Causal models of item format- and gender -related differences in performance on a large -scale mathematics assessment for grade three to grade ten

U.S. and international studies have reported that gender differences in overall mathematics performance tend to favor males. In contrast, gender differences on overall mathematics assessments in Hawai'i have been consistently in favor of females. When item format is considered, the findings of the U.S. and international studies are inconclusive. For multiple-choice (MC) items, it is usually found that males perform better than females whereas, on constructed-response (CR) items, no clear pattern emerges. These findings were confirmed by a meta-analytic review.;This study examined the mathematics section of the 2002 Hawai'i State Assessment (HSA) program, the first administration of the standards-based HSA, administered to third, fifth, eighth, and tenth graders in public schools. Preliminary ANOVAs at the four grade levels were conducted to determine whether HSA mathematics performance is influenced by gender, item format, or the interaction of the two. Next, gender difference effect sizes for mathematics total, MC, and CR were obtained, and the homogeneity of the effect sizes was tested. Finally, in order to better understand gender-related differences in mathematics performance, the study investigated a causal model that accounts for gender differences in mathematics performance on MC and CR items in conjunction with the effects of reading and writing. Four separate path analyses, one for each grade level, confirmed a generalizable causal model across the four grade levels.;With linguistic factors incorporated in the causal model it was confirmed that the females' superior performance on mathematics assessments in Hawai'i is due to their advantage in reading and writing, not in mathematics. Females' advantage in reading and writing skills compensate for their relative disadvantage on either MC or CR items.;Although gender difference effect sizes in this study appear to be favoring females, girls are disfavored on both MC and CR items, and the female disadvantage is greater on MC items than on CR items after controlling for language factors. These findings suggest a need for the reexamination of the mathematics construct definition, related validity and test specification issues, and a gender-specific approach to mathematics education.