初始階段選題對電腦適性測驗試題曝光率之比較

Abstract

電腦適性測驗之題庫建立不易，如不能有效控制題庫題目曝光率，當題目過度曝光，受測者則容易施測到相同的題目使得測驗的安全性與公平性產生危機。在測驗開始階段三至五題時，由於能力估計仍未精確，此時若能有效分散試題，提升題庫中低曝光率試題的使用率，並可有效降低題庫中未使用試題題數，對整體試題曝光率達到有效控制，以期達到提升題庫經濟效益。 \n 本研究針對測驗前期三至五題採取隨機選題、McBride & Martin(1983)提出之5-4-3-2-1隨機選題法及b值分層隨機選取等方法，比較其試題曝光率控制效果，試圖找出有效控制題庫中未使用試題題數、降低試題曝光率、改善施測時前後測試題重疊率及試題重疊率的方法。 \n 研究結果：電腦適性測驗測驗前期三至五題採取b值分層隨機選取對試題題庫題目曝光率有明顯改善。以b值分層選題法控制測驗測驗前期五題選題，六至二十題採鄰近法選題在曝光率控制上的表現最佳；若以b值分層選題法控制測驗測驗前期三題選題，四至二十題採最大訊息法選題，則能力估計誤差最小。

To maintain test security and fairness of computerized adaptive testing (CAT), the exposure rate of item pool should be well controlled. The items administered to different examinees are most frequently the same in the early stage of CAT for the estimation of latent trait of examinees were not accurate enough. It can be expected to balance the exposure rate of items in the item pool and therefore efficiently usage of the item pool. \n This study compared the performance of exposure control of three methods in the early stage of CAT: random selection, 5-4-3-2-1 random selection, and b-stratified random selection. The results of percentage of unused items, exposure rate, test overlap rate and item overlap rate of these methods were presented in this study. \n It was found that the b-stratified random selection method improved the exposure control rate of the item pool significantly. To use the b-stratified random selection method combined with Nearest-Neighborhood method performs better than other combination on item exposure control. The bias of latent trait was minimized under the b-stratified random selection method combined with maximum information method.