ABSTRACT
Researchers interested in the study of organizations are becoming increasingly aware of the fact that many types of social institutions are hierarchically structured. For example, in the context of business organizations, employees are nested in work units and work units are nested in organizations. In still another example, students are nested in classrooms, which in turn are nested in schools, and so forth. Moreover, data generated from these types of systems are typically obtained through some form of clustered random sampling. Until relatively recently, common approaches to the analysis of hierarchically organized social science data would be to either disaggregate data to the individual (e.g., employee) level or aggregate data to the organizational (e.g., work unit or firm) level. Neither approach is adequate for a proper analysis of the actual structure of the data. In the former case, units within an organization will have the same values on observed and unobserved organizational-level variables. As such, the usual regression assumption of independence of errors is violated—leading to biased regression coefficients. In the latter case, where within-unit data are aggregated and analyzed at the between-unit level, a great deal of information on within-unit variation is lost. The result of aggregation is that relationships among between-unit variables may appear stronger than they really are.
