An Improved Strategy for Bayesian Estimation of the Reduced Reparameterized Unified Model

Area of Science:

• Cognitive diagnosis modeling
• Bayesian statistics
• Psychometrics

Background:

• The reduced reparameterized unified model (rRUM) is a widely used cognitive diagnosis model.
• Existing computational methods for rRUM may lack efficiency or require complex implementation.

Purpose of the Study:

• To develop a novel Bayesian formulation for the rRUM.
• To implement an efficient computational strategy for parameter estimation.

Main Methods:

• A Bayesian formulation of the rRUM using latent response data augmentation.
• A Gibbs sampling algorithm to approximate the posterior distribution of model parameters.
• A Monte Carlo simulation study to evaluate performance.

Main Results:

• The Bayesian rRUM formulation demonstrated accurate parameter recovery.
• The Gibbs sampling algorithm showed faster convergence and larger effective sample sizes compared to Metropolis-Hastings.
• The developed method is provided as an accessible R package ('rRUM').

Conclusions:

• The proposed Bayesian rRUM offers an accurate and computationally efficient alternative for cognitive diagnosis.
• The R package facilitates the application of this advanced psychometric model by researchers.