Advances in the detection of differentially functioning test items

The development and evaluation of methods for detecting potentially biased items or differentially functioning items (DIF) represent a critical area of research for psychometricians because of the negative impact of biased items on test validity. A summary is provided of the authors ' 12 years of research at the University of Massachusetts (Amherst) pursuing item response theory-based and Mantel Haenszel (MH) DIF detection methods. In addition, a set of guidelines is offered for conducting DIF studies based on these research findings. These recommendations include the following: (1) the two-step procedure recommended by P. W. Holland and D. T. Thayer (1988) is preferred to the simple procedure; (2) the criterion used for matching examinees must be approximately unidimensional; (3) larger examinee samples are preferred; (4) with limited sample size, the power of the statistic can be increased by increasing the majority group and holding the minority group constant; (5) with very large samples, measures of both statistical significance and effect size may be important in screening items; (6) examinees in the sample must represent the population of interest; (7) combining score groups in the matching criterion may be useful for increasing the power of the MH statistic; (8) items that have lower a-parameters (item discrimination parameters) are less likely to be identified by the MH method; and (9) the ME method is not blind to non-uniform DIF. (Contains 5 tables, 3 figures and 47 references.) (SLD) *************4:********************************************************* Reproductions supplied by EDRS are the best that can be made from the original document. t