各层分布近似的计算机化自适应测验分层选题策略

Similar Distributed Stratification Method for Computerized Adaptive Testing

选题策略是计算机化自适应测验重要的组成部分,其好坏直接关系到测验的准确性、安全性、效率和测验信度,而分层法又是其中极其重要的一种方法.针对在分层法中按区分度分层(a-STR)和按最大信息量分层(MIS)的曝光率依然较大的缺点,提出了按区分度近似分布分层法(A-SDS)和按最大信息量近似分布分层法(MI-SDS)2种方法.通过Matlab模拟实验表明:在测验精度和效率与原方法接近的情况下,新方法比a-STR和MIS方法较明显地降低了项目的曝光率.

Item selection method is a key part of computerized adaptive testing, it will have a direct impact on accu- racy, safety, efficiency and reliability of the testing. Besides, stratified method is a most important part of item selec- tion method. For the relatively high item exposure rate shortcoming against α-stratified （α-STR）and maximum infor- mation stratification（MIS） method, two kinds of methods are proposed ： similar distributions of discrimination param- eters/maximum information stratification （A-SDS/MI-SDS）. The results of the Matlab simulation study show that the new item selection methods obtain lower average exposure rates than α-STR and MIS methods while maintaining the approximate accuracy and efficiency of testing.