The Differential Contributions of Aural, Written Receptive, and Written Productive Vocabulary to the Guangdong Version of China's National Matriculation English Test

Abstract Vocabulary knowledge is multidimensional, yet how its component dimensions jointly contribute to integrated, high-stakes language assessments remains under-examined. This study investigated how aural receptive vocabulary (LVLT), written receptive vocabulary breadth (VST), and written productive vocabulary (PVLT) predict performance on the Guangdong version of China's National Matriculation English Test (NMET), where the long‑standing computer‑based listening‑speaking test (CELST) has been weighted at 20 points within the 150‑point total since the 2021 provincial scoring reform. Three consecutive cohorts of first-year university students in Guangdong (2023–2025; final N = 502) completed the three vocabulary measures, and their NMET total scores were analyzed separately by cohort using correlational analyses and hierarchical regressions with bootstrap validation. Aural receptive vocabulary was the only dimension significantly correlated with NMET across all three test forms ( r = .385–.562) and accounted for the largest unique variance (14.8%–31.5%; β = .382–.461), with mid-frequency (2K–3K) bands contributing most. Written receptive and productive vocabulary showed fragile, cohort-dependent associations that attenuated to non-significance in 2025 and contributed negligible incremental variance once aural vocabulary was controlled.