Skip to content
Prev 6459 / 20628 Next

How do you report lmer results?

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 07/27/2011 04:01 AM, Luke Duncan wrote:
OK.  Are these binary (sun vs shade) outcomes?

 I
This is fine; you should mention that you previously tried on r-help
(this is a more appropriate list anyway).
Hard to say.  You could try searching google scholar for references to
"lmer" or "lme4", for a start, although there aren't a lot of hits
there.  I would recommend Bolker et al _Trends in Ecology and Evolution_
2008, too (of course); it might also be useful to look in Zuur's book on
mixed models.
One thing that's clear here is that your model is a bit overfitted --
you are getting zero or near-zero variances for both 'indiv' and 'exp'.
 That doesn't necessarily mean anything is wrong, but you might want to
test whether the model performs differently if you drop these random
effects, or whether it gives significantly different results with
different combinations of random effects (again, unlikely but worth
checking).

  It's also not surprising that you get zero variance for Exp, with only
two levels.  The numbers of random-effects levels here are very small --
7 or 8 is probably the *minimum* you could expect to work.
Opinions differ (see <http://glmm.wikidot.com/faq>), but I would suggest
you make Exp a fixed factor ...
I would try drop1() [works with version ...-40 of lme4, I believe] for
more convenient output.

  What you are doing is reasonable, although likelihood ratio test
statistics are probably going to be somewhat liberal (anticonservative)
for "small" sample sizes.  *If* all the predictors were represented
equally in all the random-effect levels (i.e. individuals, zones,
experimental blocks all had similar levels of Tertile, Min/Max temp,
time) then this would be close to a randomized-block design, and in the
classical sense you would have a fairly large residual N (as opposed to
a nested design, where at least some of the treatments don't differ
within blocks).  LRT is probably appropriate with a caveat.  If you are
paranoid, use parametric bootstrapping (see the examples for
?"simulate-mer").
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk4xZjIACgkQc5UpGjwzenM5QACfSfnnneERl3Q4NN936N8i2/cl
TM8AniEQeCArTkuSf18p8qoTepFdEVQC
=Xc5e
-----END PGP SIGNATURE-----