Incorporating the Q-Matrix Into Multidimensional Item Response Theory Models.

Area of Science:

• Psychometrics
• Educational Measurement
• Psychological Measurement

Background:

• Multidimensional Item Response Theory (MIRT) models estimate multiple latent traits from item responses.
• Current MIRT models may not explicitly define which items measure specific traits.
• Items in MIRT can measure varying numbers of latent traits, complicating interpretation.

Purpose of the Study:

• To integrate Q-matrices from diagnostic classification models into MIRT.
• To develop a hybrid MIRT model for explicit trait-item relationships.
• To demonstrate the utility and ease of this hybrid model in data analysis.

Main Methods:

• Incorporation of Q-matrix concept into existing MIRT framework.
• Development of a hybrid MIRT model.
• Application of Bayesian inference with the NUTS algorithm for parameter estimation.

Main Results:

• The proposed hybrid MIRT model effectively clarifies item-trait associations.
• Simulation studies validated the benefits of the Q-matrix integration.
• The Bayesian approach facilitated straightforward modeling of complex datasets.

Conclusions:

• The Q-matrix enhanced MIRT model offers a structured approach to defining item-trait relationships.
• This hybrid model improves clarity and applicability in educational and psychological assessments.
• Bayesian methods, particularly NUTS, provide an efficient way to implement these models.