| John Bielinski
University of Minnesota
Describing and accounting for gender differences in mathematics achievement test item performance
FINAL REPORT:
This study seeks to replicate a sex difference by item difficulty interaction reported by Bielinski and Davison, 1998. Using multiple-choice mathematics items from a minimum competency test, that study demonstrated that easy items were easier for females than males and hard items were harder for females than males. They reported a significant interaction on seven of nine forms of the test administered to an 8th grade sample in a mid-western state. The present study used multiple-choice mathematics items from the NELS:88, 1992 NAEP, and the TIMSS. Participants spanned grades from 4th to 12th.
Using the three-parameter logistic model and the DIF option in BILOG-MG, separate item difficulty estimates for males and females were generated. The sex difference by item difficulty interaction was evaluated by computing the Pearson correlation between the item difficulty difference (bm - br) and item difficulty estimated on the combined male/female sample. In all nine item pool the correlation was negative, and in five, it was significant. The findings imply that gender DIF is related to item difficulty, which runs contrary to the usual assumption that DDF is a function of construct irrelevant item characteristics. The interaction also suggests that the size and direction of sex differences in mean math achievement may depend on the composition of item difficulties.
The effect of sex differences in math course-taking on the interaction was also examined. If sex differences in math course-taking alone account for the interaction then the interaction should disappear after controlling for math course-taking. Examinees were assigned to one of three mutually exclusive subgroups based on the highest level of math they had taken in high school: Basic (less than algebra); Standard (algebra to trigonometry); and Advanced (above trigonometry). Males were more likely than females to be in the Basic and Advanced group; whereas females were more likely to be in the Standard group. The sex difference by item difficulty interaction was reanalyzed within these mutually exclusive groups. Of the six groups within which the interaction was re-computed, three remained statistically significant. The results suggest that math course-taking differences, alone, cannot account for the interaction.
Back to Funded Dissertation Grants Page |