Measuring Road Safety with Surrogate Events

By Andrew Tarko

Measuring Road Safety Using Surrogate Events provides researchers and practitioners with the tools they need to quickly and effectively measure traffic safety. As traditional crash-based safety analyses are being undermined by today’s growing use of intelligent vehicular and road safety technologies, crash surrogates--or near misses--can be more effectively used to measure the future risk of crashes. This book advances the idea of using these near-crash techniques to deliver quicker and more adequate measurements of safety. It explores the relationships between traffic conflicts and crashes using an extrapolation of observed events rather than post-crash data, which is significantly slower to obtain.

Readers will find sound estimation methods based on rigorous scientific principles, offering compelling new tools to better equip researchers to understand road safety and its factors.

• Consolidates the latest updates/ideas from disparate places into a single resource
• Establishes a consistent use of key terms, definitions and concepts to help codify this emerging field
• Contains numerous application-oriented case studies throughout
• Includes learning aids, such as chapter objectives, a glossary, and links to data used in examples

• Language: English
• Publisher: Elsevier Science
• Release date: Nov 7, 2019
• ISBN: 9780128105054

Andrew Tarko

Andrew P. Tarko is Director of the Purdue Center for Road Safety at Purdue University. He is Chair of the Transportation Research Board Subcommittee on Surrogate Measures of Safety, an Advisory Board Member of Accident Analysis and Prevention, and has written numerous transportation research reports, conference papers, journal articles, and book chapters.

Book preview

Measuring Road Safety with Surrogate Events

Andrew P. Tarko

Lyles School of Civil Engineering, Purdue University, IN, United States

Table of Contents

Cover image

Title page

Copyright

Chapter 1. Introduction

Chapter 2. Crashes in safety analysis

Human factor in crash occurrence

Safety factors in Haddon's periods

Crash probability and frequency

Crash severity

Crash-based safety analysis

Other challenges

Chapter 3. Traffic conflicts as crash surrogates

Historical perspective

Etiological consistency

Separation between road users (crash nearness)

Detecting conflicts by evasive action

Other characterizations of traffic conflict

Chapter 4. Techniques and technologies of observing traffic conflicts

Historical perspective

Video-based roadside technique

LiDAR-based roadside technique

Naturalistic driving study

Summary

Chapter 5. Studies on the conflict-crash relationship

Conflict and crash counts

Outcomes of safety-relevant events

Separation between road users (crash nearness)

Summary

Chapter 6. Probabilistic connection of traffic conflicts with crashes

Proposed probabilistic connection

Conflicts grouped by severity and inclusion of non-conflicts

Crash probability in heterogeneous conditions

Reducing the threshold separation under heterogeneous conditions

Example variability of crash probability

Practical implications for analysis and estimation results

Generalization to any reasonable measure of separation

Chapter 7. Estimating crash frequency from traffic conflicts

Existing estimation methods

Proposed estimation method

Computer simulation

Evaluating the estimation performance

Summary

Chapter 8. Challenges and treatments in estimating crash frequency

Claiming traffic conflicts

Selecting separation thresholds

Linearity of log-log curve

Separation estimation errors

Small sample

Censoring

Summary

Chapter 9. Road departures – driving simulator study

Instrumentation and experiment design

Speed behavior and near-departures

Departure risk analysis

Summary

Chapter 10. Right-angle collisions – a lesson learned

PET data collection

Data collection method

PET data summary

PET data analysis

Evaluation of PET-based crash estimates

Chapter 11. Rear-end collisions – naturalistic driving study

Data

Initial analysis

Studied types of drivers

Claiming and analyzing traffic conflicts

Evaluating the results

Summary

Chapter 12. Traffic conflicts of autonomous vehicles

Autonomous vehicles and traffic safety

Autonomous navigation and its potential shortcomings

Application of traffic conflict analysis

Chapter 13. Summary and future research directions

Crash nearness versus traffic conflicts

Method limitations and potential solutions

Observing conflicts with in-vehicle and roadside technology

Estimating safety effects

Index

Copyright

Elsevier

Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands

The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom

50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States

Copyright © 2020 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).

Notices

Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-12-810504-7

For information on all Elsevier publications visit our website at https://www.elsevier.com/books-and-journals

Publisher: Jonathan Simpson

Acquisition Editor: Brian Romer

Editorial Project Manager: Emily Thomson

Production Project Manager: Paul Prasad Chandramohan

Cover Designer: Mark Rogers

Typeset by TNQ Technologies

Chapter 1

Introduction

Abstract

Road safety is affected by a plethora of intertwined factors. A great number of studies have been published on traffic safety by authors representing various disciplines. Yet, it still is not possible to confidently measure safety and attribute safety improvements to particular countermeasures because of the scarcity and poor quality of crash data. Although years of progress have produced advanced statistical models, serious concerns remain, including questionable causality, high level of aggregation, and, frequently, too small data samples. The crash-based estimation issue may continue to grow with the increasing presence of autonomous and other vehicles equipped with advanced technology. One possible solution, borrowed from medical sciences, is using surrogate events that are more frequent than crashes but related to them.

Keywords

Safety factors; Multidisciplinary analysis; Crashes; Surrogate event; Traffic conflict

References

People leave their homes and offices every day to travel on roads. They maneuver in traffic as drivers and passengers of buses, cars, and trucks, or they walk and ride bicycles near moving traffic. The unfortunate reality is that some of them become involved in traffic crashes, get injured and even die. Searching through the great number of journal articles and books published on traffic safety, one may be inclined to conclude that a good understanding of the subject must have been reached. Indeed, our general understanding of safety-related matters is much greater than ever before; and thanks to research initiatives, this topic is continually being studied from many angles, including its psychological, social, economic, medical, statistical, and engineering aspects. Psychologists continue to identify various types and sources of human errors on roads and propose theories that explain why road users take risks even when they are aware of their imperfections. Computer engineers and human factors specialists are collaborating to design man-machine interfaces (instrumented dashboards in less fancy terms) and adding active safety systems that warn about imminent dangers and even take control over the vehicle. Mechanical and structural engineers are redesigning vehicle bodies to make them sturdy around humans but flexible elsewhere to absorb part of the energy and reduce the impact on the humans. Biomechanical engineers are optimizing seatbelts, boosters, and various types of airbags to provide additional restraint and cushion for car drivers and passengers. Improved access to medical services and creation of trauma centers that specialize in mechanical injuries are saving lives and improving the wellbeing of traffic crash victims.

From an economic standpoint, the losses due to crashes are now estimable for countries and regions based on crash records, and their tangible components are identifiable, such as loss of wages, medical treatment and rehabilitation costs, property damage, and time lost by road users due to the crashes. In addition, the willingness of road users to pay to avoid injury and death is helping in the estimation of safety benefits to support investing taxpayer money to improve road safety. Transportation engineers are improving road design; inventing new types of intersections and interchanges; improving road and median barriers to prevent crashes; and when crashes do happen, to reduce the dangerous impact of the roadside objects. These accomplishments most certainly are borne out of an improved understanding of safety and injury factors. Unfortunately, although the crash rates have been falling over the last 20 years, they remain high: 32.7 thousand people were killed and 2.34 million were injured on US roads in 2014 (FARS, 2015; Fig. 1.1). These staggering statistics persist partly because humans become more confident and have lower risk perception once they are aware of improvements in vehicles, so there is still a lot to learn about road safety.

It may sound surprising, but one of the greatest challenges faced by those involved in safety research and management is estimating safety itself. Even with the abovementioned numerous accomplishments, it still is not possible to confidently attribute a resulting safety improvement to any particular implemented countermeasure because of the fundamental difficulty in measuring countermeasures' effects. Ezra Hauer, one of the renowned researchers in road safety admitted in 1986, When it comes to managing road safety, we're in the Dark Ages. There's a lot of arm waving but very little knowledge of what works (Hauer, 1986). We dare to say that this evaluation of our knowledge in the road safety area may be valid today. The first statistical models that estimated annual crash frequency and its factors were simple and inadequate. While the nearly 50 years of progress, especially the last 20 years, have produced econometric models that more adequately account for both the statistical properties of crashes and the data limitations, serious concerns remain. Unfortunately, often the performance of even these newer models still does not meet the needs of crash-based safety analysis, which is briefly discussed later in this book.

Our inability to estimate safety becomes obvious when we want to know the level of safety on a certain road in a short period, such as 1   h. Crashes are the ultimate demonstration of unsafety, but does that mean a road is safe during an hour that passes without a crash? Some drivers who were driving on that road and pedestrians who crossed it during that hour may disagree, as they may have been concerned about the possibility of a crash based on their general knowledge and their own experience. Although a feeling is not the most reliable way of estimating safety, it reflects the undeniable non-zero risk of a crash. Counting crashes in short periods yields zero crashes most of the time. Yet, considering zero as a good estimate of the expected number of crashes in a short interval is questionable. Why? The answer is in the potential of an outcome so severe that even a very low risk of this outcome cannot be neglected. This particular combination of an extremely infrequent occurrence with an extremely severe outcome presents a major challenge to measuring safety. To further emphasize this difficulty, let us consider a different case: estimating traffic quality, which is measured with speeds, delays, and queues that frequently take non-zero values. Averaging such observations during a 1-h period yields a reasonable estimate of the expected values. Even when certain periods are free of delays or queues, the sheer presence of zero values indicates that the expected value indeed should be close to zero. Since the value of lost seconds or even minutes by a road user cannot be compared to the cost of a severe crash, a zero estimation of an almost-zero true value is acceptable.

Fig. 1.1 Fatalities and injuries 1990–2014 (FARS, 2015).

Several decades of struggling with the challenge of estimating safety have brought forth an impressive number of statistical techniques that yield the expected number of crashes and their severity in relatively long periods, typically years. Given the nature of highly infrequent and random events, a large number of observations are needed to obtain a reasonable estimation of the crash frequency. Accumulating observations is typically accomplished by allowing long observation periods and including large regions with many roads. For many years, safety estimation relied on highly aggregated data collected routinely by transportation agencies and law enforcement. The annual frequency of crashes, sometimes measured by outcome severity, were estimated based on the road characteristics and the average traffic.

The serious downfalls of using aggregate data for safety are well known. Data aggregation can be mitigated by increasing the geographical resolution. A good example of this attempt to mitigate aggregation is the use of safety predictive models for evaluating safety based on road geometry (AASHTO, 2010). Since their focus is on geometry, which does not change often, the annual average frequency of crashes seems to serve the purpose well, and thus prompts focusing on increasing the geographic differentiation (different types of roads in different regions) while using crash counts during long observation periods. This approach requires developing separate models for various types of roads and various types of crashes under precisely defined base road conditions. The low numbers of crashes for different types of roads in the base condition comprise the serious hurdle. To make matters worse, a number of studies have indicated that these models were not transferable between regions and countries (Persaud et al., 2002; Brimley et al., 2012; Farid et al., 2016). Regions with different weather and driving cultures require custom models developed with their own crash data. Thus, in spite of the enormous effort, the development of safety predictive models is compromised considerably by the differences between regions and road types, and requires geographical disaggregation of the data, while low crash frequency prevents confident safety estimation with the statistical models.

The temporal differences in safety are commonly expected to be even stronger than the spatial differences. Roads instrumented with vehicle detectors have been providing disaggregate traffic flow data at designated locations for years. The availability of such data with the advancements in statistical modeling techniques has encouraged attempts to estimate safety in short intervals of several minutes based on the data aggregated in these intervals. Although probabilistic models can be estimated, the shortage of crashes in the development stage affects their performance in identifying high-risk traffic conditions. The confirmed spatial differences in safety makes these models difficult to transfer to other locations.

Although both instrumented vehicles in naturalistic driving studies and stationary road instrumentation, such as image detection and high-resolution sensors, allow traffic observations of individual road users in short intervals of seconds, growing data disaggregation leads to greater difficulty in measuring safety based on observed crashes. Chapter 2 discusses crash-based safety model estimation and the concerns related to the quality of the data, and more importantly, the omission of data in these analyses.

The safety estimation problem is particularly challenging on safe roads, which seems to be a peculiar statement, but it has substance. The risk of crash on a county road in a short period, say one day, is extremely small due to the low traffic volume. A major safety problem on county roads was exposed when the average risk of fatality faced by a driver on these roads was found to be much higher than on arterial roads. For example, the number of fatalities in South Dakota in the 2001–05 period was nearly five times higher on rural local roads than on rural freeways when adjusted for the number of miles traveled on those roads (Marshall, 2007). A similar comparison for the entire US indicated the disparity to be 2.5 times. Consequently, the number of fatal crashes on the vast mileage of county roads is staggering. Although the problem is known, identifying the specific conditions of the excessive risk is not easy.

The estimation issue, already a serious one, will continue to grow for all roads with the increasing presence of modern vehicles equipped with technological marvels such as self-driving automata. Driverless vehicles, interconnected vehicles, and vehicles communicating with instrumented roads are in the foreseeable future. A transition period from all human-driven to all driverless vehicles may take many years, but it will undoubtedly bring considerable changes in safety as well as new safety challenges. One of those challenges is its making most of our current knowledge about safety obsolete. Although the number of crashes will be gradually (hopefully) reduced, they will not go away even if all human drivers are replaced with machines. There still will be imperfect pedestrians and bicyclists interacting with vehicles, all subject to the laws of physics, acting in the environment and exposed to detrimental weather and imperfections in the new technology and its designers. Crashes may become even rarer than they are now, but possibly may be more severe due to higher speeds and the physical closeness of vehicles. The need for more efficient methods of estimating safety can only grow.

Medical science faces the same problem of estimating the risk of rare and potentially severe events in the form of diseases. The available statistics are often insufficient to identify the factors of a disease in ecological studies or to evaluate the effectiveness of a treatment in clinical studies. Although experimentation through clinical studies is not easily workable for road safety research, one idea that has attracted the attention of safety specialists that is being successfully implemented in medical science is the use of surrogate measures. A surrogate measure is used in medical science to predict the long-term clinical outcome of a treatment; for example, the level of lipoprotein cholesterol in patients with coronary heart disease has long been considered a suitable surrogate for the increased likelihood of disease-inflicted death. A useful surrogate measure must be associated with a clinical outcome, and it also should be sensitive to the studied treatment if this treatment indeed improves the chance of a patient's survival. A good surrogate measure allows estimating the risk of death much faster and, more importantly, does not require the actual death to occur. The cost of the study is reduced, the life-saving results are obtained more quickly, and the treatment may be approved and implemented sooner. From the road safety perspective, a certain condition that increases the risk of a crash is like a disease that needs to be identified and remedied. A surrogate measure is needed to help identify this condition as dangerous; and a treatment meant to eliminate the condition or neutralize its effect then must be quickly evaluated in a before-after study. To be useful, a surrogate measure must be sensitive to the studied condition and the evaluated treatment, and can be linked to the risk of crash.

This book attempts to systematically introduce the concept of estimating road safety with surrogate measures, it discusses traffic conflicts as a proper surrogate measure in light of the proposed paradigm, proposes an adequate method of estimating safety with traffic conflicts, and illustrates the method with examples that span several data collection approaches, including observing vehicles with non-intrusive roadside techniques, observing human drivers in equipped vehicles in naturalistic driving studies, and observing subjects in driving simulators. It is argued that the best surrogate measures must be the elements of the traffic events leading to a crash. Thus, anything learned while observing these events is valuable in furthering our understanding of the safety-related behavior of road users. Such knowledge is vital for improving the tools for evaluating road safety, for proposing adequate safety countermeasures, and for evaluating and improving emerging solutions.

Although this book focuses on observing surrogate events and using the collected data to evaluate road safety, it occasionally discusses human behavior and its associated factors in order to provide a deeper understanding of the underlying mechanism, which allows claiming that what is observed carries useful information about safety and safety factors. Nevertheless, it is not the intention of this book to contribute to the fields of psychology, human factors, and similar disciplines. The purpose of referring to these fields is to support the main task at hand – measuring safety – with substantiation that serves this purpose. Nevertheless, potential readers who may find this book beneficial are those whose interest is in measuring safety. Given the wide range of research and engineering disciplines involved in the area of safety, the pool of potential readers is large and may include researchers, professionals, and students in psychology, human factors, transportation engineering (including operations, management, design, and planning), industrial engineering, public safety, information and control technology, and even computer science and software engineering, which have recently become involved in developing new solutions for autonomous and connected vehicles that communicate with each other and with the transportation infrastructure.

This book is not a typical overview of past research on roadway safety, and it also does not present all the approaches proposed through the years, for the following reason: the literature on the subject is vast with sometimes inconsistent concepts, terminology, and results. The book covers as much of the existing knowledge as needed; presenting the entirety of the body of work would lead to an incoherent presentation with many contradicting points. Instead, a consistent system of concepts and terms is introduced that is complete insofar as the chosen approach, but not in a global sense. This practice of selective presentation is frequently used in science; however, the author of this book is far from claiming the same level of rigor followed by some scientific publications. Safety-related performance measures, such as reaction time and situation comprehension, and safety-related behaviors, such as speed selection and traffic violation, are sometimes used to evaluate the effects of safety countermeasures. Although these measures may be useful in certain conditions, they do not meet the definition of surrogate measures of safety adopted in this book, and are therefore not covered.

References

AASHTO.  Highway Safety Manual  Washington, DC. 2010.

Brimley B, Saito M, Schultz G. Calibration of highway safety manual safety performance function, development of new models for rural two-lane two-way highways.  Transportation Research Record: Journal of the Transportation Research Board . 2012;2279:82–89.

Farid A, Abdel-Aty M, Lee J, Eluru N, Wang J.-H. Exploring the transferability of safety performance functions.  Accident Analysis & Prevention . 2016;94:143–152.

FARS. FARS 1990-2012 Final File, 2013-2014 ARF. 2015. https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812219.

Hauer E.  Quoted in the Toronto Globe and Mail  November 12, 1986 issue. 1986.

Marshall R. Local rural road safety in SD - what does the data tell us? In:  Presented at the 2007 Transportation Safety Conference . South Dakota Department of Public Safety; February 21, 2007 Posted at. https://dps.sd.gov/enforcement/highway_safety/2007_conference.aspx.

Persaud B, Lord D, Palmisano J. Calibration and transferability of accident prediction models for urban intersections.  Transportation Research Record: Journal of the Transportation Research Board . 2002;1784:57–64.

Chapter 2

Crashes in safety analysis

Abstract

A crash occurs as a result of human error of some sort. Among safety factors, human mistakes are the leading cause of a crash. Haddon's matrix splits safety factors into three periods: pre-crash, crash, and post-crash. An event is reported as a crash if it is sufficiently harmful. A surrogate event definition must provide consistent results. The majority of safety models estimated with crash data assume the Poisson-Gamma mix of distributions. Crash statistical models typically carry quite limited safety information and are subject to multiple concerns, including long periods of data collection and related strong data aggregation, poor quality of crash reports, endogeneity, crash underreporting, and limited ability to estimate temporal safety effects.

Keywords

Human error; Haddon's matrix; Poisson-gamma distribution; Estimation confidence; Data aggregation; Endogeneity

Human factor in crash occurrence

Safety factors in Haddon's periods

Crash probability and frequency

Crash