[R-sig-ME] valid estimates using lme4?

[Ryan King]

> When the reviewer is bemoaning the use of one
> integration point they are not taking into account the fact that the
> approximation is being evaluated at the conditional mode of the random
> effects.


Maybe they are; that approximation can be terrible if there are many
weakly-identified or very correlated-given-data REs.
Expectation propagation and variational-Bayes are (often claimed)
substantially better in that regime; as best I know there is no R
package for those approaches.
A Gelmen's in-prep package "stan" seems like it will include something
along those lines, and there is the glm-ie package for Matlab/octave.
For the practical question,


>It is impossible to determine if SAS, Stata, or SPSS are implementing the steps they claim to implement since the source code is not available. It is >one thing to be able to write out the algebraic expression for solving mixed models, whether using Henderson's mixed model equations (SAS) or any >other approach.

For Mixed at least, I have produced identical output on VC estimates,
FE estimate, RE estimates, likelihood values, and other output I felt
like checking using my own from-scratch supplementation given the same
data for a few design structures. I could probably do the same with
GLIMMIX, though as noted by D Bates GLIMMIX is very restricted in the
models it will fit.


C Ryan King