Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Field validation for surrogate safety assessment methodology (SSAM) using a multi-purpose micro-simulation El-Basyouny, Karim 2006

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2006-0442.pdf [ 10.28MB ]
Metadata
JSON: 831-1.0063300.json
JSON-LD: 831-1.0063300-ld.json
RDF/XML (Pretty): 831-1.0063300-rdf.xml
RDF/JSON: 831-1.0063300-rdf.json
Turtle: 831-1.0063300-turtle.txt
N-Triples: 831-1.0063300-rdf-ntriples.txt
Original Record: 831-1.0063300-source.json
Full Text
831-1.0063300-fulltext.txt
Citation
831-1.0063300.ris

Full Text

FIELD VALIDATION FOR SURROGATE SAFETY ASSESSMENT METHODOLOGY (SSAM) USING A MULTI-PURPUSE MICRO-SIMULATION by KARIM EL-BASYOUNY B.Sc, United Arab Emirates University, 2 0 0 3 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE STUDIES (Civil Engineering) THE UNIVERSITY OF BRITISH COLUMBIA July 2006 © Kar im E l - B a s y o u n y , 2006 Abstract Several approaches exist for estimating safety ranging from using accident rates to accident prediction models which relate the expected accident frequency at a road location to its traffic and geometric characteristics. In recent years, the usefulness and reliability of accident records has led several researchers to consider surrogate safety measures. Among the most common surrogate measures is the Traffic Conflict Technique (TCT). However, there some shortcomings related to the cost of collecting conflicts data and the reliability of human observers. Several efforts were taken to incorporate safety, in terms of surrogate measures such as TCT, into multi-purpose micro-simulations (MPMS). As a result, the Surrogate Safety Assessment Methodology (SSAM) was developed. The main objective of this thesis is to perform a field validation of the SSAM approach using VISSIM 4.1-12 program. This thesis provides the results of the field validation plan for the Surrogate Safety Assessment Methodology (SSAM) with an aim to compare the predictive safety performance capabilities of the SSAM approach with actual accident experience at Canadian signalized intersections. The validation plan consisted of five tests aiming to quantify the relation between the recorded accidents and simulated conflicts. The first validation test, safety ranking analysis, compares the ranking of intersections from SSAM according to predicted average conflicts rates (ACR) and the ranking of the same intersections using actual accident frequency. The second test repeats the same comparative ranking procedures as test 1, but for sub-sets of accident/conflict types. The third validation test, conflict/accident paired comparison, compares the conflict frequency predicted by SSAM to the actual accident frequency at each intersection. A regression equation that relates actual accidents to the predicted conflicts was developed. The test determines the strength of the relationship between predicted conflicts and actual accidents. ii The fourth validation test, conflict/accident prediction model comparative analysis, determines whether the conflict prediction model can predict risk in a manner similar to the accident prediction model for intersections with the same characteristics. The comparison included several model applications such as the identification and ranking of accident prone locations. The fifth test repeats the same comparative ranking procedures as test 4 , but for sub-sets of accident/conflict types. The results of the five validation tests indicated that the safety measures computed from the simulated conflicts were poorly related to those of actual accidents. In terms of model applications, the results indicate a poor agreement between the identification and ranking of accident prone locations obtained from the conflict/accident prediction models. Furthermore, it was concluded that traffic volumes alone can explain more variation in the occurrence of accidents than simulated conflicts obtained from SSAM. The poor relation between simulated conflicts and actual accidents could be associated with SSAM's sensitivity to the manner by which an intersection was modeled in VISSIM. A number of validation issues were investigated to demonstrate the sensitivity of SSAM's output by varying some of the design parameters. The effects of redefining the priority rules, changing the minimum allowable gap size and the effects of changing the lateral clearance parameters were investigated. The results of these investigations showed that the number of conflicts produced by SSAM varied considerably as a result of changing these parameters. As well, it was found that the lane changing logic in VISSIM had a significant impact on the number of simulated conflicts. In many cases, an abrupt lane changing behavior was noticed. To this time, there is no clear justification for some of the abrupt lane-changing behaviors experienced in some intersections. Although certain measures were taken into account to reduce the effect of that unusual maneuver, this behavior continued to occur and affect the number of conflicts produced by SSAM. iii Table of Contents ABSTRACT » TABLE OF CONTENTS IV LIST OF TABLES Vll LIST OF FIGURES IX ACKNOWLEDGEMENTS X 1.0 INTRODUCTION 1 1.1 BACKGROUND 1 1.1.1 Traffic Safety Problem 1 1.1.2 International Safety Awareness 2 1.1.2.1 Global Awareness 3 1.1.2.2 North American Awareness 4 1.1.2.3 European Awareness 6 1.1.2.4 Summary of International Safety Awareness 7 1 7 . 3 Estimating Road Safety 7 1.2 STATEMENT OF THE PROBLEM 1 0 1.3 THESIS STRUCTURE 1 2 2.0 LITERATURE REVIEW 13 2.1 TRAFFIC CONFLICT TECHNIQUE ( T C T ) 1 3 2.2 TRAFFIC CONFLICT TECHNIQUE ISSUES 1 9 2 .3 TRAFFIC CONFLICT TECHNIQUE MEASURES 2 0 2.4 MICRO-SIMULATION & TRAFFIC SAFETY 2 2 2.5 SURROGATE SAFETY ASSESSMENT METHODOLOGY (SSAM) 2 3 3.0 SIMULATION TOOLS 26 3.1 SSAM SOFTWARE 2 6 3.2 VISSIM: TRAFFIC & TRANSIT SIMULATION 2 7 3.2.1 Modeling Process 28 iv 3.2.2 VISSIM Assumptions 29 3.2.2.1 Intersection Geometry 29 3.2.2.2 Signal Control & Detectors 29 3.2.2.3 Speed Profiles, Vehicle Type Characteristics & Traffic Compositions 30 3.2.2.4 Priority Rules 30 3.2.2.5 Miscellaneous Assumptions 31 3.2.3 Modeling Issues 31 3.2.3.1 Modeling Schemes 32 3.2.3.2 Simulated Accidents 33 3.2.3.3 Lane Changing Behavior 33 3.2.3.4 Modeling Left/Right Turn Tapers 35 4.0 VALIDATION DATA 38 4.1 DATA ASSEMBLY 38 4.2 SELECTION BIAS 39 5.0 METHODOLOGY FOR FIELD VALIDATION 41 5.1 VALIDATION TEST 1: SAFETY RANKING ANALYSIS 41 5.2 VALIDATION TEST 2: SAFETY RANKING ANALYSIS FOR SPECIFIC INCIDENT TYPES.. 44 5.3 VALIDATION TEST 3: CONFLICT / ACCIDENT PAIRED COMPARISON 44 5.4 VALIDATION TEST 4: CONFLICT AND ACCIDENT PREDICTION MODEL COMPARATIVE ANALYSIS 46 5.5 VALIDATION TEST 5: CONFLICT AND ACCIDENT PREDICTION MODEL COMPARATIVE ANALYSIS FOR SPECIFIC INCIDENT TYPES 48 5.6 GOODNESS OF FIT MEASURES 49 6.0 RESULTS AND DISCUSSION 50 6.1 VALIDATION TEST 1: SAFETY RANKING ANALYSIS 50 6.2 VALIDATION TEST 2: SAFETY RANKING ANALYSIS FOR SPECIFIC INCIDENT TYPES.. 52 6.3 VALIDATION TEST 3: CONFLICT / ACCIDENT PAIRED COMPARISON 54 v 6.4 VALIDATION TEST 4: CONFLICT AND ACCIDENT PREDICTION MODEL COMPARATIVE ANALYSIS 56 6.4.1 Development of Conflict Models 56 6.4.2 Identification of Accident Prone Locations (APL) 58 6.4.3 Ranking Locations 59 6.5 VALIDATION TEST 5: CONFLICT AND ACCIDENT PREDICTION MODEL COMPARATIVE ANALYSIS FOR SPECIFIC INCIDENT TYPES 60 6.5.1 Development of Conflict Models for Specific Incident Types 60 6.5.2 Identification of A ccident Prone Locations for Specific Incident Types... 62 6.5.3 Ranking Locations for Specific Incident Types 63 6.6 VALIDATION RESULTS SUMMARY 64 7.0 VALIDATION ISSUES 65 7.1 EFFECT OF REDEFINING THE PRIORITY RULES 65 7.2 EFFECT OF VARYING THE GAP SIZE 68 7.3 EFFECT OF CHANGING THE LATERAL CLEARANCE PARAMETER 70 8.0 CONCLUSIONS 72 9.0 FUTURE RESEARCH 77 10.0 REFERENCES 79 APPENDIX 85 APPENDIX A - CANADIAN (83) INTERSECTIONS DATA SUMMARY 86 APPENDIX B - DEMONSTRATION OF ABRUPT LANE-CHANGING BEHAVIOR 89 APPENDIX C - SUMMARY OF AVERAGE CONFLICTS PER HOUR - INCLUDING AND EXCLUDING SIMULATED ACCIDENTS 94 APPENDIX D - EFFECT OF VARYING THE MINIMUM GAP SIZE 97 vi List of Tables Table 1 - Total Accidents Model for Safety Prediction and Related Applications.... 51 Table 2 - Spearman Rank Correlations based on Total/Severe ACR & Total Accidents Model 52 Table 3 - Crossing Model for Safety Prediction and Related Applications 53 Table 4 - Rear-End Model for Safety Prediction and Related Applications 53 Table 5 - Lane-Changing Model for Safety Prediction and Related Applications.... 53 Table 6 - Spearman Rank Correlations based on ACR & Accident Models for Crossing, Rear-End and Lane-Changing 54 Table 7 - Regression Models for Actual Accidents given Predicted Conflicts 54 Table 8 - Total Accidents as a Function of Total Conflicts (Simulated Accidents Included) 55 Table 9 - Total Accidents as a Function of Total Conflicts (Simulated Accidents Excluded) 55 Table 10 - Total Conflicts Model (Simulated Accidents Included) 57 Table 11 - Total Conflicts Model (Simulated Accidents Excluded) 57 Table 12 - Severe Conflicts Model (Simulated Accidents Included) 57 Table 13 - Severe Conflicts Model (Simulated Accidents Excluded) 58 Table 14 - Number of Intersections Identified as Accident Prone Based on Accident/Conflict Models 59 Table 15 - Spearman Rank Correlations based on Total Accidents Model and Total/Severe Conflicts Models 59 Table 16 - Crossing Conflicts Model (Simulated Accidents Included) 60 Table 17 - Crossing Conflicts Model (Simulated Accidents Excluded) 61 Table 18 - Rear-End Conflicts Model (Simulated Accidents Included) 61 Table 19 - Rear-End Conflicts Model (Simulated Accidents Excluded) 61 Table 20 - Lane-Changing Conflicts Model (Simulated Accidents Included) 62 Table 21 - Lane-Changing Conflicts Model (Simulated Accidents Excluded) 62 Table 22 - Number of Intersections Identified as Accident Prone Based on Accident/Conflict Model Types 63 vii Table 23 - Spearman Rank Correlations based on Accident/Conflict Models for Rear-End and Lane-Changing 64 Table 24 - Number and Type of Conflicts based on Different Modeling Schemes.. 67 Table 25 - Number and Type of Conflicts based on Gap Size 69 Table 26 - Percentages of Locations Exhibiting a Decrease in the Number of Conflicts Due to a Change in Gap Size (Including Simulated Accidents) ... 70 Table 27 - Percentages of Locations Exhibiting a Decrease in the Number of Conflicts Due to a Change in Gap Size (Excluding Simulated Accidents)... 70 Table 28 - Effect of Varying Lateral Clearance (Simulated Accidents Included) 71 Table 29 - Effect of Varying Lateral Clearance (Simulated Accidents Excluded).... 71 viii List of Figures Figure 1. Hyden's proposed relationship between safety critical events according to the Traffic Conflict Technique 15 Figure 2. Potential number of Conflict Points (32) in an Intersection with Full Median Opening 22 Figure 3. First Taper Modeling Setup 36 Figure 4. Second Taper Modeling Setup 36 Figure 5. Queuing Problem due to the Second Taper Setup 37 Figure 6. Effect of Redefining the Priority Rules on Total Conflicts (Including Simulated Accidents) 66 Figure 7. Effect of Redefining the Priority Rules on Total Conflicts (Excluding Simulated Accidents) 67 ix Acknowledgements I wish to thank my supervisor Professor Tarek Sayed, Civil Engineering Department, University of British Columbia, for his guidance, encouragement, patience, and financial support. Indeed, he has been a tremendous help for the past two years. The many hours of discussions, in which he showed his enthusiasm and positive attitude towards science in particular and life in general, kept me on the right track. I would also like to acknowledge the help of my professors: Dr. Jacqueline Jenkins, Dr. Terje Haukaas and Professor Allan Russell, at UBC, who impacted my graduate study through their coursework. Special thanks are due to Dr. Suleiman Ashur, Dr. Yasser Hawas, Dr. Essam Zaneldin, Professor Abdel Mohsen Mohamed, Professor Mohsen Sherif and Professor Aly Namzy, at the Civil Engineering Department, UAE University, for fostering me during my undergraduate study and for their encouragement to pursue my Masters of Applied Science degree abroad. In addition, I wish to thank my colleagues and friends here in Vancouver: Amr Abo El Enein, Mohammed Ammar, Samer El-Houssani, Mohammed El Esawey, Yehya Madkour, Aly Omran, Tarek Abo El Seoud and, especially, Wael Ekiela who not only helped me start up my social life here in Vancouver, but also helped me learn VISSIM, which was a necessary tool to complete a major part of this thesis. I would also like to thank Ramsey Selim my life long friend who continues to show his support and amity, despite living in a different part of the world. Finally, but foremost, I am deeply grateful for my parents and sister for their unconditional support. Their prayers, help and love continue to inspire me to achieve higher goals. x 1.0 Introduction This section is composed of three main parts. The first subsection presents the background information that is necessary for understanding the significance of the research problem. The second subsection describes the research problem and states the research goals. The third and final subsection describes the thesis structure. 1.1 Background This section presents information that reflects the magnitude of the traffic safety problem, a brief overview of how international safety awareness came into place and describes various techniques to estimate road safety. 1.1.1 Traffic Safety Problem The problem of deaths and injury as a result of road accidents is now acknowledged to be a global phenomenon. Authorities in all countries of the world are concerned about the growth in the number of people killed and seriously injured on their roads. In recent years there have been two major studies of causes of death worldwide that were published in 'Global Burden of Disease' (World Health Organization, World Bank and Harvard University, 1996) and in the 'World Health Report - Making a Difference' (World Health Organization, 1999). These publications show that in 1990 road accidents as a cause of death or disability were by no means insignificant, ranking in ninth place out of a total of over 100 separately identified causes. However, by the year 2020 forecasts suggest that as a cause of death, road accidents will move up to sixth place. In terms of 'years of life lost' (YLL) and 'disability adjusted life years' (DALYs) road accidents are expected to rank in second and third place, respectively. The 1999 WHO publication reported that the leading injury related cause of death among people aged 15 to 44 years is traffic injuries. Of the 5.8 million people who 1 died of injuries in 1998, approximately 1.75 million died as a direct result of injuries sustained in a motor vehic le accident. Wor ldwide, the W H O reports that 38 million injuries were received by people involved in motor vehicle acc idents in 1998 ( W H O , 1999). Transportat ion C a n a d a ' s 2004 Annua l report lists that in 2003 (most recent statistics) injury acc idents were reported as 222,260 with 2,766 fatalit ies. Accord ing to the 2004 report by the R o a d Safety and Motor Veh ic le Regulat ion Directorate in C a n a d a , there were approximately 1,670 motor vehicle acc idents every day during 2001, of which about 75 percent resulted in property damage only and 25 percent involved an injury or a death. It was est imated that 7.6 people died on Canad ian roads every day and 606 were injured. The cost to society w a s forecasted to be approximately $26 million in current dollars each day. Accord ing to the 2004 Annua l Traffic Col l is ion Statist ics of the province of British Co lumb ia , a total of 49,478 traffic accidents were reported in the province. A total of 20,300 accidents resulted in injuries and 398 resulted in fatalities. Recogn iz ing the huge socia l costs and loss in human life, several government agenc ies and municipali t ies initiated a number of road safety programs. T h e s e programs were a imed at improving traffic safety and conduct ing research on var ious e lements of the safety program such as : identification of hazardous locations, cho ice of appropriate countermeasures, analysis procedures, etc. However, international awareness regarding road safety has only begun recently, as descr ibed in the next subsect ion. 1.1.2 I n t e r n a t i o n a l S a f e t y A w a r e n e s s In terms of road safety, visible improvements were observed in the 70's and the early 80 's . S u c h minimal improvements were mainly a consequence of local and national awareness of the traffic safety problem. However, the improving trend 2 stabilized when it appeared that new approaches to road safety management and the design of preventive measures were required. Usually, various areas of road safety (such as infrastructure improvements, regulatory measures, education etc.) were not directly incorporated or accounted for in traffic manuals and guidelines. As a result, inconsistent safety measures were implemented as seen fit by specialists in one area or another (Hauer, 1988). In the late 80's, urban road safety studies and policy-making recognized the need to establish a new road safety approach. Not only did this approach recognize the importance of safety based designs and its impacts, but also it did set a standard avoiding the use of irrelevant measures & redundancies and addressing the lack of consistency (Muhlrad, 1987). The approach recognized the fact that road accidents result from multi-factorial processes and that preventive measures addressing the road, vehicles and drivers are dependent on each other (OECD, 1984). The new approach was generally referred to as "integrated safety management". The name is derived from the fact that at some point of the study and decision-making process, safety measures of different kinds are considered, compared and their trades-off discussed. Furthermore, the approach provides explicit examination of safety impacts and quantifies them where possible. Several initiatives were undertaken to provide safety conscious planning. Traffic agencies have modified their manuals and guidelines to reflect these changes. The target of these modifications was to move away from the old rigid standards and to provide adaptable guidelines based on past safety studies and research. The following subsections describe some of the global and continental safety awareness. 1.1.2.1 G L O B A L A W A R E N E S S In 2004, the World Health Organization (WHO) developed a World Report on Road Traffic Injury Prevention (Peden M, et al. 2004). The report presents a 3 comprehensive overview of what is known about the magnitude, risk factors and impact of road traffic injuries, and about ways to prevent and lessen the impact of road accidents. The document is the outcome of a collaborative effort by institutions and individuals. Coordinated by the World Health Organization and the World Bank, over 100 experts, from all continents and different sectors - including transport, engineering, health, police, education and civil society - have worked to produce the report. In conjunction with this effort, WHO has made "Road Safety" the theme of World Health Day on April 7, 2004 and developed informational and educational programs worldwide focusing on road safety for a yearlong campaign. The momentum was fostered by a number of partners including the Permanent Mission of the Sultanate of Oman to the UN, WHO, the World Bank, United Nations Children Fund (UNICEF), the United Nations Development Programme (UNDP), the FIA Foundation for the Automobile and Society, the Bone and Joint Decade, Association for Safe International Road Travel (ASIRT), Global Road Safety Partnership (GRSP) and the Task Force for Child Survival and Development who formed a Global Road Safety Steering Committee to address global road traffic safety. The aim of these initiatives was to create world-wide road safety consciousness. 1.1.2.2 N O R T H A M E R I C A N A W A R E N E S S In the United States, the Intermodal Surface Transportation Efficiency Act of 1991 (ISTEA) was announced with an aim to change the focus of highway and transit programs. The act shifted the focus from construction to preservation with emphasized mobility and environmental protection. The main purpose of the legislation was to produce a transportation system that provides safe and efficient mobility and accessibility as well as protection of the human and natural environments. However, nowhere in the legislation was safety specifically mentioned or mandated in the planning process (Depue, L. et al. 2001). ISTEA was reauthorized in 1998 by a bill titled Transportation Equity Act' for the 21st Century (TEA-21). The legislation required the incorporation of safety and 4 security as a priority factor in their respective transportation planning processes and activities. This marked the first time safety was explicitly included as a planning factor for consideration in developing metropolitan and statewide transportation plans and programs. Prior to TEA-21, safety was sometimes a prominent factor in project development and design, but this legislation called for safety consciousness in a more comprehensive, system-wide, multimodal context (Depue, L. et al. 2001). Approximately 40 professionals convened in May 2000 to initiate a discussion on the TEA-21 safety-planning factor. The experts explored the independent planning processes, to identify data, tools, partners and other resources that are currently available or need to be developed for implementing the safety requirement. The following organizations were responsible for planning and organizing the workshop: Federal Highway Administration (FHWA), the Federal Transit Administration (FTA), the Transportation Research Board (TRB), the Roadway Safety Foundation (RSF), the National Association of Governor's Highway Safety Representatives (NAGHSR), the Federal Motor Carrier Safety Administration (FMCSA), the National Highway Traffic Safety Administration (NHTSA), and the National Safety Council (NSC). Furthermore, the transit Cooperative Research Program (TCRP) and the National Cooperative Highway Research Program (NCHRP) were delegated the task of issuing proposals to develop model impact estimates for different types of safety improvements (Depue, L. et al. 2001). In Canada, the Canadian Council of Motor Transport Administration (CCMTA) was initiated in 1987. The organization included representatives of the provincial, territorial and federal governments of Canada. CCMTA are responsible for making decisions on administration and operational matters dealing with licensing, registration and control of motor vehicle transportation and highway safety. In the fall of 2000, CCMTA imitated the Road Safety Vision (RSV) 2010. The RSV was a national effort at making Canada's roads the safest in the world, and its road 5 safety targets were officially endorsed by all ministers responsible for transportation and highway safety (CCMTA, 2002). The RSV should provide Canada's road safety community with benchmarks against which to develop new strategies and measure intervention efforts. The RSV comprises a broad range of initiatives that focus on users, road networks and vehicles and is specifically aimed to raise public awareness of road safety issues. Furthermore, the RSV attempts to improve communication, cooperation and collaboration among road safety agencies, enhance enforcement measures and improve national road safety data quality and collection (CCMTA, 2002). The RSV is supported by all levels of government, as well as some public and private sector stakeholders, including the enforcement community. To this end, the Royal Canadian Mounted Police (RCMP), Canadian Association of Chiefs of Police (CACP), provincial and municipal police forces have incorporated Road Safety Vision 2010 targets into their business plans (CCMTA, 2002). 1.1.2.3 E U R O P E A N A W A R E N E S S The European Transport Safety Council (ETSC) is a Brussels-based independent organization dedicated to the reduction of transport deaths and injuries in Europe. Founded in 1993, ETSC provides an impartial source of expert advice on transport safety matters to the European Commission, the European Parliament, and Member States. ETSC seeks to identify and promote effective measures on the basis of international scientific research and best practice in areas which offer the greatest potential for a reduction in transport crashes and casualties. Current ETSC activities focus on improving road safety in the countries of Southern, Eastern and Central Europe, whose road risk is higher than the European average (the so-called SEC Belt). ETSC is also working to improve the enforcement of traffic law, which is one of the most effective measures to reduce the risk of traffic accidents across the EU-25 (Maier, R. 2006). 6 In 2004, the Organization for Economic Co-operation and Development (OECD) and the European Conference of Ministers of Transport (ECMT) established the Transport Research Centre. The centre was created by merging the OECD's Road Transport and Intermodal Linkages Research Programme with the ECMT's Economic Research activities. All 50 OECD and ECMT countries are full members of the Joint Transport Research Centre (JTRC). In 2005, the JTRC started working on a project titled "Achieving Ambitious Road Safety Targets". Recognizing the importance of establishing a road safety philosophy, the project aims to assist high level decision making, in Europe and a number of countries around the world, on road safety priorities and strategies by providing reliable targets (OCED/ECMT Transport Research Center, 2006). 1.1.2.4 SUMMARY OF INTERNATIONAL SAFETY AWARENESS All these initiatives have been put in place in the last 10 or more years. Consequently, several studies and research programs were initiated to examine the safety of a particular entity. These studies employed several techniques to estimate the safety or risk levels. These techniques were developed based on several years of research and validations. The subsequent section describes several of the road safety measures that are currently used by engineers and practitioners. 1.1.3 Estimating Road Safety Several approaches exist for estimating safety ranging from simply using accident rates to accident prediction models which relate the expected accident frequency at a road location to its traffic and geometric characteristics. Several researchers (Hauer 1986; Higle and Witkowski 1988; Sayed and Abdelwahab 1996) have shown that the relationship between accident frequency and exposure is frequently nonlinear which indicates that accident rates are not appropriate representatives of safety. This finding has led most safety researchers to discard the use of accident rate as a measure of road safety and currently, accident prediction models constitute the primary tools for estimating road safety. 7 There are several regression techniques to develop accident prediction models. The model development and subsequently the results are strongly affected by the choice of the regression technique used. The earlier models were developed using ordinary or normal linear regression. These models assume a normal error structure for the response variable, a constant variance for the residuals, and the existence of a linear relationship between the response and explanatory variables. Such models were criticized by several researchers (Jovanis and Chang 1986; Hauer et al. 1988; Miaou and Lum 1993) indicating that traffic accidents are discrete, nonnegative and rare events that could not be adequately modeled using conventional linear regression. Such limitations led to the use of a Poisson or a negative binomial error structure when modeling the occurrence of traffic accidents. The main advantage of the Poisson error structure is the simplicity of the calculations. However, this advantage is also a limitation. Kulmala (1995) has shown that most accident data are likely to be over-dispersed (the variance is greater than the mean) which indicates that the negative binomial distribution is usually the more appropriate assumption. In most models, the predicted accident frequency varies as a function of traffic volumes, geometry and traffic features. In the majority of road safety studies accident records were used as the primary source of data. However, this was not always the case as accident statistics are still unreliable. In many cases accidents are not being attended and/or recorded in a systematic manner by police officials. Consequently, the reliability and usefulness of the accident data are often considered suspect. If accident data are faulty, caused either by a reduction in reporting levels or from the deterioration in the quality, then the ability of road safety professionals to engineer solutions to address problems may be severely jeopardized (de Leur and Sayed, 2001). Even in the more industrialized countries where safety information systems have been developing 8 over a long period of time, data is still often found unsatisfactory and a thorough safety diagnosis requires complement measures (Nilsson G., 1989). Due to the lack or poor quality of accident data several other surrogate measures were proposed as traffic safety estimates, de Leur and Sayed (2001) suggested the use of claims data from auto insurance companies. The study concluded by recommending the use of claims data as an alternative to the degrading accident records. A number of alternative traffic safety measures were proposed by several researchers (Thompson and Perkins 1983; Perkins and Bowmen 1986; Fitzpatrick 2000) such as: delay, travel time, approach speed, percent stops, queue length, stop-bar encroachments, red-light violations, percent left turns, spot speed, speed distribution and deceleration distribution. No attempt was made to relate these measures quantitatively to accident rates, but rather to assert such rules-of-thumb as "more stop-bar encroachments indicates higher probability of accidents", "longer queues indicate higher probability of accidents," and so on. Nevertheless, such measures were referred to in the literature as possible surrogates for the deteriorating accident records. After recognizing the importance of traffic safety, a number of governmental agencies and authorities have initiated traffic safety programs. The success of these programs depends mainly on estimating traffic safety for a specific entity. For practical reasons traffic accidents are considered the best traffic safety estimates. However, in the case where traffic accidents data are unavailable or corrupted other surrogate measures of traffic safety were developed. A widely common surrogate measure for traffic safety is the Traffic Conflict Technique (TCT). Several studies (Perkins and Harris 1968; Hyden 1975; Migletz et al. 1985; Sayed 1995) have shown the effectiveness of using TCT in performing safety studies. Yet there are a number of limitations for the use of the TCT. The process is not only expensive; it includes appraisals by unreliable subjective observers. 9 Consequently, a new less expensive approach for assessing the safety of a particular entity has been developed. Various researchers (Cooper and Ferguson 1976; Hernandez 1982; Sayed et al. 1994; Mehmood et al. 2001) proposed the incorporation of safety surrogate measures into micro-simulations. The use of micro-simulation to assess the safety of a particular design or configuration would provide valuable insights into the relative safety impacts. Nevertheless, despite the twenty-five years span of research into the topic of micro-simulation and safety, the concept is still not fully developed. Most of the research on this topic has been using special-purpose simulations. Recently, several efforts were made to adjust multi-purpose micro-simulations (MPMS) to account for traffic safety. Perhaps the most comprehensive work in that regard was carried out by Gettman and Head, in 2003. The authors developed an application for visualization, analysis and assessment of surrogate measures of safety for intersections from existing traffic microscopic simulation models. The process was denoted as Surrogate Safety Assessment Methodology (SSAM). The SSAM software includes a set of predefined surrogate safety measures, such as the Minimum Time-To-Collision (MTTC) and Minimum Post-Encroachment Time (MPET). After analyzing an MPMS output file, the SSAM software would extract, summarize and present the results. The next section provides the main purpose of the thesis and states the research problem. 1.2 Statement of the Problem The main purpose of this thesis is to compare the predictive safety performance capabilities of the SSAM approach with actual accident experience at Canadian signalized intersections modeled in VISSIM. VISSIM is one of the MPMS software that was adjusted to be compatible with SSAM. It is a microscopic, time step and behavior based simulation model developed to analyze the full range of functionally classified roadways operations. VISSIM was chosen for its open interfaces compatibility, aptitude to analyze networks of all sizes and its ability to provide its 10 users with the capability to model any type of geometric configuration or unique operational/driver behavior encountered within the transportation system. The aforementioned purpose is accomplished by performing a field validation for the SSAM approach to investigate whether SSAM has capabilities to predict safety performance beyond what can be obtained through a simple relationship based on exposure. The validation plan consists of five tests aiming to quantify the relation between the recorded accidents and simulated conflicts. The first validation test, safety ranking analysis, compares the ranking of intersections from SSAM according to predicted average conflicts rates (ACR) and the ranking of the same intersections using actual accident frequency. The second test repeats the same comparative ranking procedures as test 1, but for sub-sets of accident/conflict types. The third validation test, conflict/accident paired comparison, compares the conflict frequency predicted by SSAM to the actual accident frequency at each intersection. A regression equation that relates actual accidents to the predicted conflicts was developed. The test determines the strength of the relationship between predicted conflicts and actual accidents. The fourth validation test, conflict/accident prediction model comparative analysis, determines whether the conflict prediction model can predict risk in a manner similar to the accident prediction model for intersections with the same characteristics. The comparison included several model applications such as the identification and ranking of accident prone locations. The fifth test repeats the same comparative ranking procedures as test 4, but for sub-sets of accident/conflict types. At the end, the strengths and weaknesses of the SSAM approach to predict safety performance are discussed. 11 1.3 Thesis Structure There are ten main sections to this thesis. The first section provides an introduction to the thesis by presenting background information, stating the problem and describing the thesis structure. The second section presents a comprehensive literature review about the Traffic Conflict Technique as a surrogate safety measure. The review provides a timeline of the most important studies conducted on the TCT as well as some of its limitations and present applications. The third section provides an overview about the simulation tools that were used in the analysis namely; SSAM and VISSIM. The fourth section describes the methodology guidelines followed by the data assembly procedures in section five. The sixth section presents the validation results with discussion. The seventh section discusses some important validation issues. Section 8 concludes the thesis, while section 9 states the future research. Finally, section 10 lists the references. 12 2.0 Literature Review The objective of this section is to provide a review of the Traffic Conflict Technique (TCT). The first subsection provides a historic timeline about the TCT starting from its introduction in the late 1960's to its current state-of-the-art form. The studies reviewed provide valuable insight about the development of the technique. These studies include, but are not limited to, motivation of TCT, various definitions of TCT, the severity hierarchy of accidents/conflicts, relating conflicts to accidents or volumes, using conflicts as surrogates to traffic accidents, using simulation models to predict safety, guidelines and manuals demonstrating the appropriate use of the TCT and an overview of the evolution of the TCT. The second subsection describes some important issues related to the usability, validity and reliability of the technique. The third subsection presents some of the common measures employed in the TCT. The fourth subsection demonstrates the use of micro-simulation as a traffic safety tool. The subsection sheds some light on the efforts conducted in incorporating safety in micro-simulation. The fifth and final subsection describes the SSAM approach to estimating safety using surrogate measures based on the TCT. 2.1 Traffic Conflict Technique (TCT) Different road safety measures were developed to address the shortcomings associated with the quality and availability of accident data. Of the surrogate measures proposed to address road safety, the Traffic Conflict Technique (TCT) stands out. The TCT was developed from the original research at General Motors laboratory in the late 1960's for identifying safety problems related to vehicle construction (Perkins & Harris, 1968). The technique was initially developed to form a set of definitions and procedures for observing traffic conflicts at intersections. The study by Perkins & Harris related conflict patterns to accident types. They found that the rate of occurrence of serious conflicts is a more useful measure of risk than accident rate. This was also confirmed by William (1972) who analyzed a large set of field data to 13 conclude that the factors contributing to accident causation could be explained more readily by using a number of possible conflicts than conventional accident analysis techniques. Further motivation for the use of TCT was provided by Zegeer and Dean (1978) who studied traffic conflicts as a diagnostic tool in highway safety. They suggested using TCT as a surrogate measure for safety, indicating that in recent years accident data have been degrading in quality. Moreover, accidents records are subject to human error in recording and most importantly the waiting time to obtain a sizable sample to perform a safety study might be too long. As a result of such a long waiting time, several collisions may occur before any correction could be made to treat or correct a deficiency. Several definitions exist for the term "traffic conflicts". In the early 70's, Spicer (1971) defined a serious conflict as "a situation where a vehicle makes a sudden rapid deceleration or lane change to avoid accidents". Spicer's definition of traffic conflicts was among the first attempts at identifying a conflict and its severity. Perhaps the most adopted and widely used definition of traffic conflict, was that of Hyden in 1975. Hyden defined a conflict as "a situation where two road-users would have collided had neither of them made any kind of aversive maneuver". The point at which the aversive action is taken is recorded through observation of Time-to-Collision (TTC) (Hyden, 1975). The TTC value together with the conflicting speed is used to determine the conflict severity. Hyden assumed that the shape of the severity hierarchy is a three sided pyramid as shown in Figure (1). To date there has been no further publications on the proposed severity hierarchy. 14 Figure 1. Hyden's proposed relationship between safety critical events according to the Traffic Conflict Technique. There were several studies relating conflicts to accidents or volumes and using conflicts as surrogates to traffic accidents. For example, Spicer (1972) examined the variation of accidents, conflicts, and the product of intersecting flows with time of day. The study recorded a high correlation among the various entities. Continuing his initial work, Spicer (1973) correlated serious conflicts with reported injury accidents. The study made use of a data set from six intersections, and concluded with a positive correlation between serious conflicts and reported injury accidents. However, the technique was based on observer judgments using time-lapse filming, thereby proving costly and time-consuming. Hyden (1975) collected accident and conflict data for 50 intersections and attempted to find a correlation between reported injury accidents and observed serious conflicts. The study concluded by finding a positive correlation based on many parameters such as speed and so forth. Darzentas et al. (1980) attempted a study linking the driver age and gender to the conflict rate. The study used empirical data to develop an event-stepping discrete simulation model of a non-urban T-junction. The model predicted the number and severity of conflicts as a function of various traffic and behavioral parameters. In the 15 early 1990's, Crowe (1990) used a number of un-signalized three leg intersections to determine the average number of conflicts. The study used a negative binominal distribution to rank hazardous intersections. The findings of the study show a relationship between the traffic volumes at the intersection and the number of conflicts observed. Sayed (1997) conducted a study to estimate the safety of un-signalized intersections using TCT. He developed standards that allow for a relative comparison of conflict risk using an Intersection Conflict Index. Additionally, regression analyses have been used to develop predictive models that relate the number of conflicts to traffic volume and accidents. This work was followed up with two more studies (Sayed and Zein 1998, 1999) where traffic conflict models and standards for signalized and un-signalized intersections were examined. Rao and Rengaraju (1997) used a probabilistic model to determine the number of conflicts at urban uncontrolled intersections. The model was validated using field data. The study produced a model that could be used for quick estimation of possible conflicts at low volume urban uncontrolled intersections. The study concluded by finding an increase in the number of conflicts when there is an increase in flow on the cross road. Katamine (2000a,b) preformed two TCT studies using the same dataset composed of 15 four-leg un-signalized intersections, captured by a video camera. The first study examined the correlation between various volume definitions and conflicts. The study showed how various volume classifications are related to the number and severity of conflicts. The author indicates that these relationships could aid highway and traffic engineers with the type of volume that they should consider, when they are dealing with evaluating a specific conflict type that occurs most, for example, at similar intersections. The objective of the second study was to correlate primary conflicts with secondary ones. The concept of secondary conflicts was first introduced by Gluaz and Migletz (1980) and was later on expanded by Parker and Zegeer (1989). Katamine defined a secondary conflict as an evasive maneuver occurring when a second road user showed an obvious braking or swerving, as a result of a previous evasive maneuver. The study found a significant correlation between primary and secondary conflicts. Therefore, the reduction of total and severe conflicts would not only reduce the number of consequent accidents, but also 16 reduce the frequency of secondary conflicts and their related accidents. The study showed that through-cross traffic conflicts, same direction conflicts, and other types of conflicts can produce severe secondary conflicts. The study also showed that through-cross traffic conflicts are more closely associated with secondary conflicts than same-direction conflicts with their secondary conflicts. Other studies advocated the use of conflict based simulation models to predict safety. For instance, to assess T-junctions, Cooper and Ferguson (1976) developed a simulation model using conflicts as the road safety performance measure. Although their work was promising in terms of using simulation models to predict the safety of a particular entity, they neglected to consider the importance of uncertainty and the complexity of human driving behavior. To predict the conflict rate in turning maneuvers, McDowell et al. (1983) used a traffic conflict simulation to measure the gap acceptance behavior of drivers. The study is widely popular for its relationships based on gap acceptance and is considered to be among the very early attempts to incorporate safety into a simulation model. Moreover, Migletz et al. (1985) developed a methodology to predict traffic accidents from observed conflicts. The study concluded that the numbers and severities of conflict events have an established statistical relationship with accidents, and in some cases can be a better predictor of the expected number of accidents than historical accident data. Sayed et al. 1994 developed a simulation software called "TSC-SIM" to simulate traffic conflicts at T-and 4-leg un-signalized intersections. The software examined traffic conflicts as critical-event traffic situations, the effect of driver and traffic parameters on the occurrence of conflicts. The model was unique for combining an importance sampling technique while storing the traffic conflicts that occur during the simulation. A graphical animation display was used to show how these conflicts occurred and the values of critical variables. Following their initial work in 1997, Rao and Rengaraju (1998) modeled conflicts of heterogeneous traffic at urban uncontrolled intersections. The study describes the methodology of simulating the traffic flow and thereby estimating the number of conflicts in varying traffic flow conditions. The arrival pattern of vehicles was represented by a multivariate distribution to generate 1 7 input to the simulation model. The model was validated using field data. The study found that incremental increases in traffic volumes resulted in an increase in the number of conflicts. In an attempt to provide general guidelines for performing TCT studies, Parker and Zegeer (1988, 1989) developed manuals suggesting procedures to execute TCT studies. In 1988, Parker and Zegeer published the TCT Safety and Operations: Observers Manual. The manual provides basic background information and step-by-step procedures for conducting traffic conflict surveys at signalized and un-signalized intersections. The manual was prepared as a training aid and reference for practitioners and engineers. Additionally, the manual contains definitions with illustrations and examples of conflict types, and instructions for conducting the field activities, including the time schedules, forms, and other details. In 1989, Parker and Zegeer published the TCT Safety and Operations: Engineer's Manual. The manual discussed issues regarding the observer training techniques, as well as procedures for analyzing and interpreting the results of conflict surveys. The authors indicate that in order for practitioners and engineers to diagnose safety and operational problems and to evaluate the effectiveness of treatments, a proper methodology has to be established for the TCT. Also, in the mid-1990's, Hummer et al. (1994) provided guidelines and procedures for various aspects of the TCT. The study suggested procedures for training observers, proper choice and size of sample, survey methodology and data analysis techniques. In the manual, Hummer defined a traffic conflict as "interactions between two or more vehicles or road users when one or more vehicles or road users take evasive action, such as braking or weaving, to avoid an accident." Furthermore, the manual suggests using brake lights, squealing tires, or vehicle front ends that dip or dive as indications that braking occurred and a conflict was possible. Moreover, the manual indicates that an accident or near miss during which no evasive actions were observed also counts as a traffic conflict. Glennon et al. (1974) provided a detailed overview about the evolution of the TCT, summarizing and discussing the evaluation of earlier studies. The authors efforts 18 were among the very first to provide a comprehensive summary about the TCT, which evidently led more researchers to recognize and start applying this technique in their studies. Another extensive research on conflicts was conducted by Glauz and Migletz (1980). Their research provided standard definitions, refined data collecting procedures and a guideline to the application of the TCT. The study examined accident records and conducted conflict analysis. The review indicated that the predominant types of intersection accidents (opposing left-turns and cross-traffic, or side-angle types) were associated with their conflicts counterparts. Also, Zein et al. (1995) performed a study summarizing the advancement in traffic conflicts research. The study contained a review of traffic conflict standards and simulation. The above studies summarize approximately 30 years of research on the TCT. The next section attempts to answer some of the questions regarding the usability, validity and reliability of the TCT (Archer, 2001). 2.2 Traffic Conflict Technique Issues In its initial implementation, a number of researchers questioned the reliability of the TCT, as the process seemed to be highly dependent on the subjective individual observations by trained professionals. Using data from five different countries, Malaterre and Muhlrad (1979) found a substantial measure of agreement in identifying conflicts. However, there was no agreement on definition of severity classes. This work was among the first attempts to examine the application of the TCT in different countries. The conclusions derived from this study provided further proof about the usability of the TCT. Additionally, several studies (Glauz and Migletz 1980; Parker and Zegeer 1988, 1989) showed rates of up to 80% agreement between different observers. In terms of validity, several researches questioned the TCT ability to be an accurate proximal measure of accident occurrence. Once again, a number of studies (Migletz et al. 1985, Svensson 1992) showed that observed conflicts are good indicators of traffic accidents. Further concerns about the process validity were raised. The term 19 process validity means that the processes which determine conflicts are the same as those that determine accidents (Archer, 2001). A study by Hyden (1987) concluded that conflicts and accidents did in fact share the same severity distribution (see Figure 1). Generally, conflicts result from a lack or misunderstanding of communication between the different road users. Therefore, several studies (Risser 1985; Archer 2000) have shown that higher rates of traffic conflicts can indicate lower levels of safety, as it is likely for that particular entity to be accident prone. 2.3 Traffic Conflict Technique Measures Several TCT measures have been developed and used to assess safety at particular facilities. Among the very common TCT measures is the Time-to-Collision (TTC). Hayward (1972) and Hyden (1987) were among the first researchers to use the TTC measure. Hayward (1972) defined TTC as "the time it takes for two vehicles to collide if they continue on their present trajectory at the same speed". These authors have indicated that TTC is the surrogate measure of traffic safety. The measure has been widely accepted due to its simple computation procedure and its ability to indicate the severity of an accident. It was argued that lower TTC values would indicate higher accident severity with a TTC of zero being an accident. However, Kruysse (1991) and Tiwari et al. (1995) refuted that lower TTC indicates higher severity of accidents, primarily because speed is not included in the measure. The argument was although a lower TTC could indicate a higher probability of accident; it fails to recognize the severity of the accident. Cooper and Ferguson (1976) and Darzentas (1980) derived another TCT measure, called deceleration rate (DR). The DR was defined as "rate at which crossing vehicle must decelerate to avoid an accident", and was recommended as the primary indicator of severity instead of TTC. 20 Allen et al. (1978) and Ghaffari (1990) summarized a number of other TCT measures; such as the gap time (GT), encroachment time (ET), post-encroachment time (PET), initially attempted post-encroachment time (IAPET) and proportion of stopping distance (PSD). The definitions for each of these measures are given below: o Gap Time (GT): time lapse between completion of encroachment by turning vehicle and the arrival time of crossing vehicle if they continue with same speed and path. o Encroachment Time (ET): time duration during which the turning vehicle infringes upon the right-of-way of through vehicle, o Post-Encroachment Time (PET): time lapse between end of encroachment of turning vehicle and the time that the through vehicle actually arrives at the potential point of collision, o Initially Attempted Post-Encroachment Time (IAPET): time lapse between commencement of encroachment by turning vehicle plus the expected time for the through vehicle to reach the point of collision and the completion time of encroachment by turning vehicle, o Proportion of Stopping Distance (PSD): ratio of distance available to maneuver to the distance remaining to the projected location of collision. Recently, Minderhood and Bovy (2000) developed two new TCT measures based on the TTC. The first measure is the Time Exposed TTC which measures the length of time that all vehicles involved in conflicts spend under a designated TTC minimum threshold during a specified time period. The second measure is the Time Integrated TTC which uses the integral of the TTC profile of drivers to express the level of safety over the specified time period. Together these values can be used to derive average values per vehicle and the probability of safety critical situations per time unit. This new approach provides valuable comparative safety data and also provides a very useful measure to calibrating simulation models (Archer, 2001). 21 2.4 Micro-simulation & Traffic Safety All of the above TCT measures require field observer crews to collect the data. This data collection process could be carried out by sending field observers to record the number of conflicts at the study intersection. An alternative data collection method is to install video cameras at a particular intersection using certain setup guidelines such as camera angle, elevation, etc. The video footage is captured for a specific period of time. These recordings are reviewed by a crew of professional field observers to determine the potential number of conflicts and their severity. The process is not only expensive, but also involves appraisals by unreliable subjective observers. The data collection process requires expensive instrumented vehicles, high-resolution multi-view video footage and extensive human analysis. Figure (2) shows the potential number of conflicts to be analyzed for a typical 4-leg intersection with a full median opening. Figure 2. Potential number of Conflict Points (32) in an Intersection with Full Median Opening. As a result, a new less expensive approach for assessing the safety of a particular entity has been developed. The use of micro-simulation as a tool to assess the traffic safety of an entity has been proposed by a number of researchers (Darzentas 1980; Hernandez 1982; Ghaffari 1990; Sayed et al. 1994; Persaud and Mucsi 1995; Huang and Pant 1996; Roa and Regaraju 1998; Mehmood et al. 2001). These simulation programs have used the TCT as a safety measure. 22 The use of micro-simulation would provide valuable insights into the relative safety impacts brought about by changes of traffic flow, various ITS devices placed inside the vehicle or on the roadway, different signaling strategies, and many other dependent aspects (Archer 2001). Nevertheless, despite the twenty-five year span of research into the topic of micro-simulation and safety the concept is still not fully developed. Most of the research on this topic has been using special-purpose simulations. Also the level of detail and variety of modeling variables has been study specific. Some researchers (Cooper and Ferguson 1976; Rioux and Lee 1977; Fazio and Rouphail 1990; Fazio et al. 1993) have attempted to modify multipurpose traffic simulation Models to include the TCT or other surrogate measures. However, adjustments to multipurpose traffic simulation models have their restrictions and limitations. Most of these simulation models do not account for the diverse and less predictable driver behavior in real road traffic. Additionally, there is little or no lateral vehicle movement. Currently underdevelopment is the SINDI project (the name is derived from Safety INDIcators). The primary objective of the project is establishing a representative model of road-user behavior that allows for a sufficient degree of variance in road user perception, decision making and action. The second objective is to identify suitable and relevant safety indicators that can be used as a foundation for safety assessment (Archer 2000). As part of the SINDI project, Kosonen (1999) has preformed some preliminary work using the HUTSIM micro-simulation system that allows a high level of detail in the geometric design of road networks. The work has been designed primarily for use with the testing of traffic control systems and strategies. 2.5 Surrogate Safety Assessment Methodology (SSAM) More recently, Gettman and Head (2003) conducted a study to investigate the potential for deriving surrogate measures of safety from existing microscopic traffic 23 simulation models for intersections. The study provided the guidelines for developing a process of computing the measures in the simulation, extracting the required data, and summarizing the results. The process was denoted as the Surrogate Safety Assessment Methodology (SSAM). The study examined nine multi-purpose micro-simulations (MPMS) for potential use as tools for assessing traffic safety. The nine MPMS that were reviewed in the study were: CORSIM, SIMTRAFFIC, VISSIM, HUTSIM, PARAMICS, TEXAS, AIMSUN, WATSIM and INTEGERATION. For the MPMS to be used in assessing safety using surrogate measures, the review concluded that all models required some internal modifications to the source code or programming interface. At the end, only four MPMS were made compatible with the SSAM approach, namely, VISSIM, PARAMICS, TEXAS and AIMSUN. Gettman and Head (2003) used conflict events as the primary indicator of the relative safety of different intersection designs. The authors defined conflict events as "an event occurring between two vehicles that are on an accident course, but do not collide due to evasive action (by either one or both of the vehicles)." The conflict event had two classifications as either a single point or line. A conflict point occurs in a particular single location in time and space. A conflict line occurs for a range of times and spaces. Conflict points,occur as results of crossing flows. These conflict points model the potential for angle accidents due to the acceptance of a gap that is too small by the opposing traffic. On the other hand, conflict lines occur due to merging crossing flows. These conflict lines model the potential for rear-end accidents (or angle accidents from the rear) due to the acceptance of a gap that is too small by the opposing traffic. Two special cases exist for conflict lines, namely, rear-end and lane-changing rear-end conflict lines. Rear-end conflict lines are events where the leader vehicle makes a right (or left) turn causing the following vehicle to decelerate to avoid the accident. Lane-changing rear-end conflict lines are events where the leader vehicle changes lanes abruptly requiring the vehicle in the 24 adjacent lane to brake to avoid accident (The corresponding alternative lane changes from outer lanes to inner lanes are also possible). A number of other potential accidents that were not included in the modeling of the SSAM approach are: sideswipe, head-on, u-turn, swerve-out-of-lane and unrepresented (aside from conflict points and lines) evasive maneuvers. The surrogate measures proposed by Gettman and Head to evaluate the severity of a conflict event were Minimum Time-to-Collision (MTTC) and Minimum Post-Encroachment Time (MPET). Three more surrogate measures, namely, Maximum speed of the two vehicles involved in the conflict event (MaxS), Maximum relative speed of the two vehicles involved in the conflict event (DeltaS) and Deceleration Rate (DR), were also proposed to assess the severity of the potential accident that would result if, in fact, the vehicles collided. The authors pointed out that it is important to distinguish between the severity of the conflict and the severity of the resulting accident. A location with many conflict events of severity exceeding the thresholds for TTC, PET and DR but exhibits a low severity on the DeltaS and MaxS scales may not be of high interest in terms of safety implication. That is, the resulting accidents would be more likely to be property-damage-only when MaxS and DeltaS are low. Locations that may experience fewer total conflict events, but those conflict events that do occur exhibit a very high potential severity (i.e., result in injury and fatality accidents), are probably of more interest to analysts and engineers deciding how to prioritize safety upgrades amongst a number of candidate locations. Gettman and Head (2003) concluded by offering suggestions for validation activities to support the potential analysis of surrogate measures and to compare surrogate measures from simulation models with field data from previous safety studies. 25 3.0 Simulation Tools This section describes the various simulation tools utilized in the analysis. The SSAM software was used to analysis the multi-purpose micro-simulation output file to provide a summary of the number of conflicts experienced in the simulated intersection. The VISSIM software was used to model the 83 Canadian intersections. 3.1 SSAM Software Surrogate Safety Assessment Methodology (SSAM) is a software system designed to perform statistical analysis of data output from traffic multi-purpose micro-simulation (MPMS). The software is developed by Siemens ITS and is funded by the Federal Highway Administration (FHWA). The software's logic and algorithm is based on the study conducted by Gettman and Head in 2003. This section explains the configuration and usage of the SSAM software. The design and development of the SSAM software is based on the Unified Software Development Process (USDP). The functional requirement of the software is based on the "use case" scenario. The Unified Modeling Language defines the "use case" as an interaction of an external actor with the system. In the present case, the system is the SSAM software and the actor being a VISSIM's output file. In order to obtain the required statistics from SSAM, the user has to model and run an intersection on VISSIM. An event file (TRJ format) is produced from VISSIM and is inputted into SSAM for further analysis. The SSAM software uses a set of surrogate measures as the primary estimates of traffic safety. The MTTC and MPET measures were used to evaluate the severity of a conflict event. The surrogate measures MaxS, DeltaS and DR are used to assess the severity of the potential accident that would result if, in fact, the vehicles collided. 26 The software contains a large number of algorithms designed to compute four types of conflicts: crossing, rear-end, lane-changing and unclassified conflicts. The SSAM software reads the VISSIM output file (TRJ format) to identify the total number of conflicts and their corresponding types based on geometry, volumes, signal and driver behavior parameters that were previously defined in VISSIM. 3.2 VISSIM: Traffic & Transit Simulation VISSIM is a microscopic, time step and behavior based simulation model developed to analyze the full range of functionally classified roadways operations (from freeways to driveways). Additionally, VISSIM has the ability to simulate multi-modal traffic flows, including cars, trucks, buses, heavy rail, trams, LRT, bicyclists and pedestrians. Its flexible network structure provides its users with the capability to model any type of geometric configuration or unique operational/driver behavior encountered within the transportation system. The current providers of VISSIM in North and South America are PTV America Inc. The software is based on decades of intensive research at various academic institutions. Its open interfaces provide compatibility with external software. Its link-connector topology allows the highest versatility combined with vehicle movements in a detailed 1/10 (sec) resolution. There are a number of advantages that VISSIM possesses over other MPMS. VISSIM has the ability to obtain detailed state variable information on each vehicle on time scales with better than second-by-second accuracy. Furthermore, VISSIM has been interfaced to other external codes before, including hardware signal controllers, thus facilitating its compatibility with SSAM. The priority rules feature of VISSIM appears to allow complex modeling of junction behavior, including friendly merging (situations where following vehicles will slow for merging vehicles to create a gap), as it occurs in the real world. It is not apparent that other MPMS are able to represent such behavior. Another advantage of VISSIM is the representation of on-street parking behavior and double parking. VISSIM has NEMA controller models 27 available (i.e., using the VAP macro language); adaptive algorithms and real controllers can be integrated and evaluated rather easily with the real-time interface. There is complexity issues involved in setting up the multitude of priority rules at each junction. Nevertheless, this flexibility allows for very detailed modeling of location- and vehicle-specific interactions. So, VISSIM appears to support most of the modeling features required for obtaining surrogate measures at a reasonable level of reliability. Because of its competitive advantage, combined with its aptitude to analyze networks of all sizes (ranging from individual intersections to entire metropolitan areas), VISSIM was chosen to perform the simulation for the field validation plan. Accordingly, the validation effort in this thesis is focused on using the VISSIM 4.1-12 simulation program. 3.2.1 Modeling Process The process of modeling a single intersection in VISSIM starts by tracing an aerial photo of the intersection, specifying each approach number, width and length of lanes. Once the geometry of the intersection is defined, traffic flows (vehicle per hour) for each approach are allocated for all directions. The next step is to encode the signal control parameters. The NEMA controller model included with VISSIM was used for all intersections in this study. Detector locations were defined for intersections with full- or semi- actuated control strategies. A key modeling feature of VISSIM, related to the evaluation of surrogate safety measures, is the implementation of priority rules for permissive left and right turns on red and the modeling of reduced speed areas for turning movements. The inclusion of reduced speed areas is important not only for realistic modeling of traffic but it also impacts the measurements of yield points for priority rules. The effects of priority rule modeling on the surrogate measures of safety outputs from the simulation system are discussed in subsequent sections. Finally, speed profiles, 28 vehicle type characteristics and traffic composition parameters are configured for each intersection. Each intersection was modeled in VISSIM and tested for realistic and reasonable vehicle behaviors. After this nominal verification, the intersection was simulated five times for a period of one hour with different random seed values. Once the VISSIM runs were completed, the TRJ output files from VISSIM were imported to the SSAM application for surrogate measures analysis of the traffic conflicts. 3.2.2 V I S S I M Assumptions Some assumptions had to be made pertaining to intersection geometry, signal control, speed profiles, vehicle type characteristics, traffic compositions, and priority rules. These assumptions are described in the following subsections. Some of these assumptions were made in conjunction with PTV America Inc. Their technical support team provided valuable insight and guidelines when modeling the intersections on VISSIM. 3.2.2.1 INTERSECTION G E O M E T R Y Only 19 intersections had aerial photos on file. The other 64 intersections were based on schematic photos showing some dimensions and certain provisions were made to trace these intersections into VISSIM. For all 83 intersections no information was provided on the number or size of the departing traffic lanes. In some cases, this information can be estimated from the aerial photos, in other cases, assumptions had to be made. 3.2.2.2 S IGNAL C O N T R O L & D E T E C T O R S The simulated intersections included both pre-timed and actuated signal control. Several assumptions were made when modeling the actuated signalized intersections. In cases where a signal has actuated protected/permitted left turning phases, the opposing left turning phases were linked together in order to start and end at the same time. No information on detector locations was available from the safety studies files. Therefore, the detectors were assumed to have a length of 3 (m) 29 and were placed 5 (m) before the stop line. Both assumptions are based on the standard "practice" employed in the province of British Columbia. 3.2 .2 .3 S P E E D PROFILES, VEHICLE T Y P E CHARACTERISTICS & TRAFFIC COMPOSITIONS The desired speed profile was assumed to range from 50 to 65 (km/hr). The traffic was assumed to be entirely composed of passenger cars. Since no information was provided on the percentage of trucks present in each intersection, none were included in the analysis, as the standard practice in British Columbia restricts heavy trucks from using main arterial roads during peak hours. All vehicle types were defined with a number >2,000,000,000 to make use of the latest driving behavior logic in VISSIM. The new driving behavior logic had vehicles always traveling parallel with the center of the link, i.e., not tilted even during a lane change. Additionally, the vehicle doesn't start a lane change unless the gap in front of the trailing vehicle on the new lane is sufficient to perform the maneuver. 3.2 .2 .4 PRIORITY R U L E S A priority rule in VISSIM is the mechanism with which the user can define the yielding and gap-acceptance behavior of vehicles in the simulation. A priority rule is defined through three parameters; minimum gap size, minimum headway and maximum speed. The VISSIM 4.1-12 manual suggests defining priority rules for only permissive left turns and right turns on red, although in certain geometric situations it is necessary to add priority rules to be consistent with real driving behavior. As a rule of thumb, for free-flow traffic on the main road, the minimum gap time of the priority rule is the most relevant condition to calibrate the performance of crossing vehicles. For slow moving or queuing traffic on the main road, the minimum headway becomes the most relevant parameter in calibrating the priority rule. 30 The default parameters as suggested by PTV for minimum gap, min headway and maximum speed are 3 (sec), 5 (m) and 180 (km/hr), respectively. For the 83 intersections used in this study, a minimum gap of 3 (sec) resulted in a large number of simulated accidents. Therefore, the minimum gap size was increased to 5 (sec) with an additional 0.5 (sec) for any additional crossing lane. More discussion on the effect of varying the minimum gap size is presented in Section 7.2. To replicate real life behavior, additional priority rules were added for exclusive left turn and right turn bays. For intersections with no exclusive left turn and right turn bays, priority rules were defined to eliminate run-over accidents between through movement vehicles right turn vehicles and vehicles waiting for a left turn in the middle of the intersection. More discussion on the effect of changing the parameters and the introduction of new or the removal of certain priority rules is presented in Section 7.1. 3.2.2.5 M I S C E L L A N E O U S ASSUMPTIONS No on-street parking information was available in the safety studies data; therefore, this behavior was not modeled. It is a standard practice in the province of British Columbia to prohibit on-street parking during peak hours, thus justifying this assumption. Furthermore, no pedestrian volumes were available; therefore, pedestrians' behavior and interaction were not accounted for in the analysis. Car-following and lane changing behavior was set to model urban (motorized) using the Wiedemann 74 model with all default parameters used. All traffic flows input to VISSIM were based on AM peak volumes as recorded in the safety studies database for each intersection. 3.2.3 Modeling Issues This section describes some of the modeling issues encountered. The issues discussed are: the adopted modeling schemes, the appearance of simulated accidents in VISSIM, the abrupt lane changing behavior and the modeling of the left and right turn taper bays. 31 3.2.3.1 MODELING S C H E M E S This section describes two proposed modeling schemes, developed in conjunction with PTV. The main difference between the two modeling schemes used in the analysis was the number of priority rules that were defined. In addition to the assumptions mentioned in Section 3.2.2, Scheme (1) consisted of 16 priority rules. These were composed of 8 priority rules representing the governing rules for permissive left and right turn on red maneuvers. Eight additional priority rules were added for the exclusive left turn and right turn bays. Throughout the analysis, Scheme (1) was adopted and certain refinements were proposed. First, reduced speed areas for right (15-20 km/hr) and left (30-35 km/hr) were added. This is important not only for realistic modeling of traffic but it also impacts the measurements of yield points for priority rules. Second, when defining the priority rules, the headway values were defined to minimize the conflict area, thus preventing vehicles from yielding to other vehicles that are yielding to a different priority rule. Third, the minimum gap for run-over priority rules were changed from 1 (sec) to 0 (sec) and the maximum speed was reduced to 10 (km/hr) to avoid a deadlock resolution state. Fourth, all two or more lane connectors located in the middle of the intersections were closed to prevent lane changing for all vehicle types as suggested by PTV. Fifth, whenever a problem of queue balance existed, the lane change distance for the connectors were altered from 200 (m) to 800 (m), allowing vehicles more space to perform the lane changing maneuver. Finally, the number of observed vehicles in the driver behavior tab was increased from two to four, allowing vehicles to predict each other movements and react accordingly. It should be noted that any entity, including the state of the signal, is considered a "vehicle" in this feature of VISSIM. Scheme (2) had 32 priority rules defined. In addition to the 16 priority rules defined for Scheme (1), 16 additional rules were added to compensate for the lack of yellow extension and red clearance. Several situations occurred where the signal state 32 turned from green to red with a left turn vehicle waiting to perform its turning maneuver in the middle of the intersection. Due to the lack of yellow extension and red clearance times, these vehicles were run-over by opposing traffic. In real-life behavior, opposing through or other left turning vehicles will wait for traffic to clear before proceeding into the intersection. The additional 16 priority rules in Scheme (2) were designed to allow for such a behavior. Scheme (2) is used to demonstrate the effect of redefining the priority rules on SSAM output. A comparison between the SSAM results for the two modeling schemes is presented in Section (7.1). 3.2.3.2 S I M U L A T E D A C C I D E N T S Despite all the modeling techniques that were employed, simulated accidents remained in the model. These accidents result from insufficient minimum gap size (see Section 7.2), a vehicles' failure to yield to a priority rule, or as a result of an abrupt lane change of a vehicle in an intersection or during queuing. All necessary precautions were taken to minimize the number of simulated accidents while maintaining a level of consistency in modeling all intersections. Therefore, certain parameters were adjusted to reduce the number of simulated accidents experienced in VISSIM. As a result, the number of simulated accidents recorded at each intersection decreased considerably. However, simulated accidents continued to occur, specifically for intersections with high volumes. These intersections experienced a large number of simulated accidents due to the abrupt lane changing behavior, described in detail in the following subsection. Thereby, for each validation test, two sets of results including and excluding simulated accidents are displayed to demonstrate their impacts on the results. 3.2 .3 .3 L A N E C H A N G I N G B E H A V I O R A large number of conflicts/simulated accidents was observed during this study as the number of cars queued up waiting to perform a right or left maneuver increased. These conflicts/simulated accidents were recorded by SSAM as either rear-end or lane-changing. This increase in simulated accidents is due to queued cars changing lanes abruptly. It should be noted that this abrupt lane changing behavior continued to occur in situations where there was no heavy traffic or for through vehicles 33 queuing up at a red signal. This behavior is demonstrated with an example in Appendix (B). At this time, there is no clear justification for the unusual lane-changing behavior in VISSIM. Several measures were taken to reduce the effect of such unusual behavior. First, the lengths of each approach to the intersection were extended to provide vehicles with significantly more time to decide on their downstream path. Second, the driver behavior model was adjusted to allow an additional 0.5 (m) for the minimum lateral clearance. This parameter is defined as the minimum distance for vehicles passing each other within the same lane. This change, suggested by PTV, is supposed to reduce simulated accidents resulting from lane changes. However, this change can have an impact on increasing the capacity of the intersection. The impact of this parameter on simulated conflicts/accidents is presented in Section 7.3 Third, the lane changing algorithm in VISSIM allows for two types of lane-changes: necessary and free lane-changing. In case of a necessary lane change, the driving behavior parameters include the maximum acceptable deceleration for the vehicle and the trailing vehicle on the new lane. This maximum acceptable deceleration depends on the distance of emergency stop position of the next connector route. In case of a free lane change, VISSIM checks for the desired safety distance of the trailing vehicle on the new lane. This safety distance depends on the vehicle's speed. As mentioned earlier, several situations occurred where free lane changing resulted in simulated accidents due to an abrupt lane changing behavior. Unlike the necessary lane change behavior, there is currently no way for the user to manage or change the behavior of free lane changes. 34 3.2.3.4 MODELING L E F T / R I G H T T U R N T A P E R S Generally, left and right turn bay tapers can be modeled in different ways in VISSIM. Therefore, two configurations were proposed to simulate vehicle movements for exclusive left turn and right turn bays. The first configuration proposes the use of one through link and one link for each of the left and right turn storage bays. The through link is then connected to the left and right turn storage bays by a connector emulating the shared roadway and providing a smooth transition from the through movement to the left or right. During congestion, because the connector may overlap the through lane, a priority rule should be placed on the connector that prohibits vehicles from moving forward due to inadequate space to enter the taper (in VISSIM, vehicles traveling on separate links and connectors aren't recognized by other vehicles even though they are traveling in the same direction). Likewise, a priority rule should be placed so that vehicles queued on the connector are recognized by the through vehicles on the link. This modeling configuration is demonstrated in Figure (3); blue lines indicate links while pink lines indicate connectors. The second modeling configuration proposes the use of two links and a connector in between. The first link will have as many lanes as the through movement requires. The second link will group the left, through and right lanes with the connector joining the through movements of both links. This configuration will allow vehicles to respond to the internal lane changing logic to yield to conflicting vehicles for the through, left and right turn movements. Figure (4) demonstrates this modeling configuration. However, the second configuration has led to a number of problems. Figure (5) shows how the second configuration resulted in an increasingly high number of conflicts, with vehicles not queuing up normally. Therefore, the first configuration was used to model all left and right turn tapers. 35 Figure 3. First Taper Modeling Setup M Figure 4. Second Taper Modeling Setup 36 3 7 4.0 Validation Data This section provides information on the validation data provided for the analysis. The data assembly, guidelines and the choice of locations are explained. Furthermore, the selection bias issue is thoroughly discussed. 4.1 Data Assembly Since the goal of the field validation is to compare the performance of SSAM with actual accident experience at Canadian signalized intersections, guidelines for the selection of these intersections need to be determined. There are at least three guidelines to be followed in the identification of field data: o A sufficient number of locations must be used to obtain the range of parameters necessary to test the methodology. o A sufficient number of locations are needed to establish some statistical measure of significance. o The locations should not be selected based on their accident performance otherwise the analysis will be subject to the regression-to-the-mean (RTM) bias (Hauer, 1980). This is, of course, difficult to achieve since safety studies are not typically conducted at locations that do not have any safety performance issues. This bias referred to in the final selection guideline is the tendency of a randomly high accident frequency occurring at a location during a specific time period to be followed by a smaller accident frequency during a consecutive period of equal duration, even if no change took place in the field configuration that generates the accidents at that location. Based on these guidelines, eighty-three 4-leg signalized intersections were selected from a much larger database of Canadian intersections. The information was collected from safety studies performed by Hamilton Associates, Vancouver, BC. The following data were available for each intersection: 38 o Intersection layout (e.g., number and width of approach lanes and intersection design). o Traffic volumes for peak and off-peak periods (including turning movements), o Signal timing plans. o Data on accident frequency, accident type and accident severity. Three years of accident data for each location was used. Accident records were assembled from the auto insurance claims data files collected by the Insurance Corporation of British Columbia (ICBC). The auto insurance claims data maintained by ICBC is very current, comprehensive and considered quite reliable for intersections in British Columbia (De Leur and Sayed, 2001). Appendix (A) provides a summary of the 83 intersections used in this study. The appendix shows the intersection's district, number of lanes in each approach and signal type. It is evident from Table (A1) that, although all 83 intersections were 4-leg signalized, they still represent a wide range of traffic characteristics. 4.2 Selection Bias The majority of the intersections on file were selected to be studied due to a perceived high accident risk that needed to be corrected. Therefore, the majority of intersections on file could be considered "high accident locations" and thus may bias the analysis results. However, the following points should be noted: o Many intersections were selected for study due to a high frequency of accidents. After correcting for exposure, many times the intersections studied did not, in fact, represent a location with disproportionately high risk of accidents. o Several corridor studies are included in the Hamilton Associates' and ICBC files. These corridors were selected for study because some intersections represented a high accident risk. Other intersections along the corridor, which were not considered high risk, were also studied as part of the corridor 39 review, and information about these non-high-risk intersections is also included. o Some intersections were selected for study because they were locally perceived to be "high risk". Compared to a larger population of intersections from other jurisdictions, these intersections would not be considered high risk. Overall, while several intersections on-file can be considered high accident locations, many others would not qualify as high risk locations. An appropriate balance between high-risk and low-risk locations was considered while selecting intersections for executing the validation study. 40 5.0 Methodology for Field Validation The methodology for the field validation is based on relating the surrogate measures produced by SSAM to the actual accident occurrence observed in the field. The main part of the validation effort focuses on using the VISSIM simulation program. The validation study with SSAM using VISSIM involves the following five tests: o Validation Test 1 o Validation Test 2 o Validation Test 3 o Validation Test 4 Safety Ranking Analysis Safety Ranking Analysis for Specific Conflict Types Conflict / Accident Paired Comparison Conflict and Accident Prediction Model Comparative Analysis o Validation Test 5: Conflict and Accident Prediction Model Comparative Analysis for Specific Conflict Types The following sections describe the methodology for each of these tests. 5.1 Validation Test 1: Safety Ranking Analysis In this test, the ranking of intersections from SSAM according to predicted total conflicts was compared to the ranking of the same intersections using actual accident frequency. The following steps were performed: Step A: Conflict Prediction and Ranking In this step, SSAM was used to predict the total number of conflicts at each intersection. A conflict risk score was then allocated to each intersection according to the average conflict rate (ACR) calculated as: ACR = AHCITEV where AHC is the average hourly conflict and PEV is the square root of the product of the hourly entering volumes from major and minor approaches. The locations were ranked based on descending order of their ACR value. 41 Step B: Accident Ranking For this purpose, a meaningful road safety ranking indicator needs to be used. It is recommended to use the "potential for improvement" (PFI) indicator (Sayed and Rodriguez, 1999), measured as the difference between the existing accident frequency and the expected accident frequency at a location. The expected accident frequency is determined by applying a valid accident prediction model and refined by applying the Empirical Bayes technique (Hauer, 1997, Sayed et al. 2004). The difference between the observed and expected accident frequencies facilitates the ranking of sites (a greater difference having a higher rank). The starting point for this evaluation is the development of an accident prediction model to determine the expected accident frequency at each site. The generalized linear modeling approach GLM (Hauer, 1997, Sawalha and Sayed, 2001) was used to calculate the expected accident frequency at each intersection. The expected accident frequency was combined with actual accident counts using the Empirical Bayes (EB) refinement technique as follows: EB safety estimate E(A) x {K + count) (2) where: E(?) = predicted accident frequency (accidents/3years) count = observed accident frequency (accidents/3years) ? = model dispersion parameter The PFI was then calculated as the difference between the Empirical Bayes safety estimate obtained from Equation (2) and the predicted accident frequency. The rank was established based on descending order of the PFI. 42 Step C: Ranking Comparison The ranking derived from SSAM's conflict prediction was compared to the ranking from the actual accident risk. The Spearman rank-correlation coefficient was used to determine the level of agreement between the two ranks. The Spearman rank-correlation coefficient is often used as a non-parametric alternative to a traditional coefficient of correlation and can be applied under general conditions. The Spearman rank-correlation coefficient (?s) is calculated as shown in Equation (3). A score of 1.0 represents perfect correlation and a score of zero indicates no correlation. An advantage of using (?s) is that when testing for correlation between two sets of data, it is not necessary to make assumptions about the nature of the populations sampled. 6 Y d 2 n(n -1) ( 3 ) where: d = differences between ranks n = number of paired sets Under the null hypothesis of no correlation, the ordered data pairs are randomly matched and thus the sampling distribution of (?s) has a mean of zero and a standard deviation (s s) as given in Equation (4). (4) Since this sampling distribution can be approximated with a normal distribution even for relatively small values of n, it is possible to test the null hypothesis on the statistic given in Equation (5). This value can be compared to a critical z-value taken to represent a 90% level of significance. A V C W - I ) (5) 43 5.2 Validation Test 2: Safety Ranking Analysis for Specific Incident Types Test 2 repeats the same comparative ranking procedures as test 1, but for sub-sets of accident/conflict types. Indeed, the analysis was repeated for the following types: o Crossing o Rear-End o Lane-changing 5.3 Validation Test 3: Conflict / Accident Paired Comparison Test 3 compares the conflict frequency predicted by SSAM to the actual accident frequency at each intersection. A regression equation that relates actual accidents to the predicted conflicts was developed. The goodness-of-fit was tested using the R 2 value, which determines the strength of the relationship between predicted conflicts and actual accidents. This test could be conducted using either frequency or rate, although exposure can be excluded due to the paired nature of the test. Since both actual accidents and predicted conflicts are discrete random variables with long right-tail distributions, such an analysis is justified by (i) using natural logarithms to transform both variables, and (ii) conducting a conditional analysis of actual accidents given predicted conflicts. Thus, a regression equation that relates the logarithms of actual accidents (Y) to the logarithms of predicted conflicts (X) was developed. The goodness-of-fit was tested using R 2. Another regression technique was also employed to relate the actual accident frequency to the conflict frequency predicted by SSAM. It is generally accepted that both accidents and conflicts are discrete random events with a non normal error structure. Therefore, the technique assumes that accidents follow a negative binomial distribution while the simulated conflicts follow a Poisson distribution. 44 Moreover , it was a s s u m e d that s imulated conflicts follow the same distribution as real confl icts. Hence , for the i t h intersection, the number of accidents 7, is a s s u m e d to fol low a negative binomial distribution with shape parameter K and sca le parameter / / , . /K, where jj.i = E(Yi), and the number of conflicts x-, is a s s u m e d to follow a Po i sson distribution with parameter Further, it is a s s u m e d that the regression model relating conflicts ( x ) to acc idents ( 7 , ) is given by ln(Jui) = j3 + yln(6>i). (6) Maximiz ing the log l ikelihood with respect to the parameters K , , y and #, leads to two sets of condit ions. The first relates to the accident model and cor responds to the usual negative binomial model condit ional on the conflicts' est imates § j . The second relates to the conflicts model and can be written as e^Xi+rfY.-M,^-). (7) Thus , the maximum likelihood est imates can be determined using the following iterative procedure: 1) Initial est imates of are calculated as = x • 2) The latest est imates of #, are used to fit the accident (negative binomial) model and obtain est imates of K, p and y. Hence //, is est imated from Equat ion (6). 3) The latest est imates of y, ju, and K are used, together with xt and 7 , , to obtain new est imates of from Equat ion (7). Return to step 2. The procedure cont inues, carrying out alternate appl icat ions of the accident negative binomial model and the conflicts Po i sson model , until convergence is reached, normally after just a few iterations. 45 5.4 Validation Test 4: Conflict and Accident Prediction Model Comparative Analysis This test is composed of four steps. The details of these steps are listed below. Step A: A conflict prediction model was developed relating the conflicts calculated by SSAM and the traffic volume characteristics of the intersections (standard GLM procedures were used). Step B: An accident prediction model was developed relating actual accidents and the traffic volume characteristics of the intersections (standard GLM procedures were used). Step C: The two prediction models were compared to determine whether the conflict prediction model can predict risk in a manner similar to the accident prediction model for intersections with the same characteristics. The comparison included several model applications such as the identification and ranking of accident prone locations. An accident prone location is defined as any location that exhibits a significantly higher number of accidents/conflicts as compared to a specific normal value. The EB refinement method identifies problem sites according to the following four-step process. 1) Estimate the predicted number of accidents/conflicts and its variance for the intersection,.using the accident/conflict prediction model. This prediction is assumed to follow a gamma distribution (the prior distribution) with parameters a and p, where: B = £ ( A ) =-g— and a =B -E(A) = K Var(A) E(A) H ( 8 ) and Var(?) is the variance of the predicted accidents/conflicts. 46 2) Determine the appropriate point of comparison based on the mean and variance values obtained in step (1). Usually the 50th percentile (P 5 0) or the mean is used as a point of comparison. P 5 0 is calculated such that: I —— UA = 0.5 0 r(K) 3) Calculate the EB safety estimate and the variance from (9) EB safety estimate K K + E(A) E(A) + / E(A) ^ K + E(A) (count) (10) Var(EB sajety estimate) £(A) K + E(A) (K) + E(A) /c + E(A) (count) (11) This is also a gamma distribution (the posterior distribution) with parameters a i and P i defined as follows: EB K B, = = h 1 and a, = B. • EB = K + count Var(EB) E(A) HX ( 1 2 ) Then, the probability density function of the posterior distribution is given by fEB W = ~ 1 (K + count) ^2) 4) Identify the location as accident/conflict-prone if there is significant probability that the location's safety estimate exceeds the P5o value (or the mean). Thus, the location is prone if: 47 ^50 {KI E(A) + \ j ( K + c o u n t ) fl<+count-\E-(KI' E(A)+l)A r(K + count) dX >S (14) where: 5 is the desired confidence level. Once accident/conflict-prone sites are identified, it is important to rank the locations in terms of priority for treatment. Ranking problem sites enables the road authority to establish an effective road safety program, ensuring the efficient use of the limited funding available for road safety. Two techniques that reflect different priority objectives for a road authority can be used (Sayed and Rodriguez, 1999). The first ranking criterion is to calculate the ratio between the EB estimate and the predicted frequency as obtained from the GLM model (a risk-minimization objective). The ratio represents the level of deviation that the intersection is away from a "normal" safety performance value, with the higher ratio representing a more hazardous location. The second criterion, the cost-effectiveness objective, is to use the PFI as described in test 1. Step D: Two comparisons were undertaken. The first compared the locations identified as prone by the accident and conflict prediction models. The second compared the ranks obtained from the accident and conflict prediction models. 5.5 Validation Test 5: Conflict and Accident Prediction Model Comparative Analysis for Specific Incident Types Test 5 repeats the same comparative analysis as test 4, but for sub-sets of accident/conflict types. Indeed, the analysis was repeated for the following types: o Crossing o Rear-End o Lane-change 48 5.6 Goodness of Fit Measures Three statistical measures were used to assess the goodness of fit of the various accident/conflict prediction models to the 83 Canadian intersections. The first measure is the Pearson chi-squared computed by: 2 ^ Lv,-£(A,)? Pearsonx = X Var(Yt) (15) The second measure is the scaled deviance. This is the likelihood ratio test statistic measuring twice the difference between the log-likelihood of the data under the developed model and its log-likelihood under the full or saturated Model. The scaled deviance is computed by: SD = 2£ v,ln ' y, ^ *(A,), - 0 \ + * , - ) l n i J (16) The third measure is the R-squared defined by Miaou (1996). The R-squared goodness of fit measure is computed as: a (17) where a = the model dispersion parameter; a m a x = the maximum dispersion parameter estimated in the model with only the constant term and no predictor variables. 49 6.0 Results and Discussion After modeling the 83 intersections on VISSIM using Scheme (1), each intersection was run five times for one hour of simulated time with different random seeds. The output files (.TRJ format) from VISSIM were then imported into the SSAM software for identification of conflicts and computation of the surrogate measures of safety for each conflict event. Prior to running SSAM the default time-to-collision (TTC) and post encroachment time (PET) values were defined. These default thresholds were selected as 1.5 seconds for TTC and 5.0 seconds for PET. The results of the SSAM analysis consists of the number of total conflicts and the number of conflicts of each type of vehicle-vehicle interaction: crossing, rear-end and lane-changing. As mentioned earlier, the number of simulated accidents (defined by TTC = 0 in SSAM) and their types were also recorded to allow for two types of analysis; including and excluding simulated accidents. Appendix (C) displays the number and types of conflicts recorded for each intersection, including and excluding simulated accidents. The following sections describe the results of the five validation tests that were performed in this analysis. 6.1 Validation Test 1: Safety Ranking Analysis In this test, the ranking of intersections using simulated conflicts was compared to the ranking of the same intersections using accident frequency. The first step of this test included the calculation of a conflict risk score allocated to each intersection according to the average conflict rate (ACR) calculated as shown in Equation (1). The locations were ranked based on descending order of their ACR value. Once the ACR was computed, the second step was to develop an accident prediction model for total accidents. This model was used to determine the expected accident frequency at each site. The generalized linear modeling approach GLM (Hauer, 1997, Sawalha and Sayed, 2001) was used to calculate the expected accident frequency at each intersection. The development of accident prediction models was carried out using the GENMOD procedure in the SAS 9.1 statistical software 50 package. The variables present in the models were: V M i = AADT on the minor approach and V M a = AADT on the major approach. Table (1) shows the estimates of the parameters for the total accidents model, Model (1). The t-ratio was used to assess the significance of these estimates. As shown in the table, the estimates are all significant at the 90% confidence level. Furthermore, the table shows that both the Pearson chi-squared and the scaled deviance values were not significant at the 90% confidence level, indicating a good fit. Moreover, the result of the R-squared goodness of fit test agrees with to those of the Pearson chi-squared and the scaled deviance. Table 1 - Total Accidents Model for Safety Prediction and Related Applications Model (1) Total 13yr = 0.000005473 x ^ x C " DF 79 R2 Scaled Deviance 0.6277 8 5 . 8 2 Pearson % -£0.1,79 79.93 95.48 Shape Parameter K 5.447 Variable Constant AADT Minor AADT Major Coefficient 5.473E-06 0.7442 0.8960 t-ratio 7.700 8.623 5.918 The expected accident frequency was combined with actual accident counts using the Empirical Bayes (EB) refinement technique as shown in Equation (2). The potential for improvement (PFI) was then calculated as the difference between the Empirical Bayes safety estimate and the predicted accident frequency. The rank was established based on descending order of the PFI. The third step of this test compared the ranking derived from SSAM's conflict prediction, based on ACR, and the ranking from the actual accident prediction model. The Spearman rank-correlation coefficient was used to determine the level of agreement between the two ranks. The Spearman rank-correlation coefficient (?s) is calculated as shown in Equation (3). 51 As mentioned earlier, the results of the SSAM software are presented into two parts: including and excluding simulated accidents. Table (2) shows the correlation between the ranking derived from SSAM's conflict prediction, based on ACR, and the ranking from the accident prediction model Table 2 - Spearman Rank Correlations based on Total/Severe ACR & Total Accidents Model Models Simulated Accidents Correlation Total ACR / Total Accidents Included Excluded 0.045 0.132 Severe ACR / Total Accidents Included Excluded 0.134 0.008 The correlation coefficients provided in Table (2) are very weak, indicating poor and insignificant agreement between the rankings of accidents from the regression model and the average conflict rates. 6.2 Validation Test 2: Safety Ranking Analysis for Specific Incident Types Validation test (2) repeats the same comparative ranking procedures as for validation test (1), but for the sub-sets of accident/conflict types: crossing, rear-end and lane-changing. The first step included calculating the average conflict rate for each intersection and ranking them in descending order. The second step involved developing accident prediction regression models for the three accident types mentioned earlier. Tables (3)-(5) show the estimates of the parameters for each of the three regression equations, Models (2)-(4). The t-ratio was used to assess the significance of the estimates. It was found that the regression coefficients were all statistically significant at the 90% confidence level. Also, both the Pearson chi-squared and the scaled deviance indicate a good fit. The R-squared goodness of fit test was high for both rear-end and lane-changing models but was rather low (0.2853) for the crossing model. 52 Table 3 - Crossing Model for Safety Prediction and Related Applications Model (2) Crossing / 3 yr = 0.0008988 xV^xV™19 DF r 2 Scaled Deviance Pearson % £0 .1 ,79 Shape Parameter AT 79 0.2853 91.12 71.02 95.48 2.648 Variable Coefficient t-ratio Constant 8.988E-04 3.025 AADT Minor 0.5524 4.312 AADT Major 0.4319 1.935 Table 4 - Rear-End Model for Safety Prediction and Related Applications Model (3) Rear End / 3yr = 0.00000009186 x VMfm x VMAam DF ^2 Scaled Deviance Pearson £ £0 .1 ,79 Shape Parameter K 79 0.6106 87.87 85.29 95.48 3.887 Variable Coefficient t-ratio Constant 9.186E-08 8.491 AADT Minor 0.8179 7.880 AADT Major 1.1589 6.336 Table 5 - Lane-Changing Model for Safety Prediction and Related Applications Model (4) Lane Changing/3yr = 0.00000005043 x VMfm x VM°a600 DF ^2 Scaled Deviance 2 2 Pearson % £0 .1 ,79 Shape Parameter K 79 0.6106 94.74 66.59 95.48 1.886 Variable Coefficient t-ratio Constant 5.043E-08 5.912 AADT Minor 0.8172 5.373 AADT Major 1.0600 3.932 The third step compared the ranking of each intersection based on the average conflict rates derived from SSAM to the ranking of each intersection derived from the accident prediction regression models for the three types of accidents: crossing, rear-end and lane-changing. The Spearman rank-correlation coefficient was used to determine the level of agreement between the two ranks. Table (6) shows the correlation between the two ranks. 53 Table 6 - Spearman Rank Correlations based on ACR & Accident Models for Crossing, Rear-End and Lane-Changing. Type Simulated Accidents Correlation Crossing Included Excluded 0.232 0.048 Rear-End Included Excluded 0.020 0.023 Lane-Changing Included Excluded -0.012 0.058 Similar to the results of validation test (1), the correlation coefficients provided in Table (6) are weak, indicating poor and insignificant agreement between the rankings of various accident types and the corresponding rankings of simulated conflict types. 6.3 Validation Test 3: Conflict / Accident Paired Comparison Validation test (3) compared the conflict frequency predicted by SSAM to the actual accident frequency at each intersection to determine the strength of the relationship between accidents and simulated conflicts. Thus, following the methodology outlined in Subsection 5.3, a regression equation was developed relating the logarithms of actual accidents (y) to the logarithms of predicted conflicts (x). The results appear in Table (7). The goodness-of-fit of several regression equations were tested using R 2. Table 7 - Regression Models for Actual Accidents given Predicted Conflicts. Simulated Accidents Equation R2 Included Excluded Ln(Total Accidents) LnfTotal Accidents) = 0.737*Ln(Total Conflicts) + 1.399 = 0.538*Ln(Total Conflicts) + 1.013 0.248 0.267 Included Excluded LnfTotal Accidents) = Ln(Total Accidents) = 0.363*Ln(Severe Conflicts) + 3.492 0.171*Ln(Severe Conflicts) + 4.478 0.210 0.155 Table (7) reveals that total number of conflicts explains about 25% of the variations in the total number of accidents. In contrast, severe conflicts explain 21% of the 54 variations, when simulated accidents were included, while explaining only 15.5%, when simulated accidents were excluded. Another regression technique was then employed to relate the actual accident frequency to the conflict frequency predicted by SSAM. Tables (8) and (9) show the estimates of the parameters of the nonlinear regression Equation (6), including and excluding simulated accidents, Models (5)-(6). These estimates were obtained using a special SAS macro that was developed to iteratively update the likelihood equations as described in Subsection 5.3. Table 8 - Total Accidents as a Function of Total Conflicts (Simulated Accidents Included) Model (5) Crashes = 1.906 x Conflicts™12 DF 80 R 2 0.179 Scaled Deviance 88.32 Pearson % 102.76 •#0.1,79 96.58 Shape Parameter K 2.471 Variable Constant Conflicts Coefficient 1.906 0.672 t-ratio 2.931 4.587 Table 9 - Total Accidents as a Function of Total Conflicts (Simulated Accidents Excluded) Model (6) Crashes ---0.0903 x Conflicts1 .1686 DF 80 R 2 0.303 Scaled Deviance 87.48 Pearson % 93.98 •#0.1,79 96.58 Shape Parameter K 2.908 Variable Constant Conflicts Coefficient 0.0903 1.1686 t-ratio 0.126 6.674 As shown in Tables (8) and (9), the coefficients of conflicts were significant at the 90% confidence level, for both Models (5) and (6). In addition, the scaled deviance values for both models were not significant at the 90% confidence level, indicating good fits. However, the Pearson chi-squared was marginally significant for Model (5) with the simulated accidents from VISSIM included in the analysis. 55 To assess SSAM's capabilities to predict safety performance beyond what can be obtained through a simple relationship based on exposure, Models (5) and (6) were compared with Model (1). The regression equations were compared using the R-squared goodness of fit statistic, Equation (17). The results of this comparison show that traffic volumes can explain more variation in the occurrence of accidents than simulated conflicts, since the R-squared value was recorded to be 0.6277 for the accident prediction model based on volumes, Model (1), as opposed to 0.179 and 0.303 for the regression models based on simulated conflicts, Models (5) and (6), respectively. 6.4 Validation Test 4: Conflict and Accident Prediction Model Comparative Analysis 6.4.1 Development of Conflict Models The first step of this validation test included developing a number of conflict prediction models relating the conflicts calculated by SSAM to the traffic volume characteristics of the intersections using standard GLM procedures. Previously, the regression Model (1), shown in Table (1), was developed for total actual accidents. Four more regression models with the same parameter structure were developed to predict severe and total conflicts with simulated accidents included and excluded. A severe conflict was defined as a conflict with a time-to-collision (TTC) of 1.0 sec or smaller. The variables used in each model were: V M i = vehicle per hour (VPH) on the minor approach and V M a = vehicle per hour (VPH) on the major approach. Tables (10)-(13) show the estimates of the parameters for total/severe conflict models with/without simulated accidents, Models (7)-(10). The t-ratio was used to assess the significance of the parameter estimates and they were all found to be significant at the 90% confidence level. The tables also show that the scaled deviance values for all 56 models were not significant at the 90% confidence level, indicating good fits. However, the Pearson chi-squared was marginally significant for the severe conflicts model in Table (12). Table 10 - Total Conflicts Model (Simulated Accidents Included) Model (7) Total Conflicts /1 hr = 0.03408 x V ° f 9 5 x F l 7 5 7 5 DF r 2 Scaled Deviance Pearson £ £0.1,79 Shape Parameter K 79 0.5876 84.37 92.02 95.48 10.040 Variable Coefficient t-ratio Constant 3.408E-02 4.456 VPH Minor 0.3295 5.069 VPH Major 0.7575 7.383 Table 11 - Total Conflicts Model (Simulated Accidents Excluded) Model (8) Total Conflicts /1 hr = 0.2301 x V°?654 x J^ f 8 9 DF r 2 Scaled Deviance Pearson £ £ 0 1 7 9 Shape Parameter K 79 0.6262 85.11 89.15 95.48 21.505 Variable Coefficient t-ratio Constant 2.301 E-01 2.546 VPH Minor 0.2645 5.431 VPH Major 0.5089 6.566 Table 1 2 - Severe Conflicts Model (Simulated Accidents Included) Model (9) Severe Conflicts /1 hr = = 0.00005475 x V ^ x C ' 4 DF r 2 Scaled Deviance Pearson % £ 0 1 3 Shape Parameter K 79 0.4912 87.36 106.92 95.48 2.815 Variable Coefficient t-ratio Constant 5.475E-05 6.688 VPH Minor 0.4934 3.861 VPH Major 1.3314 6.637 57 Table 13 - Severe Conflicts Model (Simulated Accidents Excluded) Model (10) Severe Conflicts /1 hr = 0.00001593 x V ° f 7 7 x F ^ 1 5 9 4 DF 79 ^ 2 Scaled Deviance 0.7041 77.11 2 2 Pearson x -#0.1,79 90.40 95.48 Shape Parameter K 7.536 Variable Constant VPH Minor VPH Major Coefficient 1.593E-05 0.5677 1.1594 t-ratio 7.009 4.470 5.577 6.4.2 Identification of Accident Prone Locations (APL) The second step of this validation test determined whether the conflict prediction equation can predict risk in a manner similar to the accident prediction equation for intersections with the same characteristics, by identifying accident prone locations (APL). An accident prone location is defined as any location that exhibits a significantly higher number of accidents/conflicts as compared to a specific normal value (Sayed and Rodriguez, 1999). There are two types of clues to the safety performance of a location: its traffic and road characteristics, and its historical accident data. The Empirical Bayes (EB) approach makes use of both clues. The EB approach is used to refine the estimate of the expected number of accidents at a location by combining the observed number of accidents at the location with the predicted number of accidents obtained from GLM models to yield more accurate, location-specific safety estimate. The EB refinement method identifies problem sites according to the four-step process described in Subsection 5.4. The results in Table (14) show that fewer locations were identified as accident prone by the regression Models (7)-(10) based on simulated conflicts than those identified by the regression Model (1) based on the actual accident records. Table (14) also shows that only a few APLs were common between the accident/conflict models. This indicates a poor agreement between the conflicts and actual accident models in 58 identifying accident prone locations. The agreement is even less when the simulated accidents were excluded. Table 14 - Number of Intersections Identified as Accident Prone Based on Accident/Conflict M odels # APLs identified based on Total Accidents # APLs identified based on Total Conflicts Simulated Simulated Accidents Accidents Included Excluded # APLs identified based on Severe Conflicts Simulated Simulated Accidents Accidents Included Excluded 31 21 15 8 5 # of sites common with those identified based on total accidents 7 4 3 1 6.4.3 Ranking Locations Once accident/conflict-prone sites were identified, the next step was to rank the locations in terms of priority for treatment. Two techniques that reflect different priority objectives for a road authority were used as described in Subsection 5.4, namely, the ratio between the EB estimate and the predicted frequency obtained from the GLM regression equation (a risk-minimization objective) and the PFI (a cost-effectiveness objective). Table (15) shows the Spearman rank-correlations for both the PFI and ratio rankings. Table 15 - Spearman Rank Correlations based on Total Accidents Model and Total/Severe Conflicts Models Simulated PFI Ratio Accidents Correlation Correlation Total Accidents / Total Included 0.044 0.165 Conflicts Excluded 0.131 0.194 Total Accidents / Severe Included 0.033 0.195 Conflicts Excluded 0.064 0.123 59 For both the PFI and ratio techniques, Table (15) shows that the correlations between the rankings based on the accident and conflict prediction models are weak, indicating poor agreement. 6.5 Validation Test 5: Conflict and Accident Prediction Model Comparative Analysis for Specific Incident Types 6.5.1 Development of Conflict Models for Specific Incident Types Validation test (5) repeats the same process as for validation test (4), but for each conflict type (rear-end, lane-changing, and crossing). In Section 6.2, Models (2)-(4) were developed for predicting the three types of accidents. Tables (16)-(21) show the estimates of the parameters of the crossing, rear-end and lane-changing conflict models including and excluding simulated accidents, Models (11)-(16). The t-ratio was used to assess the significance of these estimates and they were all significant at the 90% confidence level, except for the crossing Model (12). The tables also show that the scaled deviance values for all models were not significant at the 90% confidence level, indicating good fits. However, the Pearson chi-squared was marginally significant for Models (11), (12), (15) and (16). Table 16 - Crossing Conflicts Model (Simulated Accidents Included) Model (11) Crossing Conflicts /1 hr = 0.0001424xV^, 4 3 7 5 x ^ 8 2 9 DF r 2 Scaled Deviance 2 2 Pearson % -#0.1,79 Shape Parameter K 79 0.2857 90.57 100.38 95.48 1.403 Variable Coefficient t-ratio Constant 1.424E-04 4.490 V P H Minor 0.4375 2.325 V P H Major 1.1829 4.343 60 Table 17 - Crossing Conflicts Model (Simulated Accidents Excluded) Model (12) Crossing Conflicts /1 hr = 0.0009562 xV 0.1626 T^O.8747 Mi X " Ma DF 79 r 2 Scaled Deviance 0.3659 81.09 Pearson % 96.59 >f 0.1,79 95.48 Shape Parameter K 4.013 Variable Constant VPH Minor VPH Major Coefficient 9.562E-04 0.1626 0.8747 0.944 t-ratio 3.412 - Not Significant 3.210 Table 18- Rear-End Conflicts Model (Simulated Accidents Included) Model (13) Rear End Conflicts /1 hr -1.020 x V^ 1 8 4 7 x V^243 DF 79 R 2 Scaled Deviance 0.3212 87.20 Pearson £ 83.18 £ o . l , 7 9 95.48 Shape Parameter K 13.369 Variable Constant VPH Minor VPH Major Coefficient 1.020E+00 0.1847 0.3243 0.027 t-ratio - Not Significant 3.058 3.333 Table 19 - Rear-End Conflicts Model (Simulated Accidents Excluded) Model (14) Rear End Conflicts /1 hr = 1.077 x V ° f 48 prO.3165 X * M a DF 79 r 2 Scaled Deviance 0.3162 87.19 Pearson % 83.23 £ o . l , 7 9 95.48 Shape Parameter K 13.405 Variable Constant VPH Minor VPH Major Coefficient 1.077E+00 0.1848 0.3165 0.102 t-ratio - Not Significant 3.065 3.253 61 Table 20 - Lane-Changing Conflicts Model (Simulated Accidents Included) Model (15) Lane Changing Conflicts /1 hr = 0.0003881 x V^, 4 7 0 4 X F , . 0 8 ! 8 Ma DF 2^ Scaled Deviance 2 2 Pearson % -#0.1,79 Shape Parameter K 79 0.5456 83.17 102.97 95.48 4.647 Variable Coefficient t-ratio Constant 3.881 E-04 6.643 VPH Minor 0.4704 4.800 VPH Major 1.0818 6.804 Table 21 - Lane-Changing Conflicts Model (Simulated Accidents Excluded) Model (16) Lane Changing Conflicts /1 hr = 0.001208 x V ° f 0 9 T / " 0.8860 Ma DF p2 Scaled Deviance Pearson % -#0.1,79 Shape Parameter K 79 0.5223 82.25 107.80 95.48 5.316 Variable Coefficient t-ratio Constant 1.208E-03 5.816 VPH Minor 0.4509 4.596 VPH Major 0.8860 5.727 The analysis of crossing conflicts in Table (17) revealed that the estimate associated with the minor approach was not significant, when simulated accidents were excluded. A possible cause for such a result is the lack of such conflicts at numerous simulated intersections (i.e., many locations had no crossing conflicts). 6.5.2 Identification of Accident Prone Locations for Specific Incident Types Once the parameters of the conflict prediction equations were estimated, the models were then used to identify the conflict prone locations for specific conflict types (crossing, rear-end and lane-changing). The results in Table (22) show that fewer locations were identified as conflict prone by the conflict prediction Models (11 )-(16) than those identified as accident prone by the accident prediction Models (2)-(4). There was only one exception, where 19 intersections were identified as APLs by the conflict prediction Model (15) (when 62 simulated accidents were included), as opposed to 18 intersections identified as APLs by the accident prediction Model (4). For each incident type, the agreement between the accident/conflict models on the identification of accident prone locations is rather poor since the number of common locations, identified by both approaches, is low. Table 22 - Number of Intersections Identified as Accident Prone Based on Accident/Conflict Model Types Model Type Simulated Accidents # APLs by Accident Model Type # APLs by Conflict Model Type # Common APLs Crossing Including Excluding 21 18 NA* 4 NA* Rear-End Including Excluding 28 14 13 5 5 Lane-Changing Including Excluding 18 19 13 4 2 * Not available as there were too many zeros in the "crossing" data. 6.5.3 Ranking Locations for Specific Incident Types The PFI and ratio rankings (as explained in Subsection 6.4.3) were then obtained using the crossing/rear-end/lane-changing conflict prediction models and the corresponding accident prediction models. Table (23) shows the Spearman rank-correlations for both the PFI and ratio rankings. 63 Table 23 - Spearman Rank Correlations based on Accident/Conflict Models for Rear-End and Lane-Changing Simulated Accidents PFI Correlation Ratio Correlation Crossing Included Excluded 0.359 NA* 0.276 NA* Rear-End Included Excluded 0.000 0.002 0.062 0.063 Lane-Changing Included Excluded 0.052 0.114 0.091 0.113 * Not available as there were too many zeros in the "crossing" data. The results in Table (23), indicate that the rankings derived from actual accident type models were poorly correlated with the rankings derived from the predicted conflict type models. It is worth noting that the crossing accident/conflict models recorded the highest correlation in terms of PFI and ratio ranking when compared to rear-end and lane changing. This is due to the abrupt lane changing behavior in some of the VISSIM models, which affects the number of simulated lane changing and rear-end conflicts. 6.6 Validation Results Summary For the five tests performed, the results of the validation generally indicate poor correlation between actual accidents and predicted conflicts obtained from SSAM. As well, the results indicate that the traffic volumes alone can explain more variability in the occurrence of accidents than simulated conflicts. 64 7.0 Validation Issues The results of the field validation showed that there is generally poor correlation between actual accidents and the simulated conflicts produced by the VISSIM simulation system and analyzed with the SSAM. However, it should be noted that there are several modeling issues related to VISSIM that have a significant impact on the results. The impact of these issues on the number and types of conflicts produced by the simulation system is discussed in the following sections. 7.1 Effect of Redefining the Priority Rules Several researchers have shown that both real-world accidents and real-world conflicts (as measured by field observers using the traffic conflicts measurement techniques) are strongly related to traffic volumes (Zegeer and Deen, 1978; McDowell et al. 1983; Migletz 1985; Sawalha and Sayed, 2001; Sayed and Rodreguiz, 1999). Therefore, the higher the traffic volumes at an intersection, the more likely that conflicts and accidents will occur. When modeling an intersection in VISSIM, two factors govern the discharge of traffic flow at an intersection, namely; signal control and priority rules. The signal design is based on fixed parameters that were provided directly from the safety studies. These values were used in this validation to truly represent real-life conditions. In "typical" VISSIM traffic modeling for capacity/performance analysis, priority rules are only necessary for permissive left turns and right turns on red. However, additional priority rules are necessary to reduce the simulated accidents that occur due to the driver behavior logic of VISSIM. As mentioned earlier, two modeling schemes were adopted. The main difference between Schemes (1) and (2) was the addition of 16 more priority rules in Scheme (2) to compensate for the lack of yellow extension and red clearance as discussed in Subsection 3.2.3.1. For the 83 intersections modeled in this study, Figures (6) and (7) show plots of the total number of conflicts produced from each scheme, including and excluding simulated accidents, respectively. As shown in Figure (6), the number of conflicts including simulated accidents for Scheme (1) is higher than Scheme (2). 65 This is expected because the addition of 16 priority rules in Scheme (2) has led to a reduction in simulated conflicts. Figure (7) shows the number of simulated conflicts excluding accidents. It is interesting to note that the addition of the 16 priority rules in Scheme (2) resulted in a reduction of accidents but increased the number of conflicts. This increase is due to some 'would be' simulated accidents becoming conflicts with low TTC values. O - K , , , , , 1 O 50 100 150 200 250 300 S c h e m e (1) Figure 6. Effect of Redefining the Priority Rules on Total Conflicts (Including Simulated Accidents) 66 0 20 40 60 80 100 120 140 S c h e m e (1) Figure 7. Effect of Redefining the Priority Rules on Total Conflicts (Excluding Simulated Accidents) The effect of redefining the priority rules on the total and type of conflicts is demonstrated by comparing three modeling schemes. Schemes (1) and (2) were previously described in Subsection 3.2.3.1. The third scheme uses a base case scenario where only 8 priority rules were used for permissive left and right turns on red maneuvers. To demonstrate the effect of redefining the priority rules a typical intersection was selected and simulated using the three different schemes. The selected intersection was composed of three lanes on all approaches with a pre-timed signal controller. The 4-leg intersection had a traffic flow of 1520 (vph) and 1660 (vph) on the minor and major approaches, respectively. Table (24) shows the total and type of conflicts recorded for each scheme at the selected intersection. The results are based on the average value of five simulated runs with different random seeds. Table 24 - Number and Type of Conflicts based on Different Modeling Schemes Total Crossing Rear-End Lane-Change Simulated Accidents Base Case 121 22 73 26 28 Scheme (1) 114 22 70 22 27 Scheme (2) 95 4 73 18 6 67 The base case recorded 22 crossing conflicts, 73 rear-end conflicts and 26 lane-changing conflicts, for a total of 121 conflicts over a one-hour simulation with the AM peak period volumes provided. Scheme (1) adds 8 more priority rules (for the exclusive left turn bays) to the base case. These additional priority rules decreased the number of rear-end and lane-changing by 3 and 4, respectively. However, the number of crossing conflicts remained unchanged. Overall, there were seven less conflicts than the base case modeling scheme. Scheme (2) adds another 16 rules to compensate for the lack of yellow extension and red clearance, thereby allowing the left turning cars to complete their turning movements. The additional rules decreased the number of crossing and lane-changing conflicts by 18 and 8, respectively, but increased the number of rear-end conflicts by 3. Overall, there were 26 less conflicts than the base case. Also, compared with the base case, the additional rules of Scheme (1) decreased the simulated accidents by only one, while the additional rules of Scheme (2) were much more effective in reducing these simulated accidents, resulting in a drop of 22 accidents. However, even with these additional rules defined, the simulated accidents cannot be completely eliminated. This comparison shows that the way priority rules are defined can have a significant impact on SSAM output. 7.2 Effect of Varying the Gap Size Since most intersections operated under free flow traffic, the minimum gap size becomes the dominant parameter in defining the priority rules. The default gap size used for the 83 simulated intersections was 5 (sec) with an additional 0.5 (sec) for any extra crossing lane. It was noted that varying the minimum gap size had a significant effect on the discharge flow rate of an intersection. Therefore, it is expected to have an effect on 68 the number and type of the simulated conflicts. Using the same intersection described in Subsection 7.1, the effect of varying the minimum gap size on the numbers and types of conflicts was examined. Thus, Scheme (2) was adopted and the intersection was modified in VISSIM, to represent minimum gaps of 4, 5 and 6 seconds. Table (25) shows the numbers and types of conflicts obtained from SSAM. Table 25 - Number and Type of Conflicts based on Gap Size Minimum Gap (sec) Total Crossing Rear-End Lane-Change Simulated Accidents 4 111 3 89 19 7 5 93 6 71 16 5 6 104 4 80 20 9 The results show that as the minimum gap size increased from 4 to 5 (sec) the number of conflicts were reduced. However, when the minimum gap size was increased from 5 to 6 (sec) there was an increase in the number of conflicts with a marginal change in conflict types. This increase is due to the queuing of vehicles waiting to perform a right or left maneuver, which induces the following vehicles to change lanes abruptly causing the increase in conflicts and simulated accidents. Appendix (D) shows the total number of conflicts produced by each gap size for all of the 83 locations. Examining Table (D1) in the appendix shows that increasing the gap size doesn't necessarily reduce the total number of conflicts. On the contrary, increasing the gap size from 4 (sec) to 5 (sec) decreased the total number of conflicts in only 46% of the locations. Furthermore, increasing the gap size from 4 (sec) to 6 (sec) decreased the total number of conflicts in only 35% of the locations. Similar trends were noted for crossing, rear-end and lane changing conflicts. Tables (26) and (27) show the percentages of locations exhibiting a decrease in the number of conflicts due to a change in gap size, with and without simulated accidents. 69 Table 26 - Percentages of Locations Exhibiting a Decrease in the Number of Conflicts Due to a Change in Gap Size (Including Simulated Accidents) Gap Change from 4 Gap Change from 4 Gap Change from 5 to 5 (sec) to 6 (sec) to 6 (sec) Total 46% 35% 35% Crossing 40% 43% 30% Rear-End 42% 34% 35% Lane- 46% 37% 39% Changing Table 27 - Percentages of Locations Exhibiting a Decrease in the Number of Conflicts Due to a Change in Gap Size (Excluding Simulated Accidents) Gap Change from 4 Gap Change from 4 Gap Change from 5 to 5 (sec) to 6 (sec) to 6 (sec) Total 37% 34% 36% Crossing 29% 30% 27% Rear-End 42% 34% 36% Lane- 37% 30% 40% Changing 7.3 Effect of Changing the Lateral Clearance Parameter As mentioned earlier, the lateral clearance parameter had to be increased from 1.0 (m) to 1.5 (m) to prevent some of the abrupt lane changing behavior. Logically, this parameter has a pronounced effect on the capacity of an intersection. This section demonstrates the effect of varying the lateral clearance parameter on the numbers and types of conflicts. Ten intersections were randomly selected to study this effect. The results shown in Tables (28) and (29) are based on the average value of five simulated runs with different random seeds. When the simulated accidents were included in the total conflicts, Table (28) shows that four locations exhibited an increase in the total number of conflicts, five locations exhibited a decrease and one location was unaffected by increasing the lateral clearance from 1.0 (m) to 1.5 (m). When simulated accidents were excluded from the analysis, Table (29) reveals an increase in four locations, a decrease in four locations, while two locations were unaffected. As expected, most of the increase or decrease in the total number of conflicts is due to changes in the numbers of rear-70 end and lane changing conflicts, with minimal changes in the number of crossing conflicts. Table 28 - Effect of Varying Lateral Clearance (Simulated Accidents Included) Conflicts (Lateral Clearance 1.5 m) Conflicts (Lateral Clearance 1.0 m) ID Total Crossing Rear End Lane Changing Total Crossing Rear End Lane Changing 4 43 3 35 5 43 3 35 5 7 130 43 33 54 132 40 39 53 8 101 15 46 40 107 14 48 45 15 116 14 54 48 110 11 55 44 19 120 40 53 27 117 42 54 21 43 62 19 28 15 58 14 32 12 57 215 82 34 99 227 68 35 124 69 25 4 17 4 26 4 18 4 70 36 2 27 7 34 2 26 6 80 27 4 13 10 28 4 14 10 Table 29 - Effect of Varying Lateral Clearance (Simulated Accidents Excluded) Conflicts (Lateral Clearance 1.5 m) Conflicts (Lateral Clearance 1.0 m) ID Total Crossing Rear End Lane Changing Total Crossing Rear End Lane Changing 4 40 1 35 4 40 1 35 4 7 73 4 33 36 76 4 38 34 8 76 3 46 27 81 3 48 30 15 79 1 51 27 76 1 54 21 19 74 2 53 19 70 3 54 13 43 41 2 28 11 41 1 32 8 57 88 4 34 50 87 3 34 50 69 22 1 17 4 23 1 18 4 70 34 0 27 7 32 0 26 6 80 22 0 13 9 24 1 14 9 The results in Tables (29) and (30) show that the lateral clearance parameter does not have a uniform impact on the number of conflicts produced by SSAM. 71 8.0 Conclusions This thesis provides the results of the field validation plan for the Surrogate Safety Assessment Methodology (SSAM) aiming to compare the predictive safety performance capabilities of the SSAM approach with actual accident experience at Canadian signalized intersections. Using certain plausible data assembly guidelines, eighty-three 4-leg intersections located in British Columbia were selected from safety studies conducted by Hamilton Associates. Accident records were assembled from the auto insurance claims data files collected by the Insurance Corporation of British Columbia (ICBC). The VISSIM 4.1-12 program was chosen to perform the simulation because of its many well-recognized advantages. The VISSIM output files were read by the SSAM program to obtain the conflict data. The Generalized Linear Models (GLM) procedures required to estimate the accident/conflict prediction models were run using the SAS 9.1 statistical software. The modeling process and some key modeling assumptions in VISSIM were documented. Several modeling assumptions were presented including intersection geometry, signal control, detectors, speed profiles, vehicle type characteristics, traffic composition, priority rules and other aspects. As well, a number of important modeling issues were discussed. These issues include the modeling schemes, the occurrence of simulated accidents, the abrupt lane changing behavior experienced in VISSIM, and the modeling of left and right turn bay tapers. Five validations tests were proposed and they include the following five statistical tests: o Validation Test 1: Safety Ranking Analysis o Validation Test 2: Safety Ranking Analysis for Specific Incident Types o Validation Test 3: Conflict / Accident Paired Comparison o Validation Test 4: Conflict and Accident Prediction Model Comparative Analysis 72 o Validation Test 5: Conflict and Accident Prediction Model Comparative Analysis for Specific Incident Types In the safety ranking analysis test, a conflict risk score was allocated to each intersection according to the average conflict rate (ACR). In contrast, an accident prediction model was fit to determine the expected accident frequency at each site, which was combined with the actual accident counts to obtain Empirical Bayes (EB) safety estimates. The Potential for Improvement (PFI) was then calculated as the difference between the EB safety estimate and the predicted accident frequency. The ranking derived from SSAM's conflict prediction was compared to the ranking from the actual accident risk. The Spearman rank-correlation coefficient was used to determine the level of agreement between the two ranks. The correlation between the two ranks was very poor and non-significant at the 90% confidence level. Safety ranking analysis tests were also developed for three accident types: crossing, rear-end and lane-changing and tests similar to those outlined above were conducted producing similar results. The accident/conflict paired comparison involved fitting a regression equation that relates actual accidents to predicted conflicts. The goodness-of-fit was tested using the R-squared value. This test determines the strength of the relationship between predicted conflicts and actual accidents. It turned out that, albeit the correlation is weak, actual accidents are significantly related to predicted conflicts. The conflict/accident prediction model comparative analysis entails the development of two prediction models relating the conflicts predicted by SSAM and the real-world accidents to traffic volume characteristics of the intersections. The two prediction models were compared to determine whether the conflict prediction model can predict risk in a manner similar to the accident prediction model for intersections with the same characteristics. The comparison included several model applications such as the identification and ranking of accident prone locations. The identification results show that fewer locations were identified as accident prone by the conflict 73 models. Further, only a few intersections were identified as accident prone by both the accident and conflict models. The PFI and ratio rankings were obtained using the total/severe conflict prediction model on one hand and the accident prediction model on the other hand. The Spearman rank-correlations for both the PFI and ratio rankings were very poor and non-significant at the 90% confidence level. Accident/conflict prediction models were also developed for the three accident types: crossing, rear-end and lane-changing, and tests similar to those outlined above were conducted producing similar results. Since the way an intersection is modeled into VISSIM has a pronounced effect on the SSAM output the following validation issues were explored: o Effect of redefining the priority rules. o Effect of varying the gap size. o Effect of varying the lateral clearance. Three modeling schemes were used to examine the effects of redefining the priority rules on the number of simulated conflicts. Compared with a base case having a total of 8 priority rules for permissive left and right turn on red; adding 8 more priority rules (Scheme 1) for the exclusive left turn and right turn on red movements increased the number of conflicts. To compensate for the lack of yellow extension and red clearance, 16 priority rules (Scheme 2) were added to allow the left turning cars to complete their turning movements, resulting in a decrease in the number of conflicts. Furthermore, the number of conflicts including simulated accidents for Scheme (1) was higher than Scheme (2). This was expected because the addition of 16 priority rules in Scheme (2) led to a reduction in simulated conflicts. But these additional 16 priority rules have increased the number of conflicts due to some 'would be' simulated accidents becoming conflicts with low TTC values. Minimum gaps of 4, 5 and 6 seconds were used in VISSIM to study the effect of varying the gap size on the numbers and types of conflicts. The results showed that as the minimum gap size increased from 4 to 5 (sec), the number of conflicts seem 74 to drop. However, increasing the minimum gap size to 6 (sec) resulted in an increase in the number of conflicts. This increase is due to the queuing of cars waiting to perform a right or left maneuver, which induces the following cars to change lanes abruptly causing the increase in conflicts and simulated accidents. To this time there is no clear justification for that abrupt lane-changing behavior. Certain measures were taken into account to reduce the effect of such unusual behavior such as extending the lengths of each approach to provide vehicles with significantly more time to decide on their downstream path. The driver behavior model was adjusted to allow an extra 0.5 (m) for the minimum lateral clearance. As suggested by PTV, the lateral clearance parameter was increased from 1.0 m to 1.5 m to prevent some of the abrupt lane changing behavior. Logically, this parameter has a pronounced effect on the capacity of an intersection. Thus, ten intersections were selected to study the effect of varying the lateral clearance parameter on the number and type of conflicts. The results of the comparison yielded inconsistent results, as some intersections exhibited either a random increase or decrease in consequent conflicts. In summary, the safety measures computed from the simulated conflicts were poorly related to those of actual accidents. Although VISSIM has been widely praised for its simulation abilities, it was not developed to perform traffic safety analysis. For the time being, VISSIM does not seem to be the most appropriate tool to use in the validation of the SSAM approach. VISSIM's lane changing algorithm needs to be modified to eliminate the abrupt lane-changing conflicts/simulated accidents that occurred during low and high traffic flows. Such an abrupt behavior increased the number of conflicts significantly. Furthermore, to ensure proper and consistent results certain guidelines regarding modeling schemes, issues and default parameters need to be established. The analysis showed SSAM's sensitivity towards a number of VISSIM's design parameters. This sensitivity hinders SSAM's ability to achieve what it was designed to perform. SSAM was developed to conduct traffic 75 safety studies using multi-purpose micro-simulation programs, in this case VISSIM. It is argued that the use of micro-simulation would provide valuable insights into the relative safety impacts brought about by changes of traffic flow, various ITS devices placed inside the vehicle or on the roadway, different signaling strategies, and many other dependent aspects. However, it was shown that relative changes in certain VISSIM design parameters seem to yield different conflicts from SSAM. 76 9.0 Future Research Based on the findings of this thesis the following future recommendations are proposed: o Perform the field validation plan for the other three compatible multi-purpose micro-simulations (MPMS), namely; PARAMICS, TEXAS and AIMSUN. If these software yield similar results then other surrogate safety measures should be explored, as discussed below. However, if better or more promising results were provided by any of these softwares then further research is to be conducted to identify problems involved in the other MPMS programs. o Regarding VISSIM, it is crucially important for the developers of VISSIM to investigate and probably modify VISSIM's lane changing algorithm to eliminate the abrupt lane changing behavior. This behavior increased the number of conflicts during free flow conditions and its impact was more pronounced during congested traffic flows. It is difficult to manually disregard such conflicts as it violates SSAM's primary purpose in reducing the time and level of human involvement. o Certain modeling guidelines have to be provided for VISSIM to allow for consistent and reliable estimates to be computed from SSAM. Currently, SSAM is very sensitivity to certain design parameters. During the course of this thesis a number of parameters were changed from their default settings to eliminate or provide a solution to some unrealistic driving behaviors. If SSAM is to be used by the general public to identify safety solutions or provide comparative analysis between different design options then it is likely for two different users to come up with contradicting conclusions based on their own assumptions and choice of design parameters. Once again, such a situation would violate SSAM's primary objective of providing the public with a reliable traffic safety tool. 77 Perhaps another future research topic would be the exploration of other surrogate safety measures to be incorporated into SSAM that are more likely to yield better results and do not require a significant amount of coding changes in terms of compatibility with MPMS programs. The work by Minderhood and Bovy ( 2 0 0 0 ) provides two valuable Traffic Conflict Technique (TCT) measures based on the Time-To-Collision (TTC). This new approach provides valuable comparative safety data and is useful in calibrating simulation models. A new approach to predict the safety of an intersection using traffic conflicts is being developed in the University of British Columbia. The process is denoted as 'Automated Traffic Conflict Analysis' (ATCA). The new approach utilizes the concept of Traffic Conflict Technique that was used in this thesis. Yet, the process uses real-time footage from intersections with mounted video cameras. The approach utilizes two algorithms to detect and identify potential conflicts from video images. The detection algorithm is used to identify vehicle trajectories as they approach an intersection, while the identification algorithm is used to determine wither a traffic conflict has occurred. The success of this approach would provide field observers with a number of potential traffic conflicts for examination. The process would reduce the level of human involvement and thus their subjectivity. Furthermore, the process would eliminate hours of vehicle interactions with no traffic conflicts occurring, thus, reducing the overall cost associated with hiring a team of professional observers. 78 10.0 References Allen, B. et al. (1978). Analysis of traffic conflicts and collisions. Transportation Research Record, No. 667. Annual Traffic Collision Statistics. (2004). Police attended Injury and Fatal Accidents. British Columbia. Archer, J . (2000). Developing the potential of micro-simulation modeling for traffic safety assessment. Proceedings of 13th Annual International Cooperation on Theories and Concepts on Traffic Safety Workshop. Corfu, Greece. Archer, J . (2001). Traffic Conflict Technique Historical to current State-of-the-Art. Effektmodeller for vagtrafikanlaggningar (EMV), TRITA-INFRA 02-010, ISSN 1651-0216, ISRN KTH/INFRA-02/010-SE, Stockholm, September, 2001. Canadian Council of Motor Transport Administration. (2002). Road Safety Vision 2010 Making Canada's Roads the Safest in the World, 2002 Annual Report. Transport Canada. Cooper, D. and Ferguson, N. (1976). A conflict simulation model. Traffic Engineering and Control, Vol. 17, pp. 306-309. Crowe, C. E. (1990). Traffic Conflict Values for three-leg, un-signalized intersections. Trans. Res. Rec , 1287, 185-194, Washington, DC. Darzentas, J . et al. (1980). Minimum acceptance gaps and conflict involvement in a single crossing maneuver. Traffic Engineering and Control, Vol. 21, pp. 58-62. de Leur, P. and Sayed, T. (2001). Using claims prediction model for road safety evaluation. Published on the NRC Research Press Web site at http://cjce.nrc.ca. pp. 804-812. Depue, L. et al. (2001). Safety Conscious Planning. Transportation Research Circular, Number E-C025. Fazio, J . and Rouphail, N. (1990). Conflict simulation in INTRAS: Application to weaving area capacity analysis. Transportation Research Record, Vol. 1287. Fazio, J . et al. (1993). Use of freeway conflict rates as an alternative to accident rates in weaving section safety analysis. Transportation Research Record 1401, 61-69. Fitzpatrick, K. et al. (2000). Accident mitigation guide for congested rural two-lane highways. NCHRP Report 440. National Academy Press, Washington, DC. 79 Gettman, D. and Head, L. (2003). Surrogate Safety Measures from Traffic Simulation Models, FHWA Report No. FHWA-RD-03050. http://www.tfhrc.gov/safety/pubs/03050/ Ghaffari, M. (1990). A stochastic computer simulation of traffic conflicts. Ph.D. Dissertation. University of Illinois at Chicago. Glauz, W. and Migletz, D. (1980). Application of traffic conflict analysis at intersections. NCHRP Report 219, Transportation Research Board, Washington, DC. Glennon, J.C. et al. (1974). Critique of Traffic Conflict Technique. Hwy. Res. Rec. 460. pp. 32-38. Hauer, E. (1980). Bias-by-Selection. Accident Analysis and Prevention, Vol. 12, pp. 113-117. Hauer, E. (1986). On the estimation on the expected number of accidents. Accident Analysis and Prevention, Vol 18, No. 1, pp. 1-12. Hauer, E. (1988). Case for Science-Based Road Safety Design and Management. Highway-Safety: At the Crossroads; pp. 241-267. Hauer, E. (1997). Observational Before-After Studies in Road Safety. Book, Oxford, Pergamon. Hauer, E., Ng, J . C. N., and Lovell, J . (1988). Estimation of Safety at Signalized Intersections. Transportation Research Record 1185, Transportation Research Board, Washington, D.C., 48-61. Hayward, J . (1972). Near-miss determination through use of a scale of danger. Report TTSC-7115. University Park, The Pennsylvania State University. Hernandez, V. (1982). A microscopic digital computer simulation of traffic conflicts at four-way fixed-time signalized urban intersections. Ph.D. Dissertation, University of California at Los Angeles. Higle, J.L. and Witkowski, J.M. (1988). Bayesian Identification of Hazardous Locations. Transportation Research Record 1185, Transportation Research Board, National Research Council, Washington, D.C., pp. 24-36. Huang, P. and Pant, P. (1996). Simulation neural-network model for evaluating dilemma zone problems at high-speed signalized intersections. Transportation Research Record, Vol. 1456. 80 Hummer, J.E. (1994). Intersection and Driveway Studies and Traffic Conflict Studies. Manual of Transportation Engineering Studies, Institute of Transportation Engineers (ITE), Washington, DC. Hyden, C. (1975). Relations between Conflict and Traffic Accidents. Department of Traffic Planning and Engineering, Lund Institute of Technology, Sweden. Hyden, C. (1987). The development of a method for traffic safety evaluation: The Swedish Traffic Conflicts Technique. Bulletin 70. Institute for Trafikteknik, LTH, Lund Jovanis, P.P., and Chang, H. L. (1986). Modeling the Relationship of Accidents to Miles Traveled. Transportation Research Record 1068, Transportation Research Board, Washington, D.C, 42-51. Katamine, N. (2000)a. Nature and Frequency of Secondary Conflicts at Un-signalized Intersections. Journal of Transportation Engineering, Vol. 126, No. 2. Katamine, N. (2000)b. Various Volume Definitions with Conflicts at Un-signalized Intersections. Journal of Transportation Engineering, Vol. 126, No. 1. Kosonen, I. (1999). HUTSIM: Urban traffic simulation and control model: Principles and applications. Ph.D. Dissertation, Transportation Engineering, Helsinki University of Technology. Kruysse, H. (1991). The subjective evaluation of traffic conflicts based on an internal concept of dangerousness. Accident Analysis and Prevention, Vol. 23, No. 1. Kulmala, R. (1995). Safety at Rural Three-and Four-arm Junctions: Development of Accident Prediction Models. Technical Research Centre of Finland, VTT 233, Espoo. Maier, R. (2006). Safety in rural Europe - Reducing Casualties on country roads in the EU. 8 t h European Transport Safety Lecture. European Transport Safety Council. Malaterre, G. and Muhlrad, N. (1979). International experience of conflict observation at a single location. Proceeding of 2nd International Workshop on Traffic Conflict Techniques, Transport and Road Res. Lab., SR 557, Berkshire, England. McDowell, M.R.C. et al. (1983). Gap acceptance and traffic conflict simulation as a measure of risk. SSR 776, Transport and Road Res. Lab. pp. 1-27. Mehmood, A. et al. (2001). Simulation of road accidents using systems dynamics. Paper presented at the 80th Annual Meeting of the Transportation Research Board, Washington, DC. Miaou, S., and Lum, H. (1993). Modeling Vehicle Accident and Highway Geometric Design Relationships. Accident Analysis and Prevention, 25(6), 689-709. 81 Miaou, S.P. (1996) Measuring the Goodness-of-Fit of Accident Prediction Models. Report No. FHWA-RD-96-040, Federal Highway Administration, McLean, VA. Migletz, D. et al. (1985). Relationships between traffic conflicts and accidents. Report No. FHWA-RD-84-042, Federal Highway Administration, Washington, DC. Minderhoud, M. and Bovy, P. (2001). Extended time to collision measures for road traffic safety assessment. Accident Analysis and Prevention, Vol. 33, pp. 89-97. Muhlrad, N. (1987). Road Safety Research for Developing Countries: Methodologies and first results. INRETS, Synthese number 7, Arcueil, France. Murray, C. and Alan, D. L. (1995). The Global Burden of Disease. World Health Organization/Harvard School of Public Health/World Bank. Nilsson G. (1989). Safety for different road-user groups: Data collection, problems, consequences and costs. Proceedings of the Second European Workshop on Recent Developments in Road Safety Research, Paris, 25-27 January 1989. Actes INRETS no. 17, Arcueil, France. ISBN 2-85782-269-3. OECD, (1984). Integrated road safety programmes. Organization for Economic Cooperation and Development, Paris, France. ISBN 92-64-12620-1. OCED/ECMT Transport Research Center. (2006). 17 t h International ECMT/OECD Symposium on Transport Economics and Policy Berlin. Parker, M. and Zegeer, C. (1988). Traffic conflict technique for safety and operations: Engineers guide, Report No. FHWA-IP-026, FHWA, Washington, DC. Parker, M. and Zegeer, C. (1989). Traffic conflict technique for safety and operations: Observers guide, Report No. FHWA-IP-027, FHWA, Washington, DC. Peden M, et al. (2004). World report on road traffic injury prevention. Geneva: World Health Organization (WHO). Perkins, D. and Bowman, B. (1986). Effectiveness evaluation by using non-accident measures of effectiveness. Transportation Research Record, Vol. 905. Perkins, D. and Harris, J . (1967). Criteria for traffic conflict characteristics. Report GMF 632, General Motors Corp., Warren, Ml. Persaud, B. and Mucsi, K. (1995). Microscopic accident potential models for two-lane rural roads. Transportation Research Record. No. 1485, pp. 134-139. Rao, V. and Regaraju, V. (1997). Probabilistic Model for Conflict at Urban Uncontrolled Intersections. Journal of Transportation Engineering, Jan./Feb. 82 Rao, V. and Regaraju, V. (1998). Modeling conflicts of heterogeneous traffic at urban uncontrolled intersections. Journal of Transportation Engineering, Jan./Feb. Rioux, T. and Lee, (1977). Texas: A microscopic traffic simulation package for isolated intersections. Center for Highway Research, University of Texas at Austin. Risser, R. (1985). Behavior in Traffic Conflict Situations. Accident Analysis and Prevention, Vol.17:2, pp. 179-197 Road Safety and Motor Vehicle Regulation Directorate. (2004). Road Safety in Canada 2001. The Canadian Council of Motor Transport Administrators (CCMTA) Standing Committee on Road Safety Research and Policies, Transport Canada. Sawalha, Z. and Sayed, T. (2001). Evaluating the Safety of Urban Arterial Roadways. Journal of Transportation Engineering, ASCE, Vol. 127(2), pp. 151-158. Sayed, T. (1997). Estimating the Safety of Unsignalized Intersections Using Traffic Conflicts. Third International Symposium on Intersections without Traffic Signals, Transportation Research Board, Oregon, July 21-23. Sayed, T. and Abdelwahab, W. (1997). Using Accident Correctability to Identify Accident Prone Locations, Journal of Transportation Engineering, ASCE, Vol. 123(2). Sayed, T. and Rodriguez, F. (1999). Accident Prediction Models for Urban Unsignalized Intersections in British Columbia. Transportation Research Record,. Vol. 1665, pp. 93-99. Sayed, T. et al. (1994). Simulation of Traffic Conflicts at Unsignalized Intersections with TSC-Sim. Accident Analysis and Prevention, Vol. 26(5), pp. 593-607. Sayed, T. et al. (2004). Evaluating the Insurance Corporation of British Columbia Road Safety Improvement Program. Transportation Research Record, Vol. 1865, pp. 57-63. Sayed, T., and Zein, S. (1998). Traffic Conflict Models and Standards for Signalized and Unsignalized Intersections. Proceedings of the CSCE 2nd Transportation Specialty Conference, June 10-13, Halifax. Sayed, T., and Zein, S. (1999). Traffic Conflict Standards for Intersections. Transportation Planning and Technology , Vol. 22, pp. 309-323. Spicer, B. R. (1971). A Pilot Study of Traffic Conflicts at Rural Dual Carriageway Intersections. Road Res. Lab. LR 520. Berkshire, England. Spicer, B. R. (1972). A Traffic Conflict Study at Intersections on the Andoversford by-pass. Transport and Road Res. Lab. LR 551. Berkshire, England. 83 Spicer, B.A. (1973). Study of traffic conflicts at six intersections. TRRL Report LR551. Transport and Road Research Laboratory, Berkshire, UK Svensson, A . (1992). Vidareutveckling och validering av den svenska konflikttekniken (Development and validation of the Swedish traffic conflicts technique). LTH, Lund. Thompson, H. and Perkins, D. (1983). Surrogate measures for accident experience at rural isolated horizontal curves. Paper presented at the Annual Meeting of the Transportation Research Board, Washington, DC. Tiwari, G. et al. (1995). Conflict analysis for prediction of fatal accident locations in mixed traffic streams. Proceedings of the 29th Annual Association for the Advancement of Automotive Medicine, Oct 17-18. Chicago, IL. Transportation in Canada 2004 Annual Report. (2004) Transport Canada, Minister of Public Works and Government Services, Canada. William, T.M. (1972). An Evaluation of Traffic Conflict Technique. Hwy. Res. Rec. 384, 1-8, Washington, DC. World Health Organization. (1999). World Health Report 1999: Making a Difference. Geneva. Zegeer, C. V. and Deen, R. C. (1978). Traffic conflict as a diagnostic tool in highway safety. Transp. Res. Rec. 667, Transportation Research Board, Washington, D.C, 48-55. Zein, S. et al. (1995) Advancements in Traffic Conflict Research: Standards and Simulation. 1995 Annual Conference, Transportation Association of Canada (TAC), Victoria, Canada, October 22-25. 84 APPENDIX A - Canadian (83) Intersections Data Summary B - Demonstration of Abrupt Lane-Changing Behavior C - Summary of Average Conflicts per hour, Including and Excluding Simulated Accidents D - Effect of Varying the Minimum Gap Size 85 APPENDIX A - Canadian (83) Intersections Data Summary Tab e A1- Data Summary ID Intersection District N u m b e r of L a n e s for E a c h A p p r o a c h N B W B S B E B Intersection Image S igna l T y p e 1 Gi l ley & K ingsway Burnaby 2 3 2 3 Aer ia l S e m i Ac tua ted 2 C o a s t Mer id ian & Prair ie Port Coqu i t lam 4 3 3 3 Aer ia l Ful ly Ac tua ted 3 C o a s t Mer id ian & Rober t son Port Coqu i t lam 1 1 2 1 Aer ia l Pre- t imed 4 Oxford & Prair ie Port Coqu i t lam 1 1 1 1 Aer ia l Ful ly Ac tua ted 5 128 & 96 Sur rey 3 3 3 3 Aer ia l S e m i Ac tua ted 6 Griff i ths & K ingsway Burnaby 2 3 2 3 Aer ia l S e m i Ac tua ted 7 Wi l l ingdon & M o s c r o p Burnaby 4 3 4 3 Aer ia l S e m i Ac tua ted 8 E d m o n d s & C a n a d a Burnaby 3 3 4 3 Aer ia l S e m i Ac tua ted 9 Sprott & Doug las Burnaby 1 1 1 1 Aer ia l S e m i Ac tua ted 10 132 & 88 Sur rey 2 3 2 3 Aer ia l S e m i Ac tua ted 11 128 & 88 Sur rey 3 3 3 3 Aer ia l S e m i Ac tua ted 12 C o a s t Mer id ian & L o u g h e e d Port Coqu i t lam 2 4 2 4 Aer ia l Pre- t imed 13 Mar iner & C o m o L a k e Coqu i t lam 3 3 4 3 Aer ia l Pre- t imed 14 J o h n s o n & Gui ldford Coqu i t lam 3 3 4 3 Aer ia l Pre- t imed 15 J o h n s o n & Barnet Coqu i t lam 5 6 4 6 Aer ia l Pre- t imed 16 J o h n s o n & Dav id Coqui t lam 4 3 4 3 Aer ia l Pre- t imed 17 M c C a l l u m & Marsha l l Abbots ford 4 3 4 4 Aer ia l S e m i Ac tua ted 18 Mar iner & D e w d n e y Trunk Coqui t lam 4 3 4 4 Aer ia l P re - t imed 19 128 & 76 Sur rey 3 2 3 2 S c h e m a t i c S e m i Ac tua ted 20 Y a l e & Airport Chi l l iwack 4 2 4 2 S c h e m a t i c Ful ly Ac tua ted 21 152 & 104 Sur rey 4 4 4 4 S c h e m a t i c S e m i Ac tua ted 22 G love r & Logan Lang ley 3 3 3 3 S c h e m a t i c S e m i Ac tua ted 23 Y a l e & Hodg ins Chi l l iwack 3 3 3 2 S c h e m a t i c Ful ly Ac tua ted 24 Y a l e & Hock ing Chi l l iwack 3 2 3 2 S c h e m a t i c Ful ly Ac tua ted 25 Bo rden & M c K e n z i e S a a n i c h 2 3 2 3 S c h e m a t i c S e m i Ac tua ted 26 152 & 56 Sur rey 4 3 4 2 S c h e m a t i c S e m i Ac tua ted 27 152 & 88 Sur rey 3 3 3 3 S c h e m a t i c S e m i Ac tua ted 28 Q u a d r a & M c K e n z i e S a a n i c h 3 4 3 3 S c h e m a t i c Pre- t imed 29 S a a n i c h & M c k K e n z i e S a a n i c h 2 2 2 2 S c h e m a t i c Pre- t imed 86 ID Intersection District N u m b e r of L a n e s for E a c h A p p r o a c h N B W B S B E B Intersection Image S igna l T y p e 30 She lbou rne & M c K e n z i e S a a n i c h 4 3 3 3 S c h e m a t i c S e m i Ac tua ted 31 Glanford & M c K e n z i e S a a n i c h 3 4 3 4 S c h e m a t i c S e m i Ac tua ted 32 Doug las & Kelv in S a a n i c h 3 2 3 2 S c h e m a t i c S e m i Ac tua ted 33 Doug las & S a a n i c h S a a n i c h 4 4 3 3 S c h e m a t i c S e m i Ac tua ted 34 G o r d o n & Harvey K e l o w n a 4 5 4 5 S c h e m a t i c S e m i Ac tua ted 35 G o r d o n & Spr ingf ie ld K e l o w n a 3 3 3 3 S c h e m a t i c S e m i Ac tua ted 36 G o r d o n & Bernard K e l o w n a 3 3 3 3 S c h e m a t i c S e m i Ac tua ted 37 Whar f & Dolphin Sech le t 2 2 2 2 S c h e m a t i c S e m i Ac tua ted 38 Bounda ry & Lougheed V a n c o u v e r 5 4 4 4 S c h e m a t i c S e m i Ac tua ted 39 G a t e n s b u r y & C o m o L a k e Coqu i t lam 1 2 1 2 S c h e m a t i c Ful ly Ac tua ted 40 Poir ier & C o m o L a k e Coqu i t lam 1 2 1 2 S c h e m a t i c Ful ly Ac tua ted 41 Harr is & H a m m o n d Pitt M e a d o w s 3 2 3 1 S c h e m a t i c S e m i Ac tua ted 42 B ryne & Mar ine Burnaby 2 3 3 3 S c h e m a t i c Pre- t imed 43 Pat te rson & K ingsway Burnaby 2 3 1 3 S c h e m a t i c S e m i Ac tua ted 44 B r o o k e & Nordel Del ta 2 2 2 2 S c h e m a t i c S e m i Ac tua ted 4 5 5 6 & 1 2 Del ta 3 3 3 2 S c h e m a t i c Pre- t imed 46 N o . 5 & S teves ton R i chmond 3 3 2 4 S c h e m a t i c S e m i Ac tua ted 47 No . 5 & Wes tmins te r R i chmond 4 4 3 4 S c h e m a t i c S e m i Ac tua ted 48 G a r d e n Ci ty & Westmins te r R i chmond 4 4 3 3 S c h e m a t i c S e m i Ac tua ted 49 G a r d e n Ci ty & Alderbr idge R i chmond 4 3 4 4 S c h e m a t i c S e m i Ac tua ted 50 No . 4 & Alderbr idge R i chmond 3 3 3 4 S c h e m a t i c S e m i Ac tua ted 51 She l l & A lderbr idge R i chmond 4 4 4 4 S c h e m a t i c S e m i Ac tua ted 52 F rase r & 96 Sur rey 3 4 2 4 S c h e m a t i c S e m i Ac tua ted 53 K ing G e o r g e & 72 Sur rey 4 4 4 4 S c h e m a t i c S e m i Ac tua ted 54 140 & 72 Sur rey 1 3 2 3 S c h e m a t i c S e m i Ac tua ted 55 184 & Fraser Sur rey 3 1 3 2 S c h e m a t i c Pre- t imed 56 64 & Fraser Sur rey 4 3 3 3 S c h e m a t i c Pre- t imed 57 Wi l lowbrook & F rase r Sur rey 1 4 3 4 S c h e m a t i c Pre- t imed 58 Scot t & 72 Sur rey 3 4 3 3 S c h e m a t i c S e m i Ac tua ted 59 Scot t & 80 Sur rey 3 3 3 3 S c h e m a t i c S e m i Ac tua ted 60 Scot t & 64 Sur rey 3 3 3 3 S c h e m a t i c S e m i Ac tua ted 61 Scot t & 58 Sur rey 3 4 3 4 S c h e m a t i c S e m i Ac tua ted 62 S h a w & Hwy 101 G i b s o n s 1 2 2 2 S c h e m a t i c S e m i Ac tua ted 63 S c h o o l h o u s e & Aus t in Coqu i t lam 1 2 1 2 S c h e m a t i c S e m i Ac tua ted 87 ID Intersection District Number of L a n e s for E a c h A p p r o a c h N B W B S B E B Intersection Image S igna l T y p e 64 Marmon t & Aust in Coqu i t lam 2 3 1 3 S c h e m a t i c S e m i Ac tua ted 65 Ruper t & 22 V a n c o u v e r 2 1 2 1 Aer ia l S e m i Ac tua ted 66 R u p e r t & 1 V a n c o u v e r 3 2 4 3 S c h e m a t i c Pre - t imed 67 Ruper t & G randv iew V a n c o u v e r 3 4 4 4 S c h e m a t i c S e m i Ac tua ted 68 Ker r & 49 V a n c o u v e r 3 3 3 3 S c h e m a t i c Pre - t imed 69 S h a u g h n e s s y & L ions Port Coqu i t lam 4 2 3 1 S c h e m a t i c S e m i Ac tua ted 70 S h a u g h n e s s y & Pitt R iver Port Coqu i t lam 2 2 3 2 S c h e m a t i c S e m i Ac tua ted 71 Spa l l & Spr ingf ie ld K e l o w n a 1 3 3 3 S c h e m a t i c Pre- t imed 72 272 & Fraser Lang ley 2 2 2 2 S c h e m a t i c S e m i Ac tua ted 73 240 & Fraser Lang ley 2 2 2 2 S c h e m a t i c S e m i Ac tua ted 74 216 & Fraser Lang ley 4 3 4 4 S c h e m a t i c S e m i Ac tua ted 75 B radner & F rase r Lang ley 2 2 1 2 S c h e m a t i c S e m i Ac tua ted 76 V e d d e r & S p r u c e Chi l l iwack 3 3 3 1 S c h e m a t i c S e m i Ac tua ted 77 V e d d e r & W a t s o n Chi l l iwack 3 3 3 2 S c h e m a t i c S e m i Ac tua ted 78 V e d d e r & Knight Chi l l iwack 3 2 3 2 S c h e m a t i c S e m i Ac tua ted 79 V e d d e r & Luckakuck Chi l l iwack 4 4 4 4 S c h e m a t i c S e m i Ac tua ted 80 3 4 & 2 5 V e r n o n 3 3 3 3 S c h e m a t i c S e m i Ac tua ted 81 Wi l l iam & lynn North V a n c o u v e r 3 3 2 4 S c h e m a t i c S e m i Ac tua ted 82 Mounta in & Lynn North V a n c o u v e r 2 2 2 3 S c h e m a t i c S e m i Ac tua ted 83 227 & Dewdney Map le R idge 2 3 2 3 S c h e m a t i c S e m i Ac tua ted 88 APPENDIX B - Demonstration of Abrupt Lane-Changing Behavior This sect ion demonstrates the abrupt lane-changing behavior exper ience in V I S S I M . Figure (B1) shows two vehic les stopped at the red light on the eastbound approach (EB) , with two more vehic les arriving after each other. The vehicle with the red circle will be referred to as 2 n d vehicle and the vehicle just ahead will be referred to as the 1 s t vehic le. Figure B1 - 1 and 2 n d Veh ic les Arriving 8 9 In Figure (B2) the 1 s t vehicle has come to a complete stop, while the 2ND vehicle have decided to change lanes and queue up on the outer left lane of the eastbound approach. B • * Figure B2 - 1 s t Vehicle Stops and 2NA Vehicle decides to Change Lanes 90 In Figure (B3) the 2 n d vehicle changes lanes and stops diagonally. Si i f QQ Ai \ Figure B3 - 2"1 Vehicle Changes Lanes & Stops 91 In Figure (B4) a 3 vehicle (with a red circle around it) is arriving at the eastbound intersection. Figure B4 - ^ V e h i c l e Arriving 92 In Figure (B5) the 3 r d vehicle fails to recognize the presence of the 2 n d vehicle and crashes into it. £S Vfe* • Syaf Corrtjcsl EvaJua&cn awwaiofi ^ejeriatoi 0 \ Figure B5 - 3 r d Vehicle crashing into 2 n d Vehicle 93 APPENDIX C - Summary of Average Conflicts per hour -Including and Excluding Simulated Accidents Table C.1 - Average Conflicts per hour for each Intersection based on the Total and Type of Conflicts (Results are based on the average of five 1-hour simulation runs) Simulated Accidents Included Simulated Accidents Excluded ID Total Crossing Rear End Lane Changing Total Crossing Rear End Lane Changing 1 130 25 63 42 84 1 61 22 2 55 5 39 12 49 0 39 10 3 67 7 39 20 54 1 39 14 4 43 3 35 5 40 1 35 4 5 52 3 35 14 43 0 35 7 6 155 43 61 51 85 1 61 22 7 131 43 33 54 73 4 33 36 8 101 15 46 40 75 3 46 27 9 69 14 47 8 56 6 47 4 10 80 12 54 14 67 2 54 11 11 69 9 38 22 55 1 38 16 12 55 1 30 24 45 0 30 15 13 174 13 44 118 119 2 44 74 14 78 13 32 32 57 1 32 23 15 116 14 54 48 80 1 51 27 16 50 4 16 30 43 0 16 26 17 159 58 54 47 90 5 53 31 18 108 6 50 52 76 1 49 26 19 119 40 53 27 74 2 53 19 20 117 27 41 50 75 3 41 31 21 93 19 38 36 63 1 38 24 22 15 0 7 8 13 0 7 6 23 184 81 29 75 87 9 28 49 24 190 85 49 56 95 9 46 40 25 106 41 47 18 62 3 47 12 26 170 60 46 63 92 6 46 40 27 96 19 49 28 71 2 49 20 28 79 0 38 41 56 0 38 18 29 69 13 37 19 48 1 37 10 30 67 1 46 19 58 0 46 12 31 106 6 51 49 80 0 51 29 32 86 19 41 27 57 2 41 15 33 137 21 37 78 70 2 37 30 9 4 ID Simulated Accidents Included T - . . ^ Rear Lane Total Crossing ,- . ~. a End Changing Simulated Accidents Excluded -r . . ~ Rear Lane Total Crossing ._ , ~. a End Changing 34 123 24 57 42 82 2 57 24 35 60 6 41 13 49 0 41 8 36 61 6 43 13 54 1 43 10 37 55 10 37 9 46 2 37 7 38 193 94 50 50 92 8 50 34 39 74 5 49 19 58 1 49 8 40 84 5 52 27 65 0 52 13 41 27 1 14 11 24 1 14 9 42 254 51 60 143 153 3 60 90 43 61 19 28 15 41 2 28 11 44 64 7 45 13 50 1 45 4 45 33 3 22 8 28 0 22 6 46 52 7 31 14 39 0 31 8 47 94 26 45 23 63 3 45 16 48 92 2 49 42 73 1 48 23 49 104 18 53 34 75 1 53 21 50 83 22 39 22 57 2 39 15 51 130 46 41 43 69 8 41 20 52 66 2 30 35 57 1 30 27 53 142 38 61 42 89 2 61 26 54 104 39 37 28 55 2 37 16 55 188 57 47 83 88 2 47 38 56 115 11 31 73 76 3 31 41 57 215 82 34 99 88 4 34 50 58 49 3 35 11 44 0 35 9 59 58 13 33 12 44 2 33 9 60 145 21 84 41 107 1 84 22 61 75 14 36 25 55 2 36 16 62 41 4 23 14 30 1 23 5 63 76 16 28 31 45 2 28 15 64 57 16 24 16 38 2 24 12 65 144 62 66 16 79 5 66 8 66 135 32 50 53 87 4 50 33 67 148 57 59 32 84 3 59 22 68 114 23 70 22 87 2 70 15 69 25 4 17 4 21 1 17 4 70 36 2 27 7 33 0 27 7 71 62 9 22 31 43 1 22 20 72 78 43 13 22 40 6 13 21 73 54 8 38 9 47 3 38 6 74 82 2 44 36 63 0 44 19 95 ID Simulated Accidents Included -,. , , ~ Rear Lane Total Crossing , ~. M End Changing Simulated Accidents Excluded -T- . . ~ Rear Lane Total Crossing ._ . ~. a End Changing 75 61 15 34 13 43 1 33 9 76 73 12 19 43 61 1 19 40 77 78 11 43 24 63 1 43 20 78 98 35 44 18 64 4 44 15 79 125 17 22 86 70 1 22 47 80 28 4 13 10 22 0 13 9 81 33 1 26 6 32 1 25 6 82 79 20 36 24 58 4 36 19 83 45 1 37 7 44 0 37 6 96 APPENDIX D - Effect of Varying the Minimum Gap Size Table D.1 - Gap Sizes of 4, 5 and 6 sec and their corresponding total Conflicts. Gap Size 4 sec Gap Size 5 sec Gap Size 6 sec ID Including Excluding Including Excluding Including Excluding Simulated Simulated Simulated Simulated Simulated Simulated Accidents Accidents Accidents Accidents Accidents Accidents 1 96 76 92 80 122 97 2 41 38 36 34 38 35 3 76 70 80 79 74 72 4 47 45 22 21 52 49 5 42 41 42 42 42 38 6 206 106 217 122 201 125 7 122 90 121 90 130 107 8 88 77 95 85 95 79 9 54 53 66 64 57 57 10 84 80 79 70 81 71 11 77 72 72 67 74 68 12 205 130 181 115 176 126 13 55 53 59 58 65 64 14 77 68 76 65 84 76 15 158 87 153 88 151 83 16 55 54 59 58 56 55 17 122 102 120 102 140 101 18 147 119 158 124 152 122 19 70 65 65 60 70 62 20 81 68 114 100 100 91 21 84 67 70 58 71 62 22 25 23 28 26 23 21 23 121 99 89 76 115 99 24 88 71 134 114 158 131 25 128 108 118 100 95. 89 26 152 124 168 126 175 139 27 91 82 88 76 92 82 28 70 56 74 65 71 60 29 76 68 81 69 77 67 30 82 69 83 69 83 69 31 79 73 100 94 94 84 32 87 71 99 81 131 93 33 99 65 109 80 114 88 34 110 89 106 93 100 89 35 47 46 45 44 45 44 36 84 77 85 82 85 76 97 Gap Size 4 sec Gap Size 5 sec Gap Size 6 sec ID Including Excluding Including Excluding Including Excluding Simulated Simulated Simulated Simulated Simulated Simulated Accidents Accidents Accidents Accidents Accidents Accidents 37 56 56 52 51 73 71 38 111 86 137 102 165 120 39 75 64 62 59 77 69^ 40 93 77 109 86 97 73 41 19 16 24 23 22 21 42 141 112 187 146 155 119 43 52 50 42 38 63 61 44 64 63 67 63 75 69 45 30 27 33 31 34 32 46 54 52 53 48 61 52 47 73 64 73 67 72 64 48 101 85 22 20 102 92 49 98 92 81 75 88 83 50 74 71 63 57 67 62 51 79 62 86 70 85 73 52 84 79 72 69 70 66 53 114 107 122 100 108 93 54 59 52 63 53 83 75 55 196 141 238 159 172 131 56 134 103 124 100 98 91 57 289 128 287 134 286 146 58 51 50 54 53 55 54 59 51 47 51 45 51 45 60 130 120 134 125 143 131 61 61 56 64 57 66 61 62 37 32 36 32 35 31 63 58 40 76 58 93 58 64 46 43 48 45 58 53 65 103 101 84 82 96 96 66 143 117 132 107 162 134 67 103 78 101 90 119 99 68 115 106 94 89 106 95 69 33 32 34 34 38 36 70 34 34 36 35 35 34 71 43 38 42 36 50 43 72 43 43 47 47 64 64 73 51 50 61 60 61 59 74 83 67 83 67 82 68 75 45 45 48 48 54 53 76 52 48 68 62 76 71 77 70 65 63 61 66 61 78 94 74 79 59 85 70 98 ID Gap Size 4 sec Including Excluding Simulated Simulated Accidents Accidents Gap Size 5 sec Including Excluding Simulated Simulated Accidents Accidents Gap Size 6 sec Including Excluding Simulated Simulated Accidents Accidents 79 87 65 92 72 87 67 80 30 25 35 32 36 30 81 26 23 24 23 32 32 82 61 56 57 54 59 58 83 44 43 39 38 30 30 99 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0063300/manifest

Comment

Related Items