what is the reason why the anova() method for coxph objects does not
work when robust standard errors have been requested, e.g., A robust variance estimate is normally used in situations where the model may be mis-specified, e.g., multiple events per subject. In this case a comparison of partial-likelihood values does not make sense, and anova.coxph will refuse to print results (which are a comparison of PL values). This was my decision, with the goal of decreasing misinterpretations or confusions that would arise when anova for a model was identical both with and without the robust option. There is certainly room for disagreement. Opinions welcome. Terry Therneau