An investigation of differential item functioning in PISA 2015 math subtest in terms of the educational zones and gender

In large-scale exams, the presence of differential item function (DIF) is considered as an indication of possible bias. The aim of the study is to determine whether the 2015 PISA math subtest includes differential item functioning (DIF) in terms of statistical regional units and gender variables. Rasch model method of item response theory was used for DIF detection analyses. The data of the work includes the answers of 2409 students who have applied all mathematics booklet from Turkey attending PISA 2015. Before passing on to DIF analysis, it has been examined whether the data satisfy the Rasch model's qualities. The normality graph and the skewness-kurtosis coefficient are examined for the assumption of the normal distribution of the data for each cluster. The skewness and kurtosis coefficients of the clusters were found to be between +1 and -1. Confirmatory Factor Analysis (CFA) was applied for a total of sixty-three mathematics questions in each cluster for the assumption of unidimensionality. After that, RMSEA, GFI, NFI, RFI, IFI, AGFI, CFI, and SRMR values were examined for both validity and model fit. Modification index values of confirmatory factor analysis were examined and then eight items were omitted du to the error in model conformity. The math subtest is considered to be unidimensionality. The data were analyzed in the Winsteps 3.80.1 program. In the analysis for math subtests, five items showed DIF related to the gender. Each item in the math subtest has been found to contain DIF at different values according to the statistical region unit classification. DIF was observed in at least Q4 according to statistical region unit classification, but Q47 and Q50 were observed at most.