We are conducting a variance decomposition using a hierarchical linear random effects Bayesian model to investigate the variance in a DV that is affected by three nested layers. Because the DV is right skewed and zero inflated we have used a Tweedie link function (e.g.). Is this a good approach given our setting?

Sorry this didn’t get answered earlier, @James_Wittington. You’re asking more of a modeling question than a Stan software question, which sometimes goes beyond our list’s expertise.

For example, I don’t know the Tweedie link function! Nevertheless, I would recommend the methodology we tend to recommend when comparing models, which is to try it. Specifically, in this case, I would recommend using posterior predictive checks on quantities of interest to see if you are capturing the relevant aspects of the data for your task or using cross-validation. Both are described in the third section of the Stan User’s Guide and also at much more length in the Bayesian workflow paper.

Thank you so much for your comment and help Bob.

I consolidated the posts and some posterior checks in the following post: Control for right skewed DV leading to well fitting posterior