Implementation of an EM-MMLE-Based Excel Item Response Theory System and Comparison of Parameter Estimation Performance with the R-mirt Package
The Korean Society of Human and Nature · 30.07.2026 · 7 (2) · 571-595
DOI: 10.54913/hn.2026.7.2.571 ↗Kurz gefasst
Fachartikel in der Zeitschrift The Korean Society of Human and Nature, online erschienen am 30.07.2026. Verfasst von Changho Han.
Abstract (Originalsprache)
The purpose of this study is to develop an Excel-based Item Response Theory (IRT) analysis system, referred to as E-IRT, capable of estimating item parameters and ability parameters based on the Two-Parameter Logistic (2PL) model, and to evaluate its parameter estimation performance through comparison with the R *mirt* package. Existing IRT analysis tools often rely on statistical software environments or commercial applications, which may limit their accessibility and practical use in educational and operational settings. To address this limitation, a lightweight IRT analysis system was implemented using Visual Basic for Applications (VBA) within the widely available Microsoft Excel environment. The E-IRT system employs the Marginal Maximum Likelihood Estimation (MMLE) method based on the Expectation-Maximization (EM) algorithm for item parameter estimation. BayesiAn prior distributions were incorporated to improve the stability of discrimination and difficulty parameter estimates. In addition, the Expected A Posteriori (EAP) method was applied for ability estimation. To evaluate the performance of the proposed system, both simulated datasets with a uniform score distribution representing a non-normal distribution and real examination datasets with score distributions approximating normality were analyzed. The item parameter and ability estimates obtained from E-IRT were compared with those produced by the R *mirt* package. The experimental results showed that, for real examination datasets, the difficulty parameters and ability estimates produced by the two systems were generally similar, and the reliability coefficients (Cronbach’s α) also demonstrated comparable values. In contrast, relatively large differences in discrimination parameter estimates were observed for the non-normally distributed simulated datasets. These differences appear to be attributable to the characteristics of the estimation algorithms under non-normal distribution conditions. Furthermore, as the sample size increased, the parameter estimates produced by the two systems tended to converge more consistently and stably. This study demonstrates that IRT-based item analysis can be effectively implemented within a general Excel environment without relying on commercial IRT software. The findings suggest the feasibility of a lightweight item analysis system that can be readily applied in educational institutions and practical assessment settings.
Zitieren
APA
Han, C. (2026). Implementation of an EM-MMLE-Based Excel Item Response Theory System and Comparison of Parameter Estimation Performance with the R-mirt Package. The Korean Society of Human and Nature, 7(2), 571-595. https://doi.org/10.54913/hn.2026.7.2.571
Vancouver
Han C. Implementation of an EM-MMLE-Based Excel Item Response Theory System and Comparison of Parameter Estimation Performance with the R-mirt Package. The Korean Society of Human and Nature. 2026;7(2):571-595. doi:10.54913/hn.2026.7.2.571
BibTeX
@article{Han2026,
author = {Han, Changho},
title = {Implementation of an EM-MMLE-Based Excel Item Response Theory System and Comparison of Parameter Estimation Performance with the R-mirt Package},
journal = {The Korean Society of Human and Nature},
year = {2026},
volume = {7},
number = {2},
pages = {571--595},
doi = {10.54913/hn.2026.7.2.571}
}
EUVIMED