[Application of four procedures for detecting differential item functioning in polytomous items].

Area of Science:

• Educational Measurement
• Psychometrics
• Statistical Modeling

Context:

• Differential Item Functioning (DIF) is crucial for ensuring fairness in educational assessments.
• Polytomous items, common in large-scale assessments, present unique challenges for DIF detection.
• The PISA 2000 Reading Comprehension Test provides a relevant dataset for evaluating DIF methods.

Purpose:

• To describe and apply four statistical methods for detecting Differential Item Functioning (DIF) in polytomous items.
• To evaluate the performance of Mantel, Generalized Mantel-Haenszel (GMH), Ordinal Logistic Regression (RLO), and Discriminant Logistic Regression (RLD) methods.
• To assess the agreement between these DIF detection procedures using statistical significance and effect size.

Summary:

• Four DIF detection methods (Mantel, GMH, RLO, RLD) were applied to polytomous items from the PISA 2000 Reading Comprehension Test.
• A cross-validation design using American and Spanish samples was employed.
• Total agreement among the four methods was observed for two items when considering both statistical significance and effect size.

Impact:

• Provides empirical evidence on the comparative performance of common DIF detection methods for polytomous items.
• Highlights the importance of considering both statistical significance and effect size in DIF analysis.
• Informs test developers and researchers on robust methods for ensuring item fairness in international assessments.