This study did the comparison between four statistics for stopping rule in CAT. In previous studies, the stopping rules usually were investigated the performance of single statistic in the precision for estimate of ability. When we want to compare different statistics for stopping rule, the question is what criterion of cut off for each statistic is fair for comparison. It is impossible to find a common value for each statistic because different statistics have different meaning and unit. In this study, they use the empirical simulation method to find a cut off value appropriate with each statistic. The statistics for the situation that two abilities were zero and the test length equal to 10 in CAT were set as the criterion. Then the four statistics can be compared at the performance in the precision for extreme ability. See weather each statistic can hold the precision well cross all theta level.
1. Because we want to control the precisions of estimate of ability, the criterion should be set from the desirable RMSE or bias. The table 3 showed the stable precisions cross every statistics. It showed this method for setting cutoff value is successful. Of course test length related to precisions one by one. However, we just can set different criterions for different ability level by this empirical method to fixed the precision of each ability level.