Comparison of Methods for Handling Missing Data When Fitting a Latent Growth Model

• In Event: Methodological Considerations in Modeling Latent Growth

Abstract

In longitudinal studies, missing data is ubiquitous and usually under missing not at random (MNAR) mechanism, which can bias parameter estimates, even distort the study results. This article compares three techniques handling MNAR missing data (i.e. maximum likelihood approach, Diggle-Kenward selection model and pattern-mixture model) through a Monte Carlo simulation study based on a five-wave longitudinal dataset. Estimates of parameters and standard errors using each method are contrasted under four levels of missingness (5%, 10%, 15%, 20%) and four levels of sample size (100, 300, 500, 1000). Results support Diggle-Kenward selection model for parameter estimates, especially under high level of missingness and large sample size. Better standard error estimates are obtained from ML approach under most conditions.

Authors

• Nan Chen, Beijing Normal University

• Hongyun Liu, Beijing Normal University

• Yuan Liu, The Chinese University of Hong Kong

• Huan Liu, Beijing Normal University