chapter  9
Moving beyond bivariate indicators
The role of multi-level modelling in international comparisons
ByRoel J. Bosker and Tom A.B. Snijders
Pages 12

Like almost all social systems, education is structured hierarchically. Grouping, thus, is a key feature, which in itself may be important to study, but which can also be somewhat troublesome for researchers attempting to study the impacts of education. Two pupils from one class will look more alike than two pupils from different classes because they share the same teacher, the same instructional group and probably many more things such as the same neighbourhood. Conventional statistical models like multiple linear regression, however, assume that observations (or, more precisely, residuals in the model) are sampled independently. Constructing education indicators (possibly with margins of errors attached to these) therefore should be based on models that represent this multi-level nature adequately. This chapter briefly introduces the multi-level statistical model and illustrates how it can be put to use in education research endeavours, such as the calculation of indicators of the performance of education systems using international assessment data.