ABSTRACT

In longitudinal studies, multiple observations are collected over time on each subject, which is different from cross-sectional studies where a single observation is obtained. The study is said to be balanced if all subjects share a common set of observation times and thus have the same number of measurements. Methods for balanced data such as repeated measures ANOVA, repeated measures MANOVA, and summary measure analysis are not readily applicable to unbalanced data, which occur if the subjects have irregularly spaced observations and/or unequal numbers of measurements. In some studies, unbalanced data are caused by missing observations when some subjects miss one or more intended visits. In other cases, irregular timing of measurements arises when there is random variation of the actual measurement date around the scheduled visit, or the timing is defined relative to a subject-specific benchmark event during follow-up. Both types of unbalanced data can easily be handled using the methods reviewed in this chapter, assuming the observation times are non-informative. In Chapter 4 we discuss methods to analyze unbalanced data when the observation times are informative.