Exploring factors contributing to heavy vehicle crash severity in Bangladesh using alternative severity models

Abstract

Heavy vehicle crashes and consequent causalities have been victimizing the society and its people for decades. Realizing the gravity of this catastrophe, developed countries have always been prompt in taking efficacious safety measures to mitigate heavy vehicles’ crash injury severity. However, developing countries like, Bangladesh, who experience 80% of road traffic deaths, are being challenged highly in this sector. Studies have shown that heavy vehicles are involved in approximately 60% of the crashes in Bangladesh. Although these crashes have made an enduring place in the electronic and print media, very few in-depth researches have been conducted on heavy vehicle crash severities. So far, the best attempts made to face this deadly issue mostly include descriptive-based works. Hence, the focus of this study is to widen the existing horizon of heavy vehicle safety in Bangladesh in the foremost way possible. A prominent way to deal with heavy vehicles’ crash injury severity is by using statistical modeling techniques. The selection of these suitable methods often depends on the nature of data, especially the response variable. In relation to the genre of heavy vehicle crash data (2012 – 2015) collected from Accident Research Institute (ARI) of Bangladesh University of Engineering and Technology (BUET), four different established models namely, Multinomial Logit (MNL), Ordered Logit (OL), Ordered Probit (OP), and Partial Proportional Odds (PPO), have been selected for this thesis. All of these severity models were then applied on this crash data to investigate heavy vehicle safety mechanism prevalent in Bangladesh. This study mainly pivoted around geometric and environment-related features concerning heavy vehicle crashes. The significant outcomes apropos to these features include traffic control, collision type, movement, divider, weather, light condition, roadway geometry, surface condition, road class and location type. More specifically, traffic control systems principally police-controlled system in urban areas have been identified as persuasive in reducing fatal accidents (31% fatal accidents with traffic control system, whereas, 48% fatal accidents when there is no control). Moreover, urban city roads were found to mitigate fatal propensity in heavy vehicle crashes (36% fatality with police-controlled system vs 44% fatality with no control). Altogether urban regions have been estimated as safer areas compared to rural regions.

Analyzing pedestrian condition, heavy vehicles colliding with pedestrians have been identified as a direct cause to fatal consequences. This situation is even much worser in rural areas, where 90% accidents result in pedestrian fatality. National roads have also been found to heighten the propensity of pedestrian fatal crashes (91% fatality). Furthermore, pedestrians were also vulnerable to heavy vehicle crashes during dawn/dusk (89% fatal consequence) and at night (89% fatal crashes). In addition to areas where there are no standard traffic control systems, heavy vehicles are responsible for 90% pedestrian fatal injury severity. Head-on collision is found to be a prominent parameter in investigating heavy vehicle safety in national roads (83% fatal crashes). One-way roads were much safer compared to two-way roads. In case of two-way roads, the presence of divider impacted heavy vehicle safety to many folds. It was observed that, a two-way road with no divider generated almost 91% head-on collisions whereas a two-way road with no divider was accountable for almost 9% head-on collisions. Moreover, a rainy weather also increased the chances of fatal injury severity and was responsible for 74% of head-on crashes. Although wet surface has been identified as a mitigator of fatal injuries, yet it was behind 63% of fatal crashes compared to other injury severities. The severity models (i.e., MNL, OL, OP, PPO) have been compared at the end in terms of relevant comparative parameters (viz., log-likelihood at convergence, Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC) and feature significance). The comparative analysis showed that the PPO model is more effective compared to others in the context of available crash data in Bangladesh.