Fit of Responses to the Model III—Differential Item Functioning

Abstract

DIF occurs when items do not function in the same way for different groups of people who otherwise have the same value on the trait. DIF can be identified graphically through the ICC. Different locations of the curves for the groups but similar slopes indicate uniform DIF. Different slopes for the groups indicate non-uniform DIF. DIF can be confirmed statistically through analysis of variance of the residuals for the groups. An ANOVA group main effect indicates uniform DIF. A group by class intervalinteraction effect indicates non-uniform DIF. It may be possible to deal with DIF by deleting or resolving the item(s) with DIF. Resolving an item means creating from the item as many pseudo items as there are groups, each item having just one group responding to it. The principle of artificial DIF refers to DIF that can appear in an item as a result of real DIF in other items. To investigate whether DIF is real or artificial, deleting or resolving items are done sequentially one at a time, starting with the item with the largest DIF.