Multilevel models for hierarchically nested data: potential applications in substance abuse prevention research.

Abstract

This chapter reports on an application of a multilevel analysis. A multilevel analysis is a data analysis that uses variables that are measured at different levels of the hierarchy. A hierarchy can have many levels, such as student level, class level, school level, and State or country level, where students are nested within classes, classes are nested within schools or school districts, and school districts can be nested within towns, States, or countries. As soon as one pays attention, hierarchies are present in all data. In large-scale prevention research, researchers usually have information about two or more levels involved, for instance, variables describing individuals (such as achievement, drug use, gender, and measures of socioeconomic status or home environment); variables describing schools (such as school environment, urban versus rural, and type of treatment administered); and perhaps variables describing districts, States, or countries. It is well known that the analysis of variables (i.e., measures at different levels of the hierarchy) on any of these levels separately can be misleading, as will be shown in this chapter. It is more satisfactory to construct a model and technique that simultaneously take information on all levels into account. This chapter introduces such a multilevel model for hierarchically nested data by evaluating the effect of a drug prevention program, Normative Education (NORM), wherein data are collected on students nested within schools. The model is a linear regression model. The difference between this model and the traditional linear regression model is that it takes the intraclass correlation into account and treats variables measured at different levels of the hierarchy in a more appropriate way.