Evaluation of MIMIC-Model Methods for DIF Testing With Comparison to Two-Group Analysis.

Area of Science:

• Psychometrics
• Educational Measurement
• Statistical Modeling

Background:

• Differential item functioning (DIF) is critical for ensuring test fairness across diverse groups.
• Existing DIF detection methods, like 2-group item response theory (IRT), have limitations, particularly with small focal groups.
• The accuracy and sample size requirements for multiple-indicator multiple-cause (MIMIC) models in DIF testing are not well-established.

Purpose of the Study:

• To examine the accuracy of MIMIC structural equation models for DIF testing when the focal group sample size is small.
• To compare the performance of MIMIC models against traditional 2-group IRT methods for DIF detection.
• To provide recommendations for applying MIMIC methods in DIF testing scenarios with limited focal group data.

Main Methods:

• Utilized multiple-indicator multiple-cause (MIMIC) structural equation models, parameterized as item response models.
• Employed simulation studies to assess the accuracy of MIMIC methods under small focal group conditions.
• Compared MIMIC model results with those from 2-group item response theory (IRT) analyses.

Main Results:

• MIMIC models demonstrated superior accuracy in detecting uniform DIF compared to 2-group IRT, particularly with small focal-group samples.
• The accuracy of MIMIC methods was validated for both binary and 5-category ordinal response data.
• Results support the utility and robustness of the MIMIC approach for DIF testing in challenging sample size situations.

Conclusions:

• The MIMIC modeling approach is a valuable and accurate tool for differential item functioning (DIF) testing, especially when dealing with small focal groups.
• Findings suggest that MIMIC models provide more reliable DIF detection than 2-group IRT under these conditions.
• The study offers practical guidance for researchers and test developers on implementing MIMIC methods for enhanced measurement fairness.