A Comparison of the Separate and Concurrent Calibration Methods for the Full-Information Bifactor model

Area of Science:

• Psychometrics
• Educational Measurement
• Statistical Modeling

Background:

• Multidimensional item response theory (MIRT) models require linking for item parameter estimation across nonequivalent groups.
• Existing research on linking methods primarily focuses on full MIRT models, with limited investigation into bifactor models.

Purpose of the Study:

• To describe separate and concurrent calibration methods for bifactor models.
• To compare the effectiveness of three linking methods for bifactor models via simulation.

Main Methods:

• Detailed descriptions of two separate calibration methods and one concurrent calibration method for bifactor models.
• Simulation study to compare the linking accuracy of the described methods.

Main Results:

• Concurrent calibration demonstrated superior accuracy in linking item parameters compared to separate calibration methods.
• Concurrent calibration showed better recovery of item parameters, item characteristic surfaces, and expected score distributions.

Conclusions:

• Concurrent calibration is recommended for bifactor models to ensure accurate and consistent item parameter estimation.
• The findings provide valuable insights for researchers and practitioners using bifactor models in item response theory.