Hi, The following code, from Angelo Canty article on line "Resampling
Methods in R: the boot Package, 2002", works fine for Angelo Canty using R
2.6.0 on Windows XP.
It also works for me using R 1.2.1 and S-PLUS 2000 on Windows XP after
installing the S-PLUS bootstrap library, with slight differences in my outputs.
> library(boot)
> library(survival)
> set.seed(12345)
> mel <- melanoma[melanoma$ulcer==1,]
> mel$cens <- 1*(mel$status==1)
> mel.cox <- coxph(Surv(time, status==1)~thickness,
+ data=mel)
> mel.surv <- survfit(mel.cox)
> mel.cens <- survfit(Surv(time-0.001*(status==1),status!=1)~1,
+ data=mel)
> mel.fun <- function(d) {
+ cox <- coxph(Surv(time, status==1)~thickness,
+ data=d)
+ cox$coefficients}
> mel.boot.con <- censboot(mel, mel.fun, R=999, sim="cond",
+ F.surv=mel.surv, G.surv=mel.cens,
+ cox=mel.cox, index=c(1,8))
> mel.boot.con
CONDITIONAL BOOTSTRAP FOR CENSORED DATA
Call:
censboot(data = mel, statistic = mel.fun, R = 999, F.surv = mel.surv,
G.surv = mel.cens, sim = "cond", cox = mel.cox, index = c(1,
8))
Bootstrap Statistics :
original bias std. error
t1* 0.09967665 0.03579701 0.04973614
I want to apply the Fast bootstrap method from Salibian-Barrera and Zamar
(2003) and Salibian-Barrera, M., Van Aels, S. and Willems, G. (2007) to the
previous example, i.e., to produce a confidence interval for the exponent of
the coefficient of tumour thickness in the Melanoma dataset . Moreover, I
want to compare the performance of the Fast bootstrap with that of the
classical bootstrap, which requires of course computing power and time. How I
can adjust the previous code to do what I want. I asked Angelo Canty for
helping me to do this, but he told me that he is afraid that he does not know
anything about the Fast Bootstrap to which I refer. He suspects that one could
force his boot package to do something like this but he is not sure if that
would be possible through censboot or not. Although my problem does seem
interesting for him, he is afraid that he is not currently in a position to
take on any new collaborations. He remains available to answer any
questions about the boot package itself as it is currently written, he just
does not have the time to consider including new elements into the library at
this time. Salibian-Barrera and Zamar ( 2003) have studied in their paper the
problem of estimating the distribution of statistics defined by estimating
equations. In particular, they have considered two cases: robust regression
estimates and quasi-likelihood estimates.Their approach applies in principle to
the wider class of estimates defined by estimating equations. The Fast
bootstrap povides an inference procedure that is notably faster than the
classical bootstrap (where the estimating equations have to be fully solved for
each bootstrap sample). Salibian-Barrera and Zamar simulation studies have
shown that this Fast bootstrap method is more efficient and more robust to
model departures. I think that the paper of Salibian-barrera and Zamar (2003)
is not published till now. Now I do not have an electronic copy of
this paper draft but I have a hard copy. And so I can send you an attachment
including some written sections from this paper "Fast and Stable Bootstrap
Methods for Statistics Defined by Estimating Equations, Salibian-Barrera and
Zamar (2003)". Please tell me If you want to send you this attachment, which
can also includes what I want to do, and what I suggest.
I need the adjusted code bad. I hope you help me. If you can not help me,
please guide me to anyone who can help me. Thank you in advance.
Alyaa Mohammad El-wakf
Assistant Lecturer
Department of Applied Statistics and Insurance
Faculty of Commerce
Mansoura University
Egypt
---------------------------------
[[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.
