[R] Interpreting output of coxph with frailty.gamma

Terry Therneau therneau at mayo.edu
Tue Apr 27 14:36:20 CEST 2010

> (1) Compared to a "normal" coxph model, it seems that I obtain two
> standard errors [se(coef) and se2].

   At the time the code was written there were two prososed estimates
for the se from a penalized model.  I didn't know which was better, so
offered them both.  My best guess was for method 1, so I called it "se"
and based the tests on it by default.

> (2) Again compared to a "normal" coxph model, the z/p-values are
> replaced by] a chi-squared test (Chisq, DF, p).
> What is the reason for this? Does a standard z-test not work once a
> frailty term is included?

  The z-test is a chisquare on 1 degree of freedom.  The frailty terms
don't usually have 1 df, so one can't use the special case.

Terry Therneau

More information about the R-help mailing list