<![CDATA[This study aimed to evaluate the psychometric quality of a science assessment instrument designed for fifth-grade elementary school students in a deep learning instructional context using the Rasch model. This study used a descriptive quantitative approach with a one-parameter logistic Rasch model (1-PL) analyzed using Winsteps software version 5.10.4.0 applied to 20 dichotomous multiple-choice items administered to 29 fifth-grade elementary school students; the analysis covered item difficulty, construct validity, unidimensionality based on PCA of standardized residuals, person ability, person fit, and instrument reliability. The results showed strong reliability with a KR-20 coefficient of 0.89, person reliability of 0.84, and item reliability of 0.81; the raw variance explained by the measures reached 42.9% exceeding the 40% Rasch standard and all unexplained variance contrasts remained below 10% confirming that the instrument is unidimensional; 19 of the 20 items were valid while item A10 showed misfit (Outfit MNSQ = 2.45; ZSTD = 2.15); and the distribution of student ability indicated 20.69% in the high category, 62.07% moderate, and 17.24% low with six students exhibiting inconsistent response patterns. This study concludes that the science assessment instrument developed within the deep learning framework demonstrates adequate psychometric quality for measuring fifth-grade elementary school students' science competence. Keywords: Deep Learning, Item Response Theory, Rasch Model, Science Assessment, Unidimensionality, Construct Validity. ]]>
Copyrights © 2026
