The Performance of Diffrernt Estimation Methods under Multidimensional Item Response Theory

• Main Authors: Kuo, Hsiu-Fen, 郭秀芬
• Other Authors: Kuo, Bor-Chen
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/95824695563257046316

碩士 === 國立臺中教育大學 === 教育測驗統計研究所 === 101 === 　　Currently, the relevant research of possible values is based on UIRT. Reaserch that focus on the influence of auxiliary variables on the estimation of population statistics and item parameter based on MIRT is rare. The purpose of this study is to explore the performance of different methods of estimation under MIRT, using simulated data. This study is based on multidimensional random coefficients multinomial logit model (MRCMLM). EAP, EAP_AV, MLE, WLE, PV_noAV and PV are used to compare the estimation efficiency of individual ability and population statistics under both of the situations that ancillary variables are used or not. 　　The result shows that when the number of items increase, the estimation error decrease, especially in the individual ability estimation. In multidimensional item response theory, when the correlation between dimensions increases, the accuracy of estimation does not improve. In the condition that ancillary variables are used, the impact of the correlation between the ancillary variables and the ability is limited by the item parameter setting. 　　When ancillary variables are incorporated, the parameters can be estimated well. Even if the item parameters and ability parameters have different distributions. In the condition that ancillary variables are not incorporated, plausible value still performances well in the recovery of standard deviation.