Incorporating Safety into Long-Range Transportation Planning (2006)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Incorporating Safety into Long-Range Transportation-Planning APPENDIX C SAFETY TOOLS INTRODUCTION A variety of tools for safety analysis exist that can be used at various levels: some at the regional level, others on a project level, as well as some tools for corridor level safety planning. The use of these tools is essential to allow for the integration of safety into long range transportation-planning. This appendix focuses on the safety tools that are widely available for conducting safety analyses as well a new tool— safety forecasting at the planning level—which has been developed as part of this research. The aim of this chapter is to provide individuals involved in decision-making and those involved in long-range transportation-planning with enough information to make an appropriate selection of tools for integrating safety into planning in their region or state. The tools described in this appendix require data with varying levels of detail, ranging from TAZ level information in the Planning Level Safety Prediction Model to fairly detailed input in the IHSDM. The tools also vary in terms of purpose: the Planning Level Safety Prediction Model is used to perform safety prediction by TAZ area (pro-active), Intersection Magic analyzes historical accident data, etc. Finally, the tools vary in their required levels of expertise. All of these important characteristics of available analytical tools are described in this appendix. Analytical tools that are likely to be applicable to a wide audience (i.e., all states) are provided with examples to illustrate an application, whereas tools that are more limited (i.e., serve only a few states) are merely described. The more limited tools are provided mainly to demonstrate the kinds of analytical tool development efforts that are possible through cooperation between states and agencies within the state (DOT, university, etc.). In all cases references are provided so additional information can be found regarding the tools and their successful application. OVERVIEW Exhibit 53 lists available analytical safety tools by name, primary purpose, level of detail needed to apply the tool, and required expertise to apply the tool. This initial overview is followed by subsections providing more detailed descriptions of each of the tools. Appendix C: Safety Tools 91

Incorporating Safety into Long-Range Transportation-Planning SUMMARY OF TOOLS Exhibit 53: Purpose, level of detail, and required expertise for tools available to incorporate safety into long-range transportation-planning TOOL PRIMARY PURPOSE LEVEL OF DETAIL REQUIRED EXPERTISE Arizona Local Government Safety Project Analysis Model (LGSP) Reactive: Generate list of most hazardous locations using user-specified criteria, Provides summary data for other sites for use in Before-After studies, Ranking alternatives by benefit-cost ratios Generation of project details to supplement HES eligibility applications High: Accident data (detail for vehicle, driver, and passengers) Roadway data: grade, number of lanes, lane width, control type, road alignment, etc. Environmental: traffic volume, weather, terrain, etc. Basic computer skills, familiarity with Microsoft Access desirable. Before-After Studies as described in “Observational Before-After Studies in Road Safety”, Hauer (1997) Reactive or Proactive tool to assess the safety effectiveness of a given improvement or countermeasure Moderate: Accident data Geometric, traffic, weather, and human attributes Ranges from fundamental algebra and statistical knowledge, to the more complicated empirical Bayes (EB) approach Crash Outcome Data Evaluation System (CODES) Reactive: Generate medical and financial outcome information related to motor vehicle accidents High: Accident Data Emergency Service Data Hospital Inpatient Data Death Certificate Data Vehicle Identification Number Data Trauma Registry Data Statistical analysis, use of the CODES linkage software Interactive Highway Safety Design Model (IHSDM) Pro-active and reactive. Assess the safety of two-lane roadway designs (model for multi-lane roadways in development) High: General data (terrain, volumes, functional classification, speed) Horizontal elements (curves, station equations, intersections) Vertical elements (curves) Cross-sectional data (cross- slopes, pavement type, shoulder detail) Lane dimensions Roadside elements (detailed) Roadway data (accident data, bridge elements, decision sight distance) Basic understanding of geometric design concepts, ability to input data in Microsoft Windows environment through conversion of detailed geometric designs from other software or comma-separated file format (*.csv) Intersection Magic Reactive: Analysis of accident data Moderate: Accident data Basic computer skills Appendix C: Safety Tools 92

Incorporating Safety into Long-Range Transportation-Planning TOOL PRIMARY PURPOSE LEVEL OF DETAIL REQUIRED EXPERTISE Level of Service of Safety (LOSS) Qualitative assessment of safety performance of existing facility planning major corridor improvements Moderate: Accident Data Geometry of existing roadway Basic understanding of traffic engineering and computer skills Multimodal Transportation- planning Tool (MTPT) GDOT Analysis of operational performance of transportation system, includes analysis of transit, bicycle and pedestrian plans. High: Road Characteristics Database Accident Database Statewide modal transportation plans Basic computer skills Pedestrian and Bicycle Accident Analysis Tool (PBCAT) Reactive: Development and analysis of pedestrian and bicycle related accident database, assist in the selection of countermeasures Moderate: Accident data with geometry, time, weather, location, age, gender, subject actions, and other attributes Basic computer skills Pedestrian Safety Guide and Countermeasure (PEDSAFE) Reactive: Analysis of pedestrian related accident data Assist in the selection of countermeasures or treatments: engineering, education, or enforcement Moderate: Accident data Basic computer skills Roadside Safety Analysis Program Pro-active and reactive. Cost effectiveness analysis to assess effectiveness of roadside safety improvements Moderate: Accident Data Geometry of existing roadway Basic understanding of traffic engineering and Monte Carlo simulation technique. SafeNET Pro-active and reactive, Traffic accident prediction for intersections and sections Differs depending on purpose: Basic: traffic flows averaged over day More detailed: vehicle flow, pedestrian flow, site characteristics, specific geometric features, junction turning flows, and other design features Basic traffic engineering, accident modeling, 4-step planning models SafetyAnalyst Reactive but some pro- active applications: Analysis of accident data: by site, by section or systemwide Moderate: Accident data Geometric, traffic, weather, and human attributes Statistical analysis & basics of traffic engineering Appendix C: Safety Tools 93

Incorporating Safety into Long-Range Transportation-Planning TOOL PRIMARY PURPOSE LEVEL OF DETAIL REQUIRED EXPERTISE TRansportation ANalysis and Simulation System (TRANSIMS) Pro-active: Evaluate transportation alternatives and reliability to determine benefits and adverse effects; predict volumes along the network : are used as input in other tools High: Census data of household surveys Origin/Destination matrices Transportation network data for major intersections, and Other information used to produce pseudo-activities for trip generation Transportation Network Modeling, Software: TRANSIMS software, Oracle, C++ programming language, or ArcView Avenue programming language Forecasting Accidents at the Planning Level Proactive and Reactive. Prediction of accidents by Traffic Analysis Zone (TAZ) Moderate to High: Accident data Census data Bicycle and transit facility locations Functional Classification of road network Statistical Analysis and the use of GIS software: expertise required for GIS analysis will depend on the nature of existing GIS information and databases. Appendix C: Safety Tools 94

Incorporating Safety into Long-Range Transportation-Planning ARIZONA LOCAL GOVERNMENT SAFETY PROJECT ANALYSIS MODEL (LGSP) Exhibit 54: Summary of the LGSP tool LGSP Vendor name and address: Arizona Department of Transportation, 206 South 17th Avenue, Phoenix, Arizona 85007. Brief description of transportation safety applications: The Arizona LGSP is a useful tool to facilitate site identification and safety project selection by local jurisdictions and planning organizations. It could feasibly be adapted for use with non-Arizona databases; however, alternative tools might be selected for conducting ‘hot spot’ identification in non-Arizona states. Based on a database containing information regarding accidents and highways, it can automatically generate a list of the most hazardous locations in terms of user-defined parameters (e.g., alcohol involvement, location reference, distance, weighting method, etc.). It provides not only the total and annualized accident details, but also those details limited to a specific subset. In addition, for the sake of facilitating before-and-after comparisons and estimating regression-to-the-mean potential at a given site, the Arizona LGSP model can create a comparison site list report containing a summary of additional sites in a jurisdiction that have similar characteristics to the site location being analyzed. Finally, the model’s project evaluation routine allows multiple projects to be analyzed simultaneously, with minimum run time, providing opportunities to revise site selection and project characteristics throughout the programming process. These project alternatives are ranked by benefit-cost ratios and project details are formatted to supplement HES eligibility applications. With these features, the Arizona LGSP model supports local governments in Arizona to address their highway safety needs on a timelier basis, and ensure that more attention is directed at the most hazardous locations, thereby improving the overall safety of the roadway system. Types and sources of data needed: To develop appropriate parameters for implementation strategies, a substantial body of data is required to support this model. These data can be divided into the following groups: Human attributes: number of injuries and fatalities, age, gender, alcohol involvement, driver state, seat belt use, etc. Vehicle: number of vehicles involved, vehicle type, axles, plate number, etc. road: grade, number of lanes, lane width, control type, road alignment, etc. Environmental conditions: traffic volume, weather, terrain, etc. Expertise required: The Arizona LGSP model was created in MS ACCESS 97. Due to a user-friendly interface and automated processes for site identification and improvement strategies selection, there is no need for special knowledge to run it; however, familiarity with MS ACCESS is a desirable. Hardware requirements: Windows work station. Because the model consists of a self-contained query and reporting database, and a supplemental database of accident records on CDROM, running the model requires a CDROM drive or network access and approximately 32Mb RAM and 100Mb hard disk space. Due to the computational intensiveness of this model, a higher-speed processor is recommended. Example application of tool: Exhibit 55 shows that the model can provide the users the list of hazardous sites and evaluations of project alternatives. For example, by clicking the Button 1, the inputs form shown in Exhibit 56 appears. Once the user inputs have been specified, the GET RESULTS button will run the prioritization procedure and return the priority list report containing the 25 most hazardous locations. Appendix C: Safety Tools 95

Incorporating Safety into Long-Range Transportation-Planning Exhibit 55: Arizona LGSP analysis set-up window Exhibit 56: Arizona LGSP analysis parameters window Appendix C: Safety Tools 96

Incorporating Safety into Long-Range Transportation-Planning BEFORE-AFTER STUDIES AS DESCRIBED IN “OBSERVATIONAL BEFORE- AFTER STUDIES IN ROAD SAFETY”, HAUER (1997) Exhibit 57: Summary of before-after studies tool as described in Hauer (1997) BEFORE-AFTER STUDIES Vendor name and address: Although various transportation agency personnel apply before-after analysis methods to estimate the effectiveness of safety countermeasures, one detailed before-after methodology has emerged as “state of the practice”. The currently accepted method is described in Hauer, (1997, Observational Before After Studies in Road Safety, Pergamon). Brief description of transportation safety applications: This particular tool to assess the effectiveness of safety strategies or countermeasures that have been implemented in a state or region. Example countermeasures might include shoulder widening, signalization, culvert installations, pedestrian crossing improvements or installations, and the addition of bicycle lanes. Hauer develops and describes a detailed methodology which defines target accidents for which the before-after study will be applied. The book also provides guidance on various refinements of a before-after study, including the “Naïve before-after study”, the “comparison group” method, the multivariate method, and the most advanced Bayesian before-after method. Each of these refinements are meant to deal with a variety of shortcomings that arise in before-after studies of road safety. A synthesis on statistical methods in highway safety analysis presented a more elaborate statistical treatise on conventional before-after studies [Griffin and Flowers, 1997]. The report describes six different evaluation designs to determine the impact of selected highway strategies on the accident record. The six evaluation designs covered in the report are: a) simple before and after design, b) multiple before and after design, c) simple before and after design with yoked comparison, d) multiple before and after design with yoked comparison, e) simple before and after design with yoked comparison and check for comparability and f) multiple before and after design with comparisons and check for comparability. Types and sources of data needed: For this type of analysis, accident data with geometric, traffic, weather, and driver behavior attributes are necessary. The data requirements also depend on the type of treatments and accidents of interest, and can become quite demanding in a thorough and reliable analysis. Expertise required: The expertise needed for this tool ranges from simple algebra and statistical knowledge to work with before-after comparisons, to the more complicated empirical Bayes (EB) approach. Some background in statistical methods is desirable. Hardware requirements: Any computer offering spreadsheet capabilities will support the application of this methodology. Example application of tool: An example of the before and after study tool is illustrated below. The effectiveness of a change is determined by comparing the change in the value of the performance measure (e.g., frequency or rate of accidents) given the change with what would have occurred without the change. This approach is appropriate whether one is evaluating the application of strategies at a particular site, or applied to different accident characteristics (e.g., driver types). The biggest challenge in this effort is estimating what the change would have been if there had not been a treatment. It is especially difficult because all other factors do not remain equal in the after period, including environmental, traffic, and other factors. Exhibit 58 demonstrates the use of control sites to help address the critical question of what would have happened if no treatment had been made (which is not observed since the site received the treatment). Control sites are locations (or population groups) not receiving a treatment that are considered sufficiently similar in character to the one(s) being treated that any change in performance over the before-and-after time frame can be assumed to be natural maturation of the phenomenon. The period between before and after measurements is shown in Exhibit 58 by the vertical bar. This time period can be fairly short (e.g., the time to install a countermeasure) to a much longer time period (e.g., two years or more after Appendix C: Safety Tools 97

Incorporating Safety into Long-Range Transportation-Planning implementation). Performance measurements are taken periodically (e.g., monthly or annually) in the before and after periods for both the control and treated sites. Exhibit 58: Depiction of before-after study using control sites Exhibit 58 suggests that one could estimate the change in performance at the control sites by using averages for the before and after period. The effect of treatment on performance (e.g., fatal crashes, pedestrian crashes, etc.) is estimated as the difference between expected (predicted) and actual crashes. It also shows that individual site values may be used to perform a regression or trend analysis. More recent developments suggest that use of the Empirical Bayes approach may be more appropriate in many instances for estimating the expected value in the after period— due to the often present regression to the mean effect caused by site-selection bias (sites are selected for treatment due to observed high crash counts—part of which may be due to random fluctuations). While this figure merely shows one aspect of the before-after methodology, it demonstrates, in general, the methodology for assessing countermeasures. Further information is provided in Hauer, 1999. Appendix C: Safety Tools 98

Incorporating Safety into Long-Range Transportation-Planning CRASH OUTCOME DATA EVALUATION SYSTEM (CODES) Exhibit 59: Summary of the CODES tool CODES Vendor name and address: U.S. Department of Transportation, National Highway Traffic Safety Administration. Brief description of transportation safety applications: CODES was designed to generate crash statistics that merge medical and financial outcome information with motor vehicle accidents. The information is used to estimate costs associated with crashes under a variety of circumstances (e.g., rollover crashes, pedestrian crashes, tire blowouts, etc.). The state maintained databases in turn help to conduct analysis toward the prevention of deaths and injuries, the reduction of injury severity and health care costs, and improvement in the basis for decisions related to highway traffic safety investments. CODES is perhaps the most valuable tool available (regarding motor vehicle crashes) to state and federal legislators, since it is the only known software capable of linking accident costs with accidents in a rigorous and defensible way. Types and sources of data needed: The main aim of CODES is to link various data related to traffic accidents. The data necessary for CODES includes accident data, emergency medical service (EMS) data, hospital inpatient data, death certificates, vehicle identification number data, and trauma registry data. Expertise required: Knowledge about statistical data analysis and the CODES linkage software. Training in CODES software is absolutely necessary to use this software, and expertise is available already in many states in the U.S.. Hardware requirements: Windows work station running MS Access Example application of tool: The CODES is a software system that enables probabilistic linkage of accident data from various sources. The data sources usually consist of police-recorded accidents, hospital inpatient data, emergency medical services data, trauma registry data, and death certificate data. Probabilistic linkage enables the linking (association) of accident records with the highest probability, and is needed because accident records lack a unique identifier throughout the accident process. An example application is to link all 2001 motor vehicle accident data in the state of Arizona with emergency medical service records, hospital information, trauma registry information, and death certificate data. Then, analysis can be conducted to determine what the costs in the state of Arizona are associated with safety restraint use violations, lack of child-seat usage compliance, or motorcyclists not wearing helmets. Similarly, an analysis can be conducted to estimate the impact of emergency response times on fatality probabilities. Finally, the types of injuries and associated costs of SUV rollover accidents can be examined, as well as other analyses. Appendix C: Safety Tools 99

Incorporating Safety into Long-Range Transportation-Planning CRITICAL ANALYSIS REPORTING ENVIRONMENT (CARE) Exhibit 60: Summary of the CARE tool CARE Vendor name and address: CARE Research and Development Laboratory (CRDL), Department of Computer Science, University of Alabama, Box 870290, Tuscaloosa, AL-35487-0290. This is free software and can be downloaded from internet, and is set up for the analysis of crashes in the state of Alabama. Brief description of transportation safety applications: CARE is a data analysis software package designed for problem identification and countermeasure development. CARE can be used to retrieve subset of any specific interest from the entire crash dataset in a few seconds, providing the feedback necessary to allow the user to make subsequent queries based on preliminary results. The user can apply CARE to get started immediately without having to do any programming or sophisticated analysis. The information mining capability (IMPACT) of CARE generates information through the comparison of subsets of data (e.g., weather-related vs. non-weather-related cases), and graphically demonstrates possible potential areas for countermeasure implementation. In addition to its capability of identifying high crash locations CARE supplies corrective measures in terms of countermeasure selection. Another attractive feature of CARE is its ability to generate collision diagrams through the popular software “Intersection Magic” that was incorporated into it. The reports produced by CARE can be directly exported to Microsoft Office products such as Word and Excel. Types and sources of data needed: It is necessary to have the crash data in a specific format to allow CARE to perform the analysis. Users can transform existing datasets to the required format by following the easy steps described in the software manual. If needed CRDL staff can prepare the dataset for a fee. Expertise required: Although CARE uses advanced analytical and statistical techniques to generate valuable information from the data, users do not have to be familiar with any special knowledge. This user-friendly software can be used efficiently by just following step-by-step menus outlined on screen. Interpretation of the results, however, requires an understanding of the crash database and associated variables. Hardware requirements: CARE can be used on a desktop or through the internet. The CARE desktop operates in the Microsoft Windows environment (including Windows 95, 98, NT, 2000, ME, XP). Example application of tool: CARE prepares a variety of canned reports. An example of such a report is shown in Exhibit 61. In this example the user selected a frequency report based on the time of day. For further examples, refer to the software website: http://care.cs.ua.edu. Exhibit 61: Example output from CARE Appendix C: Safety Tools 100

Incorporating Safety into Long-Range Transportation-Planning INTERACTIVE HIGHWAY SAFETY DESIGN MODEL (IHSDM) Exhibit 62: Summary of the IHSDM tool IHSDM Vendor name and address: Federal Highway Administration / Turner-Fairbank Highway Research Center. Brief description of transportation safety applications: Currently available for testing and evaluation, the IHSDM can assess the safety implications of two-lane roadway designs. A model to evaluate multi-lane roadways is in the development stages. The 2003 (2-lane version) IHSDM consists of 5 modules: the Accident Prediction Module, Design Consistency Module, Intersection Review Module, Policy Review Module, and Traffic Analysis Module. The multi-lane IHSDM (yet to be released) will also feature a Driver/Vehicle Module, which will consist of a Driver Performance Model linked to a Vehicle Dynamics Model. The IHSDM can review designs in both metric units and U.S. customary units. Types and sources of data needed: The data required for the IHSDM are numerous: General data Terrain type (level, rolling, mountainous, null) Volumes (daily and hourly) Functional classification Speed (design, 85th percentile, and posted) Horizontal elements Curves Station equations (if any) Intersections Vertical elements Curves Cross-sectional data Cross-slopes Pavement type Shoulder (slope, width, material, category) Lane dimensions Roadside elements (slopes, ditch, obstruction offset, bike facilities, driveway density, hazard rating) Other relevant roadway data: Accident data (based on accident records) Bridge elements Decision sight distance Design Vehicle Expertise required: Basic understanding of geometric design concepts and ability to input data into the Windows-based interface, either through detailed geometric elements or by conversion of highway design data to *.csv (comma-separated) format. Hardware requirements: Windows work station. Example application of tool: The IHSDM is useful for determining the safety implications of either an existing or planned roadway alignment/configuration. For instance, a two-lane roadway with extensive curvature and other geometric intricacies can be evaluated for a variety of issues, including compliance with federal policy (such as the 1994 or 2001 AASHTO Policy, metric or English units), its expected accident rates or frequencies, how well the roadway design meets with driver expectations, the policy consistency and operational performance of intersections, and various aspects of traffic analysis. The IHSDM can also be used to present graphical representations of an analyzed roadway showing plan, profile, and cross-sectional views. Appendix C: Safety Tools 101

Incorporating Safety into Long-Range Transportation-Planning INTERSECTION MAGIC Exhibit 63: Summary of the INTERSECTION MAGIC tool INTERSECTION MAGIC Vendor name and address: Pd’ Programming, Inc 725 Aegean Drive Lafayette, CO 80026 Phone Number: Main number: (303) 666-7896 R&D: (303) 666-6035 Fax: (303) 666-7347 Brief description of transportation safety applications: Intersection Magic is efficient software for accident record analysis. From the accident database, it generates automated collision diagrams, pin maps of high accident locations, high accident location lists, frequency reports, and much more. A transportation engineer can easily extract a particular type of accident say, left turning accident, or angle accident at the intersection, and that accident type will be displayed on the screen with the total number of such accidents. Similarly, to locate the accident-prone intersections, pin maps can be generated, along with the name of the intersections and frequency of accidents. Besides these spatial features, the temporal features such as time, month, or year of accidents can also be displayed with the help of presentation graphics. With all these advantages, Intersection Magic is no doubt a useful tool for traffic engineers and planners to identify hazardous locations, which could then be treated to enhance safety. Types and sources of data needed: For the purpose of analyzing various attributes of accidents, the database can be developed within this software. In addition to this, Microsoft Excel database can also be used. Expertise required: This is user-friendly software and does not need any special knowledge to use. The software is linked with Arc GIS. This linkage helps in displaying pin maps from high accident location reports, traffic volume maps, corridor line maps from sliding spot reports, comparisons of various kinds of accidents, and so on. However, plotting these special features does not require learning GIS software. Hardware requirements: Windows work station. Example application of tool: Refer to the examples provided below. Intersection Magic (IM) is software that is used to perform accident analysis based on historic accident data at intersections. It generates automated collision diagrams, pin maps of high accident locations, and frequency reports. The software is primarily used as a reactive tool to analyze safety. IM’s powerful query system helps an analyst to investigate various spatial and temporal attributes of each accident occurring at an intersection along with a schematic representation of each. This special feature of IM enhances the analysis power, as visual representation of data greatly improves the ability of the individual to evaluate the dataset. While an intersection can be identified as a hot spot, based on high accident statistics, IM can provide a powerful tool to the professional to analyze the intersection to determine the nature of the accidents occurring at the site. In these cases, a quick analysis in IM would give the analyst all possible information associated with the accident; for example, the time of accident occurrence, the maneuver before the accident occurred, the severity of the accident, and much more. Exhibit 63 provides a summary of the IM Tool. Appendix C: Safety Tools 102

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools Particular Advantages In addition to the existing strong analytical capabilities, IM can also be linked to ESRI’s Arcview (Geographic Information System software). This enhances the analytical ability of IM as it brings the analysis tools and benefits of the GIS environment to the analysis process. The web browser facility of IM enables the user to perform analysis via the inter- or intranet and to use all of the analysis capabilities of the software from any computer on the network. Updates are applied automatically, the results of an analysis is automatically saved until the user decides to discard it, and all reports, filters, and charts are stored centrally. Examples of Analysis using IM To demonstrate the various accident analysis features available in IM, a set of analysis examples are provided using accident data from City of Chandler, Arizona. It should be noted that this software description do not intend to substitute for the IM manual; it merely intends to demonstrate by example some of the analysis capabilities of the software. To access a user guide for the software, consult the Intersection Magic User manual or visit HTUwww.pdmagic.comUTH. The identification of accident hot spots in a jurisdiction is a critical element of a hazard elimination program. IM enables the user to identify the hot spots, to evaluate the accident statistics, and to identify possible countermeasure treatments. Identify locations In this example, a list of intersections is generated with at least one accident. This is done by selecting: Select Reports / Listings / High Accident Locations and by following the steps to extract the information from the database: • First the number of intersections in the dialogue box are specified, say in this case 100 intersections with at least an accident are selected. • In the same dialogue box, the user has the option to specify a date range, sorting criteria of accidents such by counts/rates and so on. • There are also options as to display the volume, rate, counts and other records for the intersection. This procedure generates a list of intersections with accident counts that can be compared to each other, used to identify ‘high risk’ locations (most likely along with additional numerical analysis), or to trigger field audits. Two screen captures from the analysis showing list and diagram are shown in Exhibit 64 and Exhibit 65 respectively. An analyst could also target specific types of accidents, such as accidents that occurred during night time or rear end crashes. This is accomplished by clicking on the Diagram/Settings tab and then selecting the desired accident labels. Accident labels can be made visible or not; however, a couple of aspects should be kept in mind: • If labels are visible, the diagram as shown in Exhibit 65 will include specific information. • If the number of accidents becomes large, the diagram becomes congested and the labels result in a cumbersome diagram rather than improving the analysis capability. In this case, the various attributes of accidents are represented by different colors. For example, bad weather accidents such as during rain in red, accidents during dawn/dusk in blue, and so on. This is done by selecting the Intersection Magic has a powerful query system to investigate spatial and temporal attributes of each accident along with a schematic diagram. 103

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools option Diagram/Accidents/Toggle highlighter and then choosing specific colors to represent typical accident types. Exhibit 64: Screen grab for top 100 Intersections with at least one accident Exhibit 65: Screen grab for Alma School & Warner Road intersection accident diagram 104

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools Filters help the analyst to extract specific attributes of interest. For example, after getting the intersection list from the previous step, a filter is used to identify intersections that have an associated accident rate above a certain value, or accidents resulting in incapacitating injuries or fatalities. To create a filter, to the user selects “record filter” and then “edit filter” to write the filter expression. Once the filter expression is given, the conditions to qualify the accidents to be included in the list/diagram is complete. For example: • Suppose the analyst wants to know about the top 100 intersections with at least 1 incapacitating/fatal accident. Hence, a filter is created that extracts only those intersections in the database as shown in Exhibit 67. The list yields 92 intersections; so there are 92 intersections in the city of Chandler where at least one severe/fatal accident occurred during the first 11 months of 2003. The list also shows the count of such accidents in descending order, helping to characterize the safety problem at different intersections. Identification of Problems Next, the intersections on the top 100 list are examined to identify possible underlying problems associated with each. To accomplish this, the analyst completes the following: • A schematic diagram is created of the accidents at each intersection to visualize the accidents occurring each of the sites. • During an analysis of the diagram, the analyst can click on each accident position (precisely the junction of the two arrows indicated as accident) to get access to all available details of the accident record about the circumstances in which the accident occurred. Exhibit 66: Screen grab for Evergreen St & Warner Road intersection accident diagram with labels IM’s filter tool helps analyst extracting specific attribute of accident from the database. 105

Incorporating Safety into Long-Range Transportation-Planning Exhibit 67: Screen grab for top 92 Intersections with at least one severe accident Filters can also be saved for future analyses. All generated diagrams can also be saved for future reference. Another useful command is Sliding Spot Listing. It provides the user with a means of locating high accident locations on roads with hundred block or milepost data. Specifically, for each specified road, this function examines the entire length and sorts the high accident locations into the list of all roads specified for processing. Appendix C: Safety Tools 106

Incorporating Safety into Long-Range Transportation-Planning LEVEL OF SERVICE OF SAFETY (LOSS) Exhibit 68: Summary of the LOSS tool LOSS Vendor name and address: National Cooperative Highway Research Project 17-18(4) Brief description of transportation safety applications: The concept of Level of Service of Safety was first introduced by Kononov and Allery (2003). As an effort to develop Highway Safety Manual (HSM) under the NCHRP project 17-18(4), they have developed a detailed procedure to identify the existing level of service of safety for highways. According to the authors, the concept of level of service uses qualitative measures that characterize the safety of a roadway segment in reference to its expected performance. They also explain that the level of safety predicted by the Safety Performance Function (SPF) will represent the normal or expected number of accidents at a specific level of AADT, and the degree of deviation from this norm or expected value can be stratified to represent specific levels of safety. In the case of roadway safety, both frequency and severity are important. Hence it is necessary to calibrate two kinds of SPFs, one for the total number of accidents, and another for injury and fatal accidents only. When the magnitude of the safety problem is assessed using the LOSS methodology, it is done so from frequency and severity standpoints. Four Levels of Service of Safety (LOSS) were proposed by Kononov and Allery (2003), these are: LOSS-I - Indicates low potential for accident reduction LOSS-II- Indicates better than expected safety performance LOSS-III - Indicates less than expected safety performance LOSS-IV – Indicates high potential for accident reduction The LOSS concept is widely used by the Colorado Department of Transportation for system-level planning, as well as project scoping. Kononov and Allery assert that this approach will bring about badly needed consensus in the transportation engineering profession on the subject of the magnitude of safety problems for different classes of roads. In addition, the classification will also make it possible to take the following critical steps in effective and responsible resource allocation directed at improving road safety: Qualitatively describing the degree of safety or un-safety of a roadway segment Effectively communicating the magnitude of the safety problem to other professionals or elected officials Bringing the perception of roadway safety in line with reality of safety performance reflecting a specific facility Providing a frame of reference for decision making on non-safety motivated projects (resurfacing or reconstruction, for instance) Providing a frame of reference from a safety perspective for planning major corridor improvements. Types and sources of data needed: Information about the accident history and the geometry of the existing roadway, as well as exposure (traffic volumes or flows) information to support the development of SPFs. Expertise required: No special expertise is needed, although basic understanding of statistics is desirable. Hardware requirements: Windows workstation with spreadsheet and/or database capabilities. Example application of tool: To identify the existing LOSS, a highway is divided into a number of segments. Accident counts per segment are then obtained from which accidents per mile per year are identified. After this process, the existing LOSS is easily identified using the SPF graphs and the existing Annual Average Daily Traffic (AADT). A similar procedure can be used for forecasting future accidents if the traffic engineer has predicted traffic volumes. Kononov and Allery (2004) have shown an example in their recent paper to illustrate the appropriate use of LOSS as a proactive tool to predict safety. A more detailed discussion of LOSS is described in the following section. Appendix C: Safety Tools 107

Incorporating Safety into Long-Range Transportation-Planning Exhibit 68 briefly introduces the concept of Level of Service of Safety (LOSS) that was developed as a part of Highway Safety Manual (HSM) under National Cooperative Highway Research Program (NCHRP) Project 17-18 (4). Although the concept of the LOSS is new, it is similar to the Level of Service concept used in the Highway Capacity Manual (HCM). The LOSS reflects the performance of a site, project, or facility in terms of expected accident frequency and severity at a specific level of AADT. The LOSS concept was developed by Jake Kononov and Bryan K. Allery. In their most recent papers (2003, 2004), the concept of Safety Performance Function (SPF) is discussed in great detail as well as the development of LOSS using SPFs. Brief descriptions of these terms is provided here; however for additional details, readers are encouraged to refer the previously mentioned papers. Information about HSM is also available at www.hsm.fhwa.com. LOSS is very similar to that of Level of Service in Highway Capacity Manual. However, LOSS concept is intended to reflect the performance in terms of expected accident frequency and severity at a specific level of AADT, as opposed to the measure of delay, conventional in the Level of Service Analysis. Safety Performance Function (SPF) The SPF is simply a function that relates expected crash frequencies to exposure. Different facilities and traffic control situations deserve their own SPFs; for examples SPFs are appropriate for signalized intersections, stop controlled intersection, 2-lane highway segments, 4-lane highway segments, etc. Typically these relationships are fitted using Negative Binomial or Poisson regression models. In many cases these relationships are not straight lines (linear functions), and as a result provide evidence against the use of accident rates for assessing the safety performance of sites. Details about dataset preparation and model fitting for the development of the Safety Performance Functions (SPF) are described by Kononov and Allery (2003), whereas discussion about the non-linearity of SPFs is provided in Hauer (1997). Level of Service of Safety (LOSS) LOSS is developed using the concept of the SPF. The LOSS concept uses qualitative measures that characterize safety of a roadway segment in reference to its expected performance. If the level of safety predicted by the SPF represents normal or expected number of accidents at a specific level of AADT, then the degree of deviation from this expected count can be stratified to represent specific levels of safety. Kononov and Allery calibrated two kinds of SPFs, one for the total number of accidents and another for injury and fatal accidents. There are four level of service for safety as follows: LOSS-I - Indicates low potential for accident reduction LOSS-II- Indicates better than expected safety performance LOSS-III - Indicates less than expected safety performance LOSS-IV – Indicates high potential for accident reduction To illustrate how LOSS is applied in practice, a case study conducted by Kononov and Allery (2004) in the Denver Metro Area is presented. In this case study, the authors examined a segment of a major 6 lane urban freeway in the Denver Metropolitan Area, shown in Exhibit 69. Appendix C: Safety Tools 108

Incorporating Safety into Long-Range Transportation-Planning Exhibit 69: Project area map, Denver Colorado metropolitan area corridor study The analysis begins with a LOSS analysis that reflects average safety performance of the section for three years (2000-2002), followed by a diagnostic investigation of accident causality. A running average of the 3 years was used to smooth out peaks related to annual fluctuations in accident frequency. The results of the LOSS total frequency analysis of the urban 6-lane freeway in the study area are shown in Exhibit 70, while the results of the LOSS injury and fatal only analysis are presented in Exhibit 71. The models shown in Exhibit 70 and Exhibit 71 reflect 14 years of accident data. The models represented by dark blue and dark red curves in the figures represent expected crashes predicted by the Negative Binomial regression models estimated using the observed crash data. Noteworthy observations are as follows: • Segments #1, 3, 4 and 5 performed more or less as expected for an aging urban freeway. Observed frequency and severity are in the LOSS-II and LOSS-III range. • Segment #2, however, showed highly undesirable safety performance in the high range of the LOSS-IV for both frequency and severity, which suggested a high potential for accident reduction. At this stage of the diagnostic investigation, the researchers concluded that the site experienced significantly more accidents than expected for some unknown Appendix C: Safety Tools 109

Incorporating Safety into Long-Range Transportation-Planning reason. Based on this observation, they followed up by examining the accident type distribution observed on the study segment over a period of three (3) years. This examination led to the distribution presented in Exhibit 72. The distribution shows a high percentage of rear-end collisions followed by sideswipes in the same direction on the segment under study. Exhibit 70: LOSS-injury and fatal accident frequency in the study area . Exhibit 71: LOSS-total accident frequency in the study area Rear-end collisions represented 73% of the accident types. This was higher than the expected 44.5% level, which is typical for 6-lane urban freeways. Same direction sideswipe accidents represented 18% of the accidents, which is also higher than the expected 12.6% for similar types of segments. Based on these findings, the authors concluded that elements in the roadway environment possibly triggered a deviation from the random process of accident occurrence in the direction of reduced safety. More specifically, it triggered rear-end and sideswipe collisions. Appendix C: Safety Tools 110

Incorporating Safety into Long-Range Transportation-Planning Subsequent plan reviews and site visits by the researchers revealed the existence of a highly constrained weave type C section within segment #2 in the southbound direction (Exhibit 73 and Exhibit 74). Specifically, vehicles entering the freeway on the left side were attempting to exit on the right side while crossing three highly congested through lanes of traffic and one auxiliary lane over a very short distance. Operational Level of service (LOS) analysis procedures outlined in the Highway Capacity Manual (TRB) showed a LOS-F in the weaving section in the southbound direction. In this case, a traffic operational problem related to the highly constrained weave type C translated into a significant safety problem manifested by the high frequency and severity of rear-end and sideswipe collisions. They concluded that the high number of rear-end and sideswipe accidents is the reason behind the highly elevated accident frequency and severity on this segment. They recommended the removal of the type C weave by reconfiguring the interchange and constructing a flyover ramp or a tunnel to facilitate the conflicting vehicular movements. Exhibit 72: Breakdown by accident type in the study area Appendix C: Safety Tools 111

Incorporating Safety into Long-Range Transportation-Planning Exhibit 73: Wave type C- LOSS analysis for total accidents Researchers identified a highly constrained weave type C section within segment #2 in the southbound direction which contributed to the higher number of rear-end and sideswipe collisions Exhibit 74: Wave type C- LOSS analysis for injury and fatal accidents Appendix C: Safety Tools 112

Incorporating Safety into Long-Range Transportation-Planning The researchers concluded that removing the weave type C section from segment #2 will improve safety performance to an average 6-lane freeway segment in an urban environment (the comparison group mean). To determine the expected accident frequency and severity they used the SPF graphs and estimated that at the current AADT level of 188,000, approximately 90 accidents per mile are expected, of which approximately 20.5 collisions will result in injuries or fatalities. The improvement, they estimated, would result in a reduction of approximately 88 accidents, including 19.5 injuries, during the first year following construction. It is important to note that within segment #2, each injury accident results in injuries to 1.3 people on average. This suggests that removal of the type C weave section could potentially prevent injuries to 25 people in the first year following construction. Exhibit 73 and Exhibit 74 graphically illustrate the anticipated accident reduction resulting from the elimination of the constrained weave in the southbound direction. In addition to the safety improvements, a corridor expansion from 6 to 8 lanes was also planned and the researchers estimated that the improvement would prevent 34 accidents, 11.5 of which result in injury. This example illustrates how safety can be explicitly addressed while planning long-range major transportation improvements in urban corridors. The analysis is achieved with fairly simple tools, and with some basic knowledge in the estimation of statistical models (Negative Binomial or Poisson Regression), some spreadsheet functions, and as always sound engineering judgment. Appendix C: Safety Tools 113

Incorporating Safety into Long-Range Transportation-Planning MULTIMODAL TRANSPORTATION-PLANNING TOOL (MTPT) GDOT Exhibit 75: Summary of the MTPT tool MTPT (GDOT) Vendor name and address: Georgia Department of Transportation: Office of Planning, 2 Capitol Square, Atlanta, GA 30334 Brief description of transportation safety applications: This tool is a first-step analysis of the operating performance of the Georgia’s rural transportation system. The tool consists of modules corresponding to different transportation modes, including, Highway – The highway module incorporates several roadway analysis elements. Essentially, the module determines level of service for existing and future conditions, evaluates delay associated with interrupted flow conditions (in the form of signalized intersections), identifies high risk accident locations, and prioritizes possible improvement strategies for the individual road. Improvement recommendations may range from “No action required” to “Requires Immediate Action”. Below are overviews of the individual modules incorporated in the highway analysis. The accident analysis component of this module permits the program user to identify regions within the study area where the number of accidents exceeded statewide averages for that specific roadway functional classification. Rural Transit – The rural transit analysis module performs several tasks. First, it eliminates urbanized and FTA Section 5311 public transit service provider regions from analysis due to the focus of the module on rural conditions. Next, the module evaluates the socioeconomic characteristics of a region to determine transit needs, prioritizes the identified needs, and then estimates implementation costs. Commuter and Passenger Rail – The Georgia State Commuter Rail Plan is available to the MTPT as a two-phase plan. The current program analysis includes a list of proposed station locations and the recommended implementation phase from the Commuter Rail Plan. Phase 1 indicates service proposed for initial plan implementation. Phase 2 represents rail service proposed for later implementation. Aviation – The MTPT aviation analysis is based upon two GDOT data sources. The five-year aviation capital improvements plan (CIP) provides anticipated improvement projects based upon local input. A separate database record is maintained for each airport, and the airports are categorized according to “associated city”. FAA-funded airports are also available to the program. This resource permits the MTPT to include recommendations for the appropriate airport (current status, availability, etc.) when a city or county analysis is undertaken. Bicycle and Pedestrian – Two separate analyses are performed for the statewide bicycle plan. First, if the program user selects bicycle analysis then the program queries the bicycle lane database and identifies corridors that are common to the proposed bike plan. Next, the program evaluates the specific road characteristics (lane width, surface type, shoulder width, etc.) to determine what improvements are required before the bicycle lane can be accommodated. An upgrade designation ranging from Minor 1 (essentially only an overlay required) up to Major 2 (full roadway reconstruction and widening necessary) is assigned to the road and an estimated improvement cost is applied. The second bicycle analysis feature is implemented during highway analysis. At that time, when a road level of service has been analyzed, an “action priority code” is assigned to the road that indicates if the road is currently suitable. One step in assigning this code is evaluation of the corridor for suitable conditions to accommodate future bicycle facilities based on physical road characteristics. Intercity Bus – The Georgia intercity bus plan provides information on where current services are potentially vulnerable to abandonment, and where new services should be considered. The MTPT includes a database query that identifies potential new routes and all routes vulnerable to abandonment. Types and sources of data needed: The data are derived from several sources at the Georgia Department of Transportation. The Road Characteristics Database and the Accident Database are the two most important data sources for the highway and pedestrian/bicycle analyses. Other information is provided from current statewide modal transportation plans. Expertise required: The package comes with a simple tutorial on how to use the system. No special expertise is needed to use the package. Appendix C: Safety Tools 114

Incorporating Safety into Long-Range Transportation-Planning Hardware requirements: The package is currently available in Windows 98, Windows 2000, NT Service Pack 5, and Windows XP. Example application of tool: One of the more innovative applications of this tool is found in the Environmental Justice module, which relates motor vehicle/ pedestrian accidents to socio-economic characteristics of the surrounding area. Pedestrian/bicycle accident rates (from the state’s accident database) are linked to the Road Characteristics database through GIS and where high accident rates are found to occur on high-volume, high-speed roads, the corridors are flagged for further attention. Appendix C: Safety Tools 115

Incorporating Safety into Long-Range Transportation-Planning PEDESTRIAN AND BICYCLE ACCIDENT ANALYSIS TOOL (PBCAT) Exhibit 76: Summary of the PBCAT tool PBCAT Vendor name and address: Federal Highway Administration (FHWA). See http://www.fhwa.dot.gov/ or http://www.hsrc.unc.edu/ for additional information. Brief description of transportation safety applications: The PBCAT is utilized to develop and analyze databases containing details associated with accidents between motor vehicles and pedestrians or bicyclists. The tool includes “accident type”, which describes the actions of the parties involved immediately prior to the accident. The software was developed based on NHTSA’s development of “typing” methodology in the 1970’s to better describe the sequence of events and precipitating actions that led up to the accidents. Once the supporting database is developed, the tool can be utilized to select countermeasures to address identified problems. Types and sources of data needed: Accident type data with geometric, time, weather, location, age, sex, subject actions, and other attributes. Expertise required: Basic computer knowledge. The software contains user-friendly, online instructions and help features, along with a user's manual. Hardware requirements: Windows work station. Example application of tool: The accident-typing methodology included in the PBCAT allows the user to determine the accident type through a series of on-screen questions about the accident, accident location, and the maneuvers of the parties involved. The PBCAT enables practitioners to generate information for promoting bicycle and pedestrian safety and designing safer facilities where bicyclists, pedestrians, and motor vehicles interact. The software is designed with recommended countermeasures linked to specific bicycle and pedestrian accident types and has related resource and reference information. Countermeasures may include physical roadway improvements such as raised pedestrian crossings or other measures, or may include targeted enforcement. Users also have the ability to customize the database in terms of units of measurement, variables, and location referencing, as well as import/export data from/to other databases, as shown in Exhibit 77 and Exhibit 78. Users can produce a series of tables and graphs defining the various accident types and other factors associated with the accidents such as age, gender, light conditions, etc. Appendix C: Safety Tools 116

Incorporating Safety into Long-Range Transportation-Planning Exhibit 77: PBCAT style and navigation window Version 1 Crash Typing Exhibit 78: PBCAT style and navigation window Appendix C: Safety Tools 117

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools PEDESTRIAN SAFETY GUIDE AND COUNTERMEASURE (PEDSAFE) PEDSAFE Vendor name and address: University of North Carolina, sponsored by Federal Highway Administration (FHWA). See HTUhttp://www.fhwa.dot.gov/ UTH or HTUhttp://www.hsrc.unc.edu/ UTH for additional information. Brief description of transportation safety applications: The PEDSAFE prototype is currently under beta testing and will incorporate the content of the FHWA Pedestrian Facilities User Guide into a system that allows the user to select appropriate countermeasures or treatments to address specific safety problems for pedestrians. PEDSAFE also includes a large number of case studies to illustrate treatments implemented in several communities throughout the United States and Europe. PEDSAFE was designed to enable practitioners to select engineering, education, or enforcement treatments to help mitigate a known accident problem and/or to help achieve a specific performance objective. Types and sources of data needed: Accident type data with geometric, time, weather, location, age, sex, subject actions, and other attributes. Expertise required: Basic computer knowledge. The software contains user-friendly, online instructions and help features, along with a user's manual. Hardware requirements: Windows work station. Example application of tool: PEDSAFE uses known characteristics of the environment and permits the user to either view all countermeasures associated with a given objective or accident type or to view only those that are applicable to a defined set (as input by the user) of geometric and operating conditions. While the majority of the specific treatments are engineering countermeasures, many of the case studies include supplemental enforcement activities (e.g., neighbor speed watch programs) and/or educational approaches (e.g., in conjunction with school route improvements). The objectives of the product are as follows: • Provide user with information on which countermeasures are available for prevention of specific categories of pedestrian accidents or to achieve certain performance objectives. • Outline considerations to be addressed in the selection of a countermeasure. • Provide a decision process to eliminate countermeasures from the list of possibilities. • Provide case studies, statistics, and other resources for the short list of countermeasures. Upon completion of beta testing and continued revision, the PEDSAFE Guide and Countermeasure Selection System will be incorporated into the Pedestrian and Bicycle Accident Analysis Tool (PBCAT). Exhibit 79: Summary of the PEDSAFE tool 118

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools ROADSIDE SAFETY ANALYSIS PROGRAM (RSAP) RSAP Vendor name and address: American Association of State Highway and Transportation Officials, 444 North Capitol Street N.W., Suite 249, Washington, D.C., 20001. Brief description of transportation safety applications: Provides a software tool to perform economic analysis of roadside safety feature or treatment alternatives. Types and sources of data needed: Traffic related information (including traffic volumes, expected traffic growth); highway characteristics (such as type, horizontal and vertical alignment); roadside safety feature impact characteristics; expected crash costs for various injury severity levels; and the costs associated with installation, maintenance and repair of roadside safety feature or system. Expertise required: Knowledge of roadside safety features considered, associated costs, and interpretation of benefit/cost analysis. The user-friendly interface requires a basic computer knowledge. Hardware requirements: Minimum requirements: a Pentium III PC or equivalent platform, 128MB RAM, 8.5MB hard disk space for program files and an additional 1MB for temporary data files, mouse for navigation within the software, Microsoft Windows operating system (98, NT, ME, 2000, or XP) RSAP was developed under NCHRP Project 22-9 and was incorporated in the AASHTO Roadside Design Guide (2002). An Engineer’s Manual (2003) and User’s Manual are available from the Transportation Research Board. The purpose of this section is to briefly describe the methodology used by RSAP using Appendix A of the AASHTO Roadside Design Guide as reference. RSAP is used for the evaluation of alternatives of roadside safety-related projects. It supports the principle that investments in the roadside, whether it be the selection of roadside features or a particular roadside design, be made based on maximizing the benefits of public funding. The software defines benefits as the savings in societal cost from a reduction in the frequency and/or severity of roadside-related crashes. Costs refer to the direct costs related to the installation, maintenance and repair of the particular device or system. The incremental benefit cost ratio is calculated during the analysis and refers to the increased benefit and cost related to the improvement option selected over another alternative or existing condition. The software uses an Encroachment Model with the following basic form: )()|()|()()( ICCIPECPEVPCE = Where E(C) = Estimated accident cost V = Traffic volume P(E) = Probability of encroachment (encroachment rate) P(C|E) = Probability of accident given encroachment P(I|C) = Probability of injury given accident C(I ) = Cost of injury. In the encroachment probability-based-cost-effectiveness analysis procedure, four different modules are used: the Encroachment Module, the Crash Prediction Exhibit 80: Summary of the RSAP tool 119

Incorporating Safety into Long-Range Transportation-Planning Module, the Severity Prediction Module, and the Benefit/Cost Module. The procedure calculates the estimated accident cost by calculating: a) the encroachment frequency with the Encroachment Model, b) the likelihood that the encroachment will result in an accident with the Crash Prediction Model, c) the estimated severity in the event that an accident occur with the Severity Prediction Module, and d) the annualized crash cost (AC) with the first part of the Benefit/Cost Module. The annual direct cost (DC) related to the roadside safety feature is calculated in the Benefit/Cost Module by adding the following: a) the annualized initial installation cost: annualized over the lifetime of the project by using the discount rate, b) the annual general maintenance cost, and c) the estimated annual accident maintenance repair cost: estimated by using the likely damage in the event of an impact. In the last part of the Benefit/Cost Ratio Module all alternatives are compared in a pair wise manner by using the following equation: B/C Ratio2-1 = (AC1 – AC2 ) / (DC1 – DC2 ), where B/C Ratio2-1 = incremental benefit/cost ratio of Alt. 2 compared to Alt. 1. AC1 , AC2 = annualized crash or societal cost of Alternatives 1 and 2. DC1 , DC2 = annualized direct cost of Alternatives 1 and 2. REFERENCES American Association of State Highway and Transportation Officials. Roadside Design Guide, Washington, D.C., 2002. King, K.M, and Sicking, D.L. Roadside Safety Analysis Program (RSAP) – Engineer’s Manual, National Cooperative Highway Research Program Report 492, Washington D.C., 2003. King, K.M, and Sicking, D.L. Roadside Safety Analysis Program (RSAP) – User’s Manual, National Cooperative Highway Research Program Project 22-9: Improved Procedures for Cost-Effectiveness Analysis of Roadside Safety Features, Transportation Research Board, Washington D.C., 2002. Appendix C: Safety Tools 120

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools SAFENET SAFENET Vendor name and address: UK Department for Transport e-mail: HTUsoftwarebureau@trl.co.ukUTH Brief description of transportation safety applications: SafeNET is an interactive software package developed under UK Department for Transport for safety management. SafeNET includes various traffic accident prediction models for different types of intersections as well as roadway segments. This system is used as a stand-alone product to assess safety and predict total as well as specific types (e.g., pedestrian accidents, rear end or head on accidents) of accidents in a transportation network. Additionally, SafeNET is used with a traffic assignment model “CONTRAM”, from which SafeNET can extract traffic flow data on the transportation network. This specific feature enables SafeNET to produce more information by accounting for safety and congestion issues simultaneously. The graphical display allows the engineer to visualize the effect on accident frequency of any change in junction design, form of control, and traffic assignment. The various types of road networks that can be modeled by SafeNET include: Roundabouts Mini-roundabouts Signalized intersections Urban and rural priority T-intersections Urban crossroads and staged intersections Urban collector roads Urban roads including minor intersections Traffic calming measures Types and sources of data needed: In SafeNET, models are possible at various “levels” with different input requirements. The most basic levels require simple traffic inflows averaged over the average day (ADT or AADT). More detailed analysis requires information on vehicle flows, pedestrian flows, site characteristics, specific geometric features, junction turning flows, and other design features. Expertise required: Basics of traffic engineering and knowledge about accident modeling as well as 4-step planning methods. Hardware requirements: Windows work station. Example application of tool: A description of the software as provided on the website of UK Department for Transport under “Traffic Advisory Leaflet, 08/99: Urban Safety Management: Using SafeNET” is provided below. A road network in a typical SafeNET window is shown in Exhibit 82. This network consists of two east-west routes into a town center and a number of connecting north-south roads. The above east-west route has been designated as A road and it is a wide single carriageway, whereas the bottom east-west route, designated as B road, is connected with a school, a number of shops, and some residential locations. All of the north-south routes connected with the above and bottom east-west roads are largely residential areas. The main problem in the area is the B road, as it carries local traffic as well as flow of traffic to and from the town center which would be more suited to the A road. A considerable portion of drivers also use one or other of the residential north-south roads. As a result, the speeds and flows through parts of the network are inappropriate for the character of the roads concerned. The main treatments being considered are: • to close some of the junctions between the north-south and the east-west routes to reduce the opportunities for ‘rat running’, and to traffic calm those north- south roads which would remain fully open; • to install traffic calming on the B road near the shops and the school; Exhibit 81: Summary of the SAFENET tool 121

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools • to convert a number of junctions from major/minor priority junctions to mini- roundabouts.” As described on the manufacturer’s website, one of the key aims of this scheme was to achieve a re-allocation of traffic to more suitable routes. Hence, it was necessary to use a traffic assignment model to input the traffic flow along the various links of the network. As mentioned previously, SafeNET can extract the total daily flows from the CONTRAM assignment program model outputs. Consequently, CONTRAM is used as an assignment model and is modified to represent the proposed network with the relevant closures and changes in intersection control. The impact on safety of the plan can then be assessed using SafeNET. In practice, it is wise to compare and contrast the results of several ‘build’ alternatives. Frequently it will become apparent that some modifications to the plan are necessary to achieve the desired safety (and other) objectives. For example, features may be needed to reduce congestion, cause a further re-distribution of traffic, or to achieve greater safety. A modification for one purpose might have unintended, possibly undesirable, effects on other aspects of performance. The link between the traffic assignment and impact on safety makes it easy to account for interactions between traffic volumes, safety, and mobility. In particular, the process allows rapid adjustment of the flows, which yield key inputs to the accident prediction models. According to the website, in this case, the initial proposal proved to be flawed due to a significant number of vehicles using an alternative route to avoid traffic calming installed outside the shops. The extent of this fresh “rat running” was apparent from the CONTRAM run and the accident predictions from SafeNET showed the expected effect on safety. When additional measures were included in the proposed plan and modeled by CONTRAM, the revised flows were immediately available to SafeNET, which can then be used to predict the safety effects of the network changes. Exhibit 82: Road network in SafeNET 122

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools SAFETYANALYST SAFETYANALYST Vendor name and address: Federal Highway Administration, HTUwww.fhwa.dot.govUTH Brief description of transportation safety applications: SafetyAnalyst addresses site-specific safety improvements that involve physical modifications to the highway system. SafetyAnalyst is not intended for direct application to non-site-specific highway safety programs that can improve safety for all highway travel, such as vehicle design improvements, graduated licensing, occupant restraints, or alcohol/drug use programs. However, SafetyAnalyst has the capability not only to identify accident patterns at specific locations and determine whether those accident types are overrepresented, but also to determine the frequency and percentage of particular accident types system-wide or for specified portions of the system (e.g., particular highway segments or intersection types). This capability can be used to investigate the need for system-wide engineering improvements (e.g., shoulder rumble strips on freeways) and for enforcement and public education efforts that may be effective in situations where engineering countermeasures may not. Types and sources of data needed: Accident data with geometric, traffic, weather, and driver demographics. Expertise required: Knowledge about statistical analysis and basics of traffic engineering is sufficient to handle this tool. Hardware requirements: Windows work station Example application of tool: SafetyAnalyst consists of six software programs to analyze the safety performance of specific sites, to suggest appropriate countermeasures, quantify their expected benefits, and to evaluate their effectiveness. Planning for SafetyAnalyst development began in April 2001. The software to implement the SafetyAnalyst tools will be developed in a two-stage process. Interim tools with some, but not necessarily all, of the planned capabilities are planned for release in 2004. The interim tools will be revised based on user experience and expanded to include all planned capabilities. The final software tools are planned for release in 2006. SafetyAnalyst is envisioned as a set of software tools used by state and local highway agencies for highway safety management. The website for safety analyst HTUwww.safetyanalyst.org UTH provides considerable information about how to use the suite of software as when they are appropriate. According to the website, “SafetyAnalyst will be used by highway agencies to improve their programming of site-specific highway safety improvements. SafetyAnalyst will incorporate state-of-the-art safety management approaches into computerized analytical tools for guiding the decision-making process to identify safety improvement needs and develop a system wide program of site-specific improvement projects.” In addition, SafetyAnalyst can be used for cost-effectiveness analysis of safety improvements to ensure that highway agencies get the greatest possible safety benefit from each dollar spent in the name of safety. SafetyAnalyst consists of six software programs to analyze the safety performance of specific sites, to suggest appropriate countermeasures, quantify their expected benefits and to evaluate their effectiveness. These six tools are • Network Screening Tool • Diagnosis Tool • Countermeasure selection Tool • Economic Appraisal Tool • Priority Ranking Tool Exhibit 83: Summary of the SAFETYANALYST tool 123

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools • Evaluation Tool Network Screening Tool: The main aim of this tool is to identify sites that need safety improvements. This requires advanced data management as well as appropriate statistical methodology. As a result of extensive research on highway safety and statistical analysis over last 20 years, SafetyAnalyst software will implement these new approaches in its network screening. For example, the Empirical Bayes (EB) approach is included in the tool. EB combines observed and expected accident frequencies to provide estimates of the safety performance of specific sites that are not biased by regression to the mean, which is a drawback of traditional methods. The EB approach also incorporates nonlinear regression relationships between traffic volume and expected accident frequency. The sites identified by the network screening methodology are referred to as "sites with promise", as they are sites that have promise as locations where improvements can result in substantial accident reductions. Another new measure that has been proposed for network screening application is the potential for safety improvement (PSI) index. PSI is a measure of the excess accident frequency, above the expected frequency, that might be reduced if a safety improvement were implemented. An example of the application of PSI and how it is beneficial in safety improvement is demonstrated on the SafetyAnalyst website and is also presented here. Exhibit 84 shows a group of signalized intersections that have been ranked according to their accident frequencies during a five-year period. The last column in the table shows the ranking based upon the PSI. Based on accident frequency ranking alone, one might improve the highest-volume location first. Alternatively, using the PSI to rank sites, the highest-ranking intersection is a lower- volume intersection, ranked sixth in accident frequency, showing a greater potential for accident reduction. Intersection Total Accident Frequency (1995-99) Average Annual Daily Traffic (veh/day) Accident Frequency Ranking Potential for Safety Improvement (PSI) Ranking A 131 63502 1 2 B 104 35284 2 3 C 77 57988 3 11 D 75 46979 4 6 E 66 51933 5 10 F 51 48427 6 1 G 51 20423 7 15 H 46 34759 8 5 I 42 53396 9 61 J 38 25223 10 17 In Exhibit 85, intersections in a city have been ranked according to accident rate. The last column in the table shows the ranking based upon the PSI. If the five highest-ranking intersections based on accident rate were selected for improvements, the three highest-ranking intersections based on the PSI would not receive attention. The figures also indicate that scarce financial resources are allocated to sites ranked 33rd and 35th in PSI, while over 30 intersections with greater potential for safety improvements will not receive attention (application of countermeasures). Exhibit 84: Comparison of rankings by accident frequency and PSI for signalized intersections in a particular city 124

Incorporating Safety into Long-Range Transportation-Planning Exhibit 85: Comparison of rankings by accident rate and PSI for signalized intersections in a particular city Intersection Total Accident Frequency (1995-99) Average Annual Daily Traffic (veh/day) Accident Frequency Ranking Potential for Safety Improvement (PSI) Ranking N 18 5063 1 33 M 22 7009 2 9 L 27 8152 3 8 R 14 4402 4 35 K 33 10458 5 4 B 104 35284 6 3 O 18 4242 7 14 A 131 63502 8 2 P 16 7815 9 19 J 38 25223 10 17 Diagnosis Tool: This tool is used to diagnose the nature of safety problems at specific sites. Although highway agencies use various different software packages to generate collision diagrams (see for example Intersection Magic), these tools are independent and do work seamlessly with Safety Analyst. The diagnosis tool in SafetyAnalyst generates collision diagrams that identify collision types that are overrepresented at specific locations. The software will also examine common factors that might exist among similar crash outcomes. As a result, the software serves as an expert system to guide the user through field investigations of particular sites. As described on the SafetyAnalyst website, the software generates site-specific lists of questions to be asked during a field visit to the site based on the generated collision diagram, available data about the accident experience, geometric design features, as well as traffic control at the site. The field investigation will then serve to aid in the identification of appropriate countermeasures for improving safety at the site. Countermeasure Selection Tool: This tool assists users selecting countermeasures to reduce accident frequency and severity at specific sites. It aids investigators to identify appropriate countermeasures for a particular site from lists of potential countermeasures incorporated in the software. The logic that identifies appropriate countermeasures considers the accident patterns and related site conditions investigated in the diagnostic process. The user can select one or more of the suggested countermeasures for further consideration or can add other countermeasures that they consider appropriate. When two or more countermeasures are selected by the user, a final choice among them is made using the economic appraisal and priority-ranking tools. Economic Appraisal Tool: SafetyAnalyst permits users to conduct economic appraisals of the costs and safety benefits of countermeasures selected for a specific site. The economic appraisal results are used to compare alternative countermeasures for a particular site and to develop improvement priorities across sites. SafetyAnalyst includes an optimization program that is capable of selecting a set of safety improvements that maximizes the system wide safety benefits of a program of improvements with a specific improvement budget. Safety effectiveness measures are estimated from data on the observed, expected, and predicted accident frequency and severity at the site, the accident patterns identified in the preceding tools, and accident modification factors (AMFs) for specific countermeasures. The AMFs representing the safety effectiveness of particular countermeasures are based on the best available safety research. The analyses will include appropriate consideration of the service life of the countermeasure and the time value of money. This tool is capable of performing economic analyses consistent with the requirements of the Federal Highway Safety Improvement Program (HSIP) so that analysis results are readily acceptable to FHWA for implementation with federal funds. Appendix C: Safety Tools 125

Incorporating Safety into Long-Range Transportation-Planning The Priority Ranking Tool: This tool provides a priority ranking of sites and proposed improvement projects based on the benefit and cost estimates determined by the economic appraisal tool. The priority-ranking tool compares the benefits and costs of projects across sites and ranks the projects on the basis of cost effectiveness, benefit-cost ratio, or net present value. This comparison will allow users to fund projects in priority order, with the highest-ranked projects being funded first. The priority-ranking tool also determines an optimal set of projects to maximize safety benefits. Evaluation Tool: Most highway agencies do not routinely conduct evaluations of implemented countermeasures, and few evaluations that are conducted are well designed. SafetyAnalyst provides a tool to enable the design and application of well- designed before/after evaluations. These evaluations are highly desirable to increase knowledge of project effectiveness and supplement or improve the safety effectiveness measures for improvements available for use in SafetyAnalyst. This tool is capable of performing before-after evaluations using the Empirical Bayes (EB) approach. As mentioned previously, the EB approach is a statistical technique that compensates for regression to the mean, and allows for the proper accounting of changes in safety that may be due to changes in other factors, such as traffic volumes. This tool will also provide, where appropriate, users with the ability to perform before-after evaluations using statistical techniques other than the EB approach. Appendix C: Safety Tools 126

Incorporating Safety into Long-Range Transportation-Planning TRANSPORTATION ANALYSIS AND SIMULATION SYSTEM (TRANSIMS) Exhibit 86: Summary of the TRANSIMS tool TRANSIMS Vendor name and address: TRANSIMS technology is being developed under U.S. DOT and EPA funding at the Los Alamos National Laboratory (LANL). IBM Business Consulting has created a commercial version of TRANSIMS named TRANSIMS-DOT. See http://transims.tsasa.lanl.gov/ for more details. Brief description of transportation safety applications: TRANSIMS is an integrated system of travel forecasting models designed to give transportation planners accurate, complete information on traffic impacts, congestion, and pollution. The design of TRANSIMS is based on requirements in the Intermodal Surface Transportation Efficiency Act (ISTEA), the Transportation Equity Act for the 21st Century (TEA-21), and Clean Air Act Amendments. The software consists of mutually supportive simulations, models, and databases that employ advanced computational and analytical techniques to create an integrated regional transportation system analysis environment. By applying advanced technologies and methods, it simulates the dynamic details that contribute to the complexity inherent in today's and tomorrow's transportation issues. The integrated results from the detailed simulations will support transportation planners, engineers, decision makers, and others who must address environmental pollution, energy consumption, traffic congestion, land use planning, traffic safety, intelligent vehicle efficiencies, and the transportation infrastructure effect on the quality of life, productivity, and economy. Although safety is not currently integrated into the tool, it is possible and/or likely that safety considerations may be added in future revisions to the software. Types and sources of data needed: TRANSIMS uses census data of household surveys such as production/attraction (PA) tables, origin/destination (OD) matrices, the transportation network data for major intersections, and other information to produce pseudo-activities for trip generation. Expertise required: TRANSIMS contains an easy-to-use Graphical User Interface for the transportation modeling function, a GIS-based network editor, a 3D data visualization and animation software, and a reporting system. Hence knowledge of Oracle database, C++ programming language, or the ArcView Avenue programming language is preferred to handle this tool, although not essential. The essential expertise, of course, is needed in the field of transportation network modeling. It is also likely that full-time maintenance and operation of models is needed due to the sophistication and complexity of the simulation characteristics, inputs, and outputs. It is also anticipated that running TRANSIMS will require significantly greater resources—both human and computer—than traditional 4-step travel demand models. Hardware requirements: Programs in the TRANSIMS-DOT software are distributed applications with components running on different hardware/software platforms. In order to install and run all of the components of the TRANSIMS-DOT, it is necessary to have the following three types of computer systems: a) UNIX or LINUX servers for hosting the core LANL TRANSIMS software, and Oracle database and server-side components of the TRANSIMS-DOT Modeling Interface. To execute large-size problems, it is necessary to install a multi-server Linux computing cluster or an equivalent multi-processor UNIX- based framework. b) Windows workstation(s) for running the Network Editor, the client-side GUI Modeling Interface, and Crystal Reports. c) Optional Linux workstation(s) for running the Visualizer. Alternatively, it is possible to equip the Linux server with a high-end graphics card to use it as the Visualizer platform. Example application of tool: TRANSIMS models create a virtual metropolitan region with a complete representation of the region's individuals, their activities, and the transportation infrastructure. Trips are planned to satisfy the individuals' activity patterns. TRANSIMS then simulates the movement of individuals across the transportation network, including their use of vehicles such as cars or buses, on a second-by-second basis. This virtual world of travelers mimics the traveling and driving behavior of real people in the region. The interactions of individual vehicles produce realistic traffic dynamics from which analysts using TRANSIMS can estimate vehicle emissions and judge the overall performance of the transportation system. Previous transportation-planning (usual four step methods) surveyed people about elements of their Appendix C: Safety Tools 127

Incorporating Safety into Long-Range Transportation-Planning trips such as origins, destinations, routes, timing, and forms of transportation used, or modes. TRANSIMS starts with data about people's activities and the trips they take to carry out those activities, and then builds a model of household and activity demand. The model forecasts how changes in transportation policy or infrastructure might affect those activities and trips. TRANSIMS tries to capture every important interaction between travel subsystems, such as an individual's activity plans and congestion on the transportation system. For instance, when a trip takes too long, people find other routes, change from car to bus or vice versa, leave at different times, or decide not to do a given activity at a given location. Also, because TRANSIMS tracks individual travelers—locations, routes, modes taken, and how well their travel plans are executed—it can be used to evaluate transportation alternatives and reliability to determine who might benefit and who might be adversely affected by transportation changes. In addition, it can make better volume predictions along the network, which in turn is useful for safety analysis. Appendix C: Safety Tools 128

Incorporating Safety into Long-Range Transportation-Planning PLANSAFE: PLANNING LEVEL SAFETY PREDICTION MODEL Introduction to PLANSAFE The researchers involved with NCHRP 8-44 developed a planning level safety prediction model, dubbed PLANSAFE, which is intended to serve as a useful tool for regional level safety planning. The model is intended to support and supplement some of the planning level activities described in this guidance. The reader should be aware that the majority of tools described in this appendix are corridor or project level analysis tools, and are not suitable for forecasting crashes at a planning scale. Planning level safety decisions, unlike corridor and project level analyses, do not involve considerations about design details of facilities. The PLANSAFE model, in keeping, is inappropriate for supporting decisions regarding design details of facilities. As an example, the congestion impacts of signal timing schemes are not considered in travel demand models nor are the safety impacts of signal timing schemes estimated when using PLANSAFE. The PLANSAFE model uses typical planning level information: socio-economic, demographic, and transportation-related data to predict the safety of TAZs or larger sub-areas of a jurisdiction. The intent, of course, is to enable straightforward analyses using travel demand model output and planning level data to support the PLANSAFE models and ultimately to guide decision making at the planning. The PLANSAFE model is extremely useful for a variety of planning level activities, which are described in detail later in this section. For example, setting safety performance targets requires an estimate of what safety will be in some future time period in the absence of ‘additional’ safety countermeasures. The PLANSAFE model supports this type of forecast, where the smallest analysis unit is a traffic analysis zone (TAZ), and the largest unit of analysis is an entire region (say a non-attainment area or metropolitan planning region). These and other types of analyses are described in this section. There are few, if any, planning level safety prediction models available for use and relatively little research on them has been conducted as of the date of this report. The need for these models has arisen from the ISTEA legislation, which requires the explicit consideration of safety at the planning level, but has left the profession lacking a complete set of tools. Planning level safety prediction models are fundamentally different in nature to corridor or site specific crash prediction models because; 1) The input data are aggregate and not site or project specific; 2) The focus is prediction and not explanation of safety; and 3) The model should not be used to choose between investment alternatives but instead to inform the user of safety impacts of alternative investments and to establish future performance targets. Exhibit 87 summarizes the PLANSAFE model characteristics and attributes. Appendix C: Safety Tools 129

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools PLANNING LEVEL SAFETY PREDICTION MODEL Vendor name and address: Simon Washington and Ida van Schalkwyk, Arizona State University. Tempe, Arizona, 85287. HTU simon.washington@asu.eduUTH . Brief description of transportation safety applications: The Planning Level Safety Prediction Model is a planning-level model used to predict motor vehicle accidents per traffic analysis zone (TAZ) area or larger sub-areas of a jurisdiction. Thus, the smallest unit of analysis is the TAZ, whereas the largest unit of analysis is collections of TAZs such as neighborhoods, those TAZs affected by a major transportation project, etc. Crashes of various types are modelled as functions of various predictors such as the distribution and mileage of the functional classifications of highways, vehicle miles traveled, socio-economic and demographic factors, and population characteristics. For development of the models under NCHRP 8-44, data from Pima and Maricopa Counties in Arizona and the state of Michigan were used. These regions represent a fairly diverse range of geography and driving populations in order to derive models that may approximate aggregate relationships across the U.S.. Types and sources of data needed: TAZ level data regarding population, travel, schools, infrastructure (e.g., residential units, commercial units, etc.), and crashes. Expertise required: Knowledge of GIS, some statistical modelling, or statistical model interpretation skills. Hardware requirements: Desktop PC with database and GIS software. Example application of tool: The tool can be used to forecast the projected increase (over baseline totals) in fatal, injury, pedestrian, and total crashes expected in 10 years given population growth, the provision of new schools, and other changes under the ‘no-build’ scenario and various ‘build’ scenarios (refer to the section titled When to use the PLANSAFE (and when not to). This remainder of this appendix provides details of the operation, assumptions, and output of the PLANSAFE model, whose core models are estimated using data from the states of Arizona and Michigan. Appendix D, in contrast, is targeted to agencies with the resources and desire to develop models that are based on local, regional, or statewide data. The remainder of this section is divided into five subsections: • Why TAZ level safety prediction models are logically feasible and defensible; • When to use the PLANSAFE (and when not to); • What data are needed to apply the PLANSAFE models; • The PLANSAFE set of forecasting models; • How to apply PLANSAFE models; and • Examples of PLANSAFE applications. Why TAZ level safety prediction models are logically feasible and defensible The safety profession is replete with models that predict crashes at the microscopic level—say for intersections or for road segments. Many of the analysis tools described in this appendix include microscopic safety crash models. A reasonable question to ask is “Are macroscopic, or TAZ level statistical models defensible and logically feasible?” The following arguments, based on accepted principles and logic from the road safety and statistics communities, support the use of aggregate level safety prediction models. Exhibit 87: Summary of the PLANNING LEVEL SAFETY PREDICTION MODEL tool 130

Incorporating Safety into Long-Range Transportation-Planning 1. Crashes are largely random events. Much research has shown that crashes are largely caused by human errors, with estimates ranging between 60% and 90% of crashes being caused by human errors. Thus, many crashes are more a function of human-related factors rather than roadway-related factors. As simple examples, crashes that result from of a driver tuning a radio, answering a cell phone, following another vehicle too closely, speeding, and running a red light are events that occur somewhat randomly on a network. It is easy to understand, then, that modelling crashes at the segment or intersection level is challenging, because there is a large random component to crashes that is not explained by local road characteristics. At a more aggregate level, in contrast, crashes are related to aggregate predictors, such as population demographics, ‘high risk’ driving populations, the general classes of road facilities, etc., and assigning crashes to specific links or segments is not necessary. Thus, by aggregating the transportation system at the TAZ level, some of the difficulties caused by ‘lumpiness’ of random events that we see across intersections or across road segments are reduced. 2. Aggregate safety differences are substantiated by research. Much research supports ‘aggregate’ or average safety differences across groups. Older drivers suffer from reduced reaction and perception times, as well as reduced vision and flexibility. Younger drivers suffer from inexperience and aggressiveness. Minorities have been shown to wear safety restraints less than whites, and restraint use in rural areas is less than in urban areas. Interstates are associated with relatively low crash rates, while rural roads with high speeds are associated with more serious injury crashes. Crashes in urban areas are attended by emergency medical services more quickly than crashes in rural areas. Intersections are locations of complex traffic movements and thus are associated with greater numbers of crashes than road segments. Increasing traffic congestion tends to reduce crash severity. School zones are associated with bicycle and pedestrian crashes. These well supported aggregate relationships, and others not listed here, are the relationships captured in aggregate level prediction models. The aggregate relationships described above form the basis for the statistical modelling at the TAZ level. It is the reliance on these ‘average’ relationships, and characteristics measured at the TAZ level, on which model predictions are based. 3. Models for predicting have fewer restrictions than models for explaining. Intersection and road-segment level accident prediction models are usually held to a high standard, as they are often used both to predict the expected performance of such facilities but also to explain relationships between variables. Often, and sometimes wrongly, these microscopic models are used to infer the effects of countermeasures, such as the safety effect of the presence of a left-turn lane on angle crashes. When a model is used simply for prediction, however, and not inference, there is greater flexibility in model estimation and variable selection choices. The PLANSAFE model is intended only for prediction, and not explanation. Thus, for example, if a population variable is used to predict fatal crashes per TAZ, its estimated coefficient is used solely in the prediction equation but is not interpreted to have specific explanatory marginal effects. Appendix C: Safety Tools 131

Incorporating Safety into Long-Range Transportation-Planning These three arguments, or justifications, form the basis for the development of aggregate level accident prediction models. A consequence of these arguments, however, is that the models cannot be used for explanation of crash causation or for the assessment of roadway-specific countermeasures. The aggregate relationships modeled are suitable for predicting a hypothetical or future outcome should the set of predictors be changed. This restriction is not too dissimilar from the restriction placed on travel demand models, whose primary purpose is to predict demand for roadway space of motor vehicles in hypothetical or future scenarios. When to use the PLANSAFE (and when not to) As described in the previous section, the PLANSAFE model has limitations and assumptions. Most importantly, PLANSAFE is fundamentally different from many other safety prediction models that have been presented and discussed previously in this appendix (e.g., IHSDM, SafetyAnalyst, etc.). Exhibit 88 lists appropriate and inappropriate uses of PLANSAFE models. The appropriate uses fall squarely in the domain of planning, prediction, or forecasting, while the inappropriate uses fall in the domain of traffic and safety engineering. An important assumption of the PLANSAFE model is that ‘new’ safety countermeasures are not applied in future scenarios. In other words, the ‘average’ set of design standards with respect to safety are assumed to exist in the future, while innovative, newly adopted, or progressive safety countermeasure investments are analyzed independently by some other model or research study. As a result, an investment in innovative safety countermeasures in the future will yield improvements in safety over and above those predicted by PLANSAFE models. Appropriate and inappropriate uses of PLANSAFE are provided in Exhibit 88. Exhibit 88: Appropriate and inappropriate uses of PLANSAFE models Appropriate Applications of PLANSAFE Setting safety targets or performance measures Safety targets serve as milestones for accomplishment. For example, a region may want to achieve a measurable decrease in pedestrian involved crashes in a future time period, say 5 years hence. The PLANSAFE model is suitable for establishing the expected number of crashes in some future period in the absence of targeted safety countermeasures. PLANSAFE is useful because crashes in the future are expected to change as a result of population growth, new road mileage, new schools, changing of the driving population, etc. Using simply the baseline (e.g., the current year’s) crash frequencies (e.g., fatal crashes, injury crashes, etc.) to set performance targets is strictly incorrect, since the impacts of growth alone will have an impact on the expected safety of a region or sub region. Understand the safety impacts of large scale projects (corridor level or higher) Large-scale projects that may affect VMT, future growth, and other planning related factors will affect safety. The PLANSAFE model is appropriate for forecasting the future expected safety performance of these projects in the absence o targeted safety counte measu es. f r r Given that a future project will influence the forecasting variables in the PLANSAFE model, the PLANSAFE model will produce a prediction of the effect of the project on safety (i.e., crashes of various types). Compare and contrast growth scenarios Growth scenarios are often compared looking 5, 10, and 20 years into the future. PLANSAFE is suitable for predicting the safety performance of the region under different growth scenarios (e.g., infill development, sprawl, interstate vs. highway, population and demographic shifts, new schools, etc.) in the absence of new or innovative and targeted safety countermeasures. These types of analyses are informative to determine how much safety Appendix C: Safety Tools 132

Incorporating Safety into Long-Range Transportation-Planning investment is needed to meet safety performance targets. For example, three different growth scenarios will produce three different estimates of future safety (in say the affected TAZs). These different growth scenarios would imply then, three different levels of safety investment required to meet regional safety performance targets. Additional analysis, through other means (say IHSDM or SafetyAnalyst software), would then be used to meet the safety objectives under the different growth scenarios. Inappropriate Applications of PLANSAFE Select land- use/transportation investment strategies based on model results Different growth scenarios will yield different estimates of future safety; however, the PLANSAFE models are predictive models and cannot account for the safety-related complexities present in real life growth scenarios. A future scenario with relatively ‘worse’ predicted safety does not mean it is a bad project, it may simply mean that more serious attention to safety investments may need to be made if that particular growth scenario is adopted. There are many factors other than safety to consider in land use/transportation investment, such as maintenance costs, air quality impacts, congestion, and environmental impacts (e.g., water, wetlands, endangered species, and archaeology). Evaluate or select safety countermeasures PLANSAFE models do not contain variables that are proxies for countermeasures. PLANSAFE models predict but do not explain crashes. Thus, PLANSAFE models are not suitable for evaluating roadway- or intersection-specific countermeasures. What data are needed to apply the PLANSAFE Models? Application of the PLANSAFE model requires forecasting data from the region where the model is applied. For example, a model that uses a particular population characteristic, percentage of a particular functional road class, and density of households would require estimates of these variables in both the ‘base’ and ‘future’ scenarios. To allow local calibration (to enable the model to reflect local conditions), the particular accident variable that is predicted also needs to be known in the base year. The required input (forecasting) variables needed for the base year and future year/proposed project for the set of affected TAZs are shown in Exhibit 89. The table shows the abbreviated name of the variable, the units of measure, and the source of the data. In many cases, variables were extracted from U.S. census block group data. All variables are calculated by TAZ. For example, the Total Accident Frequency Model requires as predictors the population density of the TAZ (persons per acre), the total population aged 16 to 64 in the TAZ, and the total mileage of all federal road functional classifications in the TAZ. Exhibit 90 shows the predictor variables required for eight safety-related outcome variables; total crashes, property damage only crashes, fatal crashes, incapacitating and fatal injury crashes, nighttime crashes, pedestrian crashes, injury crashes, and bicycle-involved crashes. Thus, at this time the PLANSAFE model includes the ability to predict eight safety-related outcome variables as a function of various predictor variables. Appendix C: Safety Tools 133

Incorporating Safety into Long-Range Transportation-Planning Exhibit 89: Variables and descriptions for the PLANSAFE models VARIABLE DESCRIPTION (all units are calculated per TAZ) Total Accident Frequency Model POP_PAC Population density (population estimates from U.S. Census SF1) in persons per acre POP16_64 Total population of ages 16 to 64 (from U.S. Census SF1) TOT_MILE Total mileage of all functional classes of roads Property Damage Only Accident Frequency Model PH_URB Number of urban housing units (U.S. Census SF1) as portion of all housing units POP_PAC Population density (population estimates from U.S. Census SF1) in persons per acre VMT Vehicle miles traveled (it is estimated using road section lengths and section traffic counts) Fatal Accident Frequency Model INT_PMI Number of intersections per mile (using total mileage in the TAZ) PNF_0111 Total mileage of urban and rural interstates as a portion of the total mileage (federal functional classifications 01 and 11) PNF_0512 Total mileage of other freeways and expressways (i.e., not interstate and also not principal arterials) as a portion of the total mileage POP00_15 Total population of ages 0 to 15 (from U.S. Census SF1) PPOPMIN Total number of minorities (from U.S. Census SF1) as a portion of the total population. Incapacitating and Fatal Accident Frequency Model INT_PMI Number of intersections per mile (using total mileage in the TAZ) PNF_0111 Total mileage of urban and rural interstates as a portion of the total mileage (federal functional classes 01 and 11) PNF_0512 Total mileage of other freeways and expressways (i.e., not interstate and also not principal arterials) as a portion of the total mileage POP00_15 Total population of ages 0 to 15 (from U.S. Census SF1) Nighttime Accident Frequency Model MI_PACRE Total mileage of the TAZ per acre of the TAZ PNF_0111 Total mileage of urban and rural interstates as a portion of the total mileage in the TAZ (federal functional classes 1 and 11) PNF_0214 Total mileage of urban and rural principal arterials as a portion of the total mileage in the TAZ (federal functional classes 2 and 14) PNF_0512 Total mileage of other freeways and expressways (i.e., not interstate and also not principal arterials) as a portion of the total mileage PPOPMIN Total number of minorities (from U.S. Census SF1) as a portion of the total population. WORKERS Total number of workers 16 years and older (from U.S. Census SF3) Accidents Involving Pedestrians Frequency Model HH_INC Median household income in 1999 (P053001 from U.S. Census SF3) POP_PAC Population density (population estimates from U.S. Census SF1) in persons per acre POPTOT Total population (P001001 from U.S. Census SF1) PWTPRV Proportion of workers 16 years and older that use a car, truck, or a van as a means of transportation to work (from U.S. Census SF3) Injury Accident Frequency Model HU_PACRE Number of housing units per acre: (H001001 from U.S. Census SF1)/Acres PPOPURB Urban population (P002002 from U.S. Census SF1) as a portion of the total population. VMT Vehicle miles traveled (it is estimated using road section lengths and section traffic counts) Accidents Involving Bicycles Frequency Model HU Number of housing units (from U.S. Census SF1) TOT_MILE Total mileage of all functional classes of roads VMT Vehicle miles traveled (it is estimated using road section lengths and section traffic counts) WORK_PAC Total number of workers 16 years and over (from U.S. Census SF3) per acre Appendix C: Safety Tools 134

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools The PLANSAFE set of Forecasting Models This section describes the statistical modelling results of the PLANSAFE models that are available for forecasting crashes by TAZ, and describes two of the models in greater detail. The application of these PLANSAFE models is described in the next section. The statistical modelling results presented here are based upon data from: • Pima Assocation of Governments (includes City of Tucson), Arizona. • Maricopa Association of Governments (Phoenix metropolitan area), Arizona. • The state of Michigan. Exhibit 90 shows the variables, the estimated coefficients, and the associated t- statistics with the PLANSAFE set of eight models. VARIABLE COEFFICIENTS t-STATISTIC Total Accident Frequency Model* POP_PAC 0.474 x 10P-1P 9.067 POP16_64 0.196 x 10P-3P 36.373 TOT_MILE 0.151 x 10P-2P 3.482 Property Damage Only Accident Frequency Model* PH_URB 0.515 13.626 POP_PAC 0.566 x 10P-1P 11.894 VMT 0.392 x 10P-5P 37.554 Fatal Accident Frequency Model* INT_PMI -0.924 x 10P-1P -18.535 PNF_0111 1.762 8.958 PNF_0512 1.389 4.755 POP00_15 0.263 x 10P-3P 26.340 PPOPMIN 0.319 5.577 Incapacitating and Fatal Accident Frequency Model* INT_PMI -0.659 x 10P-1P -9.864 PNF_0111 3.328 11.892 PNF_0512 3.674 8.723 POP00_15 0.512 x 10P-3P 36.793 Nighttime Accident Frequency Model* MI_PACRE -19.167 -12.126 PNF_0111 3.524 14.661 PNF_0214 1.414 5.393 PNF_0512 3.588 10.038 PPOPMIN 0.861 11.261 WORKERS 0.238 x 10P-3P 37.741 Pedestrians Accident Frequency Model* HH_INC -0.706 x 10P-5P -7.040 POP_PAC 0.129 27.101 POPTOT 0.884 x 10P-4P 24.520 PWTPRV -0.902 -3.808 Injury Accident Frequency Model* HU_PACRE 0.153 11.669 PPOPURB 0.768 18.401 VMT 0.443 x 10P-5P 39.250 Accidents Involving Bicycles Frequency Model** HU 0.252 x 10P-3P 10.394 TOT_MILE 0.162 x 10P-2P 2.012 VMT 0.292 x 10P-5P 9.730 WORK_PAC 1.539 15.600 NOTE: * indicates models developed using data from the State of Michigan, ** indicates that model was developed using data from Maricopa County (AZ). Exhibit 90: PLANSAFE Models with variable coefficients and t-statistics 135

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools The standard form of the models is a log linear regression model. The expressions for the PLANSAFE models are provided in Exhibit 91. To transform any of the models into original scale units, both sides of the equation are exponentiated, then 1 is subtracted from both sides. For example, the prediction equation for the Total Accident Frequency Model is: Acc_Freq = exp(5.020 + 0.0474(POP_PAC) + 0.00196(POP16_64) + 0.00151(TOT_MILE)) – 1 The log linear regression was chosen over the negative binomial form because: 1) It is known a priori that TAZs are of different size and therefore the underlying process is not a Poisson process with gamma heterogeneity of means; 2) Goodness of fit statistics are more intuitive and comparable using ordinary least squares estimated coefficients; and 3) Predictions and of non-integer values are acceptable for aggregated data. MODEL FORMS Total Accident Frequency Model )1_( +FrequencyAccidentLog ( ) ( )64_1610196.0_10 0.474020.5 3-1 POPPACPOP −×+×+= ( )MILETOT _10151.0 2−×+ Property Damage Only Accident Frequency Model )1__( +frequencyaccidentPDOLog ( ) ( ) ( )VMTPACPOPURBPH 51 10392.0_10566.0_ 0.515762.4 −− ×+×++= Fatal Accident Frequency Model )1__( +frequencyaccidentFatalLog ( ) ( ) ( )0512_389.10111_762.1_ 10924.0-652.0 1 PNFPNFPMIINT ++×= − ( ) ( )PPOPMINPOP 319.015_0010263.0 3 +×+ − Incapacitating and Fatal Accident Frequency Model )1____( +frequencyaccidentFatalandingIncpacitatLog ( ) ( ) ( )0512_674.30111_328.3_ 10.6590-257.2 1 PNFPNFPMIINT ++×= − ( )15_0010512.0 3 POP−×+ Nighttime Accident Frequency Model )1__( +frequencyaccidentNighttimeLog ( ) ( ) ( )0214_414.10111_524.3_ 167.19092.4 PNFPNFPACREMI ++−= ( ) ( ) ( )WORKERSPPOPMINPNF 10238.0861.00512_588.3 3−×+++ Pedestrians Accident Frequency Model )1____( +spedestrianinvolvingaccidentsoffrequencyLog ( ) ( ) ( )POPTOTPACPOPINCHH 45 10 884.0_129.0_ 10.7060-443.1 −− ×++×= ( )PWTPRV902.0− Injury Accident Frequency Model )1___( +accidentsinjuryoffrequencyLog ( ) ( ) ( )VMTPPOPURBPACREHU 510 443.0768.0_ 153.0108.3 −×+++= Accidents Involving Bicycles Frequency Model )1____( +bicyclistsinvolvingAccidentsoffrequencyLog ( ) ( ) ( )VMTMILETOTHU 5231 10 292.0_10 162.010 252.010655.0 −−−− ×+×+×+×= ( )PACWORK _539.1+ Exhibit 91: PLANSAFE model forms 136

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools It is worthwhile to discuss and assess a couple of the PLANSAFE models to illustrate how relationships are captured by the model and how predictions of these models are made. Discussion 1: Frequency of Incapacitating and Fatal Accidents This discussion focuses on the PLANSAFE model that predicts the frequency of incapacitating and fatal injury accidents within TAZs. The model prediction equation is given by: )1____( +frequencyaccidentFatalandingIncpacitatLog ( ) ( ) ( )0512_674.30111_328.3_ 10.6590-257.2 1 PNFPNFPMIINT ++×= − ( )15_0010512.0 3 POP−×+ The model predictor variables include intersections per mile of road, mileage of rural and urban interstates as a proportion of the total mileage, mileage of other freeways and expressways as a proportion of total mileage, and proportion of the population aged 0 to 15. Because the logarithm is a monotonically increasing function, a positive coefficient in the logarithm implies a positive effect of the predictor variable on crashes. As the number of intersections per mile increases, the predicted count of incapacitating and fatal accidents decreases, suggesting that greater urbanization is associated with greater congestion, lower travel speeds, and less serious crashes on average. The interstate and primary arterial mileage represents the exposure of vehicular traffic on relatively higher speed roads, and as the proportion of these facilities increase so do the predicted counts of incapacitating and fatal crashes. As the number of individuals between ages 0 and 15 increases, so does the predicted number of incapacitating and fatal accidents. Exhibit 92 shows the relationship between the predicted count of incapacitating and fatal accidents and the persons aged 0 to 15, with other variables held constant. Exhibit 93 shows the relationship between the predicted count of incapacitating and fatal accidents and the number of intersections per mile, with other variables held constant. The population variable is one of four exposure based variables, with two road mileage variables and one intersection exposure variable. Young children typically represent ‘active’ households with respect to VMT, and also represent an increased exposure to pedestrian and bicycle involved serious injury crashes. It is likely that this population based variable captures both the aggregate population effect as well as the ‘activity’ factor associated with families with young children. Discussion 2: Frequency of Accidents Involving Pedestrians The pedestrian crash prediction model is given as: )1____( +spedestrianinvolvingaccidentsoffrequencyLog ( ) ( ) ( )POPTOTPACPOPINCHH 45 10 884.0_129.0_ 10.7060-443.1 −− ×++×= ( )PWTPRV902.0− Four predictor variables are included in the model predicting the frequency of pedestrian involved accidents. The first is median household income; as median household income increases, predicted pedestrian involved accidents decrease. The effect of income captures many facets of pedestrian crashes: lower income neighborhoods are less likely to have sidewalks, are more likely to have unattended children walking in the streets, and are more likely to 137

Incorporating Safety into Long-Range Transportation-Planning Exhibit 92: Predicted number of incapacitating and fatal injury crashes by population count ages 0 to 15 by TAZ:- PLANSAFE incapacitating and fatal model PREDICTED NUMBER OF INCAPACITATING AND FATAL ACCIDENTS PER TAZ FOR A TAZ WITH 5 INTERSECTIONS PER MILE; 20% OF INTERSTATE MILEAGE; AND 20% OF OTHER FREEWAYS OTHER THAN PRINCIPAL ARTERIALS 0.2 0.4 0.6 0.8 1 1.2 1.4 0 50 0 10 00 15 00 20 00 25 00 30 00 35 00 40 00 45 00 50 00 55 00 60 00 65 00 70 00 75 00 80 00 85 00 90 00 95 00 10 00 0 10 50 0 11 00 0 11 50 0 12 00 0 IN C A PA C IT A TI N G A N D F A TA LA C C ID EN T FR EQ U EN C Y TOTAL POPULATION BETWEEN AGES 0 AND 15 (POP00_15) Exhibit 93: Predicted number of incapacitating and fatal injury crashes by intersection count per mile by TAZ:- PLANSAFE incapacitating and fatal model PREDICTED NUMBER OF INCAPACITATING AND FATAL ACCIDENTS PER TAZ FOR A TAZ WITH 8000 INDIVIDUALS AGE 0 TO 15; 20% OF INTERSTATE MILEAGE; AND 20% OF OTHER FREEWAYS OTHER THAN PRINCIPAL ARTERIALS 0.94 0.96 0.98 1 1.02 1.04 1.06 0.3 0.7 1.1 1.5 1.9 2.3 2.7 3.1 3.5 3.9 4.3 4.7 5.1 5.5 5.9 6.3 6.7 7.1 7.5 7.9 8.3 8.7 9.1 9.5 9.9 10 .3 10 .7 11 .1 11 .5 11 .9 IN C A PA C IT A TI N G A N D F A TA LA C C ID EN T FR EQ U EN C Y INTERSECTIONS PER MILE (INT_PMI) have workers commuting by walking among other possible aspects on average. The second predictor variable, density of the population in a TAZ, is another exposure variable as higher population densities typically indicate more urban environments where a greater amount of walking takes place and also where the available walking destinations increase, and therefore lead to an increased likelihood of walking as a transportation mode. Thus, the variable captures pedestrian exposure. The number of individuals living in a TAZ is another measure of exposure, and as such increases the likelihood of accidents involving pedestrians. The fourth variable, the portion of workers age 16 and older that use private transportation to travel to work, is also an exposure-related variable. A worker is less likely to be injured in a pedestrian accident when traveling by vehicle then by walking, bicycling, or when taking public transit. Exhibit 94 and Exhibit 95 illustrate the predicted relationships between pedestrian-related Appendix C: Safety Tools 138

Incorporating Safety into Long-Range Transportation-Planning accidents and population density and workers aged 16 and older, with other predictor variables held constant. Exhibit 94: Predicted number of crashes involving pedestrians by TAZ by population count per acre by TAZ:- PLANSAFE pedestrian model PREDICTED NUMBER OF ACCIDENTS INVOLVING PER TAZ FOR A TAZ WITH A MEAN HOUSEHOLD INCOME OF $45,000; TOTAL POPULATION COUNT OF 40,000; AND 91% OF WORKERS AGE 16 AND OLDER TRAVELLING TO WORK BY PRIVATE CAR, TRUCK, OR VAN. 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 FR EQ U EN C Y O F A C C ID EN TS IN VO LV IN G P ED ES TR IA N S POPULATION COUNT PER ACRE (POP_PAC) Exhibit 95: Predicted number of Crashes involving pedestrians by TAZ by portion of workers age 16 and older traveling to work by car, truck, or van, by TAZ:- PLANSAFE pedestrian model PREDICTED NUMBER OF ACCIDENTS INVOLVING PER TAZ FOR A TAZ WITH A MEAN HOUSEHOLD INCOME OF $45,000; POPULATION OF 5 PER ACRE; AND A TOTAL POPULATION COUNT OF 40,000.. 0.47 0.49 0.51 0.53 0.55 0.57 0.59 0.61 0.63 0.65 0. 2 0. 25 0. 3 0. 35 0. 4 0. 45 0. 5 0. 55 0. 6 0. 65 0. 7 0. 75 0. 8 0. 85 0. 9 0. 95 FR EQ U EN C Y O F A C C ID EN TS IN VO LV IN G P ED ES TR IA N S PORTION OF WORKERS 16 YEARS AND OLDER TRAVELLING TO WORK BY CAR, TRUCK, OR VAN (PWTPRV) This section describes the PLANSAFE accident prediction models with various summary statistics. Two of the PLANSAFE models are illustrated and fair detail, describing the nature of the modeled relationships. It is re-emphasized that although valid explanations are provided for the predictor variables in the models, the models are not used for explaining but instead for predicting crash outcomes by TAZ. Thus, one would not interpret a change in one of the predictor variables as a marginal Appendix C: Safety Tools 139

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools change in accidents to reflect the outcome of a countermeasure application. For example, one would not want to increase the number of workers taking an auto, truck, or van to work (in a TAZ) in order to reduce pedestrian crashes—this is an incorrect use of this model, and there exist far more effective methods and tools for assessing a pedestrian crash ‘problem’ once a problem is identified. Instead, one should simply use the model to forecast whether a pedestrian crash problem may exist in some future time or given a hypothetical growth scenario, and then plan to remediate pedestrian crashes with specific countermeasures as required by the local conditions and further study. How to Apply PLANSAFE Models The PLANSAFE Models are used to forecast safety in future periods or for various project/build scenarios at the TAZ level, as described previously. The same variables (data) used to estimate the models are also needed in order to make forecasts. Application of the PLANSAFE models proceeds by applying the following six analysis steps. 1. Collect variables needed to run models: All model variables need to be collected for TAZs in the affected analysis area for the base and forecast years or scenarios. The analysis area could be a set of TAZs affected by a large-scale project or the entire metropolitan region. The relevant crash data for the base year or scenario are also needed. All information should be manipulated in a GIS environment to allow for assignment of data to the TAZs or group of TAZs (the data collection and generation process is described in Appendix D). The most recent census block group data will constitute a major portion of the explanatory variables. 2. Generate the expected crash counts in a spreadsheet program (such as Microsoft Excel) or database management software program (such as Microsoft Access): The simple equation derived from the logarithmic linear regression model estimation results presented in the previous section is used to calculate the expected crash counts (e.g., pedestrian, total, fatal, etc.) by TAZ for the selected crash outcome. The model inputs are current crash counts by TAZ and independent variables for the baseline as well as forecasted independent variables for the future year scenarios. 3. Compute baseline correction factors, BCF: The baseline correction factors (BCFs) are obtained using the expected crash counts generated in step 2 to predict crashes in the baseline scenario. The BCF is an essential component of the analysis, as it corrects for differences between model calibrated safety and safety in the local region or state and is used to assess the goodness of fit of the model. In effect, the BCF is used to adjust for differences in expected accident frequencies observed in the states of Arizona and Michigan used to estimate the PLANSAFE models and the forecast state or region where the model is being applied. The asymptotically unbiased BCF, used to correct future predictions from the PLANSAFE models, is obtained using ∑ ∑ ∑ ∑ = = = = == toNi i toNi i toNi i toNi i unbiased P O N P N O BCF 1 1 1 1 . It should be noted that this asymptotically unbiased estimate of the BCF is not the average of the BCFs across TAZs, but instead the ratio of the average of observed crash frequencies divided by the average predicted frequencies. 140

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools To assess model fit, BCFs need to be calculated for individual TAZs. The BCF for TAZ i is calculated as, BCF BiB = OBiB/PBiB, where OBiB is the locally observed crash frequency for TAZ i, and PBiB is the predicted crash frequency using the PLANSAFE model for TAZ i. The next step is to compute the average BCF across TAZs, using ∑ = = toNi i i average P OBCF 1 . The standard deviation and coefficient of variation of individual BCFs are then calculated to enable goodness of fit assessment and comparison across PLANSAFE models. These two summary statistics are obtained using 2/12 1 ⎟⎟ ⎟⎟ ⎟⎟ ⎠ ⎞ ⎜⎜ ⎜⎜ ⎜⎜ ⎝ ⎛ ⎜⎜⎝ ⎛ ⎟⎟⎠ ⎞− = ∑ = N BCF P O SD toNi average i i BCF ; average BCF BCF BCF SDCV = . The standard deviation is the simple population standard deviation of the TAZ level BCFs, and the coefficient of variation is the standard deviation divided by the mean. 4. Predict future crashes: The baseline (comparison scenario) data for the TAZs of interest are used to calculate the BCFs described in step 3. The model is then used with forecasted independent variables to predict future crashes. The model predictions for all TAZs are then multiplied by the unbiased BCF computed previously to obtain the ‘best’ estimate of crashes in the future/scenario forecast. These estimates reflect the forecast of the PLANSAFE models adjusted for local/regional conditions. 5. Compare BCF coefficient of variations: To assess goodness of fit of the PLANSAFE model or to compare goodness of fit of several models, the TAZ level BCFs are used. The coefficient of variation (CV)—the standard deviation of BCF divided by the average BCF (not the unbiased BCF) gives a measure of the unexplained crash variation from the PLANSAFE model. A CV near zero suggests that the model fits the observed data perfectly, a CV equal to 1 suggests that there the standard deviation is as large as the mean, and CV values much greater than 1 would suggest that there is significant unexplained variation in the local data. CV values equal to or greater than 1 may indicate a problem of lack of model fit, and preferably values considerably less than 1 are preferred. 6. Incorporate modelling results into planning process: The modelling results are now used to inform decision making in the transportation process. The modelling results provide a prediction of expected safety in a TAZ, a collection of TAZs, or an entire region because of growth in various forms. Growth can affect population, road mileage, and intersection density. The modelling results provide the planner information about the expected future safety, assuming that similar roadway design standards and no new safety initiatives are implemented. Using the forecasts, the 141

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools planner can then estimate how much safety investment is needed to attain regional or project level safety targets. Example: Application of PLANSAFE: Incapacitating and Fatal Injury Crashes Using the 6-step procedure described previously and the PLANSAFE incapacitating and fatal injury models, an application of the models is illustrated. UStep 1 U: An analyst has decided to apply the PLANSAFE Incapacitating and Fatal Injury Crash Frequency Model to make predictions across 10 TAZs within a jurisdiction. A major corridor improvement is being considered, which will bring about new residential and commercial development to the 10 TAZs, as well as traffic volumes and associated activity. A host of new intersections will be added because of the project, as well as new road mileage. Of course, interest focuses on what changes to safety are anticipated as result of this project—assuming similar road designs and no innovative safety countermeasures. The baseline data are shown in Exhibit 96 for the 10 TAZs under consideration. The count of incapacitating and fatal crashes in the base year is known, whereas incapacitating and fatal crashes will be predicted for the future ‘project build out’ year. Increases in road mileage and intersections are forecasted for each TAZ as a result of the major project, as shown in TExhibitT 96. TAZ NUMBER INT_PMI PNF_0111 PNF_0512 POP00_15 Base Year Data for Existing Conditions 1 1 0.12 0.15 2500 2 4 0.09 0.12 6500 3 5 0.12 0.16 2780 4 2 0.17 0.2 8000 5 4 0.03 0.04 5400 6 6 0.023 0.035 2000 7 2 0.095 0.1 3526 8 1 0.045 0.06 4578 9 2 0.014 0.025 3278 10 7 0.021 0.3 6900 Data for Future Conditions at Implementation of Planned Project 1 3 0.15 0.15 6500 2 5 0.09 0.15 10000 3 6 0.15 0.16 6400 4 2 0.17 0.25 12000 5 5 0.03 0.04 5400 6 7 0.028 0.044 2600 7 4 0.095 0.1 3526 8 3 0.045 0.075 4578 9 4 0.018 0.025 9500 10 7 0.021 0.3 6900 UStep 2 U: MS Excel is used to set up a spreadsheet equation for predicting crashes in the baseline and future years. The appropriate prediction equation for the PLANSAFE Incapacitating and Fatal Injury Model is given as frequencyaccidentFatalandingIncpacitat ____ ( ) ( ) ( )0512_674.30111_328.3_ 10.6590-257.2exp( 1 PNFPNFPMIINT ++×= − ( ) ) 115_0010512.0 3 −×+ − POP . Exhibit 96: Base Year data for PLANSAFE example application 142

Incorporating Safety into Long-Range Transportation-Planning Each of the independent variables is used to forecast the base and future year scenarios. Of course, the impact of the future scenario on the value of the independent variables needs to be forecast. In this particular model the number of new intersections, road mileage of various types, and new population aged 0 to 15 are needed to forecast crashes. Step 3: The (asymptotically unbiased) BCF is calculated to be 1.594 for the base year conditions and is shown in Exhibit 97. The table also shows the BCF calculations across the 10 TAZs impacted by the major project. The spreadsheet is used to calculate predicted crashes, and the BCF is calculated for each TAZ prediction. Then, the average and standard deviation of the BCF is calculated for the PLANSAFE Incapacitating and Fatal Injury Model to assess model fit. If multiple models are being considered (say a fatal and incapacitating fatal and injury model), then the coefficient of variations of the BCFs should be compared to see if one prediction model is significantly outperforming another—the significantly smaller coefficient suggesting better fit of the PLANSAFE model to the local data. In this example, BCF CV is about .18 or 18%, which means that the standard deviation is about 18% of the mean value. The unbiased average BCF reflects the average bias between the PLANSAFE Incapacitating and Fatal Injury Model and the region where the model is applied. In this example, the PLANSAFE model is under-predicting incapacitating and fatal injury crashes, on average, by a factor of about 1.6. This under-prediction is the result of multiple potential factors that are not included in the prediction models, including differences in weather (e.g., wet, ice, snow, and fog conditions), driver population differences, and other factors between the application and calibration data. The BCF, therefore, is used to adjust pedestrian crash predictions in future years, which would otherwise be biased low in this particular example. Base Year Data for Status Quo TAZ Observed Crashes Predicted Crashes BCF 1 4 3.4207 1.169 2 8 5.0598 1.581 3 5 3.3369 1.498 4 10 6.5194 1.534 5 7 4.0033 1.749 6 3 2.0798 1.442 7 8 5.9589 1.343 8 8 3.8539 2.076 9 6 2.9276 2.049 10 9 5.4950 1.638 Totals 68 42.6552 unbiased BCF 1.594 average BCF 1.607 std.dev. BCF 0.287 CV BCF 0.179 Exhibit 97: BCF calculations for PLANSAFE example application Step 4: The PLANSAFE Incapacitating and Fatal Injury Model is applied again to forecast future pedestrian crashes under the project scenario. To make these forecasts, future values of explanatory variables, road mileage and intersections, are forecasted using knowledge of the project and its impact on these variables. Although in this example these variables are provided, considerable discussion and additional modelling may be required to forecast the predictor variables. The forecasted incapacitating and fatal crash frequencies (derived from the PLANSAFE model) are Appendix C: Safety Tools 143

Incorporating Safety into Long-Range Transportation-Planning Appendix C: Safety Tools multiplied by the unbiased BCF to obtain the forecasted estimate of incapacitating and fatal crashes in the future project scenario. Exhibit 98 shows the results of this step. New road mileage, intersections, and additional residential development will lead to predicted increases in crashes in the absence of new, innovative, or progressive safety interventions. In other words, if the project is built using roadway design standards commonly used in the base year, an increase in incapacitating and fatal injury crashes from 68 observed crashes to 87 crashes is expected. Exhibit 98: Predicted future incapacitating and fatal crashes for PLANSAFE example application TAZ Predicted Project Scenario Crash Frequency BCF Adjusted Project Scenario Crash Frequency 1 5.70 1.594 9.09 2 7.39 1.594 11.79 3 5.36 1.594 8.54 4 9.02 1.594 14.37 5 4.34 1.594 6.91 6 3.28 1.594 5.24 7 3.83 1.594 6.11 8 3.84 1.594 6.13 9 6.25 1.594 9.96 10 5.76 1.594 9.18 Total 87.31 The increase in expected crashes that results from the project is not an argument in of itself for or against the project, and in fact is merely an informative statement regarding safety and not a value statement about safety. That incapacitating and fatal crashes are predicted to increase from 68 to 87.31 merely represents an increase in injury severity risk expected by increases in the number of intersections, residential development, road mileage, and local population increases. Steps 5 and 6. Since only one PLANSAFE model is being considered in this example, a comparison of CVs across models is not appropriate in this case (in addition the CV is considerably less than 1). The results of this analysis might be coupled with the analysis of other projects to compare and contrast the expected change in safety. For example, one might apply the same procedure using a total crash model and the fatal injury model. If the CVs are comparable across models then all models might be used; however, if one of the CVs is considerable lower than the other models it might be preferred for prediction than the models with relatively high CVs. The information provided through this analysis suggests that the project will bring about a sizeable increase in incapacitating and fatal crashes because of the project and new population growth. If a regional safety goal was to reduce fatal and incapacitating injury crashes by 20%, then one would obtain a target for this project using [87.31*0.80] = 69.84 crashes. The next step would be to examine design policies and safety investments, using software such as the IHSDM (for example) to seek reductions in crashes. To meet the regional safety goal the project would need to demonstrate through additional safety investments and strategies an expected reduction of [87 – 70] = 17 fatal and incapacitating injury crashes. A point of note here is that a statewide or regional safety objective of a 20% reduction in incapacitating and fatal crashes does not correspond with a reduction of the current level of incapacitating and fatal crashes, because growth in population and other factors will necessarily lead to increases in crashes in most cases. Thus, the PLANSAFE Incapacitating and Fatal Crash Model in this example provides planners 144

Incorporating Safety into Long-Range Transportation-Planning with a tool for setting targets for meeting safety objectives and performance milestones. Appendix C: Safety Tools 145

Incorporating Safety into Long-Range Transportation-Planning Page left intentionally blank. Appendix C: Safety Tools 146