Deviance, Priors, and Fitted Values of a brms Multinomial Mixed Model


I’ve been using the point estimates from fixef(mod).

So in order to manually obtain the same fitted values as fitted.brmsfit() , I would take the 4000 posterior samples of logit coefficients, use them to construct 4000 matrices of fitted values (transformed back into probabilities), and finally calculate the average over the 4000 matrices for every observation?


Not 4000 matrices, but one matrix of dimension 4000 (i.e. nsamples) x nobservations.


This is an amazing interchange and learning exercise. Paul what you do for the community is awesome.