It was found that the majority groups tend to perform better in easier items and the minority groups tend to perform better in harder items in SAT. The relationship was referred to Feedle phenomenon and has been confirmed by some previous research.
However, the findings have been criticized: (1) the association may due to the confounding of DIF and impact which could not successfully discriminated by the standardization approach (e.g., MH procedure); (2) the relationship may due to the random guessing, which is not detected by the approach used in the previous studies. That is, the Feedle phenomenon may due to the incorrect methods used.
Hence, the present study aim to investigate whether the Feedle phenomenon could still be observed when using IRT methods. The 1PL and 3PL DIF model in which the DIF effect is expressed by the interaction between the item parameters and memberships were used. The correlation analysis between item difficulty and DIF estimated was computed. The relationships between the item difficulty and the difference of item parameters (b and c parameter) were computed as well. The results obtained generally support the existence of Feedle phenomenon.
Questions:
(1) Some problems have been found for the IRT models used in the present study in detecting DIF. Therefore, the DIF estimates may not reliable. Consequently, the correlation between the item difficulty and the DIF estimate may not be true.
Future study:
It is desirable to investigate the potential cause of Feedle phenomenon.