Authentic Mathematics Assessment Using an Integrated Deep Learning and Adiwiyata Testlet Model for Elementary Schools and Madrasah Ibtidaiyah

Hamdi, Syukrul
Yuliana, Lia
Oktarina, Anisa Dwi
Hidayati, Kana
Arliani, Elly
Mu'minin Mz, Nurul

Jurnal Pendidikan Islam
Vol. 9 No. 1 (2026): Transformative Islamic Education in Pesantren and Madrasah

This study aimed to develop an authentic mathematics assessment for elementary schools in the form of a testlet-based instrument integrated with deep learning principles and the Adiwiyata context within the framework of transformative Islamic education in the Special Region of Yogyakarta. Employing basic research with an embedded mixed-methods design, the development process adapted the Plomp model and the instrument development framework of Oreondo and Antonio. Data were collected through teacher needs surveys, expert validation, readability testing, and field trials involving 462 fifth-grade students from elementary schools and Islamic elementary schools. Quantitative analyses included Content Validity Index (CVI), Aiken’s V, Cronbach’s Alpha, Classical Test Theory (CTT), and Item Response Theory (IRT) using the 2-PL Graded Response Model. The results indicate that the developed instruments meet acceptable psychometric standards, with Aiken’s V values ranging from 0.75 to 1.00 and high internal consistency for both the testlet instrument (Cronbach’s Alpha = 0.845) and the environmental awareness questionnaire (Cronbach’s Alpha = 0.850). Item analysis shows adequate discrimination and a structured progression of difficulty, although one item exhibited low discrimination in the 2-PL GRM, highlighting the importance of IRT-based diagnostics for testlet refinement. Descriptive findings reveal that students demonstrate high levels of environmental awareness, particularly in the knowledge and attitude dimensions, while mathematical achievement remains low on non-routine items. Correlation analysis shows no significant relationship between environmental awareness and mathematical ability. Methodologically, this study contributes a validated and contextually grounded assessment framework that integrates expert judgment, reliability analysis, and complementary CTT–IRT procedures. Theoretically, the findings reconceptualize authentic assessment as a diagnostic bridge rather than a direct causal link between affective values and cognitive performance, demonstrating that environmental concern functions as a potential cognitive resource only when explicitly activated within mathematical tasks.