A typical rule of thumb has been that the number of covariables should
be less than 10*n. I'm not sure if this has been formally investigated
for survival analysis. An additional issue may be if you have missing
values for some covariables. The usual complete cases algorithm will
reduce the sample size to much smaller than you think. In one study
(regression) although no more than 20% were missing on any one variable,
when doing a regression on 20 variables, the sample size was reduced to
50% of n.
Tony
Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Colleen
Murphy
Sent: Saturday, September 27, 2008 5:05 PM
To: .
Subject: st: Covariate selection for survival analysis
Hi
I have a question regarding selection of covariates for survival
analysis. I have a data set from a cohort study, following a group of
dogs exposed to antimicrobials for the occurrence of resistance-
measuring antimicrobial resistance to specific antimicrobials at 6
discreet time intervals. My concern at the moment is that I have very
small sample sizes in each of the antimicrobial cohort group. The sample
size in my largest group is 39 dogs and my smallest cohort has 9 dogs.
Are there any guidelines in how many covariates I can consider when
doing a survival analysis even at the univariate stage (like there are
for linear and logistic regression). There is a lot data available on
these dogs, but given the very small sample sizes, I feel that I need to
be very prudent in covariate selection for univariate and multivariable
analysis.
Thanks
Colleen
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
