Regularized Latent Class Analysis with Application in Cognitive Diagnosis.

Area of Science:

• Psychometrics
• Statistical Modeling
• Educational Measurement

Background:

• Diagnostic classification models (DCMs) offer interpretable insights into latent attributes but often suffer from poor data fit due to oversimplification.
• Parameterized DCMs struggle to capture complex data patterns, leading to a trade-off between interpretability and goodness of fit.

Purpose of the Study:

• To develop a regularized latent class model that achieves a balance between model interpretability and goodness of fit.
• To enhance the flexibility of diagnostic classification models without sacrificing their practical meaning.

Main Methods:

• A regularized latent class model approach was proposed, starting with minimal data assumptions.
• An expectation-maximization-type algorithm was developed for efficient computation of the proposed model.
• The method incorporates regularization to reduce model complexity and improve data pattern capture.

Main Results:

• The proposed regularized latent class model demonstrates improved goodness of fit compared to traditional parameterized models.
• Simulation studies and a real-world application validated the model's effectiveness and theoretical properties.
• The approach successfully balances interpretability with enhanced flexibility in capturing data patterns.

Conclusions:

• The regularized latent class model offers a viable solution to the interpretability-fit trade-off in diagnostic classification.
• This method provides a more flexible yet interpretable tool for analyzing latent attributes and response data.
• The developed algorithm ensures efficient computation, making the approach practical for various applications.