Comparison of exposure controls, item pool characteristics, and population distributions for CAT using the partial credit model
by Lee & Dodd
Sometimes, there is mismatch condition between item pool and ability distribution, for example, normally distributed item pool with negatively skewed distribution. The situation happens when the examinees with high ability level such as gifted students join the CAT. In addition, the effect of various exposure control procedure is under investigation.
The study examined three exposure control procedures (randomesque procedure, progressive-restrited procedure, and maximum information procedure) under various combinations of item pool characteristics and ability distribution based on the partial credit model. The item pool included easy, medium and hard item pool; ability distribution consisted of normally distributed and negatively skewed data. It was found that the medium item pool produced better precision of measurement than both the easy and hard item pools. The progressive-restricted procedure performed better in terms of maximum exposure rates, item average overlap, and pool utilization than both the randomesque procedure and the maximum information procedure. The easy item pool with the negatively skewed data, which is mostly not matched produced the worst performance.
It is an good start point to learn different exposure controls for CAT. It used the IRTGEN SAS macro to generate data and a SAS program for CAT simulation. We can further incorporate other exposure controls procedure such as Sympson-Hetter procedure, alpha-stratified procedure, etc to investigate their effects.