Beyond Provincial Averages: A Spatial Diagnostic Framework for Educational Inequality in Vietnam’s National Upper Secondary Graduation Examination

Provincial average scores are widely used to monitor educational performance, but they may obscure how inequality is distributed within and across administrative units. Using annual data for all 63 provinces in Vietnam's national upper secondary graduation examination from 2017 to 2024, this paper develops a sequential spatial diagnostic framework for three compulsory subjects: Mathematics, Literature, and Foreign Language. The framework first evaluates the adequacy of provincial averages and then examines the spatial structure, persistence, and neighborhood-conditioned mobility of the between-province signal that remains visible at the administrative scale. Theil decomposition indicates that provincial mean scores capture only 6-18% of total examination-score inequality; the dominant 82-94% lies within provinces. This finding establishes the limits of average-based monitoring while motivating a more precise diagnosis of the provincial signal that policymakers observe and compare. The subsequent diagnostics show a clear subject-specific gradient. Mathematics displays the strongest rank stability, spatial clustering, and regime persistence; Literature is more mobile but increasingly spatially differentiated; and Foreign Language occupies an intermediate position. A 3×3 Spatial Markov analysis shows differences in observed transition profiles across neighborhood contexts, although estimates based on small conditional origin counts remain imprecise. The findings suggest that province-level educational monitoring would benefit from supplementary spatial diagnostics that track within-province dispersion, persistent low-performing clusters, and subjectspecific mobility profiles.