Injury severity is a key aspect in road safety, especially at railway crossings. Traditional statistical methods such as ordered probit or ordered logit models provide adequate approaches to these analyses when plenty of covariates are available. However, these traditional approaches can be combined with other techniques, for instance, spatial methods and Bayesian analyses to provide better estimations by including the effects of covariates that are not explicitly included in the models, among other benefits. Bayesian analyses are useful when low crash frequencies and small sample sizes are present in the data, which is common for railway crashes. A case from San José, Costa Rica is used to demonstrate the use of the proposed methodology. Spatial analyses were implemented through conditional autoregressive (CAR) models up to third-ordered neighbors. A Bayesian approach was used to obtain more accurate estimations for the model coefficients and parameters. The results show how this methodology provides better overall results than a non-spatial ordered probit model. The first-order CAR model was determined to be the best overall model and therefore is preferred.
Bayesian spatial models of injury severity at railway crossings
Journal of Transportation Safety & Security ; 13 , 6 ; 680-693
2021-06-03
14 pages
Article (Journal)
Electronic Resource
Unknown
Bayesian Spatial Models of Crash Frequency at Highway–Railway Crossings
Transportation Research Record | 2017
|Transmission line railway crossings
Engineering Index Backfile | 1909
Amiesite paved railway crossings
Engineering Index Backfile | 1928
|