Joint modeling of pedestrian and motorcycle crashes in urban setting using copula approach

Abstract

Road traffic accidents is a growing concern in Bangladesh. The statistics revealed that around 40% of the reported accidents and 35% of fatalities took place in urban areas among which 58% of crashes are hitting pedestrians. Pedestrians are the most vulnerable road users in the context of Bangladesh. Rapid growth in personal vehicles is considered responsible for this growing accident rate. However, the boost in motorcycle users due to the introduction of the ride-sharing business should not be overlooked. So, it has become a vital issue to examine if there is any relation between the increasing motorcycle rates and pedestrian accidents. Motorcycles usually ride along the leftmost lane of the roadways and at times they jump onto the footpaths to cut through congestion, thereby sharing the same alley with the pedestrians. To address the effect of such interaction, this study has focused on developing a joint model of pedestrian and motorcycle accidents based on similar geometric features in the accident scene. Two different scenarios have been developed to observe the effect of dedicated pedestrian facilities; one for highways (where there are no footpaths) and another for urban areas (where sufficient mobility facilities for pedestrians are expected). Accidents happening on the national and regional highways have been considered under highway label and for urban area accidents happening inside Dhaka has been selected. The objective is to identify the factors that trigger accidents for pedestrian and motorcycle riders who are designated as vulnerable road users in the context of Bangladesh because of their small footprint on the roadway. The joint model intentions to find the underlying dependence structure between these two road users. The bivariate copula-based negative binomial modeling approach is adopted for modeling the dependence structure since Copula allows for nonparametric modeling, hence, making the dependence structure free from the structure of the marginal distributions. The data for this study has been collected from the Accident Research Institute (ARI), BUET for 2013-2015. Both univariate and bivariate models show that collision types play an important role in motorcycle accidents whereas pedestrian accidents are more influenced by the geometric properties of the roadways. Although the univariate negative binomial model finds staggered T junction and roundabout as significant factors but their effects on accident numbers are negative (coefficient values for staggered T: -1.06 and -1.57; for roundabout: -1.82 and -2.1 for pedestrian and motorcycle respectively) in case of highway accidents whereas the joint model captures the real scenario and their effects both increased and are positive (6.59 and 6.79 for staggered T; 2.89 and 4.07 for roundabout). For motorcycle accidents, right angle collisions are found to have a negative effect (-1.42) compared to the base case (head-on collision) as per the univariate model; in contrast, the joint model suggests right-angle collisions to have a high positive coefficient value (11.34), which appears to be coherent with the movement of motorcycles such as, sudden entry from blind spot to the front line at junctions. The joint model, therefore, apprehends the effects that stay hidden in the univariate scenario. For accidents inside Dhaka, although the coefficients have same direction (increase or decrease) for both univariate and bivariate cases, increase in their values (for example, for not straight section the coefficient values change from -1.2794 and -1.43 to -2.72 and -3.63 for pedestrian and motorcycle accidents respectively) in joint scenario signifies their impact and can help to rank them. The dependence parameter is 0.23 which suggests a positive dependence and increases to 1.02 and 0.84 respectively when the factors are incorporated in the joint model. The joint plot and the type of copula, Clayton both signify a lower tail dependence.