Using multilevel models is not just a matter of statistical adjustment. Illustrations in the educational field

This article examines the core elements of multilevel modeling for the study of hierarchically structured data. It shows how to design experiments in such cases, paying special attention to the notion of minimum detectable effect (MDE). The article also shows how to specify a multilevel model and how to interpret its estimations. It is argued that multilevel models are not just a means of better statistically adjusting the data; they are also a flexible and powerful heuristic tool that encourages researchers to explore issues that were ignored or considered as nuisances by traditional models. Two empirical examples in the educational field are presented: the first one utilizes experimental data on class size reduction and the second one utilizes mathematics acquisition survey data.