校准
费希尔信息
计算机科学
算法
跟踪(心理语言学)
对角线的
数学
计算机化自适应测验
统计
反向
基质(化学分析)
数据挖掘
特征向量
数学优化
估计理论
主对角线
人工智能
计分算法
统计假设检验
停止规则
作者
Hengxin Pang,Yuanyuan Qi,Yinhong He,Yiming Zhao
摘要
Abstract Multidimensional computerized adaptive testing (MCAT) has gained increasing attention, as has online calibration for MCAT. As an essential component of online calibration, the calibration stopping rule determines the conditions for ending calibration and affects its accuracy and efficiency. Current MCAT online calibration studies predominantly employ fixed‐length calibration stopping rules. This study introduces four novel variable‐length calibration stopping rules (designated as D‐rule, E‐rule, T‐rule, and DI‐rule), all derived from the Fisher information matrix of item parameters. The D‐rule and E‐rule quantify calibration information through the determinant and the minimum eigenvalue of the Fisher information matrix, respectively; the T‐rule and DI‐rule quantify item parameter estimation error variances using the trace and the maximum diagonal element of the inverse of the Fisher information matrix, respectively. To obtain appropriate termination thresholds, two methods were employed: the empirical method and calibration reliability. Results showed that the four variable‐length calibration stopping rules could terminate calibration as expected and achieve satisfactory calibration accuracy.
科研通智能强力驱动
Strongly Powered by AbleSci AI