Safety analysis of highway merging and diverging areas using advanced econometric methods

Abstract

Considering the substantial social costs of road crashes, it is important to identify the road safety hazards at crash-prone locations, such as transition segments where vehicles merge onto or exit the main highway. Therefore, effective road safety measures can be developed to mitigate the crash risk. Conventional highway safety studies rely on statistical models based on historical crash data. However, they may have limitations because of the rare and random nature of crash occurrences. Additionally, some crucial information like dynamic traffic characteristics is often not available in the crash dataset. Furthermore, there are shortfalls in existing statistical models. Therefore, this study aims to develop advanced statistical and econometric methods for the safety analysis of highway merging and diverging areas, which have relatively high crash risk among all road entities. First, effects of possible influencing factors on the crash injury severity are different between single-vehicle and multi-vehicle crashes. Therefore, separate crash injury severity models should be established for single- and multi-vehicle crashes. Additionally, transferability assessment of the crash injury severity models across different time periods should be considered. To this end, the random parameter multinomial logit regression model with heterogeneity in means and variances is adopted to measure the association between crash injury severity and possible influencing factors at the highway ramp areas (including acceleration and deceleration lanes), with which the effect of unobserved heterogeneity is accounted for. Furthermore, partially constrained and temporal unconstrained modeling approaches are adopted to account for temporally shifting parameters. Results indicate that there are considerable differences in the effects of aberrant driver behavior, vehicle class, lighting condition and crash location on injury severity between single- and multi-vehicle crashes. Last but not least, out-of-sample prediction performance for the models of single- and multi-vehicle crashes are assessed. Second, a key assumption for multiple regression model is the exogeneity of the independent variables. However, some influencing factors that affect crash injury severity may correlate with other unobserved factors. As such, there could exist an endogenous effect of crash type on crash injury severity. Hence, a random parameter recursive bivariate probit model is proposed to model the crash type (hit-object and rollover) and crash injury severity at ramps simultaneously and to account for any endogenous effect of crash type. The results indicate that the proposed model is capable of capturing the correlation between unobserved factors and crash type. Endogeneity effect of crash type is incorporated in the crash severity model using a correlation parameter. Furthermore, other exogenous variables including road environment, crash location, and driver characteristics that affect the crash type and crash injury severity at the highway ramp areas were also identified. Third, information on dynamic traffic characteristics is often not available in historical crash dataset. To resolve this problem, vehicle trajectories can be captured using videos and image processing techniques in real-time safety analysis. In this study, a modified conflict risk indicator is developed to assess the safety of diverging, merging, and weaving movements of traffic, with which vehicles’ dimensions (width and length), and longitudinal and angular movements are considered. A correlated grouped random parameter logit model with heterogeneity in the means is established to capture the unobserved heterogeneity, with additional flexibility, at road user level for the association between conflict risk and influencing factors. Prevalence and severity of both rear-end and sideswipe traffic conflicts are examined. Results indicate that toll collection type, vehicle location, average longitudinal speed, angular speed, acceleration, and vehicle class all affect the risk of traffic conflicts. Overall, contribution of this study is twofold. First, advanced statistical and econometric methods are developed to account for several prevalent issues in safety analysis. Second, alternate safety indicators are adopted to proxy the crash risk at the highway merging and diverging areas. Findings should shed light on effective traffic management and control measures like variable speed limits and message signs that can mitigate the crash risk at the hazardous locations.