[R] lmekin bug

[Terry Therneau]

Version 2.2-3 of the coxme package has been posted to CRAN. It should 
propogate to the mirrors over the next 1-2 days.

Version 2.2-2 had a serious bug in the lmekin function.  If the variance 
matrix of the random effects was not diagonal the answers were wrong: at 
one spot an upper triangular matrix U was being used when tr(U) was 
called for.  This affects genetic models in particular.  A new test case 
with known solution has now been added as to the package as well.

  There was no issue with any of the other functions, e.g., coxme.

This pointed out to me by Paola Sebastiani just after 2-2 was posted and 
just before I left on a trip.  I really appreciate the bug report, but 
the timing was clearly Murphy's law at work.

Terry Therneau