Applied Missing Data Analysis in the Health Sciences

Chapter 4Cross-sectional Data Methods

4.1 Overview of General Methods

As the name suggests, cross-sectional data analysis refers to the analysis that looks at the data collected on subjects at one time point, or without regard to time differences (a single-occasion snapshot of the system of variables and constructs), and analyses typically consist of comparing differences between subjects. Even with longitudinal data, cross-sectional analysis is often performed by looking at individual time points. For example, in the IMPACT study described later, investigators randomized patients into two groups and examined depression outcomes at 6 months after baseline to determine if average depression scores differed between the two groups. Many of the missing data methods developed for this setting have been discussed in Chapter 2. Among the recommended approaches that are further described in this chapter, there are four main themes: maximum likelihood approach, Bayesian methods, multiple imputation (MI), and inverse probability weighting (IPW). In addition, we discuss some more advanced techniques such as doubly robust estimators toward the end of the chapter.

4.2 Data Examples

We will illustrate the methods using simulated data as well as data from real-world studies. Three real-world data sets are considered: the NHANES study, the IMPACT study, and the NACC study. The last two examples have already been described in Chapter 1. Next, we briefly describe the simulation study and the ...