Multilevel Linear Models for the Educational System

During the last eight years, interest in assessing the comparative performance of teachers, schools and educational policies has greatly increased in the USA, the UK and other countries. Particularly in the context of declining budgets for education, a serious concern for accountability and effectiveness of the school systerr1 has led to attempts to rate teachers, schools, education authorities and even states using simple SUlTlmaries of administrative data and aggregate scores on standardized tests. These attempts at ratings have aroused anxiety, anger and even legal challenges. Some of the simple methods of comparison or rating used in the past have been seriously defective methodologically, and unwarranted conclusions have been drawn from thenL In this paper I describe a model for a hierarchically structured educational systern_, the analysis of which provides a methodologically sound description of differences among teachers, schools, authorities and states. The implications of this model for educational policy issues are then discussed, and two general conclusions are drawn: i) in general, it is impossible to assess in a methodologically sound way the effect of a change or "intervention" in educational policy on outcome variables from this model, unless the change is embedded in some form of randomized experiment; ii) the assessment of such effects is necessary over an extended period of tirn_e, requiring longitudinal analysis of data frmn the model.