why is there no the REML option for lme4::glmer

Can anyone explain to me why the REML option is not available when using lme4::glmer?  I believe the recommendation for fitting linear mixed effects models using lmer is to first determine the random effects structure using REML (with the full fixed effect structure), then to conduct LR tests (or model selection) for fixed effects structure using ML, and finally to make inferences from the final model selected fit with REML.  However, when fitting a GLMM in lem4 there is no longer any REML option.  I am confused about why this is and would really appreciate knowing some more of the theory behind it.

Well, it needs to be implemented before it can be made available; it needs to be defined before it can be implemented.

As my recollection goes (must admit that it's been a while), there are only select cases where there's an exact parallel to the Gaussian situation, i.e. a transformation of data which has a distribution depending only on the variance parameters where the model for the conditional distribution given the transformation is a saturated by the mean parameters. Various ideas exist, e.g. modified profile likelihood, but I am not aware that a clear consensus has ever emerged. The famous GLIMMIX SAS macro allowed you to fit mixed models with glm-style data by an algorithm that called PROC MIXED internally, and you could turn on its REML option, but I wonder if anyone fathomed exactly what that implied.

-Peter D.

Eric Stolen




	[[alternative HTML version deleted]]

_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

Peter Dalgaard, Professor
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com