Spatiotemporal analysis of crash severity on rural highway: A case study in Anhui, China.

Traffic crashes are the result of the interaction between human activities and different socio-economic, geographical, and environmental factors, showing a temporal and spatial relationship. The temporal and spatial correlations must be characterized in crash severity studies, for which the geographically and temporally weighted ordered logistic regression (GTWOLR) model is an effective approach. However, existing studies using the GTWOLR model only subjectively selected a type of kernel function and kernel bandwidth, which cannot determine the best expression of the spatiotemporal relationship between crashes. This paper explores the optimal kernel function and kernel bandwidth considering the aforementioned problem to obtain the best GTWOLR model to analyze the crash data based on the crash data of rural highways in Anhui Province, China, from 2014 to 2017. First, the GTWOLR models with Gaussian or Bi-square kernel function and fixed (the spatiotemporal distance remains constant of local sample) or adaptive (the quantity of the local sample is constant) bandwidth are compared. Second, the log-likelihood and Akaike information criterion are used to compare the GTWOLR model with the ordered logistic regression (OLR) model. Finally, the spatial and temporal characteristics of the contributing factors in the best GTWOLR model are analyzed, and corresponding countermeasures for improving traffic safety on rural highways are proposed. Model comparison results reveal that although the difference was insignificant, the Bi-square kernel function with fixed bandwidth (BF)- GTWOLR model has a better goodness of fit than the GTWOLR models with other types of kernel function and bandwidth and the OLR model. The BF-GTWOLR model estimation results showed that eight factors, including pedestrian-vehicle crash, middle-aged driver, hit-and-run, truck, motorcycle, curve, slope and mountainous, passed the non-stationary test, indicating their varying effects on the crash severity across space and over time. As a crash severity modeling approach that effectively quantifies the spatiotemporal relationships in crashes, the BF-GTWOLR model, which adapts to crash data, may have implications for future research. In addition, the findings of this paper can help traffic management departments to propose progressive and targeted policies or countermeasures, so as to reduce the severity of rural highway crashes.

Applying latent class analysis to investigate rural highway single-vehicle fatal crashes in China.

Rural highways are an important component of highway networks in developing countries. The high fatality rates of single-vehicle crashes in these highways recently attracted increasing attention. Given that most studies on the factors that affect the severity of single-vehicle crashes in rural highways were conducted in developing countries, the present study investigated this issue in a Chinese setting by analyzing the single-vehicle crash data of rural highways in Anhui Province, China from 2014 to 2017. First, in consideration of the unobserved heterogeneity of crash data, a method that combines latent class analysis (LCA) and binary logistic regression (BLR), which is called LC-BLR, was applied to identify the significant factors that affect the severity of single-vehicle crashes in rural highways. Second, the goodness-of-fit and prediction accuracy of the LC-BLR model and the BLR model were compared. Results revealed that the performance of the former was more satisfactory than that of the latter. Finally, countermeasures were proposed based on the analysis of the main factors that affect each sub-class crash in the LC-BLR model. The LC-BLR model results indicated that collision typewas significant in all three sub-class models considered in the analysis, but the effects on crash severity varied. Several variables (e.g., driving license state, time of week, driver age) demonstrated a significant effect in a specific sub-class model, thereby indicating that these factors were only effective in mitigating the crash severity of one sub-class. The findings of this study can facilitate the development of cost-effective policies or countermeasures for reducing the severity of single-vehicle crashes in rural highways.