[R] multinomial MCMCglmm

Ben Bolker bbolker at gmail.com
Mon Oct 8 03:42:00 CEST 2012

Vaniscotte <vanamelie <at> gmail.com> writes:

> Dear all,
> I would like to add mixed effects in a multinomial model and I am trying
> to use MCMCglmm for that.

> > "ill-conditioned G/R structure: use proper priors if you haven't or 
> rescale data if you have"
> I guess that the problem comes from the nature of my observations which
> are frequencies instead of 0/1 per unit
> Does someone know if a multinomial model fitted with MCMCglmm can handle
> those frequencies table and how to specify the good G/R variance
> structures?

  (1) you should probably post this to r-sig-mixed-models at r-project.org
instead (and try to post only once ...)

  (2) multinomial models generally require *integer* responses (although
not necessarily 0/1); if you don't have a natural "denominator", you may
have to think a bit harder about this (a Dirichlet distribution would be
the natural way to model frequencies that sum to 1)

  (3) have you read the course notes vignette that comes with MCMCglmm?

  Ben Bolker

More information about the R-help mailing list