Avoiding Omitted Variable Bias in Exploratory Differential Item Functioning Assessment: A Propensity Score Adjustment Approach

• In Event: AERA Roundtable Session 10
  In Roundtable Session: Missing Data in Measurement

Abstract

When there is omitted variable bias (OVB), the estimation of DIF effects would be biased, resulting in inflated type I error rates and/or reduced power in DIF assessment. Traditionally, the OVB can be avoided by controlling all possible confounding variables; however, it would generate an overfitting model. This study employed the propensity score (PS) method to improve the parsimony of the DIF model and incorporated the DIF-free-then-DIF strategy to improve the power in exploratory DIF assessment. Based on the results of simulation studies, assessing DIF without the control of confounding variables would result in mistaken outcomes. By contrast, the PS method appeared promising in terms of its well-controlled type I error rates and acceptable power in most conditions.

Authors

• Hsiu-Yi Chao, National Taiwan Ocean University

• Jyun-Hong Chen, National Cheng Kung University