Comparison of Methods of Handling Missing Time-Invariant Covariates in Latent Growth Models Under the Assumption of Missing Completely at Random
通过计算机模拟比较了列表删除、均值替代、期望最大化算法、多重插补和全信息最大似然法在潜增长模型中处理完全随机缺失的时间不变协变量的效果，发现全信息最大似然法和列表删除表现最佳。
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.