Multilevel regression models: applications in School Psychology

Data collection procedures in the behavioral sciences do not always follow the rules of simple random sampling, an assumption of ordinary least square s regression. This is the case in cluster sampling designs which contain mo re than one type of experimental unit, such as subjects that are nested wit hin classes, schools or companies. Failure of taking into consideration the special structure of the data results in estimation bias for the standard error and in an increase in the probability of inference errors. Multilevel analysis models this relationship among the observations while providing u nbiased standard error and estimates of the contextual variability of regre ssion coefficients. The purpose of this paper is to simplify multilevel ana lysis as a generalization of the analysis of covariance on the basis of com monly used statistical concepts such as repeated measures designs, fixed an d random effects models. Several application examples in school psychology are reviewed.