A regularization approach for the detection of differential item functioning in generalized partial credit models

Area of Science:

• Psychometrics
• Statistical modeling
• Educational measurement

Background:

• Traditional differential item functioning (DIF) detection methods in item response theory (IRT) are limited to single covariate analyses.
• Analyzing multiple covariates requires repetitive, independent application of existing methods, increasing complexity and potential for error.

Purpose of the Study:

• To propose a novel regularization approach based on the lasso principle for detecting uniform DIF.
• To extend DIF detection to polytomous item response models, including the generalized partial credit model.
• To develop a method that simultaneously corrects for DIF effects from multiple covariates.

Main Methods:

• A joint model is specified to parameterize DIF effects for all items and covariates.
• A penalized likelihood approach is employed for model estimation.
• The method utilizes the lasso principle for regularization to automatically detect DIF effects.

Main Results:

• The proposed regularization approach effectively detects uniform DIF across multiple covariates simultaneously.
• The method provides trait estimates adjusted for simultaneously detected DIF effects.
• Simulation studies demonstrate the robustness and accuracy of the proposed approach.

Conclusions:

• The regularization approach offers a powerful and efficient tool for detecting uniform DIF in the presence of multiple covariates.
• This method enhances the accuracy of trait estimation by accounting for complex DIF patterns.
• The approach is applicable to a wide range of polytomous IRT models, including the generalized partial credit model.