Examination of differential item functioning in PISA through univariate and multivariate matching differential item functioning

The present research aims to examine whether the questions in the Program for the International Student Assessment (PISA) 2009 reading literacy instrument display differential item functioning (DIF) among the Turkish, French, and American samples based on univariate and multivariate matching techniques before and after the total score, which is the matching variable, is purified of the items flagged with DIF. The study is a correlational survey model research, and the participants of the study consist of 4459 Turkish, French, and American students who took booklets 1, 3, 4, and 6 in the PISA 2009 reading literacy measure. Univariate and multivariate (bivariate, trivariate, and quadrivariate) DIF analyses were performed through logistic regression before and after purifying the matching variable off the items displaying DIF. Literature was used to detect extra matching variables, and multiple linear regression analysis was carried out. As a result of the analyses, it was discovered that using extra matching variables apart from the total score reduces type I errors. It was also concluded that the exclusion of DIF items (removal of items with DIF) while calculating the total score led to variation in the number of questions detected as DIF and DIF levels of the items, although it did not yield consistent results.