Skip to main content
. Author manuscript; available in PMC: 2013 Apr 15.
Published in final edited form as: Neuroimage. 2012 Feb 10;60(3):1843–1855. doi: 10.1016/j.neuroimage.2012.01.123

Figure 3.

Figure 3

The bootstrap simulation demonstrated that available case analysis over-estimated variance while mean replacement over-estimated variance. Imputation error increased for those approaches with higher levels of missingness or smaller sample size. Multiple imputation performed best in the 10-30% missing range, although the regression model appeared to suffer from having too few degrees of freedom when N = 25 and missing = 50%. Multiple imputation preserved variability at the 50% missing point, when N = 49. Neighbor replacement appeared to preserve the original variability of simulated-complete data well, even for smaller sample sizes and high levels of missing data.