Thanks for the reply. However I am not entirely sure it works; given that the formulation of a multinomial logit is
y = \beta_{x_i,k_i} \cdot x_i
- I am not 100% sure that we can use the same hyper prior over betas of different output states (in your model \beta_{x1,k0} shares the same hyperprior of \beta_{x1,k1})
- How are you giving information about the panel nature of the data to the model? I am afraid that if you model it this way we are simply losing the “panel” level (i.e. multiple observations per person).