Skip to content
Prev 17248 / 20628 Next

Denominator df in glmer

I can't remember any more where I saw the commentary, but I'm pretty
sure that in Walt Stroup's book he mentions that Kenward-Roger seems to
work pretty well for GLMMs, even though it doesn't have much theoretical
justification (it was developed for LMMs).  However, he also mentions
somewhere that in order to make it work (i.e. to be able to extract the
quantities on which the K-R approximation is based) you have to use a
pseudo-likelihood approach (which lme4 doesn't).

  I agree that the URL Mollie provides should give you useful
information (especially the "Df alternatives" section) ...
On 2018-12-06 5:57 a.m., Mollie Brooks wrote: