1
$\begingroup$

During my analysis that I perform with a collaborator we get consistently different ouputs in cox regression between R and SPSS. Betas and p-values differ minutely, usually somewhere in the third or fourth decimal, though sometimes more. This occurs both for factors and continuous variables, so it does not appear to be a rounding issue in covariates.

In essence the entire database is prepared in SPSS, which includes calculation of follow-up. The file is then loaded into R via get.spps in HMISC. The follow-up does not consist of whole numbers (for days), could it be that one of the software packages rounds follow-up to whole numbers? Or are the methods calculation different? We use SPSS version 22, and 3.12 with coxph() from the survival package.

Did anyone encounter something similar?

$\endgroup$
3
$\begingroup$

Ah, I found the answer, it has to do with the method with which ties in follow-up duration are evaluated. SPSS uses the Breslow method, whereas R uses the Efron approximation.

$\endgroup$
  • 1
    $\begingroup$ You can specify the choice of tie-handling method in R if you wish. $\endgroup$ – EdM Aug 28 '15 at 14:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.