Differential Item Functioning via Robust Scaling.

Area of Science:

• Psychometrics
• Educational Measurement
• Statistics

Background:

• Differential Item Functioning (DIF) is crucial for test fairness.
• Current DIF detection methods often require pre-specified anchor items, limiting their applicability.
• Item Response Theory (IRT) provides a framework for analyzing item and person characteristics.

Purpose of the Study:

• To propose a novel method for assessing DIF in IRT models.
• To develop a DIF detection approach that does not require anchor items.
• To enhance the robustness and efficiency of DIF analysis.

Main Methods:

• Re-formulating DIF as an outlier detection problem within IRT scaling.
• Utilizing robust statistics, specifically a redescending M-estimator, for parameter estimation.
• Tuning the estimator to control the asymptotic type I error rate for DIF detection.

Main Results:

• The proposed redescending M-estimator demonstrates efficiency in the absence of DIF and robustness in its presence.
• Simulation studies indicate favorable comparisons with existing DIF detection methods.
• A real data example showcases the method's practical application where anchor items are not feasible.

Conclusions:

• The proposed method offers a viable alternative for DIF assessment, particularly when anchor items are unavailable.
• This robust statistical approach enhances the reliability of DIF detection in IRT.
• The findings have implications for improving the fairness and validity of educational and psychological assessments.