My question: Assume sampling was was organized as a multi-site cluster randomized trial blocked on teachers (conceptualized as “site” in the model) with students nested in classes (classes represent clusters). Classes were randomized within teachers, with about 30 teachers and 120 classes, each teacher with about 4 classes each, 2 treatment and 2 control. I want to estimate the overall effect of treatment adjusted for teacher effects and a covariate (a pretest). In HLM context, this is pretty straightforward.In translating to MPLUS, I have it as a multilevel model with complex sampling, where %within% is within classes and %between% is between classes and teacher effects are controlled using the clustering command. Does this sound about right?
You're right, it is three levels and cross sectional, but not sure it's fully (or typically) nested, given that we blocked on teachers. So it's students nested in classes (classes represent treatment condition) and classes (or conditions) randomized within teachers. My primary concern with the teacher level is ICC...I think. If possible, I'd like to stay in MPLUS for this part of the analysis (which is one part of a larger analysis that will benefit from MPLUS), but am having some trouble conceptualizing and accounting for the "blocking" featere of the design. Any recommendations welcome.