Determining Differential Item Functioning with the Mixture Item Response Theory

Abstract

Purpose: Studies in the literature have generally demonstrated that the causes of differential item functioning (DIF) are complex and not directly related to defined groups. The purpose of this study is to determine the DIF according to the mixture item response theory (MixIRT) model, based on the latent group approach, as well as the Mantel-Haenszel method, based on the observed group approach, compare the results, and determine the possible causes of the DIF. Research Methods: As this study is contributing to the production of information to develop the theory, it is considered basic research. In accordance with the purposive sampling method, the

Latent Class Approach to Detect Differential Item Functioning: PISA 2015 Science Sample

“…Looking at item difficulties, items were at medium level for the members in first latent class, but they were difficult for members in the second class. This finding is consistent with the results of Yalcin (2018). So, we can say that conducting reference and focal groups in terms of the country may not be sufficient to represent equal ability level groups.…”

“…In these cases, it is argued that DIF results obtained from manifest groups may be biased when the ratio of overlap is less than 70%. In this context, it is proposed that DIF studies should be examined among unknown groups/ latent classes (Bilir, 2009;Choi et al, 2015;Cho, 2007;Cohen & Bolt, 2005;De Ayala et al, 2002;De Mars & Lau, 2011;Finch & French, 2013;Karadavut, 2017;Maij-de Meij et al, 2010;Oliveri, Ercikan, & Zumbo, 2013;Samuelsen, 2005, Uyar, Kelecioglu & Dogan, 2017Yalcin, 2018). Kelderman and McReady (1990) agree with the idea that a latent class approach to detect DIF can be productive.…”

“…Studies in this field revealed there was a weak correlation between gender and latent classes. This means that DIF analysis conducted with gender groups may produce misleading results (Cohen & Bolt, 2005;Yalcin, 2018). Some members in one group can have the advantage to respond to an item correctly, but other members in this group can be disadvantaged (Cohen & Bolt, 2005;De ayala et all, 2002).…”

“…Therefore, according to previous studies, it can be said that the performance of manifest DIF analysis may be lower than latent DIF analysis. Studies, related to MixIRT DIF were carried with simulated and real data (Cohen & Bolt, 2005;Maij-de Meij, 2010), with simulated data (Bilir, 2009;Samuelsen, 2005, Uyar et al, 2017Yuksel, 2012) or only with real data (Finch & Finch, 2013;Karadavut, 2017;Van Nijlen & Janssen, 2008;Yalcin, 2018). According to Cho & Lee (2016), it is required to examine the performance of manifest DIF detection methods for studying latent DIF approach in future studies.…”

Rasch Calibration of Achievement Test: An Application of Item Response Theory

“…It deals with measurement assumptions about test items difficulty level, examinee performance in the test, relation of performance with knowledge and skills measured by an individual item in a test. Yalcin (2018) discussed the characteristics of IRT: first of all, IRT must specify the relationship between the observed response and the underlying unobservable construct. Secondly, the model must facilitate the ways about observed responses.…”