Citation

Aguero-Valverde J. Transp. Res. Rec. 2018; 2672(30): 90-98.

Affiliation

Programa de Investigación en Desarrollo Urbano Sostenible, Universidad de Costa Rica, Barrio Los Profesores, San José, Costa Rica Corresponding Author: Address correspondence to Jonathan Aguero-Valverde: jonathan.aguero@ucr.ac.cr

Copyright

(Copyright © 2018, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.1177/0361198118782249

PMID

unavailable

Abstract

The presence of spatial correlation in crash frequency data has being well documented. Most of the studies in highway safety used conditionally specified spatial errors, such as Conditional Autoregressive (CAR) random effects. An alternative to conditionally specified models is to choose directly a parametric functional form for the covariance matrix of the error term, also known as jointly specified models of spatial correlation. The main advantage of these models is that they do not depend on an arbitrary defined proximity matrix. The purpose of the study is to demonstrate the use of jointly specified spatial priors for Bayesian models of crash frequency using four types of correlation functions: exponential, rational quadratic, Gaussian, and Matérn. The frequency of equivalent-property-damage-only crashes in 162 intersections on Route 34 in Costa Rica was modeled. The results show that the four jointly specified models are similar in goodness-of-fit and significantly better than the CAR model. The estimated mean range varies from about 20 m to more than 60 m and the mean effective range varies between 100 m and 200 m. Most of the information on the posterior of the range comes from the data, since the posteriors diverge significantly from the uniform priors. There is not enough evidence to select a correlation function over the other three; further studies are, therefore, recommended to shed light on this issue.

Language: en