Anchor Detection Strategy in Moderated Non-Linear Factor Analysis for Differential Item Functioning (DIF).

Area of Science:

• Psychometrics
• Educational Measurement
• Psychological Assessment

Background:

• Measurement invariance is essential for fair assessments and detecting Differential Item Functioning (DIF).
• Moderated Non-linear Factor Analysis (MNLFA) is a flexible method for DIF detection, but anchor item selection remains a challenge.
• Incorrect anchor selection in MNLFA can introduce bias into DIF detection results.

Purpose of the Study:

• To propose a refined constrained-baseline anchor detection approach for MNLFA-based DIF analysis.
• To enhance the accuracy and reliability of DIF detection by improving anchor item selection.
• To provide a robust method for identifying DIF-free anchor items crucial for fair assessments.

Main Methods:

• A three-step anchor detection procedure using information criteria (IC) is proposed.
• Bayesian Information Criterion (BIC) and Weighted Information Criterion (WIC) are used for initial DIF item identification.
• Akaike Information Criterion (AIC) is employed for selecting DIF-free anchor items.

Main Results:

• The proposed method effectively controls Type I error rates in DIF detection.
• The approach maintains adequate statistical power for identifying true DIF.
• Simulation studies and empirical data analysis validate the method's effectiveness.

Conclusions:

• The refined constrained-baseline anchor detection approach offers an accurate and computationally efficient solution for MNLFA-based DIF analysis.
• This method addresses the critical challenge of anchor item selection, leading to more reliable measurement invariance.
• The findings contribute to fairer and more precise psychological and educational assessments.