John Maindonald
2007-Jan-28 11:55 UTC
[R] [R-pkgs] New version of lme4 and new mailing list R-SIG-mixed-models
Dear Douglas -
I have tried several fits that relate to ch 10 of the 2nd edition of
Maindonald & Braun. In two cases a fit that does not currently
converge with lmer does now converge with lme2. In one case where I
took logs in order to get convergence (the data did not really demand
it), I can now get away without taking logs. Basically, none of the
problems that I had to work around when I did the calculations for
that chapter have surfaced. An exercise that had been problematic
now converges without apparent problem.
In one case CPU time was reduced by a factor of around 3. In one
case where I formerly could not get convergence, the reduction (to
termination of the calculations) was by a factor of around 10. The
results seems similar, with a small increase in the loglikelihood
(perhaps ~0.5; I have not checked very carefully) in some cases. If
you want more detail, I will provide it.
It looks a huge improvement. I am sure that R users will be duly
grateful for your efforts.
Regards
John Maindonald.
John Maindonald email: john.maindonald at anu.edu.au
phone : +61 2 (6125)3473 fax : +61 2(6125)5549
Centre for Mathematics & Its Applications, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.
> sessionInfo()
R version 2.4.1 Patched (2007-01-17 r40518)
i386-apple-darwin8.8.1
locale:
C
attached base packages:
[1] "stats" "graphics" "grDevices"
"datasets" "utils"
"methods"
[7] "base"
other attached packages:
DAAG MASS lme4 Matrix lattice
"0.91" "7.2-31" "0.9975-11"
"0.9975-8" "0.14-16"