These actually aren't terribly different from each other. I suspect
that lmer is slightly closer to the correct answer, because lme
reports a log-likelihood (really -1/2 times the REML criterion) of
49.30021, while lmer reports a REML criterion of -98.8 - slightly
better fit at -R/2 = 49.4.
Ben,
I doubt the very small difference in log likelihood gives much, if any
information about which model is a better fit. Even if we overlook the
limited precision of the estimate of the REML criterion, the difference
is so small as to me of minimal importance.
John
John David Sorkin M.D.,
2 matches
Mail list logo
