Information matrix estimation procedures for cognitive diagnostic models.

Area of Science:

• Psychometrics
• Educational Measurement
• Statistical Modeling

Background:

• Cognitive Diagnosis Models (CDMs) are essential for understanding student mastery of attributes.
• Accurate estimation of the asymptotic covariance matrix is crucial for reliable parameter inference in CDMs.
• Existing covariance matrix estimators may not adequately account for all model parameters.

Purpose of the Study:

• To introduce and evaluate two novel methods for estimating the asymptotic covariance matrix in CDMs.
• To compare the performance of the new methods against existing approaches, particularly under varying degrees of model misspecification.
• To enhance the precision and robustness of standard error estimation for item parameters in CDMs.

Main Methods:

• Development of two new estimators: the inverse of the observed information matrix and a sandwich-type estimator.
• Analysis of the relationships between various information matrices and existing estimation approaches.
• Simulation studies to assess estimator performance under different model specification conditions.

Main Results:

• Both the observed information matrix and sandwich-type estimator provide consistent standard errors for item parameters when the CDM and Q-matrix are correctly specified or slightly misspecified.
• The sandwich-type estimator demonstrates superior robust performance compared to other methods when substantial model misspecification occurs.
• The new methods effectively incorporate both item and structural parameters, unlike some previous estimators.

Conclusions:

• The proposed inverse observed information matrix and sandwich-type estimators offer improved accuracy for asymptotic covariance matrix estimation in CDMs.
• The sandwich-type estimator is recommended for its robustness, particularly in scenarios with potential model misspecification.
• These advancements contribute to more reliable statistical inference in cognitive diagnosis.