. Author manuscript; available in PMC: 2016 Jan 5.

Published in final edited form as: Multivariate Behav Res. 2015 Sep-Oct;50(5):504–519. doi: 10.1080/00273171.2015.1068157

Table 5.

Simulation Study 3, MSE Ratios Comparing FIML Approaches to Item-Level Imputation

			MSE Ratio
Parameter	Items Per Scale	Item-Level Missing Data Rate	All But One Item from Each Scale as Auxiliary Variables	Incomplete Items and Composite of Complete Items as Auxiliary Variables	Half of the Items from Each Scale as Auxiliary Variables
Mean of X	8	15%	1.0000	1.0000	0.9643
	8	25%	0.9655	0.9655	0.8182

	16	15%	1.0000	1.0000	0.8800
	16	25%	0.9565	0.9565	0.6875

Mean of Y	8	15%	0.9600	0.9600	0.8889
	8	25%	1.0000	1.0000	0.7742

	16	15%	1.0000	1.0000	0.8571
	16	25%	0.9600	0.9600	0.6857

Variance of X	8	15%	0.9825	0.9825	0.9180
	8	25%	0.9672	0.9833	0.8116

	16	15%	1.0000	1.0000	0.8824
	16	25%	0.9787	0.9787	0.7541

Variance of Y	8	15%	1.0000	1.0000	0.8871
	8	25%	0.9655	0.9655	0.7746

	16	15%	1.0000	1.0000	0.8621
	16	25%	1.0200	1.0200	0.7500

Covariance	8	15%	0.9459	0.9459	0.8974
	8	25%	0.9737	0.9737	0.8333

	16	15%	1.0000	1.0000	0.8788
	16	25%	1.0000	1.0000	0.7895

Correlation	8	15%	0.9474	0.9474	0.9000
	8	25%	0.9500	0.9500	0.8182

	16	15%	1.0000	1.0000	0.8947
	16	25%	0.9444	1.0000	0.7391

Regression Coefficient	8	15%	0.9500	0.9500	0.9048
	8	25%	0.9524	0.9524	0.7917

	16	15%	1.0000	0.9474	0.8571
	16	25%	1.0000	1.0000	0.7917

Note. The table contains MSE ratios for conditions with a sample size of 500. We can interpret the MSE ratios by saying that the FIML approach is (MSE Ratio × 100)% as efficient as item-level imputation.