Comparison of Methods for Handling Missingness in Incomplete Longitudinal Data Using Growth Model

• In Event: 67.040 - Missing Data Considerations, Solutions, and Issues

Abstract

This simulation study examined parameter recovery in multi-level growth model under a variety of manipulated conditions including: the number of repeated observations (3 times and, 5 times), missing patterns (missing completely at random_general [MCAR_G], missing completely at random_monotone [MCAR_M] and missing at random_monotone [MAR_M]), missing ratios (10%, 20%, 30%, 40%, 50% and 60%) and sample sizes (250, 500, 1,000 and 2,000). In general full information maximum likelihood (FIML) stably estimated the fixed effects, while multi-level multiple imputation 10(MLMI 10) stably estimated random effect variance components.

Authors

• Seunghiey Baek

• Hyewon Chung, Chungnam National University