EXAMINING DRIVER RISK FACTORS IN ROAD DEPARTURE CONFLICTS USING SHRP2 DATA. Thesis. Submitted to. The School of Engineering of the

Transcription

1 EXAMINING DRIVER RISK FACTORS IN ROAD DEPARTURE CONFLICTS USING SHRP2 DATA Thesis Submitted to The School of Engineering of the UNIVERSITY OF DAYTON In Partial Fulfillment of the Requirements for The Degree of Master of Science in Civil Engineering By Danah Alshatti Dayton, Ohio May 2018

2 EXAMINING DRIVER RISK FACTORS IN ROAD DEPARTURE CONFLICTS USING SHRP2 DATA Name: Alshatti, Danah Ahmad APPROVED BY: Deogratias Eustace, Ph.D., P.E., PTOE Advisory Committee Chairperson, Associate Professor, Department of Civil and Environmental Engineering and Engineering Mechanics Peter Hovey, Ph.D. Committee Member, Associate Professor, Department of Mathematics Paul Goodhue, P.E., PTOE Committee Member, Traffic Key Services Leader, LJB, Inc. Robert J. Wilkens, Ph.D., P.E. Associate Dean for Research & Innovation Professor School of Engineering Eddy Rojas, Ph.D., M.A., P.E. Dean, School of Engineering ii

3 Copyright by Danah Ahmad Alshatti All rights reserved 2018 iii

4 ABSTRACT EXAMINING DRIVER RISK FACTORS IN ROAD DEPARTURE CONFLICTS Name: Alshatti, Danah Ahmad University of Dayton Advisor: Dr. Deogratias Eustace USING SHRP2 DATA When a vehicle leaves the roadway, the likelihood of a roadway departure (RwD) crash can be deadly. Roadway departure conflicts usually involve a single vehicle, which occurs after a vehicle crosses an edge line, a centerline, or otherwise leaves the designated traveled way and collide with another vehicle or with a fixed object or overturns, etc. This study investigates the nature of the interrelations between roadway, vehicle, and driver (characteristics and behavior) risk factors in roadway departure conflicts. The purpose of this thesis study was to examine which factors increase the risk of roadway departure conflict and increase the likelihood becoming a roadway departure crash, using the Second Strategic Highway Research program (SHRP2) data. SHRP2 include Naturalistic Driving Study (NDS) and Roadway Information Database (RID), which were collected from six different states in Stepwise logistic and generalized linear regression models were estimated to provide insights as to those factors that have association with roadway departure conflicts and more importantly to those that are more likely to lead the conflict into crashes. iv

5 The results revealed that drivers pre-incident maneuvers, judgment maneuvers, secondary tasks (distracted drivers), road alignment (curves) were significant factors. Driver education, average mileage driven per years were also significant factors. However, driver gender and age were non-significant risk factor of roadway departure conflicts in the current study. v

6 DEDICATION I dedicate this research to God Almighty my initiator, my source of inspiration, wisdom, knowledge and understanding. He has been the source of my strength throughout this research. I also dedicate this thesis to my family, especially my mother, May Alhasan that supported my dream and help accomplish it through ups and downs. In addition I would like to thank my brothers and sister, who where thousands of miles away, and never stopped supporting and encouraging. Last but not least, my deepest gratitude goes to my beloved fiancé, for all the things that can never be quantified. God bless you. To those who indirectly contributed in this research, your kindness means a lot to me. Thank you very much. vi

7 ACKNOWLEDGMENTS I am deeply thankful to my research supervisor Dr. Deogratias Eustace, whose continuous guidance, feedback, advice and encouragements have been truly exceptional. Thank you for providing the time and equipment necessary for this research, and for directing this thesis and bringing it to its conclusion with patience and expertise. I thank Dr. Peter Hovey, the committee member, who provided his statistics insights and expertise that greatly assisted in my research study. He is always there whenever I needed his clarification on statistics matters. I also thank Mr. Paul Goodhue, a committee member who also reviewed my manuscript and provided comments that greatly improved my thesis manuscript. I sincerely thank Candise J. Powell, Chair of the Institutional Review Board (IRB) for assistance with data acquisition, and sincere graduate to Whitney Atkins of Virginia Tech Transportation Institute (VTTI) for answering my questions for everything about the data. In addition, I also thank Ron Sanders of UDit who assisted in securing the data and provided all the needed assistance and equipment. vii

8 TABLE OF CONTENTS ABSTRACT... iv DEDICATION... vi ACKNOWLEDGMENTS... vii LIST OF FIGURES... x LIST OF TABLES... xi LIST OF SYMBOLS/ABBREVIATIONS... xii CHAPTER INTRODUCTION Background Problem Statement Research Objective Thesis Structure... 4 CHAPTER Literature Review General Overview Relationship between Roadway vs. Roadway Departure Risk Factors Average Miles Traveled Number of Lanes Alignment Locality Relationship between Driver and Roadway Departure Risk Factors Characteristics Behavioral Risk Factors Relationship between Vehicle and Roadway Departure Risk Factors CHAPTER viii

9 Data and Methodology Data Description Second Strategic Highway Research Program (SHRP2) Data Collection Data Request Scope Merging Files Creating of Roadway Departure-related Traffic Event Database Description of Selected Variables for Study Methodology Introduction Statistical Analysis CHAPTER RESULTS AND DISCUSSION Introduction Results of the Stepwise Logistic Regression Results of Generalized Linear Regression CHAPTER CONCLUSIONS AND RECOMMENDATIONS Conclusions Recommendations REFERENCES APPENDIX ix

10 LIST OF FIGURES Figure 2.1 Female injury risk compared to males (Carlsson, 2010) Figure 3.1 DAS four continuous video view from the vehicle interior Figure 3.2 DAS schematic view Figure 3.3 SHRP2 Insight Query Builder Figure 3.4 Researcher Dictionary for Video Reduction Data: Event Nature, Incident Type, and Vehicle Configuration Figure 3.5 Description of incident types Figure 4.1 Receiver Operating Characteristics (ROC) - stepwise selection x

11 LIST OF TABLES Table 2.1: Ohio Average Annual Roadway Departure Crash Outcomes, Table 3.1 Description of crash and event severity, and vehicle classification Table 3.2 Description of driver demographics Table 3.3 Description of Driver behavior risk factors Table 3.4 Description of roadway characteristics Table 4.1 Summary of Stepwise Selection in the Logistic regression model Table 4.2 Odds ratio estimates Table 4.3 Analysis of the GEE parameter estimates xi

12 LIST OF SYMBOLS/ABBREVIATIONS AASHTO ADT CDS FARS FHWA GES IRB NASS NASS-GES NCHRP NDS NHTSA RID ROR RwD SHRP2 SVROR TRAID VTTI American Association of State Highway and Transportation Official Average Daily Traffic Crashworthiness Data System Fatality Analysis Reporting System Federal Highway Administration General Estimates System Institutional Review Board National Automotive Sampling System National Automotive Sampling System General Estimates System National Cooperative Highway Research Naturalistic Driving Study National Highway Traffic Safety Administration Roadway Information Database Run-off-Road Roadway Departure Second Strategic Highway Research Program Single Vehicle run-off-road Canadian Traffic Accident Information Databank Virginia Tech Transportation Institute xii

13 CHAPTER 1 INTRODUCTION 1.1 Background The Federal Highway Administration (FHWA) defines a roadway departure (RwD) crashes as a crash in which a vehicle crosses an edge line, a centerline, or otherwise leaves the traveled way (FHWA, 2017). Various studies have reported that roadway departure (RwD) crashes are a major contributor of highway fatalities (e.g., Neuman et al. 2003; Liu and Subramanian, 2009; Jalayer. and Zhou, 2016; FHWA 2017). Neuman et al. (2003) estimates that 39% of national fatal crashes are single-vehicle run-off-road (SVROR) crashes. According to the FHWA (2006), over 25,000 people in 2005 were killed because drivers left their lane and crashed with an oncoming vehicle, rolled over, or hit an object located alongside the highway. Furthermore, based on more recent traffic crash data, the FHWA (2017) estimates that between 2014 and 2016, 53% of roadway fatalities in the United States involved lane departure crashes, which resulted into an average of 18, 779 annual fatalities. Moreover, statistics show that the proportion of fatal run-off-road (ROR) crashes has been steadily increasing (e.g., SAIC, 2005; Liu and Subramanian, 2009). RwD and ROR crashes essentially mean the same type of crashes, i.e., in principle the vehicle leaves its lane and gets involved in a collision of some kind or overturns (Eustace et al., 2016). 1

14 1.2 Problem Statement When a vehicle leaves the roadway, the likelihood of a RwD crash can be deadly. Different reasons could cause such an event, for instance, a driver attempting to avoid another vehicle, an animal, or even an object in the travel lane. Major known reasons include inattentive driving due to fatigue, sleepiness, alcohol, drugs or distraction. In addition, the effect of weather and the effect of speed traveled through a curve, or down a grade of a roadway. There is a need to identify important factors that contribute highly or increase the likelihood of the occurrences of ROR crashes in the United States. Although most previous studies examined the interrelationship between crash risk and the characteristics of both roadway and driver by using police-reported crash data, yet still lack capturing minor crashes that not reported ore recorded in police reports or near misses where police officers are not involved. Most prior studies are limited with specific road segments data and environment-related factors only, missing the behavioral risk factors, which make it difficult to prove all driver behaviors or actions that contributed to the occurrence of the crash. The Strategic Highway Research Program (SHRP 2) safety study collected an excellent amount of data on that can provide researchers the ability make comprehensive assessment on driver behavior while driving. In essence, driver behavior, performance and interact with roadway, environmental, vehicular, and human factors can be studied from the videos taken inside and outside the vehicle while the driver is driving. These data can enable a researcher to get additional factors that otherwise will not available from traditional police-reported crash records. 2

15 1.3 Research Objective In order to understand better the interrelationship between driver characteristics and other factors causing crash events, a detailed assessment of individual driver information and behavior is essential. This accounts to the real-time factors observed from driver behavior and response while failing to interact with the roadway or the vehicle, and their impact to crash severity. The goal of this study is to leverage intelligent reports to investigate the nature of the interrelations between driver and roadway departure risk factors, using the Second Strategic Highway Research program (SHRP2). SHRP2 provides unique resources of data, which include Naturalistic Driving Study (NDS) and Roadway Information Database (RID) to demonstrate and develop real-time procedures to examine driver risk factors in road departure conflicts. The integration of the SHRP 2 data provides a detailed driver driving behaviors that would otherwise not easy to observe or obtain from the traditional crash data sources. The advantage of the SHRP2 data is that besides getting information on real crash events, they could capture near misses of lane departure events and situations where the driver was able to steer the vehicle back and continue with the trip as such events can be recorded as departure conflicts. The objectives of the study attempt to answer the following research questions: 1- What is the relation between driver and risk of roadway departure conflicts? 2- What is the relation between roadway departure risk on driver behaviors, vehicle, and roadway characteristics? 3- How driver, vehicle, roadway and environmental factors affect driver behavior? 3

16 This proof-of-concept study focuses primarily on driver behavior on roadway departure conflicts. That is, this study includes events where the actual crashes happened, potential situations where crashes could have happened but did not happen, and near misses whereby for some reasons, the crash was avoided. The research begins with an analysis of driver, vehicle, and roadway geometry and crash/near-crash risk as they relate to the driver behavior. A series of statistical models are estimated to determine whether significant differences in these measures exists, with respect to vehicle types, roadway geometry, and driver characteristics. Additional data are collected for the purpose of a more detailed comparison of vehicle dynamics and driver behavior during crash, near crash, and baseline driving events. 1.4 Thesis Structure This thesis is structured into five chapters; organized in a detailed manner, introducing a topic of interest, provide the background of the research problem, illustrate a perspective of the study, outline the data and select the applicable methodology, demonstrate results to the fundamental research questions addressed, and presenting conclusions and recommendations for further studies. Chapters are arranged as follows: Chapter 1: Introduction This chapter provides background of road departure conflicts. A problem statement of the study follows the background section. The chapter concludes with the research objective to the fundamental research questions addressed. Chapter 2: Literature review This chapter summarizes previous studies with the interest of roadway departure crashes. That includes previous studies on 4

17 contributing risk factors that interrelate with the vehicle, roadway and driver characteristics and behaviors. Chapter 3: Data Description & Methodology This chapter provides a general description of the data collection from the Second Strategic Highway Research Program (SHRP2). This chapter includes a description of the data requested for the purpose of this study, followed with the general formulation of the statistical method selected for this study based on the nature of the data used. Chapter 4: Results and Discussion This chapter presents the results of the study, which include of descriptive statistics and the statistical regression model that was developed using the data utilized in this study. These results are followed by a discussion as to the practical implications of the findings. Chapter 5: Conclusions and Recommendations- This chapter represents the conclusions of this research study and provides some recommendations for further research.. 5

18 CHAPTER 2 LITERATURE REVIEW 2.1 General Overview The American Association of State Highway and Transportation Officials (AASHTO) estimates that in the United States, a lane-departure fatality occurs after every 21 minutes (AASHTO, 2008). Roadway departure single-vehicle crashes are a foremost cause of severe injuries and fatalities along the state highways (Lee and Mannering, 2002; ODOT 2013; Eustace et al., 2016; FHWA 2017), and account for nearly one-third of all highway fatalities. (Lee and Mannering, 2002; ODOT 2013) The Ohio Department of Transportation (ODOT) reports that in the state of Ohio roadway departure crashes account for about 52% of roadway-related fatalities and 37% of serious injuries each year. In addition, ODOT (2013) reports that between 2006 and 2012 on average 11 people died or seriously injured each day due to roadway departure crashes. Table 2.1: Ohio Average Annual Roadway Departure Crash Outcomes, Injury Type Frequency Fatalities 593 Serious Injuries 3,673 Total 4,266 6

19 According to ODOT (2018) roadway departure crashes usually occur due to high speeds, poor weather conditions such as rain or fog, fatigue, and alcohol and eventually a vehicle crosses an edge or center line causing a head-on collision or sideswipe, or runs off the road and hits a fixed object such as a tree or pole. Roadway departure crashes occur when a vehicle leaves its lane, either moving into another lane or off the roadway itself. The result is often disastrous when a vehicle leaves the traveled way, and the risk of injury or fatality during a roadway departure conflict is strongly correlated to the conflict conditions. Conflict conditions are crash factors or actions that contribute to the occurrence of the crash and are considered causes and risk factors. In order to better understand these causes, recorded crash data should be significant to investigate everyday driving under real world circumstances. Many researchers analyze available crash data, however, the crash databases are not always complete and they lack essential information such as non-reported crashes, that is, crashes that are never reported to crash databases or missing crash location information (Jamil, 2006). Police crash reports are the most permeating source of traffic crash data used in many studies. Shinar et al. (1983) found that most police reports provide minimal information concerning driver conditions and states, vehicle defects, and roadway and environmental deficiencies that may be present at scene of the crash. Shinar et al. (1983) presents a study of 124 crashes linked with 207 drivers investigated by police officers and the Multi-Disciplinary Accident Investigation (MDAI) teams. The analyzed police reports show that police officers failed to report view obstructions in 30 crashes and the police 7

20 improperly identified driver age in 11.6% of the total crash drivers, where driver information was simply not recorded (Shinar et al., 1983) An improved approach that empowers the lack of reliable traffic crash records is the traffic conflict technique that relies on observations of critical circumstances to enhance traffic safety analysis. In 1968 two General Motors engineers first introduced the concept of traffic conflicts (Perkins and Harris, 1969). Using the 1966 reported traffic crashes they observed traffic movements at intersections and found that most drivers reacted to possible conflicts by either braking or weaving (Perkins and Harris, 1969). These conflict conditions appeared to be very similar to the condition in an accident; they appeared to have the same characteristics except of the final component, an impact. (Perkins and Harris, 1969). Another principal associated with this study, data where imprecise and not reliably collected. Crash surrogates greatly increase these judgments, since crash frequencies at individual sites are often low and conflicts occur much more frequently than crashes. One needs to collect sufficient data to make the suitable statistical judgment given that traffic conditions change through the years. On the other hand, the Naturalistic Driving Study (NDS) is an innovative research for examining driver behaviors in traffic safety. The NDS data present a sample of driving behaviors, demographic distribution, and a variety of road types. Research results from NDS data may be used to develop strategic solutions that can reduce the number of roadway departure crashes and decrease the severity of crashes, by enhancing the knowledge and providing appropriate solutions and procedures to examine roadway 8

21 departure crashes. The naturalistic driving method promotes two advantages: an accurate pre-crash information that includes information about driving behavior in normal conditions, as well as a detailed context of contributing factors leading to a crash or conflict (Dingus et al. 2016). The NDS study examined video and kinematic data to analyze crash, near crash, and critical-incident events. A crash event was recorded if there was any contact between the subject vehicle and another vehicle, fixed object, animal, etc. A near-crash event was recorded if a conflict situation requiring a quick evasive maneuver judgment involvement to avoid a crash was observed. A study by Guo et al. (2010) assesses the use of near crashes as a crash surrogate measure to examine the impact of driver behaviors and other significant risk factors. That study concludes that the use of crash surrogate (near crashes) provides a definite benefit especially given that naturalistic studies are not large enough to produce sufficient numbers of crashes for statistical tests. The Guo et al. (2010) study indicates that a positive relationship exists between the occurrences of the contributing factors in crashes and near crashes. McLaughlin et al. (2009) investigated roadway departure events to identify significant conditions and contributing factors by using data generated from the Virginia Tech Transportation Institute (VTTI) s 100-car Naturalistic Driving Study. Results from McLaughlin et al. (2009) found that roadway departure events occur more frequently in low-visibility and low-friction conditions than in clear and dry conditions. Their results also indicate that approximately half of the events (56%) occurred on straight roadways, (30%) occurred on curves and (14%) at intersections. The most frequently identified contributing factor was driver distraction, due to the comprehensiveness of the VTTI s 9

22 data, the findings introduced the consideration of factors such as changes in roadway boundaries, encroaching vehicles, and late route selection. A study by Jovanis et al. (2011) used the same dataset from (VTTI) 100-Car Naturalistic Driving Study data to study roadway departure events. They estimated the probability that a crash or a near crash would occur. Their study concludes that the data should have included variables that described the attributes of the event, the driver, and the context to reduce the likelihood of bias. They note that understanding how driver decisions are executed, that is, drivers with particular characteristics (one level) find themselves in contexts in which they execute specific driving maneuvers (second level), would lead to certain outcomes (Jovanis et al., 2011). The Second Strategic Highway Research Program (SHRP2) safety data leverage information from the Naturalistic Driving Study (NDS), which provide the opportunity to extra real-time behavioral information from more than 5 million trips (Hamzeie, 2016). The videos in the data are given at a very high resolution in which researchers can observe and extract information about drivers on what happens when a crash occurs, when they experience a near crash, and when they drive without incidents. The SHRP2 study, completed in 2014, is considered the largest existing naturalistic driving study in the United States (Campbell, 2012). The SHRP2 data as well include related information from Roadway Information Database (RID), which is a geospatial database providing detailed information for 25,000 miles of roadway across the NDS study sites (Hamzeie, 2016). The professional data reductionists at Virginia Tech Transportation Institute (VTTI) coded SHRP2 data, including driver performance, behavior metrics, situational characteristics, and environmental characteristics for SHRP2 data for statistical analyses (Hankey et al. 10

23 2016). The central plan of the study was to address the essential role of driver behavior by learning more about driver decisions in traffic conflict, in order to make driving safer by understanding how the driver interacts with the vehicle, traffic environment, and the roadway (Hamzeie, 2016). The current study seeks to understand how drivers adapt with conflict related factors that influence the single vehicle, roadway departure conflicts. The scope of the current study focuses on understanding how one or multiple contributing factors could lead to roadway departure conflicts. Contributing factors in this literature review are group into four categories: 1) Roadway geometry: number of lanes, work zones, alignment (with curves and/or grades) of the roadway, locality, etc. 2) Driver characteristics: age, gender, ethnicity of the driver, etc. 3) Driver and driving behaviors: alcohol, distraction, secondary task, pre-incident maneuvers, speed choices, average miles traveled, etc. 4) Vehicle characteristics: vehicle type, etc. Researchers use various statistical models to examine which of the contributing factors play significant role in causing road departure conflicts and crashes. This chapter on literature review discusses various variables that have been found to be significant factors in causing roadway departure events and severity in past studies. 11

24 2.2 Relationship between Roadway vs. Roadway Departure Risk Factors Average Miles Traveled Massie et al. (1997) studied the effects of four major contributing factors on crash involvement rates. The tested factors include driver age, driver gender, time of day, and average annual mileage of travel. The Massie et al. (1997) study found that all four predictor variables examined were found to be highly significant in contributing to the driver crash involvement rates. In addition, their rates showed that men have a higher risk of crash involvement per mile driven when comparison to women. The National Highway Traffic Safety Administration (NHTSA, 2017) found that the number of vehicle miles traveled on U.S. roads increased by 2.2 percent in 2016, and resulted in a fatality rate of 1.18 deaths per 100 million vehicle miles traveled (VMT) and a 2.6 percent increase from the previous year Number of Lanes Roadways with fewer numbers of lanes are more likely to experience more roadway departure-related crashes as compared with divided or undivided roadways with three or more lanes (Liu. and Subramanian, 2009). Their study results show that roadway departure crash rates that occur on two-lane roadways are higher than those on multi-lane (three and more lanes) roadways, for instance they found that 76.6% of crashes that occurred on twolane undivided roadways were run-off-road crashes while only 40% of crashes that occurred on undivided multi-lane roadways were ROR crashes. 12

25 2.2.3 Alignment It is generally logical that a vehicle will more likely leave its lane on a curve than on a tangent segment due to the centrifugal force that acts on the vehicle when it enters the curve. Roadway alignment plays a significant role in the occurrence of roadway departure crashes (Zegeer et al., 1988). A study by Zegeer et al. (1988) found that terrain, grade, and horizontal curvature were positively associated with the number of crashes, for instance, tighter curves usually experience more crashes. Young drivers are more likely to get involved in ROR crashes when negotiating curves or grades (Dissanayake, 2003). Neuman et al. (2003) provide a summary of previous studies on the effectiveness of flattening horizontal curves and they conclude that by reducing the degree of curve (that is, increasing horizontal radius) the number of roadway departure crashes decreases. Lacy et al. (2004) analyzed the safety effectiveness of horizontal curves by modifying curve features such as increasing the length of radius, addition of spiral transition curves, and elimination of compound curves. Their study conclude that the addition of spiral transition curves, which in essence reduce the curve tightness without changing its radius, reduce run-off-road (roadway departure) crashes. When analyzing single-vehicle run-off-road fatal crashes using the Fatality Analysis Reporting System (FARS) data for a period from 1991 to 2007, Liu and Subramanian (2009) found that 90.2% of the crashes that occurred on curved roads were ROR crashes while for crashes that occurred on straight roadways, only 62.1% were ROR crashes. Lord et al. (2011) analyzed roadway departure crashes on two-lane rural highways in the state of Texas that occurred between 2003 and The number of crashes occurring on 13

26 horizontal curves were nearly two-thirds (261) of all of the ROR crashes. The majority of these crashes were a result of a vehicle attempting to negotiate the curve at a speed that was too high for the curve or too fast based on the weather or roadway conditions present at the time of the crash (Lord et al., 2011) Locality A study by Dissanayake (2003) found that young drivers are more likely to be involved in ROR crashes when driving on rural roadways than in urban roadways. According to Liu and Subramanian (2009), roadway departure crashes are more likely to occur on rural roads as compared to urban roads whereas in their study they found that roadway departure crashes represented 80.6% of all crashes that occurred on rural roadways while 56.2% of the crashes that occurred on urban roadways roadway departure crashes. Kusano and Gabler (2012) analyzed road and driver characteristic from 851 crashes utilized in NCHRP study and found that 70% of road departure collisions occurred in areas classified as rural land use. 2.3 Relationship between Driver and Roadway Departure Risk Factors Characteristics Previous studies have narrated on the existence of the relationship between driver characteristics and the risk factors related to traffic crashes and injuries. This section reviews some of such previous findings. 14

27 Gender Many studies have shown that female drivers have higher injury risk due to a crash impact than males (e.g., Dissanayake and Roy, 2014; Eustace et al., 2016). Carlsson (2010) summarized 13 different previous studies (refer to Figure 1) comparing relative injury risk factors for females and males. According to these studies the injury risk vary from approximately 1.5 to 3 times higher for females when involved under similar crash conditions. Figure 2.1 Female injury risk compared to males (Carlsson, 2010) Vehicles driven by male drivers are more likely to be involved in ROR crashes as compared to the vehicles driven by female drivers (Liu and Subramanian, 2009). 15

28 Age Zhang et al. (2000) used the he Canadian Traffic Accident Information Databank (TRAID) crash data to examine fatal motor vehicle traffic crashes between 1984 and They studied risk factors among young and elderly drivers compared with middle-aged drivers. Their results show significant differences in risk factors where young drivers demonstrated excessive risk taking behaviors and conditions such as alcohol use, speeding, no seat belt use, fatigue or sleepiness, and inexperience. Most crashes occurred during the summertime, on weekends and at night. While for elderly drivers their excessive risks were due to medical and physical conditions, inattention; improper turning, and most crash occurred on weekdays and during the day (Zhang et al., 2000; Liu and Subramanian, 2009). A study by Preusser et al. (1998) found that vehicles driven by young drivers aged 24 years old and younger are more likely to be involved in fatal crashes. Dissanayake (2003) used crash data from Florida to study ROR crashes involving younger drivers between the ages of 16 and 25 years old and reported that several factors were positively associated with ROR crashes including the influence of alcohol or drugs, gender, and speed. Motor vehicle crashes are the leading cause of death for teens (Seacrist, 2016). Distracted driving amongst teens is a much greater problem than previously thought (NHTSA, 2016). Therefore, injury risk is higher among men in some age groups, and among women in other groups, but these age groups vary depending on mode of transport and severity (Santamariña-Rubio, 2014). 16

29 2.3.2 Behavioral Risk Factors Driving behavior has been identified as a life-threatening factor in nearly all traffic crashes. Driver inattention, in its various forms, is probably the most prevalent cause of traffic crashes. (Wang, et al. 1996). Some of the driver behavioral risk factors are discussed in this section Distraction Drivers with risk factors such as sleepiness, inattentiveness, over- correction of the vehicle, or crash avoiding issues are more likely to be involved in ROR crashes as compared to other driver contributing factors (Liu. and Subramanian, 2009). Drivers get distracted during any activity that diverts their attention from the task of safe driving, including talking or texting over the phone, talking to other people in the vehicle, adjusting stereo, or navigation system, etc. In 1995, NHTSA initiated a Crashworthiness Data System (CDS) study that provided in-depth information in the role of driver inattention into three principle forms of crash causing factors, which are: distraction (11.7%), looked but did not see (8.9%), and sleepy/fell asleep (3.1%) (Wang et al., 1996). Distraction and inattention contribute to 80% of traffic crashes by mainly delaying or hindering driver s ability to respond to stimulus (Dozza, 2013). However, Dozza (2013) states that distraction and inattention are not the only factors increasing response times. Driver impairment primarily due to alcohol, distraction, drowsiness, are responsible for significant increases in crash risk. NHTSA (2016) reports that in 2015 the number of people killed was 3,477 and 391,000 were injured involving distracted drivers in all traffic crashes in the United States. In addition, NHTSA found that distracted driving and drowsy driving 17

30 fatalities weakened, but other fatal contributing factor related to other irresponsible behaviors; including speeding, alcohol limit, and not wearing seat belts continued to grow. In 2016, NHTSA (2017) report of distraction-related deaths of 3,450, a slight decrease from the previous year and drowsy-driving accounted for 803 fatalities, a decrease of 3.5% from the previous year. In addition, 10,497 fatalities were due drunk driving, an increase of 1.7% from the previous year; speeding-related deaths were 10,111, an increase of 4.0% from the previous year; unbelted deaths accounted for 10,428 fatalities, an increase of 4.6% from the previous year Alcohol Drivers who drive while having alcohol contents in their blood above the legal limit are more likely to be involved in fatal roadway departure crashes as compared to sober drivers (Liu. and Subramanian, 2009). Driving under the influence of alcohol is more prominent at nighttime than during the daytime. A study by Massie and Campbell. (1993) using the 1990 Nationwide Personal Transportation Survey data report that although 2.0 % of the daytime traffic crashes involved drivers who were driving under the influence of alcohol, however, about 22.8% of the traffic crashes that occurred during nighttime involved drivers who were under the influence of alcohol. In addition, for all fatal nighttime roadway departure crashes for which drivers were considered to be under the influence of alcohol, 37.5% were females, 49.4% were males, and just about 13.1% were non-alcohol related crashes (Massie and Campbell, 1993). 18

31 Speeding The Minnesota Department of Transportation (MnDOT) reviewed 338 fatal crashes that occurred on two-lane two-way highways due to run-off-the-road (ROR) crash type from and these crashes resulted in 365 fatalities (Leuer, 2015). In addition, 197 of these crashes were the result of drivers drifting off the roadway and 141 crashes were the result of the driver losing control of the vehicle. Behaviors such as inattention and distraction are both contributing significantly to ROR crashes. A study by Knipling and Wang (1994) that utilized single-vehicle roadway crashes from the General Estimates System (GES) and the Fatality Analysis Reporting System (FARS) examined driver-related actions that significantly contribute to run-off-road crashes. They mention that run-off-road crashes are primarily caused by the following six factors: excessive speeding (32.0%), driver lost control (20.1%), drowsiness (16.0%), evasive maneuvers (15.7%), i.e., driver steers off road to avoid obstacle, driver inattention (12.7%), and vehicle failure (3.6%) Pre-Incident Maneuver A study by McLaughlin et al. (2009) examined roadway departure maneuvers by capturing the driver s driving actions such as steering, braking,, etc., that lead in ROR events and driver s driving actions when attempting to regain control. Their results show that drivers applied brakes in an attempt to avoid ROR crashes in approximately half of the events (McLaughlin et al., 2009). A study by (Kostyniuk et al., (2002) states that nearly 80% of the crashes resulted in the vehicle departing the roadway on the right side of the road. 19

32 Where almost half (47.6%) of the crashes involved a vehicle that crossed the centerline and the rest of ROR crashes (52.4%) resulted in the vehicle leaving the road on the left side Secondary Task A secondary task occurs when the operator performs two tasks: the primary task, the operational task of interest (steering, for a car), and a secondary task which is imposed to occupy the part of the operator's "capacity" not required by the primary task (such as adjusting the car radio). There are specific environmental conditions in which engaging in secondary tasks is more dangerous than other tasks. Secondary tasks and eyes-off-road are major indicators of driver's distraction and inattention; in fact, they significantly delay driver response times by 29% and 16%, respectively (McLaughlin, et al., 2009). Response times also depend on the incident and evasive maneuver types. 2.4 Relationship between Vehicle and Roadway Departure Risk Factors Liu and Subramanian (2009) report that among all passenger vehicles (passenger cars, vans, pickup trucks, and utility vehicles), passenger cars are more likely to be involved in fatal roadway departure crashes as compared to other vehicle types; and compared to passenger cars, trucks usually have a high center of gravity. Hence, trucks have a greater risk of rolling over in the event of an ROR crash. McGinnis et al. (2001) used FARS database to analyze traffic crashes and noted that the number of light trucks involved in a rollover crash increased by 130% between 1975 and 1997; this increase was also attributed to the growing use of light trucks over the years. In addition, Farmer and Lund (2002) examined FARS crash data for the period and found that light trucks (pickups, 20

33 vans, and SUVs) were twice as likely as cars to roll over, following a roadway departure crash. 21

34 CHAPTER 3 DATA AND METHODOLOGY 3.1 Data Description Second Strategic Highway Research Program (SHRP2) The Second Strategic Highway Research Program (SHRP2) was initiated to find strategic solutions in improving highway safety, reducing congestion, and improving national roads, etc. Extensive data collection effort was performed to address unique fundamental research questions that were not possible to address before. The SHRP 2 safety study collected an extraordinary amount of data, moreover on that it provides comprehensive assessment on understanding driver behavior in traffic safety. In specific, how driver behavior performance interact with roadway, environment, vehicular, and human factors. Data were collected from six different states around the United States. The largest collection sites were in Bloomington, Indiana; Durham, New Carolina; State College, Pennsylvania; Seattle, Washington; Tampa, Florida; and Buffalo, New York. This chapter includes details of the data sources used to conduct this study, how these data sources were collected for the purpose of this study and details of the statistical methods used to analyze the data. 22

35 Naturalistic Driving Study The Naturalistic Driving Study (NDS), which was done as part of SHRP2, NDS recruited nearly 2,360 participants. The volunteers were men and women in various age groups, from different socioeconomic backgrounds, driving different types of light vehicles. Participants were distributed in sites hosting 150 to 450 participant vehicles, and were evaluated by their, driving knowledge, reaction time, psychological traits, visual perception, medical history, etc. As a compensation for the participation, each driver was paid $300 per year. The NDS-SHRP2 data were collected using onboard data acquisition systems (DAS) whenever the vehicles were in use. To capture data of their usual driving tasks in their real world driving behavior, video images of the view out the front and rear windshields, the passenger side view, the driver s face and hands, and the cabin were recorded as shown in Figure 3.1. Figure 3.1 DAS four continuous video view from the vehicle interior 23

36 That included a record of vehicle kinematics such as speed, acceleration, lateral and vertical motion. In addition, inside the vehicle there was also a sensor that could detect the presence of alcohol in the vehicle, use of seatbelt, etc. Various sensors were also recorded through radar to identify objects in the front of the vehicle, including rates at which the range changes. An incident push button allowed participants to record critical events and emergencies. Figure 3.2 DAS schematic view Roadway Information Database Roadway information is necessary to relate driver actions to the roadway characteristics. In conjunction with the NDS data, the Roadway Information Database (RID) was developed as part of SHRP2 project. Roadway Information Database (RID) is a geospatial database that provides detailed data for 24,000 miles of roadway across the six study states. The roadway database identifies number of lanes, lane width, grade, alignment, superelevation, speed limit and their locations, the location of intersections, etc. 24

37 3.1.2 Data Collection Before requesting the data, a detailed assessment identifying the desired driver, roadway, and vehicle characteristics is necessary to answer the fundamental research questions previously addressed. The data needed for the purpose of the current study were requested from the Insight query builder available portion of the data collected as part of SHRP2 study. Conflicts that occurred between 2010 and 2012 were extracted from SHRP2 database to identify the important factors that affect roadway departure event severity. This subset data includes aggregated summary data of the NDS that excludes all personal identifying information, but includes information as to event details, trip details and vehicle characteristics. Figure 3.3 SHRP2 Insight Query Builder Driver-specific information was also added to the dataset, which includes demographic characteristics for each study participant. In addition, information about behaviors and 25

38 perceptions that was captured through several participant surveys and questionnaires was included. Additional data included surveys about driving knowledge and questionnaires regarding driver behavior, driver demographic characteristics, driving history, risk perception, and risk-taking. Factors related to roadway departures were used to guide the data requests Data Request Scope The Data requested specified for all crash and near-crash events identified over the duration of the SHRP2 Naturalistic Driving Study (NDS). The datasets consisted of the following InSight tables: Driver Demographic Questionnaire: Driver gender, ethnicity, education, income, and average age group, etc. Driving History Questionnaire: The estimated participant s average annual mileage over the past five years. The number of years the participant has been driving. Severity of the first crash, whether the participant was at fault or not at fault for the first crash, etc. Event Table: Driver seatbelt use, hands on the wheel, alignment, grade, locality, presence of construction zone, fault and intersection influence, pre-incident maneuver, maneuver judgment, precipitating event, event nature, incident type, secondary task and secondary task outcome, event severity, crash severity, etc. Trip Summary Table: 26

39 Maximum vehicle speed recorded during the trip, and mean speed of the vehicle while moving calculated over the duration of the entire trip, etc. Vehicle Detail Table: Vehicle classification, etc Merging Files A single-to-many merging technique in IBM SPSS Statistics (Version 24.0) software was used to merge the Insight tables requested for the scope of the current study. The selected Insight tables were then merged into one file, using their Participant ID, Vehicle ID, Event ID, and Trip ID. Participant ID is a numeric identifier assigned to a participant randomly when the participants were recruited in the study and it was used in place of any other driver identity information. The Vehicle ID is the system identification number used to refer to a specific vehicle. Event ID is an identification number randomly assigned to this event, which can be used to link this event to other data from the website. Trip ID is a unique numerical identifier for the trip. The number of records in this dataset represents 3,352 vehicles, 3,546 drivers, 5,414,063 trips and 36,832 events Creating of Roadway Departure-related Traffic Event Database To create a roadway departure event database a clear understanding of the naturalistic driving video analysis and a thorough understanding of all variables within the categories is essential. It is specifically important to have a clear concept of the variable precipitating event, because the proper coding of many other variables depends upon an accurate identification of the precipitating event. The precipitating event is the state of environment 27

40 or action that began the event sequence under analysis. This is the critical event, which made the crash, or near crash possible, based on what the action of the vehicle does, not the driver behavior action. After identifying the precipitating event, a clear understanding of the event nature is another essential variable; it identifies the other object(s) of conflict (e.g., conflict with lead vehicle, or single vehicle conflict) for the crash or near-crash event. To manifest the goal of this study, a single-vehicle roadway departure conflict is then selected through the incident type variable. Targeting roadway departure conflicts for the most severe type of crash, near crash, or safety-related incident that occurred. Incident type variable identifies the type of conflict occurred based on the precipitating event and the event nature. Example: subject vehicle avoids rear-ending a lead vehicle (near crash) by steering off the road into a ditch (a crash). 1 = rear-end, striking (the near crash); 2 = runoff-road (the crash). Figures 3.4 shows the Research Dictionary for Video Reduction Data (RDVRD), referenced to when coding the incident type variable. 28

41 Figure 3.4 Researcher Dictionary for Video Reduction Data: Event Nature, Incident Type, and Vehicle Configuration Figure 3.5 describes the type of incidents that occurred in SHRP2 NDS three-year data collection where 26% of the incidents are roadway departure conflicts. Roadway departures are grouped in two categories: left or right roadway departure and end roadway 29

42 departure. A left or right roadway departure occurs when any tire on the subject vehicle leaves the roadway, beyond the shoulder or onto median, on the left or right side of the roadway, including interaction with roadside barriers and curbs. This category resulted in 1007 conflicts that are 23.7% of total incidents. The second category, end roadway departures occur when any tire on the subject vehicle leaves the end of the roadway and these types of departures resulted in 2.3% of total incidents. Figure 3.5 Description of incident types Description of Selected Variables for Study After selecting road departure cases from the merged insight tables requested for the study, variables were selected that involve any driver behavioral and characteristic, vehicular characteristics and roadway characteristic, that may be responsible for causing a roadway departure conflict. To be able to examine these risk factors, it is essential to describe the 30

43 outcome of the event/incident, which is the dependent variable, in this in case, it is either event severity or crash severity. Event severity is a general term that denotes the outcome of each event/incident type as a crash, near crash, or non-subject conflict. Crash severity is a ranking of crash severity for the referenced event based on the magnitude of vehicle dynamics, the presumed amount of property damage, and the level of risk posed to the drivers and other road users. The characteristics of roadway departure conflicts that occurred on SHRP2 study roads and highways are summarized in Tables 3.1 through 3.4 for vehicle classification, driver demographic characteristics, driver behavior characteristics risk factors, and roadway/environmental characteristics, respectively. Table 3.1 also describes the events severity and crash severity. Table 3.1 Description of crash and event severity, and vehicle classification. Variable Description Frequency Percent Vehicle Classification Car Pickup Truck SUV Minivan Average Annual Mileage Up to 10,000 miles ,000-15,000 miles Above 15,000 miles Event Severity Crash Near-Crash Non-Subject Conflict Crash Severity Most Severe Police-reportable crash Minor crash Low-risk Tire Strick Not a crash

44 Table 3.2 summarizes driver demographic characteristics that include gender, ethnicity, education level, income and their average age groups. Table 3.2 Description of driver demographics. Variable Description Frequency Percent Gender Female Male (null) ethnicity Hispanic or Latino Not Hispanic or Latino (null) Education Level High school Some Education beyond HS College degree Advance degree Driver Income less than $50, $50,000-$100, $100,000 & Above Average Age Group & Above Table 3.3 summarizes driver behavior risk factors, that include where the driver had his hands on wheel or not, whether the driver used the seatbelt during an incident or not, and whether the driver was distracted during a secondary task or not. The data also recorded the pre-incident maneuver, vehicle evasive maneuver and the driver judgment to the maneuver. 32