Standard errors for predictions of logistic GLMM

BACKGROUND

Hello, Dr. Bolker. In the R-Sig-ME discussion below, you describe
how to construct crude standard errors for predictions of GLMMS
from lmer and glmmadmb.

https://stat.ethz.ch/pipermail/r-sig-mixed-models/2011q2/016356.html

 You referenced the wiki below in stating that you could simply add
in the among-group variance to the fixed effects variance.

http://glmm.wikidot.com/faq

QUESTION

Is this still the case for a model from a binomial family with a
logit link function?

I prefer not to answer questions like this offline, so I'm cc'ing
back to r-sig-mixed-models.

  To the extent that we are willing to assume that the error structure
on the logit scale is Normally distributed, and that we are not
worried about ignoring the uncertainty in the random effects
parameters (i.e. we are conditioning on the estimates of RE
variance-covariance matrices), if you do this procedure on the link
(logit) scale and inverse-link-transform the resulting confidence
intervals, I would think you would get reasonable answers.

  You might want to do a quick check against the results of bootMer() ...

  As always, corrections/caveats welcome from the rest of the community.

  Ben Bolker