Applied Missing Data Analysis in the Health Sciences

Chapter 5Longitudinal Data Methods

5.1 Overview

In contrast to the cross-sectional studies discussed in Chapter 4, longitudinal studies have the defining feature of repeated measures collected on individuals over time, enabling a direct study of temporal patterns or trajectories. Although both cross-sectional and longitudinal studies can look at differences among individuals in their baseline values (called cohort effects in population studies), only a longitudinal study can look at changes over time within an individual (called aging effects in population studies). Longitudinal data can be collected prospectively, following individuals forward in time, or retrospectively, looking back at historical records. The methods described in this chapter can be used for either data collection method.

We begin this chapter by introducing two data examples in Section 5.2 and carry out some basic descriptive analysis. Section 5.3 reviews the modeling approaches and statistical inferences for longitudinal data without missing values. We then introduce the settings of missing longitudinal data as well as simple methods to deal with missingness in Section 5.4. When only the response variable is subject to monotone missingness (e.g., dropout), Section 5.5 presents the likelihood-based method and Section 5.6 describes the inverse probability weighted generalized estimating equation approaches. Section 5.7 extends the WEE to the situation of intermittent missingness of the outcome. The multiple ...