Bayesian Adaptive Lasso for Detecting Item-Trait Relationship and Differential Item Functioning in Multidimensional...

Area of Science:

• Psychometrics
• Statistical Modeling
• Educational Measurement

Background:

• Accurate measurement of latent traits is essential in multidimensional tests.
• Differential item functioning (DIF) analysis is critical for ensuring fairness and validity across different groups.
• Existing literature often addresses item-trait relationships and DIF separately, limiting a comprehensive understanding.

Purpose of the Study:

• To develop a unified framework for simultaneously identifying item-trait relationships and DIF within multidimensional item response theory (MIRT) models.
• To integrate DIF effects into MIRT models, treating the problem as a variable selection task for latent/observed variables and their interactions.
• To provide a robust statistical method for analyzing complex psychometric data.

Main Methods:

• Development of a Bayesian adaptive Lasso procedure for simultaneous variable selection.
• Incorporation of DIF effects directly within the MIRT framework.
• Utilizing simulation studies to evaluate the proposed method's performance.

Main Results:

• The proposed Bayesian adaptive Lasso method effectively estimates parameters, recovers item-trait relationships, and detects DIF effects.
• Simulation studies demonstrate the method's accuracy and reliability in identifying both item-trait associations and DIF.
• The unified framework successfully addresses the simultaneous detection of these psychometric properties.

Conclusions:

• The developed unified framework offers a powerful approach for analyzing item-trait relationships and DIF concurrently in MIRT.
• The Bayesian adaptive Lasso procedure provides an efficient and accurate tool for psychometricians and researchers.
• This method enhances the validity and interpretability of multidimensional assessment data, as demonstrated by the Eysenck Personality Questionnaire application.