ABSTRACT
Over the past thirty-five years, longitudinal data collection has become a ubiquitous design element in epidemiology, clinical trials, program evaluation, natural history studies, and related areas of biomedical and public health investigation (Diggle, 2002). Longitudinal data are powerful because with them, we can model individual trajectories of growth or can compare trajectories of, say, disease progression or remission between treated (or exposed) versus untreated (unexposed) groups of individuals. Such designs and accompanying methods of analysis are well described in several key texts (Diggle, 2002; Fitzmaurice, 2012). In addition, longitudinal designs permit the study of within-subject covariation of disease response and time-varying predictors, and this feature allows study participants to 262“serve as their own control” when examining time-varying treatments or exposures, thereby opening potential to control confounders that may not be measured or even known (Neuhaus and Kalbfleisch, 1998; Begg and Parides, 2003). In many designs, longitudinal data yield increases in statistical efficiency relative to cross-sectional data, owing to beneficial effects of within-subject correlation of responses over time.
