Abstract


THE EFFECTS OF DIF ON TEST PARAMETERS ESTIMATES, DECISION STUDIES, G AND PHI COEFFICIENTS

This study aims to reveal the effect of removing items with low, medium and high Differential Item Function (DIF) from numerical and verbal ability tests, which were determined by using different (DIF) detection methods, on the test parameters estimates, the G and Phi coefficients of the decision studies, and the G and Phi coefficients of the test without DIF, according to the gender of the students, their weekly pocket money amount and whether their pocket money amount is sufficient or not. In this respect, the study is a basic research. It is also a descriptive research in terms of revealing the current situation. Mantel-Haenszel and Logistic Regression Methods based on Classical Test Theory (CTT) were used as DIF detection methods, and the SIBTEST, Lord's Chi-Square and Raju's Area Measures methods were used based on Item Response Theory (IRT). In order to reveal the general abilities of the students, 45-item numerical and 45-item verbal ability tests were used in the multiple-choice test item type by the researcher. Research data were collected from 2304 sixth grade students. As a result of the research, it is seen that there is no change that creates a significant difference in measurement comparisons as a result of removing items with low DIF from the test by using different DIF detection methods. However, as a result of removing the items showing medium and high DIF from the test, it was determined that there was a significant difference in some test parameters, some reliability coefficients, G and Phi coefficients of decision studies, and G and Phi coefficients of the test without DIF. In this context, it can be said that in order to positively increase the psychometric properties of a test, such as validity and reliability, the test should be freed from items showing medium and high DIF.



Keywords

Differential item function, ability, generalizability coefficient, test parameter estimates


Reference