When research is theory-driven, the “proper” research method is the one that best tests one's theory. An often unnoticed byproduct of theory-driven research is the ability to “discover” appropriate data. I illustrate this point with the many instances when I recognized that the data I needed to test a theory were already available in some rather odd places.
A number of disciplines in the social and behavioral sciences address data that are quantitative and hierarchical. Examples include quantitative data on students who attend various schools, psychiatric patients who are treated by different mental health specialists, and workers who are employed by different types of firms. Longitudinal data are also hierarchical in the sense that data on individuals are collected at different time points. Similar to students being nested in schools, observations may be nested within individuals. This paper discusses a set of very useful statistical tools, known as multilevel models, that may be used to examine hierarchical data. It begins with a general description of these models and then provides specific examples that address common social science research issues. It also discusses software that makes these models available for even the novice social statistician.