Is there an R package for a multinomial repeated measures logistic regression with 3 levels of the dependent variable?

Question

dependent variable: one of three forms of a word-ending independent variables: age, language, word stem with 3 levels random effects: speaker, sentence Each speaker produces many sentences, so it repeated measures is required.

Answer 1

I'm not aware of any. It's unlikely any are going to arise either. The hope is that lme4 will be expanded from handling just simple logistic models to multinomial models. Then you can analyze your within subjects design.

One thing you might try is random coefficients analysis (RCA). This was essentially the main analysis of early FMRI work. You estimate coefficients of the model for each individual subject and then do an analysis across subjects of the coefficients.

Answer 2

You should have a look at MCMCglmm, see for example this blog post:

http://hlplab.wordpress.com/2009/05/07/multinomial-random-effects-models-in-r/

Answer 3

There is the drm package that implements "[l]ikelihood-based marginal regression and association modelling for repeated, or otherwise clustered, categorical responses using dependence ratio as a measure of the association," but I have not tried it.