[R-sig-ME] polychotomous response data

[Gregor Gorjanc]

Daniel Ezra Johnson <danielezrajohnson at ...> writes:
> Can anyone suggest what I might do to analyze data where:
> 
> - response has more than two values (nominal)
> - observations are grouped by subject
> - predictors are both categorical and continuous
> - some predictors are within-subject, some are between-subject
> 
> If it wasn't for the within-subject predictors, I thought this was
> going to be a compositional data analysis problem, so I was looking
> into the "compositions" library (which I found confusing).

Nope, compositions can be used when you measure the composition of a particular
things, say you have a soil and you measure % os sand, % of, ...

> However, treating each subject's data as a single composition, while
> it makes sense, is not going to enable the analysis of within-subject
> predictors.
>
> Is there a way to use lme4 or another library to extend GLMM to a
> multi-category response?

I would first fit a "threshold" model with polr() in MASS. Then I would try to
add "mixed" effects with use of BUGS. Latest Book by Gelman and Hill is a very
good resourse about this topic.