定錨試題分佈對測驗等化之影響

• Main Authors: HUANG CHIH-CHIEH, 黃志傑
• Other Authors: 郭伯臣
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/28204293575040412254

碩士 === 臺中師範學院 === 教育測驗統計研究所 === 92 === The purpose of this research is to investigate the effects of different anchor item distributions on test equating using the 3 parameter logistic model and non-parametric IRT. The impacts of four factors will be discussed: 1. the number of anchor item, 2. discrimination parameter of anchor item, 3. difficulty parameter of anchor item, 4. the distributions of ability at testees. Three cases of discrimination parameter of anchor item (a=0.5, 1.0, 1.5) are discussed in this study and the simulation experiment result show that a=1.0 and 1.5 cases have better performances of estimatings. The result also reveals that the best decision of the locations (difficulty parameters) of anchor items depends on the number of anchor items and the ability distributions. In summary, 1. Difficulty parameter is better in the range of 0 and 1.5. Estimating will be precise as the number of anchor item is large and items distribute even. 2. The number of anchor item is 2 or 3 supposing the number of testees is 3000 at having 50 items in test. 3. Discrimination parameter of anchor item is greater or near 1.0. Difficulty parameter of anchor item ranging between 0 and 1.5 will be fine. Three cases of discrimination parameter of anchor item (a=0.5, 1.0, 1.5) at non-parametric IRT are surveyed in this study. Comparing with parameter logistic model, non-parametric model is unacceptable obviously. Filtering of data may be the cause of result for ingoring some out-ranged samples trigger biases, and simulating data using parameter logistic model arise bigger margin of error.