| Computerized Adaptive Testing(CAT) item selection strategy can be divided into two broad categories according to the differences of constraint conditions. One is the item selection strategy with which only statistical constraints are considered. It has better accuracy. The other is the item selection strategy that considers to add the non-statistical constraints( such as content balancing, answer key balancing, exposure control and so on). The Maximum Priority Index Method(MPI) is the typical representative method, which is able to accommodate various non-statistical constraints simultaneously. However, its utilization ratio of item bank is less than50%, leading to a waste of resources. It is shown that a-stratified item selection strategy can effectively improve the utilization ratio of low degree of differentiation program. Therefore, the present study constructed the MPI Method used the mean inequality by combining the MPI Method and a-stratified item selection strategy.The author puts forward four new item selection strategies that based on the arithmetic square root form,which can accommodate non-statistical constraints under the 3PLM in the paper. According to the long end and variable length termination rules to do the Monte Carlo experiment. The results showed that: Firstly, all the new item selection strategies improved a lot on non-utilization ratio. But there are no significant differences between the new method and the existing methods in evaluation indexes such as measurement accuracy and constraint control. Secondly,Thirdly, after attaching discrimination constraints to the new methods and the original methods, the utilization ratio of item bank increases far more than that without using a-stratified strategy. Finally,each item selection strategy in the variable-length CAT works as well as, if not better than, that in the fix-length CAT. |
