Citation

Olufowobi O, Ivan J, Zhao S, Wang K, Eluru N. Transp. Res. Rec. 2024; 2678(2): 537-547.

Copyright

(Copyright © 2024, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.1177/03611981231175901

PMID

unavailable

Abstract

Traffic safety research will continue to create new and improved methods for the analysis of safety data. Even if these models perform well, the precise underlying crash mechanism remains unknown. The missing piece is a tool that may be used to evaluate how well a method identifies the cause-and-effect relationship in the data. To meet these safety analysis needs, a high-resolution disaggregate data generating process called realistic artificial data (RAD) was developed. This tool simulates crash incidence on transportation facilities, capturing real-world causal links between individual roadway characteristics and crashes. The objective of this study was to check if the stochasticity embedded in the RAD generation process will be consistent for different random seeds and miles of data generated from the tool. To accomplish this, 10 different datasets were generated from the RAD tool and estimated using the negative binomial model; parameter estimates from the model were checked using a revised Wald statistic. The t-statistic estimates showed that the differences among the parameter values across the dataset are within a statistically acceptable level. Given the stability of the tool, the RAD framework can be useful in addressing the known limitations and knowledge gap in assessing the extent to which a statistical method succeeds in identifying the cause-and-effect relationship in the data. This in return can help guide and improve the practical application of statistical methods and eventually lead to more effective safety countermeasures that can reduce highway-related injuries and fatalities.

Language: en