MH,SIBTEST,LR－DIF方法檢測成效之比較研究

• Main Authors: Wu, Pei-Ling, 伍珮鈴
• Other Authors: 曾建銘
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/05490552186748053332

碩士 === 國立臺中教育大學 === 教育測驗統計研究所 === 99 === This research investigated gender and region Differential Item Functioning (DIF) on science subject of the Basic Competence Test for Junior High School Students (BCTEST) from 2006 to 2008 administrations. Three types of software were used in this study: SIBTEST, SPSS and EZDIF including five DIF detection methods: Crossing SIBTEST, SPSS MH, EZDIF MH, EZDIF ETS and EZDIF LR. The DIF items detected by the first three DIF detection methods were classified “serious DIF”. In order to presume the cause for DIF, the “serious DIF” items were sorted by knowledge subject classification and analyzed by the item option analysis. The comparisons between DIF detection methods were also discussed in this study. In the part of region DIF analysis, the results showed that total 7 (2.01%) of the 348 items were categorized as “serious region DIF”. There was no favorite tendency to any region. The reasons to cause items DIF were extrapolated as different activities and life experience based on living conditions and background. Regarding to gender DIF analysis, the results discussed that total 78 (22.41%) of the 348 items were categorized as “serious gender DIF”, 44 items tended to benefit boys; 34 items tended to benefit girls. Through the analysis of the DIF items, the results indicated that items of “biology on earth”, “composition and characteristic of substances”, “structure and function”, “the continuation of life” tended to favor female, because most of the contents involved in biology and chemistry. Moreover, the items of “earth environment”, “change and balance”, “ecological conservation”, “environmental protection” tended to favor male, because most of the contents were related to physics and geosciences. Under the standard of significance level of 0.01, the results showed that the consistency was above 77% by using three types of software to detect DIF. The similarity between the results of serious region DIF detection by Crossing SIBTEST method was the highest (44%). The similarity between the results of serious gender DIF detection by MH method along with ETS classification rules was the highest (100%). Furthermore, the similarity between the results of serious DIF detection by LR method was the lowest (4.76%, 37.50%) both at the region or gender DIF detection. Finally, some suggestions based on the findings of this study were proposed for test developers, researchers and science teachers.