Answers are grounded in this paper and cite their sources.
Road traffic casualties remain a major challenge for transportation safety governance, particularly in countries undergoing rapid motorization, transportation expansion, and changing climatic conditions. In China, provincial traffic accident casualties show pronounced temporal fluctuations and regional disparities, suggesting coupled associations with climate–environmental conditions, socioeconomic activity, and transportation expansion. Existing studies often examine these dimensions separately, with limited attention to their joint macro-level associations, nonlinear interactions, and regional heterogeneity. This study develops an integrated framework combining econometric benchmarks with interpretable machine learning. Using a temporally aligned provincial-quarterly data set covering 31 Chinese provinces from 2001 to 2020, a fixed-effects linear model with province, year, and quarter fixed effects, fixed-effects Poisson pseudomaximum likelihood (PPML), and common correlated effects models are used as statistical benchmarks, and an extreme gradient boosting (XGBoost) framework interpreted using Shapley additive explanations (SHAP) identifies key predictors, nonlinear responses, interaction patterns, and province-level association profiles. Results indicated that health personnel, private vehicles per household, wind speed, road mileage, nighttime lights, mean relative humidity, and passenger volume are the leading predictors. Predicted casualty levels tended to be higher when motorization, transportation supply, mobility demand, and regional activity intensity were simultaneously high, whereas climate contributions depend on exposure representation. Province-level SHAP clustering revealed distinct regional association profiles. These findings support differentiated traffic safety governance, transportation-system management, and climate-resilient transportation planning.
No takes yet. Share an insight, caveat, or question.
张嘉陵 et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: