Can a survival analysis model with poor model fit still be useful in reporting? I have set up a time-varying survival analysis model (Cox regression in SAS), and have found it to have poor model fit. I tested for goodness of fit using Cox-Snell residuals, and also the Gronnesby and Borgan test. Assuming proportional hazards assumptions hold, and other diagnostics (multicollinearity, confounding, interaction) are being accounted and controlled for, what are the implications of this model? Are there any suggested next steps in either reporting or potentially resolving fit issues?

  • $\begingroup$ It means predictions from your model and estimates will likely be far off from what you would see in reality, but it depends. For example, it may be that your model fits well over a certain range, but then breaks down in another range. In this case, you model may perform quite well under the first range, but do terribly in the second. $\endgroup$ – StatsStudent Mar 14 '16 at 7:23

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.