s-news
[Top] [All Lists]

Re: FW: Weibull questions

To: Chris Wilkinson <chrisw@wam.umd.edu>
Subject: Re: FW: Weibull questions
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Date: Mon, 17 May 2004 20:39:17 +0100 (BST)
Cc: S-News List <s-news@lists.biostat.wustl.edu>
In-reply-to: <004801c43c44$95d455e0$b23e0281@RM3200>
You previously asked about Weibull in S-PLUS and not about survReg.  The
latter fits an accelerated life model, and is documented in the S-PLUS
manuals, V&R and elsewhere.

exp(5.200735) = 181.4, as the coefficient is a mean on - log(T) scale.

1/0.1673101 = 5.98, as `scale' is the standard deviation of the error on 
that scale, 1/alpha in the standard Weibull parametrization.

That's all in V&R (2002, pp.360-1).

You can also use fitdistr, as already suggested to you.


On Mon, 17 May 2004, Chris Wilkinson wrote:

> I got some different results from S+ and an independent method (Statistica)
> for the same data (10 uncensored failure times)
> 
> > survReg(Surv(ttf) ~ 1, data=data1, dist="weibull")
> Call:
> survReg(formula = Surv(ttf) ~ 1, data = data1, dist = "weibull")
> 
> Coefficients:
>  (Intercept) 
>     5.200735
> 
> Scale= 0.1673101 
> 
> Loglik(model)= -48.8   Loglik(intercept only)= -48.8
> n= 10
> 
> whereas Statistica (2-p Weibull) gives scale = 181.4 and shape = 5.98
> 
> Now I am confused as to what parameterization is used in survReg().
> Statistica uses exp(-(t/scale)^shape)) as Spencer suggests S+ does but the
> results suggest survReg() uses something different but what?
> 
> 
> Chris Wilkinson
> 
> -----Original Message-----
> From: s-news-owner@lists.biostat.wustl.edu
> [mailto:s-news-owner@lists.biostat.wustl.edu] On Behalf Of Terry Therneau
> Sent: Sunday, May 16, 2004 9:13 PM
> To: kamiltoth@yahoo.com
> Cc: s-news@lists.biostat.wustl.edu
> Subject: Re: [S] Weibull questions
> 
> 
>  
>   2. How to fit a weibull?
>       Although survReg and censorreg were designed to allow for censored
> data,  they work just fine with ordinary Weibull data.
>  
>       fit <- survReg(Surv(y) ~ x1 + x2 + ...., data=mydata,
> dist='weibull')
>       
>       
>               Terry Therneau
>               
> 
> --------------------------------------------------------------------
> This message was distributed by s-news@lists.biostat.wustl.edu.  To
> unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with the
> BODY of the message:  unsubscribe s-news
> 

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


<Prev in Thread] Current Thread [Next in Thread>