Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 337 - Approaches for Modeling Clustered and Longitudinal Data
Type: Contributed
Date/Time: Wednesday, August 5, 2020 : 10:00 AM to 2:00 PM
Sponsor: Biometrics Section
Abstract #312913
Title: Exploration of Misspecification in Latent Class Trajectory Analysis and Growth Mixture Modeling: Correlation Structure Matters
Author(s): Megan Neely* and Jane Pendergast and Natasha Dmitreava and Carl Pieper
Companies: Biostatistics & Bioinformatics Dept., Duke University and Duke University Department of Biostatistics and Bioinformatics and Duke University Medical Center and Duke University Department of Biostatistics and Bioinformatics
Keywords: latent class trajectory analysis; growth mixture modeling; proc traj; Mplus ; longitudinal ; error structure
Abstract:

Modeling longitudinal trajectories and identifying latent classes of those trajectories is common in biomedical research. Software to identify latent classes is readily available, leading to increased use of LTCA and GMM. LCTA assumes observations are independent conditional on class membership. However, within-person correlation is often non-negligible and can be accounted for in GMM using group-specific variance models. Despite this important difference in assumptions, we found LCTA (via PROC TRAJ in SAS) is widely used. In this work, we investigated if LCTA is robust to the presence of within-person correlations and how correlation misspecification impacts findings from LCTA and GMM. Using simulation, we varied correlation structures and strength. We found, even in the presence of weak correlation, LCTA performed poorly in class enumeration resulting in biased estimation of class trajectories, incorporation of the correct correlation structure was crucial for GMM in class enumeration, and as expected, under correlation misspecification both LCTA and GMM gave unbiased estimates of class trajectories when the number of classes was correctly specified.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program