1) In figure 3, the condition of the numberof IP size equal to zero should be the ordinary CAT condition. We can expectthe Bias on zero IP size to close to zero. But see the middle figure, the performanceof Bias on zero IP size is almost 0.1. It is a little large in the CAT,especially the test length achieved to 40 items and pool size was large as 500items.
2) insimulation2, the test taker can find the item challenging of 50 % of time whenthe difficulty was 0.5 logit larger than his true ability. This fixedprobability can be changed to be a varied probability. We can try to model theprobability of correctly judging equal to a logit model of (difficulty -ability - 0.5), that is, the more difficulty item lead to larger probability ofputting to PI. if (difficulty - ability) is equal to 0.5, the probability isequal to 50% which is correspond to the previous study (Vispoel et al, 2005).