Using Hierarchical Logistic Regression to Study DIF and DIF Variance in Multilevel Data
Link to Resource: Using Hierarchical Logistic Regression to Study DIF and DIF Variance in Multilevel Data
Authors: Benjamin R. Shear
Citation: Shear, B.R. (2018). Using hierarchical logistic regression to study DIF and DIF variance in multilevel data. Pre-print: Journal of Educational Measurement, v. 55, no. 4, pp. 513
Abstract:
When contextual features of test-taking environments differentially affect item responding for different test-takers and these features vary across test administrations, they may cause differential item functioning (DIF) that varies across test administrations. Because many common DIF detection methods ignore potential DIF variance, this paper proposes the use of random coefficient hierarchical logistic regression (RC-HLR) models to test for both uniform DIF and DIF variance simultaneously. A simulation study and real data analysis are used to demonstrate and evaluate the proposed RC-HLR model. Results show the RC-HLR model can detect uniform DIF and DIF variance more accurately than standard logistic regression DIF models in terms of bias and Type I error rates.