I am new to survival analysis, and I am trying to understand how model diagnostics work. Let's say I have a categorical variable with 2 categories 0 and 1 in my Cox Proportional Hazard Model. Does it make sense to plot dfbeta residuals for this covariable? Intuitively, I would say no, but I would like to make sure I am correct about this.

Many thanks!


1 Answer 1


There isn't that much difference in this respect between a Cox regression and other regression models. If you want to see if there are any individual cases with very large contributions to the regression coefficient for that predictor, then examining the dfbetas is a way to go.

You should also consider doing a bootstrap-based evaluation of the entire Cox model, as provided by the validate() and calibrate() functions in the rms package in R. There's a danger that after finding an apparent outlier in the dfbetas you will be tempted simply to remove that case from the analysis. If that case isn't a true outlier but represents an unusual situation in the underlying population, however, that wouldn't be true to your data. In the bootstrap-based evaluation of the model with that case included, you would get an appropriately high standard error for that coefficient.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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