R help - Jun 2009 - coxph and robust variance estimation

Hello,

I would like to compare two different models in the framework of Cox
proportional hazards regression models.

On Rsitesearch and google I don't find a clear answer to my question.

My R-Code (R version 2.9.0)

coxph.fit0 <- coxph(y ~ z2_ + cluster(as.factor(keys))+ strata(stratvar_),
    method="breslow" ,robust=T  )
    
coxph.fit1 <- coxph(y ~ z_ + cluster(as.factor(keys))+ strata(stratvar_),
  method="breslow" ,robust=T  )  # marker and covariates

# Analysis of Devaince table
coxph.aov <- anova(coxph.fit0 , coxph.fit1, test="Chisq")

In the single models coxph.fit0 and coxph.fit1 I can use a robust variance
estimation.
It seems that the function anova don't use a robust estimation for the
analysis of deviance.

My question is, how can I use robust estimation for the analysis of deviance?

With kind regards,

Sven Kn?ppel
-- 
GMX FreeDSL Komplettanschluss mit DSL 6.000 Flatrate und Telefonanschluss
f?r nur 17,95 Euro/mtl.!* http://portal.gmx.net/de/go/dsl02

On 6/15/09, "Sven Kn?ppel" <sven.knueppel at gmx.de>
wrote:
>  My question is, how can I use robust estimation for the analysis of
deviance?
>
Did you check the Robust Task View and the packages listed in it?
Perhaps some would do what you need.
Liviu

> It seems that the function anova don't use a robust estimation for the
> analysis of deviance.
  You are correct, anova.coxph always uses the partial likelihood

  This is a serious oversight, particularly since some of the situations in 
which the robust variance is recommended are ones where the LR test is invalid. 
It is also a fairly obvious oversight, but I have to admit that I have never 
thought about it --- it only became obvious to me when I saw your note.
  
  I will add this to my list of things to be repaired.
  
  	Terry Therneau