[R-sig-ME] lme4.0

[Ben Bolker]

Viechtbauer Wolfgang (STAT <wolfgang.viechtbauer at ...> writes:

> 
> Thank you for the feedback. It's good to have a better idea where
> things stand.
> 
> One argument in favor of lme4.0:
> 
> In the metafor package, I use lme4 to fit a particular 
> Poisson/logistic mixed-effects model that involves
> a nonscalar random effects term. In lme4.0, this could be fitted 
> with nAGQ > 1. In lme4 (1.0.x), this is
> (currently) not possible (only nAGQ=1, i.e., Laplace approximation).
> See also:
> 
> https://mailman.stat.ethz.ch/pipermail/r-sig-mixed-models/2013q3/020556.html
> 
> Best,
> Wolfgang

  I agree that this is one of the best arguments for sticking with
lme4.0: I really hope we can implement this in the relatively near future
(unfortunately no guarantees on what "relatively near" means!), but we
need to find someone who has the time and energy to devote to this,
as well as being technically capable of understanding and implementing
multi-dimensional adaptive Gauss-Hermite quadrature within the current
R/reference class/RcppEigen framework ... (For anyone who's interested,
start from https://github.com/lme4/lme4/blob/master/src/external.cpp --
search for "glmerAGQ" ... and compare it with
https://r-forge.r-project.org/scm/viewvc.php/pkg/lme4.0
  /src/lmer.c?view=markup&revision=1725&root=lme4
[broken URL], starting at line 1355, to see how AGQ was implemented
in lme4.0 ...)

  Ben Bolker


  Ben Bolker