Dear all,
I have a couple of questions regarding the survival:::cch function.
1) I notice that Prentice and Self-Prentice functions are giving identical
standard errors (not by chance but by programming design) while their estimates
are different. My guess is they are both using the standard error form from
Self and Prentice (1986). I see that standard errors for both methods are
asymptotically identical, but in my simulation study I need to distinguish
between two standard errors evaluated at different beta coefficients. My guess
is changing the option iter.max=35 in Prentice function to iter.max=0 should do
the trick. But I wanted to hear from the experts (or the author of the program)
on this issue.
The fact that SE's are identical can be found by the R help example codes of
CCH. I'm copying and pasting them.
subcoh <- nwtco$in.subcohort
selccoh <- with(nwtco, rel==1|subcoh==1)
ccoh.data <- nwtco[selccoh,]
ccoh.data$subcohort <- subcoh[selccoh]
## central-lab histology
ccoh.data$histol <- factor(ccoh.data$histol,labels=c("FH","UH"))
## tumour stage
ccoh.data$stage <- factor(ccoh.data$stage,labels=c("I","II","III","IV"))
ccoh.data$age <- ccoh.data$age/12 # Age in years
cch(Surv(edrel, rel) ~ stage + histol + age, data =ccoh.data, subcoh =
~subcohort, id=~seqno, cohort.size=4028)
cch(Surv(edrel, rel) ~ stage + histol + age, data =ccoh.data, subcoh =
~subcohort, id=~seqno, cohort.size=4028, method="SelfPren")
2) I also notice that Lin-Ying beta estimates are quite different from
Self-Prentice estimates. But Lin and Ying (1993) 's state "the estimating
equation... reduces to the pseduolikelihood score function of Self and
Prentice" and Therneau and Li (1999, Lifetime Data Analysis) state "that [Lin
and Ying's] proposed [beta] estimates ... are identical to those of Self and
Prentice...". My understanding was that the beta estimates should be the same
(or at least very close) and only the variance estimates are supposed to be
different. Can someone shed light on why the beta estimates are different (and
it seems by design) from one another?
This also can be seen by the data in the example codes in R help.
subcoh <- nwtco$in.subcohort
selccoh <- with(nwtco, rel==1|subcoh==1)
ccoh.data <- nwtco[selccoh,]
ccoh.data$subcohort <- subcoh[selccoh]
## central-lab histology
ccoh.data$histol <- factor(ccoh.data$histol,labels=c("FH","UH"))
## tumour stage
ccoh.data$stage <- factor(ccoh.data$stage,labels=c("I","II","III","IV"))
ccoh.data$age <- ccoh.data$age/12 # Age in years
cch(Surv(edrel, rel) ~ stage + histol + age, data =ccoh.data, subcoh =
~subcohort, id=~seqno, cohort.size=4028, method="SelfPren")
cch(Surv(edrel, rel) ~ stage + histol + age, data =ccoh.data, subcoh =
~subcohort, id=~seqno, cohort.size=4028, method="LinYing")
Ryung Kim
Department of Epidemiology and Population Health
Albert Einstein College of Medicine
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
