ABSTRACT
Differential item functioning (DIF) statistics ide_B.tify SAT test items that are unusually difficult for members of particular population subgroups. Such items usually are considered undesirable because they appear to be measuring slightly different constructs for different subgroups. These items may not be removed from the test, however, when the construct being measured is an explicit part of the intended domain of the test. For example, the College Board science achievement tests require some mathematical skills, and sometimes the required math skills are differentially difficult for women or non-Asian minority students, who on average take fewer advanced mathematics courses. Although differentially difficult questions are avoided when possible, they are used when necessary to meet test specifications. Note that such results are found when the test measures more than one dimension of skill or knowledge (such as science knowledge and math skill), and those dimensions are distributed differentially in the two groups being compared.
