Figure 6.

Percentage deviations of predictions R across replicate calibrations for Approaches 1, 2, and 3 in the CLL data example, plotted versus the number of imputations used ($K=10$, 100, 1000). Results are shown separately for fully observed records (solid dots) and observations containing missing observations (open dots), but always based on models calibrated on the full data. R measures at $K=1$ are 9.6% for fully observed records and 15.3% for missing observations and identical across approaches (hence not shown in above plots)