[R] Simulating from the Weibull with right censoring

[David Winsemius]

On Mar 23, 2014, at 9:14 PM, Lucy Leigh wrote:

> Hi everyone,
> 
> I am currently attempting to simulate some survival data, based on a Weibull model. I basically need to simulate some survival data
> so that I can then test out a program I'm writing, before applying it to some real data.
> 
> I've managed to create failure time data, using the rweibull function. I.e.
> Y[i] <- rweibull(1, shape, scale)
> For a variety of shape and scale parameters etc.
> 
> I wish to add in right censoring. If I am considering my failure times as deaths, and the censoring time
> as a study cut-off date (for example, 5 years from baseline), would it be correct to do something like,
> 
> for(i in 1:nSubjects){
> if (Y[i] >  5){
> Y[i] <- 5}
> else
> {Y[i]  <- Y[i]
> }}
> 

This would be more R-ish if you did it this way (loop-free)

Y[ Y > 5 ] <- 5  # set cutoff
cens.time <- runif(length(Y), 0, max(Y) )
cens <- cens.time < Y

survreg( Surv(Y, !cens) ~ 1 )

Untested (well, lightly tested)  in absence of data object.

> I guess my question is, is it statistically sound to impose the right censoring after estimating the distribution with no censoring?
> I am leaning towards yes, because in an ideal world where we followed all participants until they died, the
> distribution would be as estimated by the rweibull above....and assuming that the right censoring is independent,
> then cutting of the measurements at a pre-specified time wouldn't affect the outcome at all. But I guess I am
> just looking for a second opinion?
> 
> Thanks in advance, this list has been immensely helpful with some previous questions I have had about
> the survival package.
> Lucy
> 
Please post in plain text.


> 	[[alternative HTML version deleted]]
-- 

David Winsemius
Alameda, CA, USA