Score-Based Tests of Differential Item Functioning via Pairwise Maximum Likelihood Estimation

Area of Science:

• Psychometrics
• Statistical modeling
• Educational measurement

Background:

• Measurement invariance is crucial in item response theory (IRT) for accurate latent construct assessment.
• Violations can lead to misinterpretations and systematic bias, particularly in diverse populations.
• Existing detection methods often require unavailable prior information on item parameters and groups.

Purpose of the Study:

• To extend recently developed score-based tests for detecting measurement invariance violations.
• To adapt these tests for two-parameter item response models, focusing on pairwise maximum likelihood.
• To evaluate the tests' efficacy in identifying problematic item parameters.

Main Methods:

• Utilizing score-based tests derived from casewise derivatives of the likelihood function.
• Estimating only the null model (assuming measurement invariance holds).
• Applying tests to two-parameter IRT models with pairwise maximum likelihood estimation.

Main Results:

• The proposed score-based tests demonstrate effectiveness in identifying problematic item parameters in simulations.
• The study details the theoretical underpinnings and practical implementation of these novel tests.
• An empirical example showcases the real-world application of the measurement invariance tests.

Conclusions:

• Score-based tests offer a practical alternative for detecting measurement invariance violations in IRT.
• These tests are valuable for ensuring scale validity and fairness across different groups.
• The extension to two-parameter models broadens the applicability of these statistical tools.