Comparative Research On Three Item Selection Approaches In Computerized Adaptive Testing

It is well-known that Fisher information method is one of the classic tools to select items in computerized adaptive testing(CAT).However,it often produces the deviation of the estimation of the ability at the beginning of the test.For this problem,some scholars introduced the Kullback-Leibler(KL)test information,and their studies show that the KL test information method reduces the possibility of the deviation of the ability,and improves the utilization efficiency of the item bank.Further more,it also leads to that the highly discriminating items tend to be overexposed seriously and thus decreases the test effciency.For this problem,a-stratified with b-blocking method is proposed,that is b stratified method.This paper has improved this method.And this improved method can be more balanced and effectively use the pool and improve the test accuracy.The thesis uses Matlab software to generate the item bank and the examinees.We make a preliminary estimation of the examinees' ability according to their responses to the initial item.Finally,we use three methods,i.e.,b stratified,the improved b stratified and KL information this three methods to estimate the ability.The simulation results show that when the testing length consists of 16 items,the estimation of the examinees' ability achieved by the KL information method is the most accurate,and the improved b stratified method is better than b stratified method.When the testing length reaches to 36 items,three methods have no significant difference on the accuracy of estimation,while KL information method is better for the control of exposure rate and makes better use of the item bank,however,the improved b stratified method is better than b stratified method for the control of item exposure rate.