A neural network paradigm for modeling psychometric data and estimating IRT model parameters: Cross estimation...

Area of Science:

• Psychometrics
• Educational Measurement
• Artificial Intelligence in Behavioral Science

Background:

• Item Response Theory (IRT) models are crucial for analyzing educational and psychological test data.
• Accurate estimation of person ability and item parameters is essential for test validity and fairness.
• Traditional IRT parameter estimation methods can be computationally intensive and may struggle with complex datasets.

Purpose of the Study:

• To introduce and evaluate a novel artificial intelligence (AI) approach, the Cross Estimation Network (CEN), for IRT model parameter estimation.
• To assess the performance of CEN in fitting datasets and estimating both person and item parameters.
• To demonstrate the adaptability of CEN for new, unseen response patterns.

Main Methods:

• Development of the Cross Estimation Network (CEN), comprising a person network (PN) and an item network (IN).
• PN processes respondent patterns to estimate latent ability; IN processes item patterns to estimate item parameters.
• Four simulation studies and one empirical study were conducted to evaluate CEN's performance on the two-parameter logistic model under diverse conditions.

Main Results:

• CEN demonstrated effective fitting of training data.
• Accurate estimation of both person ability and item parameters was achieved.
• The trained PN and IN exhibited AI principles, providing reliable evaluations for novel response patterns.

Conclusions:

• The Cross Estimation Network (CEN) offers a powerful and accurate AI-driven method for IRT parameter estimation.
• CEN shows promise in handling complex datasets and generalizing to new data, enhancing psychometric analysis.
• This approach advances the application of AI in educational and psychological measurement.