“a-分層”電腦適性測驗之曝光率控管
摘要
In the traditional computerized adaptive testing (CAT) design, items with high a-parameter values tend to be over administered while those with low values may rarely be administered. This problem could be solved by using the a-stratified design proposed by Chang & Ying (1999). In the a-stratified design, an item pool is stratified into several strata based on an ascending order of the a-parameter values; at the early stages of a test, items are administered from the stratum with lower a-parameter values, and those from the stratum with higher a-parameter values are administered during the later stages. Even though the a-stratified design can improve the utilization of the entire item pool, without tight control on item exposure rates, the maximum exposure rate observed from the the a-stratified design may be too high to ensure test security. This study attempted to control the maximum exposure rate by using online exposure control procedures. Results indicated that item exposure can be well controlled by implementing the online procedures in the a-stratified design without sacrificing much the precision of ability estimate. The efficiency of the online procedure remained when item pools or examinee populations have been changed. Thus, the online procedures would be good alternatives for controlling item exposure in the a-stratified CAT.