s-news
[Top] [All Lists]

Re: max number of variables in coxph

To: carol white <wht_crl@yahoo.com>
Subject: Re: max number of variables in coxph
From: Frank E Harrell Jr <f.harrell@vanderbilt.edu>
Date: Sun, 20 May 2007 15:24:28 -0500
Cc: s-news@lists.biostat.wustl.edu
In-reply-to: <614756.35700.qm@web62007.mail.re1.yahoo.com>
References: <614756.35700.qm@web62007.mail.re1.yahoo.com>
User-agent: Thunderbird 1.5.0.10 (X11/20070403)
carol white wrote:
Dear All,
Knowing the number of samples and censored patients in a data set, is there any way to find out how many variables (max) should be used so that a solution for the maximum partial likelihood estimation in cox regression using coxph would exist?

thanks

carol

It's an oversimplification but typically it's safe to have p < e/15 if you want the model to be reliable, where e is the number of events and p is the total number of degrees of freedom ever entertained when model building (p = # variables + dummy variables + nonlinear terms + interaction terms). But just to be able to converge and not have a singularity, p < e.

Frank Harrell


Finding fabulous fares is fun.
Let Yahoo! FareChase search your favorite travel sites <http://farechase.yahoo.com/promo-generic-14795097;_ylc=X3oDMTFtNW45amVpBF9TAzk3NDA3NTg5BF9zAzI3MTk0ODEEcG9zAzEEc2VjA21haWx0YWdsaW5lBHNsawNxMS0wNw-- > to find flight and hotel bargains.


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

<Prev in Thread] Current Thread [Next in Thread>