Comparison of methods of handling missing time-invariant covariates in latent growth models under the assumption of missing completely at random

Latent growth models implemented in multilevel models (MLM) or structural equation models (SEM) may be used to analyze longitudinal data with an emphasis on interindividual. and intraindividual differences. The main objective of this study is to compare methods of handling missing time-invariant data under the assumption of missing completely at random. Listwise deletion (LD), mean substitution (MS), the expectation-maximization (EM) algorithm, multiple imputation (MI), and full information maximum likelihood (FIML) are compared via a computer simulation study. The findings show that FIML and LD generally perform best, whereas the standard errors of EM and MI are usually underestimated. The results on MS are generally acceptable except when the percentage of missingness is large. Practical implications and directions for future research are discussed.