Driver injury severity: an application of ordered probit models

https://doi.org/10.1016/S0001-4575(01)00028-8Get rights and content

Abstract

This paper describes the use of ordered probit models to examine the risk of different injury levels sustained under all crash types, two-vehicle crashes, and single-vehicle crashes. The results suggest that pickups and sport utility vehicles are less safe than passenger cars under single-vehicle crash conditions. In two-vehicle crashes, however, these vehicle types are associated with less severe injuries for their drivers and more severe injuries for occupants of their collision partners. Other conclusions also are presented; for example, the results indicate that males and younger drivers in newer vehicles at lower speeds sustain less severe injuries.

Introduction

Traffic crash modeling is helpful for assessing risk factors and design issues in roadway travel. To this end, logit and other discrete specifications have been used to model injury severity and ratios of crash counts, while least squares, Poisson, negative binomial, and other models have been applied to total crash counts.

The severity of injuries sustained by drivers involved in crashes is of considerable interest to policy makers and safety specialists. The General Estimates System (GES), a sample of US crashes collected by the National Highway Traffic Safety Administration (NHTSA), classifies injuries according to four levels: no injury, non-severe injury, severe injury, and fatal injury. In this paper, the probability of these levels of injury severity is examined by applying an ordered probit regression model, thereby recognizing the ordinality of injury level, the dependent variable. This model is discussed in detail in Section 4.

The regression models control for various driver, vehicle, and crash characteristics, and the data come from the 1998 GES data set. Before detailing the data sets, describing the model, and presenting the results, a brief review of the extensive crash-modeling literature is provided here.

Section snippets

Literature review

In the most recent and sophisticated modeling of crash-count totals, the Poisson and negative binomial are rather common model specifications. And logit-based models (e.g., a log-linear specification of count ratios) have been used to model counts across injury severity classes. A variety of explanatory variables are typically available in crash records, but many models aggregate their data and examine the effects of only a few such variables (e.g., gender and age).

There is an extensive

Data set

This study investigates the injury severity for all crashes, two-vehicle crashes and single-vehicle crashes. Therefore, three data sets were prepared for estimation. These all derive from the 1998 National Automotive Sampling System GES which covers 0.85% of all police-reported crashes in the US. This data set is intended to be nationally representative and comes from a sample of all police-reported crash records. Such crashes include property-damaging crashes, injury crashes, and fatal crashes.

Model specification

Ordered response models recognize the indexed nature of various response variables; in this application, driver injury severities are the ordered response. Underlying the indexing in such models is a latent but continuous descriptor of the response. In an ordered probit model, the random error associated with this continuous descriptor is assumed to follow a normal distribution.

In contrast to ordered response models, multinomial logit and probit models neglect the data's ordinality, require

Model results

Ordered probit regression models were estimated for driver injury status in all crashes, two-vehicle crashes and single-vehicle crashes, with and without speed variables. The six separate sets of results are shown in Table 2, Table 3, Table 4 and are discussed below. Note that no variables were removed from the model on the basis of low statistical significance; since all variables are of interest and expected to have some effect on injury severity, all were maintained in the final models.1

Conclusions

A variety of factors can come into play when vehicles crash on the road. In terms of the severity of injuries sustained by drivers, this work suggests that the manner of collision, number of vehicles involved, driver gender, vehicle type, and driver alcohol use play major roles. Rollover and head-on collisions are particularly serious, contributing to more severe injury levels than speed increases of 50 mph and more. And males tend to fare significantly better than females. In contrast, the

Acknowledgements

The authors wish to acknowledge the University of Texas’ Department of Civil Engineering and the Luce Foundation for their support of this work, and recognize all reviewers for their comments.

References (21)

There are more references available in the full text version of this article.

Cited by (424)

View all citing articles on Scopus
View full text