DIF sources

Context is a vague notion with numerous building blocks making language test scores inferences quite convoluted. This study has made use of a model of item responding that has striven to theorize the contextual infrastructure of differential item functioning (DIF) research and help specify the sources of DIF. Two steps were taken in this research: first, to identify DIF by gender grouping via logistic regression modeling, an inventory of mostly cited DIF sources was prepared, based on which a list of demographic items was appended to the TOEFL reading paper only to be administered to the intermediate Iranian undergraduates; second, using multiple-variable matching regression (Wu & Ercikan, 2006), a built-in sequence was followed to let every potential DIF source be considered as a covariate, over and above the conditioning variable, and specify whether a particular ecological variable could reduce DIF value/status. Then, all significant variables were analyzed together to show the final DIF predictors. The same procedures, i.e., individual/collective analyses, were employed after the purification of the test. The results indicated three ecological predictors affecting DIF before and after purification: income, administration convenience, and SES. The ultimate predictors helped create an EFL configuration of the ecological model of item responding. The implications for validity arguments are also discussed.

The purpose of the present study was two-fold: (a) First, it examined fairness of Special English Test (SET) of Iranian National University Entrance Exam (INUEE) by analyzing Differential Item Functioning (DIF) with reading comprehension section of this test (b) second, it explored test takers’ attitudes towards possible sources of unfairness and DIF. In the quantitative part of the study the data from 10000 test takers (6820 females and 3180 males) were analyzed for gender DIF using Mantel-Haenszel (MH) technique. It was revealed that only 6 items in the reading comprehension skill showed DIF. Further analysis manifested that the effect size of DIF for all six items were category A or negligible. Moreover, qualitative interview results indicated that learners generally considered the test a fair one while some potential sources of bias such as topic familiarity, multiple-choice format of the test, topic interest, passage length, and complex structure of test items were mentioned.