Bayesian Approaches for Detecting Differential Item Functioning Using the Generalized Graded Unfolding Model.

Area of Science:

• Psychological Assessment
• Psychometrics
• Statistical Modeling

Background:

• Differential item functioning (DIF) analysis is crucial in psychological assessment for ensuring fairness.
• Item response theory (IRT) provides a framework for DIF analysis.
• Bayesian methods offer alternative approaches to traditional likelihood-based methods.

Purpose of the Study:

• To evaluate the performance of two Bayesian DIF methods (Bayes factor and deviance information criterion) within the generalized graded unfolding model (GGUM).
• To compare the effectiveness of Bayesian methods against likelihood-based methods (likelihood ratio test and Akaike information criterion) in detecting DIF.
• To investigate the impact of various factors (sample size, DIF characteristics, subgroup distributions) on DIF detection accuracy.

Main Methods:

• A Monte Carlo simulation study was conducted.
• The generalized graded unfolding model (GGUM) was used.
• Bayesian (Bayes factor, DIC) and likelihood-based (LR, AIC) DIF detection methods were implemented and compared.

Main Results:

• Bayesian methods (BF and DIC) demonstrated well-controlled Type I error rates and high statistical power.
• The Bayesian methods outperformed likelihood-based methods (LR and AIC) in controlling Type I error rates, especially when subgroup trait distributions differed.
• Performance was evaluated under various simulation conditions including sample size, DIF source, size, location, and baseline model type.

Conclusions:

• Bayesian DIF analysis using BF and DIC with GGUM offers a robust and effective approach for psychological assessment.
• These Bayesian methods provide superior control over Type I errors compared to traditional likelihood-based methods under specific conditions.
• Recommendations are provided for the application of these advanced DIF detection techniques in applied research.