Clustering Trajectories in the Presence of Informative Patterns of Monotone Missingness

• Activity Number: 526
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Gabrielle Flynt*+ and Howard Seltman and Rebecca Nugent and Joel Greenhouse
• Companies: Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University
• Keywords: Trajectory Analysis ; Pattern Mixture Modelss ; Missingness ; Clustering

Abstract

Growth mixture models are a method for analyzing longitudinal data that have been recognized for their usefulness in identifying homogeneous subpopulations within the larger heterogeneous population. Missing data is an inevitable obstacle present in longitudinal studies. Missingness is most often dependent on some unobserved variables and may cause biased estimates in statistical procedures that do not account for the informative missing values. Pattern mixture models are an approach to missing data analysis that models the drop-out mechanism. A common assumption used in pattern mixture models is known as the complete case missing variable restriction which uses fully observed trajectories to make parameter estimates for trajectories with missing values. Attempting to cluster growth trajectories without accounting for informative missingness can easily lead to misclassification. The goal of this work is to combine pattern mixture models and trajectory classification while investigating the validity of the complete case missing variable restriction. Results will be shown for several simulated data sets as well as a data set that measures clinical depression in subjects.