Re: st: Residuals of a Cox regression?

[Steven Samuels <sjsamuels@gmail.com>]

Contrary to your impression, Cox regression doesn't have an "error  
term" in the usual linear model sense. It never predicts individual  
(possibly censored) failure times; indeed the time values don't enter  
the estimation procedure at all, just their ranks. I suggest that you  
consult a text on survival analysis to see why (the manual is a bit  
terse).

 If you type "help stcox postestimation", you will see that you can  
calculate three kinds of residuals.  The martingale residual, for  
example, is indeed of the form outcome- predicted.  But it compares,  
at each failure time in the data, the 0/1 event indicator  d  with its  
expectation.

Steve

Steven J. Samuels
sjsamuels@gmail.com
18 Cantine's Island
Saugerties NY 12477
USA
Voice: 845-246-0774
Fax:    206-202-4783


On Dec 9, 2010, at 9:00 AM, Stata Chris wrote:

Dear all,

I would like to plot the "residuals" of a Cox regression, but when I
type -predict resid, residuals- after -stcox- Stata tells me that the
option is not allowed here.
Indeed it is also not clear how I could compute the residuals manually
by subtracting the predicted outcome (which is a hazard ratio) from
the actual outcome, since I'm not sure how to boil the outcome down to
a single number (except maybe the realized duration?).

Yet since Cox regression like any estimation has an error term, it
should be possible to say what this error term is for each
observation?

Thanks a lot for your thoughts on this,
Chris
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/