Mixed Process Model: Probit and Linear Regression


#1

Hi there,

I am trying to combine probit and classical linear regression equations in latent variable formulations so that their errors can be correlated in the Seemingly Unrelated Regression way (like in Stata’s --cmp-- command). I am struggling with how to set this up (currently I am explicitly modelling latent variables in a naive way and it doesn’t work well). I found the multivariate probit code by @bgoodri on GitHub and was hoping to tweak it. However, I am not super fluent in manipulating likelihoods, so I was hoping you could point me in the right direction of what I need to do? I checked the user manual to no avail. Any advice you could give me would be greatly appreciated, as always!

Cheers,
Christian


#2

Those data augmentation things tend not to work well in Stan. You may well have written things about as good as they can get.