Developing a Guide for Quantitative Approaches to Systemic Safety Analysis (2020)

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.

13 Summary of Literature Review A number of highway agencies, both in the United States and internationally, have been implementing systemic safety management approaches for several years. This section summarizes how agencies in the United States and abroad have implemented systemic safety management approaches within their agencies. Domestic applications are first described, followed by details of international applications. As part of this research, the research team also reviewed several guidebooks and other key resources related to systemic safety analysis. These guidebooks and key resources are summarized in the comprehensive guidance document on quantitative approaches to systemic safety analysis and are not summarized here. 2.1 Domestic Applications of Systemic Safety Analyses Systemic Safety Management of Low-Volume Rural Roads in Iowa Souleyrette et al. (2010) conducted a crash analysis of low-volume roads in Iowa, defined as paved or unpaved, undivided, two-lane roads located in rural areas with daily traffic volumes less than or equal to 400 vehicles per day (vpd). The study focused on both paved and unpaved low- volume local roads (or secondary roads) and compared the safety performance of these types of roads to each other and to the state maintained, paved, two-lane roads that carry mostly higher volumes and benefit from a more consistent maintenance and traffic control policy. Souleyrette et al. addressed the problem of rural low-volume road safety by initiating an in-depth database investigation of the history of total and fatal and serious-injury crashes on very low- volume rural roads, paved and unpaved, in Iowa. The project consisted of three principal tasks: statewide statistical assessment, statistical modeling, and site-specific field review. Descriptive statistics of available crash data (2001-2008) were prepared for rural low-volume (≤ 400 ADT) secondary road crashes, as well as for all other two-lane rural road crashes for comparison purposes. To identify commonalities among the crashes, descriptive statistics were developed for several major crash characteristic categories including:  Month  Day  Hour  Crash severity  Crash type  Terrain  Manner of crash/collision  Light condition  Weather condition  Surface condition  Location of first harmful event  First harmful event  Major cause  Environment contributing circumstances  Roadway contributing circumstances  Type of roadway/junction/feature  Younger/older driver involvement  Drug/alcohol involvement  Farm vehicle involvement Distributions of low-volume rural secondary road crashes were compared to crashes on rural primary roads to find the main factors that increase the potential for a crash on low-volume roads compared to primary two-lane roads. Tests of proportions were employed to determine various

14 crash characteristics overrepresented on different low-volume road categories. The comparison group was primary road crashes. Proportions of various characteristics of crashes on roads of different jurisdictions, traffic volume ranges, and surface types were computed and then statistically tested to determine if differences between pairs of proportions were statistically significant. The results identified the types of crashes that are particularly problematic for low- volume roads and served as the basis for the development of crash-based statistical models. Based on the test of proportions, Souleyrette et al. developed system-level models for secondary, low-volume roads using crash, driver, and roadway variables that were the best predictors of the severity of low-volume road crashes. Annual average daily traffic (AADT) and road system type were the two basic criteria utilized in setting up groups of roads for modeling, and three thresholds of daily volume and two types of jurisdictional road systems were used to categorize roads and assign crashes to them, respectively. AADT values were also used in the process to determine candidate crash sites to review and evaluate. To find the most significant factors of all crashes on rural roads in Iowa with 400 vpd or less between 2001 and 2008, Souleyrette et al. used an ordered probit model. The classes of low- volume local agency roads included 0-100 vpd paved and unpaved, 0-100 vpd all, 101-400 vpd paved and unpaved, 0-400 paved and unpaved, and all local rural roads with traffic volumes of 0-400. The DOT primary comparison roads ranged in traffic volume of 70-12,200 vpd and are all paved. The models were based on crashes rather than locations, so their application was limited to the identification of important contributing factors to crashes on low-volume roads rather than the identification of high-crash locations. Comparing the frequency of crash attributes in each local road class to those in the primary road group and testing for statistical significance using a test of proportions, Souleyrette et al. concluded the following: 1. Almost all local road classes exhibit a higher frequency of injury crashes than primary roads. 2. Single-vehicle, run-off-road crashes occur at a higher frequency on local roads than multi-vehicle events, although multi-vehicle broadside crashes occurred more often on local roads. 3. Animal-related crash frequencies were lower on local roads than primary roads. 4. Speed-related crashes were higher for all classes of local roads but lower for weather- related crashes. 5. Younger driver involvement in local road crashes was higher for most local road classes but older driver crashes were lower. 6. Impaired driving crashes occurred more frequently on local roads than on primary roads. 7. Crashes in rolling or hilly terrain were more frequent on local roads, but less on flat terrain. 8. When farm vehicles were involved, crash frequency was higher on local roads than primary roads.

15 9. Local road crashes occurred at a higher frequency in the summer and fall and were lower in winter months. Also local roads crashes had a higher frequency on weekends but lower during the week than primary roads. 10. Local road crashes were more frequent during nighttime hours and higher on unpaved roads during the day, but all classes of local road crashes were lower from about 5 p.m. to 9 p.m. and from 4 a.m. to 7 a.m. 11. Non-collision crashes involving a rollover or overturn showed a higher frequency for all local road classes, especially on unpaved roads. 12. On local roads, collisions with fixed objects involved culverts, ditch/embankment, trees, or poles indicated a higher frequency of occurrence on local roads, but lower than primary for guard rail collisions. 13. Major cause of crashes on local roads was higher than primary roads for failure to yield at uncontrolled intersections and driveways. Too fast for conditions and swerving or evasive action also showed a higher frequency on local roads, but left-turn crashes, following too close, and crossed-centerline crashes had a lower frequency on local roads. 14. Lower-volume roads exhibited less contribution to crashes from adverse weather or undesirable pavement surface conditions than did primary roads. Roadway contributing circumstances such as ruts/holes/bumps showed a higher frequency as did deficient traffic control devices on local roads, although the latter category was quite low in occurrence. Shoulder conditions contributed to local road crashes at a higher frequency than on primary roads. 15. Crashes on local roads occurred at a higher frequency at bridges, railroad crossings, farm or residence driveways, as well as T or Y configuration intersections but at a lower frequency at four-way intersections. The study findings confirmed several common assumptions regarding younger driver involvement, speed related, and impaired driving on local roads, especially unpaved roads. Due to the random nature of crashes on low-volume rural roads, Souleyrette et al. indicated that a systemic safety management approach to crash mitigation would be beneficial. Using the study results, major common crash contributing factors can be determined and mitigative steps taken without relying on the development of crash history. Types of mitigative countermeasures to consider might include improved signing at shorter radii horizontal curves, delineation of roadsides, education of younger drivers to address unique safety concerns on unpaved roads, and focused enforcement for undesirable driver behavior, such as speeding and impaired driving. Application of FHWA Systemic Safety Project Selection Tool and usRAP in Iowa Knapp et al. (2014) applied the FHWA Systemic Safety Project Selection Tool approach and usRAP tool and collected data from two counties in Iowa to evaluate systemic safety improvement crash contributing factors and countermeasures. The first task of the research was a review of five systemic safety tools and methodologies, including:

16  The Minnesota County Road Safety Plan (CRSP) approach  The FHWA Systemic Safety Project Selection Tool  usRAP  The New Jersey systemic road safety tool  Safety Analyst software The five tools/methodologies were summarized and compared through the consideration of five attributes: (1) general availability (e.g., cost, etc.), (2) required input data, (3) ease of use, (4) basis of prioritization, and (5) the potential to provide sensitivity analysis insight. Knapp et al. indicated that the data requirements and cost of these tools/methodologies varied widely; all of them used “stars” (i.e., number of safety crash contributing factors) and/or benefit-cost ratios for the prioritization of locations; the tools or methodologies were relatively easy to apply; and most of the tools or methodologies seemed likely to provide some insight if they were evaluated with a sensitivity analysis. Knapp et al. selected the Minnesota CRSP and usRAP approaches for further evaluation. Data were collected for 353 mi of paved secondary (i.e., county) roadway within two counties in Iowa to apply the Minnesota CRSP and usRAP approaches. Data were collected at the regional and county level to define the crash contributing factors used in the Minnesota CRSP approach. In addition, data were collected at approximately 80 horizontal curves, 50 stop-controlled intersections, and 58 roadway segments in each county to compare crash contributing factor criteria and the number of criteria met at each location. Overall, five crash contributing factors were considered at each horizontal curve, seven at each stop-controlled intersection, and five at each roadway segment. The locations were then ranked by the number crash contributing factors met, and rules were set to determine the rank of each location. To apply the usRAP methodology, data were collected that described 40 to 50 roadway characteristics for every 328 ft (100 m) segment. Knapp et al. also brought together a focus group to identify potential crash contributing factors for unpaved roadways. Twenty-four potential crash contributing factors were identified related to the environment, roadway design or infrastructure, driver behavior and education, and roadway maintenance. Many of the potential crash contributing factors for unpaved roadways also applied to paved roadways but others did not (e.g., dust and surface characteristics). Also, the impact these factors have on safety likely differs along paved and unpaved roadways. The final task of the project was a sensitivity analysis. Input variables were changed in each tool or methodology (e.g., Minnesota CRSP and usRAP), and the results were compared. Three approaches were used to adjust the coefficients or “weights” of the crash contributing factors applied within the Minnesota CRSP approach. To compare the results, Knapp et al. considered the percentage change in the “top 20” locations ranked and also calculated the Kendall tau-b correlation coefficient to perform a statistical comparison of the initial and alternative rankings. Overall, the percentage change in “top 20” locations ranged from zero to 50 percent, with a typical average of 10 to 20 percent (i.e., two to four locations). Knapp et al. concluded that this type of shift could affect the decision making based on the Minnesota CRSP results. However, the statistical comparison using the Kendall tau-b correlation coefficient indicated that the initial and alternative ranking lists were similar in all cases. The difference in results was not entirely surprising considering that the “top 20” shift only considered the changes that might have an impact on decision making and the statistical evaluation considered the entire ranking list.

17 For the sensitivity analysis using usRAP, the type and number of countermeasures suggested by usRAP were compared for acceptable benefit-cost ratios of both one and two. Not surprisingly, the increase in acceptable benefit-cost ratio from one to two resulted in the suggestion of fewer countermeasures and produced a higher overall benefit-cost ratio for the group of countermeasures suggested. In addition, the number of sites or mileage suggested for the implementation of countermeasures decreased. Based on the research results, Knapp et al. suggested the following: 1. The approach followed with this research should be expanded to counties with higher variability in roadway characteristics, and a wider range of “weights” and more specifically defined crash contributing factors might be applied. 2. When identifying crash-contributing factors, they should meet two objectives: (a) the identification of locations with characteristics known to affect rural roadway safety, and (b) the differentiation of locations with relatively unique crash contributing factors or combinations of factors. 3. The variability in roadway characteristics connected to potential crash contributing factors should be considered during their selection. 4. A research project should be conducted to specifically define the weights of crash contributing factors. 5. Ranking changes that result from alterations in crash contributing factor weights should be evaluated when applied. 6. More research should be conducted on the selection and application of crash contributing factors along gravel and/or rock roadways, the selection of crash contributing factors for paved roadways, the development of ranking “tiebreakers,” and the proper evaluation of systemic safety program applications. Systemic Safety Prioritization Method for Pedestrian and Bicycle Crashes in Oregon Oregon Department of Transportation (ODOT) applied a systemic safety management approach to prioritize corridors with the most potential for reducing pedestrian and bicycle crashes (Bergh et al., 2015). The two primary steps in the methodology included:  Identify crash contributing factors (i.e., roadway or location characteristics) that were present at locations where crashes were reported from 2007-2011.  Identify and prioritize locations within the state where one or more crash contributing factors are present. Crash contributing factors include a range of roadway characteristics that appear to be associated with higher frequencies of fatal and serious-injury pedestrian or bicycle crashes. Pedestrian and bicycle volume were not considered in the methodology due to the lack of consistent, statewide data. The systemic safety network screening approach was complemented by a secondary network screening method outlined in HSM Part B to account for locations with known history

18 of fatal and serious-injury pedestrian and bicycle crashes. Results from both screening methods were combined to prioritize candidate project corridors. The systemic analysis was consistent with Steps 1 through 4 of the FHWA Systemic Safety Project Selection Tool process: 1. Identify focus crash types and contributing factors. 2. Screen and prioritize candidate locations. 3. Select countermeasures. 4. Identify specific projects at prioritized candidate locations. The focus crash types of this study were defined as pedestrian and bicycle crashes. Crash contributing factors developed by this study were roadway or location characteristics that could contribute to a pedestrian or bicycle crash and were present at locations where target crashes were reported. Based on 2007 through 2011 crash data, the primary risk factors for pedestrian and bicycle crashes included: Pedestrian Crash Contributing Factors: Bicycle Crash Contributing Factors:  Presence of transit stops  Undivided 4-lane roadways in urban areas  Presence of traffic signals  Presence of pedestrian activated flashers or beacons  Posted speed limit  Average daily traffic  Reported crash frequency and severity  Driveway density  Undivided 4-lane roadways in urban areas  Lack of bicycle facility on at least one side of the roadway  Presence of traffic signals  Posted speed limit  Presence of transit stops  Reported crash frequency and severity Having identified pedestrian and bicycle crash contributing factors, ODOT screened the state and urban non-state networks separately to prioritize locations where multiple risk factors are present. The network screening process involved five general steps:  Develop and segment (0.10 mi increments) a roadway network including information about facility characteristics.  Develop a scoring process to apply to each segment, reflecting the relative potential of each factor contributing to a crash.  Apply scoring to prioritize individual roadway segments.  Combine individual priority roadway segments into candidate project corridors.  Develop list of prioritized candidate project corridors based on average corridor score. There was no quantitative data associating the identified crash contributing factors and crash frequency so a subjective scoring system was developed to account for combinations of factors.

19 Table 1 and Table 2 illustrate the segment scoring system developed to rank the priority of sites based on the presence pedestrian and bicycle crash contributing factors. ODOT applied the scoring methodology to rank 9,490 segments (0.1 mi each) on the state network. Individual segments were prioritized based on their total score and grouped into longer candidate project corridors based on the scores of up- and downstream segments within 0.5 miles. Within each region, sites were prioritized into project corridor lists using systemic and/or frequency/severity-based screening methods. Projects focusing on pedestrian and bicycle crashes were prioritized separately, as each requires different types of mitigation countermeasures. ODOT local agency staff can use the prioritized corridors and summary of crash contributing factors to identify appropriate countermeasures for individual projects. Table 1. Systemic Segment Scoring Worksheet for Pedestrian Crash Contributing Factors (Bergh et al., 2015) Pedestrian Segment Data Roadway Name: Start (Nearest Cross-Street or MP): End (Nearest Cross-Street or MP): Length (mi): 0.1 Jurisdiction: Instructions: Please complete this form for each 0.1 mi long segment within the area you would like to evaluate. To develop a score for the corridor, calculate the average score of the corridor by dividing the sum of all scores by the number of 0.1 mi long segments within the corridor. Use the final corridor score to prioritize projects Contributing Factor Data Score Score Methodology Is at least 1 traffic signal located with 100 ft of the segment? 1 point if at least 1 signal is located on the segment or within 100 ft of the segment How many transit stops are located within 100 ft of the segment? 1 point for segments with 1 transit stop located on the segment or within 100 ft of the segment; 2 points for 2 or more transit stops Is there one or more pedestrian activated beacons or flashers located on the segment? 1 point subtracted (rewarded) for the presence of an enhanced midblock crossing What is the posted speed limit? 2 points for posted speed limit of 35 or 40 mph; 4 points for posted speed limits above 40 mph Is the corridor an undivided, 4-lane segment? 2 points if segment is an undivided 4- lane segment AADT of corridor 2 points for AADT between 12,000 and 18,000; 4 points awarded for AADT above 18,000 Number of minor or moderate injuries resulting from pedestrian involvement* 2 points if a non-severe injury was reported; 1 additional point for each additional injury Number of pedestrians involved in a crash but not injured* 2 points if a pedestrian is involved in a crash but not injured; 1 additional point for each additional pedestrian involved but not injured Number of severe injuries resulting from pedestrian involved crashes* 4 points awarded if a severe injury was reported; 2 additional points awarded for each additional injury Number of fatalities resulting from pedestrian involved crashes* 4 points awarded for fatalities Total Score * All crash history should reflect the latest five years of available data

20 Table 2. Systemic Segment Scoring Worksheet for BIcycle Crash Contributing Factors (Bergh et al., 2015) Bicycle Segment Data Roadway Name: Start: End: Length (mi): 0.1 Jurisdiction: Instructions: Please complete this form for each 0.1 mi long segment within the area you would like to evaluate. To develop a score for the corridor, calculate the average score of the corridor by dividing the sum of all scores by the number of 0.1 mi long segments within the corridor. Use the final corridor score to prioritize projects Contributing Factor Data Score Score Methodology Is at least 1 traffic signal located with 100 ft of the segment? 1 point if at least 1 signal is located on the segment or within 100 ft of the segment Is the corridor an undivided, 4-lane segment? 2 points if segment is an undivided 4- lane segment How many transit stops are located within 100 ft of the segment? 1 point for segments with 1 transit stop located on the segment or within 100 ft of the segment; 2 points for 2 or more transit stops Does the left side of the road have a bicycle facility 2 points awarded for the lack of bicycle facility on the left side of the road Does the right side of the road have a bicycle facility 2 points awarded for the lack of bicycle facility on the right side of the road AADT of corridor 2 points for AADT between 12,000 and 18,000; 4 points awarded for AADT above 18,000 What is the posted speed limit? 2 points for posted speed limit of 35 or 40 mph; 4 points for posted speed limits above 40 mph How many driveways or alleys are located on the corridor? 2 points awarded for segments with 1 driveway; 3 points for segments with 2-3 driveways; 4 points for segments with 4 to 8 driveways; 5 points for segments with more than 8 driveways Number of minor or moderate injuries resulting from bicyclist involved* 2 points if a non-severe injury was reported; 1 additional point for each additional injury Number of bicyclists involved in a crash but not injured* 2 points if a bicyclist is involved in a crash but not injured; 1 additional point for each additional bicyclist involved but not injured Number of severe injuries resulting from bicyclist involved crashes* 4 points awarded if a severe injury was reported; 2 additional points awarded for each additional injury Number of fatalities resulting from bicyclist involved crashes* 4 points awarded for fatalities Total Score * All crash history should reflect the latest five years of available data Systemic Safety Management Approach in Pennsylvania Aguero-Valverde et al. (2015) conducted a study of rural, two-lane highways in Pennsylvania to identify sites with promise (SWiPs) using a variety of multivariate and spatial crash frequency models. Four models, with and without multivariate and spatial correlations, were investigated to determine model precision. After comparison of models, it was observed that the multivariate correlation and spatial correlations were considered to give the best fit models. Additionally, to

21 yield better crash frequency models in terms of different crash types, it was established that multivariate correlations plays a stronger role than the spatial correlation. The results yielded the number of crashes, by crash type, for each rural, two-lane roadway segment. For some models, the results indicate that an increase in traffic volume does not always correlate to an equal increase in crash types. The availability of better crash frequency models would ideally help identify the most deserving SWiPs, which allows corresponding countermeasures to be systemically implemented to mitigate the specific target crash types, alongside the traditional “hot-spot” approach. Systemic Safety Improvements at Intersections in South Carolina In recent efforts to improve the safety performance at both signalized and stop-controlled intersections throughout the State of South Carolina, the South Carolina Department of Transportation (SCDOT) elected to take a systemic safety management approach (Le et al., 2017). SCDOT implemented a series of low-cost intersection treatment packages, including improvements to intersection signage and roadway markings, as well as replacing and upgrading traffic and pedestrian signal heads. These treatments were implemented at more than 500 intersections, including urban and rural signalized and unsignalized intersections, with both three- and four-leg configurations. To assess the effectiveness of the implementation of these systemic safety treatments on South Carolina’s roadway network, Le et al. conducted a safety evaluation as part of the FHWA Evaluation of Low Cost Safety Improvements Transportation Pooled Fund study. Before and after crash and network data were analyzed, and the results of the evaluation were summarize in a series of CMFs (see Table 3). Table 3. CMFs of Systemic Safety Improvements at Intersections in South Carolina (Le et al., 2017) Crash Type Signalized Intersections Stop-Controlled Intersections EB Estimate of Crashes for After Period Observed Crashes in After Period CMF Std. Error of CMF EB Estimate of Crashes for After Period Observed Crashes in After Period CMF Std. Error of CMF Total 2,801 2,675 0.955 0.023 4,614 4,231 0.917 0.017 Fatal & Injury 617 551 0.893 0.045 1,434 1,290 0.899 0.028 Rear- End 1,385 1,349 0.974 0.034 1,577 1,472 0.933 0.03 Right- Angle 1,042 921 0.883 0.035 1,955 1,840 0.941 0.026 Night- Time 599 581 0.969 0.048 1,072 953 0.853 0.031 To conduct the evaluation, Le et al. developed a set of SPFs for each intersection configuration and crash type configuration. These SPFs were used in the development of the resultant CMFs, comparing the expected numbers of crashes in the after period without the treatments to the observed numbers of crashes in the after period with the treatments. The evaluation showed crash reductions in all studied crash types, indicating notable positive impacts of the low-cost systemic safety implementation.

22 South Dakota Rural Road Safety Index Mahgoub et al. (2011) conducted a study to improve the safety performance of local rural roads in South Dakota by analyzing the crash occurrence and potential safety treatments. One of the primary objectives was to develop a rural road safety index to rank roadway network locations by their safety features to identify deficiencies. Mahgoub et al. focused on developing rural road safety index for rollover and fixed object crashes as the data for South Dakota showed that rollover and fixed object crashes represented 43 percent of the crashes and 72 percent of the injuries and fatalities on local rural roads. Field studies were conducted on 26 local roads in Brookings County (SD) to identify safety issues possibly contributing to crashes along local rural roads. All of the roads were low-volume, unpaved rural arterials or collectors with right-of-way (ROW) typically between 40-60 ft. Most of the sections had 30-55 mph speed limits. Table 4 shows the list of safety issues considered during the field visits to flag features that need to be investigated further to determine what, if any, action should be taken. The list is not intended to be all-inclusive but to serve as a starting point. Table 4. Safety Questions Considered During Site Visits (Mahgoub et al., 2011) ITEM SAFETY QUESTIONS Road Overview Are there changes in land use and/or traffic or other environmental challenges such as terrain? Crash History Is there a history of crashes that points to specific problems areas? Road Alignment & Cross Section How well does the roadway serve current and future traffic? Are lane width, shoulders, and sight distances adequate? Roadside Features Are there steep slopes, drainage features, narrow shoulders or clear zones, or fixed objects (narrow bridges, mailboxes, utility poles)? Are guardrails adequate and meet standards? Gravel Road Surface Conditions Are road surfaces well maintained (proper shape, smooth surfaces, loose gravel, or edge drop- offs)? Paved Road Surface Conditions Is surface smooth, adequate skid resistance, free of edge drop-offs? Signing & Pavement Marking Are signs and pavement markings well maintained and meet the requirements of the Manual on Uniform Traffic Control Devices (MUTCD), including nighttime visibility? Intersections & Approaches Are there sight restrictions (vegetation or other) that limit visibility? And is signing adequate? Railroad Crossings Are crossings properly signed and free of sight restrictions? Pedestrians & Bicycles Are crossings clearly signed and marked, and are there areas of pedestrian activity (schools, playgrounds, parks) in need of special considerations? Provisions for Heavy Vehicles Are there operational issues due to the presence of heavy commercial or agricultural vehicles? Mahgoub et al. (2011) used the safety priority evaluation matrix to subjectively rate the potential for a crash of each site based on relative exposure, probability, and consequence of safety issues, as well as to discuss the reasons to help establish consensus on priorities. Figure 1 illustrates a sample safety priority evaluation matrix for an intersection location. Exposure refers to the number of opportunities of an event (such as a crash), generally in proportion to traffic volume. Probability refers to the likelihood of a crash occurring, and consequence refers to the severity of the crash.

23 Figure 1. Sample Safety Priority Evaluation Matrix for an Intersection (Mahgoub et al., 2011) Mahgoub et al. then developed a Rural Road Safety Index (RRSI) to rank the road network according to the safety features and identify the deficiencies in road sections with the intended purpose to help solve road safety concerns before they contribute to crashes. The RRSI provided a technique to quantify the safety gains that could be achieved by addressing the issues identified in the review process. To place a meaningful index on the safety performance of individual sections of the local rural road network, a value needed to be assigned. The safety effect was expressed by two indices, the estimated relative increase in crash probability, and the estimated relative increase in crash severity caused by the various safety issues. Five safety issues were incorporated into the RRSI. Each safety issue is graded on a scale from 4 to 1, with 4 being the best (no treatments are needed) and 1 being the worst where some remedies are expected to be applied. Table 5 provides guidance for determining the value of RRSI. For each 500 ft segment of roadway, the review team scored the detailed safety issues according to the ranking system. Deducted points are added for each section and subtracted from 100 to determine the RRSI value for the section according to the following equation: RRSI Section = 100 - Σ DP where: DP = Deduct Points

24 Table 5. Safety Items for Quantifying the RRSI (Mahgoub et al., 2011) Roadside Obstacles 30 Points Rank 1 Rigid utility poles, rigid obstacles, inadequate bridge rails ≥ 5 times 30 Rank 2 Rigid utility poles, rigid obstacles, inadequate bridge rails ≥ 3 times 20 Rank 3 Rigid utility poles, rigid obstacles, inadequate bridge rails ≥ 1 time 10 Rank 4 No roadside obstacles within the whole section. 0 Signs and Delineation 10 Points Rank 1 Curve warning missing or ineffective on severe curve 10 Rank 2 Guideposts or barrier reflectors damaged or missing 7 Rank 3 Some curve warning missing or inadequate 5 Rank 4 No deficiencies on roadside signing 0 Cross Section 20 Points Rank 1 Lane width < 9 ft. / no shoulder 20 Rank 2 9 ft. < lane width ≤ 10 ft / no shoulder 15 Rank 3 9 ft. < lane width ≤ 10 ft. / sufficient shoulder 10 Rank 4 Lane width > 10 ft / sufficient shoulder 0 Alignment and Accesses 30 Points Rank 1 Sight distance problems present ≥ 5 times 30 Rank 2 Sight distance problems present ≥ 3 times 20 Rank 3 Sight distance problems ≥ 1 time 10 Rank 4 No sight distance problems within the whole section 0 Road Surface and Maintenance 10 Points Rank 1 Very poor surface & maintenance, corrugation, and pot holes 10 Rank 2 Presence of corrugation and pot holes 7 Rank 3 Slightly deteriorated roads 5 Rank 4 Good and very good surface 0 After calculating the RRSI for the study locations in Brookings County, the major issues identified by Mahgoub et al. were related to intersection sight distance, angle of approach, signage, road alignment, vertical and horizontal curvature, culverts, table drains and location of signage, visibility and legibility of signs. If a non-frangible object on the roadside presented an unreasonable potential to cause harm to road users, it should be considered for treatment. Acceptable treatments in descending order of preference included:  Removal of the object to create a more forgiving roadside.  Relocation of the object.  Replacement of the object with a more visible and frangible type.  Shielding the object with safety barriers.  Enhancing the visibility of the object.  Warning road users of the object. With the help of RRSI, Mahgoub et al. noted that it would be easier for local governments to prioritize road safety and maintenance work. Systemic Safety Management Approach of Pedestrians Crashes in Texas Wang et al. (2016) conducted a literature review and study to investigate the limitations of current systemic pedestrian safety analysis methods and explore the relative effectiveness of roadway facility ranking methods using primary crash contributing factors. In particular, the study analyzed pedestrian crashes related to urban intersections using a crash data set from

25 Austin, Texas. Though the study proved ultimately to be inconclusive, several important insights were drawn from the research process. The study found that a few challenges tend to be encountered during systemic pedestrian safety analyses. First, due to the historic focus of engineering solutions on improved vehicular safety as opposed to the safety of other vulnerable users, current practices often do not address pedestrian safety and some measures even at times adversely affect it. Second, pedestrian crashes tend to be more sporadic and rare, often making it challenging to identify patterns in crash data. Finally, pedestrian crash patterns tend to have a complex relationship with the attributes of each facility’s surrounding area and related infrastructure, making it essential that analysts properly account for these contributing factors. While some reliable methodologies have been developed for identifying and analyzing pedestrian crash contributing factors, relatively few methodologies have been developed which utilize these contributing factors to reliably rank roadway facilities. To begin addressing this issue, this study investigated two unique roadway facility ranking procedures, analyzing their relative sensitivities to the underlying ranking methods and to different contributing factor weighting schemes. In particular, a grouping-based method was studied as well as a weighted- average method which utilized both negative binomial-based coefficient estimates and crash histories. A comparison between the two showed that, though, both appear to be sensitive to the applied weights used in the procedures, the grouping-based ranking method combining arbitrary weights produced relatively stable results. Though this study produced helpful insights into possible innovations in the methodologies used in systemic pedestrian safety analyses, the authors note that additional research is necessary. Key technical questions still remain and limit the conclusiveness of this study; however, the research furthered the industry’s understanding of these complex topics and opened doors for additional work to be conducted. Systemic Safety Management Approach of Extended Highway Corridors in Virginia In recent years, approximately one-quarter of the crashes in Virginia occurred on two-lane rural roadways (Tsyganov and Read, 2017). These rural corridors make up approximately 80 percent of the state’s roadway network with a total of more than 44,000 centerline miles. Due to the nature of such roadways, their crash distributions tend to be highly random, making traditional “hot-spot” treatment methods for highway safety relatively ineffective. This necessitates the development of a more broadly scoped, systemic method for the state to address such crashes. To help alleviate this issue of high-crash frequencies on two-lane rural roadways throughout the state, the Virginia Department of Transportation (VDOT) developed a systemic safety assessment approach for extended highway corridors, a procedure named “Corridor Safety Assessment” (CSA). The method was developed for their HSIP to assist in the identification of effective and appropriate systemic safety treatments and the promotion of a mobile methodology which empowers field review teams to safely study large network areas. The CSA method considers the entire network as opposed to being specific location-based. It utilizes crash statistics, roadway network data, as well as expert field evaluations to conduct a

26 detailed, yet extensive assessment of the entire travel environment. The CSA method takes into account characteristics of the traveled way, the roadside environment, as well as all relevant traffic control features. The approach allows for the consideration of a broad set of countermeasures, encompassing both short-term low-cost improvements as well as long-term high-cost improvements. This flexibility as well as the inclusion of a variety of crash contributing factors relevant to the corridor being studied allows for strategic and well-informed decisions to be made. Additionally, the nature of the data allows analysts to obtain planning- and programming-level cost estimates for various scenario options. VDOT’s CSA utilizes a specialized data collection method which includes a harmonized video/GPS logging process, allowing analysts to collect a great deal of roadway information with relatively little effort. The detailed network data that this method provides allows for a thorough assessment of the study corridor and a relatively high level of precision for VDOT to use in their analysis as well as in their countermeasure implementation planning. Louisiana and Mississippi Systemic Safety Project The University of Louisiana at Lafayette National Center for Intermodal Transportation for Economic Competitiveness (NCITEC) conducted a research project that used a systemic safety management approach to identify characteristics common to crashes on rural two-lane roads in Louisiana and Mississippi (Sun and Rahman, 2016). The reason for the project was that Louisiana and Mississippi have had crash and fatality rates among some of the highest in the nation. The goals of this project were to develop a quantitative approach to assessing the potential for a crash separate from a “black spot” analysis; review relevant literature; produce a prioritized map for rural roadways; and propose corresponding inexpensive crash countermeasures. The researchers provided a brief summary of the HSM Part C crash predictive method, the FHWA Systemic Tool, usRAP procedures, and a systemic methodology used by Minnesota. They then conducted an analysis of crash data from 2013-2014 on Louisiana rural two-lane roads, with the assumption that the results would be applicable to Mississippi as well. The crash tree analysis found that the key contributing factors to crashes on rural, two-lane roads were alcohol involvement, surface condition, AADT, and curve radius. 2.2 International Applications of Systemic Safety Analyses Systemic Safety Management Approach in Argentina Zini (2010) presented an approach to performing a systemic safety analysis to identify the best traffic safety treatments for inclusion in a strategic plan for Argentina and other developing countries in the absence of reliable crash data. The methodology started by identifying the primary categories of factors that impact crash likelihood and outcomes, such as factors related to the driver, vehicle, education, infrastructure, and law enforcement. Within each of these factor categories, the specific mechanisms that may have an impact on crash exposure, likelihood, or severity were identified. For example, when considering the category “vehicle,” considerations such as the age of the fleet, standard safety devices on cars sold in the country, and crash- worthiness of vehicles were important factors. Next, factors specific to the country that differ from other countries should be examined. For example, in Argentina and other developing

27 countries, the respect citizens have for the law may not be as great as in developed nations. Another example is that car manufacturers do not include the same safety features in cars in different countries. Certain user demographics and characteristics can be very country-specific as well. After the factors have been fully explored, they must be weighted. Some weightings can be more quantitative in nature, but in many cases they are qualitative. Experts from different disciplines would ideally be involved in the process of weighting factors for their likely impact on fatalities and injuries. This can be visualized in radial charts showing how one factor contributes relative to another. The radial charts can also be used to illustrate goals for the future. This methodology was intended as a complement to an analysis of traffic safety needs based on historical crash data, but it can also be used to guide decisions about traffic investments where comprehensive crash data are not available. Quantifying Road Safety Risk at Locations without Crashes in British Columbia, Canada Several years ago the Insurance Corporation of British Columbia (ICBC) created the Road Improvement Program (RIP) to partner with provincial road authorities to identify locations needing improvement and implement interventions to improve road safety with the goal to reduce the frequency and severity of collisions and resulting auto claims (de Leur and Hill, 2015). Recognizing that the RIP was effective and necessary, the reactive program did not allow for investments at locations needing improvement but did not have a significant history of crashes. Thus, the ICBC developed the Proactive Road Safety Program (PRSP) to identify sites with high-crash potential, but little crash history to supplement the RIP. The PRSP works to allow safety investments from the ICBC to be directed to locations with high-crash potential rather than just high-crash locations. The proactive safety program allows the ICBC to fund projects that cannot be shown to reduce crashes based on crash history. A road safety risk index (RSRI) that incorporates exposure, probability of crash, and consequence of crashes (i.e., severity) was developed to evaluate projects that applied for funding. The three components of index are measured on a relative basis and scored on a scale of 0 to 3 and then averaged. For example, if the expected pedestrian volume at an intersection is about half the pedestrian volume at the most heavily traversed intersection in the network, then the score would be 0.5 x 3 = 1.5 for the exposure metric. The probability component of the RSRI is obtained by referencing guidelines that were developed to focus the analyst on specific features that would likely contribute to the occurrence of a crash. Table 6 illustrates several rural and urban roadway features and guidelines for selecting a probability score. Several factors such as vehicle speed, the potential for speed differential, the interaction between modes, and roadside features can be used to gauge the consequence component of the RSRI.

28 Table 6. Evaluation of Probability Component of Road Safety Risk (de Leur and Hill, 2015) Environment Road Feature Evaluation of Probability Rural Horizontal or vertical curve  Radius or sharpness of curves  Presence of compound curves (S-curves)  Combination of horizontal and vertical curves Highway access location  Access frequency or density  Access alignment or connection  Sight distance from access location Passing  Length of passing zone  Sight distance in passing zone  Opportunity for passing Roadside hazard  Shoulder width and condition  Degree of horizontal and vertical curve  Conflict points from passing or access Road surface and superelevation  Presence of rutting, ponding, cracking, or holes  Inappropriate superelevation Design consistency or expectation  Unexpected feature requiring driver action  Inconsistent road design features Urban Intersection configuration  Oblique alignment of intersection  Level of channelization Traffic control  Inappropriate or degree of traffic control  Visibility of traffic control devices Roadway access  Access frequency or density  Access alignment or connection  Sight distance from access location Cross-sectional elements  Narrow lane widths  Facilities for alternate modes (e.g., sidewalks)  Proximity of roadside hazards (e.g., poles) Road friction or maneuverability  Features creating road friction (e.g., parking)  Ability to maneuver with ease (change lanes) Illumination and road markings  Low or inappropriate level of illumination  Consistent and clear signs and markings Road surface or drainage  Presence of rutting, ponding, cracking, and holes  Opportunity for poor drainage Pedestrian and cyclist facilities  Proximity to adjacent traffic  Conspicuity and visibility to drivers Where numeric scores are not possible, subjective scores can be assigned by someone with expertise in the area. The final RSRI—a summation of the exposure, probability, and consequences indices—is a measure of project need. The RSRI is then converted to a proactive funding index (PFI) as follows: 𝑃𝐹𝐼 𝑅𝑆𝑅𝐼 1 𝐶𝑀𝐹0.5 where: CMF = crash modification factor (i.e., estimate of the safety effectiveness of a countermeasure) The PFI is a score between zero and one and is used to determine the level of ICBC funding for a proactive road safety project on the basis of an exponential function as illustrated in Figure 2. The maximum function level is capped at $25,000. Finally, de Leur and Hill note that most project applications that do not qualify for traditional funding based on crash reduction are low- cost in nature or are related to pedestrian or bike facilities, since those crashes are relatively rare.

29 Figure 2. Function to Determine Level of Funding for Proactive Projects (de Leur and Hill, 2015) Systemic Safety Analysis for Small and Medium-Size Communities in Quebec, Canada Rondier et al. (2014) recognized that on municipal roads, the costs associated with conducting an Empirical Bayes (EB) study to identify “crash prone locations” (i.e., “hot-spots” or “black- spots”) may be too high for local agencies due to data requirements and that the low number of fatal and serious-injury crashes in these areas may limit the ability of EB methods to locate “crash prone locations”. Rondier et al. also recognized the importance of involving stakeholders from different disciplines in the safety planning process; and assessed that while their judgment of safety issues and needs may be biased and based on heuristics, partial information, or incorrect perceptions, the input of these stakeholders may still be valuable because of their in- depth knowledge of the roadway network, historical issues, and ability to identify concerns that the data does not highlight. Starting from this premise, Rondier et al. set out to compare the results of crash prone locations identified through the EB approach and by the stakeholders. The first step in the study involved developing a spatial database for road safety diagnosis. Table 7 presents the data types and sources used for analysis. The second step involved developing SPFs for segments and intersections. The SPFs did not use volume data because volume data were not available for many of the roads. Instead, Rondier et al. used information about the population density, surrounding land use/development, and other geometric information (e.g., number of intersection legs, pedestrian crossings, right-turn lanes at intersections; paved (or unpaved) and number of lanes for segments) as predictors. The SPFs were tested against SPFs that used volume data at a number of locations and were found to provide similar results. Crash prone locations were identified on the system using these SPFs and crash history.

30 Table 7. Spatial Database for Road Safety Diagnosis (Rondier et al., 2014) Data Source Most Used Attributes Hierarchical road network Quebec Ministry of Transportation  Intersections: Municipal or Municipal/MTQ responsibility  Road segments: number of lanes, paved or unpaved, functional classification of the road (local, arterial, collector, highways) Crash data (2007-2011) Quebec Ministry of Transportation  Crash characteristics (number of victims, injury severity, time, factors/causes, impact type)  Vehicle or road user characteristics (type of vehicle, including pedestrians and bicyclists)  Individual driver or pedestrian characteristics (age, gender, postal code of the home location) Traffic flow Field data collection  Traffic flow of intersections and road segments for a limited sample of road categorized according to the functional classification Road environment Field data collection  Intersections: type of intersection (T, X), type of signal control, presence of signs and markings  Road segments: exclusive right-turn lane, permitted parking on the street, presence of driveways to residential and commercial areas, presence of markings Safety Issues Stakeholder interviews and workshops  Thematic issues  Problematic segments and intersections drawn by the stakeholders on maps (categorized according to the thematic issues) Land use, census data and other traffic network Arthabaska Regional County Municipality and Statistics Canada  Land use (urban, industrial, agricultural, forest use)  Urban perimeters of each of the villages  Population density  Presence of bicycle facilities, snowmobile and quad trails The research team then held a workshop with local road safety stakeholders to identify safety issues and crash prone locations. Rondier et al. concluded that the crash prone locations and issues identified through the EB process matched with what the stakeholders identified fairly well, but the stakeholders tended to overemphasize crash types and locations with higher severities but lower frequencies. Rondier et al. found the involvement of local stakeholders to be beneficial, especially in situations where sufficient data are not available to perform an EB analysis or network screening. Comparison of the road safety stakeholder’s subjective perspective to the more objective identification of crash prone locations through an EB method shed light on the usefulness of combining qualitative and quantitative data in the identification of systemic issues and possible crash prone locations in a local and predominantly rural area. In addition, the knowledge of the stakeholders provided insight on the most important road safety issues, while the quantitative analyses tended to both confirm and nuance the crash prone locations to be further investigated.