In the field of macro-level safety studies, road traffic safety is significantly related to socioeconomic factors, such as population, number of vehicles, and Gross Domestic Product (GDP). Due to different levels of economic and urbanization, the influence of the predictive factors on traffic safety measurements can differ between cities (or regions). However, such region-level or city-level heterogeneities have not been adequately concerned in previous studies. The objective of this paper is to adopt a novel approach for traffic safety analysis with a dataset containing multiple target variables and samples from different subpopulations. Based on a dataset with annual traffic safety and socioeconomic measurements from 36 major cities in China, we estimate single-output regression models, multi-output regression models, and clustering-based regression models. The results indicate that the 36 cities can be clustered into a metropolitan city class and a non-metropolitan city class, and the class-specified models can notably improve the goodness-of-fit and the interpretability of city-level heterogeneities. Specifically, we note that the effect of primary and secondary industrial GDP on traffic safety is opposite to that of tertiary industrial GDP in the metropolitan city class, while the effects of the two decomposed GDP on traffic safety are consistent in the non-metropolitan city class. We also note that the population has a positive effect on the number of fatalities and the number of injures in metropolitan cities but has no significant influence on traffic safety in non-metropolitan cities.
City-Level China Traffic Safety Analysis via Multi-Output and Clustering-Based Regression Models