If the Martingale Residuals indicate that a predictor $X$ violates the proportional hazards assumption but the Schoenfeld Residuals indicate that it does not....which one should you trust?
1
-
$\begingroup$ In reference to @andrea answer, how do you know one violates the assumption and the other does not? $\endgroup$– GlenCommented Apr 13, 2012 at 14:28
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Martingale residuals are used to help determining the best functional form of the covariates included in the model. If you want to assess the PH assumption you should look at the (scaled) Schoenfeld residuals (or you could include time-varying coefficients in your model).
