Research And Implement On C-Stratified CAT With Variable Length

There are three major issues in developing a Computer Adaptive Test: the item selection procedure, the trait estimation, the stopping rule. To date, a popular method is to select an item that maximizes Fisher information at a current estimated ability level. It can increase the efficiency of CAT. However, it also leads to unbalanced item usage within a pool. Highly discriminating items tend to be overexposed which seriously decreases the test security. The other item selection method is a-stratified with content-blocking which stratified the item pool according to a- and b-parameters and the content specifications. It is to select an item that the difficulty of the item is matching the ability of the examinee being measured. This method results in more balanced item usage within a pool and advances the test security. It can be used in high stakes test. In this thesis a CAT based on network is programmed with Visual Basic and SQL Sever. The item selection method used in this CAT is a-stratified with content-blocking. The Modified Multinomial Model is used to balance the content and the stopping rule of variable length is used. The results of the experiment indicate that there is some advantage in this CAT: (1) there is more validity in controlling the item exposure, and the test is more security. (2) the target proportions of content area can be fulfilled for all examinees,(3) the variable length CAT is more efficient than the fixed length CAT.