Abstract:
This study attempts to identify the influencing factors for triggering public bus crash
injury severity in Dhaka city where public buses alone were involved in 23 % of all the
crashes. Though there are some descriptive-based works in Bangladesh pertinent to public
bus safety, very few in-depth studies on the crash severities of public bus have been
conducted; those are however, mostly based on an old crash data. Hence, utilizing the
recent crash data (2017-2020) collected from the Accident Research Institute (ARI) of
Bangladesh University of Engineering and Technology (BUET), the primary goal of this
study is to discover the roadway and environment-related factors impacting the public bus
crash severity in the context of Dhaka city.
A prominent way to deal with crash injury severity is by using statistical modelling
techniques; the selection of these suitable methods often depends on the nature of data,
especially the response variables. R software environment has been adopted to facilitate
the analysis. In relation to the genre of police-reported public bus crash data, four
different established models namely, Multinomial Logit (MNL), Ordered Logit (OL),
Ordered Probit (OP) and Partial Proportional Odds (PPO) have been selected for the
study. All of these severity models were then applied on this crash data to investigate
public bus safety mechanism prevalent in Dhaka city. The analysis showed that pedestrians, bicyclists and motorcyclists are the most vulnerable
road user group (around 80%), as indicated by the all selected models. Lack of efficient
police controlled traffic in all the places (in some cases, 0% fatal incidents in policecontrolled areas), absence of dividers in two way roads (38.23% fatal vs 57.78% fatal
where there are no dividers), over speeding, lack of necessary safety parameters as per the
condition/geometry of roads etc. seemed to accelerate road traffic crashes. In addition, the
severity models (i.e., MNL, OL, OP, and PPO) were evaluated in terms of relevant
comparative parameters where MNL model is found to be more effective in terms of loglikelihood (-237) and PPO model fared better in terms of Akaike Information Criterion
ii
(AIC_529) and Bayesian Information Criterion (BIC_616). The models were further
evaluated on the significance of their predictors where collision type, junction type,
movement, road class, road geometry, surface quality, surface type and time are found to
be significant for triggering public bus related accidents in Dhaka city. Some viewpoints
related to pedestrian facilities and roadway improvement (safety features) have been
recommended for the decision makers for reducing both accident frequency and severity.
