ABSTRACT
Hierarchical or multilevel models (e.g., Bryk & Raudenbush, 1992; Goldstein, 1987) have become valuable in many research domains as a way to account for naturally occurring or experimentally imposed hierarchical sampling configurations. The classic example in education is the nesting structure that arises when students are sampled from classrooms, which in turn are sampled from schools. Several desirable features are associated with multilevel models for hierarchical designs, including improved efficiency of estimation and more sensitive evaluation of treatment effects.
