And if you're interested in the Bayesian machinery behind SS-MELSM, take a look at our pre-print, which is in its (hopefully) final review stage at JEBS: osf.io/preprints/ps.... Also check the links to the package

📦 CRAN: cran.r-project.org/package=ivd 💻 GitHub: github.com/consistently...

ivd uses NIMBLE as its backend and only requires that the user be familiar with standard R formulas (e.g., brms). But be aware that it currently uses a lot of RAM, so start with simpler models if you want to test it out.

For example, in a scenario of students nested within schools, our SS-MELSM model can be used to identify which schools show unusually consistent or inconsistent academic achievement.

The core contribution of ivd is to include a spike-and-slab prior on the scale random effects, providing probabilistic evidence for which clustering units deviate from the average variability.

MELSMs are models for joint estimation of means and variances in hierarchical data. Basically, we can estimate both submodels simultaneously during model fitting, taking advantage of the fact that dependencies between location and scale can influence all other estimated parameters.

thank you! bom te ver por aqui