coxph.rvar {survival4} | R Documentation |
Let r be the matrix of infinitesimal influence functions, i.e., r <- residuals(fit, type='dbeta'). Then the robust variance is v <- t(r) %*% r. If there are correlated observations, the appropriate rows or r are first summed, and v is based on the reduced r matrix. There is an obvious connection with the ordinary and group jackknife estimates.
coxph.rvar(fit, collapse)
fit |
a coxph object, i.e., the result of fitting a Cox model. |
collapse |
if the original data contained correlated observations, e.g., multiple data rows per subject, then this argument contains the id vector that identifies the subgroups. |
robust.var |
the robust variance estimate. |
rcall |
the call to this function. |
a copy of the input, with two components added
the print and summary methods for coxph recognize and use the robust variance. The global likelihood ratio and score statistics are uneffected, but the global Wald test will now be based on the robust estimator.
data(ovarian) fit <- coxph(Surv(futime, fustat) ~ age + rx +ecog.ps, data=ovarian) fit2 <- coxph.rvar(fit) summary(fit2)