Evaluation of approaches for multiple imputation of three-level data

Table 2 Summary of the imputation approaches for handling incomplete three-level data

MI approach	Paradigm	Model	Software^a	How the two sources of clustering are handled
MI approach	Paradigm	Model	Software^a	Clustering due to higher level clusters	Clustering due to repeated measures
*JM-1L-DI-wide*	JM	Standard (single-level)	SAS [64], SPSS [36], Stata [35], Mplus [24], R [46]	DI	Repeated measures arranged in wide format
*FCS-1L-DI-wide*	FCS	Standard (single-level)	SAS, SPSS, Stata, Mplus, R, Blimp [26]	DI	Repeated measures arranged in wide format
*JM-2L-wide*	JM	Two-level MLMM	SAS [28], Mplus, Realcom-impute [23], Stat-JR [29], R	RE	Repeated measures arranged in wide format
*JM-2L-wide*	JM	Two-level MLMM	SAS [28], Mplus, Realcom-impute [23], Stat-JR [29], R	DI	RE
*FCS-2L-wide*	FCS	Two-level LMM	Mplus, R, Blimp	RE	Repeated measures arranged in wide format
*FCS-2L-DI*	FCS	Two-level LMM	Mplus, R, Blimp	DI	RE
*JM-3L*	JM	Three-level MLMM	Stat-JR, Mplus	RE	RE
*CS-3L*	FCS	Three-level LMM	R, Blimp	RE	RE

DI dummy indicators, FCS fully conditional specification, JM joint modelling, LMM linear mixed model, MLMM multivariate linear mixed model, RE random effects
^aR and Blimp are the only freely available, open-source software implementations

ISSN: 1471-2288