Application of Optimal Designs to Computerized Test

• Main Authors: Shou-Ming Yang, 楊守民
• Other Authors: Hung-Yi Lu
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/70741780828552929312

碩士 === 輔仁大學 === 應用統計學研究所 === 98 === We discuss the influence of different optimal designs to item calibration under two-parameter logistic model assumption. Our result shows that both item discrimination and difficulty are key factors to affect the accuracy of item calibration. Various optimal designs are clearly better than random design in the accuracy of estimated parameters. There is no major difference between various optimal designs in error estimates since they use the same stop criterion. However the difference was reflected in efficiency of item calibration. In short, A-optimal design requires fewer samples and thus can reduce the cost for item calibration.