Development of Critical Knowledge Gaps and Research Efforts

in Support of the Safety R & T Partnership Agenda

Prepared By:

Barry H. Kantowitz, Ph.D. University of Michigan Transportation Research Institute

Jeremiah P. Singer

Neil D. Lerner, Ph.D. Westat, Inc.

Hugh W. McGee

Warren E. Hughes BMI-SG

William A. Perez, Ph.D.

Cambridge Systematics, Inc.

Gerald L. Ullman, Ph.D. Texas Transportation Institute

Prepared for:

US Federal Highway Administration Washington, DC

Contract DTFH61-01-C-00049 Task #10

2

FHWA Document Abstract 1. Report No. 2. Government Accession No.

3. Recipient's Catalog No.

4. Title and Subtitle

Development of Critical Knowledge Gaps and Research Efforts in Support of the

Safety R & T Partnership Agenda

5. Report Date

May 2004

6. Performing Organization Code

7. Authors

Barry H. Kantowitz , Warren E. Hughes, Neil D. Lerner, Hugh W. McGee,

William A. Perez, Jeremiah P. Singer, and Gerald L. Ullman

8. Performing Organization Report No.

9. Performing Organization Name and Address

The University of Michigan Transportation Research Institute 2901 Baxter Road Ann Arbor, Michigan 48109-2150

10. Work Unit No.

11. Contract or Grant No.

Contract # DTFH61-01-C-00049 Task 10

12. Sponsoring Agency Name and Address

Federal Highway Administration

Turner-Fairbank Highway Research Center

6300 Georgetown Pike

McLean, VA 22101

13. Type of Report and Period Covered

14. Sponsoring Agency Code

15. Supplementary Notes Contrac ing Officer’s Technical Representative (COTR) – Michael Griffith t16. Abstract This report presents a set of five independent white papers on topics selected by the National Highway R&T Partnership, plus an introductory chapter that gives background information, and a final chapter that compares results across the white papers. All papers fall within the general topic of Highway Infrastructure and Operations, one of eight safety research themes in the National Highway Safety Research Agenda. Sets of potential research projects were produced for each of the following topics: run-off-road accidents, intersection safety, human factors, work zone safety, and advanced projects that cut across the preceding topics. Each potential project was rated for likelihood of success and estimates were given for cost and duration. The most important element of this report is the formal distinction between Applied and Advanced projects. Applied projects represent the standard highway safety topics historically supported by the Federal Highway Administration (FHWA). Advanced projects represent high-risk high-reward topics that have not been supported in the past; these projects include basic research that will provide a firm foundation for future applied research efforts. Examples of recommended Advanced projects include Development of a Computational Driver Model, Development and Application of a Roadside Inventory Database, and Safety Impacts of Alternative Intersection Controls. Examples of recommended Applied projects include Optimizing the Net Benefits of Delineation, Safety Effects of Alternative Left Turn Phasing, and Driving Simulator Validity. 17. Key Words

Highway Infrastructure and Operations, National Highway

Safety Research Agenda, run-off-road accidents, intersection

safety, human factors, work zone safety, and advanced projects

18. Distribution Statement

19. Security Classif. (of this report)

Unclassified

20. Security Classif. (of this page)

Unclassified

21. No. of Pages

22. Price

Table of Contents

1.0 EXECUTIVE SUMMARY .......................................................................................... 6

1.1 Development of the White Papers ........................................................................................................................6

1.2 Results.....................................................................................................................................................................7

1.3 Appendix A: Statement of Work........................................................................................................................13

2.0 RUN-OFF-ROAD RESEARCH NEEDS .................................................................. 19

2.1 Introduction .........................................................................................................................................................19

2.2 Run-Off-Road Research Topics .........................................................................................................................21

2.3 Knowledge Strongholds and Gaps .....................................................................................................................21

2.4 Critical Future Highway Issues ..........................................................................................................................24

2.5 Research Recommendations ...............................................................................................................................27

2.6 Summary ..............................................................................................................................................................43

2.7 Appendix A: Compiled Lists of Run-Off-Road Research Needs.....................................................................45

3.0 INTERSECTION SAFETY ...................................................................................... 53

3.1 Introduction .........................................................................................................................................................53

3.2 Potential Research Needs ....................................................................................................................................55

3.3 Summary ..............................................................................................................................................................72

3.4 Appendix A...........................................................................................................................................................73

3.5 Appendix B...........................................................................................................................................................76

3.6 Appendix C...........................................................................................................................................................77

4.0 HUMAN FACTORS................................................................................................. 78

4.1 Introduction .........................................................................................................................................................78

4.2 Critical Future Highway Issues ..........................................................................................................................78

4.3 Human Factors Research Topics........................................................................................................................79

4.4 Knowledge Strongholds.......................................................................................................................................80

4

4.5 Knowledge Gaps ..................................................................................................................................................82

4.6 Research Topics ...................................................................................................................................................86

4.7 Relationship Between These Proposal Projects and the F-SHRP Research Plan ........................................102

4.8 Summary ............................................................................................................................................................103

5.0 WORK ZONES...................................................................................................... 104

5.1 Introduction .......................................................................................................................................................104

5.2 Specific Research Topics...................................................................................................................................106

5.3 Knowledge Strongholds.....................................................................................................................................107

5.4 Knowledge Gaps ................................................................................................................................................107

5.5 Specific Research Projects ................................................................................................................................110

5.6 Summary ............................................................................................................................................................125

6.0 FUNDAMENTAL ADVANCED RESEARCH ........................................................ 126

6.1 Introduction .......................................................................................................................................................126

6.2 Specific Research Topics...................................................................................................................................129

6.3 Knowledge Strongholds.....................................................................................................................................132

6.4 Knowledge Gaps ................................................................................................................................................132

6.5 Research Recommendations .............................................................................................................................133

6.6 Summary ............................................................................................................................................................143

7.0 SYNTHESIS.......................................................................................................... 145

7.1 Advanced Projects .............................................................................................................................................146

7.2 Applied Projects.................................................................................................................................................149

8.0 REFERENCES...................................................................................................... 152

5

1.0 Executive Summary This report presents a set of five independent white papers on topics selected by the National Highway Research and Technology (R&T) Partnership, plus an additional chapter that attempts to integrate the set of research recommendations proposed in the white papers. The full statement of work (SOW) for this project is appended to this chapter. It is important to note that this is a small-scale project with limited hours for the white-paper authors. Indeed, the original SOW specified a maximum limit of fifteen pages for each paper; this limit was deleted to allow the authors to respond to useful feedback provided by reviewers from the R&T Partnership Steering Committee. Responses to this feedback are contained in each white paper. In most cases, since the reviewers did not receive a copy of the SOW, their helpful suggestions exceeded the scope of this project. Nevertheless, each author made a serious effort to address most of the issues raised despite an extremely limited number of hours remaining in the project for revision.

1.1 Development of the White Papers The charge for these papers originated in a Safety Research Agenda Planning Conference held by the R&T Partnership in Irvine, California Sept 17 and 18, 2002. The theme of the conference was Highway Infrastructure and Operations, one of eight safety research themes in the National Highway Safety Research Agenda. The conference brought together stakeholders, from state highway agency researchers and traffic engineers, university researchers and research administrators, private sector researchers and representatives from the Federal Highway Administration, to identify highway safety research needs. This was the initial process to identify and prioritize research needs and knowledge gaps in the following areas:

• Run-off-road accidents • Intersection safety • Intelligent infrastructure initiative • Human factors • Work zone safety.

Working groups in each of these research areas produced a tentative list of critical research needs. The goal was to produce a set of candidate projects that over the next five years could gain valuable knowledge that would improve highway safety. Since the focus of the conference was making progress in these key areas, the rationale for the selection of the areas was not debated during the conference. The topics were derived from earlier work of the R&T Partnership which has been articulated in an April 2002 report titled Highway Research and Technology: The need for greater investment. This report identified (Table 2) the following R&T themes and emphasis areas for Highway Infrastructure and Operations:

6

• Human factor safety guidelines • Consequences of leaving the road • Intersection safety • Intelligent infrastructure initiative • Work zones • Inclusion of safety in the highway design process.

Table 1 of the report estimated an annual cost of $30 million over a 5 year time period for the Highway Infrastructure and Operations R&T theme. The goal of this report is to expand on the projects suggested at the Irvine Planning Conference by giving greater detail and adding any important projects that might have been omitted by the working groups. As discussed in the SOW (Appendix A) the original plan called for a white paper on each of the five topics listed above plus a sixth chapter on Advanced projects that cut across the other areas. However, the paper on intelligent infrastructures, originally planned to be written outside this contract by a Federal Highway Administration staff member, was cancelled due to competing priorities at Turner-Fairbanks Highway Research Center. In order to minimize contracting delays, this report was assigned as Task 10 of the Federal Highway Administration IDIQ support contract, “Technical Support and Assistance for the Federal Highway Administration’s (FHWA’s) Human-Centered Systems Team” DTFH61-01-C-00049. The prime contractor is the University of Michigan Transportation Research Institute (UMTRI) supported by an all-star team of public and private sub-contractors including Battelle Human Factors Transportation Center, Bellomo-McGee, Inc., Cambridge Systematics, Center for Applied Research (CAR), Georgia Institute of Technology, University of Iowa, Texas Transportation Institute, TransAnalytics, LLC., Transportation Research Corporation, Virginia Polytechnical Institute & State University Center for Transportation Research, Westat, Inc., and William H. Levison Associates . In order to avoid contractual delays that would be required to add new sub-contractors to the team, white papers were assigned to these sub-contractors contingent upon staff availability. Table 1.1.1 shows the staffing for each white paper. Table 1.1.1 Assignment of IDIQ team members to white paper topics.

Run-Off-Road Intersection Safety

Human Factors Work Zones Fundamental Advanced Research

Westat BMI-SG UMTRI TTI Cambridge Systematics

1.2 Results Each white paper author was instructed to address statistical evidence from national data bases, knowledge strongholds and gaps, critical future highway issues (a new topic added in the revision process), and to generate a list of research topics and projects. Each topic was classified

7

as Applied or Advanced research. Likelihood of success, defined as completing the project on time and within budget and either generating useful countermeasures (Applied projects) or building a firm foundation for useful countermeasures (Advanced projects), was rated on a five-point scale. Project costs and durations were also estimated. Summary tables for each white paper are presented in the following sections. Run-off-road

Category Project Title Type of Research

Likelihood of success (1-5 scale)

Duration (months)

Cost (in millions)

Run-Off-Road

ROR 1: Use of Rumble Strips on Non-Freeways

Applied 5 36 1

Run-Off-Road

ROR 2: Development of a System of TCDs to Reduce ROR Crashes at Curves

Applied 4 18 1

Run-Off-Road

ROR 3: Optimizing the Net Benefits of Delineation

Applied 4 36 1

Run-Off-Road

ROR 4: Development and Application of a Roadside Inventory Database

Advanced 3 24+ 1.5

Table 1.2.1

8

Intersection Safety

Category Project Title Type of Research

Likelihood of success (1-5 scale)

Duration (months)

Cost (millions)

IS 1a: Magnitude, Characteristic, & Causation of Intersection Accidents

Advanced Moderate to High,

4

36 $1.5 - 2M Accident Causation

IS 1b: Establish Root Causes of Driver Error

Advanced Moderate 3

24 $0.5 – 0.75 M

Relationship of Safety to:

IS 2a-1: Safety Impacts of Alternative Intersection Controls

Advanced Moderate 3

36 – 60 $1.0 – 1.5 M

a. Traffic & Operational Features

IS 2a-2: Safety Effects of Alternative Left Turn Phasing

Applied Moderate 3

24 $0.3 M

b. Traffic Control Devices

IS 2b: Safety effects of alternative signal layouts

Applied Moderate to High, 3

36 $0.3 M

c. Design Features

IS 2c: Intersection Sight Distance

Advanced Low to Moderate, 2

24 – 36 $0.5 M

Effectiveness of Counter-measures

IS 3: Effectiveness of various countermeasures for reducing accidents

Applied Moderate to High, 4

84 for all phases

$2.0 M

Advanced Technology

IS 4: Effectiveness of & Driver Response to Automatic All-red Signal Extension System

Applied High, 5 6 $0.05 - $0.1 M

Table 1.2.2

9

Human Factors Category Project Title Type of

Research Likelihood of success

Duration (Months)

Cost ($1,000,000)

Human Factors Cognitive Models

HF 1a: ComputationalDriver Model: WE (Whole Enchilada)

Advanced 3.5 144 12

HF 1b: Computational Driver Model: Light

Advanced 4.0 60 5

Human Factors Information Overload

HF 2: Processing Multiple Sources Of Information

Advanced 4.0 60 10

Human Factors Speed Control

HF 3: Understanding Speed Selection

Applied 3.0 48 8

Human Factors Perception/Attention

HF 4: Look but not see

Applied 2.5 36 4

Human Factors Basis for Design Standards

HF 5: Design Driver

Applied 4.5 18 0.3

Human Factors Decision Rationality

HF 6: Risk Homeostasis

Applied 2.5 28 1

Human Factors Simulator Generalization

HF 7: Driving Simulator Validity

Applied Methodology

4.5 42 4.5

Table 1.2.3 Prioritized list of research projects. Likelihood of success ranges from Very Low (1) to Very High (5).

10

Work Zones

Category Project Title Type of Research

Likelihood of success (1-5 scale)

Duration (months)

Cost (Millions)

WZ 1a: Estimate WZ Exposure Characteristics from FMIS

Applied High 4

30 $1M Research Methodology –WZ Exposure Data WZ 1b: Develop VMT

Temporal Distributions to Estimate WZ Exposure

Applied Very High 5

18 $0.5M

WZ 2a: Incorporate New WZ Data Elements into CDS Crash Investigations

Advanced Moderately Low 2

60 $2M Research Methodology – WZ Crash Data

WZ 2b: Investigate Likelihood of Work Zone Crash Reporting

Applied Very High 5

18 $0.5M

WZ 3a: Feasibility and Validity of Region-wide WZ Crash Risk Estimation Techniques

Advanced Moderate 3

30 $1M Determine WZ Crash Causation

WZ 3b: Project-Level Crash Consequences of Work Zone Design Features

Applied Moderate 3

60 $2.5M

WZ 4a: Improving the Understanding and Measurement of Driver Behavior in High Driver Workload Environments

Advanced Moderate 3

36 $1.5M Identify/Evaluate Countermeasures to Mitigate WZ Crash Risk

WZ 4b: Evaluate Dynamic Queue End Warning Systems for WZ

Applied Moderately 3

60 $1.5M

Develop/Apply/ Evaluate WZ Management Procedures

WZ 5a: Analyze State WZ Monitoring and Management Programs and Procedures

Applied High 4

48 $1M

WZ = Work Zone GES = General Estimates System FMIS = Financial Management Information System VMT = Vehicle-Miles-Traveled CDS = Crashworthiness Data System Table 1.2.4

11

Fundamental Research

Category Project Title Type of

Research

Likelihood of Success

(1-5 Scale) Duration (Months)

EstimatedCost

1. Understanding the Driver

ADV 1a: Development of a Driver Modeling Structure

Advanced Very High 5

24 $2M

ADV 1b: Development of a Prototype Driver Model

Advanced High 4

36 4M

ADV 1c: Development of a Driver Model

Advanced Moderate 3

60 15M

2. Data Collection/ Analytical Tools

ADV 2a: Evaluation of Advanced Sensors and Data Mining Techniques

Advanced Moderate 3

36 4M

ADV 2b: Development of Safety Decision Aids for Planners

Advanced Moderate 3

36 5M

3. Advanced Technology for Countermeasures

ADV 3: Evaluation of Nanotechnology for Safety Countermeasures

Advanced High 5

12 250K1

Table 1.2.5

3 Note: The proposed $250K is for a project to develop a nanotechnology research program at FHWA.

12

1.3 Appendix A: Statement of Work

Task Order Statement of Work for IDIQ (DTFH61-01-R-00049): Submitted by The University of Michigan

Development of Critical Knowledge Gaps and Research Efforts in Support of the Safety R&T Partnership Agenda

Background: In April 2002, the National Highway Research &Technology Partnership published its report, Highway Research and Technology: The Need for Greater Investment, with the purpose of establishing a document of priority research, development, and technology program areas for enhancing the opportunity for all parties involved in highways to coordinate and collaborate on conducting these priority areas. The report indicated that the working group to the highway safety portion of the Partnership proposed to refine and conduct a pilot implementation of coordination and collaboration in the Highway Infrastructure & Operations safety theme. This theme was one of eight major safety theme areas in the Agenda. [The other themes are Safety Management and Data Systems, Driver Competency, High Risk Driving, Light Duty Vehicle Safety, Vulnerable Road Users, Truck/Bus Safety, and Post-Crash Management.] A Safety Research Agenda Planning Conference was help on September 17 & 18, 2002, in Irvine CA, to address how to proceed with the objective for coordination and collaboration in the Highway Infrastructure & Operations safety theme. The two fundamental goals of the Conference were to:

• Generate a high-level agenda process for the systematic identification of highway safety research based on stakeholder needs, the development/conduct of high-quality coordinated research, and implementation of results.

• Initiate the process of identifying and prioritizing critical highway safety research needs

and information gaps in the following areas: (1) Run-Off-Road, (2) Intersection Safety, (3) Human Factors, (4) Work Zones, and (5) Fundamental Advanced Research.

The safety conference attendees were comprised of state highway agency researchers and traffic engineers, university researchers and research administrators, private sector researchers, and representatives from the Federal Highway Administration. The three products emerging from the conference were: (1) a preliminary safety research agenda process, (2) recommendations pertaining to the quality of research, and (3) the identification of critical infrastructure safety research needs. It is this last product that additional assistance and input is sought via this Task Order.

Objectives: The objectives of this effort are:

• Develop writing guidelines for white paper authors,

13

• Produce five preliminary white papers identifying critical research needs in the areas of Run-Off-Road, Intersection Safety, Human Factors, Work Zones, and Fundamental Advanced Research, (A sixth white paper on Intelligent Infrastructure Initiatives will be written outside this task order).

• Produce an Integration Paper which identifies the cross-cutting aspects of the six white papers, and

• Develop a quantitative approach to evaluating the ultimate list of safety research topics which will be generated using the seven papers mentioned above as a starting point.

Tasking: Task 1 – Guidelines for White Paper Authors In order to maximize the uniformity in approach which will allow comparison between the white papers and lead to optimizing the integration of potential research topics from the six white papers, UMTRI shall develop guidelines for the white paper authors. The guidelines shall suggest a common structure for the white papers. Suggested guidelines might be:

• Length of paper – no longer than 12 to 15 pages • Breakout of applied or advanced research • Section suggestions

o magnitude of the highway safety problem (number of fatalities and injuries) o estimate of the application of knowledge o likelihood that the research area project would produce successful or intended

results All white papers will include a final list of specific research topics to be rated in Task 4. Suggested guidelines shall be submitted to the government within 4 weeks of the initiation of the contract. The government shall provide comments on the guidelines to UMTRI in 2 weeks. Within 6 weeks of the initiation of the contract, the guidelines will be finalized and sent to the white paper authors. Task 1 will be performed by UMTRI. Task 2 - White Papers The six topic areas are a mix of highway safety problem areas (run off road, intersection, and work zones), as well as a solution-based type area (intelligent infrastructure initiatives) and of a cross cutting nature (human factors and advanced technology). Accordingly, there will be a level of overlap in the white papers. The authors should address the topics individually, as well as to incorporate appropriate considerations of the intelligent infrastructure initiatives and human factor research projects into the other four problem areas. Task 2A – Run-Off-Road Run-off-road (ROR) research needs to address two major questions: why drivers run off the road and what happens when they leave the road. Specific knowledge is needed in two areas: (1) causes of roadway departures (i.e., what happens before the vehicle crosses the edge line); and

14

(2) roadside countermeasure effectiveness (i.e., what can be done to reduce crash consequences after the vehicle crosses the edge line). Of particular interest is how highway infrastructure technology can mitigate R-O-R crashes. The highest-ranked research topics rated by the ROR breakout group were:

• Use of one-pass van to inventory the roadside features –digitize roads for vehicle /roadway interactions

• Methods for choosing ROR sites, corridors, treatment programs • Rumble strips on narrow paved shoulders • Intelligent Infrastructure warning systems – field test of driver response • Relationship of simulation and crash test results to real-world crash injuries • Safer ditch design for rural two-lane roads • Tree removal tradeoff research • Vehicle/roadside interaction (Roll over, tripping effects, etc. by vehicle type)

Task 2A will be performed by Westat. Task 2B – Intersection Safety Intersection safety research was viewed as a high-priority/high-payoff area by the R&T Partnership members due to the fact that crashes are highly concentrated at intersections. The breakout groups considered a systematic review of crash data to decide what research topics deserve priority. Fundamental intersection safety issues were identified as: (1) effectiveness of stop-sign vs. signalized traffic control, (2) effectiveness of actuated vs. semi-actuated vs. fixed time signalization, (3) relationship between signal-timing decisions and safety, and (4) safety effects associated with traffic flow characteristics

Specifically identified research topics are as follows:

• Safety effects of cross-sectional elements at intersections • Safety performance of roundabouts • Analytical tools/models for traffic engineers & planners to consider the safety

consequences of intersection safety and design • Safety effects of transitional elements moving from corridors to intersection approaches • How to accommodate various users (pedestrians, bicycles, trucks, etc.) for different

scenarios • Safety effects of traffic-calming devices/perceptual measures • How does the culture of road user behavior evolve and how can it be influenced • Relationship of ISD safety • Analytical tools to identify which intersections to provide selective enforcement

Task 2B will be performed by BMI. Task 2C – Human Factors Identified research gaps were guided by the need to determine what is in the driver’s head, emphasizing driver observational and cognitive processes that affect safety-related actions. Certain modeling applications, i.e., development of a Computational Cognitive Driver Model

15

and extended work on the IHSDM, received the highest ratings for their potential to produce important results; however, the estimated costs for these applications were high. Continued research on the Design Driver emerged as the most promising study area, i.e., likely to produce important and timely results with a high probability of success while requiring minimal resources of time and funding. Specifically identified research topics are as follows.

• Understanding the driver via application of a computational cognitive model • Determining sources of driver overload from multiple sources of information • Understanding speed selection, i.e., contributing factors to driver loss of control • Causes of look-but-did-not-see crashes, i.e., why some drivers fail to notice critical

elements in the visual field • Development of a design driver as a basis for human factors standards, to assist in

understanding the effects changing demographics and related factors • Assessing the validity of homeostasis theory, an approach to risk acceptance and

compensatory behavior • Continue MDAI data collection, via updating the Indiana Tri-level study

Task 2C will be performed by UMTRI. Task 2D – Work Zones Existing work zone crash data, describing a wide variety of crash types, is highly diverse and lacks consistency. While fragmented studies have been undertaken in many states, there is no uniform database that is suitable to adequately establish causation. Additional data elements e.g., (exposure measures) are required for statistically valid cause and effect studies. Critically lacking is a consistent research methodology supporting a synthesis of the varied databases. Once such a methodology is developed, it will then be necessary to conduct controlled studies in order to develop a conclusion regarding causation. Such an effort would entail considerable time and funding. In the interim, improved work zone management procedures are necessary to maximize the safety benefit of existing traffic control techniques. Specifically identified research topics are as follows:

• Develop a research methodology to compile detailed work zone crash and exposure information

• Apply the methodology to determine crash causation • Identify and prioritize crash types and causal factors to reduce work zone injuries and

fatalities • Identify countermeasures to mitigate risk based on crash exposure information • Conduct controlled experiments to evaluate promising countermeasures • Develop and apply management procedures to control consistency and quality

Task 2D will be performed by TTI. Task 2E – Fundamental Advanced Research The fundamental/advanced research paper could include, but not be limited to, research of new information, computational analysis, methods, and modeling, materials, and technologies that

16

could be used to improve highway safety or to better understand highway safety issues or the research of those issues. While fundamental/advanced research will be a separate white paper, if there are priority research projects associated with the five topic areas from the Irvine Conference, those fundamental/advanced research projects should also be included and ranked in those topic areas. The draft of the white papers shall be due 17 weeks after initiation of contract. The government will have 4 weeks to review the white papers and make comments. Telephonic communications between government personnel and the white paper authors may be necessary to resolve certain issues. The final white papers shall be due 23 weeks after the initiation of the contract. Task 2E will be performed by Cambridge Systematics. Task 3 – Integration White Paper As mentioned in the introduction to the white papers in Task 2, there will be overlap in the white papers. For instance, it is easy to understand that research can, and perhaps should, be conducted which covers intersections, speed management, and intelligent infrastructure. This white paper needs to address how the individual white papers fit together. This white paper should be a synthesis of the ideas and research topics contained in the other white papers. It should make recommendations which will aid those who will decide on the list of research topics to be evaluated. This attempt at synthesis will be a difficult one so this integration white paper should be considered a first approximation at integration of safety research topics. The draft of the Integration white paper shall be due 25 weeks after initiation of contract. The government will have 2 weeks to review the Integration white paper and make comments. Telephonic communications between government personnel and the white paper author may be necessary to resolve certain issues. The final Integration white paper shall be due 29 weeks after the initiation of the contract. Task 3 will be performed by UMTRI. Task 4 – Quantitative Approach to Evaluating Safety Research Topics Based on the white papers (and perhaps other input), a list of safety research topics for each of the six research areas will be developed. These topics will then be evaluated by subject matter experts. Decisions on which safety research topics to pursue will be based to a large degree on the ratings of the expert evaluators. In order for the expert ratings to be interpreted in a uniform and consistent manner, a quantifiable approach must be developed. The approach must allow for the comparison between the ratings within the topic area and across topic areas. Fortunately, a methodology for rating and comparing research topics has already been developed and validated under the FHWA ATIS/CVO project (Kantowitz, Lee & Kantowitz, 1997). It is based on the general linear model. The study showed that this psychometric model is fairly

17

insensitive to criteria weightings: as long as the ranks of the weightings are unchanged the precise values are not that important. In order to apply this method we need (a) a list of research topics and (b) a set of criteria. The topics will be generated in Task 2. Criteria used in the earlier study included Congestion, Safety, Mobility, Environment, Economic, Existing Data, Suitability for Guidelines, Older Drivers, Younger Drivers, Cost and Time to Complete. This set will be modified to suit the goals of the present project. Experts will rate the topics for all criteria on a five-point scale. A spreadsheet model will calculate the most important topics within a category and, more importantly, the most important topics across categories. Among the items that UMTRI will consider are:

a. Criteria definition – for example: • size of the crash problem which will be impacted by the research • anticipated portion of the problem that countermeasures from the research would

impact • degree and length of time that the countermeasures are estimated to be deployed, • probability of successful research, • research costs, and • length of time to perform the research.

b. Scale for each criteria, c. A weighting system across criteria, d. Categorization of efforts into those that will result in:

• ready solutions or strategies/countermeasures that will potentially improve safety (applied),

• knowledge breakthroughs that can impact safety in a number of indefinable ways (fundamental), and

• new strategies/countermeasures based upon the fundamental research findings. e. List of potential project evaluators.

The draft of the “Quantitative Approach to Evaluating Safety Research Topics” paper shall be due 32 weeks after initiation of contract. The government will have 3 weeks to review the Integration white paper and make comments. Telephonic communications between government personnel and the white paper author may be necessary to resolve certain issues. The final “Quantitative Approach to Evaluating Safety Research Topics” paper shall be due 36 weeks after the initiation of the contract. Task 4 will be performed by UMTRI with expert ratings completed from all team members.

18

2.0 Run-Off-Road Research Needs

2.1 Introduction There were more than one million run-off-road (ROR) crashes in the United States in 2002, of which 387,000 caused injury and 15,458 were fatal. A ROR crash occurs when a single vehicle leaves the traveled lanes and encroaches on the shoulder, the median, or the roadside and either collides with an object, overturns, or both. ROR crashes do not include situations in which a collision that occurred on the roadway resulted in a vehicle leaving the roadway or situations in which a vehicle crosses the centerline and crashes into a vehicle traveling in the opposite direction. However, these crash types are relevant because they often involve the same circumstances as ROR crashes. The factors that contribute to vehicles leaving the roadway are varied and include inattention, drowsiness, use of drugs or alcohol, speeding, steering overcorrection, vehicle failure, and on-road avoidance maneuvers or collisions, among others. Road conditions can also contribute to ROR crashes. These include adverse weather, poor visibility, inadequate signing, and poor pavements, among others. Road type is also a factor in ROR crashes, with 85 percent occurring on non-freeways and 66 percent occurring on rural roads (Najm, Koopman, Boyle, & Smith, 2002). ROR crashes most often occur with fixed objects such as trees, shrubs, poles, curbs, guard rails, and embankments. Trees, shrubs, and poles account for 43 percent of all fatal fixed object crashes. Curbs, culverts, and ditches account for an additional 20 percent of fatal fixed object crashes. ROR crashes can also occur with non-fixed objects such as parked vehicles and pedestrians. Collisions with parked vehicles accounted for about 336,000 crashes in 2002, but were less likely to be severe than collisions with most fixed objects. Vehicle rollovers are the most severe ROR crash type. About 64 percent of rollovers result in injury or death. In recent years, many improvements have been made to reduce the likelihood and severity of ROR crashes. These improvements affect roadway infrastructure, signing, roadside design, and vehicle characteristics. Nonetheless, ROR crashes continue to occur at a substantial rate. Although ROR crashes accounted for less than 17 percent of all nonfatal police-reported crashes, they accounted for more than 40 percent of all fatal crashes. Although some of this difference may be accounted for by inconsistent rates of police reporting for different crash types, this statistic indicates that ROR crashes are more often fatal than on-road crashes. There are two primary explanations for this. First, many roadside objects such as trees and utility poles tend to cause more serious damage when struck than unfixed objects such as other vehicles. Second, vehicles that run off the road sometimes roll over. According to 1999 FARS data, 19 percent of fatal ROR crashes involved a rollover as the first harmful event and 41 percent of fatal ROR crashes involved a rollover as the most harmful event. There are three general areas in which improvements can be made to reduce the ROR problem:

19

1. Prevent the vehicle from leaving the roadway. Although the primary responsibility to prevent vehicles from leaving the roadway falls upon motorists themselves, transportation officials can provide an environment that helps motorists to drive more safely. General improvements that can reduce the likelihood of vehicles leaving the roadway include repaving and resurfacing, installing rumble strips, improving curve and edgeline delineation, and improving signage and roadway markings. Additionally, in-vehicle systems that provide a warning to drivers when they encroach on the roadside are currently being tested.

2. Prevent the vehicle from striking a roadside object or rolling over. The most effective way

to prevent a vehicle from striking a roadside object is to provide a clear zone free of obstructions. However, it is rarely possible to clear all obstructions. Alternatively, traffic engineers can focus on removing only the most hazardous objects. Rollovers can be reduced by improving ditch and slope design and by using roadside surface materials that minimize tripping (the event in which the rollover is initiated) and maximize motorists’ ability to maintain or regain control of the vehicle.

3. Minimize crash severity. The two factors that have the greatest influence on crash severity

are vehicle speed and the features of the struck object. Roadsides should be designed using surfaces, slopes, and clear zones that allow the vehicle to decelerate as much as possible before encountering roadside objects. Additionally, many roadside objects can be made more forgiving. Hazardous objects that cannot be removed or replaced can be shielded using attenuation devices such as guard rails or crash cushions. In locations where new trees will be planted on the roadside, smaller, less hazardous species can be chosen.

Although research and practice have contributed substantially to safety improvements in all three of these areas, there are many potential improvements that have not been investigated or implemented to the extent that their contribution to ROR crash reduction can be known. The purpose of this white paper is to recommend programs of priority research to investigate innovative and unproven approaches to reduce the ROR crash problem. The Future Strategic Highway Research Partnership’s (F-SHRP) vision for roadside safety is “a highway system where drivers rarely leave the road; but when they do, the vehicle and roadside work together to protect vehicle occupants and pedestrians from serious harm” (McGinnis, 2001). This vision emphasizes the interrelation of the three areas described above. Campbell, Lepofsky, and Bittner (2003) note that two current NCHRP projects are being conducted to better understand the factors that influence ROR crash severity, so “In light of these projects, the F-SHRP Safety Research Plan is focused on understanding the initial road departure event.” In concert with the goals stated by F-SHRP, this white paper focuses on the fundamental understanding of the underlying human and situational factors that are associated with ROR crashes. Such knowledge is a necessary foundation for the development of countermeasures, and will be especially helpful in targeting countermeasures to the locations and situations where they will be most cost-effective.

20

2.2 Run-Off-Road Research Topics Category Project Title Type of

ResearchLikelihood of

success (1-5 scale)Duration (months)

Cost (in millions)

Run-Off-Road

ROR 1: Use of Rumble Strips on Non-Freeways

Applied 5 36 1

Run-Off-Road

ROR 2: Development of a System of TCDs to Reduce ROR Crashes at Curves

Applied 4 18 1

Run-Off-Road

ROR 3: Optimizing the Net Benefits of Delineation

Applied 4 36 1

Run-Off-Road

ROR 4: Development and Application of a Roadside Inventory Database

Advanced 3 24+ 1.5

Table 2.2.1

2.3 Knowledge Strongholds and Gaps This section presents a brief overview of strategies to reduce the likelihood and severity of ROR crashes. The purpose of this overview is to distinguish strategies that have a known effect (knowledge strongholds) from strategies that are not well understood, but may have a positive effect (knowledge gaps). More thorough evaluation and discussion of many of these strategies can be found in sources from which this summary draws substantially. These include AASHTO’s Roadside Design Guide, NCHRP Report 500 Volume 6: A Guide for Addressing Run-Off-Road Collisions (Neuman, et al., 2003), Strategic Plan for Improving Roadside Safety (McGinnis, 2001), Detailed Planning for Research on Making a Significant Improvement in Highway Safety: Study 2 – Safety (Campbell et al., 2003). Curve Design. The effects of curves on ROR crashes are generally well-documented. Vehicles are 1.5 to 4 times more likely to leave the roadway on curves than straight roadway sections (Glennon, Newman, & Leish, 1985; cited in Neuman et al., 2003) and crash likelihood generally increases with curve sharpness. However, little is known about the speeds and angles at which vehicles tend to run off the road on curves, how these characteristics differ on straight road segments, and what effects they may have on crash likelihood and severity. Furthermore, additional research is needed to determine why driver errors continue to occur on curves, often despite the presence of TCDs. Surfacing Materials. Skid-resistant surfaces help to prevent vehicles from losing traction and control. Skid-resistance is often most important when roads are wet. New York State

21

established criteria for resurfacing roadway sections based on the proportion of crashes in wet and dry weather. Following treatment, wet-weather crashes at the resurfaced locations decreased 50 percent. NCHRP Report 486 (Harwood, Kohlman Rabbani, Richard, McGee, & Gittings, 2003) provides guidelines on how to select the most cost effective locations to resurface and to decide if geometric improvements such as shoulder paving should be made during the resurfacing effort. Shoulder Design. A paved shoulder that is free of objects can provide a “margin of error” for drivers that allows them leave the traveled roadway without losing control or striking a roadside object. Shoulder width has an inverse relationship with crash likelihood. The Federal Highway Administration has established accident modification factors (AMFs) to predict the crash reduction that would be associated by a given increase in shoulder width or an upgrade of shoulder surfacing (Harwood, Council, Hauer, Hughes, & Vogt, 2000). Although the effects of shoulder width and materials are well-established, the effect of edgedrop height (the vertical dropoff where the edge of the shoulder meets the roadside) is not. It is generally assumed that the larger the edgedrop is, the more difficult it is for a driver to safely return to the roadway once one or more tires have exceeded the edgedrop. However, there is insufficient data to back up this assumption. Rumble Strips. Shoulder rumble strips have been shown to substantially decrease the likelihood of ROR crashes on freeways with paved shoulders. However, much less is known about the effects that shoulder rumble strips may have on nonfreeways, many of which have smaller shoulders and smaller clear zones than freeways. There is also little known about the effects of rumble strips on roads with no shoulder or an unpaved shoulder. Potential treatments for these roads include rumble strips on the edgeline or in the middle of the lane. Midlane rumble strips are intended to rumble when the vehicle’s wheels closest to the centerline move too close to the edge of the roadway. Centerline rumble strips, which have been recently investigated by Kansas Department of Transportation and the Insurance Institute for Highway Safety (IIHS) may also have the potential to reduce the likelihood of vehicles crossing the centerline and colliding with oncoming traffic. Signage. Signage can be used to warn drivers of upcoming curves, hills, and other features that may be associated with ROR crashes. The MUTCD currently provides guidelines for signs to indicate the direction of curves and the safe speed. Innovative signage using ITS technology may also help to control vehicle speed. For example, a speed detection system can be installed a sign prior to a sharp curve that provides a “Slow Down” message to vehicles entering the curve at an unsafe speed. ITS can also be used to create a variable speed limit system that can be adjusted according to road or weather conditions. Pavement Markings and Roadway Delineation. The primary purpose of roadway delineation as it relates to ROR crashes is to provide drivers with advance information about the path of the roadway ahead. The MUTCD provides extensive and detailed guidance on pavement markings. It also often provides more than one marking option to engineers, and these options may differ in terms of safety benefits and costs to implement. For example, there is evidence that raised pavement markers (RPMs) to supplement edgelines can improve delineation and thereby reduce ROR crashes, but the benefits of RPMs may depend upon the type of roadway where they are

22

installed. More research is needed to determine which pavement marking options are most effective, and most cost effective, for a variety of roadway conditions. Slope and Ditch Design. There are two primary goals of slope and ditch design: 1) allow the driver to maintain control of the vehicle and 2) minimize the likelihood of a rollover. Maintaining control of an errant vehicle increases the likelihood of safely returning to the roadway or coming to a stop without striking an object or rolling over. Crash testing, simulation, and crash statistics have helped researchers to create a comprehensive picture of why rollovers occur. Off-road rollovers most often occur when a vehicle strikes an object while skidding laterally. In nearly a third of single-vehicle rollovers, the first object struck is a ditch or embankment. In general, ROR crashes are more likely to occur on wider and flatter slopes with surface materials or objects that can trip a skidding vehicle. AASHTO’s Roadside Design Guide (2002) recommends safe clear zone distances according to side slope steepness, design speed, and design ADT. Although safe slope and ditch design have been established, improving existing roadways can be difficult and expensive. It is therefore a research priority to target slope and ditch improvements to locations where they can provide the greatest safety benefit. Roadside Crash Hazards. The simplest way to ensure that vehicles do not collide with roadside hazards is to remove all hazards within a reasonable distance of the roadway. AASHTO’s Roadside Design Guide (2002) provides guidance on recommended clear zone distance for various roadway types. However, it is not always possible to create safe clear zones, especially on nonfreeways where trees, curbs, posts, and an assortment of other hazards are often located adjacent to the roadway. For a variety of reasons, removal of hazardous objects is often not an option. Alternatives to removal include installing less hazardous barriers and attenuation systems (e.g., guard rails and crash cushions) and replacing hazardous objects with less hazardous alternatives (e.g., breakaway signs and poles). NCHRP Report 350 (Ross, Sicking, Zimmer, & Michie, 1993) provides guidelines for crash testing and in-service evaluation of various highway safety features. Another alternative to object removal is improved delineation of roadside objects using reflective markings. Section 3C.03 of the MUTCD provides guidance on markings for objects and other hazards, such as drop-offs, adjacent to the roadway. However, it may be possible to develop innovative delineation markings that are more effective than the standard markings. Summary of Knowledge Strongholds and Gaps. Thanks to decades of research, highway engineers have an arsenal of ROR countermeasures available for implementation. Guidelines documents have been developed that specify how, and under what conditions, countermeasures should be implemented. Crash databases, crash testing, and simulations are being used to document how and why ROR crashes occur. One of the most recent advances in ROR crash prevention and mitigation is the development of analysis programs that can be used to target roadside improvements to the locations where they are most needed. Although ROR research has progressed substantially, there is still much to be learned. Many roadway and roadside improvements have been studied individually, but very few studies have focused on how ROR countermeasures function as a system. Past research has also not provided a strong basis for understanding the driver behavioral events that result in ROR crashes and the human factors considerations that relate to these causes. Although statistics describe the

23

situations associated with ROR crashes (e.g., speeding, alcohol/drug intoxication, slick roadway) less is known about the specific driver actions that cause (or fail to prevent) ROR crashes. F-SHRP notes that most collisions are to some extent caused by driver errors; therefore understanding the driver’s actions in the moments before and during the ROR crash is critical to understanding why ROR crashes occur and how they can be prevented and mitigated (Campbell et al., 2003).

2.4 Critical Future Highway Issues The highway environment is not static. The future may bring changes in travel patterns, traffic characteristics, driver demographics, driver behavior, vehicle characteristics, communications technologies, and other factors. As the transportation system evolves, established knowledge and practice may become dated as new problems and opportunities arise. Researchers should be aware of likely changes so that appropriate conditions can be included in their detailed study plans. As Campbell et al. (2003) noted, the “changing traffic environment both complicates and heightens the need for fundamental traffic safety research.” This section highlights those potential changes that are most relevant to ROR crashes. It draws on issues identified in F-SHRP reports and other sources. Issues are grouped under three general headings: changes among drivers; changes in vehicles; and changes in the highway environment. Changes Among Drivers Aging Population: As members of the baby boom generation enter their sixties and seventies, older drivers are likely to represent an increasingly substantial percentage of the driving public. Staplin, Lococo, Byington, and Harkey (2001) have discussed how standards and practice for sight distance, horizontal and vertical alignment, and associated TCDs are based on driver performance capabilities that are affected by age. These capabilities include diminished visual performance, slower reaction time, attentional deficits, poorer physical vehicle control capability, and less accurate perceptual judgments. Current practice may not adequately meet the needs of older drivers, and Staplin et al. provide some guidelines for design, delineation, and signage that better accommodate the older driver. Future practice related to ROR crashes should take the increased older driver population into account. Proposed research that includes driver perception and response considerations must include adequate representation of older drivers and new practices must match their capabilities Aggressive Driving: Aggressive driving is a current hot topic in highway safety, but has proven difficult to study empirically (Llaneras & Lerner, 1999). Part of the difficulty can be attributed to the lack of clear definitions of aggressive acts. For example, red light-running may be done aggressively to beat the light or accidentally as a result of driver inattention or misjudgment of signal timing. As highways become increasingly crowded, it is likely that aggressive driving will become more common and it is important to understand it’s causes and effects. For instance, speeding is a common precursor to ROR crashes, but less is known about the effects of speeders on surrounding traffic. Other aggressive behaviors that may play a role in ROR crashes include frequent or erratic lane changes, cutting off other drivers, tailgating, and aggressive interactions (i.e., road rage). There are many challenges inherent in understanding drivers’

24

intentions and motivations, but future research may reveal why aggressive behaviors occur, how they relate to ROR crashes, and how they can be prevented. Changes in Vehicles In-Vehicle Distractions: Technologies such as entertainment systems, navigation systems, and handheld devices such as cell phones are becoming increasingly common in vehicles (Llaneras & Singer, 2002). Empirical evidence suggests that in-vehicle distractions can have a negative effect on driver performance, including lane-keeping, perception-reaction time, and eyes-off-road time. The universe of in-vehicle devices is growing rapidly and the distractions that put drivers at risk in the future may not be the same as the distractions drivers currently face. Driver distraction may have implications for traffic control devices. Visual cues will be less effective, so acoustic or tactile signals may be important. Much depends on future device designs, accessibility, and regulation. Although in-vehicle devices are likely to become increasingly common in vehicles, designs that reduce driver workload and prevent drivers from using them at inopportune times may contribute to safer device use. Hazard Warnings and Alerts: With improvements in vehicle sensing technologies, hazard warnings and alerts are becoming feasible for wide scale deployment. Many vehicle manufacturers currently offer proximity sensors on their vehicles, and more advanced systems are under development. Future systems may help to prevent ROR crashes by using a combination of GPS, GIS map databases, and vehicle sensors to determine the appropriate speed for an upcoming section of road and warn drivers who are encroaching on an edgeline or centerline. In fact, the recently developed Road Departure Crash Warning (RDCW) system includes both a speed warning component and a lane departure component (University of Michigan Transportation Research Institute, Visteon Corporation, & AssistWare Technologies, Inc, 2003). The use of hazard warnings and alerts may hold great promise for reducing ROR crashes, but a number of important questions need to be answered. First, research must be conducted to choose the most appropriate warning modalities and features. Second, warnings algorithms must be calibrated to maximize real warnings while minimizing nuisance warnings (false alarms). Finally, special consideration must be given when multiple warning systems are present in the same vehicle (e.g., lane departure, speed warning, proximity alert). Drivers must be able to immediately recognize the meaning of a warning if they are to respond quickly and appropriately. The National Highway Traffic Safety Administration (NHTSA) is currently developing and testing alerts for use under conditions of multiple warnings issued individually and concurrently. Driving Automation: Systems that automate portions of the driver’s task are just beginning to penetrate the vehicle fleet. These systems sense the roadway environment and intelligently respond to changing conditions. Automation can be employed for driver convenience or for imminent crash avoidance. Although a variety of automation systems have been developed or simulated, adaptive cruise control (ACC) is the only system that is currently available as a manufacturer option in passenger cars. Automation for imminent crash avoidance involves many of the same issues as hazard warnings and alerts, but these systems go one step further by taking some degree of vehicle control away from the driver. Fundamental questions regarding the use of automation for imminent crash avoidance are yet to be answered. For example, how should alerts and warnings be integrated with automation and at what point should control be

25

taken away from the driver? The answers to these questions will likely depend on the specific system in use and the nature of the imminent crash scenario. Although automated crash avoidance may have the advantage of instantaneous response, the nature of the vehicle response may not always be appropriate for the specific threat. Furthermore, little is known about how drivers react when vehicle control is taken from them. With regard to automation for driver convenience, the primary concern is that drivers may become complacent or over reliant and fail to properly monitor the environment or respond to changing conditions. An additional concern is drivers who are unsure of the degree of automation (e.g., Will ACC respond to stopped traffic ahead?) Crash Detection and Reporting: The basis of crash detection and reporting systems is an event data recorder (EDR). EDRs, which are commonly referred to as “black boxes,” collect information about vehicle motion, inputs, outputs, and system functionality. EDRs are sometimes installed by vehicle manufacturers to aid in vehicle diagnostics, but in the case of a crash, data collected immediately before the crash can be used to reconstruct the crash scenario. Crash reconstructions can help researchers to understand the cause of the crash, the events that cause harm, and what countermeasures may have prevented or mitigated the crash. EDRs can be configured to detect a crash if some criterion is met, such as an extreme deceleration or an airbag deployment. Although current EDRs capture some useful information, additional data such as vehicle trajectory, deceleration, and lateral motion would be valuable in helping to reconstruct crashes. With additional data, EDRs could even be used to record close calls, as indicated by antilock brake activation, lateral vehicle motion, or extreme steering inputs. NCHRP Project 17-24, entitled Use if Event Data Recorder (EDR) Technology for Roadside Crash Data Analysis, which will be completed in June 2004, will recommend a minimum set of EDR data to address roadside safety. As EDRs become more common in vehicles, crash reconstructions may provide an increasingly clear picture of why crashes occur and how they can be prevented. In addition to aiding in crash reconstruction, EDRs can also be used to report crashes. If the EDR detects vehicle characteristics that indicate that a crash has occurred, a wireless communication device can be used to report the crash. When used with a GPS unit, the vehicle’s location can be reported to emergency services, even if the victim is not capable of communicating. The OnStar system in some General Motors vehicles provides this service. Automatic crash reporting is especially useful for ROR crashes because they often occur in rural areas where there may not be bystanders to report the crash. Vehicle Compatibility with Roadside Hardware and Features: The requirements for roadside hardware (e.g., guardrails, signage) and roadside features (e.g., side slope, shoulder treatments, clear zones) are related to features of the errant vehicle. The current vehicle fleet is very diverse, and includes automobiles, motorcycles, SUVs, vans, light trucks, and heavy trucks of various sorts. These vehicles differ in size and weight, stability characteristics, handling, braking, and may also be traveling at a range of speeds. Designing roadway and roadside features to be compatible with the full range of vehicles is a current challenge. This problem undoubtedly will continue into the future and the characteristics of the vehicle fleet may change in a variety of ways. The particular changes may be difficult to predict and may be influenced by economic, technology, environmental, and policy considerations. In addition to changes in vehicle size, there may be new types of passenger and commercial vehicles as well as non-traditional downsized personal transportation products. Alternative fuels and alternative power systems

26

may influence vehicle size and handling characteristics. Vehicle control and stability systems, such as steer-by-wire, brake-by-wire, or intelligent suspensions, may influence performance related to ROR crashes. Various degrees of machine intelligence and automation may also affect crashes. Intelligent cruise control and driver assist systems are currently developing examples. While predicting the evolution of the vehicle fleet decades from now may be difficult, it seems likely that diversity will continue to present challenges for design of the roadway and roadside to optimally mitigate ROR crashes. Changes in the Highway Environment Increased Traffic Volumes: Motorists traveled 2.83 trillion miles on US highways in 2002, up from 2.25 trillion in 1992 and 1.60 trillion in 1982. As the number of vehicle-miles driven increases, traffic volumes are likely to increase on many roads. Increased attention should be given to the effects of traffic volume, especially as average daily traffic begins to exceed design capacity. Increasing traffic volume is likely to have a complex effect on ROR crash likelihood and may also vary by road type. Research is needed to determine the effects of increased traffic volume on ROR crashes and to help choose appropriate countermeasures if highway improvements are needed. Intelligent Signing: Most roadway communications to the motorist, such as signs and markings, are fixed. Even those displays that are variable (changeable message signs [CMS]) are generally changed by direct intervention of a human controller. However, it is likely that future highways will take more advantage of intelligent systems to communicate warnings to the motorist. Roadway based sensors, processors, and communications, used in conjunction with display technologies such as CMS or in-vehicle display panels, can be used to dynamically communicate changing situations to motorists. Intelligent sign systems have been demonstrated or implemented for such applications as speed control, work zone messages, and route guidance. Intelligent systems have potential application to ROR crashes. They may be used to detect and warn about hazardous conditions, such as wet/icy road surfaces, queued traffic, work zones, or obstacles. They may be used to indicate safe speeds based on current conditions. They may also be used to provide targeted messages to individual drivers. For example, a message might be provided to a driver approaching a curve at too high a speed, and might even consider vehicle characteristics (e.g., size, weight). To date, there have been few efforts directed specifically at ROR crashes, although there are examples such as curve warnings when the road is wet or icy (Brisbane & Vasiliou, 2002) and freeway exit ramp warnings to large trucks if high speeds and potential instability are sensed (Strickland & McGee, 1997). More research will be needed on the benefits, drawbacks, and cost effectiveness of intelligent roadway systems directed at ROR problems.

2.5 Research Recommendations Comment from R+T Partnership Steering Committee Ongoing and Proposed Projects – It’s not possible to tell what recent ongoing or proposed projects were reviewed (e.g., F-SHRP and NCHRP 17-11). Some other examples include: For ROR 1, “Crash Reductions Following Installation of Centerline Rumble Strips on Rural Two-Lane Roads” published by the Insurance Institute for Highway Safety (IIHS) should be

27

referenced. With the results of this study, there doesn’t appear to be a significant need to do more research in this area. Ongoing NCHRP 3-61 project, “Communicating Changes in Horizontal Alignment,” has similar objectives to ROR 2. The ROR 2 write-up should explicitly acknowledge and address the relationship with this ongoing project. Also, there’s the NCHRP 5-17 project “Safety Evaluation of Permanent Raised Pavement Markers” currently being completed. Response from White Paper authors

Comments from the R&T Partnership Steering Committee were thoroughly reviewed and the responses have been fully integrated into this revised chapter. The responses were not provided on a question-by-question basis (we had not interpreted the request in this way), but rather woven appropriately throughout the text discussion and recommendations.

Comment from R&T Partnership Steering Committee Speed - The authors mention speeding as one of the contributory factors to ROR crashes but do not address the need to manage speed in the subsequent reasoning. Yet speed affects both the frequency and the severity of ROR crashes. Therefore, research into speed choice and speed control ought to be given consideration. Comment from R&T Partnership Steering Committee Median Barriers - The safety effects of median barriers are still unclear so research on this topic should be explored. Comment from R&T Partnership Steering Committee Curve Design and Safety - There are still a number of unresolved issues pertaining to curve design and safety. What are the relationships of tangent length, superelevation, pavement widening, lane/shoulder values, etc. with safety? The effects of curvature on multi-lane roads may be quite different than for two-lane roads, but requires more research. In regards to delineation treatments for curves, changing unsafe behaviors by traffic control devices has low probability. Instead, broaden to include research on effects of shoulder paving, widening, and for other factors. Comment from R&T Partnership Steering Committee Shoulder Widths - The relationship of shoulder widths and safety is still unclear. Therefore, more research is needed. Comment from R&T Partnership Steering Committee Understanding Driver Behavior – The authors of the papers indicate that past research has not provided a strong basis for understanding the driver behavioral events that result in ROR crashes and the human factors considerations that relate to these causes. However, there is no reasoned

28

description of how better “understanding the driver behavioral events that result in ROR crashes and the human factors considerations that relate to these causes” would lead to cost-effective countermeasures. A general understanding of the main factors already exists. As noted early on in their white paper: “The factors that contribute to vehicles leaving the roadway are varied and include inattention, drowsiness, use of drugs or alcohol, speeding, steering overcorrection, vehicle failure, and on-road avoidance maneuvers or collisions, among others.” Nor is there a particular shortage of research on inattention, drowsiness, alcohol, etc. Comment from R&T Partnership Steering Committee Visibility - The paper identifies that visibility (or lack thereof) and visibility issues (signage and pavement markings) are factors that should be evaluated. However, the potential for improved lighting, both in terms of absolute levels and optimal use of lighting, are not considered. While the data tend to support the claim that improved visibility, through the use of fixed roadway lighting, will result in a reduced crash rate and reduces the severity of crashes, additional research needs to be completed to calculate the actual impact of a lighting system for different highway types and intersections. Comment from R&T Partnership Steering Committee Crash Testing and Simulation Work - While the authors noted three areas of ROR improvements on page 4, all this work seems to be oriented to preventing the vehicle from leaving the road, with nothing on preventing crashes with roadside objects or lessening the severity of rollovers or crashes with objects. If FHWA is going to continue to fund crash test and simulation work, someone needs to look closely at these two areas and determine what the major research needs are. Research recommendations for four high-priority topic areas follow. A primary objective of this white paper was to identify a selected set of key research topics and develop these into formal research problem statements. Numerous possibilities for addressing ROR problems exist, and the set of four recommended topics should not be seen as comprehensively covering all of the issues and approaches. Rather, taking F-SHRP criteria into account, it represents a judgment about the advanced and applied research topics that appear particularly promising in advancing knowledge in a way that will support future efforts in ROR crash reduction. In developing these recommendations, we have drawn on existing suggestions for research as well our own review of the problem. Documents that provide substantial insight into roadside safety issues include NCHRP Report 500, Volume 6: A Guide for Addressing Run-Off-Road Collisions (Neuman et al., 2003); NCHRP Report 350: Recommended Procedures for the Safety Performance Evaluation of Highway Features (Ross, Sicking, Zimmer, & Michie, 1993); Research-Gap Identification Background Issues: Human Factors (Hanscom, 2002); and AASHTO’s Roadside Design Guide (AASHTO, 2002) The magnitude of the ROR problem has also made this crash type a focus within a number of recent studies, workshops, and conferences. Many of these efforts have produced lists of

29

research needs for ROR. Among these efforts have been: NCHRP’s Strategic Plan for Improving Roadside Safety (McGinnis, 2001); Safety Research Agenda Planning Conference (Research and Technology Partnership, 2002); and Future Strategic Highway Research Program’s (F-SHRP) Special Report 260: Strategic Highway Research: Saving Lives, Reducing Congestion, Improving Quality of Life (2001). As a result, dozens of suggestions for research topics already exist. These range from highly specific, narrow recommendations (e.g., effect of 4-inch vs. 6-inch edge line) to general targeting of broad issues (e.g., vehicle/roadside interaction). Appendix A provides the roadside safety research topics proposed by various researchers, committees, and workshop groups. In developing the research project recommendations that follow, we have tried to integrate many of these issues into systematic programs of research activity that deal comprehensively with the major issues. ROR 1. Use of Rumble Strips on Non-Freeways Comment from R&T Partnership Steering Committee The IIHS completed a comprehensive study on centerline rumble strips (referenced earlier) so additional research on this type of rumble strip shouldn’t be a priority. Evaluating other types of rumble strips on non-freeways (e.g., shoulder rumble strips, edgeline rumble strips, rumble strips on very narrow shoulders) has merit. Response from White Paper authors Problem Statement Based on research conducted by individual states, the Federal Highway Administration estimates that shoulder rumble strips on freeways can reduce ROR crashes by 15 to 70 percent (http://safety.fhwa.dot.gov/programs/rumble.htm). However, less is known about the effectiveness of rumble strips on non-freeways. There are a number of rumble strip applications that may have the potential to improve roadside safety on non-freeways. These include: Shoulder rumble strips: On roads with shoulders, rumble strips can be placed outside the edgeline. Research is needed to determine cost/benefit ratios for narrow shoulders. Although evidence does not suggest a negative effect of rumble strips on narrow shoulders, logic dictates that as the distance between the rumble strip and the edge of the paved shoulder decreases, the likelihood of recovery decreases as well. Edgeline rumble strips: Another option for roads with narrow or nonexistent shoulders is to install rumble strips on the edgeline itself. There are three primary concerns regarding this application. First, drivers may try keep their distance from the rumble strips and as a result, be more likely to encroach into the lane to the left or cross the centerline. Second, research has not investigated the possibility that a driver who has come into contact with a rumble strip will make a steering overcorrection and encroach into the lane to the left or cross the centerline. Third, edgeline rumble strips must not degrade the quality of the edgeline delineation, though edgelines painted on rumble strips were found to have excellent visibility in a Michigan study cited by Morena (2003).

30

Midlane rumble strips: Neuman and his associates (2003) discuss the potential of midlane rumble strips to prevent ROR crashes. The authors note that midlane rumble strips may be most effective when shoulder rumble strips are not feasible. Because they are located in the center of the lane, midlane rumble strips may also be an effective countermeasure against centerline encroachments. However, negative effects of midlane rumble strips may include lack of public acceptance, risk to motorcyclists, and interference with snow removal. Furthermore, to avoid a constant annoyance during lane changes, use of midlane rumble strips will most likely be restricted to two-lane roads. Centerline rumble strips: Although crashes in which vehicles cross over the centerline and strike opposing traffic are not classified as ROR crashes, the causes are often the same. Centerline encroachments have one of three results: a collision with oncoming traffic, a ROR event if the driver does not collide with oncoming traffic yet fails to make a steering correction, or a recovery into the correct travel lane. Centerline rumble strips may alert drivers who have encroached on the centerline and help them to recover safely. However, the potential negative effects of edgeline rumble strips may apply to centerline rumble strips as well (e.g., lane deviations to avoid rumble strips and degraded visibility of centerline delineation). Centerline rumble strips are of less concern for the proposed research because 1) the primary crash types for this treatment are head-on collisions and sideswipes rather than ROR, 2) the University of Kansas (in progress) and the Insurance Institute for Highway Safety (2003) have conducted recent studies of centerline rumble strips on two-lane rural highways, and 3) additional research is needed to investigate the value of highway-dividing median barriers as an alternative to centerline rumble strips. Nonetheless, centerline rumble strips have common attributes with other rumble strip treatments directly relevant to ROR and may be used concurrently with other types of rumble strips, in which case centerline rumble strips should be considered part of a larger system of countermeasures to prevent crashes. The rumble strip applications discussed above may have the potential to reduce ROR crashes in a cost effective way. However, direct evidence of their effectiveness is incomplete. Furthermore, although many studies have found that ROR crashes are less frequent after rumble strips are installed, it is not know how rumble strips actually influence driver behavior (Campbell et al., 2003). This project proposes to investigate the effects of rumble strips for non-freeway applications, compare cost to implement versus safety improvement, and provide guidance on targeting installations to locations where they will have the most positive effect. Method / Approach Task 1: Identify alternatives and key factors. Review literature, current practices, design alternatives, and existing guidance regarding the use of rumble strips. Identify the key factors related to the use of rumble strips on non-freeways, including roadway, roadside, and vehicle characteristics. Include consideration of the requirements of non-motorized roadway users, including pedestrians, bicyclists, and ADA requirements. Develop a set of alternative treatments for the use of non-freeway rumble strips and the key factors, strengths, and weaknesses of each. Task 2: Test track or other controlled evaluation of driver response and acceptability. Systematically evaluate alternatives for rumble strip design and placement under controlled driving conditions. Compare alternatives in terms of driver reaction time, speed and accuracy of

31

recovery, using distracted and/or fatigued drivers and/or induced tracking errors. Collect data on problems and acceptability for various road user groups, including various categories of passenger vehicle, motorcycle, bicycle, and pedestrian. Investigate the changes in driver behavior that are associated with rumble strips through observation and discussion. Task 3: Identify most promising treatments by condition. Based on the findings of Tasks 1 and 2, identify a set of most-promising rumble strip treatments. Organize recommendations by roadway type and characteristics. Task 4: Field evaluation. Field-implement selected rumble strip treatments for formal comparison and evaluation. Comparisons should include a no-rumble-strip control condition as well as alternatives for rumble strip implementation (e.g., shoulder, edge line, centerline placements, RPM vs. milled/rolled, spacing, and dimensions). Site considerations should include geometry, design speed, ADT, crash history, shoulder and roadside characteristics. Treatment and control conditions must be comparable in terms of weather/seasonal effects. Measures of effectiveness should include the following:

• Driver performance: Both positive and negative effects of the treatment on driver performance should be measured. Measures should include shoulder encroachments, center line or lane line encroachments, and lane position. Since lane departures may be relatively infrequent, some automated system of data recording (e.g., sensors, video image processing) should be considered.

• Crashes and incidents: Records of crashes and incidents (from sources such as police reports, road assistance) should be compared for treatment and control conditions. The analysis should also consider “accident migration” effects (reduced crashes in the treatment area, but increased crashes in surrounding areas). The extent of the treatment area, the crash history of the roadway, and the duration of data collection should capture enough incidents to provide a meaningful comparison.

• Road user acceptance: Surveys, complaints, or other methods to determine the acceptability of the treatment to all classes of road users (including motorcyclists, bicyclists, and pedestrians).

• Maintenance issues and costs Task 5: Estimate of safety outcomes, cost/benefit. Based on the findings of Task 4 and the literature of Task 1, estimate the probable safety effects (crashes, injuries, fatalities) of various rumble strip treatments for various conditions. Conduct cost-benefit analyses, including crash costs, installation, and maintenance. Task 6: Guidelines and decision aids. Provide systematic guidance on how to select the most appropriate rumble strip treatment, warrants/recommendations for the use of rumble strips, and considerations for various road user groups. Project Duration A minimum of three years should be anticipated for this project. Tasks 1 – 3 would require a year and the field evaluation at least two years, assuming a before/after design. Because ROR crashes are rare events for any given stretch of highway, crash data may need to be collected over a longer period.

32

Payoff Potential Rumble strips have established benefits for freeway applications. Because most ROR crashes occur on non-freeways, the potential exists for significant benefits from wider use on such roads. The proposed project would demonstrate and quantify the benefits associated with various rumble strip designs and roadway conditions and would provide designers and engineers with criteria and methods for implementing the most effective treatment. ROR 2. Development of a System of Countermeasures to Reduce ROR Crashes on Curves Comment from R&T Partnership Steering Committee As the authors note, “The payoff potential of this study is largely dependent upon the assumption that unsafe driving behaviors are amenable to change.” Inasmuch as the chances of changing unsafe behaviors by a system of TCDs are not large, it seems that this specific project does not hold much promise for significantly reducing ROR crashes on curves now in service. Still a program of research into the safety of horizontal curves is in order. For curves now in service a research program along the suggested lines seems appropriate. It’s recommended, as stated earlier, that pavement widening on curves, shoulder paving, and guardrail installation be added to the measures considered (and that the title to be changed from “system of TCDs” to “system of countermeasures”). For curves not now in service the issues are entirely different. The problem should not be how to bend the drivers’ expectations to be in line with the curve on their paths but how to shape the curve so that the expectations of drivers do not need to be unduly bent. Response from White Paper authors On a per mile basis, curved sections of highway experience more ROR crashes than straight sections. Najm, Schimek, and Smith’s (2001) analysis of 1998 GES data includes a breakdown of factors associated with ROR crashes on curves. The authors report that about 28 percent of ROR crashes occur on curves. Roadway factors associated with ROR crashes on curves include: two-lane undivided roads (about 95 percent of crashes in which number of lanes was reported), non-freeways (90 percent), and a posted speed limit of 55 mph (33 percent of non-freeway crashes). Slightly more crashes occur during daytime than in the dark, and slightly more crashes occur under adverse and/or slippery conditions than dry and clear conditions, though these statistics do not reflect rates of exposure to these conditions. Although roadway and environmental factors often play a significant role in ROR crash likelihood, most crashes are to some extent the result of driver error. The driver circumstances most often associated with ROR crashes on curves include speeding (35 percent) and alcohol/drug use (18 percent). Although numerous studies have evaluated methods to improve curve safety, the frequency of ROR crashes on curves indicates that current countermeasure systems are not having an ideal influence on driver behavior. “System” refers to the concern that all of the aspects of the roadway work together in a complementary fashion and result in safe and appropriate driver behavior. These system elements include all of the various signs and markings, roadway geometric features, lighting, and other aspects, both at the curve site and on the approach to it. A

33

limitation of previous research is that it often examined elements individually, whereas the driver is influenced by the entire system of features he or she encounters. The ability to reduce the likelihood of ROR crashes through improved curve design, shoulder paving and widening, or other physical aspects of the site will vary considerably, and there will be more opportunity for new facilities than for existing facilities. The research project should address all aspects of curve site as a system, including physical features as well as TCDs and lighting, and looking at how these factors interact to determine what the driver actually does. There is a need for a comprehensive look at alternatives for a systematic approach to improving driver behavior on the approach to and negotiation of a curve. This may include treatments to improve curve awareness (e.g., fixed warning signs, active warning signs, and pavement markings), curve perception (e.g., RPMs, post-mounted delineators, and arrow panels), awareness of conditions (e.g., visibility warnings, slickness warnings, oncoming traffic warning), and speed choice ( existing countermeasures and variations on existing countermeasures such as spacing delineators to give the illusion of increasing speed), and human factors solutions (e.g., variable speed warnings based on environmental conditions, speed-activated warning to speeders using radar or laser detection, or oncoming traffic warnings to motorists on blind curves). NCHRP Project 3-61, “Communicating Changes in Horizontal Alignment,” which is currently being conducted by Michigan State University, addresses a number of these issues. The goal of this project is to develop a standard system of traffic control devices (TCDs) to better communicate information about upcoming curves to drivers. The scope of this project is limited to currently existing TCDs. This project is an important step in improving signing and delineation at curves and should provide a foundation for future advanced research. The next step should be to build on the products of Project 3-61 by investigating innovative new technologies for curve communication and by expanding the focus to include roadway and roadside design factors, as well as non-TCD countermeasures such as lighting. Continuous fixed roadway lighting is associated with a reduction in nighttime crashes, but is usually too expensive for use on the rural roads where ROR crashes most often occur. Noncontinuous lighting at hazardous locations may be a more cost-effective solution, but a number of questions must first be addressed (e.g., Will noncontinuous lighting create glare and interfere with drivers’ darkness adaptation? Will noncontinuous lighting reduce the conspicuity of TCDs? Will the lighting poles be roadside hazards themselves?). An ideal treatment for a problem curve site would accomplish three interrelated goals:

• Good driver perception of curve geometry while approaching and negotiating the curve • Driver recognition and appreciation of the hazard • Proper speed selection on approach and through the curve

These needs should be considered together. For example, improved delineation may make the path clear but can sometimes lead to faster speeds. Speed control measures may be ineffective if the driver does not appreciate the potential hazard at the site. The objective of the proposed research is to quantify the effectiveness of alternative treatments, singly and in combination, as they relate to the characteristics of the site. The goal is to provide the traffic engineer with a set

34

of options, with guidance in selecting treatments based on site characteristics and crash history. The emphasis of this project is on providing drivers with the information they need to choose safely negotiate curves rather than providing warnings and countermeasures to correct or protect errant vehicles. Method / Approach Task 1: Identify alternatives and key factors. Review literature, current practices, design alternatives, innovative prototypes, and existing guidance, regarding signage, delineation, and pavement markings for curves. Particular attention should be given to NCHRP Project 3-61. Using information from research and crash studies, conduct detailed task analyses to identify driver information needs and error potential on the approach to and negotiation of curves. These analyses should be used to generate a list of typical curve types and ROR scenarios to represent the range of important environmental and traffic conditions. Develop systems of alternative countermeasures for ROR crashes on curves and identify the key factors, strengths, and weaknesses of each. Map TCD devices and functions to the driver requirements in the task analyses. Task 2: Evaluate countermeasures. Using laboratory, test track, or closed-road methods, evaluate the systems of countermeasures identified in Task 1 across a variety of representative roadway and environmental conditions. The objective of this research is to determine the effectiveness and appropriateness of each system of countermeasures. Specifically, the countermeasures must be visible, understandable, credible, and have a positive influence on behavior. Key measures may include speed selection/uniformity, recognition/tracking of path, and risk perception. Task 3: Field implementation. Based on the results of Task 2, implement the most promising systems of countermeasures at select curves. The purpose of this implementation is to determine the real-world effects of the countermeasures. Countermeasures should be implemented at curves that represent a broad range of characteristics (e.g., speeds, sharpness, sight distances, number of lanes). Measures of success may include crash history and direct observation of motorist behaviors, including lane deviations, speed entering curve, and speed in the curve. Measures should also be recorded at matched control sites to account for trends not attributable to the countermeasures. Task 4: Develop guidelines. Based on the results of Tasks 2 and 3, as well as existing guidance, develop/revise guidelines for the use of countermeasures to help drivers negotiate curves. Guidelines should address design and installation, cost, locations where they should or should not be used, and interaction with other countermeasures. The scope of the guidelines should bear in mind that a large body of guidance already exists on this topic and that the purpose of these guidelines is to provide new alternatives and to supplement and improve current best practices. Project Duration This entire project would require at least 36 months. Task 1 would require about four months. Task 2 may require six months to one year, depending upon the study methodology. Task 3 would require about eight months to install countermeasures and conduct a before-after evaluation of vehicle speeds and lane positions. If statistical comparison of crash data is a high

35

priority, Task 3 may require at least one additional year to collect a sufficient amount of data. Guidelines development would require an additional six months. Payoff Potential The majority of ROR crashes on curves are the result of preventable driver errors. The goal of this project is to develop systems of curve delineation that reduce the likelihood of ROR crashes on curves. The payoff potential of this study is largely dependent upon the assumption that unsafe driving behaviors are amenable to change. Even small changes in behavior are likely to result in a substantial reduction in ROR crashes on curves. Although the cost of treatments such as fixed roadway lighting may limit them to the most hazardous locations, treatments using relatively inexpensive devices may be sufficiently cost-effective for widespread use. The findings of this study will be most beneficial on existing roads where inexpensive alternatives to reengineering are needed to reduce the frequency and severity of ROR crashes at curves. ROR 3. Optimizing the Net Benefits of Delineation Comment from R&T Partnership Steering Committee This proposed project is important. Its subject is the behavioral adaptation to various delineation devices and asks to what extent better delineation is converted into increased confidence, reduced alertness, higher speed etc. This project would be better classified as “fundamental research” since its object is to provide the now missing understanding of adaptation. Also, the results of the recently started study NCHRP Project 17-28 “Pavement Marking Materials and Markers: Safety Impact and Cost-Effectiveness” should feed into this research project. Also, NCHRP 5-17 project “Safety Evaluation of Permanent Raised Pavement Markers” mentioned earlier, should be considered. Finally, if lab studies are considered, it is critical that the lab measures used be validated surrogates of crashes (perhaps a fundamental research issue). Response from White Paper authors Problem Statement When roadway and roadside delineation are improved, motorists are better able to perceive the proper path, track their vehicle’s position, and recognize driving demands and potential hazards. This should result in fewer ROR crashes and reduced injury. However, research on delineation has observed mixed results (Neuman et al., 2003). Findings have ranged from clear benefits to minimal effects to negative effects. For example, Kallberg (1993) found that on certain low design standard roads, the installation of raised reflector posts led to a significant increase in nighttime crashes. One reason why the full benefits of improved delineation are not realized is that drivers may alter their driving behavior as a result of the “improvement.” In other words, if they continued to drive just as they had before the enhanced delineation, the treatment would lead to reduced crashes. But what actually happens is that behavior is altered in ways that give back part, or all, of the benefit. For example, when delineation is improved, speeds often increase (although in some studies, the delineation improvement has been confounded with road resurfacing, making interpretation difficult). Increased speed can influence safety in several ways. It can make path

36

tracking more difficult. It reduces the time that a given heading error will lead to an encroachment and it makes error recovery more difficult. Speed increases crash severity, should a crash occur. Increases in speed related to improved uproad path clarity can mean that the driver has less time to react to non-delineated road elements and hazards (“selective visual degradation”). Speed also influences where drivers direct their point of gaze and at greater speeds there is less peripheral vision sensitivity to near roadside cues. In addition to the direct effects of speed, improved delineation might negatively influence driver behavior in other ways. The driver may see the road section as less ambiguous and of higher quality, reducing the perceived risk and sense of caution. The driver may devote less attention to tracking the vehicle’s position and path. There may be more willingness to engage in distracting activities that take the eyes off the road. Finally, particularly at night, numerous roadway and roadside (e.g., object markers) delineation elements sometimes can make the path confusing and lead to driver perceptual errors. These concerns are examples of a general highway safety phenomenon termed “behavioral adaptation,” whereby roadway improvements do not result in the level of safety benefit that might be predicted, due to other induced changes in driver behavior. Improved delineation can certainly help reduce ROR crashes. However, we lack an adequate understanding of the conditions under which delineation alternatives will be effective, ineffective, or even counter-productive. We also do not understand behavioral adaptation to delineation enhancements well enough to know how to minimize its effects and maximize the benefits. The purpose of the proposed research is to understand how delineation influences driver behavior, both negatively and positively, to result in some net safety benefit. The objective is to define the methods and conditions under which delineation will be most beneficial and to identify ways to minimize the negative effects. The focus of the research is on basic, traditional fixed markers and delineation, rather than more elaborate and high-cost technologies. While both day and night conditions should be considered, the issue is probably of greater concern at night. Numerous research reports, guidelines, and standards documents that have addressed delineation, but this project is not intended to provide redundant research, syntheses, or meta analyses. Rather, this project will investigate the behavioral effects of current delineation and search for solutions to increase their systematic effectiveness. While most delineation studies have focused on crash history, visibility, and conspicuity, the proposed study will focus on the human factors that influence driver behavior. Another project with complementary goals is NCHRP Project 17-28, entitled “Pavement Marking Materials and Markers: Safety Impact and Cost-Effectiveness.” This objective of this project is to investigate the safety and cost-effectiveness of existing pavement markings and markers and develop guidelines for their use. This project was recently awarded and will be completed in 2007. A synergistic opportunity may be available between the proposed study and the NCHRP study. Method / Approach Task 1: Experimental studies of driver behavior. This task should use an appropriate mix of laboratory, driving simulator, and test track methods to develop an understanding of how drivers

37

react to various delineation configurations. Information should be acquired on such measures as the speeds that drivers adapt, the speeds they see as appropriate, glance locations, lane tracking, speed profile/braking, perceived risk/roadway demand, hazard recognition/perception-reaction time, willingness to engage in distracting activity, and the reasons underlying driver beliefs/attitudes about the road. The research should include a range of driver ages and capabilities and a range of roadway and geometric conditions. The set of experiments should provide a comprehensive description of how delineation and roadway factors influence the range of driver actions. Net safety benefits of alternative practices for key roadway/geometric conditions should be estimated. Task 2: Development predictive model of delineation effects on driver behavior. Based on the findings of Task 1, develop a model that predicts how drivers will respond to a given treatment for a given application. The model should include driver attributes as well as delineation and roadway factors. The model should predict driver behavior and resultant effects on crash frequency and severity. Task 3: Field comparison and evaluation of selected treatments. Field implements selected treatments to provide direct comparisons of alternatives in terms of behavioral effects and safety-related measures of effectiveness. The roadway situations and delineation treatments selected for field evaluation should be based on a need to identify optimal treatments, clarify questions remaining from the Task 1 experiments, and validate the Task 2 model. Task 4: Guidelines for use of delineation treatments. Based on the findings and the model, develop guidance to aid traffic engineers in selecting delineation treatments that will maximize the net benefits for a given situation. The guidance should include recommendations for TCDs in addition to delineation, where that is seen as part of an effective system to generate the desired driver behavior. The guidance should focus on existing delineation and marking devices and techniques. However, if the research suggests innovative treatments that may be more effective than current practices, these should be included as possibilities or as recommendations for further research. The scope of the guidelines should bear in mind that a large body of guidance already exists on the topic of delineation. Guidance should focus on maximizing the benefits of delineation. If feasible, adapt the predictive model into a user-friendly decision tool that can be used by engineers to predict the effects of delineation enhancements on traffic behavior and safety outcomes. Project Duration The various sequential tasks of this project would require about three years. This is based on an estimate of one year for the experimental lab/simulator/test track work, six months for model development, one year for the field evaluations, and six months for finalizing the model, developing guidance, and designing a model-based decision tool. Payoff Potential Because this work addresses difficult, fundamental driver behavioral issues, there is some risk that the findings may not be definitive. But the fundamental nature of the work also means there is enormous potential for broad safety benefits and the opportunity to impact a great deal of practice.

38

ROR 4. Development and Application of a Roadside Inventory Database Comment from R&T Partnership Steering Committee The paper does not make it clear that an early step should be the determination of roadside research questions that need to be answered and cannot without this inventory, and the determination of other uses of the data for the collecting agency (e.g., use in maintenance or hardware inventory programs, use in tort cases, etc.). Current ongoing and proposed efforts (e.g., ARAN, FHWA digital highway measurement vehicle, F-SHRP) are not documented here. This is an important topic worth undertaking. Response from White Paper authors Problem Statement In recent years, advances in GPS locating and computing technology have facilitated the development of GIS databases that can be used to maintain data about roadway geometry, traffic, maintenance information, crash history, and a variety of other information types. GIS enables highway databases to be viewed as interactive maps with data displayed by location. GPS technologies can be used to collect field data and map it to the GIS database. A thorough discussion of GPS/GIS interactivity issues can be found in NCHRP Synthesis 301: Collecting, Processing, and Integrating GPS Data into GIS (Czerniak, 2002). The combination of GIS and GPS has opened new possibilities for roadside safety. For instance, a GIS database can be enhanced to include data on roadside features and safety countermeasures, including maintenance history. This roadside data can also be linked to ROR crash data from police reports. Police crash reporting procedures may also be improved to link ROR crash data more accurately with the locations where they occur. In fact, police may be able to use GPS to identify the exact location where a vehicle left the roadway and the exact location where it came to rest. Such information would greatly improve the understanding of vehicle dynamics before and during the ROR event, and would be an invaluable supplement to (and validating measure of) crash testing and crash simulation. Improved crash reporting combined with a detailed knowledge of roadside features may then aid in the identification and treatment of the most hazardous roadside locations. Campbell et al. (2003) consider a GIS database of roadway and roadside features to be a critical aspect of F-SHRP, especially to enable a large scale, long term instrumented-vehicle study of crash risk factors in naturalistic driving. Table 2.1 lists the potential uses of a GIS database for roadside safety. The table distinguishes between basic uses (possible with GIS database alone) and advanced uses (requiring additional enabling systems or technologies). There are other uses for GIS data, but this project exclusively addresses roadside safety. The US Federal Highway Administration (FHWA) is currently testing a van that uses an array of sensors to create a three-dimensional digital highway map that includes data on roadway and roadside features (P. Mills, FHWA, personal communication, April, 2004). In addition to recording its location, the van will be able to scan pavement roughness for signs of wear, record lane width, presence of markings, and location of roadside hardware and traffic control devices.

39

All of this can be done while the van is traveling at speeds up to 60 mph with other vehicles present on the road. Although the system has not yet been field tested, it is designed to provide a GIS database and map that can be used to maintain information on roadside hardware. The system is even capable of digitizing three-dimensional roadway environments and transferring them into a driving simulator. FHWA and Pennsylvania Department of Transportation (PennDOT) plan to use the van to explore possible safety improvements at curves. Real highway segments will be digitized for study using FHWA’s driving simulator at the Turner-Fairbank Highway Research Center. A similar system is marketed by Roadware Corporation (http://www.roadware.com). ARAN (Automatic Road ANalyzer) is available in three models ranging in price from $180,000 to $1,000,000. The models range in size from a minivan to a full-size van to a larger cube van. The van can contain a maximum of 15 data collection subsystems. Data can be collected at highway speed, and include pavement roughness and rutting, pavement texture, grade, pavement condition, GPS location, location of roadway and roadside features, and video of the pavement, roadway, and roadside. Users can specify the number of subsystems they want to have. Data is coded to the exact location in a GIS and users can interact with the GIS to view roadway features and observe the condition of highway assets such as signs, roadside hardware, and pavement markings.

40

Table 2.5.1 Potential Uses of GIS Data to Improve Roadside Safety

Data Class Example Elements Potential Data Uses BASIC USES Roadway Inventory Number of lanes

Lane width Lane markings/markers Speed limit Horizontal curves Vertical curves Pavement quality Sight distance

Roadway maintenance history Link to crash statistics Enhance detail of crash statistics Evaluate compliance with current

guidance Identify locations needing

improvement Aid in studies of countermeasure

effectiveness

Roadside Inventory Hardware type and location TCD type, location, visibility,

and luminance Obstacle type and location Lighting type and location Shoulder width and materials Rumble strip type and location Slope/ditch angle/width Right of way width

Roadside maintenance history Link to crash statistics Enhance detail of crash statistics Evaluate compliance with current

guidance Identify locations needing

improvement Aid in studies of countermeasure

effectiveness

ADVANCED USES Supplement crash statistics with Event Data Recorder (EDR) Information

Extreme deceleration Lateral motion Skidding Antilock brake deployment Airbag deployment Vehicle trajectory Location of control loss Location of first harmful event Location of subsequent harmful

events Location where vehicle came to

rest

Enhances understanding about crash causation

Aids law enforcement in conducting crash reconstructions

Vehicle crash metrics can be used to supplement crash testing and simulation

Advanced EDRs may provide valuable crash surrogate data (i.e., close calls)

Provides data to support insurance and tort liability claims

Use roadway/roadside database as foundation for in-vehicle ROR crash avoidance alerts, warnings, and countermeasures

Roadway/roadside database GPS-based in-vehicle navigation

system

Curve and speed advisories Lane deviation warning system May be tied in with navigation

system

Use roadway/roadside database as foundation for driving studies

Roadway/roadside database Vehicle instrumented with GPS

as well as additional sensors and recorders

Provides researchers with detailed information about the highway environment in which subjects are driving

Especially useful for long-term naturalistic driving studies

41

Another prospect made possible through the digitization of roadside features in a GIS database is the linkage of the roadside data with in-vehicle technologies. One potential use of this linkage is an in-vehicle GPS navigation system that provides a speed advisory or warning when approaching hazardous curves, as defined by crash history and/or roadway features. Before these advantages can be realized, the roadside data must be collected and merged with a GIS. The purpose of this project is to develop a cost effective method to compile roadside information in a GIS database, linking this information to current databases and crash history, and improving highway design tools and best practices for roadside improvements. Ultimately, the addition of roadside data into GIS databases should allow engineers to gain a better understanding of factors that influence ROR crashes and how improvements can be made in the most cost effective manner. This project should also consider ITS applications of the roadside database including in-vehicle technologies. Method / Approach Task 1: Survey of DOT use of GPS/GIS and other technologies for roadside data. Conduct an exhaustive survey of state DOTs to assess how many use GPS/GIS to inventory roadside features. Compile a list of roadway and roadside features that are inventoried and the methods used to collect the data. Assess the extent to which these data aid users in identifying problem roadside locations and targeting roadside improvements. Determine what roadside data that is not currently compiled may provide additional benefit to engineers seeking to make roadside improvements. Also identify areas for improvement over current data collection, storage, and retrieval methods. Task 2: Workshop on user needs and preferences. This workshop would bring together all parties with a stake in the development of a roadside database. This includes those who will participate in data collection and data management as well as those who may benefit from the use of the database, including government officials, highway engineers, safety researchers, law enforcement, and insurance industry representatives. The focus of the workshop is to allow stakeholders to have input in the design of the system so that it meets their needs. The potential uses listed in Table 2.1 may provide a starting point, but the ultimate users of the system should suggest additional possibilities and guide the focus and prioritization of resources. Task 3: Technology review. Review potential technologies for roadside inventory, including systems currently in use. Consider technologies in terms of data collection, data transfer, data storage, compatibility, flexibility, security, and user interface. Task 4: Develop data specifications and inventory methodology. Based on the results of Tasks 1 through 3, define specifications for technologies, data needs, and data collection methodology. One likely method will involve using a one-pass van to rapidly geocode roadside features. This method was the highest-rated ROR research topic in the recent Safety Research Agenda Planning Conference (Research and Technology Partnership, 2002). The one-pass van could be supplemented by direct observation to determine hazard type or to collect additional data. Additional data may be provided from engineering plans and maintenance records. Technologies should also be acquired as part of this task.

42

Task 5: Pilot test data collection procedures. In cooperation with selected state DOTs, a small sample of roadside sites should be inventoried and added to a GIS database to ensure that all systems are functional and compatible. The data should also be checked for compatibility and usefulness with roadside crash prediction models and roadside improvement cost/benefit tools, as well as vehicle-linked systems for warning, alert, and navigation. Task 6: Collect roadside data. If Task 4 indicates that the data collection methodology is sound, thorough, and cost-effective, larger-scale data collection may begin. A plan should be developed to determine which agencies will collect data and what strategy will be used to select and prioritize sites for addition to the database. Task 7: Integrate and standardize database. Ultimately, the roadside inventory data should be integrated with other roadway databases through a common GIS map and referencing system. Standard data elements and GIS systems should be used nationwide to allow maximum flexibility and compatibility of the data. Appendix A of Strategic Plan for Improving Roadside Safety (McGinnis, 2001) includes a detailed list of objectives and action items related to this task. The database should also be compatible with infrastructure- and vehicle-based intelligent technologies. Project Duration This project requires the cooperation and participation of numerous parties, as well as the acquisition and development of technologies, many of which may need to be built specifically for this project. As a result, a period of 4 – 5 years should be allowed for adequate collection and evaluation of data. Payoff Potential If successful, implementation of an improved roadside inventory method would allow more thorough analysis of crash causation and would aid decision-making by authorities regarding implementation of ROR countermeasures. Although a roadside GIS would provide many benefits as a stand-alone system, it may also be used as a platform for future research on in-vehicle navigation systems and alerts, naturalistic driving behavior, and detailed crash metrics based on EDR information.

2.6 Summary The research projects discussed in this white paper were designed to fill critical knowledge gaps and to provide valuable insight into ROR crash problems and solutions. Based on the success of rumble strips on freeways, rumble strips are expected to provide substantial benefits on non-freeways. The project on this topic has a very high likelihood of establishing the effectiveness of various rumble strip applications on many different types of highways. The project to improve TCDs at curves also has the potential for substantial safety improvements, but success in this project requires developing TCDs that have a greater influence on behavior than currently used devices. Closely related to this project is the research goal of optimizing the net benefits of delineation. This project addresses high level issues of human perception and behavior, but an understanding of these fundamental driver issues may provide a foundation for future theory,

43

research, and safety countermeasures. All of the projects proposed in this white paper are independent of one another and take different approaches to improving highway safety. Although there are many more research needs than those addressed here, these four projects have significant potential to increase safety in a cost effective way.

44

2.7 Appendix A: Compiled Lists of Run-Off-Road Research Needs This appendix provides lists of ROR research needs statements created by various working groups, committees, and researchers. These lists reflect the variety of strategies available to address roadside safety. Some of the lists emphasize narrowly focused current needs while others take a broad approach that emphasizes the need for a program of research to work toward long term safety improvements. Although the lists have been reformatted and some details have been removed for succinctness, the content of the lists have not been edited. Strategic Plan for Improving Roadside Safety (McGinnis, 2001): This document presents a detailed plan to reduce the frequency of ROR crashes and mitigate their effects. The vision of the plan is “a highway system where people do not pay with their lives when vehicles inadvertently leave the road; but when they do, the vehicle and roadside work together to protect vehicle occupants and pedestrians.” The plan “contains 5 missions, 25 goals, 78 objectives, and 359 action items, 221 of which are research-oriented.” For simplicity, this appendix only extends to the “goal” level of detail. • Mission 1 Increase the awareness of roadside safety and support for it.

• A coalition of governmental, industrial, institutional and civic partners that will work toward the improvement of roadside safety

• A heightened awareness of the importance of roadside safety by the public • Increased emphasis on roadside safety by partners and stakeholders and better

communications between them • Sufficient financial resources • On-going dissemination programs • A roadside safety component in all DOT safety management • systems • On-going process for updating the strategic plan

• Mission 2 Build and maintain the information resources and analysis procedures to support improvement of roadside safety. • Improved roadside and roadway databases • Sufficient roadside safety information resources on crashes, in service projects, research

results, ... • State-of-the-art methodologies for analysis and simulations of crashes and crash tests • On-going programs to conduct safety analyses and identify hazardous roadside locations

• Mission 3 Keep vehicles from leaving the roadway. • Improved highway designs that reduce the probability of vehicles leaving the roadway • Improved traffic operating environment that reduces the occurrences of roadside

encroachments • Sufficient maintenance of highways and vehicles to reduce the probability of loss of

vehicle control • Improved vehicle-based systems that keep drivers on the road • Improved driver performance and behavior

45

• Mission 4 Keep vehicles from overturning or striking objects on the roadside when they do leave the roadway. • Improved roadway geometrics and roadside designs that reduce the probability of

overturns • Improved vehicle designs that increase stability • Improved roadsides that reduce the number of collisions with hazardous objects • Improved driver performance in run-off-the-road situations

• Mission 5 Minimize injuries and fatalities when overturns occur or objects are struck in the roadside. • Optimum use of roadside safety features in relation to their selection, design, installation

& maintenance • Improved roadside safety hardware • Improved vehicle compatibility and crashworthiness • Increased seat belt use and effectiveness and enhanced occupant protection systems • Improved emergency team responsiveness for highway crashes

Detailed Planning for Research on Making a Significant Improvement in Highway Safety. Study 2 – Safety (Campbell, Lepofsky, & Bittner, 2003): This report, which was prepared for F-SHRP, addresses future research needs in two high-priority areas: roadside safety and intersection safety. The report takes a holistic approach by creating a plan of codependent research studies. Research is proposed in three interrelated topic areas: research tools and methods, risk studies, and countermeasure evaluation. The list below includes project titles and objectives, but substantial additional detail is not reproduced here. • Topic 2-1: Research Tools and Methods

• Project 2-1.1: Legal and Privacy Issues in Recruiting Volunteer Drivers and Vehicles for Field Studies of Driving Safety

Objective: The objective of this project is to develop recommended procedures to address privacy and legal issues inherent in a large-scale field study with volunteer drivers using their own vehicles equipped with extensive data recording capabilities.

• Project 2-1.2: Development of Analysis Methods for Site-Based Risk Studies Using Recent Data

Objective: The objective of this project is to further develop the hardware and software used for the site-based data collection approach. The goal is a semi-automated system that identifies the primary conflict situations and calculates the conflict severity incrementally as the vehicles move through the site. Continuous conflict, or traffic, events should be uniquely identified so they can be extracted as a unit for analysis. For example, the minimum value of the relevant crash margin measure could be extracted for vehicle pairs observed in a particular conflict situation, such as left-turn-across-path. The goal of this project is to enhance the performance of current systems so they can meet the needs of the site-based risk study.

46

• Project 2-1.3: Development of Analysis Methods for Vehicle-Based Risk Studies Using Recent Data

Objective: The objective of this project is to develop the analytic approach for the F-SHRP instrumented-vehicle field study and carry out a demonstration of the method using data from recent field studies for the road departure and intersection safety issues. Key aspects include the application of crash surrogate approaches (traffic conflicts, critical incidents, near-collisions), linking of roadway information based on GPS location, and the development of data storage and retrieval methods to implement these analytic approaches with the very large data sets produced by instrumented-vehicle field studies.

• Project 2-1.4: Development of Comprehensive Roadway Information in a GIS Database

Objective: Review the available sources of roadway data linked to a base map by GPS location in a geographic information system (GIS) and make recommendations to the instrumented vehicle design project (2-2.1) for each roadway data element.

• Project 2-1.5: Application of OEM Electronic Data Recorders for Risk Studies Objective: The objective of this project is to explore the extension of this technology to support studies of collision risk. Important issues include determining the appropriate data and procedures for access and use of the information. The specific objective is to represent this application in current forums discussing the use and future of this technology.

• Topic 2-2: Risk Studies

• Project 2-2.1: Instrumented-Vehicle Risk Study—Phase I: Study Design Objective: This project will develop the design for a field study involving 4,000-5,000 instrumented vehicles operated for a period of 2-3 years. The data collection package must Future Strategic Highway Research Program 68 Safety Research Plan accommodate the needs of the road departure and intersection studies to follow. The fleet is to be split between 2-4 geographic areas to provide good coverage of both urban and rural areas. Data for the road departure and intersection studies will be separated and archived for analysis as part of the data processing.

• Project 2-2.2: Instrumented-Vehicle Risk Study—Phase II: Pilot Study Objective: The objective of this project is to demonstrate the data collection system for the instrumented-vehicle risk study.

• Project 2-2.3: Instrumented-Vehicle Risk Study—Phase III: Field Study Objective: The objective of Phase III is to carry out the instrumented-vehicle field study.

• Project 2-2.4: Instrumented-Vehicle Risk Study—Phase IV: Intersection Analysis and Countermeasure Implications

47

Objective: The objective is to quantify the contribution of driver, roadway, vehicle, and environmental factors to the risk of specific intersection conflicts and assess the countermeasure implications of the findings.

• Project 2-2.5: Instrumented-Vehicle Risk Study—Phase IV: Road Departure Analysis and Countermeasure Implications

Objective: The objective of this project is to quantify the contribution of driver, roadway, vehicle, and environmental factors to lane-keeping performance and assess the countermeasure implications of the findings.

• Project 2-2.6: Site-Based Risk Study—Phase I: Study Design and Pilot Objective: The objective of this project is to: (a) develop a road-side study design employing sets of SAVME, or analogous site-based systems, that will provide for direct and systematic comparison of selected roadway and operational design variables; and (b) carry out a demonstration of the study design.

• Project 2-2.7: Site-Based Risk Study—Phase II: Field Study Objective: The objective of the present effort is to: (a) conduct a site-based study that will provide for direct and systematic comparison of selected roadway and operational design variables, and (b) demonstrate the relationship of surrogate measures to collision risk based on historical accident records.

• Project 2-2.8: Site-Based Risk Study—Phase III: Analysis and Countermeasure Implications

Objective: The objective of this project is to quantify the contribution of roadway characteristics and traffic operations to the risk of specific traffic events (conflicts, lane departure) and assess the countermeasure implications of the findings.

• Topic 2-3: Countermeasure Evaluation

• Project 2-3.1: Identify Countermeasure Evaluation Topics Objective: The objective of this project is to revisit the identification of countermeasures for evaluation. While a tentative list of high priority countermeasures was developed by the Technical Panel and used in this research plan, it is appropriate to revisit this process with broader representation at the beginning of this topic area.

• Project 2-3.2: Retrospective Countermeasure Evaluation Projects Objective: Conduct a rigorous evaluation to determine the benefits in reduced casualties, crashes, and costs of the subject countermeasure based on retrospective crash data using the best analytic approach.

Future Directions in Roadside Safety (Dearasaugh, Jr., 1998): This circular provides the results of a workshop intended to identify strategies to improve roadside safety. Workshop members included FHWA staff, state DOT officials, researchers, and hardware manufacturers. Using NCHRP Project 17-13 (Strategic Plan for Improving Roadside Safety) as a guide, workshop members considered research topics for inclusion in NCHRP Project 22-14

48

(Improvement of the Procedures for the Safety-Performance Evaluation of Roadside Features), which will be the basis for updating NCHRP Report 350: Recommended Procedures for the Safety Performance Evaluation of Highway Features.

Mission 1 - Increase the awareness of roadside safety and support for it.

• Develop communications and training techniques to improve the awareness and knowledge of roadside safety hardware installation and maintenance personnel, local law enforcement personnel responsible for accident reports, and key decision makers.

• Develop mechanisms to create a feedback loop between the research community and federal, state, local and private partners to enhance the exchange of information concerning roadside safety research results and implementation experience.

• Investigate ways to communicate research results and training techniques effectively to local and state practitioners, targeting issues and approaches suitable for local roads and county and city engineers.

Mission 2 - Build and maintain the information resources and analysis procedures.

• Develop improved methods of collecting real-world data necessary to enhance the design clear zone and side slopes, and to better design, install, and monitor roadside hardware.

• Research aimed at developing increased knowledge of: a.) the effect of modifications to the clear zone and side slopes on crash rates and severity; b.) the in-field crash performance of roadside hardware components, including the effects of installation and maintenance practices; and c.) general measures of the severity of roadside hardware impacts as affected by vehicle type, speed, and angles of impact.

• Research aimed at developing tools for state/local roadside improvement programs, such as data-driven procedures for roadside safety decisions as part of the Safety Management System.

Mission 3 - Keep vehicles from leaving the roadway:

• Develop techniques to encourage consistent designs that conform to driver expectancy. • Investigate the use of safety audits in the roadway design process. • Research the effectiveness of shoulder rumble strips, delineation of roadside hazards and

fixed objects (poles, trees, etc.), and signing and lighting systems for all conditions, including severe weather.

• Develop and implement guidelines for low-cost maintenance improvements on low volume roads to reduce run-off-the-road crashes.

• Develop in-vehicle systems to provide safety information to the driver and/or assist in avoiding collisions with roadside objects.

• Develop technologies to enhance driver visibility during nighttime and adverse conditions (head lamp design, ultraviolet headlamps, etc.)

• Research driver monitoring systems and define driver behaviors and characteristics that contribute to loss of vehicle control.

• Develop and deploy speed enforcement strategies which promote safer driving behavior.

49

Mission 4 - Keep vehicles from overturning or striking objects on the roadside when they do leave the roadway.

• Improve understanding of interaction between vehicles and roadside geometrics to develop roadside terrain model and update guidelines for slopes, drainage, edge drop-offs, curbs/gutters, and other features.

• Develop improved benefit-cost analysis methodologies for the selection of roadside features.

• Evaluate performance of roadside safety features of side slopes and develop guidelines to minimize impacts. Revise warranting criteria to reflect new knowledge about roadside crashes. (e.g., median barrier warrants).

• Develop simulation programs based upon driver response to running off the road situations to generate information needed in driver training programs and evaluate the effectiveness of such training.

Mission 5 - Minimize injuries and fatalities when overturns occur or objects are struck in the roadside.

• Develop new methods for evaluating the risk of occupant harm in roadside hardware crashes. Assess the assumption of restrained occupants in full-scale crash tests.

• Improve crash testing procedures based on assessing many factors include the field relevance of current full-scale test impact conditions, site geometry for crash tests of guardrails, crash cushions and guardrail terminals, and the performance of a variety of vehicle types in roadside hardware crash tests. Examine possible linkages between vehicle compliance tests (FMVSS tests) and roadside safety hardware testing.

• Develop improved hardware or new hardware including generic guardrail terminals, guardrail terminals that perform acceptably in side impact collisions, and energy absorbing devices for trees and rigid poles. Use new and innovative materials in roadside hardware.

• Improve the maintenance of roadside safety hardware to assure that it will perform effectively when needed. Identify owner-agencies with innovative and effective methods of ensuring that roadside hardware is properly installed and maintained. Utilize maintenance equipment effectively to minimize the time and cost for safety hardware repair or replacement.

Safety Research Agenda Planning Conference: Summary and Proceedings (Research and Technology Partnership, 2002): The Research and Technology Partnership was created by the FHWA, AASHTO, and TRB. One of the major goals of the Partnership is to improve the research process. A breakout session of safety experts was convened to address improvements. One topic of particular focus was roadside safety. The following list contains the entire set of possible roadside safety research topics and knowledge gaps, with the issues identified as highest priority at the top.

• One pass van to inventory the roadside features – incorporate digitize the roads for vehicle/roadway interactions

• Methods for choosing ROR sites, corridors, treatment programs

50

• Rumble strips on narrow paved shoulders • III Warning systems – field test of driver response • Relationship of simulation and crash test results to real-world crash injuries • Safer ditch design for rural two-lane roads • Tree removal tradeoff research • Vehicle/roadside interaction (Roll over, tripping effects, etc. by vehicle type) • Edge line rumble strip effectiveness • Methods for doing economic tradeoffs of treatment choice or trade-offs • Tire-soil interactions in rollovers including soil modeling • Develop a device and associated elements to put utility lines under ground • Special curve warning markings (e.g., on-pavement markings • Positive guidance tools – set of tools/implementation manual • Safety effects of grades and vertical curves • Shoulder paving (with and without widening) • Shoulder widening • Role and cause of non-intersection “speed choice” (see F-SHARP) • Effects of RPMs • Road safety audits for ROR – assess an existing road and its potential to lead a driver off

the road • Slope flattening and widening effects • Encroachment distance verification with real-world data • Driver reactions on sideslopes (i.e., steer, brake or both) • Refine severity indices for roadside • Vegetation that reduces the severity of the crash • Safety effectiveness of automated speed enforcement (with credible speed limits) • “Object delineation” programs (e.g., marking trees and utility poles) • Effect of 4-inch vs. 6-inch edge line • Role of inattention, fatigue, driver error on ROR crashes (potential answers from F-

SHARP) • Effects/acceptability of mid-lane rumble strips • Speed-reduction programs to reduce ROR crashes • Resurfacing and remarking without other treatments (e.g., no alignment or roadside

changes) • Effects of variable speed limits • ROR education program development and evaluation • Optimum edge line installation and performance • Barrier placement criteria (e.g., sideslope angle) • Effect of pavement edge drop (lane and shoulder edge) • Future congestion-speed effects on ROR crashes • Effects of pavement edge wedge on should drop-off crashes • Field test/evaluation of new hardware designs • Wooden guardrail posts in frozen soil • Encroachment rates per mile by road class and design characteristics • Flattening horizontal curves

51

• Vehicle angle and speed of impact by roadside design characteristics • Barrier height after pavement overlay

Research Needs for Increasing Highway Safety Through Infrastructure Improvements (Harwood, 2002): This paper, which was presented to the Safety Research Agenda Planning Conference meeting in Irvine, CA, presents a set of research needs focused on “physical changes to the highway system including improvements to roadway geometric design features, traffic control features, roadside design features, and roadside hardware.” One section of needs addresses roadway improvements. The other section addresses roadside improvements Only the roadside research needs are presented below. The original paper presents additional discussion of each research need statement.

• Determine the most effective strategies for reduce [sic] the likelihood that vehicle will encroach on the roadside.

• Improve highway agency data on cross-median collisions on divided highways and develop better median barrier warrants.

• Develop and test improved roadside hardware • Develop better estimates of the safety effectiveness and cost-effectiveness of roadside

improvements

52

3.0 Intersection Safety

3.1 Introduction A highway intersection is the junction of two or more public roads at equal grade. This is the only segment of our highway system where vehicles are in direct conflict due to opposing flows. Not surprisingly then, the intersection, whether or not it is under traffic signal control, can be a hazardous location as evidenced by various accident statistics reported below. Intersection Safety Problem—Some Accident Statistics The magnitude of the accident problem at intersections and some key characteristics are discussed below. The source of the information includes a 2001 paper by Harwood et al. entitled “Overview of Current Intersection Safety Conditions” and a compilation of accident statistics from the 2002 NASS-GES data base. The Harwood et al. analysis is drawn from two data bases: 1) three years (1998-2000) of the Fatality Analysis Reporting System (FARS), and 2) two states – Minnesota and California – within the FHWA’s Highway Safety Information System (HSIS). Some key findings are enumerated below:

• The magnitude of intersection accidents by severity is as follows: Fatality Injury PDO Total

Signalized 1,971 444,866 822,760 1,269,597 Unsignalized 4,075 479,422 959,135 1,442,632

• About 22 percent of fatal accidents on all roads including freeways are intersection related. When not including freeways, the percentage increases to about 30 percent.

• 75 percent of the fatal intersection-related accidents were multiple-vehicle accidents.

Angle/turning collisions accounted for the vast majority of the multiple-vehicle accidents.

• Signalized intersections consistently have higher percentages of multiple-vehicle accidents than stop-controlled accidents.

• 22 percent of the intersection-related fatal accidents involved alcohol compared to 39

percent for all fatal accidents.

• The percentage of fatal and serious injury accidents is generally higher at rural intersections reflecting the higher speeds and greater response time for emergency medical services.

These minimal statistics do not adequately characterize and describe the intersection safety problem and indeed, one of the key research needs that will be offered in this paper is to develop a comprehensive taxonomy of crashes at intersections.

53

Intersection Safety—A Priority In recognition of the significant accident problem, achieving a higher level of intersection safety has become a priority of the public and private safety community as evidenced by the following:

• Intersection safety is one of the emphasis areas in the American Association of State Highway and Transportation Officials (AASHTO) Strategic Highway Safety Plan.

• Intersection safety is also included in the Institute of Transportation Engineers’ Safety Action Plan.

• Achieving a significant reduction in the number and severity of intersection crashes was identified by the Future Strategic Highway Safety Plan (F-SHRP) of the Transportation Research Board as a critical strategy in making a quantum leap in highway safety

• Intersection safety is recognized as one of four priority areas in FHWA’S Performance Plan.

• With the input of numerous public agencies and private organizations FHWA has established a national agenda for intersection safety.

Intersection Safety Research Needs—the Initial List To achieve the goal of improving the level of safety at intersections, the need for research has been recognized by the safety community. While quite a bit is known about how to design and operate an intersection to maximize its capacity to handle traffic demands, the knowledge base for optimum safety is not as robust. The need for research to fill the gaps in safety knowledge was the focus of the Research and Technology Partnership—a group of public and private safety specialists—that conducted its first Safety Research Agenda Planning conference on September 17 and 18 in Irvine, CA. The participants at that conference identified four fundamental safety issues related to intersections that needed to be addressed, namely:

1. Effectiveness of stop-sign vs. signalized traffic control 2. Effectiveness of actuated vs. semi-actuated vs. fixed-time signalization 3. Relationship between signal-timing decisions and safety 4. Safety effects associated with traffic flow characteristics.

Additionally, the following specific research topics were identified:

1. Safety effects of cross-sectional elements at intersection 2. Safety performance of roundabouts 3. Analytical tools/models for traffic engineers and planners to consider the safety

consequences of intersection safety and design 4. Safety effects of transitional elements moving from corridors to intersection

approaches 5. How to accommodate various users (pedestrians, bicycles, trucks, etc.) for different

scenarios 6. Safety effects of traffic-calming devices/perceptual measures 7. How does the culture of road user behavior evolve and how can it be influenced 8. Relationship of Intersection Sight Distance to safety 9. Analytical tools to identify which intersections to provide selective enforcement.

54

Objective of Paper The purpose of this paper is to review the research needs identified from the Irvine Conference in light of: 1) intersection research needs identified elsewhere, 2) what is already known about the various topics, and 3) what research on a specific topic that is already being conducted. From this review, the more critical research needs are to be identified with an assessment of the likelihood of success and payoff potential. Summary research statements are prepared for the most critical needs; these include a problem statement, suggested approach, and estimated project duration and cost.

3.2 Potential Research Needs While the list of research needs from the Irvine conference by itself covers a wide spectrum of issues for intersection safety, the search for research needs for this effort was broadened by a limited review of key literature and contacts with key persons involved in this topic. Specifically, the following sources were reviewed:

• Safety Research Agenda Planning Conference, Summary and Proceedings—September 17-18, 2002, Irvine, CA.

• Mid-Atlantic Intersection Safety Workshop—June 18-19, 2003, Linthicum, MD • Intersection Safety Workshop—November 14-16, 2001, Milwaukee, WI and the resulting

“National Agenda for Intersection Safety” (FHWA Publication FHWA-SA-02-007. • “Detailed Planning for Research on Making a Significant Improvement in Highway

Safety, Study 2—Safety” F-SHRP Web Document 2 (NCHRP Project 20-58(2), September 2003.

• NCHRP Report 500, Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volume 5: A Guide for Addressing Unsignalized Intersection Collisions, Transportation Research Board, 2003.

• NCHRP Report 500, Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volume?: A Guide for Addressing Signalized Intersection Collisions. (draft)

• Signalized Intersection Safety in Europe, FHWA Report FHWA-PL-03-020, October 2003.

• “Task D, High-Volume Signalized Intersection Research Needs,” Draft Report under FHWA Contract by K. Courage and L. Rodegerdts, Jan. 2, 2004.

In addition, contacts were made with the following groups known to have interests and experience in highway safety:

• AAA Foundation for Safety—who were identifying research projects • AAA Club, Michigan—who has a comprehensive intersection improvement program for

cities in Michigan • State Farm Insurance Company —who sponsors intersection safety improvements

projects. • Dr. John Mason, Pennsylvania State University—who participated in the above

mentioned conference in Irvine, CA.

55

• Dr. Forrest Council—who has identified research needs to support the Highway Safety Information System (HSIS) program and chaired a subcommittee that developed a list of ranked safety countermeasures for the F-SHARP safety plan.

From these sources, a list of research statements/topics was compiled. The resulting list was lengthy, covered a wide spectrum of types of needs, and included redundant and overlapping topics. The list was streamlined by categorizing the research needs and collapsing similar topics. The resulting list is shown in Appendix A. In as much as practical and reasonable, the individual research statements are as proposed by the sources noted above. The long list of research needs in Appendix A range from very specific issues, e.g. the safety effect of late-night flash mode, to very broad issues that encompass many aspects, e.g. the first need of a comprehensive and systematic analysis of intersection accidents. Some are simply too broad or vague in scope, e.g. relationship of cross-sectional elements to safety and need more focus. Some are a specific item within a broader issue; for example, establishing the effectiveness of rumble strips on intersection approaches is one of several statements that could be included under “determining the effectiveness of various countermeasures, especially those related to red light running” since it is one of the engineering countermeasures for red light running. Research has already been performed on most if not all of these issues. However, given that the safety community has identified these as researchable topics, indicates that the research to date has not fully or satisfactorily resolved the issue. Also, it is known that some of these topics are being addressed in current on-going research or related projects. With the considerations noted above a final list of what is considered the more critical research needs are presented in Table 3.1. In reviewing the list, perhaps conspicuous by their absence are needs focused on pedestrians and bicyclists, motorcycles, and large trucks—key users of the intersection. This is so because there already is a significant research and development program for these users and it is assumed that they would be considered in the conduct of any specific research project. This list of 14 research needs can be further reduced because for some of them there is either an on-going or planned research effort. There is an on-going research effort that is examining the safety effects of roundabouts to include pedestrians. Also, an NCHRP project will commence soon that specifically deals with blind and handicapped pedestrians crossing at roundabouts and channelization lanes at intersections. Under NCHRP, there is a project that deals specifically with high-speed intersections and should address the need expressed by number 10. Also, within the New Technology Applications category, there is a research program within FHWA that is identifying how technology, either infrastructure and/or vehicle-based, can be applied to promote intersection safety. Finally, the two research needs listed under Safety Programs are already being addressed through the development of the Highway Safety Manual, the Integrated Safety Management Process, and Safety Conscious Planning. Reference is also made to the F-SHRP research plan for intersection safety. As documented in “Detailed Planning for Research on Making a Significant Improvement in Highway Safety,

56

Study 2—Safety” (Campbell et al., Sept 2003), a multi-year research program was developed with the objective of determining the interaction of driver behavior with intersection design and operation in the risk of intersection collisions. All together, 14 projects were identified with the first five designed to develop the research tools and methods that would be used in the seven subsequent studies that would examine the multiple factors related to the risk of collisions and casualties for intersection safety issues. The final two projects deal specifically to countermeasure evaluations. The 14 projects are identified by title in Appendix B. Finally, reference is also made to the research needs for high-volume signalized intersections identified by K. Courage and L. Rodegerdts. They identified numerous knowledge gaps for intersection safety (and operations as well). From that base, they developed five research project statements, all of which have safety impact as the primary or secondary issue to be addressed. The five projects are also listed in Appendix C. With consideration to on-going and already programmed research, Table 3.2 provides a final list of what is viewed by these authors as critical research needs. The table lists eight projects by category, and notes the type of research, the likelihood of success from low (1) to high (5), the duration, and estimated cost. Preliminary project statements for each follow.

57

Table 3.1. Priority Research Needs to Promote Intersection Safety by Category

ACCIDENT CAUSATION Comprehensive and systematic analysis of accidents at intersection to determine magnitude, characteristics, and causation factors that would identify priority problems and potential countermeasures. Establish the root causes of intersection accidents attributable to driver error. Why do drivers take risks at intersections, such as running red lights or stop signs or turning left in front of on-coming traffic? Special consideration given to older drivers who are purported to be over-represented in intersection crashes. RELATIONSHIP OF TRAFFIC AND OPERATIONAL FEATURES TO SAFETY Safety impacts of no control vs. yield control vs. stop sign control vs. signal control, which would involve determining the relationship of traffic volume to accident occurrence. Effects of protected/permissive left-turn signal phasing on safety. RELATIONSHIP OF TRAFFIC CONTROL DEVICES TO SAFETY Safety effects of alternative traffic signal layouts. RELATIONSHIP OF DESIGN FEATURES TO SAFETY Relationship of intersection sight distance to safety. Safety performance of roundabouts and especially regarding pedestrians. Innovative intersection designs (geometric or traffic control devices) in areas (highly urbanized) where right-of-way is limited (where desirable design standards can not be used.). EFFECTIVENESS OF COUNTERMEASURES Establish the effectiveness of various engineering countermeasures in reducing accidents. NEW TECHNOLOGY APPLICATIONS Identify and evaluate technologies to reduce accidents at high speed, rural intersections. Develop and evaluate an automated all-red traffic signal extension system. Develop and evaluate infrastructure-based technology to advise motorists of safe or unsafe maneuvers, such as left turn in front of opposing traffic. SAFETY PROGRAMS Develop strategies to identify potentially hazardous intersections, rather than wait on occurrence. Analytical tools/models for traffic engineers and planners to consider the safety consequences of intersection safety and design.

58

Table 3.2. Suggested List of Critical Research Needs

Category Project Title Type of Research

Likelihood of success (1-5 scale)

Duration (months)

Cost (millions)

IS 1a: Magnitude, Characteristic, & Causation of Intersection Accidents

Advanced Moderate to High,

4

36 $1.5 - 2M Accident Causation

IS 1b: Establish Root Causes of Driver Error

Advanced Moderate 3

24 $0.5 – 0.75 M

Relationship of Safety to:

IS 2a-1: Safety Impacts of Alternative Intersection Controls

Advanced Moderate 3

36 - 60 $1.0 – 1.5 M

a. Traffic & Operational Features

IS 2a-2: Safety Effects of Alternative Left Turn Phasing

Applied Moderate 3

24 $0.3 M

b. Traffic Control Devices

IS 2b: Safety effects of alternative signal layouts

Applied Moderate to High, 3

36 $0.3 M

c. Design Features

IS 2c: Intersection Sight Distance

Advanced Low to Moderate, 2

24 - 36 $0.5 M

Effectiveness of Counter-measures

IS 3: Effectiveness of various countermeasures for reducing accidents

Applied Moderate to High, 4

84 for all phases

$2.0 M

Advanced Technology

IS 4: Effectiveness of & Driver Response to Automatic All-red Signal Extension System

Applied High, 5 6 $0.05 - $0.1 M

Research suggestions from R&T Partnership Steering Committee The following items seem worthy of development as full research statements in the plan. Most, but not all, of these items are addressed in Appendix A of the white papers:

― Safety effects of alternative signal control strategies (i.e., are there safety

differences or is this just an issue of operational efficiency?) ― Safety effects of adaptive signal timing (i.e., are there safety benefits or is this just

an issue of operational efficiency?)

59

― Safety effects of alternative signal timing strategies (i.e., given that the cycle length determines the number of occasions of “turbulence/hour), study the safety effects of inappropriate/inefficient cycle lengths)

― Safety effects of late-night flash mode (make sure that sight distance is considered in this study)

― Safety effects of offset T intersections vs. four-leg intersections (are two T-intersections safer than one four-leg intersection, or does this depend on the magnitude of the minor-road through volume)

― Are dynamic advance warning signs effective on intersection approaches? Are they effective on the major road when tied to the start of the red signal phase on a signalized intersection approach? Are they effective when tied to the speed of the vehicle for which the sign is activated? Are they effective when tied to the presence of a potentially conflicting vehicle on the minor-road approach when an approaching major-road vehicle is detected?

― Safety effectiveness of left-turn acceleration lanes at divided highway intersections ― Safety effectiveness of right-turn acceleration lanes ― Safety effectiveness of offset right-turn lanes ― Safety effectiveness of innovative intersection designs ― Safety effectiveness of automated real-time systems to inform drivers of the

suitability of available gaps for making turning and crossing maneuvers (evaluate existing installation in Missouri and Maine)

― Safety effectiveness of dilemma-zone detection at high-speed intersections _ Intersection size and density - All research finds that the frequency of crashes

depends on (Traffic Flow)β with β usually about 0.5. This means that if one intersection with and inflow of 10,000 vpd replaces two intersections with an inflow of 5,000 vpd, we will save about 30% of the intersection accidents. However, fewer intersections mean somewhat longer link travel. Nevertheless, since we are building and re-building our environment all the time, this potential for accident reduction ought to be examined and translated into guidance to planners and engineers

Response from White Paper authors If any of the above research topics are not included in the full list in Appendix A, we will add them in the edited version. We are not in agreement that they all need a full research statement. Some are being evaluated in current or already planned research, namely:

• Safety effectiveness of innovative intersection designs • Safety effectiveness of automated real-time systems to inform drivers of the suitability of

available gaps… • Safety effectiveness of automated real-time systems…

Some could be included within one research statement prepared here, namely: • Intersection size and density…--Statement IS 2a-1—while cast as intersection control,

the variables of size, type, and volume should be considered in the design. The issue of density, i.e. frequency of intersections per distance, is a matter of accessibility and

60

roadway network design and safety, like or not, is not explicitly considered. However, some of the guidelines that have evolved from Access Management should apply.

Some we considered but passed on, namely:

• Safety effects of alternative signal control strategies—we feel that it is predominantly an operational/efficiency issue. If you improve operations, then you improve safety.

• Safety effects of adaptive signal timing—same comment from above applies. • Safety effects of alternative signal timing strategies—same comment from above applies.

For all others, they did not make our shortlist since we were limited in scope/budget as to the number of research statements to prepare, those did not make our short list. IS 1a. Accident Causation: Determine the Magnitude, Characteristics, and Causation and Contributing Factors of Crashes at Intersections Comment from R&T Partnership Steering Committee The authors assert that in order to develop countermeasures to reduce the number and severity of intersection accidents (efficiently), one has to have a good understanding of how accident frequency and severity depend on variables such as traffic volume, design and operational features, and driver and vehicles characteristics. It would then make sense to see the purpose of a research program as the development of such an understanding. However, this is not what their problem statement is about. Their problem statement is about describing how many crashes of what kind occur at intersections and what are the factors that cause and contribute to their occurrence. Description of magnitude does not amount to understanding. Determination of factors and causes is not the same as (and falls very short of) knowing how frequency and severity depend on a factor or a variable. Response from White Paper authors We agree with the reviewer’s comment. This project statement emerged from this author’s inability to find a single or even a few documents taken together that adequately described and quantified the crash problem at intersections (which admittedly does not by itself identify causation and contributing factors). We believe such an investigation would be the first step in a comprehensive research study that would subsequently explore the caused relationship of crashes and crash severity to the various design, operational, driver, vehicle and environmental factors. We did not intend that this or the other remaining project statements are “final” statements—sufficiently and adequately developed for release for a proposal—but rather a preliminary statement of what is needed and how one might approach the problem. We would expect that a group of experts would develop a more complete research statement. Hence, for the purposes of this initial “white paper,” we have not made any changes to this research statement, feeling that it captures the first step of a multi-step program that ultimately will result in better understanding. Background

61

Intersections account for a significant percentage of all crashes and resulting fatalities, injuries and property damage. A complete and comprehensive understanding of how many occur as a function of the many influencing variables related to traffic volume, design and operational features, and driver and vehicle characteristics is paramount to developing countermeasures to reduce their occurrence. While numerous accident studies have been undertaken in the past, they have been piecemeal in that they have focused on elements of the problem or have not been conclusive because of the lack of desirable databases of crashes and the influencing variables. Improvements in agency databases are at a point where this deficiency can be overcome and when coupled with supplemental data collection can provide the needed data. Problem Statement Conduct a comprehensive and systematic analysis of crashes at intersections to determine their magnitude and characteristics and the factors that cause and contribute to their occurrence. All types of intersections--as defined by its control--on all classes of roadways under state and local control should be included. Databases from multiple states representing variation in climate and terrain should be included. Method / Approach The methods and approach should be dictated by a thorough data collection and analysis plan develop to address the many issues and facets of the problem. It is expected that several data bases will be utilized to include the FARS and GES databases from NHTSA’s NASS, FHWA’s Highway Safety Information System, and special databases developed from a large sample of crash investigations. Project Duration A project of this scope and complexity will likely take at least 3 years, with up to a year for planning and site selection, a year for special data collection, and a year for analyses and reports. Project Cost Given the project duration and the likely need for substantial field data collection and crash investigation, the cost of the project is estimated at $1.5 to $2.0 million. Payoff Potential If this project can be successfully completed, it will have significant payoff, not so much in a tangible measure to directly reduce crashes, but as a sound foundation for identifying the nature of the safety problem and measures that would be effective in reducing crashes and their severity. IS 1b. Accident Causation: Establish the root causes of crashes attributable to driver error. Comment from R&T Partnership Steering Committee There’s no consensus agreement on the value of this project among the reviewers. There are many difficulties in conducting this research (e.g., what method can be used to “get into the driver’s head?”). Also, reviewers are not totally convinced that one has to know the reason why,

62

as long as one can develop a treatment that affects the behavior (e.g., whether a driver runs a signal due to “frustration at home” vs. “risk taking” may not matter as long as a red-light-camera can affect the behavior). So there needs to be some preliminary thinking done about understanding why root causes are important from a behavior-change point of view. What can be done about root causes? If willful risk taking were found to be a significant proportion of root causes, then, before countermeasures are contemplated, one will have to find out why people take risks knowingly. Thus, one will have to search for the root causes of the root causes. Perhaps it was congestion, perhaps the pressures of urban living, perhaps inherent personality traits. (On this question too there is no shortage of research findings). In none of these cases (congestion, urban living,...) it is clear what countermeasure applies; perhaps no countermeasure exists. However, if it is thought that enforcement might work on willful risks, then the question will be how well will is it likely to alter willful risk taking. This last question is one about the expected effectiveness of a countermeasure. Answering the question of expected countermeasure effectiveness is an inherent part of developing any countermeasure. Unfortunately, this is not a part of the proposed causation study. Response from White Paper authors This research statement emerged from the emphasis placed on it at the Irvine Conference. We do recognize and appreciate the issues and concerns raised by the reviewers. It will not be easy to identify what the “root” causes are and how far down the “root” does one try to go! The resolution of this issue should be left to the researchers. As to whether or not it is necessary to know the reason why drivers behave the way they do, we wish to offer the following comments. It is arguable that it is useful to know that driving violations, e.g. running red lights, are attributable to “willful” risk. In other words, possible responses why a driver takes a willful risk (i.e., “hey, I knew it was red, but didn’t want to stop) may include a) I was late, b) it’s just my personality, or c) I just lost my job and was mad that day. Likewise, it is desirable to ascertain if the driver takes an unintended risk, such as “ I didn’t realize it was red”, “ I was daydreaming”, or “ I didn’t see the red until it was too late to stop”. If we can get a handle on this issue, then we can better devise countermeasures, be they enforcement, education or engineering. This is clearly preferred, rather than implementing a countermeasure and subsequently determining if it positively affects the behavior.

Again as stated above, we believe the essence of the researchable issue ought to remain and the specific objectives, procedures and methods should be determined by the group that finalizes the project statement and eventually selects the researchers. Background Driver error is by far the most often-cited crash causation factor cited by the police. There are likely other factors related to the highway, e.g. limited sight distance; the vehicle, e.g. brakes in poor condition; and the environment, e.g. slippery conditions, that may have contributed to the accident, but the driver is typically cited for not being able to control his/her vehicle in spite of these conditions. There are many other crashes where there is no apparent contributing deficiency, and for whatever reason the driver collides with another vehicle or pedestrian in the intersection. In these cases are drivers knowingly taking risks, do they misidentify risks, or are they distracted? The root causes of intersection accidents need to be established, so that the most

63

appropriate type of countermeasure in terms of education, enforcement or engineering can be developed. Problem Statement This project will examine and identify the reasons why drivers make errors at intersections that result in crashes. It will establish why drivers knowingly or unknowingly take risks in negotiating intersections. For example, why do drivers run red lights or stop signs or turn left in front of on-coming traffic? Special consideration should be given to older drivers who are purported to be over-represented in intersection crashes. Method / Approach This project may need to take alternative approaches. One approach will be to conduct in-depth investigations, to include interviewing drivers, of a sufficient sample of crashes. Another approach would be to conduct laboratory and/or field studies to observe driver behavior under various conditions. The laboratory study may involve a driver simulator. While this project has a more specific focus than the first statement, it could be included within the approach of the first statement. Project Duration This project will require at least two years depending upon the types of studies conducted in the laboratory or field. Project Cost It is estimated that this project will cost from $0.5 to $.75 million depending upon the sample size of the crash investigations and the types of laboratory and field studies conducted. Payoff Potential Having a thorough understanding of why drivers take risks and/or make errors in negotiating intersections will have a high payoff for determining what type of countermeasures may be appropriate. IS 2a-1. Traffic & Operational Features: Safety impacts of alternative intersection controls Comment from R&T Partnership Steering Committee This is worthwhile study. The review of current literature to determine what is unknown is important. Two controls seem to be missing in the discussion within the white paper: all-way stop control and roundabout. The results of the recently completed NCHRP 17-16 study “Accident Warrant for Traffic Signals” should be examined. Response from White Paper authors We concur that all-way stop and roundabouts should be included and have made changes to reflect that in our revised version of this white paper.

64

Background Typically, right-of-way and vehicular traffic flow is controlled by either one of four ways: 1) no specific traffic control device and right-of-way is established by vehicle code, 2) yield sign, 3) stop sign including all-way stop, and 4) traffic signal. Lately, roundabouts have become a viable intersection control strategy. In general, the need for each type is based upon traffic volumes and the most efficient control for traffic operations. While safety concerns can dictate which type of control is used, there is no well-defined safety relationship for the various types of control. Intuitively, as traffic volumes increase there are more conflict opportunities that can result in vehicle accidents for each level of control. After a certain level of accidents, a more restrictive control, e.g. stop sign to signal control, may reduce total accidents or accident severity but then will eventually increase with increasing volume. The threshold accident values under the different controls have not been established. Problem Statement The objective of this research is to determine how safety, in terms of vehicular accidents and severity, are affected by the type of intersection control. All types of control should be considered to include no control, yield sign, stop sign, signal control, and roundabouts. Since traffic volume levels will likely be an influencing factor, all levels should be consider and in rural, suburban, and urban settings. Method / Approach This project will require an analysis of crash data for a large sample of intersections with the all four controls. At least 3 years of crash and traffic volume data will be needed; and if not available from data files historically, it will need to be collected over that period in the future. Standard statistical methods need to be employed to develop relationships and predictive models. A desirable product would be to develop threshold values or plots of crash statistics to traffic volumes for different control strategies. The design of this project could be integrated with problem statement #1. Project Duration This project will take at least 3 years and up to 5 years if historical data can not be obtained. Project Cost The project cost is estimated at $1 to $1.5 million to insure an adequate sample size and robust data. Payoff Potential Successful completion of this project would yield more reliable values for determining the appropriate control strategy for intersections for safety. IS 2a-2. Traffic & Operational Features: Determine the relationship between left-turn traffic signal phasing and safety. Comment from R&T Partnership Steering Committee Research on this topic should be explored.

65

Response from White Paper authors It appears that a response is not necessary as there is agreement.

Background Most traffic engineers perceive that protected only left-turn phasing is generally safer compared to protected-permitted left turn phasing. They also perceive that protected-permitted phasing generally has a higher capacity than protected only phasing. Thus, they recognize phasing decisions as a trade-off between safety and capacity. Within many agencies, traffic engineers believe that protected-permitted phasing is “safe enough” and acceptable for use if intersection sight distance for a driver turning left from the left turn lane is adequate and if there are a limited number of opposing lanes. While many traffic engineers spend substantial amounts of time performing level of service calculations to evaluate an intersection, they rarely if ever perform any analysis to estimate the impacts on safety resulting from their decisions about signal phasing. Problem Statement The primary objectives of this study would be to quantify the differences in crash experience between protected-only left turn signal phasing and protected-permitted left turn signal phasing in terms of intersection geometry, posted approach speed limits, traffic volumes, and other factors related to signal control. The secondary objectives would be to quantify the safety effects for other intersection control scenarios related to signal phasing/traffic signal control practices, such as:

• Protected-permitted and protected left turn phasing on opposite approaches • Lead/lag left turn phasing • Leading green for one approach without separate left turn lanes • High type T-intersections with continuous through movement on one approach.

Method / Approach The proposed method for this study would feature two different approaches. The first would be to conduct before-after studies of intersections where left turn signal phasing has been changed. This would result in the development of crash modification functions for the selected signal phasing treatments. The second approach would be to compile data and conduct statistical analysis of a more varied range of intersections to establish relationships between crashes and signal phasing/traffic operation strategies. Project Duration This project could be completed within 24 months if the data were readily available. However, since there is a desire to also include knowledge gained from before and after studies, it is recommended that 36 months be established as the period of performance. Project Cost A project of this magnitude is estimated to cost $300,000.

66

Payoff Potential Reductions in left-turning crashes at signalized intersections would be the payoff. IS 2b. Traffic Control Devices: Determine the effect of different traffic signal layouts on accidents at intersections. Comment from R&T Partnership Steering Committee Several reviewers question the relative size of the problem and feasibility of doing the research. The “most critical” designs in terms of possible confusion should be given consideration but again, there’s a question about the relative size of the issue. Response from White Paper authors We wonder how the reviewers know the size of the problem since we did not uncover directly relevant information that addresses the issue. Hopefully, the question of magnitude of the problem would be answered in part by the first project statement. This topic was raised by others and our own experiences indicate that there is still no nationally accepted consensus on this. Therefore, the topic is still in need of research. We couldn’t find anything that said that such and such layout for a geometric design is preferred. Hopefully such guidance would result from this research. The end result is that the research could lead to more consistency in signal layouts and design among the states and local agencies. The authors are not convinced that this topic should be deleted. Background While the Manual of Uniform Traffic Control Devices contains explicit guidance on traffic signal heads, displays and indications, there is still a wide range of acceptable practices for traffic signal layouts. Consequently, there is a wide variety of configurations involving the types and locations of traffic signal heads that are used throughout the United States. Several human factors studies have suggested that older drivers may be confused by traffic signal displays. Other studies have raised the issue that the complexity and inconsistency of traffic signal displays may contribute to intersection crashes. Moreover, some traffic signal configurations may contribute to red light running. It is recognized that some research studies have investigated the effects of traffic signal displays, although many of them may no longer be applicable since the equipment has changed dramatically (e.g., greater use of LED displays, selected view technologies, changes in mounting hardware, etc.). In addition, a greater range of non-traditional signal configurations and traffic operations practices has been implemented subsequent to those older research studies. Problem Statement The primary objectives of this research would to determine the crash experience at intersections for a full and complete range of traffic signal layouts; to identify those layouts that experience high crash experience; and to develop improved guidelines on the optimal traffic signal layouts for a range of intersection conditions.

67

Method / Approach One approach to address this topic would be to gather data on reported crashes, traffic volumes, intersection geometric conditions and traffic control equipment at a large number of intersections. Then, through statistical analysis, accident relationships could be established as a function of traffic volumes and geometric conditions for a range of different traffic signal layouts. To ensure that the appropriate types of traffic signal displays are investigated, it is suggested that a directed survey of practice be conducted of states, counties, cities and towns as part of this project. Project Duration To accommodate the data collection and analysis effort, it is estimated that three years would be needed for this study. Project Cost A project of this magnitude is estimated to cost $300,000. Payoff Potential If successful this project could yield a more consistent design of traffic signals, thereby reducing possible confusion of the driving public and improved adherence. IS 2c. Design Features: Establish the relationship of intersection sight distance to occurrence of accidents. Comment from R&T Partnership Steering Committee Reviewers feel this is an important research topic and feasible if targeted to two-way stop controlled intersections. Also, some reviewers suggest more thought needed on details of problem (e.g., sight triangle vs. nearby curves and grades) and on research method to be used. Response from White Paper authors It was intended that the two-way stop controlled intersection would be the primary intersection type to study, and that sight distance would be one of several factors to be investigated although it may not have been explicitly stated in the earlier draft white paper. Sight distance (corner sight triangle) is critical for Yield and no-control intersections, but these intersections have low volumes and/or low speeds and would be more difficult to study. We will address the “more thought” comment in the edited version. Background Adequate sight distance for the driver to see objects in the roadway and various traffic control devices is a basic design element for all road types and locations. It is important for intersections, where motorist must be able to the see the intersection as they approach and then be able to see other motorist who may be in conflict as they proceed into the intersection if turning left or right or proceeding straight through. It is especially critical for non-signalized intersections where right-of-way is not controlled by a signal and motorist must make decisions about the presence of gaps in the traffic stream to allow safe maneuvers through the intersection.

68

Problem Statement This project will examine the relationship of intersection sight distance to the occurrence of accidents. Sight distance to the intersection along the approaches as well as the corner sight triangles should be included with emphasis placed on higher speed roadways. Method / Approach The standard research method for developing a relationship of a design element to safety, i.e. accident occurrence, is to develop an accident predictive model based on the experiences of a large sample of locations with variations in the primary influencing variable—for this case intersection sight distances—while controlling or otherwise accounting for other variables to include traffic volume, speed, grade, etc. It is reasonable to hypothesize that as corner sight distance decreases from some threshold value, that accident occurrence will increase given certain set of conditions, e.g. volume level. This method may prove to be too difficult and costly to be practical. It will require measuring sight distance for each corner in both directions and then identifying accidents attributable to the conflicting vehicle scenarios. Traffic volumes at the turning movement level will be needed as well. Alternatively, and even as a supplement to the above approach, accident reconstruction and forensics of a smaller sample of crashes could be conducted. The purpose of this method would be to better establish if sight distance limitations had a causative or contributing effect to the accident. Project Duration The complexity of the research method will require a study period of 2 to 3 years. Adequate time is needed to develop a thorough research design, locate sites, obtain the necessary data, and conduct the various analyses. Project Cost It is estimated that this project will cost at least $500,000. Payoff Potential A better understanding of the role that intersection sight distance plays in accident occurrence will help designers determine the safest design and allow traffic engineers to identify problematic intersection based on deficient sight distances. IS 3. Effectiveness of Countermeasures: Determine the effectiveness of various countermeasures for reducing accidents at intersections. Comment from R&T Partnership Steering Committee All agree good evaluations are needed. Studies need to be prioritized, but not just by a practitioner survey. Priority should be based to a large extent on the size of the problem, expected effect, and feasibility of study. Response from White Paper authors This is a good point, and we have incorporated this topic into the revised version.

69

Background Few definitive relationships have been established linking specific treatments implemented at intersections and the resulting crash experience. While all State transportation departments and many local agencies have highway safety improvement programs, practitioners still struggle with the selection of appropriate safety improvements at intersections. Without an adequate knowledge base, practitioners frequently make less than optimal decisions. They also cannot convince decision makers to expend additional resources for needed safety improvements. Moreover, they cannot accurately assess the impacts on safety of a wide variety of geometric and traffic control alternatives for intersection projects. Problem Statement The primary objectives of this study to develop improved and enhanced crash modification functions (e.g., accident reduction factors) for intersection treatments. Subject to funding and data availability, crash modification functions are desired for a extensive list of treatments likely to include LED signals, red light strobes for traffic signals, internally illuminated street name signs, highly reflective street name signs with large letters, offset left turn lanes, right turn acceleration lanes, bypass lanes on shoulders at T-intersections, rumble strips on approaches to intersections, supplemental stop signs mounted over the roadway, flashing beacons mounted on STOP sign structures, and others that would be identified through this project. Method / Approach The approach would feature three phases. Phase 1 would involve identifying the countermeasures and establishing a priority for conducting further evaluations. Priority could be based on the size of the problem—which could come from the first project statement, the expected effect, the feasibility of the necessary study, and input from practioners gleaned from a survey.. Phase 2 would focus on assembling and assessing the best available information on crash reduction factors in the literature and state practice. Phase 3 would to re-analyze base data and conducted new analyses to derive improved accident modification functions for selected countermeasures. Project Duration To properly conduct this study and investigate a broad range of countermeasures, it is recommended that the study be conducted in multiple phases. Phase 1 could be completed in 12 months. Phase 2 would require an addition 18 to 24 months. Phase 3 is estimated to take 5 years. Project Cost A project of this magnitude is estimated to cost $2,000,000. Payoff Potential The payoff could be incremental reductions in crashes at improved intersection sites resulting from better decisions with respect to the selection of countermeasures at a local and state level.

70

IS 4. Advanced Technology: Determine Effectiveness of and Driver Response to an Automatic All-Red Signal Extension System Comment from R&T Partnership Steering Committee Some of the wording of the problem statement is confusing (E.g., 10% percent of red light crashes occur after the first second of the violator’s red phase” Does this imply that 90% of red light crashes involve a violation during the first second of the red?) The data for this study can come from red light running cameras. Research on this topic should be coordinated with ITS research regarding systems that warn drivers who are about to violate a traffic signal or alert them to potential collisions with traffic signal violators. Response from White Paper authors We have addressed these comments in the revised version. Background Red light running is a major cause of intersection crashes. Yet, studies have shown that 90% of all red light running violations occur in the first second of the red signal indication, which may or may not be during the all-red interval, , where no opportunity to collide with another vehicle exists. If the time for the all-red light could be extended for those motorist who are likely to “run the red light”, which would provide an extra “grace” period, then the probability of a crash with side traffic entering the intersection could be reduced. Modification of current violation detection technology and algorithms (similar to those used in red light cameras) may be used to determine violators a few seconds prior to a light turning red. These few seconds of advanced warning can provide an automatic extension of the all red signal to prevent cross traffic from entering the intersection. Problem Statement Automatic extension of the all-red phase by a couple of seconds for violators detected early can catch motorists. A major drawback is that automatic red light extension can only be used for a couple of seconds after cross traffic is programmed to receive a green ball and cannot be used to prevent crashes due to mid-cycle red light violations. Examining exact violation times will show the effectiveness of automatic red light extension as a crash prevention method. Method / Approach A study of the effectiveness would involve two measures: violations and crashes. Violation frequencies could efficiently be determined at sites with red light camera systems or special video apparatus could be installed to collect this measure. Of concern is motorist adaption to red light extensions systems and the possibility of taking advantage of the extra time knowing that it is operating at a given intersection. Crash analysis using before/after with an appropriate control or reference group should be pursued to determine the crash changes. Project Duration

71

If the measure of effectiveness was limited to violation changes, then this study could be completed within a year. Assuming that the evaluation should include crash analysis then, depending upon the analysis method employed, the project could take up to three years or more. Project Cost Expected costs depend upon the effectiveness measures used. For the violation measure of effectiveness, this project could likely be conducted for less than $100,000. The crash analysis would increase the estimated cost by at least another $100,000.

3.3 Summary

This paper has identified numerous research needs to address the gaps in knowledge of intersection safety. Starting from a long list of needs culled from various sources that individually have identified research needs, those that were felt to be the more critical and for which there is no known research project planned were identified and categorized. All of them are worthy of attention and successful conduct will offer solutions and strategies to achieve the goal of improving intersection safety when implemented. However, not all projects can be pursed at once, and perhaps nor should they be. The highest priority should be to develop a comprehensive research program that would identify the magnitude, characteristics and causation of accidents at intersections. A properly designed program will likely involve a series of specific projects that will answer two fundamental inter-related questions such as:

• What is the magnitude and characteristics of accidents at intersections? • What part does the driver, vehicle, roadway, and the environment play

individually and collectively in accident causation? Once these two basic questions (and variations to them) are answered, then future research can address the specific factors that cause crashes at intersections and determine appropriate countermeasures or changes in design and operations of intersections.

72

3.4 Appendix A

FULL LIST OF INTERSECTION SAFETY RESEARCH NEEDS BY CATEGORY ACCIDENT QUANTIFICTION, CHARACTERIZATION, AND CAUSATION

1. Comprehensive and systematic analysis of accidents at intersection to determine magnitude, characteristics, and causation factors that would identify priority problems and potential countermeasures.

2. Establish the root causes of intersection accidents attributable to driver error. For example, why do drivers expose themselves to crashes or what is the mental breakdown that e.g. allows a person to drive into the path of speeding truck; improve understanding of the underlying causes of driver distraction and error at intersections. Why do drivers take risks at intersections—why do drivers run red lights or stop signs or turn left in front of on-coming traffic. Special consideration given to older drivers who are purported to be over-represented in intersection crashes.

3. How does the culture of road user behavior evolve and how can it be influenced? 4. Does driver behavior (speeding, aggressive driving) affect the collision risk of

intersection maneuvers? 5. Does information overload exist at intersections and does it contribute to accident

causation? 6. What is the role of inattention in collision risk at intersections? 7. Does the relative risk of different intersection maneuvers vary with driver age and

gender? 8. How does the pattern of conflicts and collision vary with traffic volume? Is there a

safety relationship to intersection capacity as defined by level-of-service or volume-to-capacity ratio.

9. Why drivers sometimes confuse some two-way stops as all-way stops. Are there visual queues that lead them this? Are there visual queues that can be implemented to lessen confusion? What are the effective applications of using the “Cross-Traffic Does Not Stop” sign?

RELATIONSHIP OF TRAFFIC AND OPERATIONAL FEATURES TO SAFETY

1. Safety impacts of no control vs. yield control vs. stop sign control vs. signal control. Safety effects associated with traffic flow conditions; occurrence of accidents as a function of traffic volumes.

2. Safety impacts of alternative signal control strategies: a) fixed-time, semi-actuated, fully-actuated controllers and b) free vs. coordination. How does adaptive signal timing (e. g. SCATS, SCOOT, etc.) affect safety?

3. Safety impacts of alternative signal timing strategies: a) short vs. medium vs. long cycle lengths and b) inappropriate/inefficient signal timing effects on safety.

4. Safety effectiveness of dilemma elimination systems for intersections (i.e. extension of green for vehicles that enter dilemma zone).

5. Safety impact of clearance intervals for trucks at signalized intersections.

73

6. Safety effects of late-night flash mode 7. Effects of protected/permissive left-turn signal phasing. Effects of protected left-turn

phasing at high-speed rural intersections. 8. Determine safety impacts and countermeasures of stopped or parked vehicles to include

urban goods movement and vehicular parking. Transit issues such as bus stop location would be included as well.

9. Safety impact of right-turn-on-red on vehicular and pedestrian traffic. 10. Safety implications of bus signal priority systems.

RELATIONSHIP OF TRAFFIC CONTROL DEVICES TO SAFETY

1. (Dis)Benefits of different traffic signal layouts—a study that would lead to a standard or guideline for consistent traffic signal layout. Tie in with improving signal visibility. How would this reduce crashes?

RELATIONSHIP OF DESIGN FEATURES TO SAFETY

1. Cross-sectional elements at intersections to include lane width and median width. 2. How do turn lanes change the pattern of conflicts at an intersection? 3. Intersection sight distance. 4. Safety effects of transitional elements moving from corridors to intersection approaches. 5. Safety effectiveness of right-turn acceleration lanes. 6. Safety effectiveness of offset right-turn lanes. 7. Safety effectiveness of left-turn acceleration lanes at divided highway intersections. 8. Safety comparison of four-legged intersections vis-à-vis two T-intersections 9. Safety performance of roundabouts and especially regarding pedestrians. 10. Safety effects of alternative intersection types and especially innovative and non-standard

designs such as jughandles, median U-turns, continuous flow intersections, etc. EFFECTIVENESS OF COUNTERMEASURES General

1. Control vehicle speed thru intersections using combination of practices such as speed tables, pavement markings, automated photo enforcement and changeable-message signs.

Traffic Control Devices

1. Flashing beacons, such as those mounted on Stop sign structures, or Stop Ahead signs, or with overhead intersection beacons; flashing red beacons activated by approaching cross-street through traffic; or flashing yellow warning beacons on the through road activated by side-street traffic.

2. Pedestrian Count-Down Signals 3. Roadside markers or pavement markings to assist drivers in judging the suitability of

available gaps for making turning and crossing maneuvers 4. Determine effectiveness of various countermeasures esp. those in “Engineering

Countermeasures for Red Light Running”

74

5. Safety effects of different sized signal heads (8” vs. 12”) 6. Safety effects of providing signal per approach lane 7. Safety effects of signal placement—overhead vs. nearside. 8. Red light strobe evaluation. 9. Larger regulatory and warning signs at intersections 10. Rumble strips on approaches 11. Supplemental stop signs mounted over the roadway 12. Supplemental messages on pavement such as STOP AHEAD 13. Benefits of using back lit and disappearing legend warning signs at intersections. 14. Comparison of internally illuminated and highly reflective large street name signs. Is one

better than other and cost-effectiveness. 15. LED signals

Traffic Control Operations

1. Effects of closed-loop signal systems 2. Safety effect of detector placement and signal timing parameters for semi-actuated

signals

Geometric changes

1. Longer left turn lanes 2. Offset left-turn lanes 3. Bypass lanes on shoulders at T-intersections 4. Left-turn acceleration lanes at divided highway intersections 5. Longer right turn lanes 6. Offset right turn lanes 7. Right turn acceleration lanes 8. Full width paved shoulders in intersection areas 9. Innovative intersection designs (geometric or traffic control devices) in areas (highly

urbanized) where right-of-way is limited (where desirable design standards can not be used.) Indirect left-turn treatments to minimize conflicts at divided highway intersections

NEW TECHNOLOGY APPLICATIONS

1. Automated real-time system to inform drivers of the suitability of available gaps for making turning and crossing maneuvers.

2. How does transit signal priority affect intersection safety? 3. Enhance dilemma-zone detection at high-speed rural intersections using MOVA,

LHOVRA, and similar technologies 4. Dynamic (actuated) clearance/all-red intervals.

ENFORCEMENT

1. Attitude of Law Enforcement Community towards highway safety 2. Analytical tools to identify which intersections to provide selective enforcement

75

3. Effectiveness (short and long term) of targeted enforcement to reduce stop sign and signal violations.

SAFETY PROGRAMS

1. Evaluation of Intersection Improvement Program, e.g. Michigan’s AAA. 2. More timely process for implementing safety improvements. 3. Safety audit procedure 4. Develop strategies to identify potentially hazardous intersections, rather than wait on

occurrence. 5. Analytical tools/models for traffic engineers and planners to consider the safety

consequences of intersection safety and design; Safety as an explicit item in design. PEDESTRIAN AND OTHER NON-VEHICLE USERS

1. Guidelines for improving pedestrian safety at signalized intersections using strategies such as PUFFIN and TOUCAN crossings, countdown indicators, and audible pedestrian signals.

2. Safety benefits of accessible/audible pedestrian signals. 3. What are the most cost-effective strategies and measures for accommodating

handicapped?

3.5 Appendix B Listing of titles of research projects related to Intersection Safety proposed for F-SHRP Topic 2-1: Research Tools and Methods

1. Legal and Privacy Issues in Recruiting Volunteer Drivers and Vehicles for Field Studies of Driving Safety

2. Development of Analysis Methods for Site-Based Risk Studies Using Recent Data 3. Development of Analysis Methods for Vehicle-Based Risk Studies Using Recent Data 4. Development of Comprehensive Roadway Information in a GIS Database 5. Application of OEM Electronic Data Recorders for Risk Studies

Topic 2-2: Risk Studies

1. Instrumented-Vehicle Risk Study—Phase I: Study Design 2. Instrumented-Vehicle Risk Study—Phase II: Pilot Study 3. Instrumented-Vehicle Risk Study—Phase III: Field Study 4. Instrumented-Vehicle Risk Study—Phase IV: Intersection Analysis and Countermeasure

Implications 5. Site-Based Risk Study—Phase I: Study Design 6. Site-Based Risk Study—Phase II: Field Study 7. Site-Based Risk Study—Phase III: Analysis and Countermeasure Implications

76

Topic 2-3: Countermeasure Evaluation

1. Identify Countermeasure Evaluation Topics 2. Retrospective Countermeasure Evaluation Projects

3.6 Appendix C Listing of Titles of Research Project Statements Developed For FHWA Contract DTFH61-98-C-00075, Guidelines for Signalized Intersections

1. Development of Guidelines for Planning, Implementation and Operation of Innovative Configurations for At-Grade, High-Volume Signalized Intersections

2. Accommodation of Turning Traffic at High-Volume Signalized Intersections

3. Improving Safety and Efficiency During the Change and Clearance Intervals at High-

Volume Signalized Intersections

4. Accommodation of Pedestrians and Bicycles at High-Volume Signalized Intersections

77

4.0 Human Factors

4.1 Introduction It is difficult to estimate the impact of human factors on driver safety because existing accident databases depend upon a police officer having sufficient time and motivation to carefully complete accident forms. Thus, certain factors may be unobserved or unrecorded. Therefore, such safety estimates are conservative and may underestimate the impact of human factors. A recent report on highway safety (GAO, 2003 Figure 5) stated that GAO analysis of NHTSA data from 1997-2001 found about 2.5 million drivers of passenger vehicles that were towed away from crashes attributed to driver inattentiveness. Yet driver inattention is only one of several human factors issues that contribute to crashes. UMTRI conducted a univariate analysis of FARS data from 2000-2002 classified as Driver Related Factors; Codes 61-88, which refer to items beyond the driver’s control such as weather and visibility were omitted from the analysis which covered the remaining codes as Factors 1, or 2, or 3 or 4. The number of cases found was 104,921 out of a total of 173,308. Thus, over the three-year period, 60.5% of coded fatal accidents involved human factors. Of course, these data considered alone, are not likely to provide effective countermeasures (see Figure 4.6.2). This requires research and building models of driver performance that are described in this white paper.

4.2 Critical Future Highway Issues There is, of course, an element of risk in specifying critical future issues. A panel of experts given this charge 100 years ago might have determined that the most critical transportation issue of the future would be removing vast piles of horse manure from city streets. Technology can produce sudden changes that dramatically alter our perspective about the future. Nevertheless, there are certain constants in highway safety. Vehicles will change. Roadways will change. But human drivers will be controlling vehicles for the indefinite future. Humans have not changed. In some ways, this is unfortunate, for humans did not evolve to control machines that provide greater speed and greater weight than the unaided human. People lack the muscle power and perceptual skills to directly control high-speed vehicles. By the time a bullet-train engineer sees a stop signal on the track ahead, it is too late to halt the train. Airplane pilots have even greater control issues. But these transportation systems work reliably because of equipment and procedures that take human limitations into account. Much of these safeguards are absent in ground surface transportation. Operator training and licensing requirements are quite minimal, compared to air transportation. Fitness for duty is not evaluated a priori for car drivers, who may be fatigued or under the influence of alcohol or drugs. At first blush, the world of flight should appear to be far more dangerous than travel on the ground. Airplanes cannot pull over to the side of the road to solve a problem. Yet flying is far safer than driving. The risk of driving 18 km equals the risk of a non-stop continental flight (Sivak, 2003). Familiarity breeds contempt and the car is so familiar that most drivers are unaware of the risk they assume behind the wheel.

78

From a human factors perspective, the most critical future issue is automation. Cars and trucks are starting to look more and more like airplanes with new in-vehicle systems that are proliferating at an accelerating rate. These telematic and driver assistance systems are already changing the function of the driver. In the future, the ground vehicle operator will become more a monitor of vehicle systems that will have the ability to take over control of his vehicle. Automated cruise control will slow the vehicle if headway is too low. Crash avoidance systems will apply the brakes and override inappropriate driver tendencies, such as lifting the foot from the brake pedal when ABS causes pedal vibration. Radar and additional sensors will detect other vehicles and pedestrians and offer enhanced driver vision at night. All these technologies aim to overcome the limitations of the human driver who was designed to control his own locomotion at low speed. Cars will offer enhanced convenience and entertainment for drivers who can read e-mail, receive faxes, talk on cell phones, surf the internet, and place orders with their stock brokers from a moving vehicle. Automation is not always an innate good. Indeed, the National Highway R&T Partnership has selected the human-machine interface in light-duty vehicles as a research area for emphasis within the Light-Duty Vehicle-Safety theme (Table 2, National Highway R&T Partnership, 2002). Clumsy automation, as has been studied in aviation, can decrease vehicle safety. More in-vehicle devices can equate to more driver distraction. While each individual system offers some potential for increased safety or convenience, the effects of many such systems in a single vehicle are unknown. Will the new advanced vehicles be safer or more dangerous than current vehicles?

4.3 Human Factors Research Topics Table 4.3.1 lists the specific research topics discussed in this section and summarizes their characteristics. Projects are listed according to their potential to yield important results as evaluated by the Human Factors Working Group. Project parameters are based primarily on estimates from the Working Group. The goal of this whitepaper is to elaborate the projects specified in Table 4.3.1. This is accomplished in three steps:

1. Review each project for technical feasibility. 2. Add new projects, if any major topics were omitted. 3. Explain briefly, within the 15-page white paper limit, how each project might be

conducted. Note that the author of this white paper does not view this chapter as an opportunity to substitute his own judgment for that of the working group. His charge is to refine the suggested topics, rather than to discard the topics and start anew.

79

Category Project Title Type of

Research Likelihood of success

Duration (Months)

Cost ($1,000,000)

Human Factors Cognitive Models

HF 1a: ComputationalDriver Model: WE (Whole Enchilada)

Advanced 3.5 144 12

HF 1b: Computational Driver Model: Light

Advanced 4.0 60 5

Human Factors Information Overload

HF 2: Processing Multiple Sources Of Information

Advanced 4.0 60 10

Human Factors Speed Control

HF 3: Understanding Speed Selection

Applied 3.0 48 8

Human Factors Perception/Attention

HF 4: Look but not see

Applied 2.5 36 4

Human Factors Basis for Design Standards

HF 5: Design Driver

Applied 4.5 18 0.3

Human Factors Decision Rationality

HF 6: Risk Homeostasis

Applied 2.5 28 1

Human Factors Simulator Generalization

HF 7: Driving Simulator Validity

Applied Methodology

4.5 42 4.5

Table 4.3.1 Prioritized list of research projects. Likelihood of success ranges from Very Low (1) to Very High (5).

4.4 Knowledge Strongholds Knowledge strongholds are areas where there has been substantial research yielding decreased need for future research. They arise in two ways. First, a research area may be fully mature so that additional effort yields diminishing returns. Second, a research area may be currently popular and vigorous so that the need for future research is either filled or cannot be fully

80

evaluated until research already in progress has been completed. In the human factors domain, the first instance is denoted by Multi-Disciplinary Accident Investigation (MDAI) while the second instance is represented by current work on the naturalistic observation of on-road traffic using instrumented vehicles. MDAI The landmark MDAI study was conducted over twenty-five years ago in Monroe County Indiana. This tri-level study used teams of experts to investigate accidents and suggest possible countermeasures (NHTSA DOT-HS-034-3-535, 1975). The recognition afforded this study occasionally leads to calls for a contemporary follow-up study, perhaps involving several counties geographically disbursed across the United States. However, subsequent evaluations of the methodology of this important study revealed significant limitations that arise from the subjective nature of expert opinion. Replication is the essence of science. It is impossible to suggest effective countermeasures for results that cannot be duplicated. One subsequent evaluation (Haight et al, 1976) of the tri-level study was unable to replicate findings using the same taxonomy:

Along this line, as part of the MDAI editing and coding efforts at HSRI, an attempt was made to apply the Indiana taxonomy to a small number of in-depth accident cases. Three experienced editors were given a copy of the taxonomy and asked to record the causative factors…it was exceedingly difficult to get consistent causation reporting by the three editors –particularly in the human factors area [emphasis added p. 78].

This same report also criticized the tri-level study, among others, for issuing careful caveats about potential conclusions at the start of the report and then ignoring these caveats at the end of the report when offering conclusions that could not be supported by proper statistical analysis. When observational data have known faults, it is tricky to avoid the temptation to “suggest” when one cannot prove. It is quite difficult, and perhaps impossible (Kantowitz, Roediger & Elmes, 2001 chapter 2), to make inferences about causality from even objective observational data. When the data are subjective opinions, rigorous conclusions that meet established scientific standards are unlikely. Thus, while the questions raised by the tri-level study still are vital today, the methodology used has asymptoted and little will be gained by repeating the same methodology again. Therefore, a new MDAI study is not proposed later in this document, even though this topic was recommended, albeit with the lowest ranking, in the Irvine Safety Research Agenda Planning Conference. Naturalistic Observation From Instrumented Vehicles (NOFIV) Fatal accidents are very low-frequency events. Therefore, one approach to aggregate data is to deploy large fleets of instrumented vehicles equipped to record objective data in hopes of capturing real-life accident numbers with NOFIV technology. While there are several NOFIV studies in progress, e.g., UMTRI Automotive Collision Avoidance System, UMTRI Road

81

Departure Crash Warning System, Virginia Tech Naturalistic Driving Study, a recent F-SHRP Safety Plan (Campbell et al, 2003, p. 17) noted that

a large, nationally representative number of crashes is not presently achievable with a fleet of instrumented vehicles. That would require a single project on the order of $500M to achieve a test fleet of 100,000 vehicle-years that would be expected to produce 6,000 police-reported crashes.

Furthermore, our ability to collect NOFIV data vastly exceeds our ability to analyze and understand it. Techniques to parse and analyze the large data sets provided by NOFIV are being developed, but this is a complex task, especially if video data are also considered. Therefore, no new NOFIV studies are proposed in this document.

4.5 Knowledge Gaps Six of the seven knowledge gaps listed here are based upon recommendations of the Human Factors Breakout Group at the Irvine Planning Conference. The seventh gap replaces the MDAI gap identified by this Group. The gaps are listed in the priority establish by the Breakout Group:

1. Understanding the driver. The driver is the major cause of accidents and fatalities. Until we understand the driver our ability to propose effective countermeasures is limited.

2. Driver information overload. This is a key instance of the previous gap that is so important it justifies its own listing. Drivers are overwhelmed by vast amounts of information both within and without the vehicle. New ITS developments can provide sufficient information to exceed the processing capacity of the driver while the vehicle is parked.

3. Speed out of control. Excessive speed is a major contributor to run-off-road and other accidents.

4. See but not understand. Human perception is a complex process. Looking at a stimulus does not guaranty it will be processed correctly or that an appropriate timely response will be made by the driver.

5. Human factors basis for design driver. As demographics and vehicle mix changes, the target of highway design must also be updated.

6. Is theory of risk homeostasis valid? This important theory explains why accidents rates do not improve when better technology is added to the vehicle or to the highway. Is it the best explanation?

7. Driving simulator validation. The driving simulator is a key tool used to investigate situations that are too dangerous for road testing or too expensive to create in the world (e.g., building a new tunnel or highway for test purposes). But until proper validation tools are built, designers will not accept simulator results.

The human factors of highway safety is best characterized as an empirical enterprise where data are collected to meet very specific needs. Relatively little effort has gone into building models of driver behavior. Our ability to collect data vastly exceeds our ability to understand these data in a systematic, coherent way. There is an urgent need to support more basic research efforts

82

targeted at understanding driver behavior. It is inefficient to perform an experiment to answer every question. A good model of driver behavior can answer many different questions. There is a vital need to invest research dollars in improving methodology. The best example of this need concerns driving simulators. In the past, it has been difficult to obtain funding to validate simulators and to compare different models and characteristics of advanced simulators. Decades of research in areas such as nuclear power and aviation have shown that very expensive simulators with full physical fidelity are not required to answer many important questions. The seven research projects discussed in the next section all seek to improve our understanding of some aspect of driver behavior related to road safety. While the Advanced projects clearly imply basic research efforts, it is important to realize that the Applied projects as well can and should be conducted so as to not only gather salient data, but to also advance basic understanding of the cognitive processes that control and modulate driver behavior. Comment from R&T Partnership Steering Committee The reviewers raised many questions about whether knowing the prevalence of human errors will lead to countermeasure development. It is one of the basic tenets of ergonomics that humans will make errors and the task is to make the environment of the human operator such that errors are less frequent and its consequences less harsh. Thus, if more research is needed on human factors, it needs to be well justified in terms of what the outcome is expected to be in terms of new treatments.

The paper criticizes multi-disciplinary accident investigation (MDAI) methodology. This contrasts to some extent with the recent implementation of a national MDAI-type study on truck crash causation and a proposed NHTSA large study. Should the profession be looking at better ways to do MDAI studies? Response from White Paper authors The reviewers pose a vital question: How are good countermeasures created? Much of the following answer is taken, almost verbatim in some parts, from Sivak (2002). Many early safety countermeasures could be viewed as based on common sense. No fancy theories or models were required. But as vehicles and roads became more complex, common sense countermeasures have been supplemented or replaced by better countermeasures that required going beyond common sense. An example of a common sense based countermeasure is the dual-beam headlighting system with high beams for distance viewing when there is no oncoming traffic. But low beams lack sufficient illumination for safe driving at speeds above 70km/h and high beams create glare for oncoming traffic. This creates a human factors conflict between maximizing visibility and minimizing glare that has not been solved using common sense. A noncommon sense solution might involve use of polarized headlighting. This requires a theoretical knowledge of physics and human vision to be successful. Table 4.5.1 gives other examples.

83

Common-sense countermeasure Science-based countermeasure Low beam and high beam headlighting Polarized headlighting Getting rid of sharp fins Pedestrian airbags Minimum windshield transmittance See-through A pillars Dash-mounted displays Head-up displays No nearby trees and poles Collapsible poles and channeling guardrails Visual screening Attentional Screening

Table 4.5.1 Examples of common-sense and science-based countermeasures (Sivak, 2002)

Another more quantitative example comes from the work of Palmer and Kantowitz (1994) on daytime running lights and turn signals. Daytime running lights offer some potential safety advantages but have the disadvantage of making it more difficult for drivers to perceive turn signals that are located adjacent to headlamps. This reduction in visibility is called visual masking. Industry standards exist for increasing turn-signal intensity as turn signals are moved closer to headlamps. What should the intensity be for turn signals adjacent to reduced intensity high-beam daytime running lights (5000cd)? How should turn-signal intensity be adjusted as a function of the separation between turn signals and headlamps? The simplest and most direct way to answer these questions at first blush appears to be a series of experiments that varies separation, headlamp intensity, ambient illumination, and turn-signal intensity. If enough data are collected, nomographs can be constructed to be used by design engineers. If available resources do not permit collection of all the data needed, the most important combinations can be sampled and a regression equation can be used to fill in the missing points. So when NHTSA asked us to answer this question, it anticipated a series of experiments. Instead, we found a better solution (Palmer & Kantowitz, 1994) that was more cost effective – NHTSA did not have a lot of money for this project – and more useful. We started by building a theoretical model of masking. The model was designed to predict the intensity of a turn signal that yields a fixed level of performance as a function of other relevant variables such as separation. For example, the model can be used to estimate the intensity that makes a turn signal recognizable 90% of the time as a function of separation. The heart of the model is an analysis of glare whereby light scattered from the headlamp reduces the contrast for nearby lights. Of course, the model must also include other factors of practical interest such as viewing distance, ambient illumination, and the size and shape of the lights. To make sure that we picked reasonable separation values, we did a quick survey of vehicle lights and found that the Ford Explorer was a good worst case for proximity of turn signals and headlamps. Figure 4.5.1 shows part of the model predictions that can be used to answer questions of the following form: take a square daytime running light 10 cm on a side abutted by a turn signal 5 cm wide and 10 cm high. This gives a separation (measured from the lamp edge to the center of the turn signal) of 2.5 cm. How much would the detectability of the signal be improved if separation was increased from 5 to 10 cm? Figure 4.5.1 gives an answer of 75% decrease in masking for nighttime (lower curve).

84

Figure 4.5.1 Model predictions for relative threshold. The upper curve is daylight (10,000 lx), and the lower curve is for no ambient illumination. (From Palmer and Kantowitz, 1994) This countermeasure goes well beyond common sense. Indeed, common sense might not even realize that daytime running lights can make it more difficult to perceive turn signals. The theoretical model provides a set of equations to the headlamp designer that elegantly presents the trade-offs between separation and visibility. It allows the countermeasure of spacing headlamps and turn signals to avoid excessive visual masking. A final example used by Sivak (2002) to demonstrate that common sense countermeasure can decrease safety is based upon a theory of vision applied to driving by Leibowitz and Owens (1977). The theory states that focal (central) vision is used to detect objects, like pedestrians, while ambient (peripheral) vision is used for lane keeping. Common sense implies that better reflective lane markers would improve night driving. But the theory predicts that improvement in lane keeping would not be accompanied by a similar improvement in focal vision. Drivers confidently relying on ambient vision to remain in their lane could now drive faster, making it more likely to strike objects in the road not detected in time by focal vision. It is not random that the examples above are all drawn from vision. Basic research on vision is well developed and many theories and models of vision have been formulated and tested. Thus, it is easier to relate countermeasures to driver vision than to other aspects of the driver-roadway interaction. Trying to specify what new treatments might be developed from new theories and models presents a chicken and egg conundrum. Absent the theories, it is almost impossible to specify what new countermeasures might emerge. Thus, what is a standard operating procedure for many human factors researchers (using models), might require an act of faith from practicing highway engineers who do not normally invoke theories of human behavior. The best arguments for human factors models producing effective new countermeasures are inductive, based upon

85

examples such as were presented above. There are entire handbooks of human factors presenting hundreds of such examples in many different content areas. If aviation, nuclear power, and human-computer interaction can create better countermeasures through models, so can driving. The criticisms of MDAI are of the same form as the examples given previously: common sense versus science. MDAI studies represent common sense. They are not replicable and do not meet the most basic of scientific criteria. Their time has passed. While there may be strong political reasons for funding future MDAI studies, there is no scientific justification for such work.

4.6 Research Topics

HF 1: Computational Driver Model - Category: Human Factors Cognitive Models Comment from R&T Partnership Steering Committee The proposed research is very high risk. It needs careful consideration and justification. The authors mention two different approaches (large and semi-large). More details are needed on how the “best model” will be chosen in the “large model shoot-off”, and second, how do we decide if the large project will be twice as good as the smaller one? With respect to the former, even though there is some mention of the use of test scenarios for model evaluation, there is not enough detail to convince certain reviewers this is a valid methodology. Has it been used before? What is the nature of the alternative scenarios – are they individual-driver oriented or traffic-flow oriented? Will any be based on crash outcomes? More importantly, questions were raised by the reviewers concerning whether model of individual driver can be done or is sufficient since: 1) the profession has no accepted theory of “driver adaptation” to changes, and if adaptation occurs, developing a model that can “adapt” seems very difficult, and 2) doesn’t “group behavior” matter in traffic (as opposed to “individual behavior”)? The way we behave on the road as drivers or pedestrians is clearly a reflection not only of individual capabilities and inclinations but also of a norm of behavior that evolves through emulation and imitation and may be characteristic of a city, a region and a country. Were it not so, people in New York and in Minneapolis would behave in the same way as people in Naples or Hamburg do. Indeed, the aim of important countermeasures is to affect something like ‘the culture of driving’ or the ‘distribution of speeds’.

Basic conclusion -- This proposed research needs further assessment by international Human Factors experts who have attempted this and who have no stake in the United States program. Response from White Paper authors Further review by international experts, especially from Scandinavia where the FHWA-AASHTO scan tour will be spending substantial amounts of time, is an excellent idea. The main difference between the large and semi-large versions of model development is the scope of the model. Existing human factors driver models have very limited scope. Computational models that might be borrowed and extended have medium scope. If the desired

86

scope is large, then the large version might be worth more than twice the cost of the semi-large model. For example, if it is important to include something like the ‘culture of driving’ then a large new model would be required. Existing models tend to focus upon the behavior of single individuals. While there may be some existing computational models of group behavior (which would require some research to uncover), they most likely have not been combined with the kinds of human performance models most applicable to ‘plain old driving’. The kinds of experimental details requested by the Steering Committee go beyond the scope of this effort (see chapter 1) which is aimed at selecting specific topics and not aimed at generating a detailed statement of work (SOW) for each possibility. The cooperation of several stakeholders in defining key issues for the SOW would be essential. For a long-term basic research project such as this, there should be a panel of external experts to review and help guide the work on a continuing basis. These details would need to be elaborated in a full SOW. However, here are some possible ways to approach the issues raised. The best model in a shoot-off is one that most accurately predicts and explains driver behavior for a set of specific tasks, such as run-off road, gap acceptance in passing, etc. Each model would receive a score for each task. Models would have to score above a threshold for each task: a model cannot make up for inaccuracy in a key task by high accuracy in a different task. Extra points would be given for models that accurately explained driver behavior for tasks not used to generate the model initially. Alternative scenarios would be created based on input from the expert panel. For example, the $6 million FHWA ATIS/CVO project had to create a set of scenarios for use of in-vehicle telematic devices. These scenarios were reviewed by an industry panel to ensure that vital aspects of telematics were properly considered. Another name for adaptation is learning. Psychology, as a discipline, has over a hundred years of experience in creating valid models of learning that have been successfully applied in military, commercial and educational settings. Developing an adaptive learning model does not pose unusual technical difficulties. However, it does require a sustained long-term basic research effort that has not yet been attempted by the U.S. Department of Transportation. Narrative Description This is the highest priority project considered by the Human Factors Working Group. It is a high-risk project with a huge potential payoff. Theory is the best practical human factors tool. It offers several advantages to the user who must solve an immediate pragmatic problem (Kantowitz, 2000):

• It fills in where data are lacking. No handbook or guideline has all the necessary data.

• Computational theories provide quantitative predictions needed by engineers. • It prevents us from reinventing the wheel by allowing us to recognize similarities

among problems, such as the tendency for drivers to adopt inappropriate decision criteria in many situations.

87

• It is reusable. Once the investment has been made to build a mode for a particular domain, the theory can be recycled inexpensively to answer many system-design questions.

Theory is a powerful lens that connects user needs to sets of data (Figure 4.6.1). It does this by illuminating both the need and the data. When faced with a specific user need, e.g., why do drivers look but not see important external cues?, theory allows us to refine the question in a more detailed way that improves the odds of an effective answer. A question that may at first seem unanswerable or illogical, may through the lens of theory become tractable. Theory also illuminates data. When faced with a large set of data, e.g., from a naturalistic observation study, theory helps us to determine which aspects of the data may be most useful in solving a problem.

Figure 4.6.1 Theory links data to user needs. (From Kantowitz, 2000)

Computational models produce a quantitative output. This is accomplished through mathematical equations and computer algorithms. Traffic engineers are accustomed to using computational models both at the aggregate level, where for example principles of fluid dynamics are used to predict macro-level traffic flow, and at the micro-level to describe behavior patterns of individual drivers. Since these models are usually written by civil engineers, the components of the model that explain mental and physical driver processes are often rudimentary. While human factors experts tend to produce qualitative models, if models are used at all, there are emerging process models of driver performance that are computational (Lee, in press). This is a relatively new emphasis in human factors, although continuous control-theory models have been applied to operator control for many years. Thus, the tools and techniques to build computational driver models already exist. Problem Statement A pile of bricks is not a house. A blueprint is needed to transform the pile into a dwelling. Similarly, a pile of data will not by itself lead to effective countermeasures. A theory is needed to transform the data into, first information (selective representations of data) and second, understanding (a set of organized principles to explain the data). While it is often necessary to

88

proceed from information (or even worse from raw data) to countermeasures, as shown in the dotted lines of Figure 4.6.2, this shortcut relies upon crude extrapolation, often based upon the idiosyncratic opinions of the expert collecting the data, rather than a systematic formalization of theory. In other words, many “theories” about traffic safety are implicit and latent in the minds of their creators. Science progresses when theories are made explicit and formal to be scrutinized and improved by the research and practice community at large. Thus, the goal of this project is to produce an explicit formal model of driving behavior that can generate and evaluate potential safety and human factors countermeasures.

Figure 4.6.2 Precursors of Countermeasures HF 1a: This is a large-scale long-duration effort to build a computational driver model from scratch. While several useful computational models of human behavior exist, almost none have been constructed primarily to explain driving. They cover relevant activities such as attention and decision making but a model, for example of chess playing, may not focus on the most salient cognitive activities required for driving. In particular, models of driving require fast real-time dynamic calculations; this has been called inner-loop processing in vehicle control theory plus a more strategic goal structure that governs slower high-level cognition such as the reason for the trip. These two levels of control often interact: driving a pregnant woman to a hospital as a strategic goal might impose stronger constraints on inner-loop control than a Sunday drive through the park. Drivers can succeed or fail at each level. Failure of inner-loop processing means the vehicle may crash or depart from the roadway. Failure of strategic control imposes additional trip demands: for example, most of us have had the experience of intending to drive to some recreational setting on a weekend but winding up in the parking lot at work instead. This project would entail three successive four-year phases. Phase I would accomplish specification of overall system architecture by a detailed evaluation of alternatives. Phase 2 would be an analytic build and extension of the most successful Phase I architecture. Phase II would stress empirical validation of the full model and generation of countermeasures. HF 1b: This is a light, less expensive, version of 1a. Instead of building a new general model of driving, we would try to extend one or more existing computational models to also cover driving. Therefore Phase I and part of Phase II could be minimized. The final model would be less broad than model 1a and it would be important to understand what aspects of driving behavior would not be covered.

89

Method / Approach HF1a: Phase I would be a shoot-off between different system architectures designed by competing groups of researchers. Typical architectures to be considered would be queuing theory models (Liu 1996), control theory models (Levison et al, 1998), ACT-RPM models (Byrne & Anderson, 2001; Salvucci, 2002), and central controller models (Meyer & Kieras, 1997) among others. This list could easily be expanded as part of Phase I. An independent set of researchers would generate a set of test scenarios for model evaluation. Models usually do well on an initial test scenario. Therefore, an important part of the shoot-off would require additional evaluation on new scenarios that were unknown initially to the research groups building the competing architectures. In Phase II the winning architecture would be expanded by adding new modules designed to increase the domains of applicability. Sets of experts from different areas (e.g., road geometry, traffic safety, driver workload and distraction, etc) would decide which domains are most important. Phase III would collect new data, as well as using existing data sets, to evaluate and improve the final model. HF1b: Phase I would be limited to a one-year evaluation of only two competing existing models by researchers on a team selected by open competition. Phase II would have a smaller set of test scenarios. Phase III would emphasize using existing data sets rather than collecting new data. Project Duration HF1a: A twelve-year duration is currently unheard of in DOT research projects. This goes a long way to explaining why there are no basic research models in driving. Basic research takes a long time and a consistent long duration is far more important than adding large sums of funding to decrease project time. Spending $1 million per year for 12 years will create a better outcome than, for example, spending $24 million over 6 years. Nine women cannot produce a baby in one month. A basic research model of this complexity cannot be produced in the normal five-year research cycle. Advanced projects require advanced thinking. The long-term research plan should allow the government to terminate the project at the end of each Phase if progress does not warrant continuation but the intention should be for supporting a 12-year project. HF1b: The five-year proposed duration is consistent with past large DOT projects such as FHWAATIS/CVO. This is enough time to borrow an existing computational model that can be extended into the driving domain. Payoff Potential Both proposed projects have a large potential payoff. At present, there is no complete computational model of the driver. Thus, countermeasures are proposed based upon instinct, general expertise, and luck. There are some highly gifted engineers whose years of experience have produced worthwhile counter-measures, just as centuries ago some highly gifted Italian violin makers produced extraordinary instruments. But many of their secrets died with those Italian artisans. If the human factors of driving is to advance as a discipline, it must seek and find the formal models used by other branches of science.

90

HF 2. Processing Multiple Sources of Information - Category: Human Factors Information Overload Comment from R&T Partnership Steering Committee The reviewers generally couldn’t form a clear opinion on the merits of this program. In terms of specific questions/comments, does “workload” include only “divided attention” or also “inattention” due to fatigue, boredom, etc? Since the analytical and validation parts include instrumented vehicles, this should be coordinated with the F-SHRP safety project. How will the model help engineers make better decisions about how much information is too much in a given situation?

This proposed research is better classified as Fundamental Research. Again, this proposed research needs further assessment by international Human Factors experts who have attempted this and who have no stake in the United States program. Response from White Paper authors Figure 4.6.3 shows driver performance as a function of workload. The curve has three important regions. When workload is too low, performance is limited by inattention and boredom. Performance is optimal at moderate levels of workload. When workload is too high, problems of driver capacity and divided attention decrease driving performance. As driver fatigue increases, the entire curve shifts downward. Thus, workload is a broad concept that can be applied to a wide range of issues related to driving performance. The model should be able to identify transition levels across the three regions, helping engineers make better decisions about driver underload and overload.

Figure 4.6.3 Three driver workload regions.

The empirical phase could certainly be coordinated with an F-SHRP project since it is easy for instrumented vehicles to acquire vast quantities of data.

91

Narrative Description People have a limited ability to process information (Kantowitz & Campbell, 1996). When the rate of incoming information exceeds this ability, errors are made. Fortunately, such errors are not random. Instead, they follow from the human cognitive system architecture and so can be predicted. For example, people raise their decision criterion (beta in signal detection theory) to minimize the need to respond to stimuli, thereby increasing the risk that a crucial signal will not lead to an appropriate response. It is convenient in driving to divide information into two main classifications: in-vehicle information and out-of-vehicle information. It is common sense that a driver who is not looking out through the windshield cannot observe potential dangers, although the common statement that vision accounts for 90% of driving information is premature (Sivak, 1996). The rising presence of new in-vehicle information systems offers considerable potential to distract both the driver’s vision and attention to events inside the vehicle, leading to a concurrent decrease in ability to monitor the out-of-vehicle environment. Furthermore, increasing scientific awareness of the distraction associated with devices such as cell phones has made researchers, and perhaps the general driving public, note that even “eyes on the road” do not ensure “mind on the road.” How can we measure the complexity of an intersection with many traffic control devices and intricate road geometry in terms of information processing demands placed on the driver? Although it is clear that rate of external information flow is proportional to vehicle speed, can we find a quantitative relationship between speed and information that would be useful to a highway engineer? How can we measure the workload and distraction of in-vehicle information? And, perhaps most importantly, how can we add up these measures to determine total demands on the driver? Finally, how can total demand be related to driving safety and potential countermeasures? While DOT is currently funding research on driver workload and distraction, with an emphasis upon in-vehicle information and workload managers (e.g., SAVE-IT, CAMP), there would be substantial safety benefits to an integrated approach that tries to answer the sorts of questions raised in the preceding paragraph as part of a systems approach to driving safety. It would be especially valuable to consider and to calibrate in- and out-of-vehicle information flow in a single project. Problem Statement A plethora of secondary tasks has been used to measure driver workload, but without additional theoretical understanding, it is unclear how such measures can be combined because many are redundant (Young & Angell, 2003). So the first part of the problem is to create a driver attention model to illuminate the large set of potential measures of driver workload and distraction. Such a model must explain in- and out-of-vehicle information load as well as the switching costs associated with transfers of attention from in-vehicle to out-of-vehicle events. Once a model is formulated it will be tested against empirical data for solving Applied problems such as:

• Create a metric for out-of-vehicle information load based upon vehicle speed and environment

92

• Create a metric for in-vehicle information load, including load imposed by non-visual events

• Estimate total driver workload • Create red-line limits for excessive rates of information load • Validate model predictions • Write engineering design guidelines for avoiding information overload

Finally, once the above is accomplished, suggestions should be made for appropriate countermeasures. Method / Approach This project contains analytic, empirical, and validation phases. Analytic efforts include reviews of existing attention models and DOT research in progress and building a driver attention model. Empirical aspects will require data collection in both driving simulators and instrumented vehicles. Simulators are needed to vary in-vehicle and out-of-vehicle complexity. Vehicles are needed to validate simulator results. Finally, revised model predictions must be tested against on-road data so that design guidelines can be established. Project Duration Year 1 will be entirely analytic. Year 2 continues analytic efforts and the design of empirical data collection experiments. Years 3 and 4 perform experiments and compare results to model predictions, revising the attention model as required. Year 5 validates the final revised model against on-road data and provides design guidelines for safe information loads for in- and out-of-vehicle information. Payoff Potential The potential payoff is quite high because several products will be delivered. The most important product is a driver attention model that informs research and generates useful answers for highway engineers and vehicle designers. Several data sets will be collected in a standardized format that can be used in other projects. Finally, design guidelines will make application of these results straightforward for highway engineers and vehicle designers. There is also potential to coordinate this project with other Federal research efforts in progress as well as with projects discussed in this chapter. HF 3. Understanding Speed Selection - Category: Human Factors Speed Control Comment from R&T Partnership Steering Committee This is an important project to undertake. However, there is concern about the stated hypotheses and details. It may not be just view of road ahead, but what the road behind has been like, past experience on the road, and how other drivers are behaving. The proposed method/approach is vague. Also, simulator and road experiments to measure psychological parameters related to speed are indicated which jump immediately to implementing countermeasures – time and energy to develop and test countermeasures is necessary. A clear statement of the question and a sound detailed approach is needed. Again, this work should be coordinated with F-SHRP and NHTSA’s current research on speed choice.

93

Response from White Paper authors The comments are all reasonable and appropriate. It would be prudent to add another year and additional funds to test and develop countermeasures. Narrative Description Excessive speed is a major cause of accidents for both light passenger vehicles and heavy vehicles, contributing to run-off-road accidents and other crashes. A recent accident at the intersection of I-94 and I-75 in Michigan illustrates the serious economic impact of even a single accident. A gasoline tanker truck overturned and caught fire. The heat generated by the fire melted the steel in the overpass and has closed the intersection for several months while it is being rebuilt. It is likely that the cost of this one accident, including increased congestion and construction, may exceed the proposed cost of this research project. While a rational driver would select speed according to warning signs and road conditions, many drivers fail to do this. What are some of the other factors that influence driver speed selection and what implications might they have for effective countermeasures? One reasonable hypothesis is that perception of the roadway controls speed. For example, how might a driver select a speed to enter a curve (assuming the curve warning sign is ignored)? While radius of curvature, an important design parameter for road geometry, has a strong effect other factors, such a length of the curve, may also control driver speed selection. A shorter curve may be perceived as less difficult (holding radius constant), leading to a higher speed selection. Feedback may also control speed selection. However, just looking at the speedometer may not be sufficient since drivers adapt and may not give proper weight to this display. Preliminary data on in-vehicle devices suggest that additional auditory feedback when the speed limit is exceeded may remind drivers to slow down. Problem Statement Review the psychological parameters that may affect driver selection of speed for light passenger and heavy vehicles. Parameters may vary according to the kind of driver (professional or not), hours of service, driver fatigue, driver age, and driver skill and experience. What elements of road geometry are most salient in speed decisions? Does vehicle technology, e.g., ABS brakes, studded tires, encourage a higher speed selection? How does driver perception of traffic density within varying levels of service affect speed selection? Determine empirically which parameters exert the greatest control on speed selection. Suggest and test appropriate countermeasures. Method / Approach Start by performing a statistical analysis of national accident data to define the set of accidents associated with excessive speed. Although coding limitations will probably prevent any detailed assessment of psychological parameters, this step is useful to set the stage. Design simulator and road experiments to measure driver speed selection as a function of estimated psychological parameters. Implement countermeasures and collect new data to determine their effectiveness.

94

Project Duration The first year will be spent doing statistical analyses, reviewing literature, including research in progress in Europe on Intelligent Speed Adaptation, and designing experiments. Data collection and analysis will occur in years two and three. Note that if substantial relevant data are already available from existing field operational tests, this time period (and associated costs) could be reduced substantially. Countermeasures will be determined in year three. Year four tests a small set of potential countermeasures. Payoff Potential Studying psychological factors of individual drivers is risky but there are models of perception than can be applied. This problem is so important, that it is worth the risk. Speed selection is a problem that has been around for many years, but a solution will not be attained until a better understanding of relevant driver perception and decision processes is achieved. HF 4. Look But Did Not See - Category: Human Factors Driver Perception Comment from R&T Partnership Steering Committee The topic seems to have merit but the proposed study needs refinement. Specific questions include: 1) Will eye trackers really define what is “seen?” 2) If we learn whether the problem is (a) peripheral vision, vs. (b) visual scanning habits, what will that tell us in terms of treatments? The authors need to make it clear how the increase in understanding might translate into countermeasures. More justification is needed. 3) Shouldn’t the study look at overrepresentation in crashes rather than just lab studies (i.e., do people with poor peripheral vision or poor scanning habits get in more look-but-did-not-see crashes?) It would seem that unless one can show a relationship between the competing explanations and crashes, there is not much urgency to finding out which explanation is more likely. Would one not have to show that people who scan inefficiently are over-represented in the look-but-did-not-see accidents? For, if not, efficient scanning may not be of importance. Response from White Paper authors Eye trackers do not define what is perceived; they only resolve where the eye is looking. This is sufficient to determine which visual system, foveal or peripheral vision, is controlling perception. There already are existing devices to test peripheral versus central vision. Indeed, during my last eye exam the optometrist used such a machine because poor peripheral vision is a predictor of glaucoma. Testing drivers for peripheral vision is a good idea. I am not sure how difficult it would be to obtain a sample of drivers involved in look-but-did-not-see accidents who are willing to have their vision tested. Perhaps people would do this if the project paid for the testing. The ultimate countermeasure would have drivers tested for peripheral vision in addition to central vision as is currently done. Drivers with unacceptable peripheral vision would have limitations placed on their driving but this might be a politically difficult decision to implement.

95

Other less drastic countermeasures might involve special training for drivers with poor peripheral vision to alter visual scanning patterns. Narrative Description It is an amazing fact that drivers can be looking directly at on object yet fail to perceive it. For example, drivers entering a roundabout collide with bicyclists who must have been clearly visible, yet the drivers report consistently that they did not see the bicycle rider (Herslund & Jorgensen, 2003). A similar effect, called change blindness or change inattention, has been well documented in experimental psychology research where even gross changes in a scene, e.g., a person walking by in a gorilla suit, have gone unnoticed (Mack, 2003). Two explanations of these phenomena have received some support from data. The first explanation is the distinction between foveal and peripheral vision. Drivers are less likely to perceive objects in peripheral vision than objects in foveal (central) vision. Time to detect peripheral stimuli is increased when workload increases. The second explanation is the strategy associated with visual scanning learned from experience. For example, experienced drivers know where to look for vehicles when entering a roundabout and may not scan other areas where a bicycle is more likely to appear than a vehicle (Herslund & Jorgensen, 2003). This hypothesis is consistent with the paradoxical finding that bicyclists are safer when in the proximity of vehicles than when alone. Problem Statement Both these explanations may be illuminated by studies of eye tracking. Although eye trackers have been used for many years, they can be finicky devices that can be difficult to use, especially in bright daylight and when drivers are wearing corrective lenses. But there have been new developments in eye tracking hardware that claim to solve many of these difficulties. If eye trackers are used in dim light, where the iris is open wider, they often work better. Being able to monitor exact gaze patterns much more precisely than from video data offers an opportunity to examine both explanations. First, it is easy to determine if an object falls into foveal or peripheral vision. Second, distributions of gaze patterns over time may help to clarify visual scanning strategies used by drivers. Method / Approach Conduct literature reviews of (1) eye tracking equipment, (2) effects of foveal versus peripheral vision, and (3) change blindness. Select and purchase the best two available eye trackers and compare their effectiveness in different light levels with different drivers. Conduct driving simulator and instrumented vehicle studies to investigate detection of external objects under a variety of driving conditions, monitoring gaze patterns for precise location and duration. Conduct appropriate statistical analyses, including time series analysis and pattern analysis, of gaze patterns to infer driver scanning strategies. Suggest possible countermeasures.

96

Project Duration The first year will be spent doing reviews, designing experiments, and hardware evaluation and calibration. The next two years will be spent collecting and analyzing visual driving data. Payoff Potential This research offers the potential to solve an important class of driver error by taking advantage of new instrumentation and theories of attention. HF 5. Design Driver - Category: Human Factors Design Standards Comment from R&T Partnership Steering Committee There are a number of questions surrounding this proposed study. More clarity is needed on the true effect of the definition of “design driver” on actual design standards. Also, the study should go beyond just defining the 85th, 90th, and 95th percentile of each critical parameter to some concept of what combinations of critical parameters and percentiles that might lead to a defined risk level or failure probability. Is it feasible and appropriate to incorporate “consistent” risk of failure probability in highway designs? Response from White Paper authors These are helpful comments. There are statistical techniques for determining how groups of variables form clusters and this could be applied to evaluating combinations of critical parameters. The workshops could seek agreement on the best ways to utilize driver characteristics for actual design standards. Evaluating “consistent” risk might not be feasible within the proposed workshops, but this could be a topic for future research if the workshops were successful enough to promote renewed interest on the design driver. Narrative Description Transportation engineers have implicit standards for user characteristics when they select particular parameters, such as the 85th percentile speed, for design specifications (Dewar, 1992). Unfortunately, as has been noted by a past president of the Institute of Transportation Engineers (Pline, 1991), there is no universal acceptance of driver characteristics. Changes in the demography of drivers and the mixture of on-road vehicles also argue for better specification and study of driver characteristics and the relationships between the three components of a transportation system: the facility, the vehicle, and the driver. Problem Statement The design driver is a statistical artifact, much like the average taxpayer with 2.3 children and 0.7 pet dogs. Using the 85th percentile value characterizes a population or a sample, not an individual. It is well –known in engineering anthropometry, the science of body measurements, that the 50th percentile male is not composed of body measurements that are the 50th percentile for each dimension. In other words, if we started with a sample and then successively eliminated people by filtering on specific dimensions (e.g., 50th percentile height, weight, arm length, etc), in short order the entire sample would be eliminated. The design driver suffers from the same

97

difficulty. While selecting a design driver on a single dimension is useful, great care is required to specify an ideal design driver composed of several dimensions. Thus, the first step would be to identify which design driver dimensions would assist transportation engineers. Should the ideal design driver be based upon an average, a minimum or a maximum? For example, it does not make sense to design a ladder to hold the average weight of a population. Ladders should be designed for a maximum weight, such as the 95th (or higher) percentile. When all transportation components are considered (facility, vehicle, driver) is there a common denominator? For example, would the basis for a design driver change given the target (vehicle versus facility)? Method Approach Convene a panel of transportation engineers, vehicle engineers, and human factors engineers and hold a workshop on these issues after white papers have been distributed. Then have experts write position papers about the best dimensions for a design driver. Hold a final workshop to obtain agreement on these issues. Present findings at a national meeting. Project Duration The author of this paper has taken the liberty of decreasing the Human Factors Workshop suggested duration of 36 months to 18 months. This should offer ample time to schedule and hold two workshops with associated white papers. Payoff Potential It has been 20 years since a TRB sub-committee chaired by Dr. Vivek Bhise considered the Design Driver and so it is time to update this topic. HF 6. Risk Homeostasis - Category: Human Factors Decision Rationality Comment from R&T Partnership Steering Committee This topic should be in the fundamental research category but there’s no clear support for this project. More justification and explanation is needed. The methodology is vague. The reviewers believe it’s important to find out “how” drivers will react or adapt to changes but not so sure on “why” they change is critical to know to develop effective countermeasures as the authors have stated. Also, the reviewers are not convinced that knowing if there is a difference between assessed vs. objective (true) risk will tell us anything (particularly since true risk may have little or no influence on behavior). Response from White Paper authors The author agrees with the comments and does not consider Risk Homeostasis to be a high-priority research project. (As explained earlier, these projects were proposed by the Working Committee, not the white paper author.)

98

Narrative Description The theory of risk homeostasis states that drivers attempt to maintain a constant level of risk. Thus, when technology improves safety, for example by adding ABS brakes, drivers compensate by driving faster, thus negating safety benefits of new technology. In general, data do not support this theory (Evans, 1991 p.299). Statistical analyses that purport to show no benefit of safety standards (e.g., Peltzman, 1975) have been severely criticized for containing biased and spurious correlations (e.g., Joksch, 1976). It is remarkable how this theory has survived falsifying data, but theories tend to persist until replaced by better theories (Kantowitz, et al, 2004 chapter 1). A better theory would be the construct of bounded rationality in which only highly salient cues control the driver’s decision processes (Sivak, 2002). Drivers tend to satisfice rather than to compute an optimal rational decision. According to this view, drivers with ABS maintain shorter headways than those without ABS (Sagberg, Fosser & Saetermo, 1997) because of behavioral adaptation, a form of bounded rationality. Problem Statement For safety countermeasures to be effective, they need to address what drivers actually do as opposed to what they are capable of doing. There is great need to better understand the extent to which bounded rationality is involved in driver risk assessment, and how the contribution of bounded rationality is influenced by driver aspects (e.g., age and sex), vehicular factors (e.g., passenger vs. commercial vehicle), and environmental variables (e.g., weather, geometric design of the roadway, type of conflict situation). Furthermore, given differences in fatality and accident rates in other countries, it is important to identify cultural and social factors that influence driver risk assessment. Method / Approach This study would evaluate the extent of the mismatch between objective risk and subjective assessment of risk for the same traffic situations. The objective risk would be based on accident data. The subjective risk would be obtained by rating the risk in representations of traffic scenes (e.g., Sivak, Soler, Tränkle, and Spaghnol, 1989). A variety of traffic situations would be considered. Consequently, it would be possible to identify factors for which objective and subjective risks are similar, as well as factors for which there is a major mismatch. It would be highly desirable to compare data sets from several countries and this topic is well-suited for international collaboration for which special funding is occasionally available.

Project Duration The first four months will be spent developing data protocols that are consistent across different countries and cultures. The remaining two years are for data collection. Payoff Potential This research is a first step towards understanding why drivers in different countries perceive driving risk differently.

99

HF 7. Driving Simulator Validity - Category: Human Factors Simulator Generalization Comment from R&T Partnership Steering Committee This proposed project seems to be useful. The project should consider using instrumented vehicle drivers in NHTSA’s current work and in F-SHRP as subjects, since “real world” data on driving will be collected in those projects. Alternatives to the proposed full-block experimental design should be explored that will permit strong enough statistical analyses to achieve validation. Response from White Paper authors The author agrees with the comments. It would be more cost effective if drivers from current and F-SHRP studies were used but this will require drivers to navigate a fixed course so that the same routes can be created in the simulator. Narrative Description While driving simulators are widely used, becoming more capable and less expensive as computer prices decrease, many highway engineers view them as research tools and not necessarily as highway design tools. However, in Europe driving simulators are used by highway engineers as a way to visualize and compare different designs for road geometry, tunnel design, and locations of traffic control units; indeed, an AASHTO-FHWA scanning tour of Europe scheduled for June 2004 will investigate this use of simulators. Similar work in the U.S. has been proposed (Rosenthal et al, 2004). Since people drive real vehicles on real roads, it is important to compare simulator driving with roadway driving using the same drivers. The few experiments that have done this generally conclude that simulators have high relative validity (results on roads and simulators are qualitatively alike) but little absolute validity (results being identical). However, what is of key importance is the ability to predict on-road performance from simulator data. This can be accomplished without absolute validity if a transformation equation can be developed. For example, drivers in a simulator typically drive faster than on the road, probably because the optical flow in a simulator is less than the real world. Thus, there is no absolute validity for speed. But as long as there is some mathematical, and hopefully linear, equation that relates simulator speed to road speed, it is easy to use simulator data to predict road behavior. Since it is less expensive to build new roads in a simulator, different geometries can be efficiently compared in a simulator. Finally, no one has ever died in a simulator crash so research that might be high risk on a road can still be conducted in a simulator. We are just beginning to compare the various characteristics of different models of driving simulators (Boer et al, 2003). Simulators differ on many dimensions such as field of view, contrast, refresh rate, brightness, update time, etc. Until this is studied more carefully, it will remain difficult to compare experimental results obtained from different simulators. Furthermore, such research will help us understand how to build better simulators less expensively by learning which simulator parameters are more important, and so worth capital investment, and which are less crucial.

100

Problem Statement It has been difficult in the past to obtain FHWA funds for simulator validation, since most simulator studies have been conducted to answer very specific questions where validation is but a small part of the project. In order to develop simulator methodology, making simulators a more effective tool for highway research, a project is required that is aimed at improving the methodology in general. The problem has two parts. First, empirical data must be collected that compare performance of the same driver in the simulator and on the road. Sufficient data must be collected to allow calculation of road and simulator reliability coefficients using repeated testing in both environments. Simulator roadways need be built, to official highway standards which researchers have not always accomplished in the past due to resource limitations, that match nearby road systems. Instrumented vehicles will collect the same data as the simulator. Appropriate statistical techniques will be used to compare simulator and road data, testing to what degree it is possible to estimate road performance from simulator data. Second, since the research world uses several models of simulators of varying expense and complexity, additional work is needed to compare parameters across simulators. The ideal test would have one location with several simulators so that the same drivers could be used in all simulator tests. This could also be accomplished, but perhaps less efficiently, if different simulators were located nearby so that the same drivers could be tested. Experimental designs would have to take into account possible negative transfer effects across simulators. The goal is to determine the most important simulator parameters that effect driver behavior. The most expensive simulators with very high physical identity to vehicles may not be required for many research problems. Method / Approach Using the same drivers, collect sufficient data in a simulator and an instrumented vehicle to permit appropriate statistical analyses that validate the simulator as a predictor of on-road behavior. Extend the methodology of Boer et al (2003) to further compare varieties of driving simulator and identify critical simulator parameters so that performance on simulator X can be predicted from simulator Y. Project Duration While simulator data collection is relative fast once scenarios have been programmed, on-road data collection is relatively slow. The first 6 months will be spent building scenarios, matching simulator roadways to local roadways, and designing experiments. The next 3 years will be for data acquisition and analysis. Payoff Potential Driving simulators are a key tool that cannot be fully exploited until their methodological characteristics have been systematically studied. There is great potential to use simulators for solving many problems safely and efficiently.

101

4.7 Relationship Between These Proposal Projects and the F-SHRP Research Plan Table 4.7.1 lists the current set of projects and the closest related F-SHRP project (Campbell, et al, 2003). It is likely that data from some F-SHRP projects can be used to support the present proposed research projects. Project HF1 has the greatest overlap, which is hardly surprising since its goal is a complete model of the driver. Other smaller projects have much less overlap. One important caveat, indicated for projects HF2 and HF7 is that F-SHRP data will be useful only if the same drivers are tested in the field and in the driving simulator. Although it will be quite useful to coordinate F-SHRP work on countermeasures to the current projects, it is premature to attempt this now. F-SHRP project 2-3.2, Retrospective Countermeasures Evaluation Projects, is described as a “placeholder” in the F-SHRP document. Future comparisons must wait until the places are filled in. Project Title F-SHRP Project HF 1a: Computational Driver Model: WE (Whole Enchilada) HF 1b: Computational Driver Model: Light

2-1.2 Development of analysis Methods for Site-Based Risk Studies Using Recent Data 2-1.3 Development of analysis Methods for Vehicle-Based Risk Studies Using Recent Data 2-2.3 Vehicle-Based Risk Study – Phase I: Study Design 2-2.7 Site-Based Risk Study – Phase II: Field Study

HF 2: Processing Multiple Sources Of Information

2-2.3* Vehicle-Based Risk Study – Phase I: Study Design

HF 3: Understanding Speed Selection

2-2.5 Vehicle-Based Risk Study – Phase IV: Road Departure Analysis and Countermeasure Implications

HF 4: Look but not see

2-2.7 Site-Based Risk Study – Phase II: Field Study

HF 5: Design Driver

None

HF 6: Risk Homeostasis

?

HF 7: Driving Simulator Validity

2-2.3* Vehicle-Based Risk Study – Phase I: Study Design

*Same drivers would need to be tested in a simulator and on the road

Table 4.7.1 Relationship of proposed projects to F-SHRP Safety Projects

102

4.8 Summary The most important Advanced project is Computational Models (HF 1). While this has greater risk than some of the other projects, the risk level is still reasonable and appropriate for an Advanced project with such a large potential payoff if successful. If funds permit, the complete (HF 1a) version of this project is preferred, but that will be expensive. If less funding is available, then the condensed version (HF 1b) should be supported. The most important Applied project is not clear because simulator validity (HF 7) was not carefully considered at the Irvine Workshop. The author of this chapter believes that this project will yield the best return for research dollars. However, there are also several other worthwhile Applied projects in the list.

103

5.0 Work Zones

5.1 Introduction According to the Fatality Analysis Reporting System (FARS) data maintained by the National Highway Traffic Safety Administration (NHTSA), a total of 1,181 fatalities occurred in 1,029 fatal work zone crashes across the country during 2002. During that same year, nearly 35,000 injuries occurred in highway work zones, according to data from the General Estimates System (GES) (GES Analytical Users Manual, 2001). These numbers have grown significantly in recent years. For example, the FARS statistics represent a 70 and 73 percent increase relative to the number of fatalities and fatal crashes reported in work zones during 1997, respectively. Similarly, the 35,000 injuries estimated for highway work zones during 2002 are nearly 46 percent higher than the 24,000 injuries in work zones reported by GES in 1997. More important than the absolute numbers of work zone crashes currently being reported, various before-after studies on this topic suggest that crash likelihood is increased by 20 percent or more (sometimes much more) at work zone locations. In a pro-litigious society as now exists in the U.S., the explicit creation of situations and conditions (albeit temporary) that actually increase crash risk has significant cost and liability implications to both the public and the private sector. Certainly, it would be desirable to emphasize designs and practices that minimize additional crash risk. Unfortunately, research to date has not been sufficient to fully define what work zone designs and practices are the safest, and perhaps more importantly, why they are the safest. From a national research needs perspective, work zone safety has been correctly identified as a cross-cutting emphasis area (National Highway R&T Partnership, 2002). Indeed, work zone safety concerns overlay policy decisions, programming and planning activities, operations and mobility considerations. Furthermore, all roadways eventually require some type of maintenance and renovation to keep them capable of serving the travel needs for which they are intended. Insights gained and lessons learned relative to the design and conduct of work zones on all types of roadways has potential application to non-work zone locations as well. At a broader level, it is difficult for either the FHWA or state agencies to determine whether program efforts to improve work zone safety have been successful at either the state or national level, or whether more drastic efforts are needed. Improved monitoring and analysis of work zone data, including safety data, is a critical component of the proposed rulemaking changes to 23 CFR 630 Subpart J. If the proposed rulemaking is adopted, the states will need significant guidance and assistance from FHWA to allow them to fully evaluate and monitor the effectiveness of programs and policies adopted to help improve work zone safety at the state and regional level. In the following section, a series of research projects are outlined which provide a systematic process towards developing useful estimates of work zone exposure and work zone crash data from which work zone crash risk at a regional level can be computed. In addition, descriptions

104

of research that could help reduce work zone crash risk in the near term, given existing limited knowledge of work zone crash characteristics and trends, are also offered for consideration.

105

5.2 Specific Research Topics Summary of Research Projects within Categories

Category Project Title Type of Research

Likelihood of success (1-5 scale)

Duration (months)

Cost (Millions)

WZ 1a: Estimate WZ Exposure Characteristics from FMIS

Applied High 4

30 $1M Research Methodology –WZ Exposure Data WZ 1b: Develop VMT

Temporal Distributions to Estimate WZ Exposure

Applied Very High 5

18 $0.5M

WZ 2a: Incorporate New WZ Data Elements into CDS Crash Investigations

Advanced Moderately Low 2

60 $2M Research Methodology – WZ Crash Data

WZ 2b: Investigate Likelihood of Work Zone Crash Reporting

Applied Very High 5

18 $0.5M

WZ 3a: Feasibility and Validity of Region-wide WZ Crash Risk Estimation Techniques

Advanced Moderate 3

30 $1M Determine WZ Crash Causation

WZ 3b: Project-Level Crash Consequences of Work Zone Design Features

Applied Moderate 3

60 $2.5M

WZ 4a: Improving the Understanding and Measurement of Driver Behavior in High Driver Workload Environments

Advanced Moderate 3

36 $1.5M Identify/Evaluate Countermeasures to Mitigate WZ Crash Risk

WZ 4b: Evaluate Dynamic Queue End Warning Systems for WZ

Applied Moderately 3

60 $1.5M

Develop/Apply/ Evaluate WZ Management Procedures

WZ 5a: Analyze State WZ Monitoring and Management Programs and Procedures

Applied High 4

48 $1M

WZ = Work Zone GES = General Estimates System FMIS = Financial Management Information System VMT = Vehicle-Miles-Traveled CDS = Crashworthiness Data System

106

5.3 Knowledge Strongholds A multitude of studies documented in the literature identify project-level comparisons of the effect of introducing a work zone into the roadway environment. Estimates suggest that doing so raises the likelihood of a crash on that roadway segment anywhere from 7 to 147 percent (Graham et al., 1977; Nemeth and Migletz, 1978; Wang and Abrams, 1981; Kemper et al., 1985; Kuo and Mounce, 1985; Pal and Sinha, 1996; Ullman and Krammes, 1991). The literature is also fairly clear that rear-end crashes, sideswipe crashes, and crashes involving large trucks in work zones tend to be overrepresented relative to non-work zone crashes (AASHTO, 1987; Richards and Faulkner, 1981; Flowers and Cook, 1981; Daniels et al., 2000; Garber and Zhao, 2002). It is also generally recognized that existing crash databases have significant limitations for use in evaluating work zone safety (Daniels et al., 2000; Wang et al., 1996). Several of these limitations are discussed in appropriate problems statements later in this section.

5.4 Knowledge Gaps By far, there are more knowledge gaps than strongholds with regards to work zone safety. For instance, past studies have yielded contradictory findings as to whether work zone crashes are more severe or less severe than non-work zone crashes (Graham et al., 1977; Nemeth and Migletz, 1978; Kemper et al., 1985; Pal and Sinha, 1996; Ullman and Krammes, 1991; AASHTO, 1987; Richards and Faulkner, 1981; Flowers and Cook, 1981; Daniels et al., 2000; Garber and Zhao, 2002; Wang et al., 1996; Rouphail et al., 1988; Hall, 1989), whether work zone crash risk is more adversely affected during daylight or nighttime conditions (Graham et al., 1977; Ullman and Krammes, 1991; AASHTO, 1987; Daniels et al., 2000; Hall 1989), or even whether single-vehicle crashes (run-off-road, collisions with fixed objects or equipment) are more prevalent in work zones (Ullman and Krammes, 1991; AASHTO, 1987; Daniels et al., 2000; Garber and Zhao, 2002). Such inconsistency is not unexpected, since the term “work zone” represents a tremendously wide range of conditions and impacts to the roadway environment and to subsequent traffic operating conditions. The term “work zone” is presently too generic to be of significant use in crash causation analysis. Presently, work zone locations and layouts are not captured as part of existing crash report forms or in the electronic databases. This is a significant knowledge gap that significantly restricts abilities to assess crash risk potential or to otherwise establish causal relationships regarding work zone designs under various roadway and traffic conditions. Given that work zones may change configuration from day to day (or night to night), much more detail regarding work zone conditions actually present at the time of a crash needs to be captured to support crash causation studies. In addition to improving the understanding of how work zone decisions and practices influence safety, better data and understanding of factors influencing work zone crashes are believed to offer the potential to improve roadway safety in non-work zone locations as well. Work zones typically include use of design speeds, lane widths, lateral offsets, etc. that differ significantly

107

from “normal” conditions at that location and upstream of the work zone. In some cases, these values may even be exceptions to the minimums normally allowed for that type of facility or condition. This creates a unique opportunity to better understand the consequences of allowing such minimums or exceptions (and combinations thereof) upon crash risk and underlying driver behaviors. If highway designers are provided better data as to the safety consequences of selecting certain design element values (particularly if those values approach current minimum values), overall roadway safety may be improved while at the same time maximizing the efficiency with which public dollars spent on roadway repair and rehabilitation are utilized. Another significant knowledge gap relative to the development and interpretation of work zone crash risk estimation pertains to work zone exposure computations during times when no actual work activity is occurring. Traffic control devices, barriers, and equipment are left adjacent to the roadside in some work zones but not in others. These devices and equipment are additional objects in the right-of-way that have a non-zero probability of being impacted by an errant vehicle. The devices are also potential sources of visual clutter and distraction that may contribute to crash causation in some instances. The amount and type of such equipment that may be present at a particular location varies significantly from project to project, making it extremely difficult to define how vehicular exposure past these inactive work zones should be defined. Depending on the assumptions utilized, dramatically different estimates of exposure will be developed (Ullman, 2003). Current research is underway under the National Cooperative Highway Research Program (NCHRP Project 3-69, Design of Construction Work Zones on High-Speed Roadways) to look at develop improved design guidance under high-volume conditions. Similarly, work proposed under the FSHRP program for roadway renewal (TRB, 2003) includes studies to consider the safety ramifications of accelerated construction on highway workers (project 1-1.8) and to develop improved work zone traffic control and design guidance for high-volume roadways to emphasize consistency, visibility, and safety (Project 1-7.1). These projects further emphasize a perception, at least, that current work zone practices and standards can be improved, particularly in better accommodating the extremely high volumes of traffic that many regions now face. Comment from R&T Partnership Steering Committee General comments – One reviewer declined to review this paper due to the small relative size of the Work Zone (WZ) total crash and fatality problem. Also, another reviewer noted that WZ crashes that can realistically be affected by treatment might be only 20 percent of total crashes in WZ. Indeed, some research has shown that the risk in a work zone on at least Interstate roads is about 20% higher than on the same pieces of pavement when the zone was not there. Given that, unless we can make a work zone much safer than the target road we are building (e.g., slow all traffic down to 20 MPH and lead them through with a police car), it is wrong to imply that we can somehow reduce or eliminate all WZ crashes (unless it’s possible to make the WZ safer than the road being worked on). WZ “crash risk” can’t be defined with crash studies only – must have exposure and characteristics of WZ.

108

There’s a real question on whether work zones deserve priority attention under the R&T Safety Partnership effort. Of course, we are not asking the authors to settle this issue, but the TRB oversight committee will address it in the future. Response from White Paper authors The reviewers correctly note that the relative magnitude of crashes identified as occurring in a work zone is small compared to the total number of crashes that occur on U.S. highways nationwide. Furthermore, many of these identified crashes may indeed have occurred even if the work zone had not been present. Consequently, it is reasonable to question the value of emphasizing work zone crashes as a priority area in the R&T Safety Partnership effort. However, the author believes that the need to research work zone safety lies not in the absolute numbers themselves, but in the strategic value in understanding the causes and consequences of those crashes that may be induced by work zones. For example, the author believes that much can be learned by determining and understanding what work zone factors and combinations of factors contribute most significantly to the increased crash potential, and why those factors are so detrimental to safety. Armed with such knowledge, the consequences of future roadway design and operational decisions may more assessable than would otherwise be possible through traditional crash or epidemiological studies. At the present time, a roadway designer still must select a particular roadway design element (the actual radius of a horizontal curve to use at the end of a long tangent section, for example) to use without truly knowing what the likely consequences of that selection might be upon the expected crash risk of the roadway. Because of severe geographic or cost considerations, the designer may be interested in the implications of accepting a minimum, or even an exception to the minimum, value of a particular design element. The fact that work zones often represent such exceptions to standard practice offer a unique opportunity to better understand the consequences of such exceptions upon crash risk and underlying driver behaviors creates a de facto study environment in which to evaluate the ramifications of such decisions when made outside of work zone environments. The author believes that the other key reason that work zones are a strategic area of research importance to the Safety R&T partnership is that work zones present severe and unique liability considerations to highway agencies and to the private sector, considerations that for the most part are not as prevalent when other types of crashes occur. Most roadwork today is carried out by private contractors, working for the particular roadway agency with authority for the roadway of interest. Such arrangements create dual defendant opportunities for plaintiffs looking to assign blame and to recoup damages for crashes. At the same time, decisions are routinely made (probably rightly so) to accept a lower quality of roadway service while such work occurs so that the necessary activities can be accomplished within a reasonable amount of time and a reasonable budget. These decisions increase agency exposure for litigation in comparison to other types of decisions, such as deferring the upgrade of an existing roadway section (due to funding limitations) to meet the most current safety standards. There are significant ramifications to the private contractor as well, evidenced by the recent multi-million dollar awards for negligence or other deficiencies that have led to motorist deaths or severe injuries at

109

work zone locations. These awards, in turn, drive up insurance costs which are then passed on to the general public through higher bid prices for future work activities. Ultimately, having better data on work zone crashes, the different work zone characteristics that contribute to crashes, and the amount by which crashes are influenced by these characteristics will all lead to better work zone designs and help highway agencies and contractors address future litigation demands. Discussion of these key issues has been added to the introduction and the knowledge gaps sections of this topic area. Research suggestions from R&T Partnership Steering Committee Reviewers recommend a large well-funded study that measures crash-based effects of WZ characteristics (e.g., lane drop/narrowing methods, median crossovers). The cost will be in the collection of detailed WZ inventory data (which changes over time) and exposure data. Should there be inclusion of worker-safety in study? If so, this will require additional data on worker exposure.

Response from White Paper authors The author agrees with the reviewers that this important facet of work zone safety deserves a significant level of research attention. It was omitted in the first draft of the white paper because of a long-anticipated publication of an FHWA study on work zone crashes, and an existing NCHRP project (3-69, Design Guidelines for Work Zones on High-Volume Roadways), of which the author is a member of the research team that was already underway to examine and consider these issues to some degree. In recent months, it has become clear that the NCHRP study will not be sufficient in scale or funds to effectively address all of the questions surrounding the crash-based consequences of many of the work zone characteristics. Furthermore, the FHWA publication has still not been released, and even once released, is not expected to provide data on all types of work zone characteristics that may influence crash risk. Therefore, a new project has been introduced (WZ 3b.) into this section to focus on these questions.

5.5 Specific Research Projects WZ 1a. Exposure Data: Estimating Work Zone Exposure Characteristics from FMIS Comment from R&T Partnership Steering Committee There’s agreement that “exposure” is critical to WZ studies, but “exposure” has different meanings. More clarity on specific “outcomes” from better national WZ exposure (e.g. program changes this could lead to?) is needed. FMIS data needs to be supplemented with data on non-FMIS WZs in at least a pilot study so that “expansion factors” can be developed. “Exposure” needs to include active/inactive work periods so that we can learn how to minimize the latter.

110

Response from White Paper authors The author has included additional statements to address the possible outcomes of improved exposure data, the need to expand the study to allow extrapolation of FMIS data to non-FMIS work zones, and the need to incorporate exposure in terms of active and inactive work zones. Narrative Description The Financial Management Information System (FMIS) is the database used by FHWA to track information related to all highway projects financed with Federal-aid highway funds. FMIS data are used for planning and executing program activities, evaluating program performance, and depicting financial trends and requirements related to current and future funding. Projects are subdivided into a number of categories such as new construction, 4R construction, 4R maintenance, bridge new construction, bridge replacement, bridge rehabilitation, safety, and railroad/highway crossing. State and local agencies provide the data that populates FMIS, with FHWA personnel providing general review and oversight. Most of the information in FMIS is financial in nature, but project lengths, locations, days of activity, etc., are also requested from the states. These data are relevant for the purposes of estimating work zone exposure from a national, state, or regional perspective. Problem Statement The information included in FMIS has allowed FHWA to develop estimates of the amount of roadway that goes under contract each year (Highway Statistics Series, 2002). However, recent efforts to estimate annual work zone mileage on the NHS using field data sampling and extrapolation techniques did not correlate well with these FMIS estimates (Ullman, 2003). Rather, the FMIS estimates were much smaller than those obtained from field data, most likely because the FMIS estimate did not include several types of projects that also generate work zones (i.e., safety/traffic/traffic system management projects, environmental projects, special bridge projects, etc.). Furthermore, it was noted that FMIS does not capture much of the key data that is necessary to develop good estimates of work zone exposure. These data include the lengths of actual work activity; the times of the day or night when work is performed (i.e., capturing the difference between active and inactive work zones); the frequency, magnitude, and duration of capacity restrictions enacted at each work zone; and so on. The previous field data sampling and extrapolation effort indicates that these characteristics differ significantly depending upon project and roadway type, historical traffic volume levels, and other factors. Despite these limitations, FMIS does offer the opportunity to more thoroughly estimate and track work zone exposure at a national level than is possible through field data collection efforts alone. Better estimates of true amounts of work zone exposure will allow FHWA and state DOTs to better relate existing meta-level work zone crash statistics to changes in amount of work activity that occurs yearly. In this way, the benefits of state and regional programs and other broad-based actions implemented to improve work zone safety can be monitored and evaluated for effectiveness. Typically, such programs and broad initiatives cannot be effectively evaluated at the individual project level.

111

Research is needed to more thoroughly understand the relationships between key characteristics that accurately capture and describe work zone exposure, and information now included in FMIS. With this understanding, appropriate unit values of these characteristics could be developed that, multiplied by the FMIS data, could yield reasonable estimates of work zone extent and exposure nationally as well as by state or region. Among other things, unit values will need to be generated to relate the amount of FMIS-documented work activity to the total amount (FMIS and non-FMIS) of work zone activity that occurs in a region or at the national level. Method / Approach The proposed research would require the research team to work with FHWA personnel who manage FMIS in order to develop a strong understanding of the various project categories and other data elements currently captured. Then, a number of regions will need to be identified (in a statistically-representative manner) where the research team could travel and gather key field data for all projects in those regions (including those not documented in FMIS). Based on these data, unit values of key work zone exposure characteristics would be developed and related to the population of FMIS projects in those regions. Statistical comparisons across regions, project types, and other factors would be performed to ascertain the robustness of the unit values for use in extrapolating beyond the boundaries of the sampling regions. As a final task, researchers would need to develop a plan for periodically updating these unit values in order to use them in following years. Project Duration The research team will require time to interact closely with FHWA personnel who manage FMIS in order to establish a clear understanding of its capabilities and limitations. Several geographic regions will need to be identified where field data can be obtained and correlated back to the population of FMIS project information for that region. Payoff Potential High – FMIS is currently the most comprehensive national database of work zone projects available. A mechanism for gathering project data from the states is already in place. Establishing the correlation between information already kept in FMIS and key characteristics that truly define work zone “exposure” in a region or state will allow FHWA to more accurately benchmark and track trends in work zone exposure, crash risk, and mobility impacts. WZ 1b. Exposure Data: Develop VMT Temporal Distributions to Estimate WZ Exposure Comment from R&T Partnership Steering Committee There is agreement for the need, but questions on the method. It is noted that we do not even have good data on work zone vpd at this point, and that it is not clear what “different data sources available” means. Perhaps the need is for a project of automated collection of hourly/daily exposure in a cross-section of work zones – new, rather than existing data.

112

Response from White Paper authors The author agrees that very little data on vehicular demands through work zones has been collected to date. The authors still believe there is value in evaluating existing traffic demand data sources such as HPMS, traffic surveillance counts obtained by ITS now installed in many urban areas nationwide to develop statistically-valid temporal distributions (probably stratified by region, roadway type, and other factors) for use in estimating traffic demands through work zones during times when work activity is occurring as well as when the work zone is inactive. However, there will indeed be value in gathering ground-truth data at actual work zone locations as well. Together, a multi-pronged approach could yield a much richer dataset. Additional sentences have been added to the methodology section to include such data collection and analysis in this proposed project. Narrative Description Estimates of total vehicle exposure to highway work zones will be needed if reasonable estimates of work zone crash risk are to be developed. In most cases, only historical daily traffic volume estimates are available on the roadway sections where work zone activity takes place. These estimates are generated from state DOT program and planning divisions, using a variety of traffic count samples and extrapolation techniques. The Highway Pavement Monitoring System (HPMS) is a database of roadway and traffic volumes collected on a representative sample of roadways nationwide. In some cases, installation of Intelligent Transportation System infrastructure in large urban centers can serve as another potential source of traffic volumes. Problem Statement Work zones are typically not active during all times of the day, especially when lane closures or other significant traffic disruptions are required. In fact, such work activities are typically performed during off-peak travel periods (albeit the actual hours of activity may change from day to day over the course of the project). Methods of adjusting daily traffic volumes to reflect vehicular exposure during work zone activity need to be established before work zone crash risk can effectively be estimated and compared across project and roadway types, traffic control conditions enacted, etc. Furthermore, these adjustments must also take into consideration temporal differences in exposure over the day by different vehicle types. Large trucks, for example, may tend to be more prevalent at night or during other off-peak periods in urban regions as commercial drivers schedule their trips to avoid peak period congestion. Method / Approach Researchers will need to examine the HPMS database and a sample of traffic volume data from ITS centers in urban areas nationwide to assess the hourly temporal distributions of traffic volumes as a function of daily traffic volumes, region, roadway type, vehicle type, and other factors. Statistical techniques would need to be employed to determine which stratifications are most significant. Researchers will also need to develop a research plan to validate these distribution patterns developed through the existing HPMS and ITS databases by conducting a series of data collection efforts at work zones nationally to actually gather work zone traffic demand data. In addition, special attention will need to be given to work zones where traffic demands exceed the

113

capacity of the work zone during some portion of the day. Evidence suggests that traffic demand patterns change dramatically once demands exceed work zone capacity and congestion starts to develop (Ullman, 1996). Specifically, significant numbers of motorists will change routes (if available), reducing the amount of traffic actually approaching and passing through the work zone. The results of data collection sites are likely to yield diversion curves or other methods of adjusting historical traffic volumes to reflect actual work zone traffic demand when volumes exceed capacity. As a side note, diversion behavior does not appear to occur to any significant magnitude if traffic demands remain below the work zone capacity. As with the previous project statement, researchers will also need to prepare recommendations for updating these distributions on a regular basis so that work zone vehicle exposure estimation using these distributions can be maintained over time. Project Duration The proposed research does not require significant field data collection. Significant cooperation and coordination from state or regional agencies is also not required. It is believed that this work could be accomplished in a relatively straightforward and timely manner within the proposed duration of the project. Payoff Potential High - It is unlikely that traffic surveillance infrastructure will be installed to sufficient levels nationally (particularly in rural areas on non-interstate highways) so as to allow reasonable real-time collection of vehicle exposure in work zones in an ongoing manner. Estimation tools will be required if any type of reasonable tracking of work zone vehicle exposure at a state, regional, or national level is to be realized. WZ 2a. Crash Data: Incorporate New WZ Data Elements into CDS Crash Investigations Comment from R&T Partnership Steering Committee The white paper concentrates on CDS, as should be the case. GES should be deleted since it appears CDS is the target, and GES is extracted from existing police reports with no supplementing done. If CDS is the target, may require both changes in items and changes in sampling methodology (or a special study) since current sample of WZ crashes may be too small to be useful. Response from White Paper authors The author agrees with the reviewers. References to GES have been removed from the discussion. In addition, the potential need for a special pilot study to determine appropriate items and sampling methodology to use in assessing work zone crashes has been incorporated into the discussion. Narrative Description

114

The Crashworthiness Data System (CDS), part of the National Accident Sampling System maintained by NHTSA, is designed to allow for consistent comparison and in-depth analysis of crash factors across jurisdictional boundaries. Currently, the information in the CDS does not capture information that would be deemed useful in the analysis of work zone crash causation. The Model Minimum Uniform Crash Criteria (MMUCC) prepared in 1998 identifies several additional data fields that are recommended for inclusion on state crash report forms (USDOT, 1998). However, the extent to which individual states will eventually modify their report forms to incorporate all or some of these variables is unknown at this time. Agencies continue to struggle with the balance between requiring as much detailed information about a crash as possible, and the amount of time required by an investigating officer to properly collect and code that information. Problem Statement Work zones are highly diverse, depending on the roadway alignment and cross section, traffic demands, and required work activities. Most often, the travel path required by motorists is temporary, such that existing roadway inventory files do not represent the actual driving configuration present on the roadway at the time and location of the crash. The lack of work zone layout and condition documentation hinders attempts at ascertaining actual causal links of work zone crashes. Research is needed to establish a mechanism for obtaining the necessary work zone configuration data at the time of a crash, and to use that data in crash causation studies to better understand the features and conditions that contribute to work zone crash risk on a national basis. Method / Approach For this project, researchers will need to first define a consistent set of work zone data elements that are likely to capture causal relationships of work zone crashes. For example, the prevalence of rear-end crashes in previous work zone crash studies suggests that information about the presence and length of a traffic queue (or about estimated work zone capacity values and traffic demands which could indicate congestion) at the time of the crash could be an important potential causal link to explore and quantify. Other elements might include the length and lateral offset of lane shifts, and/or the type and proximity of channelizing devices or barriers next to travel lanes. Once the set of work zone data elements of interest is established, researchers will need to identify locations willing to pilot test the elements as part of their regular CDS data collection investigations. This pilot test may actually need to be a special study that explicitly focuses solely on the investigation of work zone crashes, so that a meaningful data sample to evaluate the protocol and quality of the information is collected in a reasonable period of time Following appropriate pilot testing and revisions, a more widespread implementation effort would be required. After sufficient time has elapsed to allow the use of the new data elements to be collected, follow-up analyses would be performed to validate expected causal relationships in work zone crashes and identify new ones as appropriate. Proposed Duration Because this project is based on establishing new data collection mechanisms to support improved work zone crash causation studies, a longer-term study period is needed. Researchers

115

will need to establish the proper data collection protocol, and work with several regions now included in the CDS sampling regions to pilot test the protocol. Then, the protocol will need to be implemented within as many regions as possible, and time allowed for enough data to be collected to allow proper statistical analysis. Obviously, the amount of time required for completion of the project will then depend on the number of regions that can be persuaded to adopt the new data collection criteria. Payoff Potential High – the collection of a new stream of data from which to assess work zone crash causation could prove highly useful in the identification of appropriate countermeasures and in the economic justification of such countermeasures on the basis of expected crash reductions. WZ 2b. Crash Data: Investigate Likelihood of Work Zone Crash Reporting Comment from R&T Partnership Steering Committee This study looks feasible if the FMIS data has location information that is the same as on crash files (not always the case for other national databases). Response from White Paper authors The author agrees that the key will be to select regions and work zones within those regions whose project location data are consistent with police crash location reporting procedures. Although it would be advantageous to utilize the same regions and data sources as would be used for project statement WZ 1a. above, it is not an absolute necessity. The author has added a sentence to the methodology section to emphasize the criticality of being able to link work zone project location and crash location data together. Narrative Description In most states, the notation of the presence of a work zone is up to the discretion of the investigating officer. Two previous studies have concluded that the way that the presence of a work zone is noted on a state’s crash report form affects the likelihood that a crash will be coded as being in a work zone by the reporting officer (Wang et al., 1996; Ullman, 2003). In another study, researchers found that only a fraction of crashes that occurred within the limits of a work zone project were actually coded as being in a work zone (Ullman and Krammes, 1991). These findings imply that the existing crash databases do not represent the full population of crashes that occur within highway work zones. Consequently, work zone crash risk estimates based solely on those crash records noted as occurring in a work zone will be lower than the risk that actually exists within work zones in the region. Problem Statement Although it is hypothesized that many crashes that occur in work zones are not coded as such, the implications of this underreporting in terms of effective use of available databases has not been determined. It is not known whether such underreporting is randomly distributed amongst

116

the population of crashes, or represents systemic trends in the data (i.e., more underreporting occurs at night, on long-term construction projects, on projects in larger urban areas, in northern climates, on non-injury crashes, etc.). Research is needed to investigate and document any such systemic trends, and to establish an estimate of the amount of underreporting that occurs nationwide. Method/Approach Researchers will need to identify a variety of work zone projects from several jurisdictions across the country. Specific information on dates and locations of work for each project will need to be identified, and all crashes on those dates and times at the identified locations extracted from the crash databases. Analyses of which crashes were and were not coded as being in a work zone will be identified, and appropriate statistical methods employed to determine what trends, if any, exist. The locations studied could be the same as those used to establish the FMIS estimates (see the first problem statement) to take advantage of previously-collected data. The use of the same locations will also allow for feasibility testing of region-wide crash risk estimation (discussed in a following problem statement). However, the most important criteria for the project to be successful is to select regions for which work zone location referencing and police crash reporting location referencing are the same (or at least can be easily translated between the referencing systems used). Project Duration It is expected that only limited field data may need to be obtained from each work zone project location. The remaining effort of this research would entail extraction and analysis of information from the electronic crash databases in those regions where the work zone projects of interest exist. Payoff Potential Moderate – This project is critical in bringing to light the magnitude and trends regarding work zone crash underreporting. The findings could prove useful to agencies in interpreting work zone crash statistics to more accurately reflect what is likely occurring from a regional or statewide perspective. However, the findings themselves are not expected to result in significant reduction in this phenomenon. WZ 3a. Crash Causation: Feasibility and Validity of Region-wide WZ Crash Risk Estimation Techniques Comment from R&T Partnership Steering Committee The reviewers are unsure of merit of study. More justification is needed - more on what programs; practices might change if this study was completed? Unclear on actual model – crashes/vmt or something else? What does it tell us (particularly since majority of crashes would be present without WZ)? If issue is really “risk” as a function of WZ characteristics, need WZ inventory data (which changes over time within the same zone). Not sure if this is present in existing FMIS database.

117

Response from White Paper authors The reviewers correctly note that any efforts to assess regional programs or countermeasure implementations will require detailed work zone inventory data over time, which is the goal of project WZ 1a above. Also stated in statement WZ 1a above is the ultimate goal of being able to assess regional program or countermeasure implementation effectiveness, something that is currently not possible through project-level analyses. Thus, the goal of this project is to examine whether it is possible to gather or estimate region-wide work zone characteristics (via FMIS unit value expansion factors or other means) and relate them to region-wide work zone crash data so as to evaluate regional effectiveness of a particular program or countermeasure. The author recognizes that the actual rates or model may not reflect “ground truth” data, due to such things as many crashes not really being related to the work zone, potential lack of consistent reporting of work zone crashes, etc. However, if this project is successful, it can serve as a mechanism for evaluating changes in conditions over time, and allow at least some degree of regional program effectiveness evaluation. As to the actual model that is anticipated, the author would expect something in terms of crashes per vmt to be the most feasible measure. However, sentences in the problem statement allow for the investigation of other model approaches that might be more appropriate for gauging changes in crashes over time at a region-wide level. Narrative Description At the national level, fatality risk is estimated as a function of vehicles miles of travel and roadway classification. Such an assessment requires both accurate crash data and realistic estimates of motorist exposure in terms of vehicle miles traveled on various types of roadways each year. For highway work zones, efforts to date have focused primarily on project-level estimates of crash rates and changes in those rates due to the introduction of a work zone into the roadway environment. Such project level investigations do not allow for assessments of broad regional, statewide, or national initiatives to improve work zone safety. Problem Statement This proposed research builds upon the expected results of the previous problem statements described in this white paper. Research is needed to explore alternative model structures for estimating region-wide work zone crash risk, to perform a sensitivity analysis of the various assumptions used in extrapolating both work zone vehicular exposure and work zone crash data, and to validate the model structure using data from regions not previously examined under this research effort. Method/Approach In this project, researchers will determine the appropriate model structure to incorporate both work zone exposure estimates (from FMIS) and vehicle exposure to each work zone (using VMT temporal distributions) into an appropriate region-wide work zone crash risk estimate. The model development would likely use the data already collected from various regions nationally in the previous research projects. Validation studies would then be performed to assess the reasonableness of the model structure deemed most appropriate. The validation would be based on data from a new set of regions not previously examined under this research initiative.

118

Project Duration The initial model structure for risk estimation developed as the first part of this project would be performed on the data already obtained through the previous projects described earlier. However, the validation efforts of the new model structure will require additional data collection from FMIS, daily traffic volume estimates, and other regional roadway characteristics for use in the analysis. These additional data collection techniques will require sufficient time to coordinate efforts with appropriate agencies, collect and reduce the appropriate data, and perform the validation analysis. Payoff Potential High – this type of data has not been available to state or local agencies (nor FHWA) previously. The ability to establish region-wide crash risk estimates for work zones sets the stage for tracking work zone safety from a regional perspective, and systematic evaluations of changes in agency policies and procedures as they pertain to work zone safety. The results, if successful, will also allow agencies to assess whether or not work zone safety needs to be given additional priority relative to the other issues and initiatives also pressing for agency attention. WZ 3b. Project-Level Crash Consequences of Work Zone Design Features Narrative Description Despite the studies that have been conducted to date pertaining to work zone safety, there is a general dearth of information as to what work zone design features and operating characteristics influence crash likelihood (positively or negatively). Other than comparisons of the relative difference between median crossover and single lane closure designs for interstate highways (e.g., Burns et al., 1989) or the influence of entrance ramp presence or design (Casteel and Ullman, 1992), very few studies have attempted to isolate the effects of particular work zone features or characteristics upon crash potential. Problem Statement Research is needed to better understand the crash consequences of the different design elements upon crash potential so that future work zones can be designed and implemented more safely. Furthermore, this understanding should extend beyond single-element analyses to the consequences of the more typical design element interactions that commonly exist of work zones, and should take into consideration the change in these elements relative to pre-work zone conditions and to conditions upstream of the work zone. Consequently, an adequate geographic distribution of sites will be necessary as well. Method/Approach A before-during study of crashes at a fairly large number of work zone locations, stratified by the design elements and conditions listed previously, would be the primary method of analysis. Control sites will be required for each project location included in the analysis to account for extraneous influences (primarily weather and traffic volume changes) that could also influence crash occurrence. Detailed daily project diaries or similar records will need to be accessed to isolate specific days of activity and dates of major traffic control changes to include in the analysis. The outcomes would be a set of crash adjustment factors associated with the design elements (and element combinations) evaluated. It may be necessary to rely on historical

119

projects more so than current projects, due to the delay that typically exists in obtaining police crash report data. Project Duration The sample size required to do this study properly will be substantial. Time will be necessary to identify appropriate design criteria, locate potential sites (work zones and control), and collect the detailed project information (including daily activity information, when necessary). The study will require researchers to become familiar with several crash record databases to extract the appropriate data from sites across the country. Payoff Potential High—The results of a well-conducted study on this topic has the potential to be immediately applied to work zone designs nationwide. As in the previous project description, the results will also allow agencies to assess whether or not work zone safety needs to be given additional priority relative to the other issues and initiatives also pressing for agency attention. WZ 4a. Crash Risk: DeterImproving the Understanding and Measurement of Driver Behavior in High Driver Workload Environments Comment from R&T Partnership Steering Committee The reviewers are not clear on the underlying “knowns” or methodology. There are several questions. Does “human factors (HF) assessment” imply we know which HF measures affect driver behavior in WZs and by how much? If not, will “lab and field” measures of behavior be sufficient to validate treatments? If not, should we do a different research project – one than links HF measures with driver behavior in WZ (fundamental research)? Response from White Paper authors After consideration of the reviewer comments, the author agrees that a more fundamental research effort is needed in this area. The problem statement title and text have been modified accordingly. Narrative Description Current work zone traffic control (WZTC) standards are rooted in fundamental human factors principles of positive guidance, traffic control device visibility and legibility, and linear models of motorist information transmission approaching various work zone configurations (Post et al., 1981; McGee and Knapp, 19979; Hostetter et al., 1982). These models may be too simplistic for certain work zone situations, particularly on high-speed urban roadways. The presence of extraneous visual cues, vehicles, other traffic control devices, glare sources, etc. in some work zones may increase driver workload and decrease driving behavior. Unfortunately, reliable and calibrated measures of driver workload currently do not exist for work zones or non-work zone locations, nor has a linkage been established between any workload measures, driving behaviors, and ultimately crash outcomes. In a high-workload environment that is hypothesized to exist at many work zones, this type of linkage could prove useful in better understanding the underlying

120

causes of certain work zone crashes, and in developing work zone design evaluation methods and criteria in terms of their expected influence on driving behavior and crashes. Problem Statement Few studies have explicitly examined driver decision-making and behavior in a high driver workload environment such as can exist in a fairly complex work zone. A recent NCHRP report provides some guidance on driver information overload issues (Lerner et al., 2003), but more work is needed to further extend those concepts to address other driver workload issues, and to eventually provide assistance in work zone design and operations decisions. Specifically, research is needed to develop a more realistic model of driver cognition and behavior approaching and traveling through high work load situations such as exist at some work zones, to determine whether there are effective driver behavior performance measures that correspond to changes in driver workload, and to determine whether crash outcomes can be correlated to these driver behavior performance indicators of high workload conditions. Method/Approach The proposed research would consist of the development of a theoretical model of driver workload based on conditions and features present at work zone locations, identification of appropriate performance measures believed to be correlated with workload, laboratory studies to calibrate the performance measures to workload estimates, and validation studies to demonstrate a correlation between workload estimates (and corresponding driver behavior performance measures) and crash risk through before-during studies at work zones installed in the field. Project Duration A fairly lengthy duration is expected for this project. It may be possible to separate the model development and calibration activities (laboratory-based) from the crash risk validation portion of the study. Payoff Potential Moderate – Existing WZTC standards have functioned sufficiently for many years, and appear to be adequate for the majority of situations faced by practitioners nationwide. However, this project does have the potential to yield substantial benefits in more complex situations that often exist in work zones. The results of the study may be transferable to other complex driving situations as well, and so could extend the benefits realized to certain non-work zone locations. . WZ 4b. Crash Risk: Effectiveness of Real-Time Queue End Warning Systems in WZ Comment from R&T Partnership Steering Committee This topic has some importance since many rear-end crashes happen near WZ but rear end accidents are less severe. There are a couple of issues with the proposed methodology. The reviewers are not sure that lab/simulation studies are sufficient unless we have the link between the lab measures and actual crashes. While before\after crash studies are possible, will require significant effort to identify a reference group, since computerized WZ inventory data is not available.

121

Response from White Paper authors Whereas rear-end crashes in general tend to be less severe, they are not necessarily so at work zones where high-speed approaching traffic (i.e., 55 mph or higher) encounters a traffic queue at a work zone where vehicle speeds in the queue are often 10 mph or less. The problem may be particularly acute for the large-truck/automobile crashes at these queues. The author is aware of several fatal crashes at Texas work zones that have occurred when a large truck came up on an unexpected queue and did not stop before running completely over the automobile directly in front. The lack of good causal relationships between rear-end crashes and the proposed countermeasure is indeed problematic. The author agrees that before-after crash studies would be the most relevant means of evaluation. However, this type of countermeasure can only be evaluated at a project level, as not all work zones in a region would benefit from such a warning system (those that do not generate unexpected traffic queues for example). The author has added a statement about the need to conduct before-after crash studies, with appropriate selection of reference sites to serve as the control, as the final level of evaluation. The author has also reduced the overall likelihood of success to further address the reviewer concerns, and has lengthened the duration of the study to allow for crash evaluations to be conducted. The author has kept the laboratory study evaluation as part of the study plan, however. Although the linkage between laboratory measures and crash reduction potential is not possible at the outset, laboratory studies do offer the potential to exclude a countermeasure from further evaluation if no changes in performance measures can be detected. If driver performance changes can be identified in the laboratory, only then would testing move to the field. An interesting benefit of this study could actually be the eventual calibration of such laboratory performance measures to the eventual level of crash reduction observed in the field. Narrative Description Several studies that have examined trends in work zone crashes have identified significant increases in the relative proportion of rear-end crashes (Graham et al., 1977; Daniels, 2000; Garber and Zhao, 2002). Most studies have then concluded that these increases are the result of non-recurrent congestion created on occasion at the work zone that violates driver expectancy and leads to crashes between motorists at the upstream end of the queue (where significant speed differentials between vehicles exists). Problem Statement Current WZTC guidance indicates that advance signing warning of the presence of a work zone and possible congestion should be placed far enough upstream so as to be beyond the limits of any congestion that develops. However, since congestion typically does not exist at all times at the work zone, these static signs are often misleading to motorists and become ignored over time. Recent advances in ITS technology now allows traffic conditions to be monitored at key points in a work zone, and dynamic warning messages to be displayed on portable changeable message signs when congestion is detected. To date, however, so systematic evaluations have been performed of the ability of this technology to reduce rear-end crash potential. Also missing are

122

effective guidelines to help practitioners determine the types of roadways and work zones for which this type of technology is most effective. Method/Approach A series of laboratory (possibly driving simulator) and possibly field studies would first be performed to evaluate whether real-time dynamic queue warning systems can alter driver performance measures (speed changes, erratic maneuvers, etc.) believed to be related to the likelihood of rear-end crashes at work zones. Assuming these studies suggest potential benefit, a series of controlled field deployments would be undertaken at appropriate work zones in several regions to conduct before-during crash comparisons. Similar work zones in each region would be monitored as well during this time to serve as the control group. Project Duration Testing of dynamic queue-end warning systems will involve significant cooperation and coordination with state and local agencies, and may also require special agreements to be established with private-sector vendors providing the technology (if the agency is not purchasing the equipment outright). Identification of locations where such technology is to be deployed may take some time as well. Payoff Potential Moderate – Rear-end crashes are believed to be a significant cause of crash risk in locations where unexpected slowdowns and congestion are created. However, many agencies have already established policies restricting hours and days when work activities can occur, in an attempt to minimize the frequency and extent of this congestion. As the likelihood of encountering unexpected congestion due to work zones decreases, though, so too does driver expectancy for such congestion. This may mean that dynamic warning systems could have an even greater impact at those locations where they are deployed, but may be required at only a limited number of work zones throughout a particular region. WZ 5a. Management procedures: Analyze State WZ Monitoring and Management Programs and Procedures Comment from R&T Partnership Steering Committee As stated by the author of the white paper, the payoff for this proposed research is limited. There are two issues. If methodology is to compare current management practices to “Best Practices,” have the latter been defined based on the effects of crashes? If study is to define which management practices are “best” or successful,” then need to have data on jurisdictions with and without different management practices linked with some agreed-to “measure of WZ safety” as outcome variable. Response from White Paper authors A set of official “best practices” has not been identified through a crash analysis or similar assessment process. That lack of systematic evaluation to date is the primary impetus for recommending this type of study be done. The author agrees that additional emphasis on the

123

effects of these procedures on crashes is important. Modifications to the project statement have been made to emphasize these points further. Narrative Description The FHWA Work Zone Operations Best Practices Guidebook (FHWA, 2000) highlights a number of programs, policies, and procedures that state DOTs have enacted to help monitor and manage work zone safety and operations in their jurisdiction. Included in this category are such activities as statewide inspection programs, quality assurance reviews, and specific safety task forces. Other techniques may be in place in other states which were not included in the guidebook. Problem Statement To date, no efforts have been made to quantitatively assess the effectiveness of work zone monitoring and management activities from a statewide or regional perspective, as the data and methods for doing so have not previously been gathered or developed. This lack of assessment makes it difficult for agencies to determine which approaches are most appropriate for their particular situation. Research initiatives proposed earlier in this white paper are expected to offer an opportunity to evaluate region-wide programs and policies for work zones. Method/Approach A survey should be performed of the state DOTs to determine which regional or statewide monitoring and management programs or procedures are in place nationally, as well as when the programs or procedures were initiated. Information as to the specific activities undertaken, data obtained (if any), and decisions made based on such data should be collated. Crash analyses should be undertaken to compare crash experiences of those states that have established specific work zone programs and procedures with those states that have not. Project Duration Depending on the quality of historical crash data for work zones, some previously-implemented programs and procedures are potential candidates for immediate evaluation. For those programs and procedures that are implemented immediately prior or as part of the study, however, two to three years of data will need to be collected to allow adequate crash samples to be obtained for evaluation. Payoff Potential Moderate – An evaluation of work zone monitoring and management programs and procedures by state agencies upon crash experiences at the state or regional level could provide significant benefits to states by helping them identify actual benefits to be achieved and to justify any costs of implementing those programs and procedures found to be effective. The development of good evaluation data that help “sell” the effectiveness of such programs and practices is likely to further enhance the degree of agency buy-in and ultimate payoff to work zone safety improvement nationwide.

124

5.6 Summary The projects described in this paper are all believed to have reasonable chance of success and payoff potential. However, those having the greatest chance to succeed relate to the development of better estimates of work zone exposure and crash data quality, and to evaluate state work zone safety management programs and procedures: Develop VMT Temporal Distributions to Estimate WZ Exposure Estimate WZ Exposure Characteristics from FMIS Investigate Likelihood of Work Zone Crash Reporting Analyze State WZ Management Programs and Procedures

Perhaps most needed is the establishment of better work zone crash data documentation via the existing GES and CDS data collection schemes. However, given the various administrative and contractual details likely to be required to move this particular research initiative forward, it appears to have only a marginal chance of being successful in the near term.

125

6.0 Fundamental Advanced Research

6.1 Introduction This white paper deals with the area of Advanced Research. The following definition of Advanced Research is from the Transportation Research Board Special Report 261 (TRB, 2001) and is the working definition used for this report: Research that involves and draws upon basic research results to provide a better understanding of problems and develop innovative solutions. Sometimes referred to as exploratory research in order to convey its more fundamental character, its broader objectives, and the greater uncertainty in expected outcomes compared to problem-solving research. The TRB Special Report 261 also recommended increased funding and an increased emphasis to advanced research projects at the Federal Highway Administration (FHWA). Planned authorization of appropriations to the Transportation Research Bill2 finds that the federal investment in R&D should be properly balanced between short-term applied research and long-term fundamental research, as well as between research areas including materials and structures research, operations research, and human factors and policy research. Also, it creates a new Exploratory Advanced Research program (Section 502(d)) to address recommendations of the TRB and others that the FHWA’s R&D program should focus on fundamental, long-term research. Also, a recent statement by Vernon J. Ehlers (Chairman, Subcommittee on Environment, Technology, and Standards Committee on Science) on the introduction of the Surface Transportation research and Development Act of 2003 calls “on the U.S. Department of Transportation to take the lead in carrying out fundamental, long-term research to achieve breakthroughs in transportation research.” There is support for the expansion of the FHWA’s Advanced Research Program at high levels of government. Also, there are planned increases to the funds allocated to advanced research. The FHWA’s Advanced Research budget has averaged at about $2 million per year. The proposed Transportation Research Bill increases this funding in 2004 to about $7 million per year3. A recent White Paper by Mitretek (Mitretek, 2003) presents a summary and review of the FHWA’s Advanced Research Program and its accomplishments over the past 20 years. In addition, this paper expands on the definition of Advanced Research as presented in TRB’s Special Report 261. Advanced Research types of projects have been conducted by the FHWA for some time. Going forward, there will be an emphasis on developing an integrated and more expansive Advanced Research Program. Also, the procedures for identifying and selecting Advanced Research projects will be modified relative to FHWA’s current process. 2 Section by Section Analysis. Science Committee Transportation Research Bill. November, 19, 2003. 3 This information is out of the Surface Transportation Research and Development Act of 2003. Published November 19, 2003. The Bill shows funding for advanced research going up to $10M in the second year, and $15M per year thereafter.

126

This White Paper focuses on advanced research as applied to the highway safety problem areas. As can be seen in the recent publication in Public Roads, A Decade of Achievement (Livingston, Mills & Oskard, 2002), the FHWA’s Advanced Research Program has focused on both the areas of Operations and Safety, and Infrastructure. The Operations and Safety advanced research has been in the areas of: operations research; sensor applications; artificial intelligence; and analytic tools. As presented in the Mitretek White Paper on Advanced Research, one of the major changes in the near term for this program is the manner in which Advanced Research projects are nominated and selected. Selection of Advanced Research projects are to proceed from the identification of knowledge gaps and external (to TFHRC) input to the program. One of the major sources of information on knowledge gaps and research directions used in the preparation of this paper, are the conference proceedings from the Safety Research Agenda Planning (Research & Technology Partnership, 2002).. This conference focused on identification of critical research needs for:

• Run off road; • Human Factors; • Intersections; • Work Zones; and • Intelligent Infastructure Initiative

From an advanced research perspective, the discussions and recommendations presented in the conference proceedings can be grouped into three major areas:

• Understanding and prediction of driver behavior. This theme was evident throughout the discussion of the various research areas and proposed research efforts.

• Better collection of data to support problem identification and evaluation of treatments (e.g., countermeasures). Also, associated with this theme was the development of analytical tools to consider the impact of specific designs/countermeasures on safety.

• Application of advanced technology to transportation. This was evident both in terms of discussions of countermeasures (e.g., Intelligent Infrastructure treatments) as well as advanced techniques for collecting and analyzing data.

The above grouping is meant to guide the discussion of advanced research and the proposal of a series of research efforts to propose in this White Paper. Advanced Research is problem oriented as is any other research activity conducted by the FHWA. However, some of the defining features of advanced research may include:

• Long-range programmatic research. Advanced Fundamental Research is not a one-shot study effort. This type of research is multidisciplinary and may require years of efforts for results to bear out. Furthermore, specific research conducted under an advanced research effort may not lead to the expected outcome. That is, the model, sensor, or specific technology may not be appropriate or applicable to the problem at hand. One may not learn this after considerable effort has been expended.

• Application of advanced technology or new theoretical models to address safety problem areas.

127

• Cross-cutting in nature. Advanced research efforts may support understanding of problems across a range of safety problem areas (e.g., run off road, intersections) as well as result in countermeasures that that can be applied across a range of problems.

The following section presents specific research topic areas for Advanced Research. The selection of these specific research topic areas were in large part driven by the recommendations and suggestions presented in the proceedings for the Safety Research Agenda Planning Conference. In addition, limited reviews of research and technology were conducted. It is assumed that if a given research topic or area is deemed of interest, additional research will be need to be conducted to refine the research topic. For example, proposed research efforts may include the study and application of new technology, i.e., new developments in artificial intelligence, nanotechnology, advanced data mining techniques and procedures, and advanced sensor technology. These efforts will require additional collaboration with other government laboratories and organizations as well as detailed technology reviews. One of the guiding principles in developing the proposed research topics was to focus on a few areas that had a potential for high pay-off in the long run and that would be cross cutting in its effects. That is, a given successful Advanced Research project could benefit our understanding and amelioration (or elimination) of safety problems across a range of areas, i.e., intersections, run off the road, work zone, etc.

128

6.2 Specific Research Topics Suggested Research Topics/Studies

Category Project Title Type of

Research

Likelihood of Success

(1-5 Scale) Duration(Months)

EstimatedCost

1. Understanding the Driver

ADV 1a: Development of a Driver Modeling Structure

Advanced Very High 5

24 $2M

ADV 1b: Development of a Prototype Driver Model

Advanced High 4

36 4M

ADV 1c: Development of a Driver Model

Advanced Moderate 3

60 15M

2. Data Collection/ Analytical Tools

ADV 2a: Evaluation of Advanced Sensors and Data Mining Techniques

Advanced Moderate 3

36 4M

ADV 2b: Development of Safety Decision Aids for Planners

Advanced Moderate 3

36 5M

3. Advanced Technology for Countermeasures

ADV 3: Evaluation of Nanotechnology for Safety Countermeasures

Advanced High 5

12 250K4

Research suggestions from R&T Partnership Steering Committee If a driver model is developed, research into whether it should concentrate on group rather than individual behavior. Other suggestions include: 1) A program that defines and validates crash “surrogates” so that they can be used in future safety assessments and evaluations (perhaps related to “advanced sensors”), 2) Fundamental research into methods for evaluating the crash-related casual effects of intervention – given the difficulties of having the ability to conduct randomized experiments, and 3) Use of forensic approaches to establishing detailed reconstruction of events – high-tech field data collection techniques, perhaps image processing and high performance computations for reconstructing events.

Response from White Paper authors 1. The White paper included doing a requirements analysis and design of a modeling architecture. However, the tone of this White paper was in the development of a driver model

3 Note: The proposed $250K is for a project to develop a nanotechnology research program at FHWA.

129

for individual behavior. This is a research question that should be addressed in the requirements phase of the work. 2. The development of safety related measures that can be used to predict crashes has been going on for some time. These measures are frequently referred to as safety surrogates. The report by Campbell, Lepofsky and Bittner (2003)5 discusses safety surrogates in detail and data collection methodology to support the development of more valid and reliable safety surrogates. Their report proposes a large scale, multi-year data collection set of efforts. The issues with developing valid and predictive safety surrogates can considered from two perspectives. First, an analytical framework needs to be developed to guide the research. Another term for analytical framework would be a theory. The second major challenge has been that one wants to predict the occurrence of crashes as a function of a host of independent variables. Crashes though numerous are rare events when one considers the total number of miles traveled (exposure). So research attempting to correlate some form of behavior and other factors to crashes requires a large amount of data. Also, the data needs to be collected at a fine level of resolution and it needs to be very reliable. This white paper focuses on the development of an analytical framework or model. Numerous papers on highway safety cite human factors as a major causal factor in the occurrence of crashes. These human factors issues range from driver distraction, inattention, workload, or other theoretical constructs used to try to explain the behavior of the driver under conditions that result in a crash. A driver model as proposed in this paper would serve to develop an analytical framework that can guide research into understanding and predicting driving behavior. Also, the framework would aid research in selecting measures for predicting crashes or the risk of a crash. The focus of this section of the White Paper was on advanced research. So the author chose to focus on the development of the model rather than on the development of correlational studies to define and validate safety surrogates. This is perhaps the more difficult side of the problem, but in the long run may lead to greater payoff. Understanding the Driver The report by Campbell, Lepofsky and Bittner (2003) presents a review of current and projected highway safety issues. They observed that “the interrelationship of driver performance and behavior with roadway design and traffic conditions to affect the risk of collisions and casualties is largely an unknown area, despite the fact that driver behavior is widely believed to be responsible for most collisions.” Reasons cited for this lack of understanding of driver behavior include a lack of detailed information on driving behavior and driving errors. That is, sufficient detailed an reliable data have not available to fully understand the relationship among multiple factors responsible for collisions and casualties. One of the factors most frequently cited as being responsible for most collisions is driver behavior.

5 This report was published and made available to this author after preparation of the first draft of the White Paper on Advanced Research.

130

The driver, vehicle, highway environment is complex and there are host of factors, and the interaction of those factors, that need to be considered when conducting research to understand the factors responsible for collisions. One of the best tools to use in such complex situations are analytical frameworks or models. Such an analytical framework would serve to guide long-range research that can ultimately result in a model can predict driver behavior under a wide range of conditions. Development of such a model would go hand in hand with the collection of detailed driving performance data. Data collection that is guided by an analytical framework or model is more likely to be of use than data that are collected because we now have the technology. The end goal here would be to develop a model that would support design and evaluation of countermeasures. The FHWA has worked on developing driver models for some time. These models have been relatively simplistic as those embedded in traffic microsimulation models. More complex driver models have been developed for the IHSDM project. However, development of driver models has not been a major focus area for the FHWA R&D efforts. Why Develop a Driver Model? In general terms, models are abstractions of reality. A blueprint for a building is a model. Furthermore, blueprints for buildings will “model” different aspects of a building – the plumbing, electrical wiring, etc. This type of model is used to design a specific building for a specific purpose. If we develop a good model (blueprint), we will have a sound, functional building that meets the requirements and all of the components work together (the elevator shaft will not be located on top of the water main, etc.). As with all models, the blueprint for the building will not be perfectly identical to the building that we build. If we do a good job of following the blueprint and follow good building practices, it will be close enough. Models help to define a problem and define requirements. In the area of research, models can help to organize the research activities (e.g., specific experiments or data collection efforts that need to be conducted) and help us to determine where we have knowledge gaps. Models serve as an organizing principle. Models can be used to make predictions and support system development. For example, NASA has a System-Wide Accident Prevention Program. The goal of this multi-year program is to foster the development of new technologies to reduce aviation accidents. One of the specific technologies being developed is a synthetic vision system. To support this program, the Human Error Modeling (HEM) element is investigating the application of human performance models to study the types of errors operators could make when using new technology such as synthetic vision. (Leiden, Laughery, Keller, French, Warmick & Wood, 2001). So the development of a driver model can

• Help in focusing our research efforts. If we have a relationship or function in the model and there are no data to support the function, it will be very evident. The model won’t run or we will need to insert some “engineering estimate” or simplification.

• Help us to integrate our research efforts. For example, we need to tackle intersection collisions, work zone related crashes, and run off the road crashes. However, the human element in terms of risk taking, choice behavior, perception, and so forth is the same across

131

these different types of crashes. Also, models can help us integrate results and support development of a knowledge base.

• Help us to design better systems. If we have a good working model of the driver, different design options for intersections (for example) could be tested well in advance of building the actual system. For example, if we have a good model that incorporates prediction of human error, we may be able to rule out designs that may prove to be unsafe.

6.3 Knowledge Strongholds Significant advances have been made in the area of cognitive modeling. There are a diverse number of models currently in existence. The reason that there exists a diverse set of cognitive models has to do with the level of fidelity that perception, attention, working memory, and decision making are addressed by the individual modeling architectures (See Leiden et al, 2001, for a brief review of relevant cognitive models). Existing cognitive models have been applied to the driving situation. For example, Salvucci employed the modeling architecture of ACT-R (Adaptive Control of Thought- Rational) to model driver distraction from cognitive tasks (Salvucci, 2002). The existing cognitive models present the opportunity to build modeling architectures for driving by re-using existing models. This area is reasonably mature and computational frameworks that incorporate built-in, well tested parameters and constraints on human cognitive and perceptual-motor abilities are available. There is also a large volume of driver performance data in existence. However, determining the degree to which existing driver performance data could support model development would need to be accomplished at a later time. The research program outlined by Campbell, Lepofsky and Bittner (2003) presents a comprehensive and extensive set of field data collection studies. This project presents a potential for providing a rich source of driver performance data. However, data to support the development of behavioral models generally works best when the data collection efforts are designed in support of model development. It is likely that additional data collection would be needed to develop a comprehensive driver model.

6.4 Knowledge Gaps Based on a limited review of the literature, it appears that the application of cognitive models to the driving situation has been limited. Furthermore, the validation and calibration of the models have been frequently done by comparing model performance to performance in high fidelity driving simulators. This can present an added level of uncertainty. That is, ultimately one would want to develop models of the driver that predict performance on the road. High fidelity simulators are also models of the real world and thus have limitations; however, the limitations are in terms of how the vehicle/driving environment are modeled and not the driver’s cognition and decision making. The use of simulators in developing computational models is probably a good idea; however, ultimately the computational model needs to be tested against real-world driving.

132

The limited review of cognitive modeling indicates that there does not exist a rigorous model of multitasking. This will need to be an area of research given that the driving task does indeed entail multitasking. Also, as additional technologies get added to vehicles and more information is provided to the driver from the infrastructure, the driving task is likely to require more multitasking than it does so currently. Though there is a large volume of data on driver performance, it is likely that much of these data may not support a modeling effort. It is not the case that the available data are not useful, rather that it may not be at the level of resolution needed for modeling studies. This is an empirical question that will need to be dealt with if one were to move forward on large scale driver modeling effort. As was mentioned in the proceedings of the Safety Research Agenda Planning Conference, the FHWA has developed a driver model for the IHSDM. This driver model focuses on a narrow question – i.e., speed control through a curve. This is not unusual in the modeling world. That is, a specific question is asked and a specific modeling solution is developed. Generally, the state of the art in driver models is that specific narrow models may be available; however, general models that can be employed across a range of problem areas are not available (this is based on a limited review of the literature). The above also presents a limitation or challenge to the development of a comprehensive driver model. That is, models are built for a given purpose. The development of very general driver model will likely require the development of an overall modeling architecture. Furthermore, since the modeling effort will need to be supported by empirical studies (e.g., simulation studies, field data collection, field experimental studies) an overall research program will also need to be developed. The research program will need to consider data collection efforts being conducted by other programs (e.g., Campbell, Lepofsky, and Bittner). It is suggested that the overall driver modeling program be broken down into three major components:

• Development of a driver modeling structure; • Development of a prototype driver model; and • Development of the general driver model.

6.5 Research Recommendations Driver Model Development It is proposed that the development of a driver model be conducted in a structured series of steps. After completion of each step, the program should be reviewed and the program should be adjusted based on results of completed work. The proposed effort will be conducted over several years and therefore one should take advantage of new developments in cognitive modeling, modeling software, and computing platforms.

133

Development of a driver modeling structure Comment from R&T Partnership Steering Committee This proposed study and the proposed Human Factors program are similar and should be combined into one research program. Another question that comes up is will any driver model be capable of predicting “driver errors” rather than just “average driving behavior”? The research should be coordinated with F-SHRP research and current NHTSA research.

This will be high-risk research (which fundamental research should be). But this idea needs external review by experts who will not benefit from the program. Response from White Paper authors 1. Good points regarding predicting driver error versus average driving behavior. The White presented the development of a driver model within the context of a research program. I see the driver model to be an analytical framework to help direct research and if successful to be used in a predictive mode. The ultimate goal would be to predict driver behavior that leads to crash scenarios. Therefore, the model would need to be able to predict “errors”. 2. I don’t understand the comment “needs external review by experts who will not benefit from the program”. It appears to me that if the program is successful anyone who uses our highways will benefit. Under this task an overall architecture for the driver model would be developed. This will entail identification of requirements for the modeling program. That is, what is to be modeled, under what conditions, and the types of measures that will be estimated. This would most likely require the use of the System Engineering Process and associated techniques and procedures for capturing the system requirements and design. A tool such as the Unified Modeling Language (UML) may prove useful for this project. The to-be developed driver model will be a system, and such following good system engineering development process will support the development of a robust model. Under this project work would also be conducted with respect to review and evaluation of existing cognitive models and other developments in the modeling field. The definition of requirements and system design will drive the model development. However, in parallel with this activity a solid review and assessment of the modeling field will expedite the process. Problem Statement Development of a robust and general cognitive driver model will require the development of an overall modeling architecture. Also, review and evaluation of existing cognitive models will be needed. The results of this project will be a detailed model design that can be used to develop a prototype driver model and an overall research program for the driver model program.

134

Method / Approach This task will entail the use of a systems engineering approach and selected tools for developing the modeling architecture. The UML represents a type of tool that may be used to capture the driver model (requirements and system design) at a detailed level. The project will also entail review of the literature (with respect to the content area of safety, cognitive models, and other modeling techniques and tools) and review and evaluation of models. It is anticipated that models will actually purchased and subjected to detailed review. Project Duration The estimate of two years is to allow for the multiple design reviews that will be needed. The proposed driver model will not be a trivial system and such will require systematic review and assessment as it is being designed. Also, this task will require extensive interactions with researchers in the modeling community as well as those conducting traditional driver performance and safety research. This level of collaboration is also a factor for the proposed project duration. Project Title: Development of a Prototype Driver Model Before moving on to develop a full blown and general driver model, a prototype model would be developed. This may entail focusing on a specific safety problem area (e.g., intersections) and developing working models. Also, the development of the prototype model would entail validation of the developed model. Therefore, this task will also include the conduct of experiments or other data collection efforts. Problem Statement Embarking on the development of a complex and general driver model will be a new endeavor for the FHWA. One way to minimize risk is to develop a prototype model that exercises the capabilities envisioned for the full model; however, focused on a more narrow domain. Based on the crash statistics (GES, 2002) and the results of the Safety Research Agenda Planning Conference, the area of intersections may be a good candidate for the prototype model. The development of the prototype model will give the research team experience in implementing a modeling architecture. Also, development of the prototype model will feed back to the overall model design (e.g., propose changes based on what works and what does not work). In addition, procedures for validating the model will be developed and tested. The developed validation procedures and datasets can also be used in the development of the full-scale model. Under this task modeling software will be developed. Even if this developed software (the prototype) is not totally re-used for the development of the full driver model, there will be an ultimate savings. Given the long range duration of this effort, we need to be aware of new developments in modeling software and more specifically computational cognitive models. It may be the case that better models and modeling tools will be made available once the project goes into the phase of developing the full-scale driver model.

135

Method / Approach This project will entail software development of the driver model. Also, research will be conducted to support model calibration and validation. The specifics of the research efforts will depend on the focus domain for the prototype driver model (e.g., intersections, run off the road, work zones). Project Duration The project duration is anticipated to be about 3 years. This will be a software development effort that will require a significant amount research. The availability of existing computational cognitive models (e.g., ACT-R) will facilitate this process. However, to date there have been very few modeling efforts of this magnitude in the driving domain. Also, the model development will be conducted in parallel with empirical research efforts for collecting the data needed to calibrate and validate the model. It is also anticipated the prototype model will be a useful model for the chosen domain. Project Title: Development of a Driver Model This project would entail development of the full-scale driver model. The products of this project would be the code for the driver model, documentation (user and system documentation), and reports and databases documenting the results of the model validation efforts. Problem Statement The objective of this project would be to develop a computational driver model that has been validated. This will be one of the major challenges in this effort given that this will be a relatively complex model used to predict performance under a wide range of conditions. A detailed model validation plan will need to be developed as part of this effort. As discussed earlier, this is a research program that will entail model development as well as the conduct of empirical studies for calibration and validation of the model. As this project spans several years, we will need to ensure that up-to-date cognitive models and modeling techniques are employed. Rapidly evolving technology or theory can present significant challenges to efforts of this sort. Method / Approach This project is similar to that proposed for development of the prototype model. However, the scale will be much greater in terms of the domains to be modeled. That is, this project will entail model development and testing, conduct of empirical studies to support the modeling effort. This project may also entail a bit of theoretical work. As was mentioned earlier, currently robust models for multitasking do not exist. Either as part of this effort or in collaboration with other researchers in the area, multitasking models for this effort will need to be developed. Project Duration This project is anticipated to be a 5 year project. This will be a major modeling and research program. In principle, well as validation effort, a significant portion of the driver and safety research conducted by the FHWA could be aimed at supporting this effort. Research would be

136

conducted to answer specific research questions and provide input to guidelines and other design support efforts. At the same time the research would support the model development as Data Collection/Analytical Tools Another major area or theme from the Safety Research Agenda Planning Conference was the need for better safety data collection and analysis tools. This was in terms of data for identification of problems (e.g., “what is the relationship of the traffic volumes at intersections and safety’, pg. 26) as well as in the development of tools for planners (Analytical tools – in models for traffic engineers and planners to consider the safety consequences of intersection safety and design”, pg. 27). The report by Campbell, Lepofsky and Bittner (2003) propose the development of research tools and methods as one of the three major focus areas for the F-SHRP Safety Plan. Campbell et al present a review of vehicle-based as well as site-based instrumentation for the collection of driver and vehicle behavior. Most of the studies reviewed by Campbell et al are still in the early stages and a significant amount of data collection or analysis has not been conducted to date. Additional research aimed at developing new data collection tools will need to be coordinated with above efforts. For this area, two research projects are suggested:

• Evaluation of advanced sensors and data mining techniques; and • Development of safety decision aids for planners.

Evaluation of advanced sensors and data mining techniques Comment from R&T Partnership Steering Committee The expected benefit of this proposed project is unclear. One potential use is to develop surrogate measures of safety. In setting the requirements for research in this area, one should also attempt to anticipate future issues and thus future data needs. There’s no mention of data collection and analysis tool development proposed in F-SHRP. Response from White Paper authors The white paper focused on evaluation and development of technology. I am not sure that I want or need to address the issue of safety surrogate measures. The notion of safety surrogates has to do with a hypothesized relationship between a set of measures (e.g., driver behaviors, traffic conditions,) and the probability of a collision. Development of safety surrogates goes well beyond the application of new technology for data collection. The idea of safety surrogates is probably more relevant to the discussion of the development of a driver model. The FHWA has on-going work in the area of sensors and data mining techniques. The report by Mitretek on Advanced Research lists over 22 related projects (in the areas of Automated real-Time Use of Information, and Improved Data Quality). The FHWA has also conducted research in such areas as automatic data collection of vehicle trajectory through the SBIR program.

137

Nevertheless, one of the major themes to come out of the Safety Research Agenda Planning Conference was the need for better data collection techniques and tools along with analytical capabilities. In a related project the National Highway Transportation Safety Administration (NHTSA) has funded the System for Assessment of the Vehicle Motion Environment (SAVME). This project has been going on for over 10 years. This is a very extensive program to develop data capture and analysis techniques of potential safety related measures. The SAVME technique involves imaging motor vehicles in normal traffic from specialized video cameras on roadside towers so as to create permanent track files. The track files quantitatively capture the vehicle trajectories and inter-vehicular clearances that prevail in normal driving. The first production runs of empirical data for this project were reported at the 2001 TRB meeting (Ervin, Bogart & Fancher, 2001; Ervin, MacAdam, Vayda & Anderson, 2001). The analysis entailed evaluation of conflicts in a vehicle string and suggestions for developing crash warning systems. Lessons learned and results from the SAVME project could provide input to this proposed effort. For this proposed research it suggested the focus be on developing, testing, and evaluating methods for safety data collection and analysis. That is, this project would not focus on developing a product per se. Promising technologies and methods identified through this research would then move forward to applied-tools development type of efforts. Some critical issues for this effort will entail development of clear requirements for data collection. Data are generally collected for a specific purpose (e.g., to test a hypothesis, fill in the gaps in a performance taxonomy) and the purposes one has in mind will impact what is collected, where it is collected, the level of resolution, and so on. This tends to be one of the major challenges for developing general tools and techniques for collection of data for complex areas such as those related to highway safety. Problem Statement Conduct research to develop advanced techniques and procedures for collecting and analyzing safety related data. This effort would support research for problem identification (e.g., identifying “unsafe” intersections and relevant performance data) as well as evaluating countermeasures. One of the major challenges will be ensuring that the data capture and analysis requirements across the safety program are met (e.g., intersections, run off the road, work zones). Method / Approach The effort would entail developing data capture and analysis requirements up front. Given the identified requirement, existing or developing technologies and methods would be reviewed. In order to develop specific requirements that can be implemented outside of advanced research, it is suggested that limited testing and evaluation be conducted in the field. That is, conduct a research effort at a demonstration and testing level. Initial focus for this effort may be directed to the area intersections. This area presents perhaps the highest potential for pay-off in terms of having an impact on safety performance. Again, this effort would not directly impact safety performance, but rather would provide tools to efforts investigating safety problems and developing and testing countermeasures. Project Duration

138

This effort will entail a significant up front effort in developing a clear set of requirements. Also, this effort will entail technology evaluation and testing as well limited field tests / demonstrations. At the same time this effort is not envisioned to be a 10+ year effort as the SAVME program. If a set of specific products and tools are to be developed, they should be developed outside of the Advanced Research effort. The results of this effort should translate in a set of system requirements and design that can be implemented as tool/product development effort. Development of safety decision aids for planners Comment from R&T Partnership Steering Committee This is probably more “applied” rather than fundamental research. While need is probably there, more justification and details are needed, along with more on the relationship to current work such as IHSDM, IDAS, NCHRP 8-44, and other initiatives. Response from White Paper authors I think that there are some fundamental issues that need to be resolved in this area. The idea for suggesting this area of work came from the Safety Research Agenda Planning Conference report. Also, experience with the use of such models as IDAs suggests the need for research in this area. NCHRP 8-44 does focus on the development of estimation techniques or models for fatal and injury crashes. The study is employing simultaneous negative binomial models. Furthermore, the type of model being developed is not sensitive to network improvements typically identified in transportation plans6. This study is on-going and reports are not currently available and so a more detailed assessment can not be made at this time. However, if additional research is conducted in this area one will need to coordinate with NCHRP 8-44. IDAS does support decision making for planners. It fits well in to the planning process and uses the results of planning models to estimate the impact of a range of ITS improvements on a network. The safety module in IDAS is extremely simplistic even for the level of analysis that it is used for. The reason for the safety module being simplistic (considers facility type and level of service) is that adequate theoretical models have not been developed to consider safety within this type of modeling environment. Also, there are not sufficient data available to develop robust relationships among highway and traffic variables and the associated probability of a crash. One of the themes out of the Safety Research Agenda Planning Conference was the development of tools to “support traffic engineers and planners in considering the safety consequences of intersection safety and design” (pg. 27). The following quote from the conference proceedings expands a bit on this theme.

6These conclusions are based on review of a presentation on NCHRP 8-44 given at the Safety Conscious Planning Leadership Conference, 2003. This study is on-going and results have not been published.

139

“a justification for the need for this is that there is a lot of development and re-development going on in our urban and suburban areas, and a lot of decisions need to be made on where you put intersections and what type of intersections do you put in. So, we feel we could help traffic engineers and planners explicitly consider safety in that decision-making process if we develop the right tools and models.” (pg. 27) The FHWA has a long history of conducting research and development for tools used by engineers and planners (e.g., CORSIM, IDAS, Quickzone). However, these types of tools do not address the safety area in sufficient detail to support the types of decisions indicated in the above quote. It is suggested that an Advanced Research effort be conducted that ultimately leads to the development of decision support tools. Though the end product of this proposed research effort are tools that can be used in a applied setting, significant fundamental research needs to be conducted to develop safety types of modules that provide valid safety impacts. Research needs to be conducted to define the range of variables to consider in estimating safety impact for models to support the planning process. As mentioned earlier, models such as IDAS have very simplistic safety modules. This is due to the lack of valid and reliable models (e.g., mathematical models) and the data to support the models. Problem Statement Based on input from the Safety Research Agenda Planning Conference and limited and informal discussions with planners7, it appears that this may be a real need that should be addressed in the near-term. The FHWA has embarked on a related project entitled Impacts Analysis Process Model and Tool Development8. The objective of this project is to develop a work zone impacts analysis process (algorithm) that will assist transportation professionals in the selection of work zone impacts mitigation strategies. This proposed project would focus other areas where decision support is needed (e.g., intersections). However, it would also coordinate and share information with other related efforts such as the above cited Work Zone effort. In addition, the proposed research effort would be coordinated with other related projects such as NCHRP 8-44 “Incorporating Safety into Long-Range Transportation Planning”. This project would ultimately result in a model usable by engineers and planners. It is being proposed as an Advanced Research effort since a significant amount of research and analysis will need to be conducted in order to develop a tool that explicitly considers safety in the decision-making process of planners and safety engineers. 4 Given the time frame of the project interviews or discussions were not conducted with Public agency personnel involved in the decision making process that were indicating the conference proceedings. Rather informal discussions were held with personnel at Cambridge Systematics who provide support to State and local Government agencies. If this effort is selected for going forward, obviously systematic interviews and discussions with the end users will need to be conducted. 5 This project is a Task Order under Contract No. DTFH61-01-C-00181. Technical Support to the FHWA Office of Operations.

140

Method / Approach This type of project will entail review of the process that planners and engineers employ in their projects. We will need to ensure that the decision support tools that are developed clearly support their jobs. So the methodology will entail process review and requirements definition. Since the objective is to develop a decision support tool (some sort of software system), a structured engineering system design process should be followed. Also, tools for support in designing the system should be employed (e.g., UML or some other type of system design tool). As part of the requirements definition, specific areas for applying the developed decision support tool should be identified. For example, based on the input from the Safety Research Agenda Planning Conference and crash statistics (e.g., GES 2002), intersections may be a good candidate for the initial decision support tool. Project Duration This project is proposed with 36 month project duration. It is anticipated that it will take approximately one year to do the requirements analysis, survey and collect data, and develop a design of the decision support tool. Year two will entail rapid prototyping and testing of the decision support tool. And finally, year three will entail development of the tool for wider use and testing. Advanced Technology for Countermeasures This area of research is suggested based on review of the current project areas for Advanced Research. The FHWA Advanced Research projects are summarized in Mitretek’s report on FHWA Mission-Oriented Advanced Research: Past-Transition- Future. Also, the article entitled A Decade of Achievement in the November/December 2002 edition of Public Roads presents additional summaries of the Advanced Research projects. A significant portion of this research is aimed at employing advanced technology and methods to solve transportation problems. The current FHWA Advanced Research program includes the area of nanotechnology. The Mitretek report on Advanced Research lists 8 projects in this area with level of funding of $430K and $704K ($150K of this funding comes from State Pooled Fund), for FY 2002 and 2003, respectively. The area of nanotechnology has been receiving a tremendous amount of attention and funding during the last few years. The National Nanotechnology Initiative (NNI) was initiated in 2000 and provided a budget of $422-million in 2001. This was a 56% increase in nano spending from a year earlier (Stix, 2001). In November 2003, the 21st Century Nanotechnology Research and Development Act (bill S.189) was passed. This bill would authorize $3.7 billion over the next four years for federal nanotechnology programs.9 The National Science Foundation predicts that nano-related goods and services could be a $1 trillion market by 2015, making it one of the fastest-growing industries in history. This would be

6 Nanotechnology bill called “historic”. In GovExec.com, December 1, 2003.

141

larger than the combined telecommunications and information technology industries (Ratner & Ratner, 2003). The area of nanotechnology is relatively new. In the late 1970s Eric Drexler developed concepts for molecular nanotechnology at MIT. In 1981 the first technical paper on molecular nanotechnology was published by Drexler (Mulhall, 2002). At the same time, Goodyear has incorporated micro sensors into tires that indicate when they need replacing. Corrosion-resistant nanoparticle coatings are also being applied to the hulls of Navy ships to reduce drag, increase speed, and reduce rust (Uldrich & Newberry, 2003). Though this is a new area and significant research and investment are needed to meet the promise of what this technology has to offer, there are currently practical applications in place. Evaluation of nanotechnology for safety countermeasures Comment from R&T Partnership Steering Committee Unclear to reviewers what nanotechnology is or could possible do for safety – more clarity or examples needed. If this is “technology looking for a safety problem,” are there other such technologies that might be better or equal? Response from White Paper authors The FHWA is currently conducting small-scale studies and projects in the area of nanotechnology. The author suggests that the FHWA develop a coordinated and long range program in this area. In order to realize significant benefits from the application of nanotechnology to the highway safety area, an integrated and long range program is needed. This is not an example of a technology in search of a problem but rather a suggestion that this technology be more fully evaluated and used if it appears that it can be employed in highway safety related applications. The area of nanotechnology is at an early stage of development and it is difficult to predict all of the potential applications at this time. Under this category a single project is being proposed. Given the increase in funding and interest in the area of nanotechnology over the last few years, and the limited amount of work in this are by the FHWA Advanced Research program, it suggested that an initial study be conducted to develop a nanotechnology research program plan focusing on the area of safety. Problem Statement Nanoscience has the potential of developing self-healing materials, nanoscale sensors, applications in medicine, and other areas such as optics, electronics, computer science and so on. The promises of nanotechnology are far reaching. At this stage in the technology development, the FHWA needs to develop a plan for including this rapidly growing technology and science into its Advanced Research Program. Establishing a collaborative relationship with organizations leading this effort in government, industry, and academia should be an early step in this process. For example, the National Science Foundation (NSF) is the coordinator of the NNI program.

142

The FHWA Advanced Research Program is currently engaged in a small number of nanotechnology projects with very modest funding. A program plan for incorporating nanotechnology into the Advanced Research Program in a much larger scale is needed. Method / Approach This effort would entail the development of a program plan in the area of nanotechnology. Review of the technology and coordination with lead organizations in this area will need to be established. Areas of safety where this technology can produce countermeasures or treatments over the near- and far-term should be identified. The results of this effort will be a research program plan with a road map, specific projects, and funding. Project Duration This project is projected to last approximately twelve months. Given the rapid growth of the area and significant increase in funding levels by the Federal Government, it behooves the FHWA to rapidly develop a plan and form the necessary relationships in the near term.

6.6 Summary As mentioned in the introduction to this White Paper, one of the guiding principles was to propose a few Advanced Research projects that were felt to be cross-cutting and with the potential for significant pay-off. The projects can be ranked ordered in terms of potential value and likelihood of success. (1) Understanding the driver Development of a driver modeling structure Development of a Prototype Driver Model Development of a Driver Model This set of projects are overall highly rated in terms of potential pay-off. However, as with any modeling effort of this magnitude there are potential risks. The suggestion of breaking the program into the above three projects serves to minimize risk. Also, this type of structure presents the opportunity of learning from research and analysis efforts and modifying subsequent steps in the research program. (2) Data Collection/Analytical Tools Evaluation of advanced sensors and data mining techniques Development of safety decision aids for planners The decision tools for planners has a very high pay-off potential. Of the above two suggested projects in this area this would be the more highly rated. (3) Advanced Technology for Countermeasures Evaluation of nanotechnology for safety countermeasures

143

The FHWA is already conducting work in this area. There are 8 projects in the area of nanotechnology in the Advanced Research Program. The suggestion here is to more fully evaluate this technology and to develop a more comprehensive research program in this area. The proposed program development task is relatively low risk, however, the potential exists for very high pay-off in the long run. Estimating the impact of nanotechnology on highway safety today would be like trying to predict the impact of modern computers to traffic control and traffic management in the 1950s.

144

7.0 Synthesis The preceding chapters present imposing lists of potential research projects. Since they were written independently, coordination across white papers came from two sources: (1) each author was given the same detailed set of instructions, and (2) during the feedback phase each author received a copy of all the draft white papers. The present chapter is one author’s attempt to integrate and harmonize the diverse set of recommendations. It is most certainly not intended to be the last word on these issues; indeed, the final task of the SOW calls for ratings and project evaluations from a large set of stakeholders. It may well be that the most important legacy of this report is not the sets of particular projects that have been sketched. Instead, the formal distinction between Applied and Advanced projects and the realization that both types of projects are required for a healthy highway R&T are the vital concepts advocated in this report. While there are several valid distinctions between applied and advanced research, one very useful distinction (Kantowitz, 1982) focuses upon uncertainty as a key feature. In applied research the goal is to eliminate, or at least minimize, uncertainty. No professional engineer would wish to be responsible for a bridge or a highway that did not achieve its design goals. Of course, to engineer is human (Petrosky, 1985) and bridges can collapse, especially when pushing the limits of design scalability, as did the Tacoma Narrows bridge fail from wind-induced oscillation. Applied projects should have a high probability of achieving their goals and creating effective countermeasures to improve highway safety. In basic or advanced research, uncertainty is sought. No professional scientist would perform an experiment with an outcome that could be perfectly predicted; this does not advance the search for knowledge. Thus, advanced research is inherently more risky than applied research: its goals are more general, more difficult to evaluate, and many outcomes are possible. So why do advanced research at all? It would be safer to limit research outlays to applied projects with their more certain outcomes. Traditionally, FHWA has followed this conservative course and has avoided most basic research. There is often a long time lag between basic results being available versus being applied to create a measurable benefit to society. For example, a study of key outcomes in research funded by the Department of Defense (Adams, cited by Kantowitz, 1982) found it took up to twenty years before basic research results produced systems deployed in the field. This long delay obscures the importance of basic research and makes it more difficult to appreciate that applied systems have roots in advanced research. But there is a limit to the countermeasures that common sense provides. Some questions are so difficult, e.g., how to model what a driver is thinking, that only basic research can provide an answer. Although rates of highway fatalities are decreasing, the number of deaths is increasing due to increased exposure. This problem was faced and solved by the aviation industry with federal government assistance. Basic research about human capabilities and limitations was used to support the applied research needed to design modern glass-cockpit aircraft that increased transportation safety despite increased exposure (see Billings 1997 for examples of human-centered design improvements). These improvements were grounded in decades of basic research funded by DoD and NASA. Ground transportation must

145

follow this example because sufficient improvements in transportation safety cannot be achieved by only more applied research. Thus, FHWA faces a serious challenge in determining what portion of research resources should be allocated to advanced research and how such projects will be selected and evaluated. Because of strongly divergent views about the disadvantages/advantages of uncertainty, it is problematic for basic researchers to evaluate applied research and for applied researchers to judge basic research. The population of scientist-practitioners who are expert in both areas is quite small. This suggests that, at least initially, decisions about advanced highway safety research will be made primarily by engineers who have not been trained in basic/advanced research techniques. Suggestions for new university programs, comparable to what currently exists in aviation, to bridge this gap are beyond the scope of this report.

7.1 Advanced Projects Table 7.1.1 lists nineteen advanced projects proposed in the preceding white papers. These projects have a mean duration of 42 months and a mean cost of $3.9 million. Advanced research requires more time than applied research. Although advanced research does not necessarily cost more per year of research than applied research, total costs tend to be higher since projects are of longer duration. The need for continuity is greater in advanced research. While applied projects can often be broken down into smaller tasks that are sometimes independent, advanced projects suffer more from interruption. It would be more effective to spend $1 million per year for ten years on a major advanced project, such as development of a driver model, than to spend $20 million over five years.

146

Advanced Projects

ROR 4 Development and Application of a Roadside Inventory Database

IS 1a Magnitude, Characteristic, & Causation of Intersection Accidents

IS 1b Establish Root Causes of Driver Error

IS 2a-1 Safety Impacts of Alternative Intersection Controls

IS 2c Intersection Sight Distance

HF 1a Computational Driver Model: WE (Whole Enchilada)

HF 1b Computational Driver Model: Light

HF 2 Processing Multiple Sources Of Information

WZ 2a Incorporate New WZ Data Elements into CDS Crash Investigations

WZ 3a Feasibility and Validity of Region-wide WZ Crash Risk Estimation Techniques

WZ 4a Improving the Understanding and Measurement of Driver Behavior in High Driver Workload Environments

ADV 1a Development of a Driver Modeling Structure

ADV 1b Development of a Prototype Driver Model

ADV 1c Development of a Driver Model

ADV 2a Evaluation of Advanced Sensors and Data Mining Techniques

ADV 2b Development of Safety Decision Aids for Planners

ADV 3 Evaluation of Nanotechnology for Safety Countermeasures

Table 7.1.1 The most important advanced project is development of a computational driver model. It is reassuring that the authors of the papers recommending this project (HF 1 and ADV 1), one from academia and one from private industry, independently reached similar conclusions about the

147

need for this work, the requirement for a computational rather than a verbal model, its scope and its cost. Two additional white papers also included advanced topics that would benefit from, and support development of, a driver model (IS 1b, WZ 4a). The modal objection raised to this project, stated bluntly, is that it would be impossible to achieve. Some engineers believe it is impossible to fathom the workings of the human mind. It is probably true that the standard tools of civil engineering are not readily adopted to the study of human behavior. However, other engineering domains, such as aviation, have benefited greatly from the tools of human factors, which can fathom and model human cognition and behavior. Indeed, the Human Factors and Ergonomics Society has recently started a new special interest group devoted to modeling human behavior in applied systems. What has been accomplished in other domains can also be accomplished for highway safety and operations. FHWA has made some initial attempts to model driver behavior in very simple ways as components of microsimulation traffic models. A more serious effort was undertaken to improve the driver model in IHSDM. But even this model is relatively simple in its goals; the author of this chapter was the initial Principal Investigator for the project developing this model and so is intimately acquainted with its limitations. For example, the IHSDM driver model is for a two-lane rural road with no other traffic and assumes a fully-functioning driver who is not distracted or fatigued. While the IHSDM modelers built in place holders for other cognitive functions, such as attention, these have not yet been exploited. The approaches taken toward building a driver model are similar in both white papers. First, a review is needed to establish preferred system architecture. There are competing computational tools for this task, and the best approach cannot be specified a priori. The next decision point concerns whether to build a full-blown model from scratch or to try to save money by borrowing from existing, primarily non-driver, computational models. Finally, specific scenarios must be generated: should the model be applied to intersections, rural roads, or to workzones? Ongoing oversight by an independent committee of stakeholders would be helpful in making these decisions. This research should be coordinated with other planned research, such as specified in F-SHRP. However, coordination will not be successful if all that happens is that the F-SHRP data get thrown over the fence to the model builders. There must be a contractual mechanism that allows the contractors doing the model to create some data specifications for the F-SHRP contractors responsible for data collection. The feedback on the draft white papers indicated that the following advanced projects were considered to be important:

• ROR-4 • IS2a-1 • IS2c

However, further evaluation by a larger group of stakeholders is needed.

148

7.2 Applied Projects Table 7.2.1 lists sixteen applied projects proposed in the preceding white papers. These projects have a mean duration of 35 months and a mean cost of $1.7 million. They are a good sample of the kinds of important bread-and-butter issues FHWA has been researching for years in order to improve highway safety. Hence, it does not seem appropriate to single out only one project as being of greatest consequence.

149

Applied Projects

ROR 1 Use of Rumble Strips on Non-Freeways

ROR 2 Development of a System of TCDs to Reduce ROR Crashes at Curves

ROR 3 Optimizing the Net Benefits of Delineation

IS 2a-2 Safety Effects of Alternative Left Turn Phasing

IS 2b Safety effects of alternative signal layouts

IS 3 Effectiveness of various countermeasures for reducing accidents

IS 4 Effectiveness of & Driver Response to Automatic All-red Signal Extension System

HF 3 Understanding Speed Selection

HF 4 Look but not see

HF 5 Design Driver

HF 6 Risk Homeostasis

HF 7 Driving Simulator Validity

WZ 1a Estimate WZ Exposure Characteristics from FMIS

WZ 1b Develop VMT Temporal Distributions to Estimate WZ Exposure

WZ 2b Investigate Likelihood of Work Zone Crash Reporting

WZ 3b Project-Level Crash Consequences of Work Zone Design Features

WZ 4b Evaluate Dynamic Queue End Warning Systems for WZ

WZ 5a Analyze State WZ Monitoring and Management Programs and Procedures

Table 7.2.1

150

The feedback on the draft white papers indicated several projects to be worthwhile:

• ROR-1 [median barriers] • ROR-3 • IS 2a-2 • HF-3 • HF-4 • HF-7

However, further evaluation by a larger group of stakeholders is needed.

151

8.0 References American Association of State Highway and Transportation Officials. (2002). Roadside Design

Guide. Washington, DC: AASHTO. Billings, C. E. (1997). Aviation Automation: The search for a human-centered approach.

Mahwah, NJ: Lawrence Erlbaum Associated. Boer, E. R., Ward, N. J., Yamamura, T., Egami, M., Kuge, N., Manser, M., & Lee, J. (2003).

Difference in car following across driving simulators: the plague of explicatory factors. Paper prepared for DSC – NA Driving Simulation Conference North America 2003. Dearborn, MI.

Brisbane, G. & Vasiliou, A. (2002). Signs of rain. VAISALA News, 159, 30-32. Burns, E.N., Dudek, C.L., & Pendleton, O.J. (December 1989). Construction Costs and Safety

Impacts of Work Zone Traffic Control Strategies, Volume I: Final Report. FHWA Report No. FHWA-RD-89-209, FHWA, U.S. Department of Transportation, Washington, D.C.

Byrne, M. D. & Anderson, J R. (2001). Serial modules in parallel: The psychological refractory

period and perfect time-sharing. Psychological Review 108(4), 847-869. Campbell, K., Lepofsky, M., & Bittner, A. (2003). Detailed planning for research on making

significant improvement in highway safety. Paper prepared for Future Strategic Highway Research Program Transportation Research Board of the National Academies. Washington, DC.

Campbell, K.L., Lepofsky, M., & Bittner, A. (2003). Detailed Planning for Research on Making

a Significant Improvement in Highway Safety: Study 2 – Safety. Retrieved November 17, 2003 from the Transportation Research Board Web site: http://gulliver.trb.org/publications/f-shrp/f-shrp_webdoc_2.pdf

Casteel, D. B. & Ullman, G. L. (1992). Accidents at Entrance Ramps in Long-Term

Construction Zones. Transportation Research Record 1352. TRB, National Research Council, Washington, DC, pp. 46-55.

Czerniak, R.J. (2002). NCHRP Synthesis 301: Collecting, Processing, and Integrating GPS Data

into GIS. Washington, DC: Transportation Research Board. Daniels J., Dixon, K., & Jared D. (2000). Analysis of Fatal Crashes in Georgia Work Zones. In

Transportation Research Record 1715. TRB, National Research Council, Washington, D.C., pp. 18-23.

152

Dearasaugh, Jr., D.W. (1998). Future directions in roadside safety. Transportation Research Circular E-C002. Retrieved November 21, 2003 from the Transportation Research Board Web site: http://gulliver.trb.org/publications/circulars/ec002 /part1.html

Dewar, R. (1992). Driver and pedestrian characteristics. In Pline, James L. editor, Traffic

Engineering Handbook, 4th Ed. Englewood Cliffs, NJ.: Prentice Hall 1-28. Ervin, R., Bogard, S., & Fancher, F. (2001). Considering Radar Detection of Vehicles in a String

for Gaining Situation Awareness of a Propagating Conflict. TRB Paper No. 01-0531, TRB 81st Annual Meeting.

Ervin, R., MacAdam, C., Vayda, A., & Anderson, E. (2001) Applying the SAVME Database of

Inter-Vehicle Kinematics to Explore the Natural Driving Environment. TRB Paper No. 01-0496, 2001. TRB 81st Annual Meeting.

Evans, L. & Wasielewski, P. (1982). Do accident involved drivers exhibit riskier everyday

driving behavior? Accident Analysis & Prevention, 14, 57-64. Evans, L. (1991). Traffic safety and the driver. Evans, Leonard, editor. New York, NY.: Van

Nostrand Reihold Fatality Analysis Reporting System. NHTSA, U.S. Department of Transportation, Washington,

DC. Available for review at http://www-fars.nhtsa.dot.gov. Federal Highway Administration. (2003). Manual on Uniform Traffic Control Devices 2003

Edition. Retrieved November 18, 2003 from the Federal Highway Administration Web site: http://mutcd.fhwa.dot.gov/pdfs/2003/pdf-index.htm

Flowers, R.J. & Cook, J.M (1981). Accident Severity in Construction Zones. Texas

Transportation Institute, College Station, TX. Garber, N.J. and Zhao, M. (2002). Distribution and Characteristics of Crashes at Different

Locations Within Work Zones in Virginia. Paper presented at the Annual Meeting of the Transportation Research Board, Washington, D.C.

General Account Office (2003). Highway Safety: Research continues on a variety of factors that

contribute to motor vehicle crashes. Report to congressional requesters. Washington, DC: Transportation Department

Glennon, J.C., Newman, T.R., & Leisch J.E. (1985). Safety and Operational Considerations for

Design of Rural Highway Curves. FHWA/RD-86/035. Washington, DC: Federal Highway Administration.

Graham, J.L., Paulsen, R.J., & Glennon, J.C. (1977). Accident and Speed Studies in Construction

Zones. Report No. FHWA-RD-77-80. FHWA, U.S. Department of Transportation, Washington, D.C.

153

Haight, F., Joksch, H., O;Day, J., Waller, P., Stutts, J., & Reinfurt, D. (1976). Review of methods

for studying pre-crash factors. Paper prepared for the U.S. Department of Transportation National Highway Traffic Safety Administration, Washington, DC.

Hall, J.W. Chacteristics of Construction Zone Accidents. Transportation Research Record

1230. TRB, National Research Council, Washington, D.C., 1989, pp. 20-27. Hanscom, F. (2003). Research-Gap Identification Background Issues: Human Factors.

Presented at the Safety Research Agenda Planning Conference. Irvine, CA: September 17-18, 2002.

Harwood, D.W. (2002). Research Needs for Increasing Highway Safety Through Infrastructure

Improvements. Presented at the Safety Research Agenda Planning Conference. Irvine, CA: September 17-18, 2002.

Harwood, D.W., Council, F.M., Hauer, E., Hughes, W.E., & Vogt, A. (2000). Prediction of the

Expected Safety Performance of Rural Two-Lane Highways. Federal Highway Administration: FHWA-RD-99-207.

Harwood, D.W., Kohlman Rabbani, E.R., Richard, K.R., McGee, H.W., & Gittings, G.L. (2003).

NCHRP Report 486: Systemwide Impact of Safety and Traffic Operations Design Decisions for 3R Projects. Washington, DC: Transportation Research Board.

Herslund, M. & Jørgensen, N. (2003). Looked-but-failed-to-see-errors in traffic. Accident

Prevention & Analysis, 35, 885-891. Highway Research and Technology: The Need for Greater Investment. National Highway R & T

Partnership, U.S. Department of Transportation, Washington, DC, 2002 Highway Statistics Series. Office of Highway Policy Information. FHWA, U.S. Department of

Transportation. Accessible at http://www.fhwa.dot.gov/ohim/hs00/index.htm, Accessed October 2002.

Hostetter, R.S., Crowley, K.W., Dauber, G.W., Pollack, C.E., & Levine, S. (1982). Determination of

Driver Needs in Work Zones. Report No. FHWA-RD-82-117. FHWA, U.S. Department of Transportation, Washington, D.C.

Joksch, H.C. (1976). Critique of Sam Peltzman’s Study; The effects of automobile safety

regulation. Accident Analysis & Prevention, 8, 129-137. Kallberg, V-P. (1993). Reflector posts – signs of danger? Transportation Research Record,

1403, 57-66. Kantowitz, B. H. (2000). In-vehicle information systems: Premises, promises, and pitfalls.

Transportation Human Factors Journal, 2(4), 359-379.

154

Kantowitz, B. H. (1982). Interfacing human information processing and engineering psychology.

In W. C. Howell & E. A. Fleishman (Eds.), Human performance and productivity. Hillsdale, NJ: Lawrence Erlbaum Associates.

Kantowitz, B. H., & Campbell, J. L. (1996). Pilot workload and flightdeck automation. In R.

Parasuraman & M. Mouloua (Eds.), Human performance in automated systems (pp. 117-136). Mahwah, NJ: Lawrence Erlbaum Associates.

Kantowitz, B. H., Roediger, H. L., & Elmes, D. (2001). Experimental psychology. 7th ed.

Belmont, Ca.: Wadsworth Thomson Learning. Kantowitz, B. H., Roediger, H. L., & Elmes, D. Experimental psychology. 8th ed. Belmont, Ca.:

Wadsworth Thomson Learning. (In press). Kemper, W.J. Lum, H.S., & Tignor, S.C. (1985). The Safety of Narrow Lanes for Traffic

Control at a Construction Site. ITE Journal, Vol. 55, No. 1. Institute of Transportation Engineers, Washington, D.C., pp. 33-38.

Kuo, N.M. & Mounce, J.M. (1985). Operational and Safety Impacts on Freeway Traffic of

High-Occupancy Vehicle Lane Construction in a Median. In Transportation Research Record 1035. TRB, National Research Council, Washington, D.C., pp. 58-65.

Lee, J. Models of driver performance. International Encyclopedia of Applied Psychology. San

Diego, CA: Academic Press (In press). Leiden, K, Laughery, K.R., Keller, J., French, J., Warwick, W., & Wood, S.D. (2001). A review

of human performance models for the prediction of human error. Prepared for: National Aeronautics and Space Administration. System-Wide Accident Prevention Program. Ames Research Center, Moffett Field, CA 94035-1000.

Lerner, N.D., Laneras, R.E. , McGee, H.W., Taori, S. & Alexander, G. (2003) Additional

Investigations on Driver Information Overload. NCHRP Report 488, TRB, National Research Council, Washington D.C.

Levison, W. H., Kantowitz, B. H., Moyer, M. J., & Robinson, M. (1998). A stopping-distance

model for driver speed decision making in curve approach. Proceedings of the Human Factors and Ergonomics Society 42nd Annual Meeting, 2, 1222-1226.

Liu, Y. (1996). Queueing network modeling of elementary mental processes. Psychological

Review. 103(1), 116-136. Livingston, R.A., Mills. M.P., & Oskard, M.S. (2002). A Decade of Achievement. Public Roads.

Turner-Fairbank Highway Research Center.

155

Llaneras, R.E., & Lerner, N.D. (1999). Develop Methods to Determine the Prevalence of Aggressive Driving: A Pilot Study. Washington, DC: National Highway Traffic Safety Administration. Project DTNH22-97-D-15018.

Llaneras, R.E., & Singer, J.P. (2002). Inventory of In-Vehicle Technology Human Factors

Design Characteristics. Washington, DC: National Highway Traffic Safety Administration. DOT HS 809 457.

Mack, A. (2003). Inattentional blindness: Looking without seeing. Current Direction in

Psychological Science. 12:5, 180-184. McGee, H.W. & Knapp, G. (1979). Visibility Requirements for Traffic Control Devices in Work

Zones. Transportation Research Record 703. TRB, National Research Council, Washington, D.C.

McGinnis, R.G. (2001). Strategic Plan for Improving Roadside Safety. Retrieved November 17,

2003 from Transportation Research Board Web site: http://gulliver.trb.org/publications/nchrp/nchrp_w33.pdf

Meyer, D.E. & Kieras, D.E. (1997). A computational theory of executive cognitive processes

and multiple-task performance: I. Basic mechanisms. Psychological Review 104(1), 3-65. Migletz, J., Fish, J.K., & Graham, J.L. (1994). Roadway Delineation Practices Handbook.

Federal Highway Administration: FHWA-SA-93-001. Mitretek (May 2003). FHWA Mission-Oriented Advanced Research: Past--Transition—Future.

Mitretek White Paper, Washington, DC. Model Minimum Uniform Crash Criteria (MMUCC). Report DOT HS 808 745. U.S. Department of

Transportation, Washington, DC, August 1998. Morena, D.A. (2003). Rumbling towards safety. Public Roads, 67,2. Mulhall, D. (2002). Our molecular future: how nanotechnology, robotics, genetics, and

artificial intelligence will transform our world. Prometheus Books. Amherst, N.Y. Najm, W., Koopman, J., Boyle, L., & Smith, D. (2002). Development of Test Scenarios for Off-

Roadway Crash Countermeasures Based on Crash Statistics. National Highway Traffic Safety Administration: DOT HS 809 505.

Najm, W.G., Schimek, P.M., & Smith, D.L. (2001). Definition of the light vehicle off-roadway

crash problem for the intelligent vehicle initiative. Transportation Research Record, 1759, 28-37.

National Automotive Sampling System General Estimates System: Analytical User’s Manual 1998-

2001. (2001) NHTSA, U.S. Department of Transportation, Washington, DC.

156

Nemeth, Z.A. & Migletz, J.M. (1978). Accident Characteristics Before, During, and After Safety

Upgrading Projects on Ohio’s Rural Interstate System. Transportation Research Record 672. TRB, National Research Council, Washington, D.C., pp. 19-24.

Neuman, T.R., Pfefer, R., Slack, K.L., Kennedy Hardy, K., Council, F., McGee, H., Prothe, L.,

& Eccles, K. (2003). NCHRP Report 500, Volume 6: A Guide for Addressing Run-Off-Road Collisions. Washington, DC: Transportation Research Board.

Pal, R. and Sinha, K.C. (1996). Analysis of Crash Rates at Interstate Work Zones in Indiana.

Transportation Research Record 1529. TRB, National Research Council, Washington, D.C., pp. 43-53.

Palmer, J. & Kantowitz, B.H. (1994) Daytime running lights and turn signal marking (Final

Rep. No DOT HS 808 221). Washington, DC: National Highway Traffic Safety Administration.

Peltzman, S. (1975). The effects of automobile safety regulation. Journal of Political Economy

85(4), 677-725. Persaud, B.N., Retting, R.A., & Lyon, C. (2003). Crash Reduction Following Installation of

Centerline Rumble Strips on Rural Two-Lane Roads. Arlington, VA: Insurance Institute for Highway Safety.

Petroski, H. (1985). To engineer is human: The role of failure in successful design. St. Martin's

Press, New York: NY. Pline, J.L. (1991). Driver characteristics for highway safety purposes. Proceedings of the Second

International Conference on New Ways for Improved Road Safety and Quality of Life, 157-159. Tel-Aviv, Israel. Israel Institute of Technology.

Post, T.J., Alexander, G.J., & Lunenfeld, H. (1981). A User’s Guide to Positive Guidance. Report

No. FHWA-TO-81-1. FHWA, U.S. Department of Transportation, Washington, D.C. Ratner, M. & Ratner, D. (2003). Nanotechnology: A gentle introduction to the next big idea.

Prentice Hall, NJ. Research & Technology Partnership (2002). Safety Research Agenda Planning Conference:

Summary and Proceedings. Irvine, California. Richards, S.H. and Faulkner, M.J.S. (1981). An Evaluation of Work Zone Accidents Occurring

on Texas Highways in 1977. Report No. FHWA/TX-81/44+263-3. Texas Transportation Institute, College Station, TX.

157

Rosenthal, T. J., Allen, R. W., Chrstos, J. P., & Aponso, B. (2004). A driving simulator for testing the visibility and conspicuity of highway designs and traffic control device placement. Paper submitted at the 2004 Transportation Research Board Annual Meeting, Washington, DC.

Ross, H.E., Sicking, D.L., Zimmer, R.A., & Michie, J.D. (1993). NCHRP Report 350:

Recommended Procedures for the Safety Performance Evaluation of Highway Features. Washington, DC: Transportation Research Board.

Rouphail, N.M. et al. Comparative Study of Short- and Long-Term Urban Freeway Work

Zones. Transportation Research Record 1163. TRB, National Research Council, Washington, D.C., 1988, pp. 4-14.

Sagberg, F.; Fosser, S.; Saetermo, I.-A F. (1997). An investigation of behavioural adaptation to

airbags and antilock brakes among taxi drivers. Accident Analysis & Prevention 29(3), 293-302

Salvucci, D.D. (2002). Modeling Driver Distraction from cognitive Tasks. Proceedings of the

24th Annual Meeting of the Cognitive Science Society. Sivak, M. (1996). The information that drivers use: Is it indeed 90 percent visual? Perception,

25, 1081-1089. Sivak, M. (2002). How common sense fails us on the road: Contribution of bounded rationality

to the annual worldwide toll of one million traffic fatalities. Transportation Research Part F, 5, 259-269.

Sivak, M. & Flannagan, M.J. (2003). Flying and driving after the September 11 attacks.

American Scientist, 91, 6-8. Sivak, M., Soler, J., Tränkle, U., and Spagnhol, J.M. (1989). Cross-cultural differences in driver

risk-perception. Accident Analysis & Prevention, 21, 355-362. Special Report 260: Strategic Highway Research: Saving Lives, Reducing Congestion,

Improving Quality of Life. (2001). Retrieved November 24, 2003 from the Transportation Research Board Web site: http://trb.org/publications/sr/sr260.pdf

Stix, G. (2001). Little big science. Scientific American.com. Scientific American, Inc. Strickland, R.R., & McGee, H.W. (1998). Evaluation Results of Three Prototype Automatic

Truck Rollover Warning Systems. Proceedings of the 77th Annual Meeting of the Transportation Research Board, Washington, DC.

Summary Report on Work Zone Accidents. Final Report, Standing Committee on Traffic Safety.

American Association of State Highway and Transportation Officials. July 1987.

158

Transportation Research Board (2001). The Federal Role in Highway Research and Technology. Transportation Research Board Special Report 261. National Academy Press, Washington, D.C.

Transportation Research Board (2003). Detailed Planning for Research on Accelerating the

Renewal of America’s Highways, Study 1-Renewal. F-SHRP Web Document 1 for NCHRP Project 20-58. National Academy Press, Washington, D.C.

Uldrich, J., & Newberry, D. (2003). The next big thing is really small: How nanotechnology

will change the future of your business. Crown Business, N.Y Ullman, G. L. (1996). Queuing and Natural Diversion at Short-Term Freeway Work Zone Lane

Closures. Transportation Research Record 1529. pp. 19-26. Ullman, G.L. and Krammes, R.A. (June 1991). Analysis of Accidents at Long-Term

Construction Projects in Texas. Report No. FHWA/TX-90/1108-2. Texas Transportation Institute, College Station, TX.

Ullman, G.L., Holick, A.J. & Turner, S.M. (July 2003). Work Zone Exposure and Safety

Assessment, Draft report submitted to FHWA under contract with Battelle, contract no. DTFH61-96-C-00077. FHWA, US Department of Transportation, Washington, DC.

University of Michigan Transportation Research Institute, Visteon Corporation, & Assistware

Technologies, Inc. (2003). Interim Report on Road Departure Crash Warning Subsystem. Retrieved March 30, 2004 from the National Highway Traffic Safety Administration Web site: http://www-nrd.nhtsa.dot.gov/pdf/nrd-12/FINALINTER9-03.pdf

Wang, J., Hughes, W.E., Council, F.M., & Paniati, J.E. (1996) Investigation of Highway Work

Zone Crashes: What We Know and What We Don’t Know. Transportation Research Record 1529. TRB, National Research Council, Washington, D.C., pp. 54-64.

Wang, J.J. & Abrams, C.M. (1981). Planning and Scheduling Work Zone Traffic Control –

Technical Report. Report No. FHWA/RD-81/049. FHWA, U.S. Department of Transportation, Washington, D.C.

Work Zone Operations: Improving Mobility and Safety on Both Sides of the Barrel, Best Practices

Guidebook. (2000) Report No. FHWA-OP-00-010. FHWA, U.S. Department of Transportation, Washington, D.C.

Young, R.A. & Angell, L.S. (2003). The dimensions of driver performance during secondary

manual tasks. Proceedings of the Second International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design. Park City, UT.

159