Determining Gender Differential Item Functioning for Mathematics in Coeducational School Culture

Purpose - In an attempt to explore item characteristics that behave differently between boys and girls, this comparative study examines gender Differential Item Functioning in a school culture that is noted to be ‘thriving’ mathematically. Methodology - Some 24 grade eight mathematics items from TIMSS 2003 and TIMSS 2007 released items, with equal number of computation and word problem items were administered on 460 boys and 445 girls studying in Grade Eight from three secondary Chinese-medium coeducational schools. Word problem items were defined as items set in a real-world context. Content validity was established by constructing a table of specifications. By employing the software WINSTEPS version 3.67.0 that is based on the Rasch Model for dichotomous responses, Differential Item Functioning analysis was conducted by using Mantel-Haenszel Chi-square method. DIF items were flagged when the Mantel-Haenszel probability value was less than 0.05 and classified as negligible, moderate or large DIF based on the DIF size suggested by Educational Testing Service DIF category. The focal and reference groups were girls and boys respectively. The main delimitation was substantive analysis by using expert judgment was not conducted to identify biased items. Findings - Using Mantel-Haenszel chi-square, two moderate DIF items that assess subtraction favoured girls. They assessed Knowing from topics Whole Number and Fraction. Sources of DIF are linguistics density and item presentation style. The findings suggest that with only two moderate DIF mathematics items, there is insufficient evidence to suggest that the mathematics items functioned differently between boys and girls in school culture noted for successful mathematics learning, even though linguistics complexities of the test language cannot be ignored. Significance - While constructing Mathematics multiple choice items, careful considerations need to be given in selecting suitable numbers in composing the content of the item so that, only the correct algorithm would produce the correct answer ‘if and only if’ those numbers are used. The modest results of detecting two moderate DIF items nevertheless inform national testing agencies and teacher educators on the principles of building fair items as a part of their test improvement practice in the 21st Century. The novelty of this study is that gender Differential Item Functioning was studied in the context of school culture, which is notable for successful mathematics learning.