Compositional Effects in New Zealand Schools.

This paper describes how compositional effects are estimated from multilevel data, and discusses the way they may be interpreted. Using data from two separate longitudinal studies of New Zealand secondary schools, the paper analyses both school effects and compositional effects as estimated from the two samples. The paper then examines the (sometimes) substantial differences between level-2 estimates and shows how these can be related to: (a) differences in some aspects of methodology; and (b) to the different samples of schools used in each study. This analysis highlights the importance of adequate sampling at the level of the school, since it is the aggregate of all school level estimates within each study that sets the "benchmark" against which each school is individually evaluated. Differences in estimates reported by the two studies are mainly related to differences in their respective samples--their size and their representativeness. [ABSTRACT FROM AUTHOR]