Injury Severity Models Research Articles

Econometric approaches to examine the onset and duration of temporal variations in pedestrian and bicyclist injury severity analysis

There is considerable evidence in existing safety literature that the exogenous variable effects are likely to be time-varying in the injury severity analysis. The majority of these earlier studies tested time-varying effects of exogenous variables by crash year. However, there might be variability in the variable effects within a year, while the same effect might carry over in some or all parts of the preceding years. Towards that end, in this study, we propose a flexible framework to identify when the time-varying effect is likely to occur (the onset of temporal variation) and how long such time-varying effect lasts (duration of temporal variation) in the model estimates. In the study design, we assume that the onset of temporal variation can be any quarter of a year under consideration, while the time-varying effect can continue over different quarters after the onset of temporal variation in a variable effect. The injury severity model is estimated by using Correlated Random Parameter Generalized Ordered Logit formulation with piecewise linear functions. The empirical analysis is demonstrated by employing active traveler (pedestrian and bicyclist) crash data from Queensland, Australia for the years 2015 through 2020. The estimation results are further augmented by computing elasticity effects. The results indicate that the time-varying effects are likely to be different across years for several variables, while for other variables, the onset of time-varying effects could be different than the start of a year. Such flexibility in model specification is likely to have significant implications for devising and implementing effective countermeasures since it allows us to understand how road traffic injuries are evolving over time and when a new road safety issue might be arising.

The role of posted speed limit on pedestrian and bicycle injury severities: An investigation into systematic and unobserved heterogeneities

The posted speed limit, as a proxy of actual speed, is one of the most fundamental predictors of active travelers’ (pedestrian and bicyclist) injury outcomes when involved in crashes with motor vehicles. Although earlier studies predominantly considered posted speed limit as an exogenous variable and provided highly insightful findings, majorities of them assume the effects of active traveler behavior to remain the same across different posted speed limits, which in turn neglect the heterogeneity in active traveler behaviors on high-speed roads vs. low-speed roads. This study proposes to develop a latent segmentation-based active traveler injury severity model to relax the homogeneity assumption of the posted speed limit by active traveler behavior. Specifically, this study proposes to estimate a latent segmentation-based correlated random parameters generalized ordered logit model to examine active travel injury severity mechanisms. The proposed model accommodates systematic heterogeneity in the effects of posted speed limit, crash year and active traveler group by using a piecewise linear function in injury severity component of the latent segment model. The model is demonstrated by using active traveler crash data from Queensland, Australia, for the years 2015 through 2019. To demonstrate the implications of the estimated models, a number of hypothetical scenario analyses are performed with a specific focus on active traveler behavior and reduction in posted speed limits. The outcomes from the hypothetical scenario analysis highlighted that a 76 % (73 %) reduction in active traveler fatalities can be achieved by converting 50–60 km/hr roadways to 10–40 km/hr roadways in the urban areas (rural areas) of Queensland. The outcomes of the study will allow us to identify effective speed management strategies while targeting those with high-risk behavior.

Assessing the factors associated with pedestrian injury severity in hit-and-run crashes in Ghana

ABSTRACT Hit-and-run crashes often have severe consequences for vulnerable road users. In light of governmental efforts to promote pedestrian-friendly urban environments, the significance of these crashes cannot be overstated. In this study, we assessed the factors that influence the injury severity of pedestrians involved in hit-and-run crashes in Ghana. Historical crash data (1469 observations) spanning from 2013 to 2020 was used in this study. An injury-severity model was developed using random parameters logit approach to assess what crash factors significantly affect the injury outcome of the crashes. It was observed that hit-and-run crashes that occur on dark and unlit roadways were more likely to result in fatal injuries. Also, female pedestrians were less likely to be killed in hit-and-run crashes. These findings provide the basis for developing and implementing appropriate countermeasures, such as punitive laws for drivers who leave the crash scene and protective laws for those who help their victims.

Pedestrian injury severities resulting from vehicle/pedestrian intersection crashes: An assessment of COVID-contributing temporal shifts

Pedestrian mobility has become an increasingly important concern in transportation system analysis because of its positive impacts on the environment and healthy lifestyles. However, pedestrian safety in a vehicle-dominated transportation network remains a concern and potential barrier to pedestrian mobility, with pedestrian intersection safety being of particular concern. In addition, it is important to understand how pedestrian safety has been affected by the COVID-19 pandemic, perhaps permanently shifting pedestrian injury risks. This research seeks to provide insight into how pedestrian injury risks at intersections have changed as a result of the pandemic by estimating a series pedestrian injury severity models. To do so, unconstrained and partially constrained random parameters multinomial logit models with heterogeneity in the means of random parameters were estimated. Using Florida data, two one-year periods (one year before and one year after the COVID-19 pandemic) were defined based on vehicle miles traveled. The sample includes 3,780 single pedestrian-single vehicle crashes (2,348 from daytime and 1,432 from nighttime). A broad range of variables was considered to assess how the parameters may have shifted between the before and after periods. A series of likelihood ratio tests were conducted to examine the stability of model parameter estimates across the predefined time periods as well as to determine the differences between the daytime and nighttime crash injury severity outcomes. The results show that the nighttime crashes experienced more temporal shifts relative to daytime crashes. The findings also showed that both pedestrian and driver behavior played key temporally-shifting roles before and after the COVID-19 pandemic period. Finally, the out-of-sample simulations suggest that pedestrian injuries have become more severe after the pandemic.

Injury-severity analysis of crashes involving defective vehicles and accounting for the underlying socioeconomic mediators

Crashes occur from a combination of factors related to the driver, roadway, and vehicle factors. The impact of vehicles on road crashes is a critical consideration within road safety analysis, even though not much studies have been conducted in this area. This study assessed how various vehicle and other crash factors are significantly associated with crash outcomes. To do this, historical vehicle defect-related crashes were obtained for the state of Alabama from 2016 to 2020. After data cleaning, a crash injury severity model was developed using the random parameters multinomial logit with heterogeneity in means approach to account for possible unobserved heterogeneity in the data. A spatial analysis was further conducted to better understand vehicle defect crashes as a broader societal issue and potentially explore their connection with the socio-demographic characteristics of the drivers of these vehicles. The preliminary data analysis showed that brake and tire defects accounted for about 65% of the vehicle defects associated with the crashes. The model estimation results revealed that improper tread depth and headlight defects were associated with major injury outcomes, while brake defects were more associated with minor injuries. Also, crashes associated with speeding, drunk driving, failure to use seatbelts, and those that occurred on curved roads left with downgrades were likely to result in major injuries. Findings from the spatial analysis showed that postal codes with higher median incomes are more likely to record lower vehicle defect-related crashes, unlike those that have higher proportions of females and African Americans. The study's findings provide data-driven evidence for sustained safety campaigns, workshops, and training on basic vehicle maintenance practices in the low-income communities in the state.

Analysis of factors that influence injury severity of single and multivehicle crashes involving at-fault older drivers: A random parameters logit with heterogeneity in means and variances approach

The number of older drivers on our roads have been increasing given the upward shift of older people in the population. It is well-known that older drivers are more susceptible to injury given a crash. They are also likely to be at-fault in certain types of crashes due to health conditions associated with old age. This paper identifies factors that influence the severity of crashes involving at-fault older drivers at a disaggregated level for single and multivehicle crashes. Two random parameters multinomial logit with heterogeneity in means and variances models of injury severity were developed to assess the impact of various crash factors on crash outcomes based on the crash type. The study used five years of crash data from Alabama. The model estimations show that rainy weather and dark roadways were associated with an increased likelihood of serious injuries for single-vehicle crashes. Older male drivers were also found to have a higher probability of being at-fault in fatal single-vehicle crashes. For multivehicle older driver crashes, the model estimation results revealed that fatal injury was more likely to occur when the airbags were deployed. In addition, drivers between 65 and 75 years are less likely to be involved in fatal injury crashes for both single and multivehicle crashes. It was further found that older driver at-fault crashes that occurred in the open country were more likely to be fatal regardless of the manner of collision. The findings of this study are expected to help policymakers and transportation stakeholders to implement countermeasures that will make the road safer for older drivers and all other road users.

A hybrid approach of random forest and random parameters logit model of injury severity modeling of vulnerable road users involved crashes

Vulnerable road users (VRUs) involved crashes are a major road safety concern due to the high likelihood of fatal and severe injury. The use of data-driven methods and heterogeneity models separately have limitations in crash data analysis. This study develops a hybrid approach of Random Forest based SHAP algorithm (RF-SHAP) and random parameters logit modeling framework to explore significant factors and identify the underlying interaction effects on injury severity of VRUs-involved crashes in Shenyang (China) from 2015 to 2017. The results show that the hybrid approach can uncover more underlying causality, which not only quantifies the impact of individual factors on injury severity, but also finds the interaction effects between the factors with random parameters and fixed parameters. Seven factors are found to have significant effect on crash injury severity. Two factors, including primary roads and rural areas produce random parameters. The interaction effects reveal interesting combination features. For example, even though rural areas and primary roads increase the likelihood of fatal crash occurrence individually, the interaction effect of the two factors decreases the likelihood of being fatal. The findings form the foundation for developing safety countermeasures targeted at specific crash groups for reducing fatalities in future crashes.

Examination of factors associated with fault status and injury severity in intersection-related rear-end crashes: Application of binary and bivariate ordered probit models

Rear-end crashes are a relatively common crash type and often occur at or near intersections. Although rear-end crashes are generally less severe than some other crash types, there were still 2,346 fatal and 595,000 injury rear-end crashes in the US in 2019 alone. These crashes are generally caused by one at-fault driver who strikes a leading vehicle, and it may be useful to assess factors associated with a driver being at-fault. Additionally, it is important to analyze factors associated with injury severity outcomes in order to develop countermeasures aimed at preventing severe injuries. As such, this study investigates factors related to fault status and injury severity (and the interrelation between the two) in two-vehicle intersection-related rear-end crashes using data from a southwestern US state. A binary probit model was estimated to assess factors associated with fault status, while a bivariate ordered probit model was estimated to assess factors associated with driver injury severity by fault status. Importantly, by modelling the injury severity of both crash-involved drivers jointly, potential within-crash correlation can be accounted for. The results of the fault status model indicated numerous factors were associated with a fault status including vehicle type, driver age, and driver impairment or distraction. The results of the bivariate injury severity model indicated numerous factors were significantly associated with injury severity and importantly, differences were observed between at-fault and not-at-fault drivers. Ultimately, the results of this study may assist in development of targeted countermeasures aimed at reducing both crash occurrence and severe injury outcomes.

A hierarchical multinomial logit model to examine the effects of signal strategies on right-turn crash injury severity at signalised intersections

The severity of right-turn crashes (or left-turn crashes for the roads in the US) at signalised intersections tends to be high because of the relatively high conflicting speeds and angle of impact. However, right-turn crash injury severity at signalised intersections was not sufficiently studied. In particular, the effects of signal control strategies on crash injury severity are not known. This study developed crash injury severity models for right-turn crashes at signalised intersections with a novel approach of linking crashes with signal strategies which enabled assessing the effects of signal strategies on crash injury severity. The study provided a comprehensive understanding of the impacts of signal strategies, intersection geometry and traffic factors on crash injury severity of right-turn crashes at signalised intersections. Crash injury severity models were estimated with crash data from 221 signalised intersections in Queensland from 2012 to 2018. To address the hierarchical structure of crash data, two-level hierarchical Multinomial Logit models were applied, hypothesising that the first level includes individual crash characteristics while the second level includes intersection characteristics. The applied hierarchical model accounts for the correlation among crashes within intersections. Results showed that crashes during Lagging right-turn and Diamond overlap turns are likely to be more severe than other signal strategies at intersections, with the Lagging right-turn signal being the most hazardous. The results also illustrate that the probability of severe injuries increases with the number of conflicting lanes, whereas the corresponding probability decreases with the occupancy of the conflicting lane.

Older driver at-fault crashes at unsignalized intersections in Alabama: Injury severity analysis with supporting evidence from a deep learning based approach

Introduction: The research described in this paper explored the factors contributing to the injury severity resulting from the male and female older driver (65 years and older) at-fault crashes at unsignalized intersections in Alabama. Method: Random parameter logit models of injury severity were estimated. The estimated models identified a variety of statistically significant factors influencing the injury severities resulting from older driver at-fault crashes. Results: According to these models, some variables were found to be significant only in one model (male or female) but not in the other one. For example, variables such as driver under the influence of alcohol/drugs, horizontal curve, and stop sign were found significant only in the male model. On the other hand, variables such as intersection approaches on tangents with flat grade, and driver older than 75 years were found significant only in the female model. In addition, variables such as making turning maneuver, freeway-ramp junction, high speed approach, and so forth were found significant in both models. Estimation findings showed that two parameters in the male model and another two parameters in the female model could be modeled as random parameters, indicating their varying influences on the injury severity due to unobserved effects. In addition to the random parameter logit approach, a deep learning approach based on Artificial Neural Networks was introduced to predict the outcome of the crashes based on 164 variables that are listed in the crash database. The artificial intelligence (AI)-based method achieved an accuracy of 76% indicating the role of the variables in deciding the final outcome. Practical Applications: Future plans are set to study the use of AI on large sized datasets to achieve a relatively high-performance, and hence to be able to identify which variables contribute the most to the final outcome.

Learning about injury severity from no-injury crashes: A random parameters with heterogeneity in means and variances approach

The traditional approach to injury-severity analyses does not allow in-depth understanding of no-injury crashes, as crash factors found to contribute to the various injury severities may have similar effects on the severity of vehicle damage even if no injury is recorded. Viewing no-injury crashes using the vehicle damage severities as sub-categories and bases for potential injuries can improve understanding of future injury crashes. To better understand the mechanism of no-injury crashes and the crash factors that contribute to the extent of vehicle damage beyond the single categorization of these crashes in injury severity analysis, this study presents a vehicle damage severity analysis for no-injury crashes. To compare the effects of crash contributing factors on crash outcomes, two injury severity models were also estimated. Random parameters multinomial logit models with heterogeneity in means and variances were developed to account for unobserved heterogeneity. Model estimation results revealed that several common factors (e.g., unsafe speed, distracted driving, driving under influence, vehicle age, and run-off-road) are correlated with both injury severity in injury crashes and vehicle damage severity in no-injury crashes. Therefore, the sub-categorization of no-injury crashes by vehicle damage severity can potentially improve estimates of injury severity considered in resource allocation decisions for traffic safety.

Evaluating gender differences in injury severities of non-helmet wearing motorcyclists: Accommodating temporal shifts and unobserved heterogeneity

With rapid growth in motorcycle use and relatively low helmet-wearing usage rates, injuries and fatalities resulting from motorcycle crashes in Pakistan are a critical concern. To investigate possible temporal instability and differences in the factors that determine resulting injury severities between male and female non-helmet wearing motorcyclists, this study estimated male and female injury severity models using a random parameter logit approach with heterogeneity in means and variances. Motorcycle crash data between 2017 and 2019 from Rawalpindi, Pakistan, were used for the model estimation. With four possible crash injury severity outcomes (injury, minor injury, severe injury, and fatal injury), a wide variety of explanatory variables were considered, including the characteristics of riders, vehicles, roadways, environments, crashes, and temporal considerations. Temporal shifts in the effects of explanatory variables were confirmed using a series of likelihood ratio tests. While the effects of several explanatory variables are relatively temporally stable, those of most variables vary considerably across the years. In addition, out-of-sample simulations underscore the temporal shifts from year to year and the differences between male and female motorcyclist-injury severity. The findings suggest that factors such as effective enforcement countermeasures and relevant educational campaigns can be implemented to reduce injury severity. The statistically significant differences between male and female non-helmeted injury severity models underscore the importance of policies that separately target male and female motorcycle rider safety.

Reliability analysis of motorcycle crash severity outcomes: Consideration of model selection uncertainty

Objective While a large amount of work has been conducted on different types of crash injury severity models, model selection uncertainty remains a critical issue in traffic safety research. The objective of this study is to handle model selection uncertainty by combining multiple models. Methods Motorcycle crashes in Michigan from 2010 to 2014 are collected for the analysis. A model averaging approach is used to integrate useful information from three commonly used crash injury severity models: multinomial logit model, ordered logit model, and ordered probit model to deal with the situation where the model selection uncertainty exists in crash data analysis. The ratios of model posterior probabilities between models are used to quantify the model selection uncertainty. In addition, the effectiveness of the method is illustrated by comparing it with the single-best model. Results The ratios of model posterior probabilities among models approximate to 1. It means that three models have the same importance in statistical analysis of motorcycle injury severity, resulting in model selection uncertainty. The comparison between the results of model averaging approach and single-best model shows that the single-best model tends to overestimate the effects of risk factors on motorcycle injury severities because of ignoring the model selection uncertainty; parameter errors and confidence intervals of model averaging are greater and wider than those of the single-best model due to between-model uncertainty included in the model averaging; some risk factors are significant in the model averaging approach while not in the single-best model. Results from model averaging approach reveal that drunk or riding under influence, angle/sideswipe/head on crashes, speed limit of 35 mph or higher, and signal control play significant roles in the motorcycle crashes. Conclusions The study contributes to the existing crash injury-severity literature by developing a model averaging approach to explore the relationship between motorcyclist’s injury-severity and its contributing factors. The model averaging approach overcomes the limitations of the current crash injury-severity modeling approaches by (1) revealing the potential model selection uncertainty among injury-severity models with model posterior probabilities; (2) more reliably accounting for the effects of risk factors on motorcyclist’ injury severities through integrating all information from the candidate models; and (3) better presenting the underlying unreliability of the analysis results from each individual model.

COVID-19 and Injury Severity of Drivers Involved in Run-Off-Road Crashes: Analyzing the Impact of Contributing Factors

This study examined the relationship between the lockdown during the COVID-19 pandemic and the severity of injuries sustained by drivers involved in run-off-road (ROR) crashes. A random parameter ordered logit (RPOL) modeling framework was utilized to account for the ordinal nature of severity outcome and capture the potential unobserved heterogeneity. The data used in this study contained ROR crashes that occurred in the state of Florida from April to September for 2019 and 2020 representing non-pandemic and pandemic time periods, respectively. Separate driver injury severity models were developed across the two time periods, and the overall stability of the model estimates was examined through likelihood ratio tests. The impacts of various potential contributing factors, including crash-, driver-, and vehicle-related variables, roadway geometric characteristics, environmental conditions, and traffic-specific factors, were assessed. Although the developed models share some common features, the analysis results showed that the model specifications indicated a strong temporal instability among the estimated parameters. Compared to the non-pandemic period, the following variables resulted in increased driver injury severity in ROR crashes during the pandemic: drivers 65 years or older, careless driving, and absence of traffic control devices.

Understanding the Factors Associated with the Temporal Variability in Crash Severity before, during, and after the COVID-19 Shelter-in-Place Order

The COVID-19 travel restriction orders have significantly reduced travel and generally lowered the risk of road traffic collisions, but many accounts suggest an increase in risky driving behaviors and consequent fatal crashes during the shelter-in-place period. Risky driving behaviors including failure to wear a seatbelt, speeding, and drunk driving were observed to be the leading contributing factors of the fatalities. Whereas the fatal crashes that characterized the shelter-in-place period has become a topical issue, the high number of crashes that occurred as a result of the panic shopping and increased travel activities in the weeks before the shelter-in-place order have not received much attention. In this study, we investigated the differences and similarities in the effects of the factors that were associated with crash injury severity before, during, and after the shelter-in-place order. The study used crash data from the state of Alabama for the 2020 calendar year. Preliminary data analysis revealed interesting variations in crash trends across the three periods. It was found that the highest weekly crash frequency occurred in the immediate week before the shelter-in-place order, and a higher proportion of crashes that occurred between 6 p.m. and 6 a.m. and those that occurred in residential areas happened during the shelter-in-place period while shopping area crashes, manufacturing/industrial area crashes, rear-end collisions, and crashes involving female drivers occurred mostly before the shelter-in-place period. Three injury severity models were developed using random parameters logit with heterogeneity in means and variances approach. The results showed that major injury crashes occurred mainly in rural areas and occurred due to speeding, fatigue driving, and failure to use a seatbelt. The effects of these factors on crash outcome did not vary across the year, indicating that the shelter-in-place order did not impact the driving behaviors of the driver population that got into major injury crashes. The results further revealed that the effects of some crash factors, such as road type and manner of collision, varied across the periods. The findings of the study provide a deeper, data-driven understanding of how driving behaviors and associated crash outcomes may be affected by extreme events such as the COVID-19 shelter-in-place.

Modelling injury severity of victims in collisions involving public transit in Dhaka, Bangladesh

Collisions involving public transit (PT) have increased significantly in the recent times. In developing countries, the scenario is much worse as the PT collision occurrence and injury severity are higher by manifold. This study investigates the injury severity of victims of PT involved crashes in Dhaka, Bangladesh. Victims include PT occupants and counterpart road users such as pedestrians who are more vulnerable. A latent segmentation-based ordered logit (LSOL) model is developed using police-reported crash records of Dhaka. The LSOL model addresses the ordinality of injury severity and captures unobserved heterogeneity by allocating victims into discrete latent segments. The LSOL model is estimated for two segments: segment two is high-risk segment which includes unconventional vehicle occupants and pedestrians, whereas segment one is low-risk segment. The injury severity model results suggest that person profile, environmental factors and built environment characteristics significantly influence injury severity. For example, female and elderly victims have higher likelihood to sustain severe injury. The model confirms the existence of significant heterogeneity. For instance, mid-block collisions are more likely to yield severe injury in the high-risk segment, and less likely to result in severe injury in the low-risk segment. Similar heterogeneity is confirmed for variables representing impaired driver, three-way intersection, and police-controlled intersection. The elasticity effect analysis suggests that female victims, younger and adult victims, unconventional vehicle occupants, pedestrians, and mid-block crashes have substantial positive impact on fatality. The identified influential factors and heterogeneity captured in the study need to be accommodated in road safety plans and policies.

A literature review of machine learning algorithms for crash injury severity prediction

Introduction: Road traffic crashes represent a major public health concern, so it is of significant importance to understand the factors associated with the increase of injury severity of its interveners when involved in a road crash. Determining such factors is essential to help decision making in road safety management, improving road safety, and reducing the severity of future crashes. Method: This paper presents a recent literature review of the methods that have been applied to road crash injury severity modeling. It includes 56 studies from 2001 to 2021 that consider more than 20 different statistical or machine learning techniques. Results: Random Forest was the algorithm with the best results, achieving the best performance in 70% of the times that it was applied and in 29% of all studies. Support Vector Machine and Decision Tree achieved the best performance in 53% and 31% of the times and in 16% and 14% of all studies, respectively. Bayesian Networks and K-Nearest Neighbors achieved the best performance in 67% and 40% of the times that were used but only achieved the best performance in 4% and 7% of all the studies analyzed, respectively. Conclusions: At this point, Random Forest revealed to be a good approach for road traffic crash injury severity prediction followed by Support Vector Machine, Decision Tree, and K-Nearest Neighbor. However, there is still a lot of room in this area to explore other techniques that can best suit this purpose as not only the model's performance should be considered but also causality issues, unobserved heterogeneity, and temporal instability. Practical Applications: This review enables researchers to understand the recent techniques applied in the analysis of injury severity modeling, and the ones that achieved the best performance results. Based on the reviewed studies, challenges and future research directions are presented.

Exploring the impacts of built environment on pedestrian injury severity involving distracted driving

Introduction: This study develops an injury severity model that demonstrates level of pedestrians’ injury severity during pedestrian-vehicle collisions, specifically those involving distracted driving. Method: It uses data from a police-reported collision database that contains pedestrian-vehicle collision information in Nova Scotia, Canada. A latent segmentation-based ordered logit (LSOL) model is developed in this paper that comprehensively examines the influence of built environment characteristics on pedestrian injury severity. It estimates a latent segment allocation model within LSOL modeling framework to capture unobserved heterogeneity across pedestrians. Two latent segments, high- and low-risk segments, are identified probabilistically based on pedestrian characteristics and action, driver action, and collision attributes. Results: Results suggest that higher mixed land-use, transit stop density, length of sidewalk in the collision locations, and collisions occurring near schools yield lower pedestrian injury severity. In contrast, pedestrian-vehicle collisions in arterial roads, curved roads, sloped roads, and roundabouts tend to result in severe injuries. Interactions between distracted driving type and built environment characteristics are examined in this study. For example, using a communication device while driving on straight roads increases likelihood of higher pedestrian injury severity. This study also confirms the existence of heterogeneity across latent segments. For instance, higher percentage of people commuting by walking in the collision areas yield severe pedestrian injury in high-risk segments and lower injury severity in low-risk segments. Practical applications: The findings of this study will assist transportation planners and road safety stakeholders in developing effective and prioritized policies to reduce pedestrian injury severity involving distracted driving incidents.

An omni-directional model of injury risk in planar crashes with application for autonomous vehicles

Objective: Automated driving systems (ADS) are actively being deployed within the driving fleet. ADS are designed to safely navigate roadways, which entails an expectation of encountering varying degrees of potential conflict with other road users. The ADS design and evaluation process benefits from estimating injury severity probabilities for collisions that may occur. Current regression models in the literature are typically bespoke analyses involving targeted principal directions of force (PDOFs) and occupant positions. It is preferable to rely on injury severity models derived from a single source to provide a continuous function of risk for all planar collisions, while also accounting for specific vehicle and occupant characteristics. The novel feature of the proposed models is continuous, parametric injury risk surfaces that encompass the full spectrum of available United States field data. Methods: We used years 2001–2015 of the National Automotive Sampling System, Crashworthiness Data System (NASS-CDS) and years 2017–2019 of the Crash Investigation Sampling System (CISS) to estimate injury risk at the maximum abbreviated injury scale (MAIS) 3 and higher (3+) and 5 and higher (5+) levels for all adult occupants traveling in 2002 or newer passenger vehicles which were less than 10 years old at the time of the crash. The models account for occupant, vehicle, and crash characteristics. Interactions with vulnerable road users (e.g., pedestrian, bicyclist) were not considered. Results: We present statistical models suitable to predict injury in all non-rollover crashes at the maximum MAIS3+ and 5+ levels, and show that these models can be comparable to similar single scenario (e.g., frontal) crash models. We discuss challenges with imputing missing field data, and discuss handling of covariates that may not be known at the time of the crash. Conclusions: Collision severity assessment is a vital component of the ADS design process. We developed a novel injury risk function that can assess occupant injury risks across the spectrum of foreseeable planar collisions. These models can provide insight on potential outcomes of counterfactual simulations, injury risk and crashworthiness considerations for human driven vehicles, and provide an evaluation tool that can be applied in ADS safety impact evaluation.

Correlated mixed logit modeling with heterogeneity in means for crash severity and surrogate measure with temporal instability

This study employs the correlated mixed logit models with heterogeneity in means by accounting for temporal instability to estimate both injury severity and vehicle damage. Two years of intersection crash data from Connecticut were analyzed based on driver characteristics, highway and traffic attributes, environmental variables, vehicle and crash types. These elements were used as independent variables to explore the contributing factors to crash outcome. The likelihood ratio test highlights that the temporal instability exists in both injury severity and vehicle damage models. The model estimation results illustrate that the means of some random parameters are different among crashes. The correlation coefficients of random parameters verify that these random parameters are not always independent, and their correlations should be considered and accounted for in crash severity estimation models. The investigation and comparison between injury severity models and vehicle damage models verify that the injury severity and vehicle damage are highly correlated, and the effects of contributing factors on vehicle damage are consistent with the results of injury severity models. This finding demonstrates that vehicle damage can be used as a potential surrogate measure to injury severity when suffering from a low sample of severe injury crashes in crash severity prediction models. It is anticipated that this study can shed light on selecting appropriate statistical models in crash severity estimation, identifying intersection crash contributing factors, and help develop effective countermeasures to improve intersection safety.