IRT Differential Item Functioning: An Examination of Ability Scale Purifications

Abstract

Item Response Theory (IRT) differential item functioning (DIF) methods were employed to determine the accuracy of item classification as biased or unbiased. The procedure involved a combination of Drasgow's iterative item parameter linking method and Lord's ability scale purification procedure. Biased items were created using a two-dimensional noncompensatory IRT model. Previous research had demonstrated that the iterative linking and ability scale purification (ILAP) method held promise for discerning biased from unbiased items in one simulated unidirectional DIF condition. The present study found that ILAP was at times more effective than iterative linking alone primarily by reducing false negatives and occasionally reducing false positive misidentifications. The choice of significance level employed for detection of DIF did influence the convergence of the ILAP method. The method does hold some promise for improving correct classification of test items as biased or unbiased.