A Diagnostic for Cox Regression and General Conditional Likelihoods

Abstract Diagnostics for changes in the maximum likelihood estimate of β due to deletion of single observations in linear and logistic regression models are equivalent to diagnostics obtained from fitting augmented regression models. For the Cox proportional hazards model and in conditional models for arbitrarily matched or stratified data, diagnostics derived from augmented regression models are again useful and easy to compute. These diagnostics are applicable to different formulations for the multiplicative covariable effects (over time or over strata), an exact or approximate treatment of tied failure times or multiple-case matching, and time-dependent covariates.