Combining Auxiliary Dimension Information to Logistic Regression Methods to Control False Positive Rates in Differential Item Functioning Assessment

• In Event: Recent Advances in Differential Item Functioning

Abstract

Self-reported questionnaires used in educational studies often measure multiple constructs. For example, a study habits inventory would measure not only the real learning attitude but a nuisance factor of socially desirable responding that is usually neglected in the data analysis. Ignoring a multidimensional structure might lead to biased conclusions when examining differential item functioning (DIF). This study recommends including the auxiliary dimension as a covariate when assessing the DIF. Two logistic regression (LR) methods for ordinal data were implemented for illustration. The results showed that the standard LR methods yielded inflated false-positive rates when there was a group difference in the auxiliary dimension; including the auxiliary dimension in LR methods can help control false-positive rates at a nominal level.

Authors

• Chi-Chen Chen, College Entrance Examination Center

• Kuan-Yu Jin, Hong Kong Examinations and Assessment Authority

• Ching-Lin Shih, National Sun Yat-Sen University