[R] Problem with coxme

[Terry Therneau]

------------- Begin Forwarded Message -------------

Date: Wed, 5 Jul 2006 09:09:14 -0500 (CDT)
From: Terry Therneau <therneau at mayo.edu>
Subject: RE: Problem with coxme
To: jhz22 at medschl.cam.ac.uk
Cc: R-help at stat.mat.ethz.ch, liulei at virginia.edu, spencer.graves at pdf.com
Content-MD5: BXKVsHtW/1I9mIUqrXBU0g==

  The original question involved a strange error message from coxme when
trying to fit a random slopes model:

     coxme(Surv(time, status) ~ rx+x2, data=rats2, random=~ (1+x2)|litter)
     
 Synopsis
   	a. coxme does not yet support random slopes models
  	b. it should recognize the request and print a sensible error, instead
 it just dies further down in the routine after it has gotton thoroughly
 confused.
        c. the actual error message is interesting, as an example of just how
 odd the error message from a confused routine can be; but has no other
 use.
 
   Thanks to Jing Zhao who forwarded the message on to me, as I prefer Splus
(blasphemy!) and don't monitor the r-help list,  and also to Spencer
Graves and others who tried to further identify the bug.


		Terry Therneau
		Mayo Clinic
		(author of coxme)

PS		 
  Random slopes are on my "to do" list, and in fact most of the infrastructure
for them is already built into the underlying C code.  Unfortunately it's
been on the list for 3 years now and may stay there until a data set crosses
my own desk that requires it (or there is sufficient clamor from without).
In my defense, I was first sidetracked by trying to understand and support
pdMat structures, something I've since abandoned all hope of.  They are great
examples of object-oriented programming: elegant, short, code resuse, and 
incomprehensible (see www.netfunny.com/rhf/jokes/98/May/stroustrup.html).


------------- End Forwarded Message -------------