Exploring Potential Quantitative Item Bias across Groups

Overview

This study used a differential item impact (DII) procedure to flag items on the Graduate Management Admission Test^™ exam that might lead to bias against subgroups of examinees. Items were flagged if they resulted in different conditional probabilities of a correct response for a subgroup of US or non-US citizens when compared to the calibration sample of examinees with similar ability. A small set of non-operational items were identified as having relatively high DII values. Specifically, quantitative items with a reading comprehension component yielded negative impact values for non-US subgroups and positive values for US subgroups. The opposite pattern was revealed when purely computational quantitative items were examined. Examples of these items are provided and reasons for potential bias are explored.