查詢結果分析
來源資料
頁籤選單縮合
題 名 | 評分客觀性與能力估計客觀性:傳統作法與試題反應理論作法之比較=Objectivity of Ratings and Objectivity of Ability Estimates: A Comparison between Traditional Approaches and Item Response Modeling |
---|---|
作 者 | 王文中; | 書刊名 | 測驗年刊 |
卷 期 | 44:1 1997.01[民86.01] |
頁 次 | 頁29-52 |
分類號 | 521.32 |
關鍵詞 | Rasch模式; 評分者嚴苛度; 評分客觀性; 能力估計客觀性; 試題反應模式; 建構反應題; Rasch model; Rater severity; Objectivity of ratings; Objectivity of ability estimates; Item response theory; Constructed-response items; |
語 文 | 中文(Chinese) |
中文摘要 | 評分客觀性指的是考生實得分數與其應得分數差異的程度,能力估計客觀性則是 對考生能力估計不受試題難度或評分者嚴苛度的影響。傳統上,對於建構反應題的評分客觀 性多以評分的一致性為依據。例如積差相關、分數差距的比例、概化力理論等。事實上,如 果考生被兩位打分非嚴苛的評分者來改卷子,他們所給的分數可能非常一致的低,這顯然低 估了考生的能力。因此評分者的一致性並不能保證評分的客觀性。除此之外,傳統作法建立 在古典測驗理論上,因此對考生能力的估計也就受到試題難度的干擾,因此也喪失了對能力 估計的客觀性。 試題反應理論成功的克服了這個缺點,如果資料吻合模式的話,對考生能力的估計也就 不受試題難度的影響。評分者嚴苛度也試題難度的一種,因此即使評分者的嚴苛程度不一, 並不會妨礙我們對考生能力的估計。過去利用試題反應理論來分析評分者嚴苛度的作法,大 多限試題本身難度和評分者嚴苛度不可以有交互作用,本研究刪去這個限制,因而形成多種 嚴苛模式。這大幅提高模式的適用性,也增進能加估計的客觀性。本研究一方面評論傳統分 析的不當,同時也利用試題反應理論,建構出各種評分者嚴苛度的模式,並說明其意義。最 後,透過大學聯考生物科建構反應題的資料分析,比較傳統方法與試題反應理論方法的差異 。 |
英文摘要 | Objectivity of ratings is referred to as the degree of agreement between given scores and deserved scores. Objectivity of ability estimates is achieved if the estimates are free from the items tested. Traditionally, objectivity of ratings in usually assessed in terms of consistency between ratings, such as correlation and percentages of agreement. Suppose an examinee is judged by two severe raters, the scores given may be consistently low. Consequently,consistency of ratings does not necessarily imply objectivity of ratings. Moreover, these traditional approaches are based on classical test theory, which mixes up ability estimates and difficulty estimates. Therefore, objectivity of ability estimates is destroyed. Item response theory successfully overcomes this drawback. If data fit the model, ability estimates and difficulty estimates are mutually independent. Objectivity of ability estimates is thus possible. For constructed-response items, where raters are involved, item difficulties can be partitioned into genuine difficulties and rater severities. If data fit the model, the ability estimates are independent of raters, meaning that ability will be objectively estimated even if raters are in different degress of severity. However, in earlier works parameterization of rater severities is too rigid to fit complexity of testing situations. In this study, unnecessary constraints are released and several advanced models are proposed. In so doing, not only their applications but also objectivity of ability estimates is increased. A real data set from the biology subject of the 1995 Joint College Entrance Examination was analyzed to demonstrate advantages of item response modeling over traditional approaches. |
本系統中英文摘要資訊取自各篇刊載內容。