Longitudinal designs and models for causal inference

Abstract

Longitudinal research designs seek to capitalize on the opportunity to observe changes in outcomes in conjunction with a manifest change in the conditions for social action. The chapter illustrates the key principle with selected applications of longitudinal designs to address either micro- or macro-level research questions in rigorous sociology, and discusses the benefits of using longitudinal models and designs for causal inference. While the latter are well appreciated in the discipline, the chapter aims to sensitize readers to a range of further complications in the implementation of longitudinal designs. Some of these are shared with cross-sectional designs, like the issue of endogenous treatment choice in observational studies. Others are more germane to longitudinal designs, for example the distinction between controlling for initial conditions or subject-specific fixed effects, the implications of rare events and local identification of treatment effects, or issues of effect timing, dynamic treatment regimes and the temporal dynamics implied in particular longitudinal models.