ABSTRACT
The multilevel analytical approach has become widely adopted in social science research, especially since software programs such as HLM (Bryk, Raudenbush, & Congdon, 1996), MLn (Rasbash, Yang, Woodhouse, & Goldstein, 1995), and VARCL (Longford, 1990) have become available. The multilevel concept refers to a data structure that is hierarchical; that is, the data are usually collected from units that are nested within another larger social context. This chapter applies a classical multilevel analytical approach to a longitudinal data set from a study of substance abuse prevention. Data were collected both from schools and from students who are nested within schools. Considering; students as the first-level (or microlevel) units, and schools as the second-level (or the macrolevel) units, the data structure contains two levels of hierarchy. Analysis can proceed at either level alone or with both levels considered in some optimal way as implemented via multilevel modeling. Although there have been a number of applications of this methodology to student and school-level data (e.g., Li, Duncan, Duncan, Harmer, & Acock, 1997), our approach may be novel because we resurrect an old methodology that uses both levels in a practical way. And we extend this methodology from ordinary multilevel regression models to more general structural equation models. In this chapter, the term multilevel analytical approach refers to all the statistical techniques developed to analyze multilevel structured data (also see Hox, 1998; Hox, chap. 2, this volume).
