Modeling non-parametric effects of two-vehicle speed on crash risk at intersections: Leveraging two-dimensional additive logistic regression beyond univariable approach

AbstractUnderstanding the relationship between vehicle speed and the risk of sustaining a life-threatening injury has garnered continual attention. This study seeks to gain deep insight into the relationship between two-vehicle speeds and the risk of serious injury at intersections. The 2016–2018 crash data that occurred at intersections from the US Crash Report Sampling System (CRSS) were examined. We present a more general framework that allows the crash risk to be simultaneously linked to a universal two-dimensional variable of two-vehicle speeds, instead of the one-dimensional variable of impact speed calculated according to crash types in the existing literature. The results indicate that the risk of serious injury for head-on crashes in the medium-speed zone is mainly influenced by the faster vehicle although having little relation to the slower vehicle. More importantly, we find that the marginal relationship between the two-vehicle speeds and the crash risk is non-monotonic for angle and rear crashes. Finally, appropriate measures are suggested to reduce the crash risk at intersections, including alerting the driver not to cross intersections at exceedingly low speed, assisting the driver in making an emergency response at a medium speed, and warning the driver not to operate at a very high speed.Keywords: two-vehicle speedsrisk of serious injuryintersection crashtwo-dimensionalnon-parametric effect Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1 Thin plate splines are a form of smoothing spline that finds utility in visualizing intricate relationships between continuous predictors and response variables (Hutchinson, Citation1998). Their versatile nature makes them particularly suitable for assessing the collective impact of two continuous predictors on a singular outcome, owing to their ability to capture multidimensional patterns (Pedersen et al., Citation2019).Additional informationFundingThis research was supported by the Key Program of the National Natural Science Foundation of China(No. 62333016) and China Scholarship Council (No.202307090082).