Chapter 2: Introduction to Multilevel Models for Organizational Data

Hierarchically organized data are commonplace in a variety of research settings. For example, data on students are nested within classrooms or teachers, teachers are nested within schools, and schools are nested within districts. Alternatively, patients are nested within doctors providing services, who may in turn be nested within clinics or hospitals. In these examples, the independence assumption required for correctly estimating single-level regression models is violated. Independence implies there is no correlation between observations, but in a hierarchical system when observations are collected within multiple clusters or contexts such as classrooms or clinics, observations from within the same cluster tend to share some similarity to each other relative to what would be expected in a non-hierarchical sample. Statistically, if the nested structure of the data were ignored, standard errors for parameter estimates, power to detect effects, and Type I error rates could all be impacted (Donner & Klar, 2000; Julian, 2001; Mo-erbeek, 2004; Murray, 1998; Shadish et al., 2002; Wampold & Serlin, 2000). Multilevel models (MLMs) have been developed to properly account for the lack of independence that occurs with nested data.