The Effects of Using Educational Background Variables in DIF Analyses.

Whether stratifying gender and ethnic groups according to educational variables in the examinees' backgrounds adds information to the differential item functioning (DIF) process for the verbal and quantitative sections of a graduate admission test was explored using data from a 1990 graduate admission test. DIF analyses were performed on the following groups using the Mantel Haenszel procedure: (1) 4,282 males; (2) 6,419 females; (3) 9,267 Whites; (4) 416 African Americans; (5) 368 Hispanic Americans; and (6) 341 Asian Americans. Males and Whites served as the reference groups for females and minority groups, respectively. Undergraduate major and coursework in English and mathematics were the educational variables selected. The study was begun with the idea that a more homogeneous population might decrease the number of items with DIF, but research does not entirely support the original hypothesis. In many cases, the number of items with DIF remained the same or increased when a more homogeneous population was analyzed. Coursework analyses were not as clear as had been expected. Some of the DIF did appear to be explainable, and the contribution of background variables to DIF may be elucidated by further study. Included are 33 tables of DIF analysis results and 9 graphs illustrating comparisons among groups. (SLD)