Discrete variables: poisson and aggregated binomials

[fusaroli]

Is there a principled reason for discrete variables, such as those generated by poisson and aggregated binomials, not to be included in BGGM?
This might be a dumb question :-) But I've been implementing some GGM-like models in brms, where I can model these outcome and I wanted to better understand how to compare w BGGM

[fusaroli]

following up on this: I couldn't find much about figuring out how well the data are modelled (e.g. posterior predictive checks / residuals checks), which is crucial when trying to use more difficult variables (aka non-gaussian) in BGGM. Am I missing something?

[donaldRwilliams]

@fusaroli I somehow did not receive a notification for this comment. Apologies, as typically I respond quickly.

And you can certainly model discrete variables, or even Poisson, or some combination. I am not really sure how to directly compare to brms, as in BGGM you would be using the rank likelihood or a semi parametric copula. This would be type = "mixed". I do not think brms can do this.

Also, as for predictive checks, I plan to implement these very soon. I agree they are important.

[donaldRwilliams]

Strange that I also did not get a notification in email (just checked).

Can you provide some further details as to what you are trying to do ?

[LindaBetz]

I'm new to BGGM and have follow-up question to that: can I model a mixture of binary and continuous variables when using the setting type = "mixed"? And is it then correct to set mixed_type to 1 for the binary variables in my data set?
Thank you in advance!

[donaldRwilliams]

Yes, you can certainly do that.

And, yes, that is correct. So if you have 10 variables, with the first 5 being binary, then you would have c(1,1,1,1,1,0,0,0,0,0). The zeros denote treat those variables as continuous.