How do you model a residual that differs by levels of a grouping variable that is not the clustering variable

Interfaces brms
1

I’m currently modelling a multilevel mediation model. That is, I was wondering if it is possible to model a residual that differs by levels of a grouping variable?

2

Hey @xsasax_13
To me it sounds like this might be what you are looking for:
y ~ 1 + (x1 | group) This gives you a varing slope per group. (1 | group) would give you a varying intercept per group.

3

Yes, this is true. That is, it does not give me a varying residual variance per group. Also, the grouping variable does not necessarily have to be the clustering variable of my model.

4

You can do with within the distributional modeling framework. Here’s a quick example:

bf(y ~ 1 + (x1 |a| group), sigma ~ 1 + (x1 |a| group))

This is also what some have called the MELSM approach. See https://twitter.com/SolomonKurz/status/1337773792588918787