Multidimensional Extension of Multiple Indicators Multiple Causes Models to Detect DIF

Area of Science:

• Psychometrics
• Educational Measurement
• Statistical Modeling

Background:

• Multiple Indicators Multiple Causes (MIMIC) models are established for detecting uniform differential item functioning (DIF).
• The MIMIC-interaction model extends this to detect both uniform and nonuniform DIF within unidimensional item response theory (IRT).

Purpose of the Study:

• To extend the MIMIC-interaction model to multidimensional IRT (MIRT) for DIF detection.
• To evaluate the performance of the MIRT MIMIC-interaction model under various simulation conditions, focusing on Type I error and power rates.

Main Methods:

• Simulated data were generated under varying conditions: DIF pattern/magnitude, test length, latent trait correlation, sample size, and latent mean differences.
• The performance of the multidimensional MIMIC-interaction model was assessed using Type I error and power rates.

Main Results:

• The model demonstrated higher power for detecting uniform DIF compared to nonuniform DIF.
• Increased anchor item length and sample size positively impacted DIF detection power.
• Equal latent mean conditions yielded higher power rates than conditions with differing latent means.

Conclusions:

• The multidimensional MIMIC-interaction model shows promise for identifying uniform DIF in MIRT.
• Its effectiveness in detecting nonuniform DIF warrants further research and validation.