Abstract：
This study investigates the factors contributing to bicycle accidents, focusing on four types of bicycle lanes and other exposure and built environment characteristics of census blocks. Using Seoul as a case study, three years of bicycle accident spot data from 2018 to 2020 was collected, resulting in 1,330 bicycle accident spots and a total of 2,072 accidents. The geographically weighted Poisson regression (GWPR) model was used as a methodological approach to investigate the spatially varying relationships between the accident frequency and explanatory variables across the space, as opposed to the Poisson regression model. The results indicated that the GWPR model outperforms the global Poisson regression model in capturing unobserved spatial heterogeneity. For example, the value of deviance that determines the goodness of fit for a model was 0.244 for the Poisson regression model and 0.500 for the far better-fitting GWPR model. Further findings revealed that the factors affecting bicycle accidents have varying impacts depending on the location and distribution of accidents. For example, despite the presence of bicycle lanes, some census blocks, particularly in the northeast part of the city, still pose a risk for bicycle accidents. These findings can provide valuable insights for urban planners and policymakers in developing bicycle safety measures and regulations.
摘要：
这项研究调查了导致自行车事故的因素，重点研究了四类自行车道和其他暴露和建成环境特征的普查街区。以首尔为例，收集了2018年至2020年三年的自行车事故现场数据，导致1,330个自行车事故现场和2,072起事故。地理加权泊松回归（GWPR）模型被用作方法论方法，以研究事故频率和整个空间的解释变量之间的空间变化关系，与泊松回归模型相反。结果表明，GWPR模型在捕获未观察到的空间异质性方面优于全局泊松回归模型。例如,确定模型拟合优度的偏差值对于Poisson回归模型为0.244，对于拟合更好的GWPR模型为0.500。进一步的发现表明，影响自行车事故的因素会根据事故的位置和分布而产生不同的影响。例如,尽管有自行车道，一些人口普查街区，特别是在城市的东北部，自行车事故仍然存在风险。这些发现可以为城市规划者和决策者制定自行车安全措施和法规提供有价值的见解。
