Abstract

Multilevel models are the standard for analysis of data that are longitudinal (i.e., repeated observations within subjects) or clustered (subjects nested within, e.g., schools, firms, hospitals). These models are essentially multiple regression models that include random effects to account for the influence of subjects or clusters on their nested observations. Standard software in all of the major statistical packages has made it relatively easy for data analysts and researchers to apply these methods, especially for normal outcomes. However, non-normal outcomes are common in many areas of research—for example, outcomes that are binary, ordinal, nominal, or counts. Multilevel models and software for analysis of such non-normal outcomes are also readily available. This entry focuses on the application and interpretation of multilevel models for these different outcome types.