Using IRT approach to detect gender biased items in public examinations: A case study from the Botswana junior certificate examination in Mathematics

Abstract

This is a quantitative study, which attempted to detect gender bias test items from the Botswana Junior Certificate Examination in mathematics. To detect gender bias test items, a randomly selected sample of 4000 students responses to mathematics paper 1 of the Botswana Junior Certificate examination were selected from 36,000 students who sat for the examination. Out of which 2,000 were males and 2000 were females. The examination paper consisted of 38 test items. To detect the gender biased test items, the study used 3PL (Multilog software) item response theory (IRT) statistical analysis. This generated the item characteristics curves (ICC for the two groups (male/female). The study compared the results generated from the ICC curves for the male and female groups, and found that, out of 16 test items that fitted the 3PL item response theory (IRT) statistical analysis, 5 items were gender biased. Key words: IRT (item response theory), ICC (item characteristics curve)