A confirmatory study of Differential Item Functioning on EFL reading comprehension

Abstract

The present study aimed at investigating DIF sources on an EFL reading comprehension <br />test. Accordingly, 2 DIF detection methods, logistic regression (LR) and item response <br />theory (IRT), were used to flag emergent DIF of 203 (110 females & 93 males) Iranian <br />EFL examinees' performance on a reading comprehension test. Seven hypothetical DIF <br />sources were examined in this regard: text familiarity, gender, topic/text interest, guessing, <br />and the social variables of location, income, and educational status. Only LR, for gender <br />and text familiarity, could preempt DIF with gender supporting the gendered-text effect <br />while text familiarity benefiting, inversely, the participants with low level of text <br />familiarity. For interest in topic, LR found a single item favoring the group with higher <br />levels of interest and the IRT model detected DIF in either extreme. Regarding guessing <br />and income, the LR indicated DIF supporting the low guessers and high-income group <br />whereas IRT, conversely, showed DIF favoring the high guessers and low-income group. <br />For location and education both methods, correspondingly, demonstrated DIF for the <br />expensive location and educated groups. Finally, the differential test functioning result <br />made it clear that only three sources of DIF (gender, income, & interests) were transferred <br />to the test level. The findings could support a proportional effect of DIF sources.