Variable Selection Approach for Exploratory Differential Item Functioning Assessment With Multiple Grouping Variables

• In Event: Marriott Roundtable Session One
  In Roundtable Session: Differential Item Functioning (Table 1)

Abstract

In exploratory DIF assessment, researchers must assess DIF for each item against all suspected grouping variables to ensure test fairness. Under the circumstances, test practitioners may encounter the small n, large p problem due to the near-saturated model. This study, therefore, proposes the comprehensive DIF assessment procedure to solve the problem. The procedure incorporates variable selection criteria into the DIF-free-then-DIF strategy, in which the lasso estimator is applied to discover DIF-free items for building a common metric, while variable selection criteria (e.g., the adaptive lasso estimator) are utilized to improve the power in detecting DIF items. The simulation results indicated that the proposed procedure can well control type I error rates under most conditions with acceptable power rates.

Authors

• Hsiu-Yi Chao, Soochow University
  Presenting Author

• Jyun-Hong Chen, National Cheng Kung University
  Non-Presenting Author