Highway Safety Analytics and Modeling

By Dominique Lord, Xiao Qin and Srinivas R. Geedipally

Highway Safety Analytics and Modeling comprehensively covers the key elements needed to make effective transportation engineering and policy decisions based on highway safety data analysis in a single. reference. The book includes all aspects of the decision-making process, from collecting and assembling data to developing models and evaluating analysis results. It discusses the challenges of working with crash and naturalistic data, identifies problems and proposes well-researched methods to solve them. Finally, the book examines the nuances associated with safety data analysis and shows how to best use the information to develop countermeasures, policies, and programs to reduce the frequency and severity of traffic crashes.

• Complements the Highway Safety Manual by the American Association of State Highway and Transportation Officials
• Provides examples and case studies for most models and methods
• Includes learning aids such as online data, examples and solutions to problems

• Language: English
• Publisher: Elsevier Science
• Release date: Feb 27, 2021
• ISBN: 9780128168196

Dominique Lord

Dr. Dominique Lord is a professor and holder of the A.P. and Florence Wiley Faculty Fellowship in the Zachry Department of Civil and Environmental Engineering at Texas A&M University. Over the last 27 years, Dr. Lord has conducted numerous research studies in the United States, Canada, and across the world in highway design and safety. Dr. Lord's primary interests are conducting fundamental research on accident analysis methodology, new and innovative statistical methods for modeling motor vehicle collisions, and before-after evaluation techniques. He has extensive experience in data analysis techniques and developed new tools that have been used by engineers and scientists across the world. His other research interests include problems associated with the crash data collection process, safety audits, and traffic flow theory. He has had more than 150 papers published in peer-reviewed journals and more than 140 papers presented at international conferences with a peer-reviewed process.

Book preview

Highway Safety Analytics and Modeling

Dominique Lord

Xiao Qin

Srinivas R. Geedipally

Table of Contents

Cover image

Title page

Copyright

Dedication

Preface

Chapter 1. Introduction

1.1. Motivation

1.2. Important features of this textbook

1.3. Organization of textbook

I. Theory and background

Chapter 2. Fundamentals and data collection

2.1. Introduction

2.2. Crash process: drivers, roadways, and vehicles

2.3. Crash process: analytical framework

2.4. Sources of data and data collection procedures

2.5. Assembling data

2.6. 4-stage modeling framework

2.7. Methods for evaluating model performance

2.8. Heuristic methods for model selection

Chapter 3. Crash–frequency modeling

3.1. Introduction

3.2. Basic nomenclature

3.3. Applications of crash-frequency models

3.4. Sources of dispersion

3.5. Basic count models

3.6. Generalized count models for underdispersion

3.7. Finite mixture and multivariate models

3.8. Multi-distribution models

3.9. Models for better capturing unobserved heterogeneity

3.10. Semi- and nonparametric models

3.11. Model selection

Chapter 4. Crash-severity modeling

4.1. Introduction

4.2. Characteristics of crash injury severity data and methodological challenges

4.3. Random utility model

4.4. Modeling crash severity as an unordered discrete outcome

4.5. Modeling crash severity as an ordered discrete outcome

4.6. Model interpretation

II. Highway safety analyses

Chapter 5. Exploratory analyses of safety data

5.1. Introduction

5.2. Quantitative techniques

5.3. Graphical techniques

Chapter 6. Cross-sectional and panel studies in safety

6.1. Introduction

6.2. Types of data

6.3. Data and modeling issues

6.4. Data aggregation

6.5. Application of crash-frequency and crash-severity models

6.6. Other study types

Chapter 7. Before–after studies in highway safety

7.1. Introduction

7.2. Critical issues with before–after studies

7.3. Basic methods

7.4. Bayesian methods

7.5. Adjusting for site selection bias

7.6. Propensity score matching method

7.7. Before–after study using survival analysis

7.8. Sample size calculations

Chapter 8. Identification of hazardous sites

8.1. Introduction

8.2. Observed crash methods

8.3. Predicted crash methods

8.4. Bayesian methods

8.5. Combined criteria

8.6. Geostatistical methods

8.7. Crash concentration location methods

8.8. Proactive methods

8.9. Evaluating site selection methods

Chapter 9. Models for spatial data

9.1. Introduction

9.2. Spatial data and data models

9.3. Measurement of spatial association

9.4. Spatial weights and distance decay models

9.5. Point data analysis

9.6. Spatial regression analysis

Chapter 10. Capacity, mobility, and safety

10.1. Introduction

10.2. Modeling space between vehicles

10.3. Safety as a function of traffic flow

10.4. Characterizing crashes by real-time traffic

10.5. Predicting imminent crash likelihood

10.6. Real-time predictive analysis of crashes

10.7. Using traffic simulation to predict crashes

III. Alternative safety analyses

Chapter 11. Surrogate safety measures

11.1. Introduction

11.2. An historical perspective

11.3. Traffic conflicts technique

11.4. Field survey of traffic conflicts

11.5. Proximal surrogate safety measures

11.6. Theoretical development of safety surrogate measures

11.7. Safety surrogate measures from traffic microsimulation models

11.8. Safety surrogate measures from video and emerging data sources

Chapter 12. Data mining and machine learning techniques

12.1. Introduction

12.2. Association rules

12.3. Clustering analysis

12.4. Decision tree model

12.5. Bayesian networks

12.6. Neural network

12.7. Support vector machines

12.8. Sensitivity analysis

IV. Appendices

Appendix A. Negative binomial regression models and estimation methods

Appendix B. Summary of crash-frequency and crash-severity models in highway safety

Appendix C. Computing codes

Appendix D. List of exercise datasets

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 © 2021 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: 978-0-12-816818-9

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

Publisher: Joe Hayton

Acquisitions Editor: Brian Romer

Editorial Project Manager: Barbara Makinster

Production Project Manager: Swapna Srinivasan

Cover Designer: Mark Rogers

Typeset by TNQ Technologies

Dedication

Dominique Lord:

To my family (Leah and Javier), my mother (Diane), my brother (Sébastien), and my two former advisors (Dr. Ezra Hauer and Dr. Bhagwant Persaud).

Xiao Qin:

To my family (Yuchen, Ethan, and Eva), my parents (Xingpo and Guangqin), my brother (Hui), and my former advisor (Dr. John Ivan)

Srinivas R. Geedipally:

To my family (Ashwini, Akshath, and Svidha), my parents (Ram Reddy and Laxmi), and my brother (Rajasekhar Reddy). Special thanks to my former advisor, Dr. Lord, for involving me in this project.

Preface

The primary purpose of this textbook is to provide the state-of-the-art knowledge about how to better analyze safety data given their unique characteristics. This textbook provides the latest tools and methods documented in the highway safety literature, some of which have been developed or introduced by the authors. The textbook covers all aspects of the decision-making process, from collecting and assembling data to making decisions based on the analysis results, and is supplemented by real-world examples and case studies to help understand the state of practice on the application of models and methods in highway safety. Where warranted, helpful hints and suggestions are provided by the authors in the text to support the analysis and interpretation of safety data.

The textbook is suitable for college students, safety practitioners (e.g., traffic engineers, highway designers, data analysts), scientists, and researchers who work in highway safety. This textbook specifically complements the Highway Safety Manual (HSM) published by AAHSTO and the Road Safety Manual (RSM) by the World Road Association. The publication of the HSM, RSM, and other safety-oriented guidelines has substantially increased the demand for training engineers and scientists about understanding the concepts and methods outlined within. Hence, the content of this textbook helps fill in this gap by describing the methods in greater depth and allows the readers to broaden their knowledge about the fundamental principles and theories of highway safety.

All three authors of this textbook have taught graduate-level courses in highway safety at different institutions. The material covered had to be used from various sources, including chapters (or part of them) of various textbooks in areas within and peripheral to highway safety, published peer-reviewed papers, class notes from the world leaders in highway safety (e.g., Dr. Ezra Hauer), research reports, and manuals published by national public agencies. Most of these materials did not contain exercises and problems that students could use to apply the knowledge acquired from these documents. Throughout the years, it became clear that a textbook was needed that could combine all these important topics into a single document. The one from which students could read and learn about theoretical principles and apply them using observed (or simulated) data. In this regard, the textbook includes more than nine datasets for more than 40 exercises. Most of these datasets have been used in peer-reviewed publications. All the datasets can be found at the lead author's website: https://ceprofs.civil.tamu.edu/dlord/Highway_Safety_Analytics_and_Modeling.htm.

The content of the textbook is based on an accumulation of more than 40 years of research and applications related to methods and tools utilized for analyzing safety data. The textbook is divided into three general areas. The first area includes chapters that describe fundamental and theoretical principles associated with safety data analyses. This area covers the nature of the crash process from the human and statistical/mathematical perspectives, as well as key crash-frequency and crash-severity models that have been developed in the highway safety literature. The second area groups chapters that describe how the various models described in the first area are applied. The chapters include methods for exploring safety data, conducting cross-sectional and before-after studies, identifying hazardous sites or sites with promise as well as tools for incorporating spatial correlation and identifying crash risk on a near real-time basis. The third area assembles alternative safety analysis tools. The methods include how to use surrogate measures of safety and data mining techniques for extracting relevant information from datasets, including those categorized as big data (e.g., naturalistic data).

It is hoped that the content will help readers to better understand the analytical tools that have been used to analyze safety data to make informed decisions for reducing the negative effects associated with crashes across the globe. This is even more important given the Vision Zero programs that have been increasingly implemented by various agencies in Europe, North America, and Eurasia among others. The content should also help improve or develop new tools aimed at estimating the safety performance of connected and automated vehicles, especially when they will be deployed in mixed-driving environments (within the next decade).

For implementing methods and techniques proposed in this textbook, the authors have provided computer codes for three advanced software languages. Of course, the methods are not restricted to just three, but many other software languages can be easily implemented to be utilized given the parameterization described in the textbook. Along the same line, Microsoft Excel provides simple, flexible, and adequate tools that can be used to implement various simpler methods, such as the graphical methods presented in Chapter 5 or before-after studies described in Chapter 7.

This textbook would never have come to completion without the significant help and input from numerous individuals, colleagues, and former and current graduate students: Zhi Chen, Soma Dhavala, Kathleen Fitzgerald-Ellis, Ali Shirazi, Ioannis Tsapakis, Yuanchang Xie, Chengcheng Xu, and Lai Zheng. After a few requests on social media, several people have offered information about getting access to safety databases or giving us permission to use datasets. They include Jonathan Aguero-Valverde (Costa Rica), Amir Pooyan Afgahri (Australia), David Llopis Castelló (Spain), Aline Chouinard (Canada), Stijn Daniels (Belgium), Thomas Jonsson (Sweden) Neeraj Kumar (Netherlands), Pei Fen Kuo (Taiwan), Emad Soroori (Australia), Shawn Turner (New Zealand), and Simon Washington (Australia).

Finally, this textbook project would not have been possible without the support from Elsevier. First, a large thank you to Brian Romer, who first approached the authors several years ago and convinced us to prepare a book on highway safety (given our reluctance about the effort needed for such an endeavor). Thanks to the two book managers who kept us on our toes for the duration of this project: Barbara Makinster and Ali Afzal-Khan. Special thanks to Narmatha Mohan for helping us manage copyright information and permission log, and Swapna Srinivasan for handling the production of the textbook. The content of this textbook has been partly funded by the A.P. and Florence Wiley Faculty Fellow provided by the College of Engineering at Texas A&M University and project 01-001 from the Safety through Disruption (Safe-D) University Transportation Center (UTC).

Dominique Lord,     Texas A&M University

Xiao Qin,     University of Wisconsin—Milwaukee

Srinivas R. Geedipally,     Texas A&M University

Chapter 1: Introduction

Abstract

Despite the concerted effort by governments around the world in reducing the number and severity of crashes, highway crashes are still a leading cause of fatal and nonfatal injuries. This textbook introduces the characteristics and problems associated with highway crashes and presents how practitioners, scientists, and researchers can make engineering and policy-based decisions using data-driven approaches. The textbook is divided into three general themes: (1) theory and background of highway safety data analysis, (2) methods and tools for conducting fundamental safety studies, and (3) alternative safety analysis methods for estimating the safety performance of highways, road users, or other kinds of entities. How researchers can overcome challenges and seize the opportunities for reducing the morbidity associated with highway crash data over the next few decades is also discussed.

Keywords

Crashdata; Data-driven methods; Global highway safety; Highway fatalities; Highway safety; Highway safety researchers; Policy-based decision; Reducing crashes; Statistics; Traffic engineers

1.1. Motivation

Although a lot of effort has been placed by agencies across the world to reduce the number and severity of crashes ¹ via improvements in highway design, vehicle technology, traffic policy, emergency services, and the like, the effects of highway crashes on road transport networks are still a major source of morbidity (Lord and Washington, 2018). Fig. 1.1 illustrates the historical statistics in roadway fatalities in the United States between 1913 and 2018 (similar trends have been observed among most industrialized countries). This figure shows that the trend in roadway fatalities has been slightly going down since early 1970s, with sharp decreases during economic recessions (further discussed later). This figure also demonstrates that when the values are analyzed by taking into account the vehicle miles traveled (a measure of exposure), the rate has been going significantly down since the beginning of official crash data collected by the federal government. Even though the crash rate shows a great reduction, the raw numbers, as a public health measure, are still the most important factor that guides the allocation of resources. For example, although the crash rate is generally going down, the number of injured people arriving at various emergency rooms located within a jurisdiction, or the patient arrival rate, is the primary metric that the hospital management uses to allocate medical services. The same information is also needed, for example, for managing first responders, such as emergency medical services, firefighters, and national, regional, and local police forces. Hence, the desired attention usually focuses on crash or injury counts for many safety interventions, although exposure in terms of vehicular traffic and/or segment length may still need to be incorporated into some of the methods utilized for assessing safety.

Figure 1.1 Number of fatalities and fatalities per 100 million vehicle miles in the United States between 2013 and 1018 (NSC, 2018).

According to the World Health Organization (WHO), between 2000 and 2016, roadway-related crashes increased from about 1.15 million to 1.35 million deaths globally (WHO, 2018). On an annual basis, about 80 million nonfatal injuries warranting medical care occur on highway networks (Word Bank, 2014). Road traffic injuries are ranked eighth as the leading cause of death (2.5%) among people of all ages, right in front of diarrheal diseases and tuberculosis (WHO, 2018). Vulnerable road users (i.e., pedestrians and cyclists) represent 26% of road injury deaths, while drivers and passengers of motorized two-wheel and three-wheel vehicles account for another 28% worldwide (WHO, 2018). Unfortunately, while a large proportion of high-income countries have observed either a reduction or no change in traffic-related deaths between 2013 and 2016, a significant number of middle- and low-income countries have observed an increase in traffic-related deaths (WHO, 2018), in large part attributed to the rapid motorization observed in developing countries (World Bank, 2014).

The economic burden of crashes significantly impacts the global economy. In the United States, for instance, highway crashes are estimated to have caused more than US$871 billion in economic loss and societal harm in 2010 (Blincoe et al., 2015). In Europe, it is estimated that crashes have cost more than US$325 billion (€280 billion) in economic harm in 2015 (this value is considered underestimated) (Wijnen et al., 2017), while in Australia the economic burden was estimated to be US$ 23.9 billion (AU$33.2) in 2016 (Litchfield, 2017). Globally, it is estimated that 3% of gross domestic product (GDP) is lost to highway crashes (all severities) and can be as high as 5% for middle- and low-income countries (WHO, 2015). In short, in addition to the pain and suffering that crashes have caused to the victims of such events, highway crashes can significantly impede a country's economic growth or viability across the globe.

As described in Fig. 1.1, the relationship that economic activity is strongly linked to the number of fatalities observed on highways has now been well established (Wijnen and Rietveld, 2015; Elvik et al., 2015; Wegman et al., 2017; Noland and Zhou, 2017; Shimu, 2019). In times of economic growth, the number of crashes increases, while during economic hardship (i.e., recession), the number of crashes decreases. Fig. 1.2 illustrates such a relationship in detail, during the Great Recession of 2007–09 in the United States (the right-hand side of Fig. 1.1). The influencing factors include unemployment level, especially among young people, mode shift for people who are unemployed and lower exposure by high-risk drivers (e.g., drivers below 25 years old) during recession periods (Blower et al., 2019). The relationship between economic activity and crash risk is very important to be understood before analytical tools are used for analyzing highway crash data. This is to avoid the potential confounding effects when treatments are implemented and evaluated for reducing the number and severity of crashes.

Figure 1.2 Fatalities trend during the great recession of 2007–09 in the United States (NCS, 2018).

Given the magnitude of the problem associated with highway crashes, numerous public transportation agencies across the world, from national to local agencies, have placed a lot of effort (i.e., labor, promotion, etc.) and allocated a large amount of funds for reducing the number and severity of crashes, especially over the last 25 years. For example, in the United States, the National Highway Transportation Safety Agency (NHTSA) has devoted US$908 million for highway-safety initiatives related to vehicle safety, driver safety, and traffic enforcement in 2016 (NHTSA, 2016). In 2019, the Federal Highway Administration (FHWA) allocated US$2.60 billion solely for safety projects, which include research, dissemination, engineering, and construction projects among others (FHWA, 2019). Similar financial investments have been placed by various transportation agencies in Europe, Middle East, Asia, South Asia, and Oceania. The strong commitment to reducing the negative effects of highway crashes by decision-makers can be seen in the Vision Zero ² movement that was first introduced by the Swedish Government in 1997. This movement consists in finding new and innovative approaches and ways of thinking (i.e., shifting the responsibility from road users to highway designers and engineers for reducing crashes) for significantly reducing, if not eliminating, fatal and nonfatal injuries on highways, especially on urban highways (Kristianssen et al., 2018). Vision Zero has been assertively implemented in various communities across the globe.

To respond to the increasing investment in safety-related projects and help with the aim of reducing, if not eliminating (as per Vision Zero) highway crashes, research into methods and tools for analyzing crash data has exponentially grown during the same time period. The testament of such increase has recently been documented in two scientometric overview publications that visually mapped the knowledge in the field of highway safety (i.e., key areas of research) and the impact of the research that has been published in the leading journal Accident Analysis and Prevention (Zou and Vu, 2019; Zou et al., 2020). These authors identified crash-frequency modeling analysis to be the core research topic in road safety studies, hence showing the relevance of the material covered in this textbook.

Although design and application manuals, such as the Highway Safety Manual (HSM) (AASHTO, 2010) or the Road Safety Manual (RSM) (PIARC, 2019), specialized textbooks, such as the one by Hauer (1997) on before-after studies or Tarko (2020) on surrogate measures of safety, and review papers (see Lord and Mannering, 2010; Savolainen et al., 2011; Mannering and Bhat, 2014), already exist, there is not a single source available that covers the fundamental (and up-to-date) principles related to the analysis of safety data. As discussed by Zou and Vu (2019), the field of highway safety covers very wide areas of research and applications (i.e., psychology, human factors, policy, medicine, law enforcement, epidemiology). Manuals and textbooks have already been published on these topics (see Dewar and Olson, 2007; Shinar, 2007; Smiley, 2015). This textbook complements these published manuals and focuses on the actual analysis of highway safety data.

The primary purpose of this textbook is to provide information for practitioners, engineers, scientists, students, and researchers who are interested in analyzing safety data to make engineering- or policy-based decisions. This book provides the latest tools and methods documented in the literature for analyzing crash data, some of which have in fact been developed or introduced by the authors. The textbook covers all aspects of the decision-making process, from collecting and assembling data to making decisions based on the results of the analyses. Several examples and case studies are provided to help understand models and methods commonly used for analyzing crash data. Where warranted, helpful hints and suggestions are provided by the authors in the text to support the analysis and interpretation of crash data.

The textbook's readership is suitable for highway safety engineers, transportation safety analysts, highway designers, scientists, students, and researchers who work in highway safety. It is expected that the readers have a basic knowledge of statistical principles or an introductory undergraduate-level course in statistics. This textbook specifically complements the HSM published by AAHSTO and the RSM by the World Road Association. The publication of these manuals has increased the demand for training engineers and scientists about understanding the concepts and methods outlined in the HSM and the RSM.

1.2. Important features of this textbook

This textbook is needed for the following reasons:

(1) There are no manuals nor textbooks that summarize all the techniques and statistical methods that can be utilized for analyzing crash data into a single document (although the words crash data are frequently used in this textbook, many methods and techniques can be used for analyzing all types of safety data, such as surrogate measures of safety (i.e., traffic conflicts), speed-related incidents, citations, driver errors or distractions, and the like). The few manuals that cover highway safety concepts usually provide basic information, such as regression equations, figures, charts, or tables that may not always be suitable for the safety analyses. For example, transferring models from one jurisdiction to another may not be feasible for methodological reasons. Furthermore, no manuals specifically explain how to develop crash-frequency models, crash-severity models, or data mining techniques from the data collection procedures to the assessment of the models using data collected in their own jurisdictions.

(2) There are no textbooks that cover all aspects of safety data analyses and can be used in a teaching or classroom environment, such as data collection, statistical analyses, before-after studies, and real-time crash risk analysis among others. This textbook can be used as a core textbook for a senior undergraduate or graduate course in highway safety. Different chapters could also be used for senior-level undergraduate courses that cover some elements of highway safety, highway design, crash data analyses, or statistical analyses.

(3) Crash data are characterized by unique attributes not observed in other fields. These attributes include the low sample mean and small sample size problem, missing values, endogeneity, and serial correlation among others (Lord and Mannering, 2010; Savolainen et al., 2011). These attributes can significantly affect the results of the analysis and are, to this day, often not considered in analyses conducted by transportation safety analysts. Not taking into account, these attributes can lead to misallocation of funds and, more importantly, could potentially increase in the number and severity caused by motor vehicle crashes. The textbook addresses the nuances and complexity related to the analysis of crash and other types of safety data as well as the pitfalls and limitations associated with the methods used to analyze such data.

1.3. Organization of textbook

The textbook is divided into three general areas. The first area includes chapters that describe fundamental and theoretical principles associated with safety data analyses. This area covers the nature of crash data from the human and statistical/mathematical perspectives, as well as key crash-frequency and crash-severity models that have been developed in the highway safety literature. The second area groups chapters that describe how the models described in the first area are applied for analyzing safety data. The chapters include methods for exploring safety data, conducting cross-sectional and before-after studies, identifying hazardous sites or sites with promise as well as tools for incorporating spatial correlation, and identifying crash risk on a near real-time basis. The third area assembles alternative safety analysis tools. The methods include how to use surrogate measures of safety and data mining techniques for extracting relevant information from datasets, including those categorized as big data (e.g., naturalistic data).

1.3.1. Part I: theory and background

Chapter 2 —Fundamentals and Data Collection describes the fundamental concepts related to the crash process and crash data analysis as well as the data collection procedures needed for conducting these analyses. The chapter covers the crash process from the perspectives of drivers, roadways and vehicles, and theoretical and mathematical principles. It provides important information about sources of data and data collection procedures, as well as how to assemble crash and other related data. The chapter also describes a four-step modeling procedure for developing models and analyzing crash data and the methods for assessing the performance of these models.

Chapter 3 —Crash-Frequency Modeling describes the basic nomenclature of the models that have been proposed for analyzing highway safety data and their applications. The chapter describes the most important crash-frequency models that have been proposed for analyzing crash count data, along with the important or relevant information about their characteristics. The models are grouped by their intended use and for handling specific characteristics associated with safety data. The chapter ends with a discussion about the modeling process related crash-frequency models.

Chapter 4 —Crash-Severity Modeling introduces the methodologies and techniques that have been applied to model crash severity in safety studies. The discussion includes the different forms, constructs, and assumptions that crash severity models have been developed as a function of the prevailing issues related to crash data. The theoretical framework and practical techniques for identifying, estimating, evaluating, and interpreting factors contributing to crash injury severities are also explored.

1.3.2. Part II: highway safety analyses

Chapter 5 —Exploratory Analyses of Safety Data describes techniques and methods for exploring safety data. They are divided into two general themes: (1) quantitative techniques that involve the calculation of summary statistics and (2) graphical techniques that employ figures or plots to summarize the data. The exploratory analyses of data help frame the selection of more advanced methodologies such as those associated with cross-sectional analyses, before-after studies, identification of hazardous sites, spatial correlation and capacity, and mobility.

Chapter 6 —Cross-Sectional and Panel Studies in Safety describes different types of data and analysis methods, as well as how models described in the previous part can be used to this effort. The discussion includes data and modeling issues and presents some techniques to overcome them. The chapter describes the characteristics of different functional forms, selection of variables, and modeling framework. Techniques for determining the required sample size, identification of outliers, and transferability of models to other geographical areas are also presented. Lastly, a brief outline of other study designs that are not commonly used in highway safety is presented.

Chapter 7 —Before-After Studies in Safety covers basic and advanced study techniques for analyzing before and after data. The chapter describes the two critical issues that can negatively influence this type of study and the basic methods for conducting a before-after study with and without control groups. Then, the empirical Bayes and full Bayes methods in the context of before-after studies are presented. The last sections of the chapter document more recent methods, such as the naïve adjustment method, the before-after study using survival analysis, and the propensity score method. The chapter ends with a discussion about the sample size needed for conducting before-after studies.

Chapter 8 —Identification of Hazardous Sites first discusses various hazardous site selection methods that rely on observed crashes, predicted crashes, or expected crashes. The discussion includes each method's strengths and weaknesses. Then, the chapter presents geospatial hotspot methods that consider the effects of unmeasured confounding variables by accounting for spatial autocorrelation between the crash events over a geographical space. This chapter also documents the list of the high crash concentration location procedures because the hazardous site selection methods may not efficiently identify the point locations where a deficiency exists. The proactive approach methods are then presented due to their nature of identifying sites before a crash could occur. Lastly, the screening evaluation methods are discussed in detail.

Chapter 9 —Models for Spatial Data is dedicated to analyzing and modeling crash data within a spatial context. The chapter begins with an overview of the characteristics of spatial data and commonly used data models. Then, spatial indicators, such as Getis G and Moran's I, are introduced to help determine the distribution of crash locations as clustering, dispersed, or random. Next, the chapter describes techniques for analyzing crash point data that are presented to facilitate the discovery of the underlying process that generates these points. Finally, spatial regression methods are introduced to explicitly consider the spatial dependency of crashes and spatial heterogeneity in the relationship between crashes and their contributing factors.

Chapter 10 —Capacity, Mobility, and Safety offers a perceptive account of one of the fastest-developing fields in highway safety analysis, involving traffic flow theory, driver behavior models, and statistical methods. The chapter first describes a theoretical car-following model to demonstrate the safety aspects of a classic driver behavior model, the modeling of relationships between crashes and traffic volume, and how to map crash typologies to a variety of traffic regimes characterized by traffic variables. The use of Bayesian theory to predict crash probability given a real-time traffic input and real-time crash prediction models (RTCPM) are also described. The chapter ends with a description about the motivation and methodology for developing RTCPM from simulated traffic data when actual traffic data are not available.

1.3.3. Part III: alternative safety analyses

Chapter 11 —Surrogate Safety Measures focuses on defining, analyzing, comparing, and applying state-of-the-art surrogate safety measures. Following a brief history of traffic conflicts, the chapter explains the basic characteristics of traffic conflicts technique and the practice of observing and collecting traffic conflicts in the field. The chapter also covers both the pragmatic approach and the theoretical development of surrogate safety measures.

Chapter 12 —Data Mining and Machine Learning Techniques introduces data mining and machine learning methodologies and techniques that have been applied to highway safety studies, including association rules, clustering analysis, decision tree models, Bayesian networks, neural networks, and support vector machines. The theoretical frameworks are illustrated through exemplary cases published in safety literature and are supplemented with implementation information in the statistical software package R. The chapter ends with a description of a means of specifying the effect of an independent variable on the output, which can assist in deciding on the appropriate safety solutions.

1.3.4. Appendices

Appendix A describes the basic characteristics of the Negative Binomial model, the most popular model in crash data analysis (Lord and Mannering, 2010), with and without spatial interactions and the steps to estimate the model's parameters using the maximum likelihood estimation and Bayesian methods. Appendix B provides a historical description, a detailed and up-to-date list of crash-frequency and crash-severity models that were previously published in peer-reviewed publications (Lord and Mannering, 2010; Savolainen et al., 2011; Mannering and Bhat, 2014). Appendix C presents useful codes for developing many models described in the textbook in SAS, WinBUGS, and R software languages. Appendix D lists the available datasets for each chapter of this textbook. Finally, datasets used for the examples described in various chapters are made available on the personal website of the lead author (https://ceprofs.civil.tamu.edu/dlord/Highway_Safety_Analytics_and_Modeling.htm).

1.3.5. Future challenges and opportunities

The methods and analysis tools documented in this textbook are the accumulation of more than 40 years of research and applications in highway safety. Many of these methods and tools have been introduced in this area when new methods were developed in other fields, such as in statistics, econometric, medicine, epidemiology, and social sciences or when methodological limitations had been identified based on the unique characteristics associated with highway safety data (see Chapters 3 and 4). Despite the foreseeable changes and uncertainties, the techniques and methods introduced in this textbook should not be outdated and will continue to be used as powerful tools for analyzing highway safety data. However, with the significant advancement in transportation technologies and computing power, existing methods may need to be adapted and new ones to be developed to properly measure the safety performance associated with these technologies over the next few decades.

Although the full development of connected and autonomous vehicles is several years, if not decades, away, the impacts of their deployment in mixed traffic conditions (a mixture of human-driven and automated vehicles) are not well understood. Automated vehicles have the potential to significantly reduce vehicle fatalities, but their safety benefits have so far mainly been based on simulation analyses and surrogate measures of safety (Morando et al., 2018; Mousavi et al., 2019; Papadoulis et al., 2019; Sohrabi et al., 2020). On the other hand, crashes involving autonomous vehicles have been reported with limited open road tests, some involving fatalities. Hence, observational tools targeted for small samples would provide a more accurate and reliable picture than simulations and surrogate events.

In the new era of Big Data, a large amount of new and emerging data are becoming more and more available (i.e., smart cities, disruptive technologies, naturalistic data, video processing, etc.). In Chapter 12—Data Mining and Machine Learning Techniques, data mining and machine learning techniques for analyzing safety data, such as those collected from naturalistic studies or from connected vehicles (e.g., basic safety messages) are discussed. With the rich data, extracting useful and meaningful information becomes essential. Many competing techniques are capable of handling conventional safety data issues, their strengths and limitations vary, so do the model performance and results. Examples include random forests versus gradient boosted trees, and convolutional neural networks versus recurrent neural networks. Understanding the fundamentals of these techniques that are developed from some of the methodological and modeling principles introduced in the textbook will help the reader to select the most appropriate method when confronted by different data issues and challenges.

Despite the superior performance in handling high-dimensional data, machine learning methods have long been criticized for operating like a black box, with no statistical inferences and model goodness-of-fit or no explicit relationships between outcomes and input variables. Hence, there is a trend to use machine learning techniques as a screening tool for a large quantity of factors, or a clustering tool for grouping data into more homogeneous dataset, and then to apply conventional statistical models. Promising results have been reported in this combination of methods. On the other hand, with the increased use of naturalistic data in safety, new tools have been and are being developed to handle datasets that include video data, social media data, and vehicle performance data that record vehicle location, position, and kinematics every second or fraction of a second. Applying artificial intelligence methods, such as those currently being used by YouTube, ³ for example, should be examined. As the authors, we hope this textbook will serve as a springboard for the reader to continue advancing the safety research frontier through better analytical methods.

References

31. AASHTO.  Highway safety manual . 1st Edition. Washington, D.C: In: American Association of State Highways and Transportation Officials; 2010.

1. Blincoe L, Miller T.D, Zaloshnja E, Lawrence B.A.  The Economic and Societal Impact of Motor Vehicle Crashes, 2010 (Revised). (Technical Report DOT HS 812 013) . Washington, D.C.: U.S. Department of Transportation, National Highway Traffic Safety Administration; 2015.

2. Blower D, Flannagan C, Geedipally S, Lord D, Wunderlich R.  Identification of Factors Contributing to the Decline of Traffic Fatalities in the United States from 2008 to 2012. Final Report NCHRP Project 17-67 . Washington, D.C.: Transportation Research Board; 2019.

3. Dewar R.E, Olson P.L.  Human Factors in Traffic Safety . second ed. Tucson, AZ: Lawyers & Judges Publishing Company, Inc.; 2007.

4. Elvik R. An analysis of the relationship between economic performance and the development of road safety. In:  Why Does Road Safety Improve when Economic Times are Hard? ITF/IRTAD, Paris . 2015:43–142.

5. FHWA. FHWA FY 2019 Budget. Washington, D.C: Federal Highway Administration; 2019. https://www.fhwa.dot.gov/cfo/fhwa-fy-2019-cj-final.pdf.

6. Hauer E.  Observational Before–After Studies in Road Safety: Estimating the Effect of Highway and Traffic Engineering Measures on Road Safety . Oxford, United Kingdom: Pergamon Press, Elsevier Science, Ltd.; 1997. .

7. Kristianssen A.-C, Andersson R, Belin M.-Å, Nilsene P. Swedish Vision Zero policies for safety – a comparative policy content analysis.  Saf. Sci.  2018;103:260–269.

8. Litchfield F.  The Cost of Road Crashes in Australia 2016: An Overview of Safety Strategies . Canberra, Australia: The Australian National University; 2017.

9. Lord D, Mannering F. The statistical analysis of crash-frequency data: a review and assessment of methodological alternatives.  Transport. Res. Part A . 2010;44(5):291–305.

10. Lord D, Washington S.  Safe Mobility: Challenges, Methodology and Solutions (Transport and Sustainability . vol. 11. Emerald Publishing Limited; 2018.

11. Mannering F, Bhat C.R. Analytic methods in accident research: methodological frontier and future directions.  Anal. Methods Accid. Res.  2014;1:1–22.

12. Morando M.M, Tian Q, Truong L.T, Vu H.L. Studying the safety impact of autonomous vehicles using simulation-based surrogate safety measures.  J. Adv. Transport.  2018;2018:11. doi: 10.1155/2018/6135183 Article ID 6135183 (open access).

35. Mousavi S.M, Lord D, Osman O.A. Impact of Urban Arterial Traffic LOS on the Vehicle Density of Different Lanes of the Arterial in Proximity of an Unsignalized Intersection for Autonomous Vehicle vs. Conventional Vehicle Environments.  In: Paper presented at the ASCE International Conference on Transportation & Development . 2019 Alexandria, VA, June 9-12.

13. NHTSA, .  Fiscal Year 2016 Budget Overview . Washington, D.C: National Highway Traffic Safety Administration; 2016.

14. Noland R.B, Zhou Y. Has the great recession and its aftermath reduced traffic fatalities?  Accid. Anal. Prev.  2017;98:130–138.

15. NSC, . Historical Fatality Trends. Washington D.C: National Safety Council; 2018. https://injuryfacts.nsc.org/motor-vehicle/historical-fatality-trends/deaths-and-rates/.

16. Papadoulis A, Quddus M, Imprialou M. Evaluating the safety impact of connected and autonomous vehicles on motorways.  Accid. Anal. Prev.  March 2019;124:12–22.

17. PIARC, .  Roadway Safety Manual . third ed. Paris, France: World Road Association; 2019.

18. Savolainen P.T, Mannering F.L, Lord D, Quddus M.A. The statistical analysis of highway crash-injury severities: a review and assessment of methodological alternatives.  Accid. Anal. Prev.  2011;43(5):1666–1676.

19. Shimu T.  Examining the Factors Causing a Drastic Reduction and Subsequent Increase of Roadway Fatalities on United States Highways between 2005 and 2016  MS Thesis. College Station, TX: Zachry Department of Civil & Environmental Engineering, Texas A&M University; 2019.

20. Shinar D.  Traffic Safety and Human Behavior . Bingley, UK: Emerald Group Publishing Limited; 2007.

21. Smiley A.  Human Factors in Traffic Safety . third ed. Tucson, AZ: Lawyers & Judges Publishing Company, Inc.; 2015.

34. Sohrabi S, Khodadadi A, Mousavi S.M, Dadashova B, Lord D.  Quantifying Autonomous Vehicle Safety: A Scoping Review of the Literature, Evaluation of Methods, and Directions for Future Research .  Working Paper . College Station, TX: Zachry Department of Civil and Environmental Engineering, Texas A&M University; 2020 (submitted for publication).

22. Tarko A.  Measuring Road Safety with Surrogate Events . Amsterdam, The Neatherlands: Elsevier Inc.; 2020.

23. Wegman F, Allsop R, Antoniou C, Bergel-Hayat R, Elvik R, Lassarre S, Lloyd D, Wijnen W.How did the economic recession (2008–2010) influence traffic fatalities 18 in OECD-countries?  Accid. Anal. Prev.  2017;102:51–59. .

24. Wijnen W, Rietveld P. The impact of economic development on road safety: a literature review. In:  Why Does Road Safety Improve when Economic Times are Hard? ITF/IRTAD, Paris . 2015:22–42.

25. Wijnen W, Weijermars W, Vanden Berghe W, Schoeters A, Bauer R, Carnis L, Elvik R, Theofilatos A, Filtness A, Reed S, Perez C, Martensen H.Crash Cost Estimates for European Countries. Research Report. The Hague, The Netherlands: Institute for Road Safety Research (SWVO); 2017. https://www.swov.nl/publicatie/crash-cost-estimates-european-countries.

26. WHO. Global Status Report on Road Safety 2015. Geneva, Switzerland: World Health Organization, WHO Press; 2015. http://www.who.int/violence_injury_prevention/road_safety_status/2015/en/.

27. WHO. Global Status Report on Road Safety 2018. Geneva, Switzerland: World Health Organization, WHO Press; 2018. http://www.who.int/violence_injury_prevention/road_safety_status/2018/en/.

28. World Bank, . Transport for Health: The Global Burden of Disease from Motorized Transport. Seattle, WA. Washington, D.C: IHME; 2014. http://documents.worldbank.org/curated/en/984261468327002120/pdf/863040IHME0T4H0ORLD0BANK0compressed.pdf.

29. Zou X, Vu H.L. Mapping the knowledge domain of road safety studies: a scientometric analysis.  Accid. Anal. Prev.  2019;132:105243.

30. Zou X, Vu H.L, Huang H. Fifty years of accident analysis & prevention: a bibliometric and scientometric overview.  Accid. Anal. Prev.  September 2020;144:105568.

---

¹  

In this textbook, we use the term crash to reflect outcome of a collision between a vehicle and a fixed object (i.e., an event where only one vehicle is involved), one or more vehicles, or one or more vulnerable road users (i.e., pedestrians, cyclists, etc.). Although some people do not like to label a crash an accident because the word accident could absolve the driver of any responsibility, the word accident could still be employed as that word refers to the probabilistic nature of the event. If accidents were coming from a deterministic system, we should therefore be able to predict with certainty when one or more crashes would occur in the future. Obviously, in the context of this textbook, this is not possible.

²  

https://visionzeronetwork.org/.

³  

https://www.forbes.com/sites/bernardmarr/2019/08/23/the-amazing-ways-youtube-uses-artificial-intelligence-and-machine-learning/#1f27802e5852.

I

Theory and background

Outline

Chapter 2. Fundamentals and data collection

Chapter 3. Crash—frequency modeling

Chapter 4. Crash-severity modeling

Chapter 2: Fundamentals and data collection

Abstract

Crashes are very complex and multidimensional events. They can be analyzed from the perspective of driver, roadway environment, and vehicle. They can also be analyzed from the perspective of mathematical equations. Various types of data can be collected for analyzing crash data. They include crash, traffic flow, roadway characteristics, vehicle occupants, site visits, Google Earth, and Streetview. A four-stage modeling framework is proposed for analyzing highway crash data. Models can be estimated using the maximum likelihood estimation or the Bayesian methods. Several goodness-of-fit methods are described for evaluating crash-frequency and crash severity models. The goodness-of-fit methods to evaluate the performance of models can be grouped as either likelihood-based or model error-based.

Keywords

Bayesian method; Crash data; Crash-frequency models; Crash-severity models; Data collection; Goodness-of-fit; Goodness-of-logic; Human factors; Maximum likelihood; Modeling objectives; Poisson process; Poisson trials; Roadway data; Traffic flow data

2.1. Introduction

Crashes are very complex and multidimensional events. Although, in police reports, a single factor may have been identified as the primary cause of the crash, usually several interrelated factors could have contributed to a crash. For example, we can have a scenario where an 18-year old driver, who is traveling late at night during very windy conditions in a 20-year old pickup truck, starts dosing off, then runs off the road in a horizontal curve with low-friction pavement due to lack of maintenance and a radius that is below the design standards (but approved via a design exemption), and then hits a tree located within the designated clear zone. In this scenario, the primary factor may have been identified as falling asleep behind the wheel, but if we remove any other factors, the crash could have been avoided (e.g., adequate maintenance, no tree, no wind, on a tangent section). In addition to showing that contributing and interrelated factors can be related to the driver, the vehicle or the roadway, this scenario highlights that crashes are complex and probabilistic events (if they were deterministic events, we would be able to know when and where a crash would happen). Hence, all relevant events need to be analyzed with appropriate tools in order to account for the complexity and randomness of crash data.

This chapter describes the fundamental concepts related to the crash process and crash data analysis as well as the data collection procedures needed for conducting these analyses. The first section covers the crash process from the perspective of drivers, roadways, and vehicles. The second section describes the crash process from a theoretical and mathematical perspective. The third section provides important information about sources of data and data collection procedures. The fourth section describes how