The package provides the function multinomINLA() for fitting multinomial
models in a simple and straightforward way, using the same syntax as e.g.
nnet::multinom().
Since INLA cannot handle multinomial likelihoods, we use the Multinomial-Poisson transformation to reparameterize the model and fit independent Poisson observations. The summaries and marginals for the multinomial probabilities are given in the resulting list object under the names summary.multinomial and marginals.multinomial respectively.
Moreover, the function admits individual observations rather than category counts. This allows using the same syntax as other packages, and more flexible modelling.
- For the moment
multinomINLAfits multinomial probabilities for one single population. In the future, it will allow group-wise multinomial probabilities.
- If you use this package please cite it
citation('multinomINLA')