The Improvement of DIF Detection on Test Validity—From the Perspective of Measurement Invariance

• Main Authors: Sung-wei Ma, 馬菘蔚
• Other Authors: Hueying Tzou
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/44657216289595316931

碩士 === 國立臺南大學 === 教育學系測驗統計碩士班 === 101 === The main purpose of this study was to investigate the benefit of DIF detection on test validity in terms of measurement invariance, and the effect of using different DIF purifying procedures (one-step, two-step, iteration). Mantel-Haenszel (MH) method was used in a simulation study with different test length, DIF magnitude and DIF item proportion. Except for chi-square difference test, several indices like NFI, IFI, RFI, TLI, CFI, Gamma hat and NCI were employed as the criteria to examine measurement invariance. The results show that NFI and other indices are less restricted than chi-square difference test. DIF detection can improve measurement invariance but there is no apparent difference among different DIF purifying procedures. Meanwhile, measurement invariance is reduced with the increasing of test length. DIF magnitude and DIF item proportion have no significant varying pattern on invariance.