I am building a cox survival analysis model with a retailer's transaction data. Almost all the variables have failed the proportionality test. Can I continue with the Cox model? Should I build a LIFEREG model instead of the Cox model?

  • 1
    $\begingroup$ LIFEREG is not the name of a model. Don't think of SAS as life. Are the predictors such that they should have a constant effect over time? If they have a diminishing effect over time you might go with an accelerated failure time model such as log-logistic or log-Gaussian. $\endgroup$ – Frank Harrell Jan 29 '14 at 23:25
  • $\begingroup$ Sorry about that.I meant Parametric models. I created time dependent variables ( model variables*log(time)) and these time dependent variables were significant. I even examined Schoenfeld Residuals to confirm and all the variables failed the test. I think I will try Parametric models. I was just concerned that how can all the variables fail the test. $\endgroup$ – Newguy Jan 30 '14 at 18:07
  • $\begingroup$ Make sure you used the time-dependent special likelihood; you can't use add product terms (with time) in the model. I recommend fitting a couple of accelerated failure time models and plotting the Kaplan-Meier estimate of the distribution of the right-censored residuals to check agreement against the assumed distribution. $\endgroup$ – Frank Harrell Jan 30 '14 at 22:53

Your Answer

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

Browse other questions tagged or ask your own question.