Interaction effects of gender and item arrangement on test and item performance

The purpose of this study was to examine the possibility of differential performance for males and females on a mathematics achievement test for different item arrangements. Previous studies of sex differences on item arrangements on mathematics tests have been limited in scope and sample size. This study used a national, standardized test for the mathematics test and a large, nationwide sample. Three item arrangements were employed in the study: (1) easy-to-hard; (2) easy-to-hard within content; and (3) spiral-cyclical. The 20-item mathematics achievement test selected for study was constructed from tryout items of the General Educational Development (GED) Experimental Mathematics Test. The three forms were spiraled into the 50 GED experimental units and administered during the 1985 fall testing sites nationally. Differential performance was examined at both the test and item levels. At the test level, Spearman rank-order correlation coefficients were obtained for all pairwise combinations between and within sex. The results indicated that the rank order of item difficulty is essentially the same between male and females on the different item arrangements. However, when looking within sex, for both males and females, the rank order of the item difficulties is relatively lower. At the item level, a chi-square full procedure was performed on each item of each form. None of the items showed significant male-female differences for the easy-to-hard form or the spiral-cyclical form. Five of the 20 items showed a significant difference between males and females on the easy-to-hard within content form. This research describes a procedure that can be used to assess differential performance between groups. The results suggest that the item difficulty statistic may not remain constant regardless of the position of the item within the test, and that item arrangement may be a factor of differential performance between males and females on mathematics achievement tests. These findings have several implications for test constructors and future research