Reexamining, Reaffirming, and Improving Application of Hierarchical Models

It has now been 14 years since Leigh Burstein's (1980) influential review of the profound inferential difficulties associated with prior quantitative inquiry on schools and classrooms. That far-ranging discussion focused on the mismatch between conventional statistical models and the realities under investigation. Models were single-level, founded on the naive assumption that persons respond independently to educational practices. This atomistic conception of social life, unsatisfying as a conceptual framework to guide inquiry, also produced a host of statistical difficulties: forced choice of unit of analysis; unnecessary trade-offs between misestimated precision and weak power; aggregation bias; and unexamined heterogeneity of regression. Burstein argued that more sensible statistical models were essential to progress in conceptualization, measurement, design, and analysis, and called for a comprehensive reform in statistical practice based on multilevel models. For similar views, see Cronbach and Webb (1975), Cronbach (1976), and Aitkin, Anderson, and Hinde (1981), among others. This special issue of JEBS marks how far the field has advanced in the 14 years since Burstein's review. Motivating the special issue is a sea change in statistical analysis during that time: