I think that you might find the following usefull:
http://www.biostat.wustl.edu/archives/html/s-news/2002-11/msg00079.html
I hope this helps.
Best,
Dimitris
----
Dimitris Rizopoulos
Ph.D. Student
Biostatistical Centre
School of Public Health
Catholic University of Leuven
Address: Kapucijnenvoer 35, Leuven, Belgium
Tel: +32/16/336899
Fax: +32/16/337015
Web: http://www.med.kuleuven.be/biostat/
http://www.student.kuleuven.be/~m0390867/dimitris.htm
----- Original Message -----
From: "Robert Kinley" <KINLEY_ROBERT@LILLY.COM>
To: <s-news@lists.biostat.wustl.edu>; <r-help@stat.math.ethz.ch>
Sent: Wednesday, September 14, 2005 3:24 PM
Subject: [S] non-central t : R v.Splus
Hi
For bureaucratic reasons beyond my control I need to rewrite an R
function
(for producing operating characteristic curves) as an Splus function
(
version 6 , windows XP ).
The R function makes extensive use of the fact that the student's t
distribution function pt() has a non-centrality parameter built in
...
sadly that parameter is not present in the Splus pt() function .
However, the Splus f distribution function pf() does have such a
parameter
, so I have tried to write my own non-central version of pt() based
around
pf() , using the relationship between the t and F distributions.
Unfortunately my success has been limited ... I can only get
correct
probabilities for part of the range of the quantile space , failing
when
the quantile becomes small ... and I'm beginning to wonder whether
it's
actually possible to do what I want at all , given that the range of
x in
F(x) is [ 0:Inf ] while that in t(x) is [-Inf , Inf ] , and the
non-central t distribution is not symmetric ...
Do any wiser heads than mine have any experience or advice to offer
...
?
thanks Bob Kinley
Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm
|