[18881] in s-news-athena
[S] coxph
daemon@ATHENA.MIT.EDU (Dr. Steven P Ellis)
Fri Aug 6 14:25:02 1999
Message-Id: <37AB2544.6097@neuron.cpmc.columbia.edu>
Date: Fri, 06 Aug 1999 14:11:16 -0400
From: "Dr. Steven P Ellis" <ellis@neuron.cpmc.columbia.edu>
Mime-Version: 1.0
To: s-news@wubios.wustl.edu
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
On Monday I inquired whether or not there was a bug in coxph which
caused the log likelihoods to be double what they should be. The
question was prompted by simulations I did in which log likelihood
ratios WITHOUT DOUBLING already had the chi squared distribution that 2
times the likelihood ratio should have (under the null hypothesis).
Since then I discovered a subtle error in my simulations which caused
the samples to not satisfy the null hypothesis. Why this error should
cause the log likelihoods to be so precisely doubled (in distribution)
is not clear.
Anyway, more careful simulations produce log likelihood ratios with the
correct distribution.
-- Steve Ellis
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
