[6313] in s-news-athena
varcomp(method="reml")
daemon@ATHENA.MIT.EDU (Steve Edland)
Mon Jan 30 14:39:04 1995
Date: Mon, 30 Jan 1995 10:52:49 -0800 (PST)
From: Steve Edland <edland@u.washington.edu>
Reply-To: Steve Edland <edland@u.washington.edu>
To: s-news <s-news@utstat.toronto.edu>
A while back (4 Mar 93) Rod Ball posted that the order of the fixed
effects estimates and the order of the columns of the estimated
covariance matrix for the fixed effects as returned by
varcomp(method="reml") do not agree, wreaking havoc, e.g., on estimates
of the variances of contrasts.
I find in at least one example that it is worse that this: in addition
to the different orderings, the columns are _mislabeled_ for the
coefficient estimates and for the covariance matrix. Hence, even if you
account for the reordering, the coefficient estimates, contrasts, and
variances of contrasts are wrong.
I am wondering if anyone working with varcomp has found a fix for this
problem.
Thanks in advance,
Steve Edland
