Predicting the Likelihood of Aging Pedestrian Severe Crashes Using Dirichlet Random-Effect Bayesian Logistic Regression Model

There is ample literature on factors that contribute to the injury severity of pedestrian-vehicle crashes. Nevertheless, coupled with a continuous growing aging population, there is limited information addressing predictors that influence the injury severity of pedestrian-vehicle crashes involving older pedestrians. As such, this study developed an injury severity model with improved prediction accuracy, and hence identified the risk factors that influence the severity of aging pedestrians. In particular, the Dirichlet random-effect logistic model (DRL) was used to account for unobserved heterogeneity across crash data. Unlike the conventional parametric random-effect logistic model (CRL), which assumes that the heterogeneity of data varies across individual observations, the approach applied herein is flexible, imposing a belief that the DRL can recognize clusters of unobserved heterogeneity of crash observations. Various predictive capability indicators were utilized to compare the basic logistic (BL), CRL, and DRL model performances. The DRL model outperformed the BL and CRL models in all performance metrics used. The accuracy of the DRL was found to be 90% versus 83% and 68% for CRL and BL models, respectively. Moreover, seven variables were found to significantly influence the severity of aging pedestrians at the 95% Bayesian Credible Interval. These variables include pedestrian age, alcohol involvement, first harmful event, vehicle movement, shoulder type, posted speed, and traffic volume. It is envisioned that the findings of this study can provide a better understanding of the contributing factors to the transportation agencies, which can assist in devising traffic crash risk reduction strategies, especially for elder pedestrians.

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ANB20 Standing Committee on Safety Data, Analysis and Evaluation.
  • Authors:
    • Kitali, Angela E
    • Kidando, Emmanuel
    • Sando, Thobias
    • Moses, Ren
    • Ozguven, Eren Erman
    • ORCID 0000-0001-6006-7635
  • Conference:
  • Date: 2018

Language

  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References; Tables;
  • Pagination: 17p

Subject/Index Terms

Filing Info

  • Accession Number: 01663327
  • Record Type: Publication
  • Report/Paper Numbers: 18-05742
  • Files: TRIS, TRB, ATRI
  • Created Date: Jan 8 2018 11:28AM