search

ropensci / dynamite

Bayesian Inference of Complex Panel Data
https://docs.ropensci.org/dynamite/
GNU General Public License v3.0
27 stars 1 forks source link

Support for multiple `random` components? #38

Closed santikka closed 1 year ago

santikka commented 2 years ago

If I understand correctly, it is currently possible to only add a single random component to the model. How does one then specify, for example, correlated random intercepts between two channels, and a separate uncorrelated random intercept for a third channel? Should we try to support this?

helske commented 2 years ago

Yes, you're right, the current random is lacking in the sense that either all or none of the intercepts are correlated. For completeness, I think supporting something like your example would be a good idea, but on the other hand, I'm not sure how realistic/common such a situation would be where you want to allow some of the channels to correlate but you know for a certain that one channel is not correlated with the others?

santikka commented 2 years ago

True, that example was probably not a very realistic scenario, but I could imagine having some intracorrelated channel "groups" that are not correlated with each other.

santikka commented 1 year ago

We now have random() for the model formulas.