For this final VCRT, we also ordered trials from easiest to most difficult based on the item accuracies produced in the second VCRT version.
Four slices (two T1; two T2) were used as stimuli from each unhealthy patients' scan and so the conspicuity varied (by-item accuracy: M = 79%, 95% CI 72 85%, with three cases at ceiling [i.e. 100%]) dependent upon the location within the scan from which the slice was taken.
Item-by-item accuracy on cycles 2, 3, 4 was significantly predicted by performance on the same items in the previous cycle (1, 2 and 3, respectively) (W = 115.320, P < 0.001).
The per-Turker accuracies were lower than the per-item accuracies owing to a number of poor-performing Turkers who submitted only a handful of judgments (seven Turkers submitted l<10 responses each).
Although I haven't checked every item for accuracy, I have checked several of the listings and found the data to be reliable.
Item recognition accuracy was estimated by the Pr measure of discriminability (p hits) − p(false alarms)).
We can see that the dynamic approach was almost as inexpensive as two Turkers per item, and accuracy was actually higher than the five Turker case (86%).
Item bank accuracy and coverage was satisfactory throughout the entire continuum of mobility, with the exception of the upper level of capacity, suggesting the desirability of replenishing the item bank with items that measure at high mobility function level.
For example, if a physician correctly answered 12 items among a total of 19 items, his accuracy rate of hypertension-related knowledge would be 12/19 = 63.2%.
