Andrew Johnson

Great to hear!

While not directly related, multivariate probit models will be supported once the Gaussian copula framework from #1317 is finished. It's something that I'll be working on intermittently for the next...

Great! My current plan is to add a `copula` argument to `set_rescor`, like: `set_rescor(rescor = TRUE, copula = "gaussian")`. This keeps the interface consistent across all correlated-outcome models and also...

At the moment it's only hard-coded for gaussian, poisson, bernoulli, and binomial marginals, but is compatible with any distribution that has a CDF defined - so these marginal functions could...

But also, this is more of a proof-of-concept at the moment (I just needed the copulas for a work project), so I'm not committed to any particular implementation

As better reprex, here's an example of the implementation accurately recovering the correlations and generating parameters for a model with gaussian, poisson, and binomial outcomes: ``` r setwd("~/brms") devtools::load_all() #>...

That's a great idea! I'll keep that in mind during the changes so I'm not doing anything too dramatic (or if I have any suggestions!)

> could this structure be used to create a sort of approximation to correlated multinomial outcomes Not an approximation, but exactly that. Gaussian copulas allow you to model the correlation...

It's mostly just a time issue, as development tends to be driven by user demand and developer interest. We have an open issue in the Math library for missing distribution...

Quick progress update for anyone that's interested. I've just about finished configuring the implementation to automatically generate the Stan code for the specified marginal distributions, so that it can be...