[R-sig-ME] thoughts about Stroup 2014

Ben Bolker bbolker at gmail.com
Thu Mar 6 04:29:55 CET 2014

Kevin Wright <kw.stat at ...> writes:

> I'll add another that just appeared:
> Rethinking the Analysis of Non-Normal Data in Plant and Soil
> Science, Walt Stroup. Published in Agron. J. 106:1-17 (2014) 
> doi:10.2134/agronj2013.0342
> One key takeaway--the linear predictor in linear mixed models may
> not be appropriate as the linear predictor in a GLMM.
> Kevin Wright

  A couple of quick thoughts about this paper:

* I thought it was very well written, and enjoyed reading it.
* It's reassuring that the view from the SAS side of the fence isn't
_too_ horribly different from the way I look at things.
* To my eye, the takeaway about linear predictors is really just
saying that one often needs to incorporate an observation-level
random effect to account for overdispersion in GLMMs (score one for
MCMCglmm here ...)
* GLIMMIX can do correlated G-side random effects -- something that
is contemplated for the 'flexLambda' development branch of lme4, but
it will be a while before we get there.
* On the use of Kenward-Roger for GLMMs: "Although the Kenward–Roger
adjustment was derived for the LMM with
normally distributed data and is an ad hoc procedure for GLMMs
with non-normal data, informal simulation studies consistently
have suggested that the adjustment is accurate."
* I thought it was interesting that Beta regression is included within
the GLMM functionality in SAS -- since Beta distributions aren't in
the exponential family I think of them as (slightly) outside the scope
of GLMMs.
* I would have liked to see Table 2 (summary of analyses) done in
a graphical format.

More information about the R-sig-mixed-models mailing list