Handling Planned and Unplanned Missing Data in a Longitudinal Study

Abstract

While analyzing data, researchers are often faced with missing values. This is especially common in longitudinal studies in which participants might skip assessments. Unwanted missing data can introduce bias in the results and should thus be handled appropriately. However, researchers can sometimes want to include missing values in their data collection design to reduce its length and cost, a method called “planned missingness.” This paper review the recommended practices for handling both planned and unplanned missing data, with a focus on longitudinal studies. The current guidelines suggest to either use Full Information Maximum Likelihood or Multiple Imputation. Those techniques are illustrated with R code in the context of a longitudinal study with a representative Canadian sample on the psychological impacts of the COVID-19 pandemic