MULTILEVEL INTERACTION MODELS AND THEIR USE IN THE ANALYSIS OF LARGE-SCALE SCHOOL EFFECTIVENESS STUDIES

In the present paper we discuss the merits of a general multilevel approach to model analysis of large scale school effectiveness studies. We briefly review previous research framework which have been used in school effectiveness research and examine critically the assumptions on which they were based and the methodological implications of these assumptions. The general interaction models presented in this paper provide an answer to many of these critiques. Findings from different types of modeling supported the claim that the modeling of school effectiveness studies must be both multilevel and interactive. Achievement is found to be dependent in a very sensitive, non-additive way, on the particular combination of pupil's home background, his or her general ability, teaching style and other teacher characteristics, the class and the school context in which the pupil learns. Comparisons between parsimonious main effect models with parsimonious interactive ones show that although the explanatory power of interaction models might be slightly smaller, the use of interaction models causes changes in the specification of the models that cannot be ignored.