Parametric modeling of longitudinal data: Introduction and overview
Chapter 1 traced the development of regression models for longitudinal observations, from the simple univariate repeated-measures ANOVA model through the versatile class of linear mixed-eﬀect models for continuous responses. Generalizations of linear models for discrete longitudinal data have followed a number of diﬀerent research directions. Two important extensions are marginal (or population-averaged) models and generalized linear mixed models (GLMMs). Marginal models and the generalized estimating equations approach to inference are the subject of Chapter 3. Generalized linear mixed models are discussed in detail in Chapter 4. Both of these broad classes of models can be considered extensions of generalized linear models for longitudinal data, albeit with somewhat diﬀerent targets of inference. Chapter 7 highlights various aspects of interpretation of regression parameters in these two classes of models, emphasizing their diﬀerent targets of inference and the distinct scientiﬁc questions that can be addressed by each class of models. Although these two classes of models have distinct targets of inference, they have proven to be very useful for analyzing longitudinal data across a wide spectrum of disciplines. In Section 2.2 we review marginal and generalized linear mixed models, emphasizing the main diﬀerences in approach.