[EE] Estimation with Multilevel Data #72
Description
Description
What I would call a "multilevel model", Boos & Stefanski put under their section titled, "Generalized estimating equations". I think I actually might be using the term "multilevel modeling" more broadly than some people use it; "multilevel modeling" seems to be reserved for random effects, but I was using it for fixed effects/random intercepts. Either way, the objective is to account for the non-independence of observations and covariates within the cluster level.
Recommendation
Replicate section 7.5.6 of Boos & Stefanski to get cluster-robust standard errors in a multilevel data context, focusing on fixed effects/random intercepts (Equation 1). Fixed effects/random intercepts is what I was asking about on Monday 7/21. I can try to find you a dataset that has data with a similar structure and outcome as the dataset described in practice. Really, you could use almost any publicly available trial data and make the argument that regions/states are clusters. I'm attaching my favorite paper so far on multilevel data (delightfully, by a political scientist). Figuring out how to put the covariance matrix together seems challenging, but I'm happy to be an interlocutor if that's helpful.
understanding-choosing-and-unifying-multilevel-and-fixed-effect-approaches.pdf