<![CDATA[The Asesmen Kompetensi Minimum (AKM) constitutes the cornerstone of Indonesia's national large-scale assessment framework, designed to measure foundational numeracy competencies across the student population. Selecting the most appropriate psychometric model for calibrating AKM items is critical for ensuring valid score interpretations, equitable measurement, and evidence-based instructional policy. This study presents an empirical comparison of three Item Response Theory (IRT) models—the one-parameter logistic (Rasch) model, the two-parameter logistic (2PL) model, and the three-parameter logistic (3PL) model—applied to a 30-item AKM numeracy instrument administered to 500 senior high school students in South Sulawesi, Indonesia. Parameter estimation, model data fit, and measurement precision were evaluated using marginal maximum likelihood (MML) methods. Results revealed that The Rasch model produced the lowest Akaike Information Criterion (AIC = 15,178.11) and Bayesian Information Criterion (BIC = 15,304.54), alongside the highest marginal test information (TIF = 5.427) and reliability (.844), indicating superior parsimony and precision relative to the 2PL and 3PL models. Item difficulty parameters ranged from b = −2.788 (Item 23) to b = 0.541 (Item 22), reflecting the adequate breadth of the numeracy construct. The 2PL yielded the smallest mean chi-square item misfit, whereas the 3PL introduced unnecessary parameter complexity without meaningful gain-in-fit. These findings suggest that the Rasch model is the preferred framework for operational AKM calibration, with practical guidance provided for contexts in which 2PL or 3PL models may be appropriate.]]>
Copyrights © 2026
