Li, MinLAN, MING-CHIH2014-10-132014-10-132014LAN_washington_0250E_13509.pdfhttp://hdl.handle.net/1773/26209Thesis (Ph.D.)--University of Washington, 2014Gender differences in mathematics performance have drawn much attention from researchers. Prior to making gender comparisons, a test with unbiased items should be used. Unbiased items are expected to have equal item correct response rates for males and females matched in their abilities. Differential Item Functioning (DIF) techniques have been developed to detect items that exhibit unequal success rates. The focus of this secondary study with the international large-scale TIMSS mathematics assessment was to investigate gender DIF items for eighth grade populations in the U.S. and Taiwan based on ordinal logistic regression and Poly-SIBTest. For DIF items flagged, the study examined the gender DIF patterns across DIF methods, cognitive demand levels, and countries. The study concluded that the total amount of DIF items was low, and the number of items identified as DIF favoring each gender was approximately equal. This gender DIF pattern was consistent across items differing in cognitive demand levels, and in both the U.S. and Taiwan samples. That is, in general, neither males nor females were favored by test items differing in cognitive demand levels. The two DIF techniques were consistent in their abilities to identify DIF and non-DIF items, but the magnitude of DIF identified by the ordinal logistic regression DIF tended to be smaller than the magnitudes identified by the Poly-SIBTest. Although the gender DIF patterns were consistent across the focal countries, the actual items identified as DIF and non-DIF differed because gender DIF items were country-dependent. This has important implications to deal with country-specific gender DIF items in any international assessment programs.application/pdfen-USCopyright is held by the individual authors.DIF; differential item functioning; gender; mathematics; TIMSSEducational tests & measurementseducation - seattleThesis