MIMIC Methods for Detecting DIF Among Multiple Groups: Exploring a New Sequential-Free Baseline Procedure.

Area of Science:

• Psychometrics
• Educational Measurement
• Statistical Modeling

Background:

• Differential Item Functioning (DIF) is crucial for fair assessments.
• Multiple Indicators Multiple Causes (MIMIC) models are used for DIF detection.
• Understanding DIF sources related to background variables like gender and ethnicity is important.

Purpose of the Study:

• To evaluate the effectiveness of different Multiple Indicators Multiple Causes (MIMIC) methods for detecting uniform and non-uniform Differential Item Functioning (DIF).
• To investigate the impact of violated assumptions, specifically equal factor variance, on MIMIC model performance.
• To assess the ability of MIMIC methods to identify the sources of DIF when influenced by background variables.

Main Methods:

• A simulation study was designed to generate data with known DIF.
• Three MIMIC implementations were compared: constrained baseline, free baseline, and sequential-free baseline.
• DIF was simulated from combinations of gender and ethnicity variables.

Main Results:

• The sequential-free baseline method demonstrated excellent Type I error control and power when the equal factor variance assumption was met.
• All tested MIMIC methods exhibited inflated Type I error when the equal factor variance assumption was violated.
• Identifying the specific sources of DIF proved challenging, particularly with interacting background variables.

Conclusions:

• The sequential-free baseline MIMIC method is a robust approach for DIF detection when factor variance equality holds.
• Violations of the equal factor variance assumption significantly compromise the accuracy of MIMIC-based DIF detection.
• While MIMIC models can detect DIF, pinpointing its origins, especially in complex multi-group scenarios, requires further methodological development.