The Problem With Multi-Level Models

When you go up the hierarchy into more complicated stuff, I think you see the problem really directly. The building blocks of everything as random effects based really starts to be a house of cards and falls under its own weight when you try to extend it into other areas that are not just schools. You can fit marginal or population-average versions of these things and they converge far more easily because people have population-average questions. It's nearly impossible unless you have 10,000 people but no one has that in this case.