A scale-location adjustment for proportional hazards deviance residuals

Abstract

We show that deviance residuals derived using the proportional hazards assumption (including Cox regression) are not asymptotically standard normal, but that a scale-location adjustment makes them nearly standard normal, even for moderate sample sizes. This adjustment should aid in outlier detection, as it allows a more exact assessment of when a deviance residual is unusually large.

Keywords:

• Cox regression
• outlier detection
• semi-parametric model
• survival data

Additional information

Notes on contributors

De Wayne R. Derryberry