Skip to content
Prev 5047 / 20628 Next

Mixed-model polytomous ordered logistic regression ?

Hi Emmanuel,

You can use the clmm() function in package ordinal to fit cumulative
link mixed models which include the proportional odds mixed model that
you mention. The non-mixed equivalent, clm() gives the same answers as
polr() and glm() on your example data.

clmm() currently only allows for one random intercept, but Ken
Knoblauch proposed a polmer() function
(http://finzi.psych.upenn.edu/R-sig-mixed-models/2010q2/003778.html)
that seem to be an extension of glmer() similar to your glm() hack,
but for mixed models. This allows for several random effects in a
glmer() fashion. On the other hand, with clmm() you get a variety of
link functions and extensions to scale and nominal effects for
modelling non-proportional effects, structured thresholds etc.

The idea of reformulating the cumulative link model as a binomial
generalized linear model is good, but I haven't seen it described
before - can I ask you which exercise in what edition of Gelman & Hill
(2007) that mentions this?

Cheers,
Rune
On 30 December 2010 23:27, Emmanuel Charpentier <emm.charpentier at free.fr> wrote: