Evaluating the long-term urban effects of high-speed rail in Japan: An integrated approach using synthetic difference-in-differences and double/debiased machine learning

• Evaluates long-term urban effects of Japan’s Shinkansen network using robust causal inference methods (SDID, DDML). • Integrates multiple machine learning models via stacked generalization to improve causal policy evaluation. • Reveals significant heterogeneity in urban responses to high-speed rail, providing insights for spatial policy. • Shows that advanced causal methods (SDID, DDML) yield more reliable estimates than traditional DID models. • Offers evidence-based policy implications and methodological guidance for large-scale transport investment. This study evaluates the long-term impacts of Japan’s high-speed rail (HSR) network on urban development, using long panel data from 1960 to 2020. To address methodological limitations of traditional econometric approaches, such as difference-in-differences (DID) and propensity score matching-DID, especially in heterogeneous urban contexts, we adopt two advanced causal inference techniques: synthetic difference-in-differences (SDID) and double/debiased machine learning (DDML). To increase estimation robustness, our DDML framework uses stacked generalization, combining multiple machine-learning models into a single predictive ensemble. This method captures complex, nonlinear relationships and strengthens causal identification in high-dimensional settings. Our findings demonstrate that, overall, the introduction of HSR has greatly promoted long-term urban expansion. However, further analyses uncover considerable heterogeneity in treatment effects depending on the timing of network expansion, specific Shinkansen routes, and station typologies. Importantly, SDID proves to be highly robust across scales, from national multi-city evaluations to micro-level assessments of individual metropolitan areas, making SDID a powerful methodological tool for both macro and localized urban impact studies. These findings highlight the value of advanced causal inference techniques in capturing the nuanced, dynamic, and context-dependent effects of large-scale infrastructure investments, offering practical implications for future HSR planning and spatial policy evaluation.