Detecting the Differential Item Functioning of Numerical Ability Test in the Gulf Multiple Mental Abilities Scale by Mental-Haenszel and Likelihood Ratio Test

Abstract

The current study aimed to examine differential item functioning (DIF) of a numerical ability test for Gulf state school students. This study examined the DIF items by gender and country using two DIF analysis methods; the Likelihood Ratio Test (LRT) and the Mantel-Haenszel (MH). The sample size was 2689 individuals throughout grades 5 and 6, and the researchers used MH with the SPSS and LRT with the BILOG-MG. The study used the classification stability coefficient kappa (κ) to compare how well the two methods agreed to examine DIF. Regarding gender, LRT yielded precise results for 30% of the items, and in terms of country, for 43.3%. Also, analyzing DIF with MH found that 26.7% of items exhibited DIF based on gender and country. For gender, there was strong concordance (0.925) between the MH method and the LRT. For the country, the MH and LRT agreement was also high (0.683). The study suggested investigating the causes of test items' differential performance and comparing the DIF in two test types, paper-and-pencil and computer-based.