Longitudinal Data Analysis and Missing Data: Last Observation Stays Put (303117)

Keywords: LOCF, Longitudinal Analysis, Imputation

Longitudinal studies track the same information on the same unit of study at multiple time points in order to measure outcomes over time. These studies are conducted across many industries in many kinds of studies. For example, a medical study collects monthly blood pressures to measure the efficacy of a hypertension drug; an educational study tests quarterly student performance to measure the impact of a new teaching method; and an environmental study counts mosquito larvae over a year to quantify seasonal changes in mosquito populations. Missing data may pose a problem in longitudinal data analysis and is often imputed. One widely used method is last observation carried forward (LOCF) because It is easy to implement and intuitive. The purpose of this presentation is point out the shortcomings of this method and to strongly discourage its use. Robust methods of imputation such as direct likelihood, multiple imputation, and the expectation-maximization algorithm methods will be discussed. While LOCF is easy to understand and easy to implement, that is no excuse to forgo newer and better imputation methods.