Probabilistic models of motorcyclists’ injury severities in single- and multi-vehicle crashes
Introduction
In the US, motorcycle crash fatalities have been on a surprising upward trend. From an historic low of 2116 fatalities in 1997, fatalities have increased each year since, reaching a projected 4315 in 2005 (National Highway Traffic Safety Administration, 2006). There are potentially many reasons for this observed increase in fatalities. Motorcycle registrations have climbed significantly during the 1997–2005 time period; many states have repealed their helmet laws; and the average age of motorcyclists has increased with reflex and skill degradation a concern. With many factors likely involved in this disturbing upward trend in fatalities, there is a clear need for research to provide some insight into the relative effect of various factors so that appropriate countermeasures can be implemented to save lives.
The intent of this study is to develop probabilistic models of motorcycle crash-injury severity in an effort to provide additional insight into the factors that determine injuries to motorcyclists. We use data from the state of Indiana, whose fatality trends have followed national trends closely. With a database of all crashes reported to Indiana police from January 1, 2003 to October 15, 2005, we develop detailed multivariate analyses to determine the factors that significantly influence crash-injury severity.
In terms of appropriate statistical methods for the multivariate analysis of crash-injury severity data, recent literature indicates that a variety of methods have been used with an emphasis on car and truck crashes. For some examples, Abdel-Aty et al. (1998) used log-linear models to examine the relationship between driver age and crash characteristics, including severity; Farmer et al. (1997) used a binomial regression model to investigate the impact of vehicle and crash characteristics on injury severity in two-vehicle, side-impact crashes; O’Donnell and Connor (1996) assessed the probabilities of four levels of injury severity as a function of driver attributes using ordered logit and ordered probit specifications; Kockelman and Kweon (2002) used ordered probability models to investigate separate datasets for all crashes, two-vehicle crashes and single-vehicle crashes; Carson and Mannering (2001) developed multinomial logit models to examine the effect of ice-warning signs on crash-injury severity for different roadway functional classes; and Ulfarsson and Mannering (2004) explored differences in injury severity between male and female drivers in single and two-vehicle crashes involving passenger cars, pickups, sport-utility vehicles, and minivans using multinomial logit models.
Among the extant literature, at least two recent studies focused exclusively on motorcycle crash severity. Shankar and Mannering (1996) considered environmental, roadway, vehicular, and rider characteristics in their multinomial logit analysis of single-vehicle motorcycle crash-injury severity. And, Quddus et al. (2002) utilized ordered probit models to analyze motorcycle damage and injury severity resulting from crashes.
The review of the literature shows that two preferred approaches have emerged in the statistical modeling of crash-injury severity data, ordered probability models (ordered logit and probit) and unordered probability models (multinomial and nested logit). Because injury levels are typically progressive (ranging from no-injury to fatality), ordered probability models would seem to be a natural choice to account for the ordering of injury severities. O’Donnell and Connor (1996), Duncan et al. (1998), Renski et al. (1999), Khattak (2001), Kockelman and Kweon (2002), Khattak et al. (2002), Kweon and Kockelman (2003), Abdel-Aty (2003), Yamamoto and Shankar (2004) are some of the many that have used this technique. However, there are at least two potential problems with applying ordered probability models to injury-severity analysis. The first relates to the fact that non-injury crashes may be under-represented in police-reported crash data since lower injury levels make reporting to authorities less likely. The presence of underreporting in an ordered probability model can result in biased and inconsistent model coefficient estimates.1 In contrast, the coefficient estimates of an unordered multinomial logit probability model are consistent except for the constant term (see McFadden, 1981, Washington et al., 2003).
The second problem is more difficult to correct and relates to the restriction that ordered probability models place on variable influences. To see this, we follow the example used in Washington et al. (2003). Consider the effect that air-bag deployment has on occupant injury with four possible injury outcomes: no injury, non-incapacitating injury, incapacitating injury and fatality. The ordered probability model constrains the airbag-deployment indicator variable to either increase the probability of fatality (and subsequently decrease the probability of no injury) or decrease the probability of fatality (and subsequently increase the probability of no injury). This precludes the possibility that airbag deployment may simultaneously increase (or decrease) the probabilities of a fatality and no-injury because the deployment may save lives but may also cause minor injuries (non-incapacitating injuries) in doing so. Unordered probability approaches do not impose this constraint and such models have been applied in crash-injury severity analysis by numerous researchers (Shankar et al., 1996, Chang and Mannering, 1999, Carson and Mannering, 2001, Lee and Mannering, 2002, Ulfarsson and Mannering, 2004, Khorashadi et al., 2005) and will be the approach adopted in this study.
Section snippets
Methodology
In applying an unordered probability model to assess motorcyclists’ injury severity, we begin by defining a linear function that determines motorcyclist n's injury-severity outcome i as,where Xin is a vector of measurable characteristics (motorcyclist characteristics, roadway characteristics, and so on) that determine the injury severity for motorcyclist n, βi a vector of estimable coefficients, and ɛin is an error term accounting for unobserved effects influencing the injury
Data
Our data is drawn from all police-reported motorcycle crashes in the state of Indiana between January 1, 2003 and October 15, 2005. We also combined the Indiana State Police crash database with rider training records obtained from the American Bikers Aimed Toward Education (ABATE) of Indiana to provide a more comprehensive dataset for analysis. Although police provide estimates of crash-injury severity at the scene of the crash in one of five categories (no-injury, possible injury,
Single-vehicle crash-injury severity
A total of 2273 single-vehicle motorcycle crashes (occurring between January 1, 2003 and October 15, 2005 in the state of Indiana) were used for model estimation. We consider only motorcycle-operator injury levels (not passengers) and, of these crashes, 20% resulted in no-injury, 62% in non-incapacitating injury, 15% in capacitating injury and 3% in fatality. Using these data, we develop models to estimate the probability of the four discrete driver-injury severity outcomes conditioned on a
Multi-vehicle crash-injury severity
Multi-vehicle crashes involving motorcyclists have been identified as major safety concern with crash characteristics that vary considerably from those found in single-vehicle motorcycle crashes (Motorcycle Safety Foundation, 2000). Because motorcycles are less conspicuous than passenger cars or trucks, they are often more difficult to detect and their approaching speed is more difficult to determine, and many have argued that this is a major contributing factor to the high crash rate of
Conclusions
Motorcycle safety in Indiana is plagued by many of the same problems faced throughout the United States. The crash-injury severity analysis presented in this paper revealed several problem areas leading to more severe injuries: poor visibility (horizontal curvature, vertical curvature, darkness); unsafe speed (citations for speeding); alcohol use; not wearing a helmet; right-angle and head-on collisions; and collisions with fixed objects. There were some findings that motorcyclists may be
References (32)
- et al.
An assessment of the effect of driver age on traffic accident involvement using log-linear models
Accid. Anal. Prev.
(1998) Analysis of driver injury severity levels at multiple locations using ordered probit models
J. Saf. Res.
(2003)- et al.
The effect of ice warning signs on accident frequencies and severities
Accid. Anal. Prev.
(2001) - et al.
Analysis of injury severity and vehicle occupancy in truck- and non-truck-involved accidents
Accid. Anal. Prev.
(1999) - et al.
Two-vehicle side impact crashes: the relationship of vehicle and crash characteristics to injury severity
Accid. Anal. Prev.
(1997) - et al.
Misclassification of the dependent variable in a discrete-response setting
J. Economet.
(1998) - et al.
Driver aging and its effect on male and female single-vehicle accident injuries: some additional evidence
J. Saf. Res.
(2006) - et al.
Differences in rural and urban driver-injury severities in accidents involving large trucks: an exploratory analysis
Accid. Anal. Prev.
(2005) - et al.
Driver injury severity: an application of ordered probit models
Accid. Anal. Prev.
(2002) - et al.
Impact of roadside features on the frequency and severity of run-off-roadway accidents: an empirical analysis
Accid. Anal. Prev.
(2002)
Predicting the severity of motor vehicle accident injuries using models of ordered multiple choice
Accid. Anal. Prev.
An analysis of motorcycle injury and vehicle damage severity using ordered probit models
J. Saf. Res.
An exploratory multinomial logit analysis of single-vehicle motorcycle accident severity
J. Saf. Res.
Statistical analysis of accident severity on rural freeways
Accid. Anal. Prev.
The role frontal motorcycle conspicuity has in road accidents
Accid. Anal. Prev.
Differences in male and female injury severities in sport-utility vehicle, minivan, pickup and passenger car accidents
Accid. Anal. Prev.
Cited by (393)
Contributing factors to right-turn crash severity at signalized intersections: An application of econometric modeling
2024, International Journal of Transportation Science and TechnologyAlternative approaches to modeling heterogeneity to analyze injury severity sustained by motorcyclists in two-vehicle crashes
2024, Accident Analysis and PreventionFactors affecting injury severity in motorcycle crashes: Different age groups analysis using Catboost and SHAP techniques
2024, Traffic Injury Prevention