Direct Estimation of Diagnostic Classification Model Attribute Mastery Profiles via a Collapsed Gibbs Sampling...

Area of Science:

• Educational Measurement
• Psychometrics
• Statistical Modeling

Background:

• Diagnostic classification models (DCMs) are widely used for assessing student mastery of attributes.
• Traditional DCM estimation methods can face challenges with parameter estimation and boundary issues.
• Efficient and accurate methods for inferring latent attribute mastery are crucial.

Purpose of the Study:

• To propose a novel collapsed Gibbs sampling algorithm for DCMs.
• To directly sample latent attribute mastery patterns, avoiding model parameter estimation.
• To address boundary problems in item parameter estimation.

Main Methods:

• Developed a collapsed Gibbs sampling algorithm that marginalizes model parameters.
• Implemented the algorithm for direct sampling of latent attribute mastery patterns.
• Conducted simulation studies to evaluate accuracy and computational efficiency.

Main Results:

• The collapsed Gibbs sampling algorithm accurately recovered true attribute mastery status across various conditions.
• The algorithm demonstrated superior computational efficiency compared to existing MCMC methods (e.g., JAGS).
• Real data analysis showed good classification agreement with previous findings.

Conclusions:

• The proposed collapsed Gibbs sampling algorithm offers an accurate and efficient approach for DCMs.
• This method effectively overcomes boundary issues in item parameter estimation.
• The algorithm shows promise for practical applications in educational assessment.