UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Sustainable road safety : development, transference and application of community-based macro-level collision.. Sun, Jianchen 2009-12-31

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
ubc_2009_spring_sun_jianchen.pdf [ 3.7MB ]
[if-you-see-this-DO-NOT-CLICK]
Metadata
JSON: 1.0063125.json
JSON-LD: 1.0063125+ld.json
RDF/XML (Pretty): 1.0063125.xml
RDF/JSON: 1.0063125+rdf.json
Turtle: 1.0063125+rdf-turtle.txt
N-Triples: 1.0063125+rdf-ntriples.txt
Original Record: 1.0063125 +original-record.json
Full Text
1.0063125.txt
Citation
1.0063125.ris

Full Text

SUSTAINABLE ROAD SAFETY: DEVELOPMENT, TRANSFERENCE AND APPLICATION OF COMMUNITY-BASED MACRO-LEVEL COLLISION PREDICTION MODELS by Jianchen Sun B.Sc. in Transportation Engineering, Tongji University, P.R. China, 1994 M.Sc. in Computer Science, Wayne State University, Detroit, USA, 2002 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF APPLIED SCIENCE in The College Of Graduate Studies (Civil Engineering) THE UNIVERSITY OF BRITISH COLUMBIA (OKANAGAN) APRIL 2009  © Jianchen Sun, 2009  ABSTRACT The enormous social and economic burden imposed on society by road collision injuries is a major global problem. As a result, it is of ongoing interest of governments to discover ways of reducing this burden. The traditional engineering approach has been to address road safety in reaction to existing collision histories. While this approach has proven to be very successful, road safety authorities are also pursuing more proactive engineering approaches. Rather than working reactively to improve the safety of existing facilities, the proactive engineering approach focuses on improving the safety of planned facilities.  Proactive  programs rely heavily on reliable empirical techniques, including macro-level collision prediction models (CPMs). The three objectives of this research were to: 1. Develop community-based macro-level CPMs for the Capital Regional District (CRD) in BC, Canada and the City of Ottawa in Ontario, Canada. 2. Perform a road safety evaluation of the Canada Mortgage and Housing Corporation’s (CMHC) recently promoted Fused Grid model for sustainable subdivision development. 3. Use these models to conduct a Black Spot analysis of each region. Results were in line with intuitive expectations of each objective. First, following the recommended development and transferability guidelines, 64 community-based macro-level CPMs were successfully developed for the CRD and City of Ottawa. These models can be used by community planners and engineers as a decision-support tool in proactive road safety improvement programs. Second, the safety level of five road network patterns was evaluated using these macro-level CPMs. It was concluded that the 3-way offset and Fused Grid road networks were the safest over all, followed by the cul-de-sac and Dutch SRS road networks. The grid network was the least safe road pattern. Finally, black spot studies were also conducted, and four black spots were selected for in-depth analysis on diagnosing safety problems, and evaluating possible remedies. The results of this research demonstrate the potential of community-based, macro-level CPMs as new empirical tools for road safety planners and engineers to conduct proactive analyses, promote more sustainable development patterns, and reduce the road collision burden on communities worldwide. ii  TABLE OF CONTENTS ABSTRACT ............................................................................................................................ ii TABLE OF CONTENTS ...................................................................................................... iii LIST OF TABLES................................................................................................................ vii LIST OF FIGURES............................................................................................................. viii ACKNOWLEDGEMENTS .................................................................................................. ix CHAPTER I INTRODUCTION ......................................................................................... 1 1.1 Background ................................................................................................................... 1 1.2 Objectives of the Research........................................................................................... 4 1.3 Thesis Structure ............................................................................................................ 4 CHAPTER II LITERATURE REVIEW ............................................................................. 5 2.1 Introduction .................................................................................................................. 5 2.2 Reactive Road Safety Improvement Programs.......................................................... 5 2.3 Proactive Road Safety Improvement Programs ........................................................ 6 2.3.1 Road Safety Audits .................................................................................................. 7 2.3.2 Combining Micro-Level CPMs and Regional Transportation Planning Models .... 8 2.3.3 Dutch Sustainable Road Safety and Transportation Demand Management............ 8 2.3.4 Proactive Road Safety Planning Framework........................................................... 9 2.4 Development of Macro-Level Collision Prediction Models .................................... 10 2.5 Transferability of Macro-Level Collision Prediction Models................................. 17 2.6 Guidelines for Macro-Level Collision Prediction Model Use ................................. 19 2.6.1 Guidelines for selecting the appropriate models ................................................... 19 2.6.2 Guidelines for Macro-Reactive Use (Black Spot Programs)................................. 21 2.6.2.1 Black Spot Identification & Ranking.......................................................... 22 2.6.2.2 Diagnosis........................................................................................................ 24 2.6.2.3 Remedy .......................................................................................................... 25 2.6.3 Guidelines for Neighbourhood-Level Safety Planning ......................................... 25 2.6.3.1 Zone of Influence .......................................................................................... 25 2.6.3.2 Data Extraction............................................................................................. 25 2.6.3.3 Interpretation of Results .............................................................................. 26  iii  2.7 Case Studies................................................................................................................. 26 2.7.1 Black Spot Program Case Study............................................................................ 26 2.7.2 Neighbourhood Road Patterns Case Study............................................................ 27 2.7.3 Transferability Case Study .................................................................................... 29 2.8 Summary ..................................................................................................................... 29 CHAPTER III MODEL DEVELOPMENT....................................................................... 30 3.1 Introduction ................................................................................................................ 30 3.2 Data Extraction Methodology ................................................................................... 30 3.2.1 Geographic Scope.................................................................................................. 30 3.2.2 Aggregation ........................................................................................................... 32 3.2.2.1 Aggregation Unit........................................................................................... 32 3.2.2.2 Aggregation Approach ................................................................................. 33 3.2.3 Data Stratification.................................................................................................. 34 3.2.4 Data Sources .......................................................................................................... 37 3.2.4.1 Collision Variables........................................................................................ 37 3.2.4.2 Exposure Variables....................................................................................... 39 3.2.4.3 Network Variables ........................................................................................ 39 3.2.4.4 Socio-Demographic Variables ..................................................................... 39 3.2.4.5 Transportation Demand Management Variables (TDM)......................... 41 3.3 Model Development Methodology............................................................................. 42 3.3.1 Groupings .............................................................................................................. 42 3.3.2 Regression Technique............................................................................................ 43 3.3.3 Model Form ........................................................................................................... 43 3.3.4 Goodness of Fit...................................................................................................... 44 3.4 Results.......................................................................................................................... 46 3.4.1 The CRD Model Development Results ................................................................. 46 3.4.2 The City of Ottawa Model Development Results................................................. 51 3.5 Transference of Macro-level CPMs from the GVRD to the CRD ......................... 55 3.6 Summary ..................................................................................................................... 59 CHAPTER IV CASE STUDY: APPLICATION OF COMMUNITY-BASED MACROLEVEL CPMs FOR SAFETY EVALUATION OF ROAD PATTERNS ....................... 60  iv  4.1 Introduction ................................................................................................................ 60 4.2 Background ................................................................................................................. 60 4.3 Model Selection ........................................................................................................... 65 4.4 Approach ..................................................................................................................... 68 4.4.1 Modular Test Networks ......................................................................................... 68 4.4.2 Barrhaven Real World Networks........................................................................... 68 4.4.2.1 Tradition Grid Road Network..................................................................... 69 4.4.2.2 Fused Grid Road Network........................................................................... 70 4.4.2.3 Dutch SRS and 3-way Offset Road Networks............................................ 71 4.4.3 Selection of Trigger Variables............................................................................... 73 4.5 Results On Module Test Networks............................................................................ 74 4.5.1 Quadrant (16 Ha) ................................................................................................... 74 4.5.2 Neighbourhood (64 ha).......................................................................................... 76 4.5.3 District (256 ha)..................................................................................................... 79 4.6 Results on Barrhaven Test Site ................................................................................. 84 4.7 Accessibility................................................................................................................. 86 4.8 Discussion and Conclusions ....................................................................................... 86 4.8.1 Grid Network ......................................................................................................... 87 4.8.2 Cul-de-Sac ............................................................................................................. 87 4.8.3 Dutch SRS ............................................................................................................. 88 4.8.4 3-way Offset .......................................................................................................... 88 4.8.5 Fused Grid ............................................................................................................. 88 4.9 Summary ..................................................................................................................... 89 CHAPTER V CASE STUDY: MACRO-REACTIVE APPLICATION OF COMMUNITY-BASED MACRO-LEVEL CPMs FOR BLACK SPOT STUDY ......... 90 5.1 Introduction ................................................................................................................ 90 5.2 Approach ..................................................................................................................... 90 5.2.1 Identification & Ranking ....................................................................................... 90 5.2.2 Diagnosis & Remedy............................................................................................. 91 5.3 Results.......................................................................................................................... 91 5.3.1 Identification & Ranking ....................................................................................... 91  v  5.3.2 Diagnosis & Remedy............................................................................................. 94 5.4 Summary ..................................................................................................................... 99 CHAPTER VI CONCLUSIONS, CONTRIBUTIONS & FUTURE RESEARCH ..... 100 6.1 Introduction .............................................................................................................. 100 6.2 Summary & Conclusions ......................................................................................... 100 6.3 Research Contributions ........................................................................................... 101 6.3.1 Development of Community-based Macro-level Collision Prediction Models for use by the CRD and City of Ottawa planners and engineers to do empirical road safety planning ........................................................................................................................ 101 6.3.2 Safety level evaluation of CMHC’s Fused Grid road pattern and verification of how the level of road safety of the Fused Grid road network compares with four other known road networks.................................................................................................... 102 6.3.3 Black Spot Analyses for the CRD and City of Ottawa........................................ 103 6.4 Recommendations for Future Research ................................................................. 104 REFERENCES ................................................................................................................... 106 APPENDIX A – Model Development GLIM4 Output Sample ...................................... 110 APPENDIX B – GLIM4 Output Sample of Model Transference................................. 112 APPENDIX C – Sample Listing of CPZ Rankings ........................................................ 113 APPENDIX D – Sample Collision Frequency Calculation ............................................. 114 APPENDIX E – Sample Variables Value Distribution ................................................... 115  vi  LIST OF TABLES Table 2.1 Model Groups ......................................................................................................... 16 Table 2.2 Checklist for Selecting Appropriate CPMs* .......................................................... 20 Table 2.3 Candidate CPM Groups *....................................................................................... 21 Table 2.4 Relative Comparison to 3-way Offset Network ..................................................... 28 Table 3. 1 Candidate Variables – Definitions & Descriptive Statistics (CRD)...................... 35 Table 3. 2 Candidate Variables – Definitions & Descriptive Statistics (City of Ottawa) ...... 36 Table 3. 3 Model Groups ........................................................................................................ 42 Table 3. 4 Exposure CPMs – Total/Severe. (CRD, 2003)...................................................... 46 Table 3. 5 Socio-Demographic CPMs – Total/Severe. (CRD, 2003)..................................... 47 Table 3. 6 Transportation Demand Management CPMs – Total/Severe. (CRD, 2003)......... 48 Table 3. 7 Network CPMs – Total/Severe. (CRD, 2003)....................................................... 49 Table 3. 8 Exposure CPMs – Total/Severe. (City of Ottawa, 2006) ...................................... 51 Table 3. 9 Socio-Demographic CPMs – Total/Severe. (City of Ottawa, 2006) ..................... 52 Table 3. 10 Transportation Demand Management CPMs–Total/Severe(City of Ottawa, 2006) ................................................................................................................................................ 53 Table 3. 11 Network CPMs – Total/Severe. (City of Ottawa, 2006) ..................................... 54 Table 3. 12 Transferred Urban Measured CPMs (CRD,2003; GVRD,1996) ........................ 57 Table 4. 1 Selected CPMs for Safety Evaluation ................................................................... 66 Table 4.2 Analysis Area Data................................................................................................ 68 Table 4. 3 Summarized Trigger Variable Values (16 Ha)...................................................... 75 Table 4. 4 Comparison of Collision Densities (16 ha) ........................................................... 75 Table 4. 5 Summarized Trigger Variable Values (64 Ha)...................................................... 78 Table 4. 6 Comparison of Collision Densities (64 ha) .......................................................... 78 Table 4. 7 Summarized Trigger Variable Values (256 Ha).................................................... 82 Table 4. 8 Comparison of Collision Densities (256 ha) ......................................................... 83 Table 4. 9 Trigger Variable Values of Each Road Network Pattern ...................................... 84 Table 4. 10 Predicted Total Collisions for Barrhaven ............................................................ 84 Table 4. 11 Collision Ratios to the Fused Grid ...................................................................... 85 Table 5. 1 CPZ Identification for the CRD............................................................................. 92 Table 5. 2 CPZ Identification for the City of Ottawa ............................................................. 92 vii  LIST OF FIGURES Figure 2.1 Conventional (Micro-Reactive) & Macro-Reactive Black Spot Methods ............ 23 Figure 2.2 Neighbourhood Access Road Network Options ................................................... 28 Figure 3. 1 The Capital Regional District, BC, Canada ......................................................... 31 Figure 3. 2 The City of Ottawa, Ontario, Canada................................................................... 31 Figure 3. 3 The CRD Emme/2 Model TAZs .......................................................................... 32 Figure 3. 4 The City of Ottawa Emme/2 Model TAZs........................................................... 33 Figure 3. 5 The CRD Total Collision Densities ..................................................................... 38 Figure 3. 6 The City of Ottawa Total Collision Densities...................................................... 38 Figure 3. 7 The CRD Population Densities ............................................................................ 40 Figure 3. 8 The City of Ottawa Population Densities............................................................. 40 Figure 4. 1 Grid and Loops and Culs-de-Sac Road Patterns .................................................. 60 Figure 4. 2 Fused Grid Model................................................................................................. 62 Figure 4. 3 Fused Grid Layout Configurations....................................................................... 63 Figure 4.4 Test Site - Barrhaven............................................................................................. 64 Figure 4.5 Dutch SRS Road Network and 3-Way Offset Road Network .............................. 65 Figure 4. 6 Traditional Grid Road Network ........................................................................... 69 Figure 4. 7 Fused Grid Road Network.................................................................................... 71 Figure 4. 8 Dutch SRS Road Network ................................................................................... 72 Figure 4. 9 3-way Offset Road Network ................................................................................ 72 Figure 4. 10 16 Ha Modular Road Networks ........................................................................ 74 Figure 4. 11 64 Ha Modular Road Networks ........................................................................ 77 Figure 4. 12 256 ha Modules of Grid and Culs-de-Sac Road Networks ................................ 80 Figure 4. 13 256 ha Modules of Dutch SRS and 3-way Offset Road Networks .................... 81 Figure 4. 14 256 ha Modules of Fused Grid Road Network ................................................. 82 Figure 5. 1 Collision Prone Zones (CRD) .............................................................................. 93 Figure 5. 2 Collision Prone Zones (Ottawa)........................................................................... 93 Figure 5. 3 Urban CPZ2 (CRD)............................................................................................. 95 Figure 5. 4 Rural CPZ23 (CRD)............................................................................................. 96 Figure 5. 5 Urban CPZ105 (the City of Ottawa) .................................................................... 97 Figure 5. 6 Rural CPZ124 (the City of Ottawa) ..................................................................... 98  viii  ACKNOWLEDGEMENTS It is a pleasure to thank the people who made this thesis possible. First of all, I would like to express my deep and sincere gratitude to my supervisor, Dr. Gordon Lovegrove, for his enthusiasm, his inspiration, and his great efforts to explain things clearly and simply. He provided encouragement, sound advice, good teaching, and lots of good ideas throughout my thesis-writing period. I want to thank the Canada Mortgage and Housing Corporation (CMHC), the Transportation Department of the City of Ottawa, the Insurance Corporation of British Columbia (ICBC), and the Transportation Department of Capital Regional District (CRD), for providing me their data, without which this thesis would not have been completed. My thanks to my Lord for His wisdom and blessings in my life. May your name be honored and glorified. To my mother-in-law, I thank you for taking care of my daughter. To Faith, my lovely daughter, she brings me so much fun in my life. Last, I want to dedicate this work to my wife, Chan, for her love, support and encouragement.  ix  CHAPTER I INTRODUCTION 1.1 Background Throughout the world, roads are bustling with cars, buses, trucks, motorcycles, and other vehicles. These roads support economic and social development in many countries. But while roads facilitate many benefits, there are also costs accompanied: environmental contamination, urban stress and deteriorating air quality. Above all, roads have also been increasingly associated with a rise in road collision frequency, severity, and premature deaths. Losses are not limited to reduced worker productivity and trauma affecting a victim’s private life. Equally significant are the rising costs in health services and the added burden on public finances due to road collisions. In 2004, World Health Day, organized by the World Health Organization (WHO), was devoted to Road Safety for the first time. Every year, according to the WHO (2004), 1.2 million people are known to die in road collisions worldwide. Millions of others sustain injuries, with some suffering permanent disabilities. No country is spared this toll in lives and suffering, which strikes the young particularly. Enormous human potential is being destroyed, with also grave social and economic consequences. Road safety is thus a major public health issue throughout the world. In Canada, the social and economic burden of road collisions is also enormous. From 2000 through 2004, 32% of injury deaths were the result of motor vehicle collisions (MVCs). In the 15 to 24 age group, MVC deaths accounted for 70% of all injury deaths (Ramage-Morin, 2008). In 2006, there were 2,889 deaths and 199,337 injuries in Canada due to road collisions, or 9.1 deaths and 630 hospitalizations per 100,000 population (Transport Canada, 2007).  The economic cost of road collisions to Canadians is estimated at $25 billion  annually when health care costs, property losses and other factors are considered. This estimate represents about 2% of Canada's 2004 Gross Domestic Product (GDP). (Transport Canada & Health Canada, 2004).  1  Recognizing the traffic safety problem and the importance of reducing the frequency and severity of road collisions, the majority of road authorities have established Road Safety Improvement Programs (RSIPs). These programs have identified many factors that contribute to collisions, related to failures in one or a combination of the three road system components: the driver, the vehicle, and the road (Sayed, Abdelwahab, and Navin, 1995). Transportation engineering programs focus on improving the safety performance of the road component of the road system. Traditional RSIPs seek to address these problems by focusing on the identification, diagnosis, and remedy of collision-prone locations (CPLs) or black spots (i.e. black spot programs) often through the use of micro-level Collision Prediction Models (CPMs) (Sayed, 1998; Sawalha & Sayed, 1999). More specifically, traditional RSIP’s often analyze intersections and/or road segments individually by comparing the predicted mean collision frequency to the historically observed mean collision frequency and carrying this throughout a region. Although the use of micro-level CPMs and black spot programs has been proven to be successful, this type of program has traditionally been reactive in nature, such that a significant collision history must have existed prior to conducting any remedial action. Moreover, retrofitting countermeasures at identified black spots in existing communities is usually costly. To prevent black spots from occurring, and to reduce their associated social and economic burdens on society, road safety authorities and researchers are also pursuing more proactive engineering approaches. Rather than working to improve the safety of existing facilities, the proactive engineering approach to road safety improvement focuses on predicting and improving the safety of planned facilities. The goal of the proactive approach is to minimize the road safety risk by evaluating safety throughout each stage of the planning process, to preclude black spots from occurring. If road safety is explicitly addressed as one of the evaluation factors before a project is built, it reduces the number and cost of reactive safety countermeasures that have to be retrofitted into existing communities. However, one main obstacle associated with the delivery of a proactive road safety measure is the lack of the necessary methodology and reliable tools to evaluate road safety in a proactive manner. Community-based, macro-level CPMs may provide the basis to overcome this obstacle.  2  Development of community-based, macro-level CPMs involves a significantly different methodology from that used for micro-level CPMs. Instead of the single-facility, individual link, and node data commonly used as inputs for micro-level CPM development and predictions, community-based macro-level CPM input data are aggregated at the zonal level. The resulting macro-level CPMs relate zonal traffic collisions to zonal traffic intensity and other zonal variables, including: vehicle-kilometres-travelled (VKT), total road lane-kms (TLKM), household density (NHD), zonal area (AREA), posted speed (SPEED), average zonal congestion level (VC), intersection density (INTD), and total employed labour force (EMP) (Hadayeghi et al., 2003; Ladron de Guevara et al., 2004; Lovegrove and Sayed, 2006). Most recently, following a comprehensive literature review and data extraction process, Lovegrove (2007) successfully developed and presented forty-seven community-based, macro-level collision prediction models for the Greater Vancouver Regional District (GVRD) using 1996 data. Each CPM was significantly associated with one or more of twenty-nine variables identified using refined GLM methods. A black spot study was successfully conducted using the developed macro-level CPMs. Khondaker et al. (2009), in an attempt to validate the procedure put forward by Lovegrove, suggested that the models generated for the GVRD be transferred to a different time-space region. Using Lovegrove’s (2007) transferability guidelines of community-based, macro-level collision prediction models, Khondaker et al. successfully transferred sixteen (eight for urban and eight for rural) macrolevel CPMs from the GVRD to the City of Kelowna using 2001 data. Lovegrove (2007) developed guidelines for applying community-based, macro-level CPMs in neighbourhood safety planning. Using the developed macro-level CPMs for the GVRD, Lovegrove evaluated the safety level of four neighbourhood road patterns, including: grid network, culs-de-sac, Dutch SRS network and 3-way offset network. Most recently, the Canada Mortgage and Housing Corporation (CMHC) has begun promoting a new model for sustainable subdivision development – The Fused Grid. CMHC (2007) research to date suggested that this model had great potential to promote increasingly sustainable development patterns by combining several redeeming features from traditional grid and culde-sac road patterns. Specifically, it improves on land use and infrastructure efficiency, 3  reduces environmental impact, increases walkability and improves the neighbourhood milieu. The Fused Grid incorporates many elements of the traditional grid and cul-de-sac road patterns that exhibit greater safety, which intuitively should result in safer road conditions. However, this inference was insufficient and may be inaccurate with regarding to road safety, as CMHC has not had the empirical tools to do so. Therefore, it was necessary to perform a road safety evaluation of the Fused Grid model and assess the results using Lovegrove’s recently developed community-based, macro-level CPMs. 1.2 Objectives of the Research The main objectives of this research were then threefold: 1. To develop community-based macro-level collision prediction models for use by Capital Regional District and City of Ottawa planners and engineers to do empirical road safety planning; 2. To apply macro-level collision prediction models to evaluate the level of safety of CMHC’s Fused Grid road pattern and verify how the level of road safety of the Fused Grid road network compares with other road networks. 3. To apply macro-level collision prediction models in Black Spot analyses for the CRD and City of Ottawa. 1.3 Thesis Structure This thesis is divided into six chapters. Chapter one provides an introduction to the topic via background information, research problems and goals, and structure. Chapter two reviews the previous work on development, transferability and application of community-based macro-level CPMs. Chapter three describes the data and methodology used to develop the macro-level CPMs and the model development results for the CRD and City of Ottawa. Chapter four discusses the safety level evaluation of five road network patterns including grid network, Culs-de-sac, Dutch SRS, 3-way offset and fused grid. Chapter five discusses the application of macro-level CPMs to conduct Black Spot analyses for the CRD and the City of Ottawa. Chapter six provides a summary of results, associated conclusions, and significant contributions of this research, along with suggestions for future research. 4  CHAPTER II LITERATURE REVIEW 2.1 Introduction This chapter describes the previous research on development, transferability and application of community-based macro-level CPMs. Eight sections constitute this chapter. Section 2.2 describes the reactive road safety improvement programs. Section 2.3 describes the proactive road safety improvement programs. Section 2.4 reviews the previous works on development of macro-level CPMs. Section 2.5 describes the research on transferability of macro-level CPMs. Section 2.6 describes the macro-level CPM use guidelines developed by Lovegrove (2007). Section 2.7 reviews some case studies conducted by Lovegrove (2007). Section 2.8 summarizes this literature review. 2.2 Reactive Road Safety Improvement Programs Road collisions are a human tragedy that result in health, environmental and social problems, and have significant impacts on national economic growth strategies. In recognition of this, the road authorities and researchers have established many road safety improvement programs (RSIPs). Most RSIPs focus on engineering, enforcement, and/or education to reduce collisions and casualty rates per vehicle mile or kilometre. The most effective, longlasting road safety strategies tend to be those that focus on engineering safer road environments that lead to a reduction in exposure and risk of collision for drivers and their passengers (de Leur & Sayed, 2003). Traditionally, the engineering of safer roads has taken a reactive approach, including the identification, diagnosis and remedy (“improvement”) of hazardous or collision prone locations (called “black spots”), implementing remedial treatments after safety problems were identified. A black spot is defined as any location that exhibits a collision potential that is significantly high when compared with some normal collision potential derived from a group of similar locations (Sayed, 1998). The collision potential of a location is commonly described by several measures, including: collision frequency, collision rate, collision severity, or a combination thereof. To ensure that resources are spent only on the locations with the highest potential for safety improvements,  5  it is vital that a sound procedure be used to screen the road network in order to properly identify and rank black spots for diagnosis and treatment. The reactive engineering approach uses techniques such as micro-level (i.e. single location) collision prediction models (CPMs) to identify black spots of the road network so that appropriate remedial measures can be undertaken to reduce the likelihood and severity of collisions at those locations (Sayed, 1998; Sawalha & Sayed, 1999). While Black Spot Programs are vital and have proven to be very successful, this reactive program approach requires that a significant collision history exist to identify black spots before any action is taken to address the previously undiscovered road safety problems. Moreover, retrofitting countermeasures at identified black spots in existing (i.e. built) communities is usually costly. To prevent black spots from occurring, and to reduce their associated social and economic burdens on society, road safety authorities and researchers are also pursuing more proactive engineering approaches. 2.3 Proactive Road Safety Improvement Programs Rather than working to improve the safety of existing facilities, the proactive engineering approach to road safety improvement focuses on predicting and improving the safety of planned facilities (de Leur & Sayed, 2003). The goal of the proactive approach is to minimize the road safety risk by evaluating safety throughout each stage of the planning process, to preclude black spots from occurring at all. If road safety is explicitly addressed as one of the evaluation factors before a project is built, it reduces the number and cost of reactive safety countermeasures that have to be retrofitted into existing communities. Lowercost RSIP strategies through proactive intervention may in the long term be a more effective and sustainable road safety engineering approach than reactive strategies. Enhanced effectiveness and sustainability can occur when road safety is explicitly evaluated throughout road planning and design, before the driver is exposed to it (Lovegrove, 2007). However, the proactive approach can only be effective if supported by reliable empirical tools. While there exist sufficient reliable empirical tools for traditional reactive road safety engineering approach (i.e. Micro-level CPMs), proactive tools are at a relatively early stage of their 6  development and implementation (de Leur & Sayed, 2003; Lovegrove, 2007). But planners and engineers need the proactive tools to estimate the level of safety of planned projects, of design changes to those projects, and of other proposed safety improvements. The state of development of the proactive empirical tools, including road safety audits, combining microlevel CPMs and regional transportation planning models, Dutch Sustainable Road Safety programs, and Road Safety Risk Indices, have been reviewed in the following sections. 2.3.1 Road Safety Audits A road safety audit (RSA) is a formal and independent safety performance review of a road transportation project by an experienced team of safety specialists, addressing the safety of all road users (Wilson, 2004). Road safety audits can be performed during any or all stages of a project, including planning, preliminary design, detailed design, traffic control planning, construction, pre-opening, and on existing roads. RSAs can also be used on any sized project from minor intersection and roadway retrofits to mega-projects. The RSA is not a means to rank or rate a project, nor is it a check of compliance with standards. Audits conducted early in the life of a project—in the planning or initial design stages—have been shown to be the most beneficial and the easiest to integrate with an agency’s existing safety program (Wilson, 2004). RSAs are a proven road safety engineering tool to provide an explicit, formalized safety evaluation of road projects. Typical costs of audits were estimated to range from $1,000 to $8,000 U.S. dollars, depending on the size of the project. Austroads described an analysis of nine audit sites reporting 250 different design stage audit findings that resulted in benefit–cost ratios ranging from 3:1 to 242:1. As for audits of existing roads, benefit–cost ratios ranged from 2:1 to 84:1 (Jordan, 2001). Until recently, forecasts on collision potential and safety improvements relied heavily on experience and professional judgment. Microlevel CPMs that relate collision frequency-exposure-geometric traits had provided an additional decision aid. However, due to empirical limitations, micro-level CPMs can only be used to aid RSAs involving single intersections or road segments, where exposure is known or can be relatively accurately estimated.  7  2.3.2 Combining Micro-Level CPMs and Regional Transportation Planning Models Two studies (Ho & Guarnaschelli, 1998; Lord & Persaud, 2004) tried to use micro-level CPMs and Emme/2 regional transportation planning models to pursue improved empirical tools to do proactive, planning level analysis. Both used forecast exposure data generated by Emme/2 software, one of the most widely used transportation planning model packages in North America (INRO, 2003). The first study attempted to conduct a safety planning evaluation using the Greater Vancouver Regional District (GVRD) Emme/2 regional transportation planning model and micro-level CPMs. Facility types included signalized and unsignalized intersections, and, two-lane, multi-lane, and freeway road segments. In a second similar study, Lord & Persaud (2004) attempted to integrate refined micro-level CPMs with an Emme/2 transportation planning model of the Toronto area. The refined CPMs included those for different road and intersection configurations (e.g. 4 lane arterials, 3-way intersections), and collision categories (e.g. PDO, injury, and fatal). However, these two studies revealed that micro-level CPMs cannot fill the gap between what is needed and what is available in terms of reliable safety planning tools for proactive road safety improvement programs. Micro-level CPM limitations relate to their prediction of the level of safety at a single location (e.g. intersection or road segment), where traffic volume levels are known or can be estimated via precise short-term projections. However, traffic forecasts at any one location derived from long-term planning-level analyses are known to be inaccurate (+/- 30%) due to their regional screenline level of calibration, and their longer-term timeframe (i.e. often twenty years into the future) (Lovegrove, 2007). 2.3.3 Dutch Sustainable Road Safety and Transportation Demand Management The Netherlands has one of the best road safety records in the world due to the lowest number of fatalities per capita (Peden et al., 2004). The Dutch sustainable road safety (SRS) was vision launched in the early 1990s. The idea was to make the Dutch road traffic system inherently safe (i.e. self-enforcing). SRS has been the guiding strategy in their successful approach to improve road safety for the last two decades (Wegman et al., 2008). The SRS program works to ensure that all parts of a community’s land use and transportation system promote safety in the long term, by planning an integrated, self-reinforcing, inherently safe, 8  community-based traffic system (Wegman, 1996; van Schagen & Janssen, 2000). The Dutch SRS program relies significantly on Transportation Demand Management (TDM). TDM strategies are in effect meant to manage demand for travel in a way that utilizes the transportation system more efficiently. It represents a demand-side strategy response to growth, as opposed to the more traditional supply-side policy approach of building more roads (Khisty & Lall, 1998; Richardson, 1999). TDM is considered an effective SRS strategy because it inherently aims to reduce traffic volumes, and in turn, collisions. Unfortunately, the lack of reliable empirical planning tools has been acknowledged for some time as a significant limiting factor in the Dutch SRS program. Moreover, all SRS program forecasts have been based on a linear exposure-collision relationship, which was an erroneous assumption. Several Dutch studies have been conducted in search of improved empirical tools (Poppe, 1995, 1997a, 1997b; Van Minnen, 1999). Although some proxy indicators, such as the number of inhabitants, the number of jobs, the surface area and neighbourhood core size, were recommended to infer the level of road safety in a Traffic Analysis Zone (TAZ), none of these studies use them to develop macro-level CPMs. 2.3.4 Proactive Road Safety Planning Framework Road safety researchers in North America have been researching proactive RSIPs, including: Road Safety Planning Frameworks (RSPF), Safety Conscious Planning (SCP), and Road Safety Risk Index (RSRI) guidelines (de Leur & Sayed, 2002, 2003; Herbel, 2004). A RSPF was recommended by de Leur& Sayed (2003). The goal of the RSPF was to explicitly incorporate road safety as early as possible into every stage of road planning and design, including road safety audits, road safety multiple account evaluations, and road safety planning guidelines. To operationalize the framework, road safety assessment guidelines was recommended to be used to quantify an RSRI for each facility (de Leur & Sayed, 2003). The RSRI combines qualitative and quantitative safety measures in an analytically hierarchical approach. Although the RSRI provided an improved empirical safety-planning tool, it was acknowledged that more improvement was needed (de Leur & Sayed, 2002).  9  Recognizing the limitations of RSAs, micro-level CPMs, Dutch SRS program initiatives, and RSRI guidelines to do proactive road safety evaluations, road safety researchers developed community-based macro-level CPMs for providing improved tools to evaluate safety during the planning process ( Hadayeghi et al., 2003; Lovegrove, 2007 ). 2.4 Development of Macro-Level Collision Prediction Models Community-based, macro-level CPMs use area-wide, neighbourhood and traffic zone level traits that can be statistically associated with zonal collision frequency. These neighbourhood-wide areas are typically comprised of multiple intersections and road segments (not just a single intersection or road segment like micro-level CPMs). Macro-level CPMs are different from, and gain advantage in road safety planning applications, over micro-level CPMs in that accurate traffic forecasts for all roads within the study area are NOT necessary. Several attempts have been made to develop macro-level CPMs. Levine et al (1995) developed a spatial lag model for the City of Honolulu, which examines the zonal relationships of motor vehicle collisions to population, employment and road characteristics. However, the resulting model was based on linear regression, which was proven to be inappropriate to represent the non-negative, non-normal, non-linear nature of the collision exposure-traits relationship. Lord (2004) attempted to develop a tool that would allow the estimation of collisions on digital or coded urban transportation networks during the planning process. A series of predictive models were created for predicting the number of collisions at nodes and on links of two sample digital networks created by Emme/2 data from Toronto, Ontario for 1990–1995. The results showed that it was possible to predict collisions on digital transportation networks, but confirmed the reality that the accuracy of the predictions was directly related to the precision of the traffic flow estimates. The collision predictions are also sensitive to how the digital network is coded. Fotheringham (2000) used a Geographically Weighted Regression (GWR) model with some explanatory variables such as population and employment, to analyze the spatial variability of collisions between zones,  10  with inconclusive results. Kim & Yamashita (2002) tried to relate police collision data with land use categories using geo-statistical techniques, again without success. Recently, three studies attempted to develop macro-level CPMs using non-linear, nonNormal Generalized Linear Regression Modelling (GLM) techniques, and results appear promising. Hadayeghi et al. (2003) developed a series of macro-level collision prediction models that estimated the number of collisions in planning zones using data aggregated across 463 traffic zones in Toronto, Canada. A generalized linear modeling (GLIM) approach was employed in which separate total and severe (fatal and non-fatal injury) collision models were generated as a function of socio-demographic, traffic demand, and network data variables using Negative Binomial regression. Instead of using data from an individual link or node, macrolevel CPMs were built using data summed across all nodes and/or links in each zone, across an entire community or region. Zonal sums of vehicle-kilometres-travelled (VKT), and zonal averages of congestion (VC) were extracted from Emme/2 output. Average zonal congestion was derived for each zone by averaging the volume/capacity ratios across all modeled links in the zone. Freeway data was not considered a good predictor of zonal collisions and excluded because freeways are typically limited access facilities. The city of Toronto provided 1996 geo-coded collision data, broken down by severity and time of day. Population data was aggregated zonally from a 1996 Toronto regional survey. After confirming data availability, final choice of variables and model development followed the usual stepwise forward procedure. Explanatory variables included vehicle-kilometrestravelled, arterial road lane-kms, number of households, area, posted speed, average zonal congestion, intersection density, total employed labour force, and total minor road kilometres. Several other possible explanatory variables were explored, including different employment sectors, certain land uses, neighbourhood geometry, driver age, gender, road conditions, collision reporting practices, and police enforcement levels. The resultant macro-level CPMs predicted either total or severe collisions, for either all day or rush hour time periods, based on the following model form: bx E(Λ) = aoVKTbo e∑ i i  (2.1) 11  Where:  E ( Λ ) = Mean Collision Frequency; VKT = Zonal total of VKT from EMME/2; xi=Explanatory variables aggregated zonally (e.g. employment, population, intersections); a0, b0, bi = Model parameters; Both Poisson and negative binomial (NB) error distributions were tested in the GLM regression process and the NB distribution was found to best represent the data. The significance of each variable was gauged by looking at several criteria, including: intuitively logical signs, Pearson χ2, t-statistics at 95% confidence level, and contribution to overall model fit. The overall model goodness of fit was judged via the shape parameter κ, Pearsonχ2, Pearson R-Square, and Rk2 statistics. The results revealed that: •  Emme/2 vehicle-kilometres-travelled (VKT) output was a reasonable proxy of actual zonal traffic patterns.  •  Increasing collisions were associated with zones of increasing VKT, households, major road kilometres, and intersection density.  •  Decreasing collisions were associated with zones of increasing average posted speed, not an intuitive result. A possible explanation was offered that higher posted speeds occur where the road environment was engineered to be ‘safer’.  •  Decreasing collisions were associated with zones of increasing average zonal congestion, not an intuitive result. A possible explanation was offered that increasing congestion produces lower operating speeds, resulting in fewer collisions.  •  Increasing morning peak hour collisions tended to be associated with zones of increasing total employed labour force and minor road kilometres.  •  Morning peak period CPMs had generally better fits, suggesting that they explained more of the data. A possible explanation was offered that the dominating model influence of the exposure variable came from an Emme/2 morning peak hour model.  •  The Severe Collision model goodness of fit was no better than that for Total Collisions.  12  Ladron de Guevara et al. (2004) created planning-level collision prediction models for Tucson, Arizona. These macro-level CPMs were developed assuming a non-linear, exponential function, with NB error distribution, using simultaneous equation and log-linear transformation techniques. Data was separated into fatal, injury, and property-damage collisions. Population density was treated as a proxy for exposure instead of using a leading exposure variable (e.g. VKT, AADT) in the model form. Although using population density as a proxy for exposure has simple and practical methodological merit for model development and data collection, it fails to sustain the empirical property of recommended CPM form with a leading exposure variable. That is, when there is no exposure, there can be no collisions. To improve population density fit, the number of minors (age ≤ 17) was separated out from total population, to remove non-drivers typically unrelated to collisions. The macro-level collision prediction models were developed based on the following mathematical form: bx E (Λ ) = e ∑ i i  (2.2)  Where: E(Λ) = Dependent variable in collisions per two years; bi= Model parameters; and, xi= Independent variables. Numerous candidate models were tested, with final model choice based on several criteria: the minimum value of Akaike’s Information Criteria (AIC), a criteria related to model loglikelihood at convergence and to the number of model variables; maximum Rp2 statistic, which is based on standardized residuals; and, minimum G2 statistic, which is based on individual deviances. Screening criteria used to verify variable significance included: the parameter’s estimated coefficient p-value < 0.05 (i.e. a 95% confidence level); and, parameter’s sign and magnitude agreement with theoretical expectations. The meaningful variables for the fatal collision model included population density, persons 17 years old or younger as a percentage of the total population, and intersection density. Significant variables for the injury and property-damage collision models were population density, number of employees, intersections density, percentage of miles of principal arterial, percentage of miles of minor arterials, and percentage of miles of urban collectors. Among 13  several conclusions it was suggested that planning-level safety models were feasible and may play a role in future planning activities. Hadayeghi et al. (2007) updated and improved the previous research (Hadayeghi et, 2003) using data comprising collision, socioeconomic, demographic, and road network characteristics, as well as road traffic volumes for the city of Toronto’s 481 traffic zones. The effects of specific land uses and different types of employment and the presence of transit facilities were new variables that were considered in this study. The analysis tool used to estimate model coefficients was still the generalized linear modeling (GLM) procedure with the assumption of a negative binomial error distribution. The model form used was the same as Equation (2.1). Three statistical measures—the Pearson chi-square, scaled deviance, and the overdispersion parameter (α) were used jointly to assess the fit of each model developed. Also, the quality of fit was investigated with the cumulative residuals (CURE) method. The CURE method has the advantage of not being dependent on the number of observations as do many other traditional statistical procedures. The decision to keep a variable in the model was based primarily on two criteria. The first criterion was that the chi-square of the variable’s estimated coefficient was significant at the 95% level. The second criterion was improvement of the goodness of fit of the model that included that variable. If a variable satisfied the two foregoing criteria but had a clearly counterintuitive sign (e.g., flow variable with a negative sign), it was omitted. Twenty-three regression models were developed to examine the relationships between several types of transportation planning variables and collision frequency. The models reveal that collision frequency increases with the increase in the following explanatory variables: total arterial road kilometres, total collector kilometres, total laneway kilometres, total ramp kilometres, total road kilometres, number of signalized intersections, number of four-leg and three-leg signalized intersections, and number of schools, number of dwelling units in each zone. The total area of open space and parks was found to have a negative but significant association with collision data. Other traffic variables such as average zonal posted speed, average zonal 85% operational speed, and average zonal volume over capacity (V/C) were found to be insignificant contributors in the occurrence of total and severe collisions.  14  Following a comprehensive literature review and data extraction process, Lovegrove (2007) developed and presented forty-seven macro-level collision prediction models, each significantly associated with one or more of twenty-nine variables identified using refined GLM methods. GLM has the advantage of overcoming the limitations associated with the use of conventional linear regression in modeling road collisions. The error structure of the models was assumed to follow the negative binomial distribution. The general model form used is shown in equation 2.3. Only exposure variables meeting the ‘zero-exposure = zero collision frequency’ principle have been used as an external exposure (i.e. Z) parameter.  E (Λ ) = ao Z a1 e ∑  bi X i  (2.3)  Where : E(Λ) = predicted collision frequency (over 3 years); ao, a1, bi = model parameters; Z = external exposure variable (e.g. VKT, TLKM); and, Xi = explanatory variables. The Pearsonχ2 statistic and the Scaled Deviance statistical measures were used to assess the model goodness-of-fit. The procedure used for selecting the model variables was a forward stepwise procedure by which the variables were added to a model one by one. The decision to retain a variable in the model was based on four criteria. First, the logic (i.e. +/-) of the estimated parameter had to be associated intuitively with collisions. Second, the parameter estimate t-statistic had to be significant at the 95 percent confidence level (i.e. > 1.96). Third, the addition of the variable to the model should have caused a significant drop in the Scaled Deviance at the 95 percent confidence level (i.e. > 3.84). Fourth, the variable had to show little or no correlation with any of the other independent variables. The models met all goodness of fit measures at the 95% confidence level, suggesting that they are reasonable for use in terms of predictive power, and were stratified into the sixteen groups shown in Table 2.1.  15  Themes Exposure  Table 2.1 Model Groups Variables Land Exposure Variable Type Use VC = Avg congestion level VKT = Veh icle/km-travelled TLKM = Tt l lane-kilo metres AREA = Neighbourhood area  Socio-Demo graphic (S-D)  POPD = Population density WKGD = Nu mber of jobs NHD = Ho me density FS = Average family size UNEM P = Unemp loy ment  Transportation Demand Management (TDM )  CORE = Nbhd core size TCM = Tt l Co mmuters SCC = Short-cut Capacity DRIVE = Drive to work  Network (SIGD, INT, I3WP, IA LP, ALKP)  SIGD = Signal density INTD = Intersection density I3WP = 3-way intersections IALP = Local-Arterial Int’ns  Urban Rural Urban Rural  Urban Rural Urban Rural  Group #  Modelled Measured Modelled Measured  1 2 3 4  Modelled Measured Modelled Measured  5 6 7 8  Modelled Measured Modelled Measured  9 10 11 12  Modelled Measured Modelled Measured  13 14 15 16  Each model predicted the level of safety in either urban or rural areas. Fifteen of the CPMs could be used relying on only measured exposure data (e.g. traffic data that can be measured off maps, such as TLKM), for the benefit of practitioners without access to modeled exposure data (i.e. VKT data derived from traditional 4-step transportation planning model software resources, such as Emme/2). Each CPM predicted the three-year total of collisions for one of several collision types - total, severe, AM and/or PM peak periods, pedestrian, or bicycle – using one or more explanatory variables related to one of four neighbourhood trait themes, including: traffic exposure, road network, socio-demographic, and transportation demand management. The models revealed that increased collisions were associated with increases in the following explanatory variables: •  Exposure-related: vehicle kilometres travelled (VKT), total road lane kilometres (TLKM), and average zonal congestion (VC);  •  S-D-related: job density (WKGD), population density (POPD), unemployment (UNEMP), residential unit density (NHD);  •  TDM-related: shortcut capacity and attractiveness (SCC, SCVC), number of drivers (DRIVE), total commuters (TCM), total commuter density (TCD); and, 16  •  Network-related: signal density (SIGD), intersection density per unit area (INTD), intersection density per lane-km (INTKD), arterial-local intersections (IALP), total arterial road lane kilometres (ALKP).  Several models revealed that decreased collisions were associated with increases in the following explanatory variables: family size (FS), core size and percentage (CORE, CRP), 3way intersections (I3WP), and, local road lane-kilometres (LLKP). The results showed that it was possible to quantify, on a macro-level TAZ scale, a statistically predictive association between traffic safety and neighbourhood characteristics pertaining to traffic exposure, road network, socio-demographic, and transportation demand management. 2.5 Transferability of Macro-Level Collision Prediction Models Transfer previously developed CPMs for use in geo-demographic regions and/or time periods (i.e. space-time regions) other than the one for which they were developed is an efficient technique to deal with lack of data or poor data quality. Lovegrove (2007) developed a 4-step macro-level CPMs transferability guideline based on that described in Sawalha & Sayed (2005) for micro-level CPMs, with minor modifications. The first step was to ensure that data used in the transferred CPM’s calibration process generally conformed to the specifications of the original dataset used to develop the models, which required careful attention to variable definitions and units. Secondly, the transfer calibration needed to be based on a sufficient number of data points, depending on both data availability and data quality, expected roughly as 15 to 20 TAZs. Thirdly, the GLM software was re-run to obtain new estimates for each CPM’s lead coefficient a0 and shape parameter κ; all other parameters in the model were retained. The last step was to test the goodness of fit of the transferred models. While the SD, and Pearsonχ2 test statistics should not exceed the target χ2, a third test statistic, the z criterion, was also recommended. The model was considered to be successfully calibrated if it had the statistical property shown in Equation 2.4, in addition to meeting the SD and Pearsonχ2 statistical measures.  17  z =  χ P2 − E ( χ P2 ) < 1.00 σ ( χ P2 )  (2.4) N  E ( χ P2 ) = N ; σ ( χ P2 ) = 2 N (1 + 3 / κ ) + ∑ i =1  1 E (Λ i )[1 + E (Λ i ) / κ ]  [ yi − E (Λ i )]2 Pearsonχ = ∑ Var( yi ) i =1  (2.5)  n  2  (2.6)  Where, χ2p = Pearsonχ2 N = the number of data points used to calibrate the model. E(Λi ) = mean collision frequency predicted by transferred model for zone i κ = shape parameter for each calibrated CPM yi, Var(yi) are derived for each individual observation in the new data set. A case study was conducted by Lovegrove (2007) to test the transferability guideline. Using 2003 Kelowna data, Khondaker et al. (2009) attempted to develop community-based, macro-level collision prediction models for the City of Kelowna (COK) following the guidelines recommended by Lovegrove (2007). The results were unsuccessful due to poor data quality. Following this, Khondaker et al. attempted to transfer the Great Vancouver Regional District (GVRD) models generated by Lovegrove (2007) rather than build new models. Using the transferability guidelines described in Lovegrove (2007) for communitybased, macro-level collision prediction models, Khondaker (2008) successfully transferred sixteen (eight for urban, eight for rural) macro-level CPMs from the GVRD to the City of Kelowna. In accordance with the transferability guidelines (Lovegrove, 2007), the macrolevel CPMs for GVRD (1996) were calibrated with the COK 2003 data. To do this calibration, all GVRD model parameter estimates were retained excluding the lead coefficient and the model’s over-dispersion, or shape parameter. The GLM software was rerun using Kelowna data to obtain new estimates for each CPM’s lead coefficient and shape parameter. The Z score, as shown in equation 2.4, was used for statistical verification of each calibrated model’s goodness of fit. At 95% confidence level, a Z score close to zero indicates 18  that a transferred model meets statistical tests for goodness of fit to the data, in addition to meeting the SD and Pearson χ2 statistical measures. The results showed that all Z scores were near zero; therefore, the macro-level CPMs were deemed to have been successfully transferred. This study revealed that transferability of macro-level CPMs was an efficient and reliable method for developing community-based, macro-level CPMs where a lack of sufficient data existed. 2.6 Guidelines for Macro-Level Collision Prediction Model Use In the review of literature, only Lovegrove (2007) developed guidelines for using the macrolevel CPMs by planners and engineers. Three main guidelines were presented in Lovegrove’s research including guidelines for selecting the appropriate models, guidelines for macroreactive applications and guidelines for proactive applications. 2.6.1 Guidelines for selecting the appropriate models While numerous macro-level CPMs have been developed, it is important to use only the appropriate number of models needed for each safety application. Select only the most appropriate models while still providing reasonably accurate and reliable results would minimize the amount of time and the level of effort required to use the complex models by practitioner. A six-step selection checklist has been recommended by Lovegrove (2007), as shown in Table 2.2.  19  Table 2.2 Checklist for Selecting Appropriate CPMs*  *Guiding Principle: Select the minimum number of CPMs that will provide the desired level of accuracy and reliability in results. Table 2.3 contains a summary of recommended candidate model groups to be considered at each of the six steps in the model selection process. Highlighted in Table 2.3 (Lovegrove, 2007) are the steps taken for traffic calming an urban neighborhood example. 20  Table 2.3 Candidate CPM Groups *  2.6.2 Guidelines for Macro-Reactive Use (Black Spot Programs) A black spot is defined as a location with collision potential that is significantly higher than that of locations with similar attributes (Sawalha & Sayed, 2001). Black spot programs follow three steps. First, the locations of black spots are identified using an appropriate collision prediction model (CPM) and statistical techniques. Second, each black spot is investigated and possible causes for the problems at the location are identified. Third,  21  appropriate possible solutions or remedies are presented for each black spot to treat the problems. The guidelines of macro-level CPMs used in black spot programs outlined two major differences between micro-reactive (use micro-level CPMs) and macro-reactive (use macrolevel CPMs) methods. The first major difference is that collision data is not required for all intersections and/or road segments within a zone; a neighbourhood-wide aggregation of the collision data is sufficient. This means that the specific location of the collisions within a neighbourhood is not as important as which neighbourhood the collision occurred in. Second, it is not intersections and road-segments that are the analytical unit of interest for a macroreactive study, but rather a Traffic Analysis Zone (TAZ). A summary of the differences between conventional micro-reactive and macro-reactive black spot programs can be seen in Figure 2.1 (Lovegrove, 2007). 2.6.2.1 Black Spot Identification & Ranking Black spot identification takes a similar form for macro-reactive as it does for micro-reactive. However, there are some important differences. First, the collision history (i.e. y = count) for an area of interest is a zonal aggregation rather than an intersection or road-segment history. Second, E(Λ) and Var[E(Λ)] are also calculated over an entire TAZ. A zonal empirical Bayes (EB) safety estimate and its variance are then calculated for each selected CPM according to Equations 2.7, and 2.8 (Hauer, 1992). ⎡ E (Λ i ) ⎤ EB i = E ( Λ | Y = count ) = ⎢ ⎥ (κ + count ) ⎣ κ + E (Λ i ) ⎦ 2  ⎡ E (Λ i ) ⎤ Var ( EB i ) = Var ( Λ | Y = count ) = ⎢ ⎥ (κ + count ) ⎣ κ + E (Λ i ) ⎦  (2.7) (2.8)  22  Figure 2.1 Conventional (Micro-Reactive) & Macro-Reactive Black Spot Methods For each selected model, using Equation 2.9 (Sayed, 1998; Sawalha & Sayed 1999) to evaluate whether or not the zone is collision prone. A location is deemed to be a black spot if, with significant probability, δ (usually greater than 0.95), its safety estimate exceeds that of the comparison group.  23  ⎡ E ( Λ ) [κ / E (Λ ) + 1](κ + count ) λ(κ + count −1) e −[κ / E ( Λ ) +1]λ ⎤ f EB (λ )dλ = ⎢1 − ∫ dλ ⎥ ≥ δ κ count ( ) Γ + ⎥⎦ ⎢⎣ 0  E (Λ)  1−  ∫ 0  (2.9)  In the case of multiple models being used per TAZ, multiple EB estimates will be found for each TAZ, when most models identify a zone as collision prone, it should be considered as a collision prone zone (CPZ). The locations that are identified as black spots are then ranked according to two criteria. First, by the calculated Potential Collision Reduction (PCR), calculated simply by taking the difference between the observed and expected collision frequency as follows: PCR = EB − E (Λ)  (2.10)  Second, the Collision Risk Ratio (CRR) is used to augment the ranking to avoid excluding locations with low collision frequencies that have observed significant collision increases (Sayed & Rodriguez, 1999). The CRR is found by taking the ratio of observed to expected collision frequency as follows:  CRR =  EB E (Λ )  (2.11)  When using a macro-reactive procedure, since the E(Λ) and EB estimates often differ for each zone, a modification was made to reconcile the differences. This was done by summing the CRR and PCR across all models to derive a total ranking score for each zone. The zone(s) with the highest ranking score are recommended as those in most need of attention. 2.6.2.2 Diagnosis Diagnosis of CPZ using macro-level CPMs enhances that of micro-level CPMs. The enhancement occurs in the existence of a second indicator, beyond the over-representation of collision types. The second indicator is the existence of trigger variables for a CPZ. That is to say, the value of one or more of the variables used in macro-level CPMs is significantly different than the regional average value. Trigger variables together with additional information obtained from collision patterns and site visits, can be used to provide evidence identifying the overall road safety problem in the subject CPZ.  24  2.6.2.3 Remedy As it is not efficient to conduct a detailed micro-level analysis of each intersection and road segment in a collision prone neighbourhood or zone, a different approach was suggested for macro-reactive black spot programs. That was, strategic level of safety analysis on a zonewide basis was conducted using macro-level CPMs. The four main themes (i.e. exposure, SD, TDM and network) were used to categorize countermeasure strategies. To promote more sustainable road safety strategies, it was recommended that at least one possible remedy be generated considering each variable theme, and ideally, that the final remedial plan includes an integrated strategy that draws on each of these themes. Collision Modification Factors (CMFs) were recommended to calculate B/C ratios for the economic and ranking analyses of potential zone-wide remedies. 2.6.3 Guidelines for Neighbourhood-Level Safety Planning Safety evaluations typically deal with a relative small geographic scope in neighbourhoodlevel planning. Lovegrove (2007) discussed zones of influence, data extraction, and interpretation of results in the guidelines. 2.6.3.1 Zone of Influence Since a small geographic scope is dealt with in neighbourhood planning, the zones of influence will be either the zone in which the planning study is conducted or the group of zones contiguous to it. However, careful assessment of which zones change significantly and therefore are zones of influence should be verified in large municipal planning projects. 2.6.3.2 Data Extraction For neighbourhood-level planning, it was recommended that data extraction rely on directly available, less costly measured data, including available census survey, land use, TDM, and road network information. For short term planning (< 5 year horizon), population and employment growth are usually extrapolated using predominant historical trends and/or economic cycles. Measured network and exposure data are usually based on approved shortterm transportation plans (e.g. 3-year plans). For longer term planning, there are usually approved municipal growth plans or Official Community Plans that provide data on future  25  land uses, roads, population, and employment projections, broken down by neighbourhood, in five or ten year increments. 2.6.3.3 Interpretation of Results As individual neighbourhoods are usually physically isolated from, and therefore not significantly influenced by limited-access highways, their macro-level CPM predictions will represent a relatively complete evaluation of safety levels on their roads. As such, all comparisons can use the collision predictions from any CPM in whatever formulation (i.e. sums, differences, percentages) provides the desired clarity for practitioners and decisionmakers. Lovegrove (2007) also developed guidelines for Macro-level Collision Modification Factor Estimation, and for Regional-Level Safety Planning. Since these guidelines are not related to this research, they are not described in this paper. 2.7 Case Studies Following the proposed guidelines, Lovegrove (2007) conducted several case studies of macro-level CPM application. 2.7.1 Black Spot Program Case Study This case study was intended to conduct a black spot study to identify, diagnose, and recommend remedies for CPZs in the GVRD using data from 577 urban and rural neighbourhoods. Using the checklist in Table 2.2 and Table 2.3, all thirty-five CPMs were selected as candidate models. Following identification and ranking, two urban zones and two rural zones identified as CPZs were carried forward for diagnosis to identify the safety problems. The results suggested that macro-reactive use had the potential to complement traditional road safety improvement programs (RSIPs) by facilitating early identification of road safety problems without the need of collision history. If adopted for normal use by practitioners, macro-level CPMs appeared to be able to facilitate improved decisions by community planners and engineers, and ultimately, facilitate improved neighbourhood road safety for residents and other road users. 26  2.7.2 Neighbourhood Road Patterns Case Study This case study was conducted to verify earlier Dutch findings on the relative safety of neighbourhood road network patterns. Four road network patterns were analyzed including grid network, culs-de-sac, Dutch SRS network, and 3-way offset network, as shown in Figure 2.2. Following the recommended guidelines, urban measured and modeled CPMs for total collisions were selected. The CPMs from the network and TDM groups were used. First, a scan was conducted across all GVRD municipalities to identify existing sample neighbourhoods that resembled the four test networks under review. Second, variable values for each of these sample neighbourhoods were used as control values for non-trigger CPM variables in evaluating each test network. Non-trigger variables included: o Exposure - VKT, VC, TLKM; o S-D - POPD, FS, UNEMP, WKGD; and, o TDM - DRIVE, TCM, CORE, CRP, SIGD. o Network – ALKP, LLKP, SIGD Third, based on the usual GVRD one-to-two block-width-to-block-length ratio, four scaled versions of each test network were plotted. Fourth, values for each trigger variable were measured from the plotted test networks, to provide the remaining input data to run the network and TDM group CPMs. Trigger variables included shortcut capacity (SCC), intersection density (INTD), and 3-way intersection percentage (I3WP). The comparison results have been shown in Table 2.4.  27  Figure 2.2 Neighbourhood Access Road Network Options Table 2.4 Relative Comparison to 3-way Offset Network  The results showed that the 3-way Offset network was predicted to be safest overall. The modified Dutch SRS network was projected to be second safest followed by the original Dutch SRS network and the cul-de-sac network. The grid road network was projected to be least safe. 28  2.7.3 Transferability Case Study Following the transferability guidelines described in section 2.5, Lovegrove (2007) successfully transferred the macro-level CPMs developed for City of Vancouver to City of Richmond and City of Langley. The overall results from testing the proposed guidelines suggested that macro-level CPM transferability was feasible and no more complicated than when done for micro-level CPMs, subject to data availability. Lovegrove (2007) also conducted regional road safety planning case study, neighbourhood planning case study - core size and macro-level collision modification factors case study. Again, they are not relevant to this research. 2.8 Summary Traffic safety problems and the importance of reducing the frequency and severity of road collisions have been recognized by road safety authorities and researchers for a long time. Many Road Safety Improvement Programs (RSIPs) have been established. The traditional RSIPs using micro-level CPMs to identify, diagnose and remedy CPLs (black spots) have been reactive. This approach is very effective, but a significant collision history must have existed. Proactive RSIPs attempt to evaluate the road safety at planning process, in turn, to prevent black spots from occurring. Early proactive empirical tools, including road safety audits, combining micro-level CPMs and regional transportation planning models, Dutch Sustainable Road Safety programs, and Road Safety Risk Indices, had empirical limitations. Community-based, macro-level collision prediction models have been developed and used as an improved empirical tool to access the road safety in planning stage. Lovegrove (2007) has developed guidelines for macro-level CPM development, transferability and application. Several case studies were conducted using GVRD models to test the proposed model use guidelines. The results of the case studies showed that the proposed guidelines appeared sound as a means to be used by practitioners for road safety evaluation in black spot programs (macro-reactive) and in planning processes (proactive).  29  CHAPTER III MODEL DEVELOPMENT 3.1 Introduction This chapter consists of three main sections. In section 3.2, the methodology of data extraction is described, including geographic scope, aggregation approach, variable definitions, and sources. In section 3.3, the model development methodology is described, including information on model groupings, regression technique, model form, and goodness of fit. In section 3.4, resulting models are discussed, including model stratifications and statistical associations. The methodology for model development and use generally followed that set out by Lovegrove and Sayed (2006) and Lovegrove (2007). 3.2 Data Extraction Methodology As data quality is a cornerstone of well-fit and reliable statistical model, a comprehensive data extraction process was pursued in several steps to maximize the likelihood of statistical associations and the identification of underlying causal mechanisms. 3.2.1 Geographic Scope The data extracted for model development relates to two geographic areas as shown in Figure 3.1 and Figure 3.2. First, the Capital Regional District (CRD) (Southern Gulf Island and Salt Spring Island are excluded in this study) in the Province of British Columbia, Canada. The CRD is the regional government for the 13 municipalities and three electoral areas that are located on the southern tip of Vancouver Island. The CRD land area is roughly 1980 km2 and had a population of nearly 325,000 in 2003, dwelling in over 140,000 households (Census Canada 2001, 2006). Second, the City of Ottawa (COO) in the Province of Ontario, Canada. Ottawa is the capital city of Canada, with a metropolitan population estimated at over 812,000 (Census Canada, 2006). Ottawa is the fourth largest municipality in Canada with over 320,000 occupied dwellings built on 2778 km2 land area (Census Canada, 2006). Ottawa is situated on the south bank of the Ottawa River, and contains the Rideau River and the Rideau Canal.  30  Figure 3. 1 The Capital Regional District, BC, Canada  Figure 3. 2 The City of Ottawa, Ontario, Canada  31  3.2.2 Aggregation Aggregation of neighbourhood data was done to enable community-based, macro-level CPM development. 3.2.2.1 Aggregation Unit The aggregation unit was based on the 494 TAZs used in the CRD’s Emme/2 transportation planning model (CRD, 2003) and 400 TAZs used in the COO’s Emme/2 transportation planning model (COO, 2006). EMME/2 is a sophisticated software package which has powerful ability to automate the classic four-step transportation demand forecasting model. The TAZs boundaries were shown in Figures 3.3 (CRD) and 3.4 (COO). All data used for developing macro-level CPMs were aggregated to each TAZ.  Figure 3. 3 The CRD Emme/2 Model TAZs  32  Figure 3. 4 The City of Ottawa Emme/2 Model TAZs 3.2.2.2 Aggregation Approach When Census data (e.g. population, dwellings etc.) was aggregated to the TAZs, the TAZ boundaries did not always match the Census unit boundaries (i.e. Census Track, Dissemination Area). Here the values associated with portions of different Census units that lay wholly or partially within the TAZ were aggregated based on area and land uses. The assumption made was that values were uniformly distributed across a Census unit. The TAZ boundaries usually aligned with roads. When the aggregated data, such as collisions, roads and Emme/2 model links, overlap on the TAZ boundaries, their values (i.e. number of collisions, road lane kilometers and length of links) were just simply split and assigned to the connected TAZs. Freeway data was not considered a good predictor of zonal collisions and excluded because freeways are typically limited access facilities, with characteristics and traffic flows that relate more to the freeway segment itself than the surrounding neighbourhoods through which it runs (Hadayeghi et al., 2003).  33  3.2.3 Data Stratification To minimize aggregation bias, stratification was performed for both independent and dependent variables. For the dependent variable, three-year mean collision frequency (2002 – 2004 for CRD, 2005 – 2007 for COO), collision severity was stratified into total, severe (fatal and injury) and PDO. The three year period was chosen for two reasons. First, three years, versus just one year, of data tends to have a smoothing effect and facilitates the study of collision types for which data is sparse. Second, 2003 and 2006 was the most recent year for which detailed, calibrated Emme/2 data was available for the CRD and COO, respectively. In keeping with the recommended model development guidelines (Lovegrove, 2007), stratification of explanatory variables was done in three levels. The first level included the four themes of: exposure, S-D, TDM, and network variables. The second level of stratification used two types of exposure variables, measured (i.e. VKT) or modeled (i.e. TLKM). Use of measured variables allowed for practitioners that did not have the time or resources to access modeled exposure data. Modeled data consisted of traffic volume, speed, and congestion output from the Emme/2 transportation planning model. The third level of stratification was based on two factors: urban or rural. Table 3.1 and Table 3.2 list the candidate variables with possible data source(s), year(s), abbreviation, units, extraction method (e.g. measured/modelled), and descriptive statistics.  34  Table 3. 1 Candidate Variables – Definitions & Descriptive Statistics (CRD) Collisions Symbo Total Collisions over 3 years T3 Severe collisions (fatal & injury) S3 Property-Damage-Only collisions PDO Exposure Vehicle km's travelled VKT Average zonal congestion level VC Average zonal speed SPD Total lane km TLKM Zonal Area (Hectares) AR Socio-Demographics Urban zones URB Rural zones RUR Population POP Population Density (=POP/AR) POPD Employed residents(Over age 15) EMP Employed(=EMP/PO15)(%) EMPP Employee Density(=EMP/AR) EMPD Unemployed residents UNEM Unemployment rate (%) UNEM Average income $ INCA Average zonal family size FS Homes (#) NH Home Density (=NH/AR) NHD Zone jobs in tourism, retail,const, govt WKG Zonal job per capita(WKG/POP) WKGP Zonal job density (WKG/AR) WKG TDM Total commuters from each zone TCM Commuter density(=TCM/hectare) TCD Core area (hectare) CORE Core area %(=CORE/AR) CRP Shortcut capacity on local roads SCC Shortcut 'attractiveness' (=SCCхVC) SCVC No. of commuters by biking (%) BIKE No. of commuters by walking (%) WALK No. of commuters as car passengers PASS No. of commuters driving DRIV No. of commuters by transit (%) BUS Mode split - drivers DRP Network No. of signals SIG Signals density (/Ha) SIGD No. of intersections INT Intersection Density(INT/AR) INTD No. of intersections/TLKM INTK No. of 3-way intersections/INT I3WP No. of arterial-local intersections/INT IALP No. of Arterial lane-km ALKM No. of Collector lane-km CLKM No. of local lane-km LLKM No. of arterial lane-km/TLKM(%) ALKP No. of Collector lane-km CLKP No. of local lane-km/TLKM (%) LLKP  Method Measured Measured Measured  Source ICBC ICBC ICBC  Modelled Modelled Modelled Measured Measured  CRD CRD CRD CRD CRD  Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured  Years 02 - 04 02- 04 02 - 04  Total ZnAvg Min Max Std Dev 15808 32 0 288 38 6431 13 0 111 17 9377 19 0 177 22  2003 440692 901 2003 n/a 0.31 2003 n/a 39 2003 4095 8.3 2003 198003 400.8  1 6243 785 0 0.92 0.19 22 50 5 0 52.8 6.9 1.9 12666 5704  CRD CRD Census Census Census Census Census Census Census Census Census Census Census CRD CRD CRD  2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003  n/a n/a n/a n/a n/a n/a 1 2995 588 0 143 24 0 2608 503 0 78 18 0 89 20 0 119 22 0 21 3 0 93130 11075 0 4.3 0.74 0 1726 269 0 92 15 0 4887 477 0 3003 246 0 673 69  Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured  Census Census CRD CRD CRD CRD Census Census Census Census Census Census  2003 151476 307 2003 n/a 9 2008 194331 393 2008 n/a 93 2003 6028 12 2003 416 3.72 2003 n/a 4.3 2003 n/a 9.3 2003 n/a 5.6 2003 100742 204 2003 n/a 18.5 2003 n/a 63.3  Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured  CRD CRD CRD CRD CRD CRD CRD CRD CRD CRD CRD CRD CRD  2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003 2003  323 171 324944 n/a 167694 n/a n/a 9426 n/a n/a n/a 141691 n/a 129770 n/a n/a  n/a n/a 658 20 340 61 10 19 4.7 31412 2.2 287 27 263 0.4 23  315 0.64 n/a 0.06 5582 11 n/a 0.33 n/a 1.55 n/a 76% n/a 15.6% 436.4 0.88 760 1.54 2899 5.87 n/a 13.9 n/a 22.3 n/a 63.8  0 1688 280 0 63 12 1.9 12666 5705 23 100 15 0 180 24 0 34 12 0 21 3 0 49 11 0 15 3 0 1035 186 0 22 5 0 96 23 0 0 0 0 0 0 0 0 0 0 0 0 0  3.5 0.83 45 1.7 4 100 100 10.8 14.5 41.3 99 100 100  0.8 0.14 9 0.3 0.8 24 23 1.3 1.9 5.6 18 22 25  35  Table 3. 2 Candidate Variables – Definitions & Descriptive Statistics (City of Ottawa) Collisions Total Collisions over 3 years Severe collisions (fatal & injury) Property-Damage-Only collisions Exposure Vehicle km's travelled Average zonal congestion level Average zonal speed Total lane km Zonal Area (Hectares) Socio-Demographics Urban zones Rural zones Population Population Density (=POP/AR) Employed residents(Over age 15) Employed(=EMP/PO15)(%) Employee Density(=EMP/AR) Unemployed residents Unemployment rate (%) Average income $ Average zonal family size Homes (#) Home Density (=NH/AR) Zone jobs in tourism, retail, const, govt Zonal job per capita(WKG/POP) Zonal job density (WKG/AR) TDM Total commuters from each zone Commuter density(=TCM/hectare) Core area (hectare) Core area %(=CORE/AR) Shortcut capacity on local roads Shortcut 'attractiveness' (=SCCхVC) No. of commuters by biking (%) No. of commuters by walking (%) No. of commuters as car passengers(%) No. of commuters driving No. of commuters by transit (%) Mode split - drivers Network No. of signals Signals density (/Ha) No. of intersections Intersection Density(INT/AR) No. of intersections/TLKM No. of 3-way intersections/INT No. of arterial-local intersections/INT No. of Arterial lane-km No. of Collector lane-km No. of local lane-km No. of arterial lane-km/TLKM(%) No. of Collector lane-km No. of local lane-km/TLKM (%)  Symbol T3 S3 PDO  Method Source Years Total Zn Avg Min Max Std Dev Measured COO 05 - 07 40093 100 3 506 70 Measured COO 05- 07 8190 20 0 106 15 Measured COO 05 - 07 31903 80 3 400 56  VKT VC SPD TLKM AR  Modelled COO Modelled COO Modelled COO Measured COO Measured COO  2006 12611 3153 2006 n/a 0.53 2006 n/a 39 2006 12538 31.3 2006 28928 723.2  URB RUR POP POPD EMP EMPP EMPD UNEMP UNEMP INCA FS NH NHD WKG WKGP WKGD  Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured  COO COO Census Census Census Census Census Census Census Census Census Census Census COO COO COO  2006 2006 2006 2006 2006 2006 2006 2006 2006 2006 2006 2006 2006 2006 2006 2006  TCM TCD CORE CRP SCC SCVC BIKE WALK PASS DRIVE BUS DRP  Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured  Census Census COO COO COO COO Census Census Census Census Census Census  2006 39685 2006 n/a 2006 14274 2006 n/a 2006 2130 2006 1129 2006 n/a 2006 n/a 2006 n/a 2006 23664 2006 n/a 2006 n/a  SIG SIGD INT INTD INTKD I3WP IALP ALKM CLKM LLKM ALKP CLKP LLKP  Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured Measured  COO COO COO COO COO COO COO COO COO COO COO COO COO  275 125 81139 n/a 42893 n/a n/a 26237 n/a n/a n/a 32016 n/a 52163 n/a n/a  124 46728 4362 0.06 1 0.17 17 69 8 0 463 56 1.5 26602 2874  n/a n/a n/a n/a n/a n/a n/a n/a 2028 14 9605 1838 24 0 243 28 1072 8 4720 979 66 39 91 8 13 1 29 3 66 0 375 66 5.7 0 30 3 45399 19522 89726 9898 2.6 1 4 0.5 800 4 3346 696 11 0 157 17 1304 0 13376 1657 6.8 0 321 26 65 0 1899 227 992 12 357 78 5.3 2.82 2.2 10.1 7.1 592 20.7 58.5  2006 757 1.89 2006 n/a 0.07 2006 11547 29 2006 n/a 0.31 2006 n/a 0.93 2006 n/a 71% 2006 n/a 17.7% 2006 3112 7.78 2006 3209 8.02 2006 6218 15.5 2006 n/a 33.0 2006 n/a 14.3 2006 n/a 52.7  7 4345 910 0 139 15 .1.5 12067 1056 42 100 18 0 141 18 0 45 5 0 14 2.5 0 64 14 0 21 3 2 2954 598 0 49 10 4 91 18 0 0 0 0 0 0 0 0 0 0 0 0 0  8 1 183 2 1.78 100 100 158 237 144 100 92 100  2 0.2 27 0.2 0.3 28 21 17 27 19 24 15 23  36  3.2.4 Data Sources The data used for this study come from several public sources. 3.2.4.1 Collision Variables For CRD, collision claims data, spanning the period 2002 to 2004, was provided by the Insurance Corporation of BC (ICBC). ICBC handles most auto collision insurance claims in BC, and has taken its central claims database. ICBC had already geo-coded claim locations as either mid-block or intersection, and as being only on road centerlines. The location accuracy assumption, by default, was to accept the ICBC geo-coding assumption as reasonably accurate. De Leur & Sayed (2001) and Lovegrove (2007) have shown that claims data can be used in development of collision prediction models to evaluate road safety and can be applied in a manner similar to that of collision records without significant loss of accuracy. Figure 3.5 shows the spatial distribution of collision density (collisions/hectare) for CRD. The City of Ottawa provided actual collision report data of the complete public road system within the City of Ottawa. The collision data was collected from collision reports generated by Ottawa Police Service, the Ontario Provincial Police (OPP), and the Royal Canadian Mounted Police (RCMP). It should be noted that collision data collected refers to reportable collisions, where reportable collisions are those defined by the Ontario Highway Traffic Act as collisions having damage of $1000 or more or where injury results. Collisions not meeting these criteria are referred to as non-reportable. The non-reportable collisions were estimated to be 25% of the total collisions, and were not included in this study. The collision locations had been geo-coded as either mid-block or intersection, and were assumed to be reasonable accurate. Figure 3.6 shows the spatial distribution of collision density (collisions/hectare) for the City of Ottawa.  37  Figure 3. 5 The CRD Total Collision Densities  Figure 3. 6 The City of Ottawa Total Collision Densities  38  3.2.4.2 Exposure Variables The CRD transportation department provided Emme/2 model output files for modeled exposure variables for a 2003 PM peak hour. The City of Ottawa transportation department provided Emme/2 output files for a 2006 PM peak hour. Each Emme/2 model output files contained an Emme/2 modeled road network with data on travel forecasts for a typical afternoon rush hour across the CRD and COO respectively, including traffic volume, speed and travel time. Zonal vehicle-kilometres-travelled (VKT) was calculated by the equation 3.1. n  VKT j = ∑ Volumei * Lengthi  (3.1)  i =1  Where: VKTj = total vehicle-kilometres-travelled in TAZj; Volumei = traffic total volume assigned to link i; Lengthi = length of link i (km) in TAZj; n = number of links in the zone. Average zonal speed (SPD) is the average of speed on each link in the zone weighted by traffic volume. Average zonal congestion (VC) is the average of volume/capacity ratio of each link in the zone weighted by VKT. 3.2.4.3 Network Variables  Network variables consisted entirely of measured data, collected both digitally (using Geographic Information System techniques) and manually. Both CRD and COO transportation departments provided geo-referenced files of digital road network and TAZ boundaries. Automatic or manual aggregation was performed to extract data on intersection types, road geometry, and road-lane-kilometres. Local roads were assumed to be one lane in each direction. 3.2.4.4 Socio-Demographic Variables  All S-D variables were measured and derived from 2001 and 2006 Census Canada databases aggregated by Dissemination Area, extrapolated to year 2003 for CRD and 2006 for COO. Figure 3.7 and Figure 3.8 shows the regional distribution of population density (POPD) for the CRD and the COO respectively, with the highest densities shaded the darkest and  39  focused around regional activity centres, and the lowest densities shaded the lightest and located in the outlying rural areas.  Figure 3. 7 The CRD Population Densities  Figure 3. 8 The City of Ottawa Population Densities  40  3.2.4.5 Transportation Demand Management Variables (TDM)  Data for all TDM variables was measured and came from several sources. Census data was extracted on the number of commuters (TCM), broken down by choice of travel mode (i.e. DRIVE, PASSENGER, TRANSIT, BIKE, WALK). Neighbourhood core area (CORE), defined as the largest portion of the traffic zone area not bisected by major roads, was derived by visual examination of each zone of the neighbourhood maps. Shortcut capacity (SCC) and shortcut attractiveness (SCVC) variables followed from Lovegrove (2007) to provide a zonal descriptor of the neighbourhood’s local road access network. The mathematical definitions of these two variables are presented in Equations 3.2 and 3.3 below respectively, with little difference from Lovegrove’s definitions. Data values were assigned to each shortcutting variable by visual examination of neighbourhood street network, land use, and zone boundary maps. Since the traffic calming data for the City of Ottawa was not available, the roads in the City of Ottawa were assumed to be with no traffic calming facilities. The impact on results of this assumption need to be verified in future research.  SCC =  ( LNS + LEW ) ⋅ C f ⋅ CTC  Ar SCVC = SCC ⋅VC  (3.2) (3.3)  Where: SCC  = Shortcutting capacity  SCVC  = Shortcutting attractiveness  L NS , LEW  = Number of (north-south, east-west) local road lanes running completely across the zone = ∑ ( L NS + LEW )  Cf  = Typical local road capacity (assumed as 150 veh/lane/hr),  CTC  = Degree of zonal traffic calming (0 if traffic calmed; 1 if no calming; 0.5 if some traffic calming)  Ar  = Zonal area and  VC  = Average zonal congestion level  41  3.3 Model Development Methodology  The model development methodology essentially followed the model development guidelines presented by Lovegrove (2007). The method is summarized below. 3.3.1 Groupings Table 3.3 shows the sixteen model groups derived from the following explanatory data stratification levels:  •  Four themes of explanatory variables (Exposure, S-D, TDM and Network);  •  Two types of land use (rural and urban);  •  Two types of exposure data derivations (modelled or measured). Table 3. 3 Model Groups  Themes  Land Use  Derivation  Group #  Urban  Modeled  1  Measured  2  Modeled  3  Measured  4  Modeled  5  Measured  6  Modeled  7  Measured  8  Modeled  9  Measured  10  Modeled  11  Measured  12  Modeled  13  Measured  14  Modeled  15  Measured  16  Exposure Rural Urban Socio-Demographic (SD) Rural Urban Transportation Demand Management(TDM)  Rural Urban  Network Rural  42  3.3.2 Regression Technique Generalized Linear Modelling (GLM) regression method was used for developing macrolevel CPMs. The GLM regression method had the advantage of overcoming the limitations associated with the use of conventional linear regression in modeling traffic collisions (Hauer et al. 1988, Sawalha & Sayed, 2001). The standard generalized linear regression modeling software, GLIM4, from NAG (1994) was used, with three user-specified functions. First, a logarithmic link function was used for the linear transformation. Second, the maximum likelihood method (MLE) was chosen to provide parameter estimates. Third, as a check confirmed that the variance of the collision data exceeded its mean value suggesting overdispersion of data, the error structure of the models was assumed to follow the negative binomial distribution, which has been shown to model over-dispersed data well (Sawalha & Sayed, 2005). A sample of a GLIM4 output file is given in Appendix A. 3.3.3 Model Form The model form used is shown in Equation 3.4 as:  E ( Λ ) = a o Z a1 e ∑  bi x i  (3.4)  where:  E(Λ) = Predicted mean collision frequency; a0, a1, bi = GLM derived parameter estimates; Z = Exposure variable (VKT for modeled; TLKM for measured );  xi= Independent, explanatory variables (e.g. VC, POPD, INTD etc.). The log-linear transformation was carried out using a logarithmic linking function in the GLM software, transforming Equation 3.4 into: n  Ln[E (Λ)] = Ln(a 0 ) + a1 Ln( Z ) + ∑ (bi xi )  (3.5)  i =1  Using this regression method and model form, model development was then conducted to optimize goodness of fit.  43  3.3.4 Goodness of Fit The method used to evaluate variables and refine overall model goodness of fit is described by Lovegrove and Sayed (2006). Following those descriptions, GLIM4 software was programmed to provide statistics for overall model fit (i.e. Scaled Deviance, Pearsonχ2, κ), and for individual parameter fit (i.e. standard error, z-statistic, co-linearity, logic). The  Pearsonχ2 and Scaled Deviance (SD) statistics were defined as seen below.  [ y i − E (Λ i )]2 Pearsonχ = ∑ Var ( yi ) i =1 n  2  (3.6)  n ⎡ ⎛ y i + κ ⎞⎤ ⎛ yi ⎞ ⎟⎟⎥ ⎟⎟ − ( y i + κ ) ln⎜⎜ (3.7) SD = 2∑ ⎢ y i ln⎜⎜ i =1 ⎣ ⎝ E (Λ i ) + κ ⎠⎦ ⎝ E (Λ i ) ⎠ These statistics provide objective measures and are asymptotically χ2 distributed with n – p  degrees of freedom. For these, and all other descriptive statistics, 95% was the desired level of confidence used to assess goodness of fit. Model assessment was done in two stages: first, as candidate explanatory variables were selected to construct the model, and second, once explanatory variable selection was completed. In the first stage, explanatory variable selection followed a forward stepwise procedure. In this procedure, each model was constructed by adding one variable at a time, and testing the change in model fit due to the added variable. The first variable tested in each model was the exposure variable, due to its usually dominating prediction influence. Additional candidate variables were then systematically added to the model. The decision to retain a variable in the model was based on four criteria. First, the logic (i.e. +/-) of the estimated parameter had to be associated intuitively with collisions.  Second, the parameter estimate z-statistic (the z-statistic is  equivalent to the t-statistic for large sample sizes, >30) had to be significant at the 95 percent confidence level (i.e. > 1.96). Third, the addition of the variable to the model should have caused a significant drop in the Scaled Deviance at the 95 percent confidence level (i.e. > 3.84). Fourth, the variable had to show little or no correlation with any of the other independent variables. Correlation between variables was checked by viewing correlation results in the GLM software. Once explanatory variables were selected for a model, overall model goodness of fit was re-assessed using the SD, Pearsonχ2, and κ measures.  44  If the final stage of model goodness of fit assessment revealed statistics which did not meet expectations, model refinement was carried out to try and improve that fit. The data quality was the main reason of a poor fit for the resulting model. Therefore, the refinement method focused on an assessment of data quality, and in particular, whether there were any outliers in the data set. The Cook’s Distance (CD) technique was used for Outlier analysis, following the method described in Sawalha & Sayed (2005b). CD was defined in Equation 3.8.  ( )  ' hi ri PS p(1 − hi )  CDi =  2  (3.8)  Where: '  ri PS is the standardized residual of data point i, calculated as ∧  ri  PS '  =  y i − yi (1 − hi )Var( yi )  =  PRi 1 − hi  (3.9)  hi is the leverage value; PRi is the Pearson Residual; and, p is the number of parameters. High CD values indicate points that are very likely outliers. The analysis was done in stepwise progression, removing the points with largest CD values first. As each high CD data point was removed, the GLM software was then re-run while fixing the value of κ a t its previous value to test if the point was an outlier. This resulted in a new SD statistic value being calculated based upon the revised database (i.e. with outlier removed). If the new SD statistic value dropped significantly when compared to its original value, the removed data point was considered an outlier and removed. At the 95% confidence level, a drop in SD > 3.84 per data point deleted was considered to be significant. This outlier search procedure was repeated until the drop in SD < 3.84, indicating insignificance. Once outliers were removed from the dataset, GLM software was re-run setting κ = 0 to determine new estimates for each parameter and κ , and descriptive statistics. If the improvement in fit was enough to bring all quantitative assessment statistic values into line with targeted values (e.g. SD, Pearsonχ2< χ2, z-statistics > 1.96, κ > 1), the model was considered well-fit and ready for use in safety applications.  45  3.4 Results  Based on the data extraction and model development procedure, the resulting models for the CRD and City of Ottawa have been presented below. 3.4.1 The CRD Model Development Results The model development results for the CRD were listed in Table 3.4 to Table 3.7. Table 3. 4 Exposure CPMs – Total/Severe. (CRD, 2003) Model Group# 1  κ  DoF  2.9  Model Form  χ2  χ2  SD  320  326  350  363  Const: -3.60 vkt: 11.06 vc: 8.92  320  336  361  363  Const: -8.50 vkt: 11.43 vc: 10.77  164  185  193  195  Const: -6.37 vkt: 10.40 vc: 3.28  164  153  161  195  1.16  162  192  182  193  Const: 1.38 tlkm: 9.65  1.01  157  158  148  187  Cons: -5.37 tlkm: 8.69  0.05, dof  z-Statistics  Urban, Modeled, Exposure  Total Collisions/3yr =0.3355 VKT 0.5685  e 2.3364vc 2.78  Severe Collisions/3yr = 0.048VKT 0.6696  2  Urban, Measured, Exposure  3  Rural, Modeled, Exposure  e 3.1072vc Unsuccessful  2.13 Total Collisions/3yr =0.1078 VKT 0.6281 e  1.4975vc  2.99 Severe Collisions/3yr = 0.0077VKT 4  Pearson  0.8204  e  2.266vc  Const: -9.42 vkt: 9.90 vc: 4.56  Rural, Measured, Exposure  Total Collisions/3yr = 1.298TLKM  0.8084  Severe Collisions/3yr = 0.1864TLKM  1.107  46  Table 3. 5 Socio-Demographic CPMs – Total/Severe. (CRD, 2003) Model Group# 5  κ  Model Form  DoF  χ  2  SD  χ2  0.05,dof  z-Statistics  Urban, Modeled, Socio-Demographic  4.10 Total Collisions/3yr = 0.2021VKT 0.742 e  Severe Collisions/3yr = 0.0246VKT 0.8657 e  344  354  360  317  353  363  360  (2.12vc + 0.002261wkgad + 0.007791popd - 0.2871fs)  6  Urban, Measured, Socio-Demographic  7  Rural, Modeled, Socio-Demographic 2.59  Total Collisions/3yr =0.069 VKT 0.6562  317  (1.296vc + 0.002621wkgad + 0.005645popd - 0.254fs)  4.12  Severe Collisions/3yr = 0.0054VKT0.8381  163  190  Total Collisions/3yr = 0.2276TLKM 0.6447 e  Severe Collisions/3yr = 0.0062TLKM 0.7007 e  vkt:15.7 wkgad: 5.45 fs: -5.34  194  Const: -7.48 vc: 2.99  163  147  158  194  Const: -9.82 vc: 3.80  160  163  173  191  Const: -3.81 nhd: 7.24  tlkm: 7.99 empp: 4.22  152  193  Const: -5.18 nhd: 6.64  tlkm: 6.0 empp: 4.42  (0.5658nhd + 2.374empp)  1.11  Const: -10.11 vc: 8.04 popd: 6.01  vkt:15.46 wkgad: 6.65 fs: -5.07  193  e (1.848vc + 0.3084nhd)  Rural, Measured, Socio-Demographic 1.58  Const: -5.13 vc: 5.38 popd: 4.69  Unsuccessful  e (1.309vc + 0.2849nhd) 3.80  8  Pearson  162  152  (0.7043nhd + 5.7928empp)  vkt: 10.82 nhd: 4.36 vkt: 10.04 nhd: 4.17  47  Table 3. 6 Transportation Demand Management CPMs – Total/Severe. (CRD, 2003) Model Group# 9  κ  DoF  Pearson 2  χ  SD  χ2 0.05,dof  3.52  317  337  351  360  Const: -8.01 scvc:4.31  360  357  Const: -14.19 scvc:4.39  192  193  Const: -7.34 vc: 2.74  vkt: 10.78 tcd: 4.17  159  196  Const: -9.53 vc: 3.97  vkt: 9.57 tcd: 4.96  173  193  Const: 0.17 tcd: 5.08  tlkm: 9.34 core: -3.49  1.13 162 158 159 (0.5181tcd – 0.0008core) 0.8056  193  Const: -3.17 tcd: 5.43  tlkm: 7.68 core: -2.13  Model Form  z-Statistics  Urban, Modeled, TDM (0.0134scvc – 0.0106core)  Total Collisions/3yr = 0.1028VKT 0.915 e  3.63 Severe Collisions/3yr = 0.0074VKT 1.18 e 10  Urban, Measured, TDM  11  Rural, Modeled, TDM  318  (0.0139scvc – 0.0146core)  vkt:21.03 core: -7.61  Unsuccessful  2.66 Total Collisions/3yr =0.074 VKT 0.6482  162  191  e (1.201vc + 0.2306tcd) 3.12  Severe Collisions/3yr = 0.0055VKT0.8229 12 Rural, Measured, TDM  165  157  e (2.025vc + 0.3184tcd) 1.54  Total Collisions/3yr = 1.035TLKM 0.8358 e  Severe Collisions/3yr = 0.287TLKM  315  vkt:22.61 core: - 9.13  162  171  (0.3784tcd – 0.001core)  e  48  Table 3. 7 Network CPMs – Total/Severe. (CRD, 2003) Model Group# 13  κ  DoF  Pearson 2  χ  SD  χ2 0.05,dof  4.64  317  346  357  360  Model Form  Urban, Modeled, Network  Total Collisions/3yr = 0.1757VKT 0.8509 e  (0.4246sigd +0.829intd–0.00405i3wp-0.0086llkp)  4.68 Severe Collisions/3yr = 0.056VKT 0.7973 e  314  366  Const: -5.5 vkt:22.0 sigd; 1.96 intd: 6.59 i3wp: -2.65 llkp:-5.30  356  Const: -9.23 vc: 7.58 alkp: 3.93  192  Const: -8.41 vkt: 13.72 sigd: 4.27 intd: 3.42 alkp: 2.59 Const: -11.05 vkt: 12.9 sigd: 4.64 intd: 4.42 alkp: 2.57  vkt:14.93 i3wp: -8.66  Unsuccessful  Rural, Modeled, Network  3.20 Total Collisions/3yr =0.056 VKT 0.7169  161  177  192  e (38.18sigd + 2.73intd + 0.007alkp) 3.94  Severe Collisions/3yr = 0.0026VKT0.9916 16  341  (1.982vc – 0.0113i3wp + 0.0081alkp)  14 Urban, Measured, Network 15  z-Statistics  164  156  162  195  e (41.83sigd + 4.216intd + 0.0084alkp)  Rural, Measured, Network  1.69 Total Collisions/3yr = 0.972TLKM 0.9227 e  164  161  183  195  (76.95sigd + 0.0164ialp – 0.0097llkp +3.76intd)  1.27 Severe Collisions/3yr = 0.2687TLKM 0.9487 e  164  150  163  195  (87.33sigd + 0.0204ialp – 0.0104llkp+4.803intd)  Const: -0.13 sigd: 6.37 llkp: -3.05 Const: -4.04 sigd: 6.05 llkp: -2.33  tlkm: 9.76 ialp: 3.49 intd: 3.55 tlkm: 7.13 ialp: -5.84 intd: 3.48  49  Eight urban and sixteen rural community-based macro-level CPMs were successfully developed, including sixteen modeled and eight measured. All model statistics were within acceptable ranges (i.e. κ> 1, Pearson χ2 < χ2 and SD < χ2). The models revealed that  increased collisions were associated with increases in the following explanatory variables:  •  Exposure-related: vehicle kilometres travelled (VKT), total road lane kilometres (TLKM), and average zonal congestion (VC);  •  S-D-related: job density per unit area (WKGAD), population density (POPD), employment rate (EMPP), residential unit density (NHD);  •  TDM-related: shortcut attractiveness (SCVC), total commuter density (TCD);  •  Network-related: signal density (SIGD), intersection density per unit area (INTD), arterial-local intersections percentage (IALP), and arterial road lane kilometres percentage (ALKP).  Several models revealed that decreased collisions were associated with increases in the following explanatory variables: family size (FS), core size (CORE), 3-way intersections percentage (I3WP), and, local road lane-kilometres percentage (LLKP). These associations support earlier research findings (Lovegrove, 2006; Hadayeghi et al. 2003). Unfortunately, the urban, measured macro-level CPMs (Groups 2, 6, 10, and 14) were not developed successfully due to the low z-statistic of Total Lane Kilometres (TLKM) which was deemed as the dominant variable for measured CPMs.  50  3.4.2 The City of Ottawa Model Development Results The community-based macro-level CPMs developed for the City of Ottawa are presented below. Table 3. 8 Exposure CPMs – Total/Severe. (City of Ottawa, 2006) Model Group# 1  κ  DoF  5.45  Model Form  Pearson 2  χ2  χ  SD  z-Statistics  273  275  286  313  Const: 0.31 vkt: 15.71  273  260  293  313  Const: -6.33 vkt: 15.93  273  271  289  313  Const: 41.43 tlkm: 7.04  273  251  293  313  Const: 22.39 tlkm: 6.57  123  124  133  150  Const: -2.77 vkt: 15.67  123  133  141  150  Const: -7.19 vkt: 14.38  123  131  136  150  Const: 20.65 tlkm: 8.59  123  128  138  150  Const: 9.39 tlkm: 7.28  0.05,dof  Urban, Modeled, Exposure  Total Collisions/3yr = 1.0924VKT 0.5995 6.41 Severe Collisions/3yr = 0.13471VKT 0.6629 2  Urban, Measured, Exposure  3.42 Total Collisions/3yr =54.87 TLKM 0.2408 3.30 Severe Collisions/3yr = 10.903TLKM 0.2465 3  Rural, Modeled, Exposure  3.65 Total Collisions/3yr = 0.4017VKT 0.6439 3.67 Severe Collisions/3yr = 0.0623VKT0.6798 4  Rural, Measured, Exposure  1.81 Total Collisions/3yr = 23.081TLKM 0.3912 1.53 Severe Collisions/3yr = 5.1577TLKM 0.3762  51  Model Group# 5  Table 3. 9 Socio-Demographic CPMs – Total/Severe. (City of Ottawa, 2006) Pearson χ2 Model Form DoF SD z-Statistics 2 κ  χ  Urban, Modeled, Socio-Demographic 7.86  Total Collisions/3yr = 0.6044VKT 0.7043 e  Severe Collisions/3yr = 0.07824VKT 0.7611 e  285  309  270  276  291  309  ( 0.000239wkgad + 0.005223popd - 0.1643fs)  Urban, Measured, Socio-Demographic 3.62  272  Total Collisions/3yr = 43.9038TLKM 0.2761 e 3.39  0.007674nhd  Severe Collisions/3yr = 9.3933TLKM 0.2693 e 7 Rural, Modeled, Socio-Demographic 5.59  0.0053nhd  Total Collisions/3yr =0.1225 VKT 0.7569  284  ( 0.000213wkgad + 0.006264popd - 0.1804fs)  8.78  6  270  0.05,dof  272  122 122  Const: 35.94 nhd: 4.30  tlkm:8.10  260  292 312  Const: 18.86 nhd: 2.73  tlkm:7.05  119  134  149  Const: -6.84 nhd: 7.46  vkt: 20.68  140  146  149  Const: -11.22 nhd: 7.69  vkt: 18.99  151  133  149  Const: 12.08 nhd: 5.81  tlkm: 11.52  135  136  149  Const: 3.46 nhd: 5.24  tlkm: 9.47  0.2928nhd  Total Collisions/3yr = 9.7767TLKM 0.552 e  122  0.3373nhd  1.86 Severe Collisions/3yr = 2.1782TLKM 0.5323 e  122  vkt:18.50 popd: 5.26  288 312  Severe Collisions/3yr = 0.0161VKT0.8082 e 8 Rural, Measured, Socio-Demographic  2.24  Const: -6.76 wkgad: 2.00 fs: -2.92  287  e 0.259nhd 6.79  Const: -1.57 vkt:20.24 wkgad: 2.00 popd: 6.89 fs: -3.64  0.3518nhd  52  Table 3. 10 Transportation Demand Management CPMs–Total/Severe(City of Ottawa, 2006) Pearson χ2 Model Group# Model Form DoF SD z-Statistics 2 κ  χ  9  0.05, dof  Urban, Modeled, TDM  7.57 Total Collisions/3yr = 0.5171VKT 0.7019 e  270  271  288  309  (0.01602scvc – 0.002175core + 0.0000012drive)  9.55 Severe Collisions/3yr = 0.0572VKT 0.7895 e  270  270  295  309  (0.0116scvc – 0.002921core + 0.0000012drive)  Const: -2.48 vkt:19.36 scvc: 6.10 core: - 5.51 drive: 2.22 Const: -9.47 vkt:19.46 scvc: 4.23 core: -6.45 drive : 2.14  10 Urban, Measured, TDM  4.16  271  276  287  310  Const: 36.56 scc: 2.69  tlkm:11.04 core: -6.68  312  Const: 19.35 core: -6.49  tlkm:9.46  (0.005204scc – 0.003391core )  Total Collisions/3yr = 37.675TLKM 0.4661 e 3.90  272  260  292  – 0.003751core  Severe Collisions/3yr = 8.3145TLKM 0.4568 e 11 Rural, Modeled, TDM 5.34 Total Collisions/3yr =0.1228 VKT 0.7529  120  134  149  Const: -6.59 tcd: 6.91  vkt: 20.07  146  147  149  Const:-10.75 tcd: 6.89  vkt:18.33  128  134  148  Const: 9.43 tcd: 4.52  tlkm: 7.21 crp: -3.23  137  148  Const: 2.07 tcd: 3.99  tlkm: 5.73 crp: -3.14  e 0.2387tcd 6.31  Severe Collisions/3yr = 0.00167VKT0.8002 12  122 122  e 0.2625tcd  Rural, Measured, TDM  2.27 Total Collisions/3yr = 31.343TLKM 0.4054 e  121  (0.2466tcd – 0.009284crp)  1.92 Severe Collisions/3yr = 7.9169TLKM 0.3707 e  121  123  (0.2502tcd – 0.01028crp)  53  Model Group# 13  Table 3. 11 Network CPMs – Total/Severe. (City of Ottawa, 2006) Pearson χ2 Model Form DoF SD z-Statistics 2 κ  χ  0.05,dof  Urban, Modeled, Network  8.06  270  269  286  309  (0.432sigd + 0.2146intd – 0.003201i3wp )  Total Collisions/3yr = 0.3687VKT 0.7362 e 8.59 Severe Collisions/3yr = 0.054VKT0.7807 e 14  270  273  293  309  (0.5428sigd + 0.3312intd – 0.002556i3wp )  Const: -8.64 sigd: 2.46 i3wp: -2.24  vkt:18.93 intd: 2.97  Const: 28.45 sigd: 2.0 i3wp: -3.67  tlkm:13.26 intd: 2.21 lkp:-7.46  Urban, Measured, Network  4.80 Total Collisions/3yr = 63.5TLKM 0.545 e  0.5625  Severe Collisions/3yr = 3.4973TLKM 15 Rural, Modeled, Network  269  258  286  308  (0.546sigd+0.298intd – 0.0047i3wp – 0.014llkp )  4.57  e  4.68  271  247  291 310  (0.0176alkp – 0.003502i3wp )  122  Severe Collisions/3yr = 0.0307VKT0.7528  122  Const: 6.55 alkp:8.41  tlkm:11.73 i3wp: -2.80  vkt: 18.60  123  133  149  Const: -5.20 intd: 5.32  139  142  149  Const: -8.89 vkt: 16.09 intd: 4.52  132  135  148  Const:17.54 intd:3.27  tlkm: 9.92 llkp: -2.60  148  Const: 7.82 intd:2.36  tlkm: 8.01 llkp: -2.02  1.3116intd  Total Collisions/3yr =0.1924 VKT 0.7199 e 4.59  16  Const: -3.55 vkt: 21.31 sigd: 2.23 intd: 5.25 i3wp: -3.22  e 1.276intd  Rural, Measured, Network  1.99 Total Collisions/3yr = 21.2849TLKM 0.4742 e 1.62 Severe Collisions/3yr = 4.9978TLKM 0.4411 e  121  (1.256intd – 0.007908llkp)  121  129  137  (1.044intd – 0.007112llkp )  54  Sixteen urban CPMs and sixteen rural CPMs were developed using the year of 2006 data from the City of Ottawa. All model statistics met the model goodness of fit testing criteria (i.e. κ> 1, Pearson χ2 < χ2 and SD < χ2) which meant the successful model development. Again, these model results support the earlier findings by Lovegrove (2006, 2007). The significant explanatory variables listed below have the positive association with the collisions:  •  Exposure-related: vehicle kilometres travelled (VKT), total road lane kilometres (TLKM);  •  S-D-related: job density per unit area (WKGAD), population density (POPD), residential unit density (NHD);  •  TDM-related: shortcut capacity (SCC), shortcut attractiveness (SCVC), number of drivers (DRIVE), total commuter density (TCD);  •  Network-related: signal density (SIGD), intersection density per unit area (INTD), and arterial road lane kilometres percentage (ALKP).  The following explanatory variables showed the negative association with collisions: family size (FS), core size (CORE), 3-way intersections percentage (I3WP), and, local road lanekilometres percentage (LLKP). 3.5 Transference of Macro-level CPMs from the GVRD to the CRD  Since the urban measured models for the CRD were not developed successfully from the scratch, in order to provide a safety evaluation tool for practitioners use in planning process without requiring the Emme/2 or other complex transportation modeling resources, the urban measured macro-level CPMs developed by Lovegrove for the GVRD (1996) were transferred to the CRD. Following the macro-level CPM transferability guidelines recommended by Lovegrove (2007), the urban measured macro-level CPMs for the GVRD were calibrated with the CRD 2003 data. To do this calibration, all GVRD model parameter estimates were retained excluding the lead coefficient, a0 and the model’s shape parameter, κ. The GLM software was re-run using the CRD data to obtain new estimates for each CPM’s lead coefficient and shape parameter. At 95% confidence level, while the SD and Pearsonχ2 test statistics should  55  not exceed the target χ2, the Z score, as shown in Equation 2.4, less than one indicates that a transferred model meets statistical tests for goodness of fit to the data. Equation 2.4 was repeated for convenience below.  χ P2 − E ( χ P2 ) z = < 1.00 σ ( χ P2 )  (2.4) N  E ( χ P2 ) = N ; σ ( χ P2 ) = 2 N (1 + 3 / κ ) + ∑ i =1  1 E (Λ i )[1 + E (Λ i ) / κ ]  [ yi − E (Λ i )]2 Var( yi ) i =1  (2.5)  n  Pearsonχ 2 = ∑  (2.6)  Where, χ2p = Pearsonχ2 N = the number of data points used to calibrate the model. E(Λi ) = mean collision frequency predicted by transferred model for zone i κ = shape parameter for each calibrated CPM  yi, Var(yi) are derived for each individual observation in the new data set. Table 3.8 shows the urban measured community-based macro-level CPMs for the CRD (2003) transferred from the GVRD (1996).  56  Table 3. 12 Transferred Urban Measured CPMs (CRD,2003; GVRD,1996) Model Group# 2  Pearson  χ2  κ  DoF  1.2  470  518  530  518  zconstant = 18 ztlkm = 7  1.22  320  327  361  363  zconstant = 59.3 |Z| = 0.116  Model Form  χ  2  SD  0.05,dof  Urban, Measured, Exposure  GVRD model  Total Collisions/3yr = 92.48TLKM 0.4321 Transferred CRD model  Total Collisions/3yr = 20.78TLKM 6  z-Statistics and Z score  0.4321  Urban, Measured, Socio-Demographic  GVRD model  1.6  Total Collisions/3yr =74.2175TLKM 0.8218  Transferred CRD model  464  437  532 515  312  343  349 354  e (0.007462popd+0.06295unempp – 0.743fs)  1.06 Severe Collisions/3yr = 6.7127TLKM0.8532  518 514  e (0.0068popd+0.07899unempp – 0.5637fs)  1.50  Total Collisions/3yr =25.5082TLKM 0.8218  508  e (0.007462popd+0.06295unempp – 0.743fs)  1.3 Severe Collisions/3yr = 8.3645TLKM0.8532  463  317  331  366 360  e (0.0068popd+0.07899unempp – 0.5637fs)  zConst = 9 zpopd =6 zfs = -5 zConst = 4 zpopd = 5 zfs = -3  ztlkm =12 zunempp= 7 ztlkm =12 zunempp= 8  zConst = 65.35 |Z| = 0.684 zConst = 33.7 |Z| = 0.253  57  Table 3.12 Transferred Urban Measured CPMs (CRD, 2003;GVRD, 1996) (Continue) 10  Urban, Measured, TDM  1.5  GVRD model  460  484  517  511  (0.02702scc – 0.0000277core + 0.000123tcm)  Total Collisions/3yr =43.7285TLKM 0.5762 e 1.2 461  433  532  512  (0.01918scc – 0.0000334core +0.0000909tcm)  Severe Collisions/3yr = 7.2283TLKM0.7205 e 1.31 305 Transferred CRD model Total Collisions/3yr =8.33TLKM 0.5762  343  348  e (0.02702scc – 0.0000277core + 0.000123tcm) 1.03  Severe Collisions/3yr = 3.078TLKM0.7205 14  336  307  315  355  349  e (0.01918scc – 0.0000334core + 0.0000909tcm)  zConst = 14 ztlkm =12 zcore = -6 ztcm = 3  zscc =7  zConst = 7 ztlkm =8 zscc =4 zcore = -6 ztcm = 2 zconst = 41.61 |Z| = 0.646 zconst = 19.36 |Z| = 0.143  Urban, Measured, Network  GVRD model  1.9  Total Collisions/3yr =29.9342TLKM 0.8675 1.5  463  e  Total Collisions/3yr =21.5635TLKM  0.8675  Severe Collisions/3yr = 5.181TLKM1.04  514  463  526  514  (4.002sigd – 0.01899i3wp + 0.01587alkp)  e  1.25  511  (4.748sigd – 0.0204i3wp + 0.007193alkp)  463  Severe Collisions/3yr = 2.7096TLKM1.04 e 1.43 Transferred CRD model  511  322  302  357  365  (4.748sigd – 0.0204i3wp + 0.007193alkp)  322  301  367  365  e (4.002sigd – 0.01899i3wp + 0.01587alkp)  zConst = 11 ztlkm =14 zsigd =6 zalkp = 3 zi3wp = -12 zConst = 3 ztlkm =15 zsigd =4 zalkp = 5 zi3wp = -10 zConst = 64.4 |Z|= 0.469 zConst = 31.4 |Z|= 0.469  58  The transference results in tables 3.8 have shown that all z scores were less than one and other goodness of fit measures met the test criteria. Although some SD values were slightly exceeding their target χ2, all of them were within 3% of the χ2 statistic. The z scores suggested that all models were transferred successfully, meeting statistical tests with a 95% level of confidence. All of the κ values in the transferred models were lower than in the original GVRD models, meaning a relatively worse fit model. This worse fit was due to the forced nature of the GLM fitting process, in which the transferred model was calibrated with all parameters pre-set at the original model values except the leading parameter. Hence, caution should be taken in using transferred models since their model fit is usually worse than models developed from scratch. The re-estimated ao’s for all but one transferred model were lower than the original GVRD CPMs. This was likely due to the fact that CRD in general had lower traffic volumes than GVRD, with correspondingly lower collisions. 3.6 Summary  Following the methodology recommended by Lovegrove (2007), data was extracted and community-based macro-level CPMs were developed or transferred for the CRD and the City of Ottawa. To facilitate successful models, errors due to aggregation bias were minimized through stratification of independent variables into sixteen model groupings shown in Table 3.3. Based on the candidate variables listed in Table 3.1 and Table 3.2, several data extraction sources have been identified and described. Based on the model development guidelines described in Lovegrove (2007), the macro-level CPM form, GLM regression process with negative binomial error distribution, explanatory variable selection procedure, outlier analysis, and goodness of fit criteria were followed.  Thirty-two  community-based macro-level CPMs for CRD and thirty-two community-based macro-level CPMs for the City of Ottawa have been developed and/or transferred, which were presented in Tables 3.4 to 3.12. Each CPM predicted the three-year total of collisions for two types of collisions: Total and Severe. The model results support the earlier findings by Lovegrove (2007).  59  CHAPTER IV CASE STUDY: APPLICATION OF COMMUNITY-BASED MACROLEVEL CPMs FOR SAFETY EVALUATION OF ROAD PATTERNS 4.1 Introduction  Having developed community-based macro-level CPMs for the CRD and the City of Ottawa, it was appropriate to apply them to evaluate the safety level of various road network patterns. In section 4.2, the characteristics of each road network pattern have been introduced. In section 4.3 and 4.4, the methodology to evaluate the safety of road networks has been described. In sections 4.5 and 4.6, the road safety evaluation results have been presented. In section 4.7, the accessibility of the road networks has been described. In section 4.8, discussions and conclusions have been presented. 4.2 Background  Roads and public spaces are a major determinant of the potential sustainability of the urban area, its quality and amenity. The design of neighborhood roads is a critical consideration for establishing safe, quiet and mobility-friendly communities. Currently, two basic road design patterns exist in North America: Traditional neighbourhood layout (Grid network) and Conventional Suburban layout (Loops and Culs-de-Sac), as shown in Figure 4.1.  Figure 4. 1 Grid and Loops and Culs-de-Sac Road Patterns  The cul-de-sac network is essentially a dead-end residential street, often but not always ending with a large circular patch of pavement allowing vehicles to turn around. The cul-de60  sac can limit through-traffic, which will minimize residents’ exposure to cut-through vehicular traffic and keep the streets safe for children. Developers liked the cul-de-sac pattern because it made it possible to build on land unsuited to a grid street pattern. Some disadvantages of the cul-de-sac pattern are that it reduces emergency, utility, and transit vehicle access, and, it increases trip length and VKT, which makes local residents rely heavily on automobiles. Overall, the conventional loop and cul-de-sac road layout has the following characteristics: •  Curvilinear road network  •  Long routes to get short distances  •  Harder to navigate  •  Less paved area  •  Safer for pedestrians  •  Open space beyond walking distance for most residents  •  Limited access streets to arterials  •  Reduced transit accessibility  By contrast, the Traditional Neighbourhood Design (TND) or grid network calls for an interconnected network of streets and sidewalks to disperse vehicular trips and to make human-powered modes of travel (such as walking and biking) practical and attractive for short trips. Motorists, pedestrians, and public officials will find the regular pattern more understandable. The traditional neighbourhood design exhibits the following characteristics: •  Compact, defined urban neighbourhoods composed of a compatible mix of uses and housing types;  •  Direct, convenient routes throughout;  •  Easy to navigate;  •  More paved area for roads;  •  More accesses to Arterials compare to cul-de-sac layout;  •  Many four-way intersections in residential areas.  Recently, Canada Mortgage and Housing Corporation (CMHC) promoted a new model for sustainable subdivision development – The Fused Grid, as shown in Figure 4.2. The fused grid combines the two pre-existing road design patterns: the conventional loop and cul-de61  sac pattern and the grid network pattern. The objective of the fused grid model is to retain the best characteristics of each and none of their disadvantages while raising the quality of the neighbourhood environment. This model uses a continuous grid of roads for district and regional connectivity and a discontinuous grid of streets to preclude neighbourhood shortcutting. The neighbourhood grid is supplemented by footpaths that connect all streets, turning a neighbourhood into a fully connected pedestrian and bicycle realm. The fused grid model has the following characteristics (Hollingsworth, 2007): •  Mixed land uses. Dominant residential land use with commercial land uses contained in the major corridors;  •  Hierarchical grid of roads. Interconnected grid network of major roads while the local roads are looping and Cul-de-Sac;  •  Frequent access points to the surrounding arterials;  •  Dominant 3-way intersections over 4-way intersections;  •  Improved pedestrian/bicyclist mobility, and transit accessibility.  Figure 4. 2 Fused Grid Model  62  Figure 4.3 shows the basic configurations of the fused-grid layout: • Quadrants: 16 hectare residential areas, which are served internally by cul-de-sacs and loops to discourage through-traffic; • Neighbourhoods: 64 hectares, made up of 4 quadrants and bounded by arterials and collectors; and • Districts: 256 hectares made up of 4 neighbourhoods and bounded by arterials.  Figure 4. 3 Fused Grid Layout Configurations  The sustainability benefits of the fused grid pattern, such as improvement on land use and infrastructure efficiency, reduction of environmental impact, increment of walkability, bikeability, and transit access, can be calculated by comparing layouts for a given site and measuring the specific site plan elements. Other performance characteristics, such as traffic flow, and propensity to walk or drive can be assessed by computer simulation and using analytical tools. Two research studies have recently been completed that reported on the questions of traffic performance and the degree to which patterns such as this will encourage  63  more walking (Hollingsworth, 2007; Frank, 2006). These two research studies carried out using the community of Barrhaven in the City of Ottawa as a test site, shown in Figure 4.4. The existing street network typifies the loops and culs-de-sac road pattern. The shaded area is referred to as Street Network Analysis Area with area size 335 ha. A railway passes through the community just beside the analysis area.  Figure 4.4 Test Site - Barrhaven  The results of the reported two studies showed that the fused grid network would perform well in both respects, particularly under a denser, mixed use land development scenario under which other concepts suffer degradation of performance. Building on these positive results it would be a constructive next step to assess the safety characteristics of the fused grid. As described in Chapter II, using the community-based, macro-level CPMs developed for the GVRD, Lovegrove (2007) compared the relative safety of various road networks based on 64  the created modules, including traditional grid network, conventional culs-de-sac network, Dutch SRS network and Off-set 3-way network. The Dutch SRS road network and Off-set 3way road network are shown in Figure 4.5. The analysis results showed that the 3-way Offset network appeared to be significantly safer than all other networks. The Dutch SRS network and the cul-de-sac network were projected to be second safest. The grid road network was projected to be least safe. According to the comparison results, in relative terms, the safety of cul-de-sac neighbourhood road networks appeared to be significantly safer than grid networks by a factor of nearly three to one. However, Lovegrove did not evaluate the relative safety of the newly promoted fused grid road network.  Figure 4.5 Dutch SRS Road Network and 3-Way Offset Road Network 4.3 Model Selection  Following the recommended guidelines by Lovegrove (2007), the first step was to select the appropriate models. In this case, Urban measured and modeled macro-level CPMs for total collisions were selected. The models from the Network and TDM groups were used. Table 4.1 shows the selected CPMs from the CRD, the City of Ottawa and the GVRD applied in this case study. In order to comprehensively evaluate the road patterns safety level and provide more reliable results, the GVRD models were also transferred to CRD and Ottawa and used in the analysis in additional to those developed from scratch. As the analysis was a relative comparison with no future forecasts desired, the time period most convenient to dataset availability was chosen, in order that both modeled and measured CPMs could be used.  65  Model Group  Table 4. 1 Selected CPMs for Safety Evaluation Model Form  URBAN, MODELED, TDM GVRD (developed from scratch)  Total Collisions/3yr =0.63052VKT 0.6887 e  (0.07924scvc-0.00207core+0.000000912drive)  κ = 1.9  CRD (developed from scratch)  Total Collisions/3yr = 0.1028VKT 0.915 e  (0.0134scvc – 0.0106core)  κ = 3.5  Ottawa (developed from scratch)  Total Collisions/3yr = 0.5171VKT 0.7019 e  (0.01602scvc – 0.002175core + 0.0000012drive)  κ = 7.6  CRD (transferred from GVRD)  Total Collisions/3yr =0.2587VKT 0.6887  e (0.07924scvc–0.00207core+0.000000912drive)  κ = 1.6  e (0.07924scvc–0.00207core+0.000000912drive)  κ = 3.2  e (0.02702scc – 0.00277core + 0.000123tcm)  κ = 1.5  Ottawa((transferred from GVRD)  Total Collisions/3yr =0.5111VKT 0.6887 URBAN, MEASURED, TDM GVRD (developed from scratch)  Total Collisions/3yr =43.7285TLKM 0.5762 CRD (developed from scratch)  N/A Ottawa (developed from scratch)  Total Collisions/3yr = 37.675TLKM 0.4661 e  (0.005204scc – 0.003391core )  κ = 4.2  CRD (transferred from GVRD)  Total Collisions/3yr =8.33TLKM 0.5762  e (0.02702scc – 0.00277core + 0.000123tcm)  κ = 1.3  Ottawa((transferred from GVRD)  Total Collisions/3yr =21.3703TLKM 0.5762  e (0.02702scc – 0.00277core + 0.000123tcm)  κ = 2.8  66  Table 4.1 Selected CPMs for Safety Evaluation (Continue) Model Group Model Form URBAN, MODELED, NETWORK GVRD (developed from scratch)  Total Collisions/3yr =0.90626VKT 0.7851  e (2.399sigd + 0.7947intd – 0.02213i3wp )  κ = 2.4  CRD (developed from scratch)  Total Collisions/3yr = 0.1757VKT 0.8509 e  (0.4246sigd +0.829intd–0.00405i3wp-0.0086llkp)  κ = 4.6  Ottawa (developed from scratch)  Total Collisions/3yr = 0.3687VKT 0.7362 e  (0.432sigd + 0.2146intd – 0.003201i3wp )  κ = 8.1  CRD (transferred from GVRD)  Total Collisions/3yr =0.5202VKT 0.7851  e (2.399sigd + 0.7947intd – 0.02213i3wp  κ = 2.0  e (2.399sigd + 0.7947intd – 0.02213i3wp )  κ = 1.9  Ottawa((transferred from GVRD)  Total Collisions/3yr =0.8712VKT 0.7851 URBAN, MEASURED, NETWORK GVRD (developed from scratch)  e (4.748sigd – 0.0204i3wp + 0.007193alkp)  κ = 1.9  (0.546sigd+0.298intd – 0.0047i3wp – 0.014llkp )  κ = 4.8  e (4.748sigd – 0.0204i3wp + 0.007193alkp)  κ = 1.6  e (4.748sigd – 0.0204i3wp + 0.007193alkp)  κ = 1.6  Total Collisions/3yr =29.9342TLKM 0.8675 CRD (developed from scratch)  N/A Ottawa (developed from scratch)  Total Collisions/3yr = 63.5TLKM 0.545 e CRD (transferred from GVRD)  Total Collisions/3yr =21.5635TLKM 0.8675 Ottawa((transferred from GVRD)  Total Collisions/3yr =27.6604TLKM 0.8675  67  4.4 Approach  Having selected the appropriate models, the approach for assessing the safety level of each of these neighbourhood road patterns has been discussed below. 4.4.1 Modular Test Networks The railway (shown in Figure 4.4) discontinuity makes Barrhaven test site a less than ideal site to analyze the safety level of road networks. To ensure that no bias was introduced, the modular test networks were designed for each road patterns. Another advantage of using modular is that the theoretical potential of safety for each road network would be evaluated. Road networks display different patterns and characteristics (e.g. INTD, I3WP) for different size modules especially for the fused grid network. Lovegrove (2007) only used a small size (i.e. 10 ha) module in GVRD study. In order to use the same methodology to evaluate the safety of larger size modules (i.e. close to real world community), a progressive safety comparison approach was used. Thus three scaled versions of each test network were plotted by size of Quadrants (16 hectares), Neighbourhoods (64 hectares), and Districts (256 hectares) based on the fused grid layout configurations. 4.4.2 Barrhaven Real World Networks In order to validate the safety evaluation result which was based on the idealized test modules, safety analysis on real world networks - Barrhaven - was also conducted. Although the Barrhaven site is not an ideal test site due to its irregular shape, it was still valuable to study for comparative purposes. Table 4.2 contains aggregated data for the analysis area. Table 4.2 Analysis Area Data  Analysis Area  Total Collisions (05 – 07)  VKT  TLKM (km)  AREA (Ha)  SCC  INTD  I3WP (%)  LLKP (%)  SIGD  188  3802  82.98  334  0  0.48  95.2  59  0.0076  In order to compare the relative safety of the five road networks, all but the existing network needed to be created for the same analysis area.  68  4.4.2.1 Tradition Grid Road Network  The traditional grid network design for the study area was created in a previous study under the direction of the CMHC (1997). The traditional layout consists of: •  A hierarchical road pattern arranged in a orthogonal manner;  •  Blocks sizes of 60-120 metres by 120-240 metres;  •  3.5-metre lanes, 1.2-metre bike lanes and 2.4-metres for parking;  •  The use of both 3-way and 4-way intersections in a ratio of 2.6 to 1;  •  A fair number (14) of connecting local roads to the surrounding arterials; and  •  Integrated pedestrian and cycling path system within the neighbourhood district.  The design for the traditional grid road network showing street layout, including road classifications, and traffic signal locations is presented in Figure 4.6.  Figure 4. 6 Traditional Grid Road Network  69  4.4.2.2 Fused Grid Road Network  The fused grid neighbourhood design for the study area was also created in a previous study under the direction of the CMHC (2005). Figure 4.7 shows the fused grid road network layout with road classifications and traffic signal locations. The fused grid layout had the following characteristics (Hollingsworth, 2007): •  Hierarchical street pattern of arterials, major collectors, minor collectors, and local streets arranged in a orthogonal manner;  •  Discontinuous loop or cul-de-sac local roads;  •  Most block lengths were under 200 metres but reach a maximum of 600 metres – larger blocks were divided by pedestrian paths;  •  Street cross sections that followed the principals set forth in Neo-Traditional design;  •  A ratio of 3-way intersections to 4-way intersections of 4.7 to 1;  •  11 major access roads to the surrounding arterials, each neighbourhood has 4 main access roads and an additional 16 secondary access roads to the major collectors that surround them;  •  Clearly defined neighbourhoods by virtue of the collectors that frame them and their opaqueness to through traffic;  •  Twinned grid (couplets of one-way streets) of major collectors.  The Fused-grid has high connectivity among the collector streets but has little road connectivity within the residential quadrants.  70  Figure 4. 7 Fused Grid Road Network  4.4.2.3 Dutch SRS and 3-way Offset Road Networks  Based on the Fused Grid layout configurations, the Dutch SRS and 3-way offset neighbourhood layout were also created as shown in Figure 4.8 and Figure 4.9.  71  Figure 4. 8 Dutch SRS Road Network  Figure 4. 9 3-way Offset Road Network  72  4.4.3 Selection of Trigger Variables After various test road networks were created, the trigger variables needed to be selected from the model groups (i.e. TDM and Network). Trigger variable are defined as variables that have different values in different road network patterns, causing significant differences in road collision predictions among the road network patterns. Trigger variables in this case study included: •  TDM – SCC;  •  Network – INTD, I3WP, LLKP, SIGD  Apart from trigger variables, the variables that hold the same values in different road network patterns were called control variables. For this analysis, the control variables included: •  Exposure - VKT, VC,TLKM;  •  S-D - POPD, FS, NHD, WKGAD;  •  TDM - DRIVE, TCD, CORE, CRP;  The values of control variables were calculated in two steps. First, two typical zones for each road network were picked from each region of the GVRD, CRD and City of Ottawa. Since exact Dutch SRS, 3-way offset and fused grid networks do not exist in these regions, the most similar zones were selected. In order to make the selected zones have similar socialdemographic and driver behavior, the zones were selected as close as possible in spatial distance. Second, averages of all control variables (i.e. VKT, VC, POP, AREA etc.) for selected zones were taken in each region as input data for CPMs. Values for each trigger variable were measured from the created road networks. It was assumed that in each direction there were two lanes on arterial roads, one lane on collector (major and minor) and local roads. The intersections with signal control were determined by engineer judgment. In order to minimize the bias of variable values on module boundaries, two rules were followed. First, if the measured element was on the corner of module, the element value was counted as one fourth. Second, if the measured element was on the boundaries of module, the element value was counted as half. For instance, the number of signals for a quadrant grid network module was counted as one instead of four. This method  73  was a slightly improved methodology from that described in Lovegrove (2007). Using the trigger variable values and the remaining input data, the network and TDM group macrolevel CPMs were run to project the collisions for each road network layout. 4.5 Results On Module Test Networks  The comparative safety level results of the five road networks for each analyzed module size were discussed below. 4.5.1 Quadrant (16 Ha) The created quadrant modules of each road network pattern have been shown in Figure 4.10.  Figure 4. 10 16 Ha Modular Road Networks  74  Table 4.3 contains the values of trigger variables for the five road networks of the 16 ha modules. Table 4. 3 Summarized Trigger Variable Values (16 ha) Road Pattern  SCC  INTD  I3WP  LLKP  SIGD  Grid Network  12  0.56  0%  67%  0.0625  Culs-de-Sac  0  0.44  56%  62%  0.1875  Dutch SRS  0  0.44  0%  64%  0.0625  3-way Offset  0  1.06  94%  67%  0.0625  Fused Grid  0  0.56  88%  67%  0.0625  Since only the grid network contained non-nil SCC values, in order to avoid the bias introduced by averaging values for the other networks, only the CPMs from the Network group were used. The safety effect of SCC for grid network was estimated through the CPMs from the TDM group. A summary of the safety evaluations for each of the five road networks is contained in Table 4.4 Models GVRD  Table 4. 4 Comparison of Collision Densities (16 ha) Grid Cul-deDutch 3-way Network Sac SRS Offset Col’ Density* 7.86 3.54 7.12 1.22  Fused Grid 1.15  (Developed)  Ratio to F.G*  6.45  3.08  6.19  1.06  1  CRD  Col’ Density  2.25  1.55  1.68  0.96  0.84  (Developed)  Ratio to F.G  2.94  1.85  2.00  1.14  1  CRD  Col’ Density  5.52  2.41  5.10  0.91  0.82  (Transferred)  Ratio to F.G  4.67  2.12  4.01  1.06  1  Ottawa  Col’ Density  1.36  1.13  1.31  1.04  0.94  (Developed)  Ratio to F.G  1.45  1.20  1.39  1.11  1  Ottawa  Col’ Density  6.39  2.75  5.78  1.03  0.93  (Transferred)  Ratio to F.G  6.87  2.96  6.22  1.11  1  Average Ratio to F.G  4.48  2.24  3.87  1.09  1  Difference from F.G  348%  124%  287%  9%  0%  t-Statistic (t8, 90% = 1.86)  3.01  3.10  2.83  1.26  -  *Col’ Density = Collision Density = collisions / hectare F.G = Fused Grid Note: Values in table were projected by Network CPMs. The values for grid network were adjusted by TDM CPMs according to SCC  75  Using a 90% confidence level to test for statistical significance, the fused grid network appeared to be safer than all other networks other than the 3-way Offset. The grid road network was projected to be the least safe road pattern. Although the Dutch SRS network was safer than the grid network, its safety level was still worse than the intuitive expectation. The possible reason was that the Dutch SRS network limited the local trips and road accesses to major roads only instead of using internal connecting roads. However, the variable that represented the number of local-arterial intersections (i.e. IALP) was not included in the models due to the low z-statistics value (i.e. <1.96). Future research on refining the models to include the IALP variable should be conducted. Except for the fused grid, the 3-way offset network was projected to be the safest.  These results were consistent with earlier  expectations and results (Lovegrove, 2007). 4.5.2 Neighbourhood (64 ha) The created neighbourhood modules of each road network pattern are shown in Figure 4.11  76  Figure 4. 11 64 Ha Modular Road Networks  77  Table 4.4 lists the trigger variable values for the five road networks of the 64 ha modules Table 4. 5 Summarized Trigger Variable Values (64 Ha) Road Pattern SCC INTD I3WP LLKP SIGD  Grid Network  12  0.56  0%  67%  0.0625  Culs-de-Sac  0  0.44  56%  62%  0.1875  Dutch SRS  0  0.44  0%  64%  0.0625  3-way Offset  0  1.06  94%  67%  0.0625  Fused Grid  0  0.63  70%  67%  0.125  A summary of the safety evaluations for each of the five road networks is contained in Table 4.5.  Models GVRD  Table 4. 6 Comparison of Collision Densities (64 ha) Grid Dutch 3-way Cul-de-Sac Network SRS Offset Col’ Density* 7.86 3.47 7.12 1.22  Fused Grid 1.73  (Developed)  Ratio to F.G*  4.54  2.00  4.12  0.71  1  CRD  Col’ Density  2.25  1.55  1.68  0.96  1.06  (Developed)  Ratio to F.G  2.12  1.46  1.58  0.91  1  CRD  Col’ Density  5.52  2.41  5.10  0.91  1.54  (Transferred)  Ratio to F.G  3.58  1.56  3.31  0.59  1  Ottawa  Col’ Density  1.36  1.13  1.31  1.04  1.06  (Developed)  Ratio to F.G  1.28  1.07  1.24  0.98  1  Ottawa  Col’ Density  6.39  2.75  5.78  1.03  1.75  (Transferred)  Ratio to F.G  3.65  1.57  3.30  0.59  1  Average Ratio to F.G  3.03  1.53  2.71  0.76  1  Difference from F.G  203%  53%  171%  -24%  0%  t-Statistic (t8, 90% = 1.86)  2.60  1.87  2.38  1.74  -  *Col’ Density = Collision Density = collisions / hectare F.G = Fused Grid Note: Values in table were projected by Network CPMs. The values for grid network were adjusted by TDM CPMs according to SCC  78  Table 4.6 showed that the 3-way offset network pattern was projected to be the safest for neighbourhood module (64 ha). The fused grid was projected to be the second safest, followed by the cul-de-sac and the Dutch SRS. The grid network was the least safe road pattern. However, the t-statistic showed that there was no significant difference between the fused grid network and the 3-way offset network in terms of average collision density. These results were not surprising since the road pattern of fused grid for a neighbourhood module had some differences from the pattern for a quadrant module. Therefore, the trigger variable values of the fused grid were changed for a neighbourhood module. Other road network patterns kept the same trigger variable values as quadrant module. The collision ratio to fused grid of all road patterns decreased due to the increased collision density of fused grid for a neighbourhood module. 4.5.3 District (256 ha) The created modules of each road network pattern for districts are shown in Figure 4.7, Figure 4.12 and Figure 4.13. Twinned grids (couplets of one-way streets) of major collectors were added in the middle of modules.  79  Figure 4. 12 256 ha Modules of Grid and Culs-de-Sac Road Networks  80  Figure 4. 13 256 ha Modules of Dutch SRS and 3-way Offset Road Networks  81  Figure 4. 14 256 ha Modules of Fused Grid Road Network  Table 4.7 lists the trigger variable values for the five road networks of the 256 ha modules. A summary of the collision densities for each of the five road networks is contained in Table 4.8. Table 4. 7 Summarized Trigger Variable Values (256 Ha)  Road Pattern  SCC  INTD  I3WP  LLKP  SIGD  Grid Network  12  0.66  19%  62%  0.0977  Culs-de-Sac  0  0.50  62%  57%  0.1914  Dutch SRS  0  0.50  15%  59%  0.0977  3-way Offset  0  1.10  91%  62%  0.0977  Fused Grid  0  0.68  79%  62%  0.1445  82  Models GVRD  Table 4. 8 Comparison of Collision Densities (256 ha) Grid Dutch 3-way Cul-de-Sac Network SRS Offset Col’ Density* 6.30 3.22 6.04 1.50  Fused Grid 1.98  (Developed)  Ratio to F.G*  3.18  1.63  3.05  0.76  1  CRD  Col’ Density  2.05  1.52  1.62  1.07  1.07  (Developed)  Ratio to F.G  1.92  1.42  1.51  1  1  CRD  Col’ Density  4.37  2.20  4.29  1.12  1.39  (Transferred)  Ratio to F.G  3.14  1.58  3.09  0.81  1  Ottawa  Col’ Density  1.51  1.29  1.46  1.25  1.20  (Developed)  Ratio to F.G  1.26  1.08  1.22  1.04  1  Ottawa  Col’ Density  5.09  2.51  4.87  1.26  1.58  (Transferred)  Ratio to F.G  3.22  1.59  3.08  0.80  1  Average Ratio to F.G  2.55  1.46  2.39  0.88  1  Difference from F.G  155%  46%  139%  -12%  0%  t-Statistic (t8, 90% = 1.86)  2.57  1.88  2.41  1.15  -  *Col’ Density = Collision Density = collisions / hectare F.G = Fused Grid Note: Values in table were projected by Network CPMs. The values for grid network were adjusted by TDM CPMs according to SCC  Table 4.8 showed that the safety level rank of the five road network patterns in District size (i.e. 256 ha) was same as it ranked in Neighbourhood size (i.e. 64 ha). The 3-way offset network was the safest road pattern. The fused grid ranked the second safest followed by the cul-de-sac and the Dutch SRS networks. The grid network was the least safe road pattern. According to these results, in relative terms, the 3-way offset network would have on average 12% fewer collisions than the fused grid network. However, this difference was not found to be significant according to the t-Statistic at a 90% level. On the other hand, the cul-de-sac network was found to have on average 46% more collisions and the Dutch SRS network 139% more collisions, than the fused grid. The grid network was predicted to have twofold more collisions than the fused grid network. As in earlier results, the safety of the cul-de-sac road network appeared to be safer than the grid network by a factor of nearly 1.7 to 1.  83  4.6 Results on Barrhaven Test Site  For each road network pattern, the trigger variable values for the Barrhaven test site, along with the corresponding values of the total lane kilometre (TLKM) variable, have been summarized in table 4.10. Table 4. 9 Trigger Variable Values of Each Road Network Pattern Road Patterns  TLKM  SCC  INTD  I3WP  LLKP  SIGD  82.98  0  0.48  95%  59%  0.0076  128.3  2.6  0.95  74%  61%  0.0076  Dutch SRS  90.33  0  0.34  51%  41%  0.0076  3-way Offset  93.62  0  0.67  99%  43%  0.0076  Fused Grid  88.82  0  0.51  83%  43%  0.0076  Loops and Cul-deSac (Existing) Traditional Grid Network  Using developed and transferred Ottawa CPMs, the projected total collisions of the analysis area for the five road networks have been given in Table 4.10. The control variable values (i.e. VKT, VC, Population etc.) were kept the same as in the existing Barrhaven road network values. A sample calculation of collision frequency is shown in Appendix D. Table 4. 10 Predicted Total Collisions for Barrhaven Road Patterns Loops and Culs-deSac (existing) Traditional Grid Network  Observed (3 years)  Developed CPMs Modeled Measured  Transferred CPMs Modeled Measured  188  174  376  129  226  n/a  208  590  304  518  Dutch SRS  n/a  194  615  309  521  3-way Offset  n/a  178  518  136  232  Fused Grid  n/a  182  521  177  308  Note: Values in table were projected by Network CPMs. The values for grid network were adjusted by TDM CPMs according to SCC  The CPM projection ranged from 129 to 376, and were within a reasonable error of the observed 188 Barrhaven collisions. This wide range can be explained by the irregular shape  84  of the Barrhaven neighbourhood, and the fact that it is not a pure culs-de-sac road network design. In any case, for this analysis, only the relative safety level of the five road networks was relevant. Hence, the CPM results could still be used for comparing road network safety levels. Table 4.11 shows the collision ratios of each road network compared to the fused grid road network. Table 4. 11 Collision Ratios to the Fused Grid Road Patterns Loops and Culsde-Sac (existing) Traditional Grid Network  Developed CPMs Transferred CPMs Modeled Measured Modeled Measured  Average  t-Statistic (t6, 95% = 2.45)  0.96  0.72  0.73  0.73  0.79  3.82  1.14  1.13  1.72  1.68  1.42  2.63  Dutch SRS  1.07  1.18  1.75  1.69  1.42  2.48  3-way Offset  0.98  0.99  0.77  0.69  0.87  2.00  1  1  1  1  1  --  Fused Grid  It was noted that, for this analysis area, the existing loops and culs-de-sac road network was projected to be the safest. This result seems not consistent with the previous results which concluded that the 3-way offset and fused grid road networks were safer than the cul-de-sac network. Looking at the existing road networks, two clues were found as to possible reasons. First, the existing road network was not a typical cul-de-sac network but a mixed loops and culs-de-sac. The loops were dominant over culs-de-sac. Referring to the safety analysis for the fused grid based on quadrant (16 ha) module in section 4.5.1, loops were projected to be a very safe road pattern. Second, the 3-way offset and fused grid road networks were created based on CMHC fused grid layout configurations which introduced more collectors than the existing road networks. This would make the 3-way offset and fused grid networks contain a lower proportion of local roads (LLKP variable), which had negative impacts in terms of safety. The same problem happened to the Dutch SRS road network. The Dutch SRS network was expected based on theoretical results to be safer than the grid network intuitively. However, since the Dutch SRS road network was created using a module pattern, and the grid network was created using a real world road pattern, these two road networks were projected to be at the same safety level, as shown in Table 4. 11.  85  The cul-de-sac road network was projected to be safer than the grid network by a factor of nearly 1.8 to 1 for this test site. The 3-way offset and fused grid road networks were safer than the Dutch SRS and grid networks. The grid network was projected to be the least safe road pattern. These results were consistent with the previous conclusions based on modules. 4.7 Accessibility  As a final check, and to be consistent with earlier Dutch road safety case studies, an accessibility analysis was carried out on each of the road networks. Accessibility is the ability to reach a desired destination; it is affected by mode of transportation and network connectivity (Hollingsworth, 2007). In a more accessible neighbourhood, people can reach many other activities or destinations quickly and conveniently; however, undesirable visitors are also more likely to access the area. In a limited accessible neighbourhood, there have been instances when disaster response teams have incurred significant delays because they had to circumnavigate poorly connected neighbourhoods. Accessibility is also related to navigability, that is, if it is difficult to orient oneself in a neighbourhood the ability to reach a destination is reduced (Hollingsworth, 2007). The study of accessibility for the road networks except Dutch SRS network was conducted by Hollingsworth using standard transportation planning modelling methodology, and the same five road networks used in this road safety application (2009). It was concluded that the three-way off-set network resulted in the lowest overall VKT of any of the road networks, and thus, provided the greatest vehicular accessibility. The grid network ranked the second followed by the fused grid network. The cul-de-sac road network showed the worst accessibility. 4.8 Discussion and Conclusions  Using community-based, macro-level CPMs developed and transferred for the GVRD, CRD and City of Ottawa, the relative safety level of five road networks (i.e. Grid Network, Culde-Sac, Dutch SRS, 3-way Offset, and Fused Grid) was evaluated and compared based on  86  the created three different size road network modules and a real world test site – Barrhaven. The results of all analyses were in line with intuitive expectations, and consistent with earlier research. They have been discussed below. 4.8.1 Grid Network The grid network was projected to be the least safe road network pattern. Two characteristics of the grid network make it a potential high collision road pattern: high value of shortcutting capacity (SCC) and low value of 3-way intersections (I3WP). Shortcutting in neighbourhoods usually occurs when a semi-direct route through a neighbourhood matches commuter desire lines, and/or when travel demand on perimeter arterials exceeds capacity (e.g. during congested rush hours). Local roads are used for lower-speed, local traffic only, and not for higher-speed, through traffic. However, through traffic could use local roads penetrating neighbourhoods to avoid the congestion on major roads. The high shortcutting capacity plays an important role in the safety level of the grid network. However, the shortcutting traffic can be reduced by traffic calming local roads. The 3-way intersections have a strong relationship with road safety levels. Increased proportion of 3-way intersections (I3WP) was associated with decreased collisions in urban areas. The grid network shows dominant 4-way intersections over 3-way intersections. Low proportion of 3way intersections also make grid network an unsafe road pattern. 4.8.2 Cul-de-Sac Unlike the grid network, there are physically no opportunities for shortcutting traffic through cul-de-sac neighbourhoods. Cul-de-sac streets patterns provide a quiet and safe place for children playing in neighbourhoods. However, Cul-de-sacs also create car-dependent zones whose inhabitants drive more than downtown dwellers due to the lack of public transit and/or pedestrian/bike routes. Since local roads are discontinuous, the driving creates traffic congestion as those local trips turn on to a limited number of major roads (i.e. collectors and/or arterials). Although intersection density for the cul-de-sac road network was relatively low, the congested traffic at local-major road intersections and excessive VKT have major  87  impacts on the safety level of the cul-de-sac road network. These impacts could be mitigated by increasing pedestrian and bicycle paths. 4.8.3 Dutch SRS The Dutch SRS road networks also permit no shortcutting traffic and limited accesses to major roads. However, the local roads in neighbourhoods are connected by 4-way intersections. The Dutch SRS network presents lower intersection density which is positive for road safety but it also presents a lower proportion of 3-way intersections. As described previously, decreased 3-way intersections are associated with increases in collisions. The relatively low 3-way intersection proportion has a major impact on the safety level of Dutch SRS road networks. Offsetting the 4-way intersections to 3-way intersections could improve the Dutch SRS network safety level. 4.8.4 3-way Offset The 3-way offset road network was deemed as the safest road pattern among the five road network patterns. Although the 3-way offset network has a higher intersection density, it also has dominant 3-way intersection proportion. 3-way intersections restrict shortcutting traffic through the neighbourhood. The number of conflict points at two offset 3-way intersections is less than at one 4-way intersection. Previous studies found that the collision rates were around 1.5 times as high at a 4-way than paired 3-way intersections (Kumula, 1997; Ogden, 1996). Therefore, the dominant 3-way intersections contribute the major part of the safety level of the 3-way offset road network. The 3-way offset road network also provides better vehicular accessibility than the fused grid and cul-de-sac patterns, due to the connected local roads. 4.8.5 Fused Grid The fused grid road network combines the characteristics of the grid and culs-de-sac neworks. It contains discontinuous inner roads but more accesses to major roads than the cul-de-sac network. Pedestrian and bicycle routes are also provided for local trips. More 3-way intersections are used in fused grid than other road patterns except for the 3-way offset  88  network. Therefore, the values of intersection density and 3-way intersection proportion for the fused grid network are less than the 3-way offset network but higher than the other three road networks. Shortcutting traffic is also restricted for the fused grid network. The road safety level of the fused grid road network was projected to be safer than all but the 3-way offset road network. Overall, it appeared that the 3-way offset and fused grid networks were the safest without significant difference, followed by the cul-de-sac and Dutch SRS road networks. The grid network was the least safe road network pattern. 4.9 Summary  To evaluate the relative road safety level of the five neighbourhood road networks, three different size modules were created for each of pattern. A real world cul-de-sac test site (i.e. Barrhaven in Ottawa) was also selected for partial validation of the comparison results. Using community-based, macro-level CPMs developed and transferred for the GVRD, CRD and City of Ottawa, the collision density for each road network were projected based on the trigger variables including SCC, INTD, LLKP, I3WP, and SIGD. Through the analysis of the projected results, it was concluded that the 3-way offset and fused grid networks were the safest over all, followed by the Dutch SRS and cul-de-sac road networks. The grid network was the least safe road network. In terms of accessibility, the 3-way offset network were still the best followed by the grid and fused grid networks. Not surprisingly, the cul-de-sac network showed the worst accessibility.  89  CHAPTER V CASE STUDY: MACRO-REACTIVE APPLICATION OF COMMUNITY-BASED MACRO-LEVEL CPMs FOR BLACK SPOT STUDY 5.1 Introduction  Community-based, macro-level CPMs have been demonstrated as a feasible and practical tool for application in black spot analyses (Lovegrove & Sayed, 2006). Following the guidelines described in Lovegrove (2007), macro-reactive black spot studies were conducted for the CRD and City of Ottawa using macro-level CPMs. In section 5.2, the methodology for macro-reactive black spot studies has been presented. In section 5.3, the results of the black study for the CRD and City of Ottawa have been discussed. 5.2 Approach  Following the recommended macro-reactive use guidelines (Lovegrove, 2007), described in Chapter II, black spot case studies were conducted for the CRD and City of Ottawa. The data for the case studies involved three years of collision data (2002 to 2004 for CRD, 2005 to 2007 for Ottawa) and zonal trait data comprising twenty variables. According to the guidelines, as both rural and urban neighbourhoods across the regions need to be evaluated and all model variables were potentially trigger variables, all sixteen model groups were selected as candidates in the black spot program. 5.2.1 Identification & Ranking Identification of Collision Prone Zones (CPZs) was done by taking the expected collisions, E(Λ), for each zone as produced by the CPMs and comparing it to the observed zonal count to produce an EB safety estimate. This resulted in eight EB safety estimates for each zone. The E(Λ) was also taken as the regional reference group norm. Once the EB safety estimates for all zones were generated, they were ranked according to their respective models. Following this, the modified ranking system, as outlined previously in section 5.2, was employed to identify the worst CPZs to be carried forward for diagnosis.  90  5.2.2 Diagnosis & Remedy A two step process was used for identifying safety problems in the CPZs identified by the black spot program. Both steps included both in-office and on-site analysis. The first indicator was determined by analysing CPZs to determine the location of the highest collision frequency. The second indicator was found through the investigation of possible trigger variables. This is to say variables used for any of the models within a CPZ that had values that were drastically different from the regional averages. The in-office analysis was supplemented by on-site investigations to ensure other safety factors (e.g. adjacent land uses, speeding, high truck volumes, and existing traffic calming levels) were not overlooked. Once all significant safety problems within a zone were identified, remedies were proposed for each CPZ. Remedies were put forward in an attempt to offer a remedy for each trigger variable theme. 5.3 Results  5.3.1 Identification & Ranking Upon applying the above procedure, at least one CPZ was identified by each CPM. However, two significant observations were made. First, the CPMs did not provide similar rankings for the zones across the model groups. This implied that no single CPZ had significant problems in all trigger variable themes but rather had trigger variables from only one or two themes and minor deviations from regional norm in other themes. Although each CPM had different top ranks, the variance between the rankings did not generally exceed 5%. Table 5.1 and Table 5.2 listed the top ranked CPZs for the CRD and City of Ottawa. The geographic locations of the top ranked CPZs have been shown in Figure 5.1 and Figure 5.2, respectively.  91  Table 5. 1 CPZ Identification for the CRD Model Exposure Exposure Socio-D Socio-D TDM TDM Group Modelled Measured Modelled Measured Modele Measured d Urban Rank 1 CPZ1 CPZ1 CPZ1 CPZ9 CPZ8 CPZ4 2 CPZ5 CPZ3 CPZ2 CPZ3 CPZ6 CPZ1 3 CPZ13 CPZ5 CPZ11 CPZ10 CPZ11 CPZ8 4 CPZ11 CPZ4 CPZ3 CPZ14 CPZ2 CPZ9 5 CPZ15 CPZ9 CPZ16 CPZ2 CPZ10 CPZ17 Rural Rank 1 CPZ24 CPZ21 CPZ22 CPZ25 CPZ22 CPZ23 2 CPZ21 CPZ22 CPZ24 CPZ21 CPZ26 CPZ25 3 CPZ22 CPZ24 CPZ28 CPZ23 CPZ28 CPZ21 4 CPZ26 CPZ23 CPZ26 CPZ30 CPZ21 CPZ22 5 CPZ31 CPZ27 CPZ21 CPZ27 CPZ32 CPZ27  Network Network Modelled Measured CPZ6 CPZ7 CPZ2 CPZ1 CPZ18  CPZ12 CPZ7 CPZ4 CPZ1 CPZ2  CPZ21 CPZ26 CPZ28 CPZ29 CPZ33  CPZ21 CPZ25 CPZ29 CPZ27 CPZ23  Table 5. 2 CPZ Identification for the City of Ottawa Model Exposure Exposure Socio-D Socio-D TDM TDM Network Network Group Modelled Measured Modelled Measured Modeled Measured Modelled Measured Urban Rank 1 CPZ101 CPZ101 CPZ101 CPZ101 CPZ103 CPZ101 CPZ102 CPZ102 2 CPZ103 CPZ102 CPZ103 CPZ102 CPZ101 CPZ102 CPZ104 CPZ103 3 CPZ105 CPZ104 CPZ107 CPZ104 CPZ107 CPZ103 CPZ103 CPZ104 4 CPZ106 CPZ106 CPZ105 CPZ108 CPZ104 CPZ104 CPZ105 CPZ101 5 CPZ109 CPZ103 CPZ110 CPZ103 CPZ105 CPZ108 CPZ101 CPZ111 Rural Rank 1 CPZ121 CPZ123 CPZ122 CPZ125 CPZ122 CPZ122 CPZ121 CPZ123 2 CPZ122 CPZ121 CPZ121 CPZ122 CPZ127 CPZ126 CPZ124 CPZ121 3 CPZ124 CPZ125 CPZ127 CPZ126 CPZ121 CPZ125 CPZ127 CPZ125 4 CPZ123 CPZ122 CPZ128 CPZ121 CPZ124 CPZ123 CPZ122 CPZ122 5 CPZ128 CPZ126 CPZ129 CPZ123 CPZ130 2933 CPZ123 CPZ131  92  Figure 5. 1 Collision Prone Zones (CRD)  Figure 5. 2 Collision Prone Zones (Ottawa)  93  5.3.2 Diagnosis & Remedy Usually, the worst CPZ should be diagnosed. However, after investigating the collision locations of each top ranked CPZ, it was found that for some CPZs, most collisions occurred at one major intersection. Obviously the intersections would need to be diagnosed and remedied, following a micro-reactive black spot analysis (i.e. individual intersection). To demonstrate the Macro-reactive method dealing with a zone-wide road safety problem and conducting a strategic level of safety analysis, other CPZs were therefore chosen instead. The diagnosis of the sample CPZs (two for the CRD and two for the City of Ottawa), has been described below, including recommended remedies for each CPZ. The first sample CPZ analyzed, urban CPZ2 for the CRD, shown in Figure 5.3, was identified by high rankings produced from the S-D, TDM, and network macro-level CPMs. Analysis of collision patterns along with the trigger variables, land use, and road maps provided four possible road safety problems. First, the zonal land use consisted of high density residential homes with grid road pattern. This means a high density of personal vehicles and high VKT in the zone. Second, the trigger variables were found to be total commuters (high), population density (high), home density (high), shortcut capacity (high), intersection density (high) and 3-way intersection percentage (low). Third, upon investigation of the collision locations, it was found that a majority of the collisions occurred on internal local roads, possibly due to the shortcutting traffic. Finally, it was found that a large portion of the collisions occurred during peak period (6am-9am and 3pm-6pm). These factors confirmed the suspicion that the safety problem might have been due to shortcutting vehicles through the neighbourhood during peak hours to avoid congestion on major roads. Another clue was that the percentage of commuters taking the bus was much lower than the zonal average. This indicated that there might be lack of transit service in this zone. As such, two alternative remedies were considered to solve these safety problems. Based on the significant factors thought to contribute to the collision frequencies within the CPZ, the first remedy considered was to reduce shortcut capacity by traffic calming shortcut routes. This would force the neighbourhood shortcutting trips to stay on the major roads and reduce the traffic on local roads. The second possible remedy was to improve bus service. 94  This option would both encourage people to use public transit, and reduce the number of personal vehicles used in the peak period. For each remedy considered in the macro-level CPM analysis, an estimate of the impact on CPM variable value was made using engineering judgment in order to generate an estimate of collision reductions for that remedy. For this CPZ, the macro-level CPMs predicted an overall collision reduction of up to 10%, or 7 collisions per 3 year period. These collision predictions were based on the following adjustments to macro-level CPM input variables in the TDM models: •  Zonal traffic volumes (VKT) are reduced by 10% as a result of more transit trips (less driving trips) and less shortcutting traffic.  •  Zonal shortcut capacity (SCC) is reduced to zero due to the traffic calming facilities.  Figure 5. 3 Urban CPZ2 (CRD)  The second sample CPZ analyzed was a rural zone for CRD, CPZ23, as shown in Figure 5.4. It was identified as collision prone by exposure, socio-demographic, TDM and network macro-level CPMs. The land-use within the zone is exclusively agricultural. The trigger variables were identified as arterial lane kilometres and local-arterial intersections. Upon investigation of the collision frequency locations, almost all collisions occurred on arterial roads in the zone. These clues suggested that the possible safety problem was that the low 95  speed local trips were in conflict with the high speed through traffic on arterial. After investigation of the collision causes, it was noticed that 30% of the collisions were caused by hitting deer. As such, two remedies were considered to reduce collisions. First, add left turn lanes on the arterial at each local road intersection exclusive for vehicles left turn off the arterial. Since all local-arterial intersections are 3-way intersections, it is possible to make the center lane without left turn vehicles off arterial as speed up buffer lane for the vehicles left turned on arterial from local roads. This remedy would reduce two types of collisions: the through vehicles on arterial with the vehicles waiting for left turn off arterial, and the through vehicles on arterial with the same direction vehicles left turned on arterial from local roads. Some intersection ahead warning signs and speed limited signs were also recommended to set up ahead of each intersection. The second remedy suggested was to install Wildlife Protection System or wildlife exclusion fence along the arterial road. If the above remedies were implemented, the predicted overall collision reductions would be up to 8% or 5 collisions per 3 years.  Figure 5. 4 Rural CPZ23 (CRD)  96  The third sample CPZ analyzed is an urban neighbourhood for the City of Ottawa, CPZ105, as shown in Figure 5.5. It was identified as collision prone by exposure, socio-demographic, TDM and network macro-level CPMs. Three clues were found to possible safety problems. First, the land use for this zone is high density residential. The road network pattern shows a typical grid network. Second, the trigger variables included population density (high), shortcut capacity (high), intersection density (high), and 3-way intersections (low). Third, half of collisions occurred along the arterial which is a boundary of this zone. All intersections on the arterial are signalized. These clues suggested that the safety problems might be caused by local trips mixed with through traffic, and together with access management and/or intersectional controls. Therefore, the following remedies were considered: •  Reduce shortcut capacity by traffic calming shortcut routes;  •  Convert the local-arterial intersection control to restricted right-in / right-out stop sign control. The vehicles on arterial do not need to stop at local road intersections, which increases arterial capacity and improves traffic smoothness on arterial.  •  Optimize the collector-arterial intersection signal timing phase. Provide left-turn phase and left turn bays for left turn vehicles.  The overall predicted collision reduction would be up to 7%, or 14 collisions per 3 years, if above remedies are implemented.  Figure 5. 5 Urban CPZ105 (the City of Ottawa)  97  The last CPZ analyzed was a rural neighbourhood for the City of Ottawa, CPZ124, as shown in Figure 5.6. It was identified by the exposure, TDM, and network macro-level CPMs. The land use within the zone is residential surrounded by agricultural land. One arterial road across the residential area connects two other arterials. The trigger variables included total commuters (high) and driver percentage (high). A review of the collision patterns revealed that the majority of collisions occurred on the arterials. From these clues, the safety problem might be caused by the arterial traffic running too high speed passing through the residential area in conflict with the local trips. Although there might be speed limit signs installed at entrance of the residential area, drivers usually ignore them after a period of high speed driving on rural arterial highway. To resolve this safety problem, the following remedies were considered: •  Change local-arterial intersections to right-in / right-out.  •  Build roundabouts at first intersections on arterials before vehicles entering residential area. Roundabouts can force drivers slow down their vehicles speed  Figure 5. 6 Rural CPZ124 (the City of Ottawa)  98  5.4 Summary  Following the recommended guidelines, black spot studies were conducted using macrolevel CPMs for the CRD and City of Ottawa. Four of the worst black spots (two urban and two rural) were carried forward and analyzed. The models successfully identified and ranked collision prone zones across the CRD and City of Ottawa and offered an efficient method for CPZ identification. The possible safety problems were diagnosed and some remedies were considered for each studied CPZ. It was concluded that the application of community-based macro-level CPMs to a black spot program was not a complicated procedure. This reinforces the conclusions put forward by Lovegrove (2007) in that macro-level CPMs provide an effective compliment to traditional RSIPs and possible early identification of collision prone zones.  99  CHAPTER VI CONCLUSIONS, CONTRIBUTIONS & FUTURE RESEARCH  6.1 Introduction  This chapter was split into three sections. In section 6.2, the summary of this thesis is presented together with the main research conclusions. In Section 6.3, the contributions of this research are highlighted. In Section 6.4, future research topics are recommended. 6.2 Summary & Conclusions  Recognizing the enormous social and economic costs of road collisions, the motivation for this research arose from the need for empirical tools to reduce the number and severity of collisions in a proactive manner. Compared to the traditional approach, the proactive engineering approach to road safety improvement has focused on predicting and improving the safety of planned facilities, to preclude black spots from occurring. Community-based macro-level CPMs have the advantage of addressing road safety in the planning stages. The first objective of this research was to develop or transfer community-based macro-level collision prediction models for the Capital Regional District (CRD) in BC, Canada, and the City of Ottawa in Ontario, Canada. Having developed the macro-level CPMs, it was appropriate to use them in practice. The second objective of this research was to evaluate and compare the safety level of various road patterns including: grid network, cul-de-sac, fused grid, Dutch SRS, and 3-way offset. The third objective was to conduct black spot studies for the CRD and City of Ottawa using these CPMs. Following the development and transferability guidelines of community-based macro-level CPMs developed by Lovegrove (2007), a total of 64 community-based macro-level CPMs were successfully developed or transferred for the CRD and City of Ottawa. These models can be used by community planners and engineers as a decision-support tool in road safety improvement programs.  100  In order to evaluate the relative safety level of the five neighbourhood road networks, three different size modules were created for each of the road network. A real world test site in Ottawa was also studied based on the created road networks. Using the community-based macro-level CPMs developed for the CRD and City of Ottawa, the collisions for each road network were projected based on the trigger variables including SCC, INTD, LLKP, I3WP, and SIGD. Through the analysis of the projection results, it was concluded that the 3-way offset and fused grid road networks were the safest over all, followed by the Culs-de-Sac and the Dutch SRS road networks. The grid network was the least safe road network. Following the recommended guidelines by Lovegrove (2007), black spot studies were conducted using macro-level CPMs for the CRD and City of Ottawa. The models successfully identified and ranked collision prone zones across the CRD and City of Ottawa and offered an efficient method for CPZ identification. Two black spots (one urban and one rural) were analyzed for each region. The possible safety problems were diagnosed and some remedies were considered for each studied CPZ. It was concluded that the application of macro-level CPMs to a black spot program was not a complicated procedure. This reinforces the conclusions put forward by Lovegrove (2007) in that macro-level CPMs provide an effective compliment to traditional RSIPs. 6.3 Research Contributions  Three main contributions of this research are offered. 6.3.1 Development of Community-based Macro-level Collision Prediction Models for use by the CRD and City of Ottawa planners and engineers to do empirical road safety planning The persistent, unacceptably high frequency and severity of road collisions has been caused enormous social and economic burden on societies worldwide. Governments everywhere are pursuing various strategies to reduce road collisions, usually termed Road Safety Improvement Programs (RSIPs). Among other strategies, traditional programs have relied heavily on a reactive road safety engineering approach, in which road safety problems are addressed in reaction to an identified, pre-existing collision history. While the reactive  101  engineering approach has been effective where applied, it suffers from having to wait for the necessary monitoring over several years to identify a hazardous location, and then for the resources to take remedial action in built residential communities, often after high costs in human life and using scarce capital funds. Moreover, the incidence and severity of road collisions remains unacceptably high. To avoid these high costs and reduce road collision frequency in a more sustainable fashion, researchers have begun to pursue a more proactive approach to road safety, one that explicitly evaluates road safety throughout the land use and transportation planning process, to avoid problems before they occur in built form. This proactive road safety engineering approach has only recently been made possible through research on the development of improved empirical tools that allow road safety planners and engineers to explicitly evaluate road safety at the planning and design stages of the development process.  These new tools, termed community-based macro-level collision  prediction models (CPMs), have shown significant potential in initial applications using realworld data. Following the development and transferability guidelines of macro-level CPM, 32 community-based macro-level CPMs for both the CRD and City of Ottawa were successfully developed or transferred. These models can be used by community planners and engineers as a decision-support tool in road safety improvement programs. For these reasons, the development of these total 64 community-based macro-level CPMs and the database for future application of the CPMs as part of this research has been proposed as being a significant contribution. 6.3.2 Safety level evaluation of CMHC’s Fused Grid road pattern and verification of how the level of road safety of the Fused Grid road network compares with four other known road networks Canada Mortgage and Housing Corporation (CMHC) is currently promoting a new model for sustainable subdivision development – The Fused Grid. This model has great potential to promote increasingly sustainable development patterns by combining several redeeming features from pre-existing models. Specifically, it improves on land use and infrastructure  102  efficiency, reduces environmental impact, increases walkability and improves the neighbourhood milieu. The Fused Grid incorporates many elements of the schemes that exhibit greater safety, which intuitively should result in safer road conditions. However, this inference is insufficient and may be inaccurate. Therefore, it was prudent to perform a road safety evaluation of the Fused Grid model and assess the results. Using the community-based macro-level CPMs developed for the CRD and City of Ottawa, the safety level of five road networks including Grid Network, Cul-de-Sac, Dutch SRS, 3way Offset, and Fused Grid, were evaluated and compared. It was concluded that the 3-way offset and fused grid road were projected the safest overall, followed by the cul-de-sac and Dutch SRS road networks. The grid network was the least safe road network. These results can be used by community planners and engineers in community neighbourhood layout design. For these reasons, the safety level evaluation of the five road network patterns as part of this research has been proposed as being the second significant contribution. 6.3.3 Black Spot Analyses for the CRD and City of Ottawa A black spot is defined as any location that exhibits a collision potential that is significantly high when compared with some normal collision potential derived from a group of similar locations. To ensure that resources are spent only on the locations with the highest potential for safety improvements, it is vital that a sound procedure be used to screen the road network in order to properly identify and rank black spots for diagnosis and treatment. Using community-based macro-level CPMs, the collision prone zones were properly identified and ranked for both CRD and the City of Ottawa. Two black spots (one urban and one rural) were analyzed for each region. The possible safety problems were diagnosed and some remedies were recommended for each studied black spot. For these reasons, the black spot studies as part of this research have been proposed as being a significant contribution.  103  6.4 Recommendations for Future Research  Four areas have been identified that warrant further research, as follows: 1) Automation of the Processes Required to Develop, Transfer, and Use Macro-level CPMs: Developing macro-level CPMs consists of two main steps. First, data extraction and manipulation into a database is required. This is a very time consuming step that requires using manual and computed assisted methods, several different softwares, and several disparate data bases. Second, the data is then inputted into regression software, and an iterative procedure followed to develop the CPMs. This includes macro-level CPM form selection, a GLM regression process with Poisson or negative binomial error distribution, forward stepwise explanatory variable selection procedure, goodness-of-fit evaluation, and, to refine the model fit if needed, outlier analysis. These two steps are very time consuming, and would need some form of automated procedures to facilitate widespread use of macro-level CPMs by practitioners. The application of macro-level CPMs involves six-step appropriate model selection, three-step macro-reactive use and three-step proactive use. In the case of transferability between two significantly distinct regions, a four-step transfer calibration process is recommended: Data Definitions, Sufficient Data Points, GLM Process, and Goodness of Fit of the calibrated models. Like the steps to develop CPMs, the steps involved in using and transferring them is very involved and can be time consuming. Therefore, automation of these processes would greatly save time and effort, while promoting technology transfer of these new empirical tools to practitioners. 2) Macro-level CPMs for pedestrian and biking: Fused grid road network has the advantage of walking facilities which would encourage people walking instead of driving. In this research, these walking and biking facilities were not considered in the development of macro-level CPMs, nor in the macro-level CPM analyses of the Fused Grid model. In future research, how the pedestrian and biking facilities affect collisions should be considered. 3) Experimental Validation of Fused Grid and 3-way offset road pattern modules: The road safety level evaluations and comparisons were based on the road pattern modules  104  designed in this study. The safety level results need to be verified through in-field built form experiments in the future. 4) Macro-level collision modification factors: Micro-level analyses can use well documented collision modification factors (CMFs) to forecast collision reductions for a given remedy at a single intersection or road segment; few exist for neighbourhood-wide remedies. In the black spot studies of this research, an estimate of the impact on CPM variable value was made using engineering judgment in order to generate an estimate of collision reductions for that remedy. In the future, research is required to generate reliable macro-level collision modification factor estimates for possible remedies.  105  REFERENCES  Canada Mortgage & Housing Corporation (CMHC). 1997. Conventional and Alternative Development Patterns Phase 1: Infrastructure Costs. Ottawa, Canada. CMHC. 2007. Assessment Of The Transportation Impacts Of Current And Fused Grid Layouts. Ottawa, Canada. de Leur, Paul, and Tarek Sayed. 2003. A framework to proactively consider road safety within the road planning process. Canadian Journal of Civil Engineering, 30(4): 711-719. de Leur, Paul, and Tarek Sayed. 2001. Using claims prediction model for road safety evaluation. Canadian Journal of Civil Engineering. 28(5): 804-812. de Leur, Paul, and Tarek Sayed. 2002. Development of a road safety risk index. Transportation Research Record 1784: 33 – 42. Frank, L., and C. Hawkens. 2006. Assessing travel and environmental impacts of contrasting levels of vehicular and pedestrian connectivity: assessing aspects of the fused grid – a review of literature and experiences. An internal report prepared by the University of British Columbia for CMHC, Ottawa, Ontario. Wegman, Fred, Letty Aarts, and Charlotte Bax. 2008. Advancing sustainable safety national road safety outlook for The Netherlands for 2005–2020. ScienceDirect, Safety Science 46: 323–343 Fotheringham, A.S., and M. Wegner. 2000. Spatial models and GIS: new potential and new models. Taylor & Francis Inc., London, U.K. Hadayghi, Alireza, Amer S. Shalaby, and Bhagwant N. Persaud .2003. Macro-level accident prediction models for evaluating the safety of urban transportation systems. Paper presented at Transportation Research Board Annual Meeting, January, TRB, Washington, D.C. Handy, Susan, Robert Paterson, and Kent Butler. 2003. Planning for street connectivity: Getting from here to there. American Planning Association Report Number 515, Washington DC. Hauer, Ezra. 1992. Empirical Bayes approach to the estimation of “unsafety”: the multivariate regression method. Accident Analysis & Prevention, 24(5): 457-477, Elsevier Ltd, Amsterdam, The Netherlands. Hauer, E., J.C.N. Ng, and J. Lovell. 1998. Estimation of safety at signalized intersections. Transportation Research Record 1185: 48-61, Journal of the TRB, Washington, D.C.  106  Ho, Geoffrey, and Marco Guarnaschelli. 1998. Developing a road safety module for the regional transportation model. Technical Memorandum One: Framework, ICBC, Vancouver, Canada. Hollingsworth, Brian. 2007. Assessment of the transportation impacts of current and fused grid layouts. Internal report, prepared by IBI Group for CMHC, Ottawa, Ontario. INRO Consultants Inc. (INRO). 2003. Emme/2 User’s Guide Manual. Montreal, Canada. Jordan, Phillip. 2001. Road safety audit – low costs, big benefits. Editor F.P.D. Navin, Proceedings from the Fourth International Conference on Accident Investigation. Reconstruction, Interpretation and the Law, hosted by the Department of Civil Engineering, University of British Columbia, Vancouver, Canada. Khondaker, B., T. Sayed, and G.R. Lovegrove. 2009. Transferability of community-based macro-level collision prediction models between different time-space regions for use in road safety planning applications. Paper presented at Transportation Research Board Annual Meeting, January, Washington, D.C. Khisty, C. Jotin, and Kent B. Lall.1998. Transportation engineering: an introduction. 2nd Edition, Prentice Hall, Upper Saddle River, NJ, USA, 664-696. Kim, Karl, and Eric Yamashita. 2002. “Motor vehicle crashes and land use: empirical analysis from Hawaii. Transportation Research Record, 1784: 73-79, Journal of the Transportation Research Board, Washington, D.C. Kumula, R. 1997. Safety at highway junctions based on predictive accident models. Paper presented at Third International Symposium on Intersections without Traffic Signals. Portland OR. Ladron de Guevara, Felipe; Simon P. Washington, and Jutaek Oh. 2004. Forecasting crashes at the planning level: A simultaneous negative binomial crash model applied in Tucson, Arizona. Paper presented at Transportation Research Board Annual Meeting, January, Washington, D.C. Lord, Dominique, and Bhagwant N. Persaud. 2004. Estimating the safety performance of urban road transportation networks. Accident Analysis & Prevention 36:606-620. Elsevier Ltd, Amsterdam, The Netherlands. Lovegrove, G. R. 2007. Road safety planning: new tools for sustainable road safety and community development. VDM Verlag Dr. Muller, Saarbrucken. Lovegrove, G.R. and T. Sayed. 2007. Macrolevel collision prediction models to enhance traditional reactive road safety improvement programs. Transportation Research Record: Journal of Transportation Research Board, 2019: 65-73. National Research Council, Washington, D.C.  107  Lovegrove, G.R. and T. Sayed. 2006. Macro-level collision prediction models for evaluating neighbourhood traffic safety. Canadian Journal of Civil Engineering 33: 609-621. Ogden, K.W. 1996. Safer roads: A guide to road safety engineering. Brookfield, Vt: Avebury Technical. Aldershot, Hants, England. Peden, M., R.Scurfield, D. Sleet, D. Mohan, A.A. Hyder, E. Jarawan, and C. Mathers. 2004. World report on road traffic injury prevention. World Health Organisation, Geneva. Pogharian, S., J. Tasker-Brown, and F. Grammenos. 2005. Residential street pattern design: a proposal. Canada Housing & Mortgage Corporation. Ottawa, Canada. Poppe, Frank. 1995. Risk figures in the traffic and transport evaluation module (EVV): A contribution to the definition study traffic safety in EVV. SWOV Report No. R-95-21, Leidschendam, Netherlands. Poppe, Frank. 1997a. Traffic models: inner areas and road dangers. SWOV Report No. R-9710, Leidschendam, Netherlands. Poppe, Frank. 1997b. Cost-benefit analysis of a sustainably safe road traffic system. SWOV Report D-97-17, Leidschendam, Netherlands. Ramage-Morin, Pamela L. 2008. Motor vehicle accident deaths, 1979 to 2004. Component of Statistics Canada Catalogue no. 82-003-X Health Reports, Vol. 19, no. 3, Statistics Canada. Richardson, Barbara C. 1999. Toward a policy on a sustainable transportation system. Transportation Research Record 1670: 27-34, Washington, D.C. Roberts, Kelvin. 1998. Designing places: planning consciousness. Recovery, 9(1):9-10. Insurance Corporation of British Columbia, North Vancouver, Canada. Sawalha, Z., and T. Sayed. 2006. Traffic accidents modeling: some statistical issues. Canadian Journal of Civil Engineering, Vol. 33(9):1115-1124. Sawalha, Z., and T. Sayed. 1999. Accident prediction and safety planning models for urban arterial roadways. Draft Report for the Insurance Corporation of British Columbia by the Department of Civil Engineering, University of British Columbia, Vancouver, Canada. Sawalha, Z., and T. Sayed. 2001. Evaluating Safety of Urban Arterial Roadways. Journal of Transportation Engineering, ASCE, Vol. 127(2): 151-158. Sayed, T., and F. Rodriguez. 1999. Accident prediction models for urban unsignalized intersections in British Columbia. Transportation Research Record 1665: 93-99. Washington, D.C.  108  Sayed, T. 1998. New and essential tools, part 1: selecting sites for treatment. Proceedings from Seminar on New Tools for Traffic Safety, Section 3, hosted by the Institute of Transportation Engineers, August 9th, ITE, Toronto, Canada. Sayed, T., W. Abdelwahab, and F. Navin. 1995. Application of fuzzy pattern recognition to the identification of accident prone locations. Journal of Transportation Engineering, ASCE, 121(4): 352-358. Transport Canada, Health Canada. 2004. Road Safety in Canada – An Overview. ISBN 0662-36440-6, Ottawa, Canada. Transport Canada. 2007. Canadian Motor Vehicle Traffic Collision Statistics: 2006. Ottawa, ON, Canada. van Minnen, J.. 1999. The suitable size of residential areas: a theoretical study with testing to practical experiences. SWOV Report R-99-25, Leidschendam, Netherlands. van Schagen, Ingrid, and Theo Janssen. 2000. Managing road transport risks: sustainable safety in the Netherlands”, Risk Management in Transport 24(2): 18-27, Leidschendam, The Netherland. Wegman, Fred. 1996. Sustainable safety in the Netherlands. SWOV Institute for Road Safety Research, Leidschendam, The Netherlands. Wegman, Fred, Letty Aarts, and Charlotte Bax. 2008. Advancing sustainable safety: National road safety outlook for The Netherlands for 2005–2020. SWOV Institute for Road Safety Research, Leidschendam, The Netherlands. Wilson, E.M. 2004. NCHRP synthesis of highway practice 336: road safety audits. Transportation Research Board, National Research Council, Washington, D.C. World Health Organization (WHO). 2004. World report on road traffic injury prevention: summary. Geneva, http://www.who.int/violence_injury_prevention/publications/road_traffic/world_report/en/ (Accessed Dec. 10, 2008)  109  APPENDIX A – Model Development GLIM4 Output Sample [o] GLIM 4, update 9 for Microsoft Windows Intel on 11 Jan 2009 at 13:20:48 [o] (copyright) 1992 Royal Statistical Society, London [e] $C Ottawa Urban Modeled S3 $ [w] -- model changed [w] -- model changed [o] scaled deviance = 293.43 (change = -626.2) at cycle 2 [o] residual df = 270 (change = 0 ) [o] [o] ML Estimate of THETA = 8.592 [o] Std Error = ( 1.101) [o] [o] NOTE: standard errors of fixed effects do not [o] take account of the estimation of THETA [o] [o] 2 x Log-likelihood = 25832. on 270 df [o] 2 x Full Log-likelihood = -1905. [o] [o] Scaled deviance is 293.4 on 270 d.f. from 275 observations [o] change is -626.2 for 0 d.f. [o] [o] estimate s.e. parameter [o] 1 -2.918 0.3379 1 [o] 2 0.7807 0.04125 LVKT [o] 3 0.5428 0.2204 SIGD [o] 4 0.3312 0.1117 INTD [o] 5 -0.002556 0.001140 I3WP [o] scale parameter 1.000 [o] [o] %PE %SE TSTATN CHI2 TARGET [o] 1 -2.917935 0.337884 -8.636 272.7 309.3 [o] 2 0.780733 0.041252 18.926 272.7 309.3 [o] 3 0.542765 0.220351 2.463 272.7 309.3 [o] 4 0.331236 0.111730 2.965 272.7 309.3 [o] 5 -0.002556 0.001140 -2.242 272.7 309.3 [o] Current model: [o] [o] number of observations in model is 275 [o] [o] y-variate S3 [o] weight * [o] offset * [o]  110  [o] [o] probability distribution is defined via the macro NB_FIT. [o] link function is defined via the macro NB_LINK. [o] scale parameter is 1.000 [o] [o] linear model: [o] terms: 1+LVKT+SIGD+INTD+I3WP [o] correlations between parameter estimates [o] 1 1.0000 [o] 2 -0.9501 1.0000 [o] 3 -0.3163 0.2302 1.0000 [o] 4 -0.2386 0.0869 -0.4022 1.0000 [o] 5 -0.2660 -0.0154 0.4476 0.1775 1.0000 [o] 1 2 3 4 5 [o] [o] working triangle [o] 1 1607. [o] 2 7.608 691.5 [o] 3 0.08433 -0.09415 36.78 [o] 4 0.4205 -0.1265 0.9809 82.71 [o] 5 68.45 10.90 -103.6 -17.40 769203. [o] 6 3.032 0.6598 1.132 0.3757 -0.002556 272.6 [o] 1 2 3 4 5 6 [o] $C Outliers Analysis [o] ZONE %CD [o] 1 151.0 0.0043656048 [o] 2 152.0 0.0015378761 [o] 3 161.0 0.0000178007 [o] 4 162.0 0.0018678149 [o] 5 163.0 0.0347770117 [o] 6 171.0 0.0093227718 (Remaining outliers have been removed from this appendix, but are still in original log file)  111  APPENDIX B – GLIM4 Output Sample of Model Transference [o] GLIM 4, update 9 for Microsoft Windows Intel on 11 Jan 2009 at 13:20:48 [o] (copyright) 1992 Royal Statistical Society, London [e] $C Ottawa Urban Modeled T3, GVRD (1996) Model Parameters, Ottawa (2003) data $ [w] -- model changed [w] -- model changed [o] scaled deviance = 296.10 (change = -31715.) at cycle 4 [o] residual df = 274 (change = 0) [o] [o] ML Estimate of THETA = 1.918 [o] Std Error = ( 0.1540) [o] [o] NOTE: standard errors of fixed effects do not [o] take account of the estimation of THETA [o] [o] 2 x Log-likelihood = 221237. on 274 df [o] 2 x Full Log-likelihood = -3118. [o] [o] Scaled deviance is 296. on 274 d.f. from 275 observations [o] change is -31715. for 0 d.f. [o] [o] estimate s.e. parameter [o] 1 -0.1379 0.04408 1 [o] scale parameter 1.000 [o] [o] %PE %SE TSTATN CHI2 TARGET [o] 1 -0.1379 0.04408 -3.129 204.0 313.6 [o] Current model: [o] [o] number of observations in model is 275 [o] [o] y-variate T3 [o] weight * [o] offset OLVKT [o] [o] [o] probability distribution is defined via the macro NB_FIT. [o] link function is defined via the macro NB_LINK. [o] scale parameter is 1.000 [o] [o] linear model: [o] terms: 1  112  APPENDIX C – Sample Listing of CPZ Rankings  CPZs Urban 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20  Exposure Modelled  Exposure Measured  SocD Modelled  SocD Measured  TDM Modelled  TDM Measured  Network Modelled  Network Measured  1 222 210 743 523 202 2151 221 1041 1070 741 533 621 1640 1523 513 512 642 1100 602 521  2 222 1523 2151 523 210 1041 1070 2152 1640 800 1100 2442 201 2101 202 2621 1461 721 513 602  5 222 210 1523 743 1640 2611 2151 1100 1070 2190 1063 221 2152 523 3022 3012 2442 2732 202 2432  6 222 1523 2151 2152 210 1100 1640 1070 2442 201 523 1041 800 1461 2101 2621 230 622 2140 2272  9 210 222 1523 2151 743 1070 621 1640 523 1041 642 513 230 2732 2272 3012 1063 2442 1841 2160  10 222 1523 210 2151 2152 2621 1640 523 201 230 2272 1070 1701 1461 1841 1100 1430 2140 2442 800  13 1523 2151 210 743 222 1640 1070 2140 1861 2732 2152 3012 1032 1041 2611 3022 5043 2272 221 3152  14 1523 210 2151 222 2272 743 221 1070 1100 2140 2442 1430 1861 2101 2611 1041 800 1462 1640 201  113  APPENDIX D – Sample Collision Frequency Calculation  Zone #  VKT  TLKM (km)  AREA (Ha)  SCC  INTD  I3WP (%)  SIGD  Projected Collisions  4342  1552  21.67  94  0  0.37  91.4  0.016  67  Network Urban Modeled Model (Group# 13):  Total Collisions/3yr = 0.3687VKT 0.7362 e (0.432sigd + 0.2146intd – 0.003201i3wp ) =0.3687 * 15520.7362 * e (0.432*0.0016+0.2146*0.37 – 0.003201*91.4) = 67  114  APPENDIX E – Sample Variables Value Distribution 1, The CRD 90  160  80  120  Num ber of TA Zs  Number of TAZs  140  100 80 60 40  70 60 50 40 30 20  20  10  0 1-100  101-300 301-500 501-1000  10011500  15012500  0  >2500  0-2  2.1-4  4.1-6  6.1-8  8.1-10 10.1-12 12.1-14 14.1-16 16.1-20  140  120  120  100  Number of TAZs  Number of TAZs  >20  TLKM  VKT  100 80 60 40  80 60 40 20  20 0  0  0-10  0-2  2.1-5  5.1-10  10.1-20 20.1-30  30.1-50  10.1-50 50.1-100  100.1200  >50  160  180  140  160 Num ber of TAZs  Num ber of TAZs  300.1400  400.1500  >500  Drivers  POPD (population / ha)  120 100 80 60 40  Average = 75%  140 120 100 80 60 40 20  20  0  0 0-0.1  0.11-0.2 0.21-0.3 0.31-0.4 0.41-0.5 0.51-0.6 0.61-0.8  0-40  >0.8  41-60  61-70  71-80  81-90  91-100  I3WP (3-way intersection percentage)  INTD (Intersections / ha)  120  300  100  250 Num b er of T AZs  Nu m b er o f T AZ s  200.1300  80 60 40  200 150 100 50  20  0  0 0-30  31-50  51-60  61-70  71-80  LLKP (Local road percentage)  81-90  91-100  0-0.01  0.011-0.03  0.031-0.05  0.051-0.1  >0.1  SIGD (Signals / ha)  115  120  120  100  100  Num ber of TAZs  Num ber of TAZs  2, The City of Ottawa  80 60 40  80 60 40 20  20  0  0 1-500  501-1000 1001-2000 2001-3000 3001-5000 5001-6000  0-2  >6000  2.1-5  5.1-10  200  70  180  Num ber of TAZs  Num ber of TAZs  80  60 50 40 30 20  40.1-60  >60  160 140 120 100 80 60 40 0  0  0-10  0-2  2.1-5  5.1-10  10.1-20  20.1-30  30.1-50  10.1-50 50.1-100  100.1200  >50  200.1300  300.1400  400.1500  >500  Drivers  POPD (population / ha)  160  140  140  120 Num ber of TAZs  Num ber of TAZs  20.1-40  20  10  120 100 80 60 40  Average = 71%  100 80 60 40 20  20  0  0 0-0.1  0.11-0.2 0.21-0.3 0.31-0.4 0.41-0.5 0.51-0.6 0.61-0.8  0-40  >0.8  INTD (Intersections / ha)  41-60  61-70  71-80  81-90  91-100  I3WP (3-way intersection percentage)  180  120  160 Num ber of TAZs  100 Nu m b er o f T AZ s  10.1-20  TLKM  VKT  80 60 40 20  140 120 100 80 60 40 20  0  0 0-30  31-50  51-60  61-70  71-80  LLKP (Local road percentage)  81-90  91-100  0-0.01  0.011-0.03  0.031-0.05  0.051-0.1  >0.1  SIGD (signals / ha)  116  3, The GVRD 80  120  70 Num ber of TAZs  Nu m ber of T AZ s  100 80 60 40 20  60 50 40 30 20 10 0  0 1-1000  1001-2000 2001-3000 3001-4000 4001-5000  0-20 21-30 31-40 41-50 51-60 61-70 71-80 81-90  >5000  120  120  100 N u m b er o f T A Z s  Num ber of TAZs  140  100 80 60 40  80 60 40 20  20  0  0 0-2  2.1-5  5.1-10  10.1-20 20.1-30 30.1-50  0-100  >50  101-500  501-1000 1001-1500 1501-2000  90  180  80  160  70  140  Nu m b er o f T AZ s  N u m b er o f T A Z s  >2000  Drivers  POPD (population / ha)  60 50 40 30 20 10  Average = 53%  120 100 80 60 40 20  0  0  0-0.1 0.11-0.2 0.21-0.3 0.31-0.4 0.41-0.5 0.51-0.6 0.61-0.8 >0.8  0-40  INTD (intersections / ha)  160 Nu m b er o f T AZ s  140 120 100 80 60 40 20 0 0-30  31-40  41-50  LLKP (Local road percentage)  41-60  61-70  71-80  81-90  91-100  I3WP (3-way intersection percentage)  180  Nu m b er of T AZs  >90  TLKM  VKT  >50  200 180 160 140 120 100 80 60 40 20 0 0-0.01  0.011-0.03  0.031-0.05  0.051-0.1  >0.1  SIGD (signals / ha)  117  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Country Views Downloads
China 47 6
United States 31 0
United Kingdom 12 0
Canada 11 7
Australia 4 0
India 3 0
Brazil 3 11
Russia 2 0
Spain 2 0
Tunisia 1 1
Kenya 1 0
Poland 1 0
France 1 0
City Views Downloads
Unknown 20 12
Hangzhou 14 0
Ashburn 14 0
Beijing 8 0
Shenzhen 7 5
San Francisco 7 0
Brisbane 4 0
Vancouver 4 4
Mountain View 4 0
Nanjing 3 0
Southend-on-Sea 3 0
Seattle 2 0
Toronto 2 2

{[{ mDataHeader[type] }]} {[{ month[type] }]} {[{ tData[type] }]}
Download Stats

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.24.1-0063125/manifest

Comment

Related Items