Artificial Neural Networks Excel in Predicting Missingness in Psychometric Data
Date:
Oral presentation at The 2025 International Meeting of the Psychometric Society, Minneapolis, Minnesota, USA
Longfei Zhang1, Minjeong Jeon2, Ping Chen1
1 Collaborative Innovation Center of Assessment for Basic Education Quality, Beijing Normal University
2 School of Education & Information Studies, University of California Los Angeles
Abstract
Missing data is pervasive in educational and psychological assessments. Naive remedies—such as discarding incomplete cases or leaving them as is—often degrade research validity and misinform subsequent decisions. Recent advances in Artificial Neural Networks (ANNs) have demonstrated their efficacy in prediction tasks by leveraging observed features to infer unknown values. Building on this potential, we developed an iterative ANN-based imputation method—Iterative Neural Network Imputation (IterNNImp)—to accurately predict missing entries in psychometric data column by column. Simulation studies showed that IterNNImp reliably estimated missing responses, yielding accurate item means, inter-item covariances, and person and item parameters under Item Response Theory (IRT). Furthermore, the method was validated using multiple empirical datasets across educational and psychological contexts. We conclude by providing guidelines for implementing ANN-driven imputation and discussing directions for refining this novel approach.
Citation
Zhang, L., Jeon, M., & Chen, P. (2025, July). Artificial neural networks excel in predicting missingness in psychometric data [Oral presentation]. The 2025 International Meeting of the Psychometric Society, Minneapolis, Minnesota, USA.