[R-sig-ME] gls for generalized linear model

[Hufthammer, Karl Ove]

> What is the rationale for testing the significance of just a random intercept?
> If you think a random intercept is needed because of the design of your
> experiment, then it should be in the model. Regardless the variance of the
> random intercept or its significance.

The paper
'Random effects structure for confirmatory hypothesis testing: Keep it maximal'
http://dx.doi.org/10.1016/j.jml.2012.11.001
has a very nice discussion of this (for linear mixed models, not *generalized*
linear mixed models, but I see no reason why it shouldn't apply equally for GLMMs).

It's also a nice, easy to read introduction to the *concept* of a mixed model,
and to why it's needed.

-- 
Karl Ove Hufthammer