Oncology Fellow, Mayo Clinic College of Medicine
200 First Street SW
Rochester, MN   55905
Phone: (507) 284-1330
Fax:    (507) 284-4951
Email:  inman.brant@mayo.edu
_________________________________________________