<![CDATA[Penelitian ini mengembangkan dan mengevaluasi secara psikometrik instrumen sikap peduli budaya mahasiswa dalam pembelajaran fisika berbasis potensi lokal dan etnosains. Instrumen awal 32 butir mencakup kesadaran, keterbukaan, pengetahuan, dan empati. Pengujian melibatkan 243 mahasiswa calon guru fisika dari tiga perguruan tinggi di Daerah Istimewa Yogyakarta. Validitas struktur dianalisis menggunakan confirmatory factor analysis ordinal dengan estimator WLSMV, dilanjutkan estimasi reliabilitas dan graded response model. Model satu faktor tidak didukung, sedangkan model empat faktor berkorelasi pada versi 22 butir menunjukkan kesesuaian sangat baik, CFI = 1,000, TLI = 1,000, RMSEA = 0,000, dan SRMR = 0,036. Alpha ordinal berkisar 0,911–0,949, composite reliability 0,911–0,950, dan AVE 0,674–0,747. Analisis GRM menunjukkan threshold berurutan, tanpa item misfit dan ketergantungan lokal, dengan informasi tertinggi pada rentang sifat laten menengah. Empat skor dimensi direkomendasikan sebagai keluaran utama, sedangkan skor total digunakan sebagai ringkasan sekunder. Development of a Cultural Care Attitude Instrument for Local-Potential and Ethnoscience-Based Physics Learning This study developed and psychometrically evaluated a cultural care attitude instrument for local-potential- and ethnoscience-based physics learning. The initial 32-item instrument covered awareness, openness, knowledge, and empathy. Empirical testing involved 243 pre-service physics teachers from three universities in the Special Region of Yogyakarta, Indonesia. Internal structure was assessed using ordinal confirmatory factor analysis with the WLSMV estimator, followed by reliability estimation and graded response modeling. The unidimensional model was unsupported, whereas the correlated four-factor model for the 22-item version showed excellent fit, CFI = 1.000, TLI = 1.000, RMSEA = 0.000, and SRMR = 0.036. Ordinal alpha ranged from 0.911 to 0.949, composite reliability from 0.911 to 0.950, and average variance extracted from 0.674 to 0.747. The graded response model indicated ordered thresholds, no item misfit or local dependence, and maximum information at moderate latent-trait levels. Four dimensional scores are recommended as the primary outputs, with the total score used only as a secondary summary.]]>
