Hi everybody,
what I'm seeking here is statistical advice. I am
afraid this has little to do with S-plus (although
I used it for my computations).
However, I see from time to time such questions
occuring in this list, and I know that there are
people kind enough to answer them, and besides,
I think that statistics is more interesting than
software (that's a bad reason though).
So, here is my story:
I was asked by a colleague to help her estimate
mean survival times. Her data was special in that
everybody was interval-censored, there was not
a single exact survival time. So I had this idea:
fit a Weibull distribution and compute the mean
from the parameter estimates. I picked up
Weibull because it is often used in that context,
but I had no way to know whether it was a good
choice. However I could not find a better idea.
Then I computed confidence intervals for the
mean by simulation, that is: simulate several
Weibull samples, use the same censoring scheme
as in the study, etc. (I admit I was lazy here, for
I could have used a linearization of the mean
as function of the parameters, together with
the information matrix.)
Now, a paper has been written and submitted for
publication, and a referee asks us for references
about this method. Before I reply that I invented it,
I am forwarding this query to you S+-folks.
Are there any references on the use of Weibull
distribution in that context? or more generally,
on the estimation of the mean when everybody
is censored?
Eric Elguero
Epidemiologie et Prevention
Institut de Recherche pour le Developpement
Montpellier - France
|