A new generation of DIF studies 

Chair

Paula ELOSUA (University of the Basque Country, Spain)

Symposium Abstract

Numerous DIF studies have been published in specialized and applied psychometric journals during the last two decades. In addition to the development of statistical procedures for detecting differential item functioning that are highly efficient in spotting problematic items, the research on DIF to date has also focused on applications of DIF analyses in a range of testing contexts. All of this work is critical because of the extent to which DIF analyses are a fundamental part of item analysis. However, it is important to note that any analysis of differential item performance should not be narrowly focused on the detection of DIF: once DIF is detected, the task turns to understanding it, the study of effects of item type on examinee performance, or the study of the practical consequences. It is this idea of extending DIF studies with new methods and approaches that forms the basis of this proposed symposium: A new generation of DIF studies. The new perspective involves multilevel latent models, mixed models, consequences and new robust procedures for the detection of DIF. The symposium consists of four presentations given by researchers from four countries. The first study illustrates a new approach to detecting DIF based on using robust statistics ; the second one uses a simulation to evaluate the effects of factorial partial invariance on group comparisons ; the third and fourth presentations incorporate mixture models to evaluate the presence of latent classes and novel applications of multilevel IRT .