Consideration of Roadside Features in the Highway Safety Manual (2022)

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.

2 BACKGROUND The concept of providing a forgiving roadside first emerged in the 1960s and started to become regularly incorporated in highway designs in the 1970s resulting in the 1974 AASHTO document Highway Design and Operational Practices Related to Highway Safety and the 1977 AAHSTO Guide for Selecting, Locating and Designing Traffic Barriers (i.e., the “Barrier Guide”). [AASHTO67, AASHTO74 and AASHTO77] The 1974 “Yellow Book” first formalized the forgiving roadside approach to roadside design. In the forgiving roadside approach, designers are given four priorities for designing the roadside: 1. Eliminate the hazard, 2. Relocate the hazard outside the clear zone, 3. Use breakaway devices to reduce the severity of striking the hazard, or 4. Shield the hazard. The 1977 Barrier Guide provided much more detailed information about how to accomplish the four objectives in the Yellow Book. The 1977 Barrier Guide was updated using a great deal of roadside safety research that was performed in the 70’s and 80’s resulting in the first version of the AASHTO Roadside Design Guide (RDG) in 1989.[AASHTO89] New research and crash testing results have resulted in four major updates of the RDG. The forgiving roadside design philosophy has guided roadside designers for more than 40 years and is the underpinning of each edition of the RDG through the present edition published in 2011. [AASHTO11] The RDG is maintained by the AASHTO subcommittee on Design, Technical Committee on Roadside Safety (TCRS). The RDG is intended to aid highway agencies in developing roadside safety standards and policies. It presents material and recommendations for the roadside. The roadside is defined as the area beyond the traveled way (driving lanes), therefore applications for the side of the road as well as the median are presented. The focus of the guide is on minimizing crash severity if a vehicle leaves the traveled way. The RDG does not address geometric decisions (i.e., horizontal and vertical alignments and cross section) which clearly affect the probability of vehicles leaving the traveled way but focus instead on mitigating the severity of crashes once vehicles leave the roadway. Like the AASHTO A Policy on Geometric Design of Highway and Streets (i.e., the “Green Book”), the RDG generally takes a warranting approach to roadside safety. For example, on a 60 mi/hr roadway with an average daily traffic (ADT) over 6000 vehicles/day and a foreslope of 1V:5H, the clear zone (i.e., the area to be kept free of fixed hazards) should be between 36 and 44 ft. This approach gives the designer a design value to work with but does not indicate how much safer 44 ft might be in comparison to 36 and if the additional cost is worthwhile. This leaves the designer in much the same situation as the Green Book user who must prepare a design without really knowing if the design objective has been met. While the “warrant” method is primarily used in the RDG, an alternative cost-benefit analysis technique to evaluate and quantify roadside design decisions has always been contained in Chapter 2. This cost-benefit approach to roadside design is computationally intensive, therefore, the RDG has included a probabilistic software tool for performing these calculations since its first edition in 1989, documented in Appendix A. The first software tool was the ROADSIDE program which was replaced by the Roadside Safety Analysis Program (RSAP). The third version of RSAP (i.e., RSAPv3) was developed under NCHRP 22-27. The software

3 embodiment of the encroachment probability model and cost-benefit technique is further discussed below. There are both similarities and differences between the approaches used by Green Book for geometric design and the RDG for roadside design. The Green Book groups roads into design categories by function, where the RDG provides guidance based on speed and traffic volume. The RDG has the stated purpose of providing guidance for the design of the roadside which will reduce crash severity. The Green Book does not provide explicit guidance on improving safety or reducing crashes, but it does provide general concepts to provide safe designs. Neither document provides guidance on how to reduce the number of vehicles which encroach on the roadside through improved geometric design. A better understanding of the safety implications of geometric design decisions was needed. The first edition of the Highway Safety Manual (HSM) compiled decades of highway safety research and presented readers with the ability to better quantify the safety of geometric design improvements. The HSM and RSAPv3 provide methods for assessing the performance of highway and roadside designs with respect to highway and roadside safety. ROADSIDE SAFETY ANALYSIS PROGRAM Since its first edition in 1989, the RDG has included a probabilistic computational tool for performing benefit-cost analyses of roadside designs. Currently, the RSAP is the tool of choice. The NCHRP22-27 protect team was tasked with creating the third version of RSAP (i.e., RSAPv3) for inclusion in the Roadside Design Guide.[Ray12] RSAP was initially a computer program for performing cost-benefit analyses of roadside designs. A key step in performing such analyses is to estimate the frequency and severity of roadside crashes. RSAPv3 now functions as a database for recent roadside safety data collection efforts in addition to modeling the expected ROR crash frequency and severity for any roadside design, and conducting a cost/benefit and risk analysis of up to five alternative roadside designs. RSAPv3 first determines the encroachment frequency using the annual average daily traffic (AADT) and highway type. The encroachment frequency is adjusted for horizontal curvature, vertical grade, number of lanes, lane width, and the presence of rumble strips. Using a database of reconstructed vehicle trajectory data, the probable vehicle trajectories are queried to determine the frequency and severity of crashes for each user modeled roadside feature (i.e., roadside slope, point hazards, various types and test levels of barriers, water, etc.). The crash cost of each feature can be found by mapping the frequency and severity into units of dollars given the average societal cost of each expected crash. A roadside design that results in a smaller societal cost is, therefore, more cost effective. If the reduction in crash costs over the design life of the improvement are greater than the construction and maintenance costs of the improvement the design is cost-beneficial and should be considered for funding. On the other hand, if the reductions in crash costs are less than the construction/maintenance cost of the improvement the project probably is not worth pursuing. HIGHWAY SAFETY MANUAL The HSM provides science-based tools and techniques for estimating the number and, in some cases, severity of crashes based on specific highway characteristics like lane width, horizontal curvature, grade and many others. [AASHTO10] The first edition of the HSM is organized into four parts: (A) Introduction, Human Factors and Fundamentals; (B) Roadway Safety Management Process; (C) Predictive Method; and (D) Crash Modification Factors.

4 Crashes are predicted for different facility types, including road segments and intersections; special facilities and geometric situations; and road networks. Road Segments and Intersections Crash types common to segments differs dramatically from the crash type common to intersections; therefore, the two basic units of analysis in the HSM are road segments and intersections. Separate predictive models are presented for intersections and road segments for each road type. The HSM defines roadway segments as crashes that occur in the section of road that begins and ends at either the center point of an intersection or where the geometry of the cross section or traffic control features change. Several segments, therefore, may exist between intersections, but a segment cannot span an intersection. Intersection related crashes which occurred on the segment are treated as intersection crash (e.g., angle crashes would be assigned to the intersection). Figure 1 shows a simplified representation of these concepts. Figure 1. Roadways and Segment Definitions. [AASHTO10] Notice that, while segments and intersection are specifically considered because the crash types common to segments differs dramatically from those common to intersections, the first edition of the HSM did not explicitly consider the differences in the crash types which occur on the roadside edges (i.e., run-off-road crashes). Run-off-road crashes where considered alongside the segment crashes in the first edition of the HSM. Special Facilities and Geometric Situations Chapter 16 of the HSM addresses countermeasures associated with the improvement to special facilities and geometric situations including highway-rail grade crossings; traffic control and operational elements; work zone design elements; two-way left-turn lane elements; and passing and climbing lanes. Many of the highway-rail grade crossing treatments do not have

5 quantified countermeasures, but are represented by suggestive trends. Work zone countermeasures have been quantified for freeways only. Road Networks The HSM includes information in Chapter 17 regarding the effects of possible treatments for network-wide crashes. These treatments apply to planning, design, operations, education, or enforcement-related decisions at the road network level. The possible treatments are generally represented by trends that suggest an effect on crash frequency or user behavior. Categories in this chapter include network planning and design approaches/elements (e.g., travel distance and frequency for the users of a network, facility type selection by network use, distance between access points, need for children to cross roads on their way to school, etc.); network traffic control and operational elements (e.g., area-wide traffic calming, conversion of two-way into one-way streets, modification of the level of access control on transportation network); and road- use culture network considerations (e.g., deployment of mobile patrols, stationary patrols, aerial enforcement, radar and laser speed monitoring, automated speed enforcement, red-light reduction by enforcement, etc.). Predicting Crash Frequency and Modifying Predictions for Roadway Segments The HSM Part C includes three chapters, each addressing crash predictive methods for a different facility type: rural two-lane two-way roads (Chapter 10); rural multilane highways (Chapter 11); and urban and suburban arterials (Chapter 12). Part D of the HSM contains a collection of Crash Modification Factors (CMF). This part is also organized into four chapters each presenting CMFs by facility type: Roadway Segments (Chapter 13); Intersections (Chapter 14); Interchanges (Chapter 15); and Special Facilities and Geometric Situations (Chapter 16). Part C of the HSM introduces predictive models based on the concept of safety performance functions (SPFs). SPFs emerge from regression analysis and are used to model the expected crash frequency for a set of base conditions, at a particular facility type. Although the variables comprising SPFs may vary depending on the facility type, typical elements include AADT and segment length as measures of exposure. Additionally, default distributions for crash severity and collision type are provided. Part D of the HSM provides a collection of CMFs including base conditions and associated statistical properties. A CMF is a multiplier or function representing the expected effect of a countermeasure or risk factor relative to a base condition. A CMF value greater than one indicates an expected increase in the number of crashes associated with the countermeasure, whereas CMFs less than one indicate an expected reduction in the number of crashes. Since CMFs provide an estimate of the safety effect of engineering treatments at sites, the potential applications include: the assessment of infrastructure treatments such as widening of lanes or shoulders, flattening of curves, roadside improvements, rumble strips, signing, pavement markings, signalization, etc. Empirically derived CMFs are linked to standard errors, because of their origin as statistical estimates. As the value of a standard error increases, the reliability of the CMF decreases. The highest quality CMFs included in the HSM are associated with standard errors less than 0.1. CMFs with small standard errors result in small confidence intervals and are considered sufficiently accurate, precise, and stable. Conversely, the less reliable CMFs in the HSM are those with standard errors between 0.2 and 0.3. The HSM excludes CMFs with standard errors greater than 0.3.

6 CMFs developed following the publication of the HSM in 2010 are available in the literature. The main source for locating CMFs is the CMF clearinghouse, an online database maintained by the Federal Highway Administration (see http://www.cmfclearinghouse.org). Although users may access and use these CMFs, they should carefully assess each CMF to verify the base conditions applicability and the potential level of uncertainty associated with the standard error. When multiple treatments are considered, in principle, the combined effects of CMFs may be estimated by multiplying individual CMFs; however, users of CMFs should be careful to assess how combining CMFs may influence the ultimate result since multiplying CMFs assumes some level of independence for each individual CMF. For instance, if two treatments are expected to reduce run-off-road crashes, such as realigning the horizontal curvature and installing rumble strips, the effects are likely interrelated and multiplying the corresponding CMFs may overestimate the benefit. The HSM generally identifies crashes based on their associated functional classification. This method of facility designation is due to the crash reporting framework for roadway designations. The HSM, therefore, captures this format to better align the procedures with the reported crash data. The predictive methods included in Part C address three roadway types as previously indicated. Each predictive method incorporates a SPF and companion CMFs. Predictive Method for Rural Two-Lane, Two-Way Roads The HSM Chapter 10 includes models for rural highway segments with two lanes and two-way traffic operations, which may also include short sections of center left-turn lanes, passing lanes or climbing lanes. The model used to determine the predicted average crash frequency for rural two-lane, two-way roads has the following general form: 𝑁𝑁𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 = 𝑁𝑁𝑠𝑠𝑝𝑝𝑠𝑠 𝑥𝑥 × �𝐶𝐶𝐶𝐶𝐶𝐶1𝑥𝑥 × 𝐶𝐶𝐶𝐶𝐶𝐶2𝑥𝑥 × … × 𝐶𝐶𝐶𝐶𝐶𝐶𝑦𝑦𝑥𝑥� × 𝐶𝐶𝑥𝑥 Where: Npredicted = Predicted average crash frequency for a specific year on site type x; Nspf x = Predicted average crash frequency determined for base conditions of the SPF developed for site type x; CMFyx = Crash modification factors specific to site type x and specific geometric design and traffic control features y; and Cx = Calibration factor to adjust SPF for local conditions for site type x. A review of the rural two-lane road SPF with particular emphasis on the prediction and modification of ROR crash frequency was conducted. “The effect of traffic volumes on crash frequency is incorporated through an SPF, while the effects of geometric design and traffic control features are incorporated through the CMFs.” [AASHTO10] The SPF predicted average crash frequency is determined from the following equation: 𝑁𝑁𝑆𝑆𝑝𝑝𝑠𝑠 𝑝𝑝𝑠𝑠 = 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 × 𝐿𝐿 × 365 × 10−6 × 𝑒𝑒−0.312 Where: Nspf rs = Predicted total crash frequency for roadway segment base conditions, AADT = Annual average daily traffic (vehicles/day) and L = Length of roadway segment (miles).

7 The SPF for rural two-lane roads is applicable across traffic volumes ranging from 0 through 17,800 vpd. The base conditions for this SPF are: • Lane width of 12 feet • Paved, six foot wide shoulders • Roadside Hazard Rating (RHR) of 3 • Five driveways per mile • No horizontal curves and vertical grade, centerline rumble strips, passing lanes, two-way left-turn lanes, lighting, or automated speed enforcement. The default distribution of crash types was derived from the Washington Highway Safety Information System (HSIS) data (2002-2006). Using the HSM default data, ROR crashes are 52.1 percent of all crashes with an additional 2.5 percent of the predicted crashes modeled as overturned vehicles. In other words, more than half (i.e., 54.6 percent) of the crashes which occur on rural two-lane roads are ROR crashes. Furthermore, 58.2 percent (i.e., 54.5 ROR and 3.7 overturned) of fatal and injury crashes are the result of ROR crashes by default. The SPF was used to predict crash frequency for the applicable range of traffic volumes (i.e., 0 through 17,800 vpd) and this distribution of crash types. The results are shown Figure 2 for the base conditions. Figure 2. Default Distribution of Crash Frequency for Rural Two-Lane Roads. ROR crashes are a fixed percentage of the total predicted crashes, therefore, any modification of the total crash frequency will affect the ROR crashes unless the CMF is only applied to specific crash types. For example, the CMF for lane width and shoulder width is applicable to “single vehicle run-off-the road and multiple-vehicle head-on, opposite-direction 0 0.5 1 1.5 2 2.5 3 0 5,000 10,000 15,000 C ra sh es p er m ile p er y ea r AADT (vpd) ROR Other Crashes Overturn

8 sideswipe, and same-direction sideswipe crashes,” however, the horizontal curve and vertical grade CMFs are applicable to the total road segment.[AASHTO10] In other words, the lane width and shoulder width CMFs modify a specific group of crash types, while the horizontal and vertical geometric CMFs modify all crash types equally. A summary of the CMFs used to modify base conditions and the crashes which are modified is provided in Table 1. Table 1. Rural Two-Lane Road CMFs and the Applicable Crash Types. CMFs Applicable Crash Types CMF Derived from Driveway Density Total Crashes [Muskaug85] Centerline Rumble Strips Total Crashes Unknown Passing Lanes Total Crashes [Harwood84, Rinde77, Nettelblad79] Two-Way Left-Turn Lanes Total Crashes [Harwood84] Roadside Hazard Rating Total Crashes [Zegeer81, Harwood00] Lighting Total Crashes [Elvik04] Automated Speed Enforcement Total Crashes Unknown Lane Width single vehicle run-off-the road and multiple-vehicle head-on, opposite-direction sideswipe, and same- direction sideswipe crashes [Zegeer81, Griffin87] Shoulder Width and Type single vehicle run-off-the road and multiple-vehicle head-on, opposite-direction sideswipe, and same- direction sideswipe crashes [Zegeer81, Zegeer88] Horizontal Curves Total Crashes [Zegeer92] Superelevation Total Crashes [Zegeer92, Zegeer 91] Vertical curvature Total Crashes [Miaou98] These CMFs predominantly modify all crash types apart from lane width and shoulder width and type CMFs. Interestingly, the RHR CMF modifies all crash types, not just ROR crashes. Recall that a little more than half of the crashes are ROR crashes. The roadside environment should have little, if any, effect on on-road crash types. The distribution of crash types, however, may change as traffic volume changes. Consider further, the RHR is used to determine a roadside design CMF. “Since this rating is a subjective value and can differ marginally based on the opinion of the assessor, it is reasonable to assume that a ‘homogeneous’ segment can have a roadside hazard rating that varies by as much as two rating levels.” The HSM suggests that an average RHR can be used as long as the values do not differ by more than two. Therefore, more than 50 percent of the total crashes are ROR crashes, yet the modifier for roadside design is a single, subjective value which can be averaged to reduce the number of segments and then is applied to all crash types. The RHR is a qualitative index that is subjectively based on an individual’s visual inspection of the roadside. The visual comparison relies mainly on sideslope and clear zone

9 impressions based on comparison of the study segment to standardized reference photographs. The RHR has a scale of 1 to 7 with 1 representing very good and 7 representing very poor roadside conditions, respectively. The photographic examples (Figure 3) and the following definitions for the RHR scale are available in Appendix 13A of the HSM. Rating = 1 • Wide clear zones greater than or equal to 9 m (30 ft) from the pavement edgeline. • Sideslope flatter than 1:4. • Recoverable. Rating = 2 • Clear zone between 6 and 7.5 m (20 and 25 ft) from pavement edgeline. • Sideslope about 1:4. • Recoverable. Rating = 3 • Clear zone about 3 m (10 ft) from pavement edgeline. • Sideslope about 1:3 or 1:4. • Rough roadside surface. • Marginally recoverable. Rating = 4 • Clear zone between 1.5 and 3 m (5 to 10 ft) from pavement edgeline. • Sideslope about 1:3 or 1:4. • May have guardrail (1.5 to 2 m from pavement edgeline). • May have exposed trees, poles, or other objects (about 3 m or 10 ft from pavement edgeline). • Marginally forgiving, but increased chance of a reportable roadside collision. Rating = 5 • Clear zone between 1.5 and 3 m (5 to 10 ft) from pavement edgeline. • Sideslope about 1:3. • May have guardrail (0 to 1.5 m from pavement edgeline). • May have rigid obstacles or embankment within 2 to 3 m (6.5 to 10 ft) of pavement edgeline. • Virtually non-recoverable. Rating = 6 • Clear zone less than or equal to 1.5 m (5 ft). • Sideslope about 1:2. • No guardrail. • Exposed rigid obstacles within 0 to 2 m (0 to 6.5 ft) of the pavement edgeline. • Non-recoverable. Rating = 7 • Clear zone less than or equal to 1.5 m (5 ft). • Sideslope 1:2 or steeper. • Cliff or vertical rock cut. • No guardrail. • Non-recoverable with high likelihood of severe injuries from roadside collision.

10 Using the text or the photographs, a single RHR is chosen to represent both sides of the road for an entire segment. Variations in the roadside are not captured over the segment using this method. For example, if removal of the utility pole shown in the photo of RHR equal to four in Figure 3 is under consideration, the RHR would likely still equal four, even after the unprotected obstacle at the start of the guardrail has been removed. The RHR photographs also tend to mix alignment and roadside issues. For example, the lower valued RHR photographs include relatively straight, flat alignments with shoulders whereas the higher RHR photographs tend to be on very curved alignments with no shoulders. Horizontal curvature, grade and shoulder width have separate CMFs but the way the photographs are presented, the user may get the impression that the on-road characteristics are part of the RHR as well; even if the user realizes that alignment issues have separate CMF they may unconsciously bias their RHR choice based on the alignment and cross-sectional features. While this can be addressed with training and explanation, the subjectivity of the scale will probably always allow alignment and cross section impressions to filter in to the assessment of RHR. Last, with the exception of the photograph for RHR=4, the photographs do not include roadside safety hardware. If a crashworthy guardrail and terminals were included in the scene depicted in RHR=6, would the RHR still be judged to be six even though the guardrail shields motorists from the trees? A review of Figure 4 and Figure 5 further illustrate the potential issues with the RHR. Figure 4 shows a ledge outcropping on the right side of the road and should be considered to have a RHR of seven due to the presence of ledge. The clear zone, however, is equal to 20 feet so the RHR should be two. There is a significant difference (i.e., more than two) using the text descriptions to determine the RHR for this figure.

11 Figure 3. Photographic Representation of Roadside Hazard Ratings (RHR). [Harwood00]

12 Figure 4. Rural Two-Lane Road Example 1. The clear zone shown on the right side of the road in Figure 5 is much different from the clear zone on the left side of the road. While both measure approximately 10 feet, one side has intermittent utility poles while the other side has a dense tree line. Clearly the probability of hitting a utility pole is much different from hitting the tree line if an errant vehicle leaves the road. Figure 5. Rural Two-Lane Road Example 2.

13 Recall the Rural, Two-lane road baseline model presented above. When the model requires modification because the base conditions of the SPF do not match the field conditions, CMFs are used as shown: Nfs= Nspf rs Cr(CMF1r, CMF2r,…, CMFnr) where: Nfs = Predicted number of total roadway segment crashes per year CMFnr = Crash modification factors of design features, and Cr = Calibration factor for roadway segments developed for a particular geographical area. The Rural, Two-lane road base condition assumes and RHR of 3. The roadside CMF modifies this as follows [AASHTO10]: 𝐶𝐶𝐶𝐶𝐶𝐶10𝑝𝑝 = 𝑒𝑒(−0.6869+0.0668∗𝑅𝑅𝑅𝑅𝑅𝑅) 𝑒𝑒(−0.4865) where: CMF10r = Crash Modification Factor for the effect of roadside design; RHR = Roadside Hazard Rating The computed modification for each RHR are shown in Table 2. Table 2. Roadside Design CMFs by RHR RHR CMF10r 1 0.875 2 0.935 3 1.000 4 1.069 5 1.143 6 1.222 7 1.306 Recall that for the two-lane rural roadway in Figure 4 the RHR value could be either a 7 (based on the rock ledge) or a 2 (based on the clear zone). The resulting CMF would then be equal to 1.306 or 0.935 and this CMF is applied to all crash types. Noting that the base condition is an RHR value of 3, the possible results are shown in Figure 6.

14 Figure 6. Example 1 All Crash Types Modified for Roadside Design Over Possible RHR. This example clearly demonstrates that each individual’s decision to apply an RHR of 2 or 7 has far reaching implications to the prediction of all crash types, not just ROR crashes and implications to the ability to provide reliable results. Predictive Method for Rural Multilane Highways Rural multilane highways are the focus of the HSM Chapter 11. These facilities include rural multilane highways without full access control, and non-freeway facilities with four through lanes (excluding the rural two-lane highways with one passing lane in each direction previously identified). Facilities with six or more lanes are not covered. [AASHTO10] The model used to determine the predicted average crash frequency for rural multilane highways has this general form: 𝑁𝑁𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 𝑥𝑥 = 𝑁𝑁𝑠𝑠𝑝𝑝𝑠𝑠 𝑥𝑥 × �𝐶𝐶𝐶𝐶𝐶𝐶1𝑥𝑥 × 𝐶𝐶𝐶𝐶𝐶𝐶2𝑥𝑥 × … × 𝐶𝐶𝐶𝐶𝐶𝐶𝑦𝑦𝑥𝑥� × 𝐶𝐶𝑥𝑥 Where: Npredicted x = Predicted average crash frequency for a specific year on site type x; Nspf x = Predicted average crash frequency determined for base conditions of the SPF developed for site type x; CMFyx = Crash modification factors specific to site type x and specific geometric design and traffic control features y; and Cx = Calibration factor to adjust SPF for local conditions for site type x. As with the other facility types, Rural Multilane Highways must be broken into homogenous segments and intersections prior to predicting the crash frequency. The rural multilane roads have two SPFs to model segment crashes, one for undivided roadways and one 0 1 2 3 4 5 6 7 0 5,000 10,000 15,000 C ra sh es p er m ile p er y ea r ADT (vpd) RHR 2 RHR 3 RHR 7

15 for divided roadways. A new segment, therefore, begins at the center of an intersection when there is a change in any of the following characteristics: • AADT; • Presence of median; • Median width; • Sideslope (for undivided roadway segments); • Shoulder type; • Shoulder width; • Lane width; • Presence of lighting; or • Presence of automated speed enforcement. The HSM suggests median width ranges which can be rounded to a single number to provide homogenous segments. Suggestions are also made for measured shoulder widths and measured lane widths. The complete recommendations are provided in Table 3, Table 4, and Table 5. Table 3. Recommended Median Width Rounding for Segmenting. [AASHTO10] Measured Median Width Rounded Median Width 1 ft to 14 ft 10 ft 15 ft to 24 ft 20 ft 25 ft to 34 ft 30 ft 35 ft to 44 ft 40 ft 45 ft to 54 ft 50 ft 55 ft to 64 ft 60 ft 65 ft to 74 ft 70 ft 75 ft to 84 ft 80 ft 85 ft to 94 ft 90 ft 95 ft or more 100 ft Table 4. Recommended Shoulder Width Rounding for Segmenting. [AASHTO10] Measured Shoulder Width Rounded Shoulder Width 0.5 ft or less 0 ft 0.6 ft to 1.5 ft 1 ft 1.6 ft to 2.5 ft 2 ft 2.6 ft to 3.5 ft 3 ft 3.6 ft to 4.5 ft 4 ft 4.6 ft to 5.5 ft 5 ft 5.6 ft to 6.5 ft 6 ft 6.6 ft to 7.5 ft 7 ft 7.6 ft or more 8 ft or more

16 Table 5. Recommended Lane Width Rounding for Segmenting. [AASHTO10] Measured Lane Width Rounded Lane Width 9.2 ft or less 9 ft or less 9.3 ft to 9.7 ft 9.5 ft 9.8 ft to 10.2 ft 10 ft 10.3 ft to 10.7 ft 10.5 ft 10.8 ft to 11.2 ft 11 ft 11.3 ft to 11.7 ft 11.5 ft 11.8 ft or more 12 ft or more The SPF for rural multilane highways is applicable on undivided roads with an AADT range of zero to 33,200 vpd and divided roads with and AADT range of zero to 89,300 vpd. The base conditions for the SPF are provided in Table 7. The SPF is as follows: 𝑁𝑁𝑠𝑠𝑝𝑝𝑠𝑠 𝑝𝑝 = 𝑒𝑒(𝑎𝑎+𝑏𝑏×ln(𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴)+ln(𝐿𝐿)) Where: Nspf r = Base total expected average crash frequency for a roadway segment; AADT = Annual average daily traffic (vehicles per day) on roadway segment; L = Length of roadway segment(miles); a,b = Regression coefficients (see Table 6) Table 6. SPF Coefficients for Various Injury Crash Frequencies on Rural Multilane Road Segments. [AASHTO10] Crash Severity a b Undivided Divided Undivided Divided 4-lane total -9.653 -9.025 1.176 1.049 4-lane fatal and injury -9.410 -8.837 1.094 0.958 4-lane fatal and injury (excluding possible injury) -8.577 -8.505 0.938 0.874

17 Table 7. Rural Multilane Road Base Conditions. [AASHTO10] Undivided Divided Lane width (LW) 12 feet 12 feet Right shoulder width 6 feet 8 feet Median width N/A 30 feet Shoulder type Paved N/A Side slopes 1V:7H or flatter N/A Lighting None None Automated speed enforcement None None The model includes these crash types: • Head-on, • Sideswipe, • Rear-end, • Angle, • Single, and • Other. ROR crashes are not explicitly modeled. Single vehicle crashes compose 23.8 percent of total crashes on undivided rural multilane roads and 76.8 percent of total crashes on divided rural multilane roads. Single vehicle crashes are often ROR events, therefore a good deal of these crashes are likely ROR crashes. The increase for divided roadways may represent the single vehicle ROR crashes experienced when a median is installed, however, this is only speculation and it is not explicitly defined in this model. After determining Nspf ru (i.e., undivided crash frequency) or Nspf rd (i.e., divided crash frequency), for the base conditions, CMFs are applied to account for variations in geometric design features. The median width is measured “…between the inside edges of the through travel lanes in the opposing direction of travel; thus, inside shoulder and turning lanes are included in the median width.”[AASHTO10] “The base condition for this CMF is a median width of 30 ft. The CMF applies to total crashes, but represents the effect of median width on reducing cross-median collisions; the CMF assumes that non-intersection collision types other than cross-median collisions are not affected by median width.” The CMF was adapted from research by Harkey et al. [Harkey08] that showed “…cross-median collisions represent 12.2 percent of crashes on multilane divided highways.”[AASHTO10] The median width CMF “…applies only to traversable medians without traffic barriers. When barriers are present, a CMF equal to 1.00 should be used for median width. Essentially, this means that a 30-ft wide median (i.e., the base conditions) is equivalent to a median of any width with a median barrier.

18 Table 8. Rural Multilane Road CMFs and the Applicable Crash Types. CMF Undivided Divided Applicable Crash Types CMF Derived from Lane Width CMFRA CMFRA single vehicle run-off-the road and multiple-vehicle head-on, opposite-direction sideswipe, and same- direction sideswipe crashes [Harkey08] Shoulder Width CMF2ru --- single vehicle run-off-the road and multiple-vehicle head-on, opposite-direction sideswipe, and same- direction sideswipe crashes [Harkey08] Right Shoulder Width --- CMF2rd total crashes [Lord08] Median Width --- CMF3rd total crashes [Harkey08] Sideslopes CMF3ru --- total crashes [Zegeer88, Harkey08] Lighting CMF4ru CMF4rd total crashes [Elvik04] Automated Speed Enforcement CMF5ru CMF5rd fatal and injury crashes unknown ROR crashes are not explicitly identified for this SPF; however, single vehicle crashes compose 23.8 percent of total crashes on undivided rural multilane roads and 76.8 percent of total crashes on divided rural multilane roads. It is not known for certain, however, if ROR crashes are included in the single vehicle crashes and/or the total predicted crash frequency for each segment. Predictive Method for Urban and Suburban Arterials Two- and four-lane undivided urban and suburban facilities are addressed by these models. Additionally, four-lane divided facilities and three- and five-lane facilities with center two-way left-turn lanes are also addressed by these models. Full access control facilities are not covered, however bi-directional facilities where the traffic is separated with raised or depressed medians are included in this definition. Painted medians are covered by this definition, however, these facilities are not considered divided. This definition is limited to arterials with five or fewer lanes. The model used to determine the predicted average crash frequency of urban and suburban arterials takes the same general form as the two previously discussed models for Rural roads; however, this model accounts for vehicle-pedestrian and vehicle-bicycle collisions separate from the roadway segment SPF as shown here:

19 𝑁𝑁𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 = (𝑁𝑁𝑠𝑠𝑝𝑝𝑠𝑠 𝑥𝑥 × �𝐶𝐶𝐶𝐶𝐶𝐶1𝑥𝑥 × 𝐶𝐶𝐶𝐶𝐶𝐶2𝑥𝑥 × … × 𝐶𝐶𝐶𝐶𝐶𝐶𝑦𝑦𝑥𝑥� + 𝑁𝑁𝑝𝑝𝑝𝑝𝑝𝑝𝑥𝑥 + 𝑁𝑁𝑏𝑏𝑝𝑝𝑏𝑏𝑝𝑝𝑥𝑥) × 𝐶𝐶𝑥𝑥 Where: Npredicted = predicted average crash frequency for a specific year on site type x; Nspf x = predicted average crash frequency determined for base conditions of the SPF developed for site type x; Npedx = predicted average number of vehicle-pedestrian collisions per year for site type x; Nbikex = predicted average number of vehicle-bicycle collisions per year for site type x; CMFyx = crash modification factors specific to site type x and specific geometric design and traffic control features y; and Cx = calibration factor to adjust SPF for local conditions for site type x. Nspf x is the determined as follows: 𝑁𝑁𝑠𝑠𝑝𝑝𝑠𝑠 𝑥𝑥 = 𝑁𝑁𝑏𝑏𝑝𝑝𝑏𝑏𝑏𝑏 + 𝑁𝑁𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏 + 𝑁𝑁𝑏𝑏𝑝𝑝𝑝𝑝𝑏𝑏𝑦𝑦 Where: Nbrmv = predicted average crash frequency of multiple-vehicle nondriveway collisions for base conditions; Nbrsv = predicted average crash frequency of single vehicle crashes for base conditions; and Nbrdwy = predicted average crash frequency of multiple-vehicle driveway-related collisions The site must be broken into homogeneous segments for analysis. Recall segments do not cross intersections, but there may be several segments between intersections. The intersection models predict crashes within the geometric limits of the intersection and the above model predicts crash frequency for the segments. A new segment begins at the center of an intersection or when there is a change any of the following characteristics: • AADT; • Number of through lanes; • Median width; • Presence/type of median; • Presence/type of on-street parking; • Roadside fixed object density; • Present of lighting; or • Speed category. The segmenting for changes related to median width are based on the rounding of measured median widths, as discussed in the Predictive Method for Rural Multilane Highways and presented in Table 3 above. The presence/type of median refers to the presence or lack of a raised or depressed median.

20 There are several segment SPFs for this highway classification and these SPFs are valid at various traffic volumes, as described here: • Two-lane undivided arterials (2U): 0 to 32,600 vpd • Three-lane arterials including a center two-way left-turn lane (3T): 0 to 32,900 vpd • Four-lane undivided arterials (4U): 0 to 40,100 vpd • Four-lane divided arterials (4D): 0 to 66,000 vpd • Five-lane arterials including a center two-way left-turn lane (5T): 0 to 53,800 vpd Unlike the other predictive methods, the arterial predictive method addresses single vehicle crash frequency (Nbrsv) separate from other crash types. The predicted crash frequency is as follows: 𝑁𝑁𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏 = exp (𝑎𝑎 + 𝑏𝑏 × ln(𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴) + ln(𝐿𝐿)) Where: Nbrsv = Expected average single vehicle crash frequency for a roadway segment; AADT = Annual average daily traffic (vehicles per day) on roadway segment; L = Length of roadway segment(miles); a,b = Regression coefficients (see Table 9) The following two equations and the regression coefficients shown in Table 9 and Table 10 are then used to predict different severities for single vehicle crashes. 𝑁𝑁𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏(𝑃𝑃𝐴𝐴𝑂𝑂) = 𝑁𝑁𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏(𝑝𝑝𝑡𝑡𝑝𝑝𝑎𝑎𝑡𝑡) − 𝑁𝑁𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏(𝐹𝐹𝐹𝐹) 𝑁𝑁𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏(𝐹𝐹𝐹𝐹) = 𝑁𝑁𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏(𝑝𝑝𝑡𝑡𝑝𝑝𝑎𝑎𝑡𝑡) � 𝑁𝑁′𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏(𝐹𝐹𝐹𝐹) 𝑁𝑁′𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏(𝐹𝐹𝐹𝐹) + 𝑁𝑁′𝑏𝑏𝑝𝑝𝑠𝑠𝑏𝑏(𝑃𝑃𝐴𝐴𝑂𝑂) � Collisions with fixed objects represent 40 to 72 percent of the Fatal and Injury single vehicle crashes depending on which facility is being examined. Collisions with fixed objects also represent 75 to 96 percent of the PDO crashes. Fixed-object crashes occur on the roadside, therefore, the crashes modeled by this SPF must be ROR crashes; however, the roadside geometry is not considered in the analysis. The analysis considered only the traffic volume and the length of the segment. After determining the total roadway segment crashes (Nspf x ) for the segment, CMFs are applied to account for variations for geometric design features and roadside design.

21 Table 9. Coefficients for Urban and Suburban Single Vehicle SPF.[AASHTO10] Road Type a b Overdispersion Parameter (k) Total Crashes 2U -5.47 0.56 0.81 3T -5.74 0.54 1.37 4U -7.99 0.81 0.91 4D -5.05 0.47 0.86 5T -4.82 0.54 0.52 Fatal and Injury (FI) 2U -3.96 0.23 0.50 3T -6.37 0.47 1.06 4U -7.37 0.61 0.54 4D -8.71 0.66 0.28 5T -4.43 0.35 0.36 PDO 2U -6.51 0.64 0.87 3T -6.29 0.56 1.93 4U -8.50 0.84 0.97 4D -5.04 0.45 1.06 5T -5.83 0.61 0.55 Table 10. Proportion of Crash Severities for Road Types within the Urban and Suburban Models. [AASHTO10] 2U 3T 4U 4D 5T Collision Type FI PDO FI PDO FI PDO FI PDO FI PDO Collision with animal 0.026 0.066 0.001 0.001 0.001 0.001 0.001 0.063 0.016 0.049 Collision with fixed object 0.723 0.759 0.688 0.963 0.612 0.809 0.500 0.813 0.398 0.768 Collision with other object 0.010 0.013 0.001 0.001 0.020 0.029 0.028 0.016 0.005 0.061 Other single vehicle collision 0.241 0.162 0.310 0.035 0.367 0.161 0.471 0.108 0.581 0.122

22 Table 11. Urban and Suburban Arterial CMFs and the Applicable Crash Types. CMF Applicable Crash Types CMF Derived from On-street parking total crashes [Bonneson05] Roadside Fixed Objects total crashes [Zegeer84] Median Width total crashes [Harkey08] Lighting total crashes [Elvik04] Automated Speed Enforcement fatal and injury crashes unknown The median width CMF is similar to the CMF discussed above; however, it has been modified to have a base condition of 15 feet and adjusted for the default distribution of various types of other crashes included in the “total crashes” description. The CMF represents the effect of median width on reducing cross-median collisions, as discussed above; however, this CMF modifies total crash frequency. The CMF was adapted from research by Harkey which found that cross-median collisions represent 12 percent of crashes on suburban and urban arterials. [AASHTO10, Harkey08] The median width CMF is only applicable to traversable medians without traffic barriers. When barriers are present, a CMF equal to 1.00 should be used for median width. Using a CMF of 1.00 when barriers are presents fails to recognize the possibility that median barriers occasionally fail and have the potential to allow cross-median crashes to occur. The roadside fixed-object density CMF has been adapted from Zegeer and Cynecki’s work on utility pole collisions [Zegeer84] to determine the effect of roadside fixed objects on total crashes. It has a base condition of no roadside fixed objects and is determined using the following equation: 𝐶𝐶𝐶𝐶𝐶𝐶2𝑝𝑝 = 𝑓𝑓𝑡𝑡𝑠𝑠𝑠𝑠𝑠𝑠𝑝𝑝𝑝𝑝 × 𝐴𝐴𝑠𝑠𝑡𝑡 × 𝑝𝑝𝑠𝑠𝑡𝑡 + (1.0− 𝑝𝑝𝑠𝑠𝑡𝑡) Where: CMF2r = crash modification factor for the effect of roadside fixed objects on total crashes; foffset = fixed-object offset factors from Table 12 Dfo = fixed-object density (fixed objects/mi) from both sides of the road combined; and pfo = fixed-object collisions as a proportion of total crashes from Table 13. This CMF is applied as follows: • Point objects should be considered if they are at least four inches in diameter and do not have breakaway design. “Points that are within 70 feet of one another longitudinally along the road are counted as a single object.” [AASHTO10]. • Continuous objects which are not obstructed by the point objects are counted as one point for every 70 feet of length. • Fixed objects in the medians of divided arterials are not considered.

23 For example, if there is 140 feet of guardrail along a roadside located in front of two utility poles located 70 feet apart, the guardrail would be counted as two point objects within the segment and the utility poles would not be considered. On the other hand, if there is 140 feet of guardrail located behind two utility poles, the guardrail would not be considered, only the poles would be considered. If there is 140 feet of guardrail on one side of the road and two utility poles located 70 feet apart on the other side of the road within the same segment, this would be considered 4 objects. Continuing on with this same example, if there was 140 feet of guardrail and two utility poles in the median, neither of these items would be considered in the analysis. Table 12 provides the fixed-object offset factors for use in the CMF for object density. Any objects located more than 30 feet from the edge of the travel lane should “…use the value for 30 feet.”[AASHTO10] Table 12. Fixed-Object Offset Factor.[AASHTO10] Offset to Fixed Objects (feet) (Ofo) Fixed- Object Offset Factor (foffset) 2 0.232 5 0.133 10 0.087 15 0.068 20 0.057 25 0.049 30 0.044 Recall that the fixed-object CMF is applied to total crashes; therefore the CMF is modified to adjust for the proportion of fixed-object crashes by roadway type, using the proportions shown in Table 13. Table 13. Proportion of Fixed-Object Collisions.[AASHTO10] Road Type Proportion of Fixed- Object Collisions (pfo) 2U 0.059 3T 0.034 4U 0.037 4D 0.036 5T 0.016 The combined effect of the offset to fixed objects (Ofo) and the density of fixed objects (Dfo) are shown in Figure 7 for the two-lane undivided urban and suburban arterials (2U). The

24 CMF is greater for tighter-spaced objects that are closer to the road. For example, when the objects are spaced 70 feet apart and offset two feet from the travel way, the CMF is 2.00. The influence of this adjustment is shown in Figure 8. Figure 8 shows a two foot offset for each road type with various spacing of the fixed objects. The proportion of fixed-object crashes for 3T, 4U, and 4D are all similar (see Table 13), hence the overlapping of results. Recall this CMF does not model median fixed-object crashes; therefore the crashes with fixed objects within the medians of four-lane divided (4D) urban and suburban arterials are not included in this proportion. Figure 7. Fixed-Object CMF for Various Offsets and Spacing Along Two-Lane Urban and Suburban Arterials. Summary of Predictive Methods and Roadside Design in the HSM The HSM uses safety performance functions (SPFs) for the different functional classifications of roads. These SPFs are based on a set of base conditions which can be adjusted using CMFs. The HSM uses different approaches for representing roadside characteristics for rural two-lane roads, rural multilane highways as well as the urban and suburban arterials. For rural two-lane roads, a subjective CMF which represents the “entire roadside design” (i.e., the RHR) is used. Based on the summaries of qualitative influences, unknown crash effects, and limited application of CMFs, there appears to be considerable opportunity to improve the assessment of the safety effects of roadside conditions included in the HSM. 0.000 0.500 1.000 1.500 2.000 2.500 70 120 170 220 C M F Fixed-object Spacing (feet) CMF (2U-offset 2') CMF (2U-offset 10') CMF (2U-offset 20') CMF (2U-offset 30')

25 Figure 8. Fixed-Object CMF for Two Foot Offset for Various Urban and Suburban Arterial Road Types. SUMMARY The HSM is a carefully developed statistical representation of highway crashes that is based on observed crash data. The HSM methods enable the estimate of average crashes based on safety performance functions which represent average conditions of each road type. Site- specific calibration is based on observed or historic crashes at a site. The CMFs used in the HSM, though limited for roadside applications, then permit users to adjust the expected number of crashes for various countermeasures. Currently, the HSM statistical models predict crashes based on what has been observed and do not explicitly address crash causation. RSAPv3 aids designers in detailed roadside design analysis, however, a tool such as the HSM is needed for scoping studies to estimate the safety benefits of corridor-level decisions early in the design process. These early decisions have the potential to reduce ROR frequency and severity. The current HSM methods for predicting and modifying ROR crash frequency do not support these preliminary decision efforts. 0.000 0.500 1.000 1.500 2.000 2.500 70 120 170 220 C M F Fixed-object Spacing (feet) CMF (2U-offset 2') CMF (3T) CMF (4U) CMF (4D) CMF (5T)