Variational Estimation for Multidimensional Generalized Partial Credit Model

Area of Science:

• Psychometrics
• Statistical Modeling

Background:

• Multidimensional item response theory (MIRT) models are increasingly important in psychometrics.
• Existing efficient algorithms primarily focus on dichotomous MIRT models.
• Robust and efficient algorithms for polytomous MIRT models are lacking.

Purpose of the Study:

• To develop a novel and efficient algorithm for estimating multidimensional generalized partial credit models.
• To address the gap in computational methods for polytomous MIRT.

Main Methods:

• A Gaussian variational estimation algorithm was developed.
• The algorithm was tested using simulation studies.
• The algorithm was applied to two real-world datasets.

Main Results:

• The proposed algorithm demonstrated fast estimation performance.
• The algorithm showed accurate results in simulations and real data analyses.
• This method provides a viable solution for polytomous MIRT.

Conclusions:

• The novel Gaussian variational estimation algorithm is effective for multidimensional generalized partial credit models.
• This approach offers an efficient and robust solution for psychometric analysis.
• The findings advance the computational methods available for complex item response theory models.