A Multilevel Perspective on the Design and Analysis of Intervention Studies

The level (pupil, classroom or school) at which an educational intervention is assigned affects both the kinds of questions which can be answered in evaluation research, and the statistical methods used to answer them. This paper sets out ways of analysing different kinds of designs using multilevel models. It also considers practical issues such as the method used to allocate interventions, leakage, integrity of delivery, and cost, and how these interact with the more technical issues of model specification. These practical issues are illustrated by two recent British intervention studies. Resume Le niveau - eleve, classe ou ecole - auquel s'adresse une intervention educative affecte tout a la fois le genre de questions qui peuvent trouver reponse dans la recherche evaluative et les methodes statistiques pour y repondre. Cet article presente des voies pour analyser differents types de "design" qui utilisent des modeles a plusieurs niveaux. Il traite aussi de certains aspects pratiques tels que la methode...