R help - Feb 2009 - Diagnostics for single-observation deletion in Cox models

Hi,
 
Storer and Crowley (JASA 1985) presented an approach for approximating the
changes in maximum partial-likelihood parameter estimates for the Cox model
when a single observation is deleted.  Is there an R implementation of this
approach?
 
Any help is greatly appreciated.  Thanks.
 
Best,
Ravi.
 
----------------------------------------------------------------------------
-------

Ravi Varadhan, Ph.D.

Assistant Professor, The Center on Aging and Health

Division of Geriatric Medicine and Gerontology 

Johns Hopkins University

Ph: (410) 502-2619

Fax: (410) 614-9625

Email: rvaradhan@jhmi.edu

Webpage:  http://www.jhsph.edu/agingandhealth/People/Faculty/Varadhan.html

 

----------------------------------------------------------------------------
--------

 

	[[alternative HTML version deleted]]

> Storer and Crowley (JASA 1985) presented an approach for approximating the
> changes in maximum partial-likelihood parameter estimates for the Cox model
> when a single observation is deleted.  Is there an R implementation of this
> approach?
 
  I found that the approximation of Reid and Crepeau (Biometrika 1985) is 
superior.  It is true that it sometimes underestimates the effect of a large 
outlier, but the Storer and Crowley method does much poorer on the smaller 
residuals, sometimes even getting the wrong sign.  The R&C values are
returned
as the 'dfbeta' residuals from a coxph model.  The same residual formula
as R&C
was derived by Cain and Lange (Biometrics 1984) in a different way, and also 
arises in the definition of a robust variance for multiple event models.
  
   
   Terry Therneau