Risk analysis of freight train collisions in the United States, 2000 to 2014

Yanlei Wang, Shuang Xu, Xiang Liu

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Train accidents damage infrastructure and rolling stock, disrupt operations, and may result in casualties and environmental damage. While the majority of previous studies focused on the safety risks associated with train derailments or highway-rail grade crossing collisions, much less work has been undertaken to evaluate train collision risk. This paper develops a statistical risk analysis methodology for freighttrain collisions in the United States between 2000 and 2014. Negative binomial regression models are developed to estimate the frequency of freight-train collisions as a function of year and traffic volume by accident cause. Train collision severity, measured by the average number of railcars derailed, varied with accident cause. Train collision risk, defined as the product of collision frequency and severity, is predicted for 2015 to 2017, based on the 2000 to 2014 safety trend. The statistical procedures developed in this paper can be adapted to various other types of consequences, such as damage costs or casualties. Ultimately, this paper and its sequent studies aim to provide the railroad industry with data analytic tools to discover useful information from historical accidents so as to make risk-informed safety decisions.

Original languageEnglish (US)
Title of host publication2016 Joint Rail Conference, JRC 2016
PublisherAmerican Society of Mechanical Engineers
ISBN (Electronic)9780791849675
StatePublished - 2016
Event2016 Joint Rail Conference, JRC 2016 - Columbia, United States
Duration: Apr 12 2016Apr 15 2016

Publication series

Name2016 Joint Rail Conference, JRC 2016


Other2016 Joint Rail Conference, JRC 2016
Country/TerritoryUnited States

All Science Journal Classification (ASJC) codes

  • Transportation


Dive into the research topics of 'Risk analysis of freight train collisions in the United States, 2000 to 2014'. Together they form a unique fingerprint.

Cite this