Assessing measurement invariance of Turkish "Central Examination for Secondary Education Institutions" for visually impaired students

Abstract

Studies that focus on measurement invariance are of significant importance in proving the validation of high-stake tests, and in order to provide fairness from the results of these exams for special needs students. The aim of this study is to examine the measurement invariance of the "Central Examination for Secondary Education Institutions" in Turkey according to participant disability status. A focus group comprised of 369 visually impaired students was formed. An equal number of non-visually impaired peers were randomly selected as a reference group. Mantel-Haenszel, logistic regression, Breslow-Day, and standardization methods of classical test theory were used in order to detect items with differential item functioning (DIF). DIF analysis results proved that 16 (17.78%) of the 90 test items indicated DIF, and that ten of the DIF detected items (62.5%) represented a disadvantage for visually impaired participants. A total of 17 experts were consulted in order to investigate item bias. As a result of the collective expert opinion, five items were found to be "biased" in the Turkish (n = 1), English (n = 2), and in Science (n = 2) subtests. Close agreement was obtained between the experts that the "biased items" favored non-visually impaired participants. Use of visuals/graphics, complex/lengthy texts and response options, the need for rereading questions, and the negative attitudes of readers/coders were pointed out as sources of item bias.