Incorporating informative priors in modeling crash severity of vulnerable road users: A case study in Dar Es Salaam, Tanzania

Abstract

Vulnerable road users (VRUs) which include pedestrians, bicyclists, and motorcyclists, face an increasing risk of severe injuries and fatalities in traffic crashes. Among factors, the inadequate infrastructure, poor enforcement of traffic laws, and adverse environmental conditions have persistently been cited. Addressing these challenges requires data-driven safety interventions that can provide accurate, probabilistic insights into crash severity factors. Traditional statistical models often struggle with small sample sizes and fail to incorporate historical crash trends, limiting their predictive capabilities. To overcome these limitations, this study applied a Bayesian logistic regression model with informative priors to investigate factors influencing crash severity among VRUs along the Kimara-Kibaha section of the Morogoro road segment in Dar es Salaam, Tanzania. Crash data from 2014 to 2022, collected manually from police reports, were analyzed to identify critical risk factors. The years 2014 and 2015 were treated as historical data to inform the Bayesian prior distributions, enhancing the model's predictive power, while crash data from 2016 to 2022 was considered for analysis. The results revealed that inclement weather conditions, angle collisions, and crashes occurring at uncontrolled junctions significantly increased the likelihood of fatal outcomes in a crash. Conversely, crashes on curved road alignments, yield-controlled junctions, and on-road impacts were associated with reduced severity. The Bayesian framework provided probabilistic insights into these relationships, offering a robust approach to understanding crash dynamics in low-income settings. These findings underscore the need for targeted infrastructure improvements, enhanced traffic law enforcement, and public safety campaigns to mitigate VRU crash severity in Tanzania.