Analysis of driver and passenger crash injury severity using partial proportional odds models

The question of whether crash injury severity should be modeled using an ordinal response model or a non-ordered (multinomial) response model is persistent in traffic safety engineering. This paper proposes the use of the partial proportional odds (PPO) model as a statistical modeling technique that both bridges the gap between ordered and non-ordered response modeling, and avoids violating the key assumptions in the behavior of crash severity inherent in these two alternatives. The partial proportional odds model is a type of logistic regression that allows certain individual predictor variables to ignore the proportional odds assumption which normally forces predictor variables to affect each level of the response variable with the same magnitude, while other predictor variables retain this proportional odds assumption. This research looks at the effectiveness of this PPO technique in predicting vehicular crash severities on Connecticut state roads using data from 1995 to 2009. The PPO model is compared to ordinal and multinomial response models on the basis of adequacy of model fit, significance of covariates, and out-of-sample prediction accuracy. The results of this study show that the PPO model has adequate fit and performs best overall in terms of covariate significance and holdout prediction accuracy. Combined with the ability to accurately represent the theoretical process of crash injury severity prediction, this makes the PPO technique a favorable approach for crash injury severity modeling by adequately modeling and predicting the ordinal nature of the crash severity process and addressing the non-proportional contributions of some covariates.     

An empirical analysis of farm vehicle crash injury severities on Iowa's public road system

Farm vehicle crashes are a major safety concern for farmers as well as all other users of the public road system in agricultural states. Using data on farm vehicle crashes that occurred on Iowa's public roads between 2004 and 2006, we estimate a multinomial logit model to identify crash-, farm vehicle-, and driver-specific factors that determine farm vehicle crash injury severity outcomes. Estimation findings indicate that there are crash patterns (rear-end manner of collision; single-vehicle crash; farm vehicle crossed the centerline or median) and conditions (obstructed vision and crash in rural area; dry road, dark lighting, speed limit 55 mph or higher, and harvesting season), as well as farm vehicle and driver-contributing characteristics (old farm vehicle, young farm vehicle driver), where targeted intervention can help reduce the severity of crash outcomes. Determining these contributing factors and their effect is the first step to identifying countermeasures and safety strategies in a bid to improve transportation safety for all users on the public road system in Iowa as well as other agricultural states.     

Analysis of left-turn crash injury severity by conflicting pattern using partial proportional odds models

The purpose of this study is to examine left-turn crash injury severity. Left-turning traffic colliding with opposing through traffic and with near-side through traffic are the two most frequently occurring conflicting patterns among left-turn crashes (Patterns 5 and 8 in the paper, respectively), and they are prone to be severe. Ordered probability models with either logit or probit function is commonly applied in crash injury severity analyses; however, its critical assumption that the slope coefficients do not vary over different alternatives except the cut-off points is usually too restrictive. Partial proportional odds models are generalizations of ordered probability models, for which some of the beta coefficients can differ across alternatives, were applied to investigate Patterns 5 and 8, and the total left-turn crash injuries. The results show that partial proportional odds models consistently perform better than ordered probability models. By focusing on specific conflicting patterns, locating crashes to the exact crash sites and relating approach variables to crash injury in the analysis, researchers are able to investigate how these variables affect left-turn crash injuries. For example, opposing through traffic and near-side crossing through traffic in the hour of collision were identified significant for Patterns 5 and 8 crash injuries, respectively. Protected left-turn phasing is significantly correlated with Pattern 5 crash injury. Many other variables in driver attributes, vehicular characteristics, roadway geometry design, environmental factors, and crash characteristics were identified. Specifically, the use of the partial proportional formulation allows a much better identification of the increasing effect of alcohol and/or drug use on crash injury severity, which previously was masked using the conventional ordered probability models.     

Analysis of crash severities using nested logit model—Accounting for the underreporting of crashes

Recent studies in the area of highway safety have demonstrated the usefulness of logit models for modeling crash injury severities. Use of these models enables one to identify and quantify the effects of factors that contribute to certain levels of severity. Most often, these models are estimated assuming equal probability of the occurrence for each injury severity level in the data. However, traffic crash data are generally characterized by underreporting, especially when crashes result in lower injury severity. Thus, the sample used for an analysis is often outcome-based, which can result in a biased estimation of model parameters. This is more of a problem when a nested logit model specification is used instead of a multinomial logit model and when true shares of the outcomes-injury severity levels in the population are not known (which is almost always the case). This study demonstrates an application of a recently proposed weighted conditional maximum likelihood estimator in tackling the problem of underreporting of crashes when using a nested logit model for crash severity analyses.     

Analyzing pedestrian crash injury severity at signalized and non-signalized locations

This study identifies and compares the significant factors affecting pedestrian crash injury severity at signalized and unsignalized intersections. The factors explored include geometric predictors (e.g., presence and type of crosswalk and presence of pedestrian refuge area), traffic predictors (e.g., annual average daily traffic (AADT), speed limit, and percentage of trucks), road user variables (e.g., pedestrian age and pedestrian maneuver before crash), environmental predictors (e.g., weather and lighting conditions), and vehicle-related predictors (e.g., vehicle type). The analysis was conducted using the mixed logit model, which allows the parameter estimates to randomly vary across the observations. The study used three years of pedestrian crash data from Florida. Police reports were reviewed in detail to have a better understanding of how each pedestrian crash occurred. Additionally, information that is unavailable in the crash records, such as at-fault road user and pedestrian maneuver, was collected. At signalized intersections, higher AADT, speed limit, and percentage of trucks; very old pedestrians; at-fault pedestrians; rainy weather; and dark lighting condition were associated with higher pedestrian severity risk. For example, a one-percent higher truck percentage increases the probability of severe injuries by 1.37%. A one-mile-per-hour higher speed limit increases the probability of severe injuries by 1.22%. At unsignalized intersections, pedestrian walking along roadway, middle and very old pedestrians, at-fault pedestrians, vans, dark lighting condition, and higher speed limit were associated with higher pedestrian severity risk. On the other hand, standard crosswalks were associated with 1.36% reduction in pedestrian severe injuries. Several countermeasures to reduce pedestrian injury severity are recommended.     

A mixed logit analysis of two-vehicle crash severities involving a motorcycle

Using motorcycle crash data for Iowa from 2001 to 2008, this paper estimates a mixed logit model to investigate the factors that affect crash severity outcomes in a collision between a motorcycle and another vehicle. These include crash-specific factors (such as manner of collision, motorcycle rider and non-motorcycle driver and vehicle actions), roadway and environmental conditions, location and time, motorcycle rider and non-motorcycle driver and vehicle attributes. The methodological approach allows the parameters to vary across observations as opposed to a single parameter representing all observations. Our results showed non-uniform effects of rear-end collisions on minor injury crashes, as well as of the roadway speed limit greater or equal to 55 mph, the type of area (urban), the riding season (summer) and motorcyclist's gender on low severity crashes. We also found significant effects of the roadway surface condition, clear vision (not obscured by moving vehicles, trees, buildings, or other), light conditions, speed limit, and helmet use on severe injury outcomes.     

Evaluating alternate discrete choice frameworks for modeling ordinal discrete variables

There is considerable debate on the appropriate discrete choice framework for examining injury severity. Researchers in the safety field have employed ordered and unordered frameworks for examining the various factors influencing injury severity. The objective of the current study is to investigate the performance of the ordered and unordered response frameworks at a fundamental level. Towards this end, we undertake a comparison of the alternative frameworks by estimating ordered and unordered response models using data generated through ordered, unordered data and a combination of ordered and unordered data generation processes. We also examine the influence of aggregate sample shares on the appropriateness of the modeling framework. Rather than be limited by the aggregate sample shares in an empirical dataset, simulation allows us to explore the influence of a broad spectrum of sample shares on the performance of ordered and unordered frameworks. We also extend the data generation process based analysis to under reported data and compare the performance of the ordered and unordered response frameworks. Finally, based on these simulation exercises, we provide a discussion of the merits of the different approaches. The results clearly highlight the emergence of the generalized ordered logit model as a true equivalent ordered response model to the multinomial logit model for ordinal discrete variables.     

Using support vector machine models for crash injury severity analysis

The study presented in this paper investigated the possibility of using support vector machine (SVM) models for crash injury severity analysis. Based on crash data collected at 326 freeway diverge areas, a SVM model was developed for predicting the injury severity associated with individual crashes. An ordered probit (OP) model was also developed using the same dataset. The research team compared the performance of the SVM model and the OP model. It was found that the SVM model produced better prediction performance for crash injury severity than did the OP model. The percent of correct prediction for the SVM model was found to be 48.8%, which was higher than that produced by the OP model (44.0%). Even though the SVM model may suffer from the multi-class classification problem, it still provides better prediction results for small proportion injury severities than the OP model does.     

A multinomial logit analysis of risk factors influencing road traffic injury severities in the Erzurum and Kars Provinces of Turkey

A retrospective cross-sectional study is conducted analysing 11,771 traffic accidents reported by the police between January 2008 and December 2013 which are classified into three injury severity categories: fatal, injury, and no injury. Based on this classification, a multinomial logit analysis is performed to determine the risk factors affecting the severity of traffic injuries. The estimation results reveal that the following factors increase the probability of fatal injuries: drivers over the age of 65; primary-educated drivers; single-vehicle accidents; accidents occurring on state routes, highways or provincial roads; and the presence of pedestrian crosswalks. The results also indicate that accidents involving cars or private vehicles or those occurring during the evening peak, under clear weather conditions, on local city streets or in the presence of traffic lights decrease the probability of fatal injuries. This study comprises the most comprehensive database ever created for a Turkish sample. This study is also the first attempt to use an unordered response model to determine risk factors influencing the severity of traffic injuries in Turkey.     

Analysis of injury severity of drivers involved in single- and two-vehicle crashes on highways in Ontario

This study analyzes driver's injury severity in single- and two-vehicle crashes and compares the effects of explanatory variables among various types of crashes. The study identified factors affecting injury severity and their effects on severity levels using 5-year crash records for provincial highways in Ontario, Canada. Considering heteroscedasticity in the effects of explanatory variables on injury severity, the heteroscedastic ordered logit (HOL) models were developed for single- and two-vehicle crashes separately. The results of the models show that there exists heteroscedasticity for young drivers (≤30), safety equipment and ejection in the single-vehicle crash model, and female drivers, safety equipment and head-on collision in the two-vehicle crash models. The results also show that young car drivers have opposite effects between single-car and car–car crashes, and sideswipe crashes have opposite effects between car–car and truck–truck crashes. The study demonstrates that separate HOL models for single-vehicle and different types of two-vehicle crashes can identify differential effects of factors on driver's injury severity.     

A mixed generalized ordered response model for examining pedestrian and bicyclist injury severity level in traffic crashes

This paper proposes an econometric structure for injury severity analysis at the level of individual accidents that recognizes the ordinal nature of the categories in which injury severity are recorded, while also allowing flexibility in capturing the effects of explanatory variables on each ordinal category and allowing heterogeneity in the effects of contributing factors due to the moderating influence of unobserved factors. The model developed here, referred to as the mixed generalized ordered response logit (MGORL) model, generalizes the standard ordered response models used in the extant literature for injury severity analysis. To our knowledge, this is the first such formulation to be proposed and applied in the econometric literature in general, and in the safety analysis literature in particular.

The MGORL model is applied to examine non-motorist injury severity in accidents in the USA, using the 2004 General Estimates System (GES) database. The empirical findings emphasize the inconsistent results obtained from the standard ordered response model. An important policy result from our analysis is that the general pattern and relative magnitude of elasticity effects of injury severity determinants are similar for pedestrians and bicyclists. The analysis also suggests that the most important variables influencing non-motorist injury severity are the age of the individual (the elderly are more injury-prone), the speed limit on the roadway (higher speed limits lead to higher injury severity levels), location of crashes (those at signalized intersections are less severe than those elsewhere), and time-of-day (darker periods lead to higher injury severity).