Analysis of injury severity in rear-end crashes on an expressway involving different types of vehicles using random-parameters logit models with heterogeneity in means and variances

ABSTRACT

To examine the difference in contributing factors of rear-end crashes of different injury severity involving different types of vehicles, this paper proposed random-parameters multinomial logit models with heterogeneity in means and variances. A three-year (2017–2019) rear-end crash data collected from Beijing-Shanghai Highways in China was used to calibrate the models. The rear-end crashes were classified as five types (Car-Car, Car-Truck, Truck-Truck, Truck-Car, Others). With two possible injury severity outcomes of medium/severe injury and light injury, a wide range of possible variables including crash, traffic, speed, geometric, and sight characteristics were considered in this study. Likelihood ratio tests revealed the rationality of adopting merged models using the data across three-year periods. Remarkably significant differences were shown in crashes involving different types of vehicles. The results accounting for the possible heterogeneity could be of value to roadway designers and traffic management departments seeking to promote highway safety and raise accurate safety countermeasures.

KEYWORDS:

• Rear-end crash type
• injury severity
• random-parameters multinomial logit approach
• unobserved heterogeneity in means and variances
• marginal effect

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The unobserved heterogeneity might be caused by misspecification of model, missing data, temporal or spatial instability, etc. To explain this problem more specifically, Mannering, Shankar, and Bhat (2016) presented examples with possible heterogeneous effects in explanatory variables affecting injury severity.

2. The model includes the mechanisms of endogenous and exogenous variables. The endogenous variable denotes one whose variation was caused by other variables in the model. And, the exogenous variable means a variable which could be expected to vary autonomously or independent of other variables. More detailed information about the endogenous and exogenous variables can refer to the handbook of Washington, Karlaftis, and Mannering (2020).

3. As Ye and Lord (2014) have concluded before, the lower number of observations might significantly affect the predictive accuracy of the random parameters model, with the minimum sample size of approximately 10,000.

Additional information

Funding

This study was supported by the Project of the National Natural Science Foundation of China (grant number 51768063 and 51868068).