Year: 2026
Type: article
Abstract: The exametrika package implements statistical models for test data engineering proposed by Shojima (2022), providing a comprehensive framework for analyzing educational test data. The package offers eleven models organized into three categories: classical test theory methods (CTT, IRT), latent structure models (latent class analysis, latent rank analysis, biclustering, and Ranklustering), and advanced network models (Bayesian network model, locally dependent LRA, locally dependent biclustering, bicluster network model, and deep bicluster network model). This paper focuses on latent structure models, including latent class analysis, latent rank analysis, biclustering, and Ranklustering, which combines rank-based ordering with clustering. For advanced network models, we demonstrate the application of Bayesian network models to test data. These models are designed to capture complex structures inherent in test data and deepen understanding of examinee abilities and item characteristics. The package employs EM algorithms for efficient and accurate parameter estimation. Through detailed explanations of theoretical backgrounds, implementation methods, and practical applications to educational test data, we demonstrate the utility of the diverse analytical approaches provided by the exametrika package.
Source: The R Journal
DOI: 10.32614/rj-2026-023
Authors: Koji Kosugi, Kojiro Shojima
Language: English
Field: Decision Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: diamond
OpenAlex ID: W7203663804