ABSTRACT
Antedependence models are useful, albeit underutilized, generalizations of wellknown stationary autoregressive models for longitudinal data. Like stationary autoregressive models, antedependence models specify parsimonious parametric forms for the conditional mean and variance of each observation, given all the observations preceding it (as well as any observed covariates) from the same subject. Antedependence models differ, however, by allowing these parametric forms to change over the course of the longitudinal study. This makes them much more flexible than their stationary autoregressive counterparts and hence, as a result, they are often able to fit longitudinal data exhibiting nonstationary characteristics (e.g., increasing variances or same-lag correlations that change over time) quite well. In this introductory chapter, we motivate these models and show where they sit in the broad spectrum of statistical models used for longitudinal data analysis. We begin by describing some important features common to many longitudinal data sets, especially the tendency for observations from the same subject to be correlated. We then briefly review how various classical methods for continuous longitudinal data analysis and a more modern parametric modeling approach deal with these correlations. Next, antedependence models are described very briefly, and an example data set is used to motivate the use of a first-order antedependence model. The remainder of the chapter outlines the scope of the book and describes several longitudinal data sets that will be used throughout the book to illustrate methodology.
