Fig. 1
From: Evaluation of approaches for multiple imputation of three-level data
Distribution of the bias in the estimated regression coefficient of interest (β1, true value =  − 0.025) across the 1000 simulated datasets for available case analysis (ACA) and the 8 multiple imputation (MI) approaches under two scenarios for missing data proportions at waves 2, 4 and 6 (10%, 15%, 20% and 20%, 30%, 40%, respectively) and four ICC combinations when data are missing at random (MAR-CATS). The lower and upper margins of the boxes represent the 25th (Q1) and the 75th (Q3) percentiles of the distribution respectively. The whiskers extend to Q1–1.5*(Q3- Q1) at the bottom and Q3 + 1.5*(Q3- Q1) at the top. The following abbreviations are used to denote different MI methods, e.g., DI: dummy indicators, FCS: fully conditional specification, JM: joint modelling