ABSTRACT
It has been long recognized that missing data are the norm rather than the exception when it comes to the analysis of real data. In this chapter we focus on follow-up studies and on the statistical analysis of longitudinal outcomes with missing data. Due to the wide use of longitudinal studies in many different disciplines, there has been a lot of research in the literature on extensions of selection and pattern-mixture models, which can be considered as the two traditional modeling frameworks for handling missing data (Little and Rubin, 2002; Molenberghs and Kenward, 2007), to the longitudinal setting; see for instance, Verbeke and Molenberghs, (2000), Fitzmaurice et al. (2004), and references therein. These models are applied in non-random missing datasettings, i.e., when the probability of missingness may depend on unobserved longitudinal responses, and require defining the joint distribution of the longitudinal and dropout processes.
