Re: [S] MLE's with censored data
Fri, 24 Jul 1998 09:28:22 -0400

Perhaps I was a little informal in my use of terms. I have a data set
of lifetimes from a population. I would like to fit the lifetimes to a
lognormal distribution, and get MLE's of the parameters. When using the
parametric survival menu in the gui portion of S+, it seems to require two
data sets. I am not interested in comparing two populations. I only want
distribution parameters estimates for a single data set. Apparently, S+
will only plot the data and give a report if I give it two data sets, not
just one. In the report, it is not clear where the MLE's for the two
distributions are. I tried to give it a dummy regression variable. That
works for getting Kaplan-Meier estimates, but does not work for parametric
survival. It still gives me an error. I have talked to technical support
about this problem. They have been able to duplicate it, but to date, have
not been able to come up with an answer. Also, they suggested I use s-news
to see if anyone out there had a solution.
I would appreciate any input you can give me.

Thanks a lot,

Charlie Whitman

Prof Brian D Ripley <> on
07/23/98 05:40:13 PM

To: Charles Whitman@LEXMARK
Subject: Re: [S] MLE's with censored data

On Thu, 23 Jul 1998 wrote:

> I sent this note out a week ago, but have not gotten any response. Can
> anyone answer the question below?
> I am having a problem trying to get MLE's for a single population of life
> data with some censored observations. It seems the gui portion of S+ 4.5
> requires two populations to do any caluclations at all. After it has
> finished, it is not clear where the parameter estimates are located in
> report screen. If I try to enter only one population, I get an error
> my design matrix is collinear. Does anyone know how get the parameter
> estimates for a given distribution?

Aren't GUI's supposed to be intuitive?

I am afraid that I at least have no idea what you are talking about.
You get MLEs for distributions, not populations. And things needing
two populations are usually testing differences, not estimating.

As far as I know you can use censorReg and its GUI interface: just
have a constant ("1" for the regression part).

Perhaps you need to talk to technical support rather than this list?

Brian D. Ripley,        
Professor of Applied Statistics,
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

----------------------------------------------------------------------- This message was distributed by To unsubscribe send e-mail to with the BODY of the message: unsubscribe s-news