國中八年級自然科定期評量之性別差別試題功能（DIF）分析

Gender Differential Item Functioning in a Science Periodical Test of Eighth Graders

本研究旨在探究國中八年級學生學校自然科定期評量之性別成就差異和差別試題功能（Differential Item Functioning, DIF）。研究資料係新北市某公立國中八年級382 名學生（男性191 名、女性191 名）之答題反應，研究先以性別效果量、女／男標準差比值與女／男人數比值三項指標分析性別成就差異，再用IRT Rasch Model 與Mantel-Haenszel法分析性別DIF 現象。研究發現如下：1. 未配組的性別成就差異部分，全體受試並無顯著性別成就差異現象，而在高成就組（前10%）和低成就組（後10%）中，男性表現略高於女性；2. 將男女能力配組後，取IRT Rasch 與Mantel-Haenszel 結果交集，顯示自然科定期評量嚴重DIF的出現率為4%且皆有利女性。筆者對DIF 試題的內容進行進一步審視，試題特徵似乎與性別DIF 有關聯，初步推論試題的試題附圖的提供是否可能影響DIF 方向，但仍需更多實徵資料驗證。整體而論，本研究並未發現和測驗目標無關的因素，尚不構成試題偏誤。最後，本研究依據研究結果，對自然科測驗命題與未來研究提出建議。

This study investigates gender differences and differential item functioning (DIF) in a science periodical test of eighth graders. we selected 382 students (191 boys and 191 girls) from a junior high school in New Taipei City. we calculated and compared the effect size, female/male standard deviation ratio, and female/male ratio. In addition, we used the IRT Rasch model and the Mantel-Haenszel procedure for gender DIF. The results of this study are as follows: (a) in unmatched analysis, no gender differences were observed among all groups; however, boys exhibited slightly better performance to girls among both high-achieving groups (top 10%) and low-achieving groups (bottom 10%); (b) in matched analysis, the intersection of the results using the IRT Rasch model and the Mantel-Haenszel procedure showed that the average percentage of items displaying gender DIF across administrations was low, at approximately 4% (in favor of girls). The follow-up review of these DIF items indicated associations of gender DIF with item characteristics. Furthermore, charts may affect the DIF direction. Finally, this study provides suggestions for items construction and future studies on science.