Safety Impacts of Intersection Sight Distance (2018)

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.

NCHRP 17-59 7 CHAPTER 2: RESEARCH APPROACH/ANALYTICAL METHODOLOGY OVERVIEW Chapter 2 presents the review of existing intersection sight distance (ISD) practices and relationships. This consists of a literature review, review of documented practices, and review of uncontrolled intersection ISD methodologies. The literature review focuses on prior research identifying factors related to ISD and associated safety performance. The purpose of the review of the documented practices was to understand how state and local agencies define visual obstructions and collect ISD measurements in the field. The purpose of the review of uncontrolled intersection ISD methodologies was to further identify factors related to ISD at low volume intersections for their inclusion in this research. The remained of this chapter is divided into two sections. The first section presents the field data collection approach and associated data summary. The second section presents a brief overview of the analytical method used to evaluate the safety impacts of available ISD from the collected data. REVIEW OF EXISTING ISD PRACTICES AND RELATIONSHIPS Introduction The project team reviewed existing practices and published literature to understand ISD collection practices and established relationships from other studies. A considerable understanding of the current and historical collection practices was necessary to develop the study design. This review aided the project team in determining the necessary information, and how such information should be collected in the field. The review also provided direction toward analytical methods; specifically, the variables that should be explored and potential relationships between variables. Finally, current state and local agency practices for collection of ISD were reviewed, which provided important practical considerations during development of guidelines. Literature Review A thorough but targeted literature review was conducted to obtain prior research related to factors that influence ISD and associated safety performance, specifically intersection type and geometry, roadway classification, traffic volumes, speed, roadway geometry, and medians. This review found that studies relating the effects of ISD to safety are limited in quantity and quality. For many subjects, either no research has been conducted or the findings vary drastically from study to study and are sometimes even contradictory. This is particularly true for studies that attempted to relate sight distance to crashes. The most relevant studies were reviewed and provided insight into the critical relationships that were explored as a part of this research study.

NCHRP 17-59 8 Traffic Volume As with most safety performance research, most crash-based evaluations indicate that traffic volume is highly correlated with crash occurrence. Two relevant studies were reviewed that have attempted to relate traffic volumes and intersection sight distance to roadway safety. David and Norman (1975) examined the effect of ISD on the crash occurrence for unsignalized intersections with various ranges of total entering average annual daily traffic (AADT). For each AADT range, it was determined that increasing sight distance on the minor approach was associated with a reduction in crash occurrence. In addition, the relationship between sight distance and the predicted reduction in crashes at intersections with restricted ISD was quantified. The expected crash reduction frequencies related to ISD are shown in Table 1. Table 1. Expected Crash Reduction Frequencies per Intersection per Year (David and Norman 1975). Stockton et al. (Mounce 1981; Stockton et al. 1981) sought to assess the effects of major roadway volume and minor roadway sight distance on driver compliance with stop signs at low volume intersections. Although over 30 years old, the study provides a rigorous look at the effects of volume. The experiment was designed as a 6×2×2×2 factorial that had the dependent variables of compliance rate and crash rate measured across the independent variables of six groups of major road volumes, two types of ISD (i.e., restricted or unrestricted), two types of locations (i.e., rural or urban), and two types of intersection geometry (i.e., four-leg intersection or three-leg intersection). Two-way average daily traffic volumes ranged between 0 and 6,000. To provide an ordinal variable, the range was broken down into 1,000 ADT segments. A minimum of 10 intersections were selected within each of the major road volume groups, which were balanced between the rural and urban traffic conditions. A mean annual crash rate was calculated from the three-year crash history (1976-1978) for property-damage, injury, and fatal crashes. Available ISD was compared with design ISD from AASHTO (AASHO 1965; AASHTO 1973) to determine whether it was restricted or unrestricted. When selecting the sample of intersections, major approaches were restricted to two-lane or four-lane undivided roadways, with geographical and climatic conditions controlled as closely as possible. Analysis of variance Average Annual Daily Traffic Entering Intersection (1000s) Increased Sight Distance (ft) 20-49 50-99 >100 < 5 0.18 0.20 0.30 5 – 10 1.00 1.30 1.40 10 – 15 0.87 2.26 3.46 > 15 5.25 7.41 11.26

NCHRP 17-59 9 (ANOVA) was used for the statistical analysis. Based on a result of a sample of 2,830 observations at 66 stop-controlled intersections, the authors concluded that ISD significantly affected the total violation rate. Total violations were higher at low volume intersections, at which the sight distance was unrestricted (p<0.05). In addition, there was a significant interaction effect between ISD and volume for total violation rate (P<0.05). For both restricted and unrestricted conditions, the total violation rate decreases as major road volume increases, and displayed the most significant trend between ADTs of 2,000 and 5,000. However, no significant relationship was established by the ANOVA model between total annual crash rate and other variables including ISD because the multiple coefficient of determination was unacceptably low (0.037). Furthermore, the result was limited because of the subjective classification of sight restriction in the study. Bias may have occurred because the intersection was rated by ISD as a whole, rather than by individual sight triangle quadrants. Seasonal Effects Souleyrette et al. (2006) investigated the difference in the safety performance of stop-controlled intersections and uncontrolled intersection on ultralow-volume rural roads. A total of 6,846 unpaved rural ultralow-volume intersections (identified as intersections with fewer than 150 daily entering vehicles) were randomly selected across the State of Iowa, of which, 56 percent were uncontrolled. For the selected intersections, the crash history was reviewed for a 10-year period, concentrating on select crash types for multivehicle crashes within 150 feet of the intersection. The crash data review revealed that for both stop-controlled and uncontrolled intersections, obscured vision was not noted as a major contributory factor. Conversely, for uncontrolled intersections, an analysis of crashes did demonstrate that, during crop season (June through October), crash frequency was approximately 12 percent higher than during the rest of the year, likely due to the ISD limitations associated with growth of crops and other vegetation. Area Type Although no studies were found that attempted to specifically explore the difference in effects between rural and urban areas and intersection sight distance, area type appeared as a prominent variable in several studies. Rural Areas: Hanna et al. (1976) studied the effects of ISD on the safety performance of 232 intersections in rural areas of Virginia. The intersections were stratified by those with poor sight distance on one or more approaches and those with adequate sight distance on all approaches. They discovered that intersections with poor sight distance had higher-than-normal crash rates. The intersections with poor sight distance had an observed crash rate of 1.33 crashes per million entering vehicles, while the average crash rate of the 232 intersections was 1.13 crashes per million entering vehicles. In addition, it was discovered that the large increase in angle collisions (30 percent) at the intersections with restricted sight distance largely contributed to the higher crash rate.

NCHRP 17-59 10 Conflicting conclusions regarding the safety effectiveness of adequate ISD in rural areas were evident in some research. Charlton (2003) suggested that unrestricted visibility of approaching traffic at a rural unsignalized intersection in New Zealand encouraged drivers to anticipate traffic gaps well in advance of the intersection, and thus drive aggressively through the intersection. An analysis of five years of crash reports revealed that 23 out of the 24 crashes involved local drivers in daytime conditions attempting crossing or turning maneuvers at the intersection. In an attempt to address this issue, a visual restriction treatment, consisting of a 1.2 meter screen, was erected along the minor approach to the intersection beginning 125 meters prior to the intersection and ending 25 meters prior to the intersection. To assess the effectiveness, an evaluation was undertaken during daytime conditions using three hours of data collected before and after installation of the screen. The primary surrogate measure was the approach speed, which was measured 25 meters prior to the intersection on the minor road. For this measure, the authors considered that the treatment would be judged effective if the 80th percentile approach speed was reduced by 10 percent. The post-treatment speed data were collected 3 weeks after the installation and 141 vehicles were involved (161 vehicles in the before period). The data indicated a 23.4-percent reduction in the 80th percentile speed (from 47 km/h to 38 km/h), 23.7- percent reduction in mean speed (from 37.98 km/h to 29.22 km/h), and elimination of all approach speeds over 57 km/h. To evaluate the long-term efficacy of the treatment as well as increase the reliability of the result, the authors conducted a follow-on test on speed data. The speed data collection was conducted 2 weeks after installation, 21 weeks after installation, and 37 weeks after installation. The result showed an identical 30 percent drop in 85th percentile speeds from before period to 37 weeks after the treatment and a progressive reduction of approach speeds in the 40-50 km/h range. The second measure of effectiveness was the driver’s traffic detection rates, which were defined as the percentage of drivers correctly reporting the presence and location of a target vehicle (or other vehicle), with a 10-percent increase in detection rate considered effective. During the data collection, a target vehicle was parked on the roadside of the major road 150 meters to the north of the intersection for the first 1.5 hours and then switched to the south for the next 1.5 hours. The drivers crossing the intersection from the minor road were stopped and surveyed at a site that was inconspicuous to the approaching traffic. The data, which was collected 3 weeks after the installation, indicated an increase from 16.8 percent to 31.9 percent, easily meeting the criteria of a 10-percent increase. In addition, no crashes had been reported 37 weeks after the treatment, leading the authors to conclude that driver behavior had been positively affected. Urban Areas: Poch and Mannering (1996) investigated 63 intersections in Bellevue, Washington, to assess interactions between geometric and traffic-related elements and crash frequencies in urban areas. Prediction models were developed using negative binomial distributions for each approach. In doing this, four types of crash variables were estimated: total crash frequency, rear-end crash frequency, angle crash frequency, and approach turn crash frequency. The 63 intersections used in the study were selected from a list of intersections that were targeted for operational improvement (e.g. phasing change, control change, channelization,

NCHRP 17-59 11 etc.) between 1988 and 1992. Each intersection was divided into separate approaches, and crash data were taken at each approach in one-year intervals between 1987 and 1993. As a result, a total of 1,385 observations provided by the 63 intersections were studied. For each observation, a total of 64 possible geometric variables were collected (e.g. control type, sight restriction, speed limit, etc.). With regard to ISD, sight distance restriction (1 if restricted, 0 otherwise) was used in the model. In this case, the intersections were considered restricted if available sight distance was provided compared to the corresponding AASHTO values. When estimating the coefficient for the model of each crash type, restricted ISD on stop-controlled approaches was highly associated with high angle crash frequency. Retting et al. (2003) sought to develop a better understanding of the crash pattern for two-way stop-controlled intersection in urban areas. To reflect a wide geographic range, four cities were included in the study: Germantown, TN, Oxnard, CA, Springfield, MO, and Westfield, NJ. A total of 1,788 crash reports were reviewed and categorized by crash type based on information in the narrative descriptions and diagrams. To avoid giving greater influence to cities contributing more police reports, all statistics were calculated separately for each city, and then averaged. Seven distinct crash types (i.e., stop sign violation, rear-end, ran off road, backing, wide/narrow turn lane change, left turn oncoming), plus an “other” type for unusual, unclassifiable events, were defined. An initial comparison showed that stop sign violations occurred in 70 percent of the 1,788 crashes. To better understand the potential pattern, stop sign violation crashes involving drivers who stopped before entering the intersection were further classified into six subsets: did not see other vehicle, obstructed vision, misperceived intent of other vehicle, saw other vehicle, other, and unknown. It was discovered that the most common reasons cited by drivers were that they did not see the other vehicle (44 percent) or their view of cross traffic was obstructed (16 percent). Intersection Configuration Pickering et al. (1986) analyzed a multitude of crash categories based on a sample of 300 rural three-leg intersections in the United Kingdom. It was concluded that visibility from the minor road was a significant predictor for one class of crashes, and the analysis indicated that increased visibility was correlated with higher crash frequency, which contradicts the findings from other ISD-related safety research. Similarly, Vogt (1999) estimated the safety impact of intersection configuration combined with traffic volume. Three types of rural intersections were discussed: three-leg stop-controlled intersection on four-lane highway, four-leg stop-controlled intersection on four-lane highway, and signalized intersection. For each type of intersection, a separate negative binomial crash model was developed in terms of intersection geometry variables, such as median width, intersection angle, and sight distance.

NCHRP 17-59 12 Over 200 intersections in both Michigan and California were selected based on the following constraints:  Three-leg rural intersections, major road four-lane, minor leg two-lane stop-controlled, median width less than or equal to 36 feet on major road.  Four-leg rural intersections, major road four-lane, minor legs two-lane stop-controlled, median width less than or equal to 36 feet on major road.  Four-leg rural signalized intersections, major and minor roads two-lane. Crashes that were either coded as intersection-related or occurred within 250 feet of the intersection from 1993 to 1995 were used in the model. ISD data were collected in the field following the guidelines of AASHTO (1994), which assumed that the height of the driver’s eye was 3.5 feet and the height of the oncoming car was 4.25 feet. Instead of following the measurement in AASHTO for decision point, the study author’s used an observer standing at 10 feet from the edge of the major road. When examining the mutual correlations of crashes versus other variables, they found that providing the appropriate ISD generally has an insignificant safety impact, with the correlation coefficient less than 0.3. Unfortunately, because of the negligible effect, ISD was not included in the final model, which resulted in no conclusive relationships established between ISD, crashes, and other geometric variables. Speed As with area type, speed at the intersections was not the specific focus of any of the reviewed studies. However, speed figured prominently in the approach methodology used by Arndt and Troutbeck (2005). Arndt and Troutbeck sought to develop a prediction model to reveal the relationship between crash rate and multiple intersection geometry factors with the sample of 206 unsignalized intersections throughout Queensland, Australia. In order to obtain a relatively even spread of the widest possible range of values, intersections were selected in a two-stage process. First, the authors randomly chose a larger number of intersections as potential candidates for the study. From these candidate intersections, a smaller sample of intersections was selected to provide the greatest range of values of the variables, including number of legs, traffic control, road type, and six other variables. An analysis period of five years was selected for the high-volume intersections, while ten-year period was selected for the low volume intersections because of the relatively low frequency of crashes. The cross product of traffic flows was used to categorize the intersections. An intersection was considered a high-volume intersection if the value of the cross product of the traffic flow per day approaching on the minor leg multiplied by the sum of the traffic volume per day approaching on the major legs exceeded 500,000. To identify the potential factors affecting one particular crash type, the crash data were divided into subsets according to the nature (considering the number of vehicles involved and the original direction of travel of the vehicles) and principal cause of the collisions. Poisson regression models were used for prediction modeling for most cases. Visibility from the minor road was one of 30 variables used in the model. The authors were cautious of the relationship

NCHRP 17-59 13 between intersection sight distance and speed. Therefore, visibility was measured in time instead of distance in the model, which was accomplished by dividing the sight distance by speed on the approach. After the analysis of the prediction model for each type of crash, the authors concluded that approach visibility time was an important predictor of all types of multiple vehicle crashes. Roadway Geometry Harwood et al. (2000) investigated the impact of inadequate ISD on the safety performance of rural two-lane highways. In the study, the authors sought to develop a crash prediction that combined the use of historical accident data, regression analysis, before-and-after studies, and expert judgment to better estimate the safety of rural two-lane roadway. For at-grade intersections, only those crashes that occurred within 250 feet of the intersection were included. To better adjust the base model for at-grade intersections, the accident modification factors (AMFs) were developed by the expert panel to account for the safety effect of skew angle, sight distance, traffic control and so on. Examining the related literature, the expert panel was unable to identify a single evaluation to be the most credible. Thus, the crash impact was estimated based on the results of multiple studies involving improvements to ISD. The expert panel concluded that, for each quadrant of the intersection that had limited ISD, the total crash frequency for an intersection with two-way stop control or yield control increased by slightly less than 5 percent. ISD was deemed limited when the sight distance available was less than the minimum specified by AASHTO (1994) for a design speed 20 km/h lower than the design speed of the major road. Therefore, in addressing the limited sight distance issue for one, two, three, or four quadrants of an intersection, AMFs were estimated at 1.05, 1.10, 1.15, and 1.20, respectively. Conclusion The relevant ISD-related safety studies performed to date are inconsistent in both the direction and magnitude of the major factors related to crashes and ISD. To date, there has yet to be a statistically sound and rigorous crash-based evaluation of stop-controlled intersections that conclusively quantifies the relationship between ISD, relative factors, and safety performance. However, the review does provide direction toward the data collection plan and study design and confirms that a wide range of factors should be collected in the field and thoughtfully considered in the evaluation. Review of Documented Practices The purpose of the review of the documented practices was to understand how state and local agencies define visual obstructions and collect ISD measurements in the field. This information helped the project team determine what methods and parameters should be followed in the field.

NCHRP 17-59 14 Collection of State and Local Policies The project team collected policies that were readily available, either in the literature or on state and local transportation websites. The original intention of the research was to conduct a survey of agencies. However, in March 2012, before this project commenced, a survey was conducted by an Institute of Transportation Engineers (ITE) Traffic Engineering Council on ISD, which collected nearly 200 ISD practice/policy documents from state and local agencies in the U.S. and internationally. After discussing this with the panel at the kickoff meeting, the project team decided to build on ITE’s effort instead of conducting a separate survey, and a sample of the collected practice/policy documents was requested from ITE. The ITE project lead provided a sample of 30 policies, with 20 municipal codes in Alabama and 10 municipal codes in Tennessee. Upon review, it became apparent that these policies did not specify methods for data collection. Instead, they address requirements associated with corner visibility in residence and business districts, by (1) setting the intersection sight distance and (2) regulating the height of fences, walls, and other obstructions within the sight triangle. Upon further investigation with the ITE survey lead, we found that the entire collected set was similar in nature and that policies specifically addressing ISD collection had already been reviewed by this project team. Reviewed Policies The project team conducted a detailed review of the following policies found to be relevant to the research effort and inclusive of established specific details for collecting ISD:  Design Standards & Policies Manual, City of Scottsdale, Arizona (2006).  Subdivision Roads Minimum Construction Standard, North Carolina Department of Transportation (2010).  Road Design Manual Appendix B--Subdivision Street Design Guide, Virginia Department of Transportation (2005).  Design Manual for Roads and Storm Drains, Carroll County, Maryland (2007).  Traffic Studies Manual, Wyoming Department of Transportation (2011).  Intersection Sight Distance Measurement Standards for On-Site Review of Approaches, Oregon DOT (2003).  Handbook for Simplified Practice of Traffic studies, Iowa DOT (2002).  Sight Distance Guidelines, City of Kirkland, WA (2014).  Traffic Signing at Intersections, Bonne County, MO (2009).  Fences & Sight Triangle Zoning Code Regulations, City of Wheat Ridge, CO (2012).  Bureau of Local Roads and Streets Manual, Illinois DOT, Division of Highways (2006).  Highway Safety Investigation Manual for the Oregon Department of Transportation, Oregon DOT, Traffic-Roadway Section (2011).  Category: 900 Traffic Control, Missouri DOT (2011).

NCHRP 17-59 15 Each of the collected policies was reviewed to answer the following questions: 1. Field Method a. What equipment (e.g., measuring wheel, survey equipment) is used to collect ISD in the field? b. Does the method specify the number of staff used? c. Anything else notable about the method (e.g., traffic control requirements, time of day, specific seasons)? 2. Parameters a. Does the presumed driver’s eye height (as it relates to ISD) differ from the AASHTO value of 3.5 feet? b. Does the presumed object height (as it relates to ISD) differ from the AASHTO value of 3.5 feet? c. Are any height parameters given for objects located within the sight triangle? d. Where is the minor road decision point (DP) located? Does the presence of a stop bar have any impact on the DP location? e. Where is the major road critical point (CP) located? Does the major road CP correlate to the roadway centerline or to the right-of-way? f. Is there a minimum width at which a roadside object is considered an obstruction, or is anything located within the sight triangle that exceeds the maximum height considered an obstruction? Does the document provide any guidance on this? 3. Other Policy Specifics a. How is the major road design speed (as it relates to ISD) determined? b. Does the document provide suggestions or countermeasures to use at locations with deficient ISD? If so, list them. c. Anything additional in the policy of interest? In addition to reviewing these published findings, the project team also conducted informal interviews with ten engineers in land development and transportation that collect sight distance or confirm that the sight distance triangle is free of obstructions in the field. These individuals were questioned about the methods they employ in the field and the practical decisions made using their own judgment. These interviews provided valuable insight into how a written policy is implemented and interpreted in the field. The following section provides the findings of the review and follows the structure of these questions. Summary of Policies and Practices Table 2, Table 3, and Table 4 provide a summary of the policies. A discussion of the key points of the review follows the tables.

NCHRP 17-59 16 Table 2. Summary of Field Method. Sources Equipment Used in the Field Number of Staff Notes Design Standards & Policies Manual, City of Scottsdale, AZ Not Specified Not Specified Subdivision Roads Minimum Construction Standards, North Carolina DOT Not Specified Not Specified Road Design Manual Appendix B--Subdivision Street Design Guide, Virginia DOT Not Specified Not Specified Design Manual for Roads and Storm Drains, Carroll County, MD Sighting rod, target rod, measuring wheel Min. of 2 Dimensions of the sighting and target rods are specified: 42" high, and 6" wide at the top Traffic Studies Manual, Wyoming DOT Sighting rod (3.5' ), target rod (4.25') 2 Specific methods outlined for both Uncontrolled and stop-controlled intersections Intersection Sight Distance Measurement Standards for On-Site Review of Approaches, Oregon DOT Not Specified Not Specified Handbook for Simplified Practice of Traffic Studies, Iowa DOT Target/Sighting rods, measuring wheel, hardhat/safety vest, sight distance diagram form 2 Specific methods outlined for both Uncontrolled and stop-controlled intersections Sight Distance Guidelines, City of Kirkland, WA Sight Plan Not Specified Traffic Signing at Intersections, Bonne County, MO Not Specified Not Specified Sight Distance Regulations, City of Wheat Ridge, CO Not Specified Not Specified Illinois’ Bureau of Local Roads and Streets Manual, Illinois DOT, Division of Highways Not Specified Not Specified

NCHRP 17-59 17 Sources Equipment Used in the Field Number of Staff Notes Highway Safety Investigation Manual for the Oregon Department of Transportation, Oregon DOT, Traffic-Roadway Section Unroll measuring tape(3.5') at DP, unroll measuring tape (3.5') with an object(e.g. a clip board)at the top at CP Min. of 2 Visibility Check Category: 900 Traffic Control, Missouri DOT Sighting target, accurate measuring device mounted on an automobile Not Specified Introduce the accurate measuring device

NCHRP 17-59 18 Table 3. Summary of Parameters. Sources Driver's Eye Height Object Height Decision Point (DP) Position Critical Point (CP) Position on Major Road Obstruction in Sight Triangle From Edge of Major Road Minor Road Stop Bar's Impact Maximum Height Minimum Width Design Standards& Policies Manual, City of Scottsdale, AZ 3.5 ft 3.5 ft 15 ft 5 ft from centerline or lane line None 5 ft from nearest lane line (for measurements to the left) or from centerline (for measurements to the right) 24 inch Trees can be considered within the triangle as long as the canopies are above 7 feet. Anything else is obstruction Subdivision Roads Minimum Construction Standards, North Carolina DOT Not specified Not specified 10 ft (from ROW) Along minor road ROW None Along major road ROW Not specified Not specified Road Design Manual Appendix B- -Subdivision Street Design Guide, Virginia DOT 3.5 ft 3.5 ft 14.5 ft 4 ft from centerline or lane line None Center of nearest lane Not specified Not specified

NCHRP 17-59 19 Sources Driver's Eye Height Object Height Decision Point (DP) Position Critical Point (CP) Position on Major Road Obstruction in Sight Triangle From Edge of Major Road Minor Road Stop Bar's Impact Maximum Height Minimum Width Design Manual for Roads and Storm Drains, Carroll County, MD 3.5 ft 3.5 ft Single Use Driveways: 10ft Use-In- Common Driveways: 15 ft Center of lane None Center of nearest lane Not specified Not specified Traffic Studies Manual, Wyoming DOT 3.5 ft 4.25 ft 15 ft Center of lane None Centerline of the roadway Any object that obstruct the driver’s view of an approaching vehicle(4.25 ft) Not specified Intersection Sight Distance Measurement Standards for On-Site Review of Approaches, Oregon DOT 3.5 ft 3.5 ft Not specified Not specified None Not specified Not specified Not specified

NCHRP 17-59 20 Sources Driver's Eye Height Object Height Decision Point (DP) Position Critical Point (CP) Position on Major Road Obstruction in Sight Triangle From Edge of Major Road Minor Road Stop Bar's Impact Maximum Height Minimum Width Handbook for Simplified Practice of Traffic Studies, Iowa DOT 3.5 ft 4.25 ft 10 ft from the stop bar or aligned with the stop sign Center of lane Yes Center of nearest lane Any object that obstruct the driver’s view of an approaching vehicle (4.25 ft) Not specified Sight Distance Guidelines, City of Kirkland, WA Not specified Not specified 14 ft Center of lane None On the center of the through lane(or in the center of the major approach if more than one lane exists) Lower limit: 3 ft Upper limit: 8 ft Not specified Traffic Signing at Intersections, Bonne County, MO Not specified Not specified Not specified Not specified None Not specified Not specified Not specified Sight Distance Regulations, City of Wheat Ridge, CO Not specified Not specified Not specified Not specified None Not specified Private streets: 42” Arterial/Collecto r: 36” Sign pole does not exceed one foot in diameter

NCHRP 17-59 21 Sources Driver's Eye Height Object Height Decision Point (DP) Position Critical Point (CP) Position on Major Road Obstruction in Sight Triangle From Edge of Major Road Minor Road Stop Bar's Impact Maximum Height Minimum Width Illinois’ Bureau of Local Roads and Streets Manual, Illinois DOT, Division of Highways 3.5 ft 3.5 ft 15 ft Center of the lane None Center of nearest lane Any object that obstructs the driver’s view (3.5 ft) Point obstacles (i.e. the driver can move slightly to avoid these obstacles) are not considered sight obstructions Highway Safety Investigation Manual for the Oregon DOT, Oregon DOT, Traffic-Roadway Section 3.5 ft 3.5 ft 14.5 ft (or from edge of crosswalk if present) Not specified None In major rd thru lane closest to (for measurements to left) or farthest from (for measurements to right) the minor approach Anything that won't block driver's view of the entire triangle at the stopped vehicle position Anything that won't block driver's view of the entire triangle at the stopped vehicle position Category: 900 Traffic Control, Missouri DOT 3.5 ft 3.5 ft 12 ft Not specified None Not specified Not specified Not specified

NCHRP 17-59 22 Table 4. Summary of Other Policy Specifics. Sources Determination of Major Road Design Speed Presence of Countermeasures in Manual Notable Items Design Standards & Policies Manual, City of Scottsdale, AZ 10 MPH over expected posted speed limit No 1) Traffic safety triangles to control height of vegetation, structures, etc. 2) ISD requirements for different type of vehicles on 6-, 4-, 3, 2-lane roads Subdivision Roads Minimum Construction Standards, North Carolina DOT Not specified No Road Design Manual Appendix B--Subdivision Street Design Guide, Virginia DOT Not specified No Design Manual for Roads and Storm Drains, Carroll County, MD Posted speed limit plus 10 MPH No 1) Created a driveway sight distance data collection form Traffic Studies Manual, Wyoming DOT Posted speed is used Remove/modify object; Reduce speeds; Install additional traffic control devices Specifies the methods of measuring ISD for uncontrolled and stop-controlled intersections Intersection Sight Distance Measurement Standards for On-Site Review of Approaches, Oregon DOT Includes table of posted speed to design speed with variations range from 5 to 15 mph. No Values of ISD correspond to the AASHTO standard of a 7.5 sec gap. Handbook for Simplified Practice of Traffic Studies, Iowa DOT Great of 85th percentile speed or posted speed limit Remove/modify object; Reduce speeds; Install additional TCDs 1) ISD Case study 2)Checklist for field Sight Distance Guidelines, City of Kirkland, WA Not specified

NCHRP 17-59 23 Sources Determination of Major Road Design Speed Presence of Countermeasures in Manual Notable Items Traffic Safety Manual, Bonne County, MO Not specified Install warning signs Different stopping sight distances for different ADTs Sight Distance Regulations, City of Wheat Ridge, CO Not specified No Illinois’ Bureau of Local Roads and Streets Manual, Illinois DOT, Division of Highways Not specified Provide traffic control devices; Design applications; Remove obstructions Values of ISD correspond to the AASHTO standard of a 7.5 sec gap Highway Safety Investigation Manual for the Oregon Department of Transportation, Oregon DOT, Traffic-Roadway Section Approach speed measured at 250 ft from the intersection Improve approach visibility; Prohibit parking; Remove obstacles; Install warning signs; Reduce speed limit on approaches 1) Visibility Check 2)Method to measure the grade Category: 900 Traffic Control, Missouri DOT Not specified No

NCHRP 17-59 24 Summary of Field Methods The collected policies were reviewed to determine the methods used in the field, including equipment, staffing, and protocol used to collect ISD in the field. The majority of the policies did not specify a method. Five agencies did specify a method, with three of these describing an approach utilizing a sighting target rod. More detailed construction guidelines and field procedures are provided in Traffic Studies Manual (TSM) of Wyoming DOT and Handbook of Simplified Practice for Traffic Studies (HSPTS) of Iowa DOT. These references provide instructions for constructing a target rod out of wood in a “T” shape, with the top portion painted fluorescent orange. The sighting rod is constructed in the same way, but painted flat black. In the field, the observer with the sighting rod stands at the decision point (DP) and positions his/her eye at the top of the sighting rod. The assistant walks away from the observer along the major road toward approaching traffic. During the walk, the assistant is instructed to stop periodically and place the target rod on the pavement for sighting by the observer. The references instruct that this procedure should continue until the top of the target rod could no longer be seen. This position, referred to as the critical point (CP) in this effort, should be marked and the distance from there back to the DP should be measured by the assistant using a measuring wheel. Those policies that described a field method indicated that at least two people should work jointly to collect the measurements in the field. Members of the Oregon DOT staff follow a similar procedure to the Iowa and Wyoming method, but replaced the target sighting rod with an unrolled measuring tape. The staff persons at the CP and DP must position the end of the unrolled tape on the roadway surface, and hold the tape vertically. The visibility is met only when the observer at the DP has full visibility of the tape; that is, the entire 42-inch height must be visible to the observer. Summary of Parameters A review of the collected policies was conducted to find out the parameters that state and local agencies use and whether they correspond to the AASHTO values. The majority of the reviewed agencies have decided to accept the value for the height of driver’s eyes and object from the AASHTO Green Book, which is 3.5 feet for each. These equal heights correspond theoretically to a simultaneous and reciprocal visibility between drivers at the CP and DP. Two policies (TSM of Wyoming DOT and HSPTS of Iowa DOT) assume that the driver’s eye height is 3.5 feet and the object to be sighted is 4.25 feet above the roadway surface, which deviates from the assumption of reciprocal visibility in the Green Book. It is stated in the Green Book that the DP location should be 14.5 feet from the edge of the major road traveled way; where practical, the distance should be increased to 18 feet. Ten agencies specify the location, all choosing a value within the Green Book range, with one agency measuring the distance from right-of-way of the major road. It is also noted that Oregon DOT may measure the setback from the crosswalk if one is present. Eight agencies specify the lateral location of the DP on the minor road, as well, with four of the policies indicating that the DP

NCHRP 17-59 25 should be located at the center of the approaching lane, and two of the policies require a lateral offset of 4 feet or 5 feet from the centerline of the roadway. The review also revealed that the location of the stop bar has limited consideration in the selection of the DP position; with only one policy instructing that the DP should be located 10 feet back from the stop bar or aligned with the stop sign. In determination of the CP, 8 out of 13 policies specify the location on the major road, with 7 policies specifying the location in relation to the centerline, and one policy using the right-of way. Among those polices that place the CP on the major road correlated to the centerline, three different approaches are used to further determine the CP location. The most common method used by three agencies is to measure the ISD along the closest lane to the minor road when measuring to the right and left. In contrast, one agency measures the ISD to the right along the center of the farthest lane from the minor road. Pertaining to the visibility within the sight triangle, the majority of the agencies specify the maximum height of objects that would be considered as obstructions. Seven policies identify the maximum height of obstructions, with four of them identifying the width of the object that is allowed in the sight triangle. Generally, an obstruction is defined as any object that obstructs an approaching vehicle from the view of a driver at the minor DP position. Three out of those seven polices provide more detailed information, with the City of Scottsdale, AZ indicating a 24 inch maximum height for an object to not be considered as an obstruction; Kirkland, WA specifying a lower limit at 3 feet and upper limit at 8 feet; and Wheat Ridge, CO specifying the maximum height at 42 inches on most roadways and 36 inches on private streets and collectors, respectively. Among the four agencies that specify the minimum width of the object, one of the policies (i.e. Sight Distance Regulations, Wheat Ridge, CO) describes the particular item and the corresponding width that is allowed within the sight triangle. The Illinois DOT policy introduces the concept of a point obstacle, which is defined as any obstacle that the driver can move slightly to avoid, thereby eliminating such objects from consideration as sight obstructions. This concept resonated with the project team, as it reflects decisions that are made in the field during ISD measurement, based on conversations with staff responsible for reviewing sight distance triangles in both land development and traffic engineering work. Other Guidance According to the AASHTO Green Book (2011), the major road design speed is a key element used to determine the required sight distance for intersections. The collected policies were reviewed to determine how design speed was considered in relation to ISD collection. As shown in Table 4, six of the policies specified a method, with three policies specifying a quantitative relationship between the design speed and speed limit. For the other three agencies, posted speed, approach speed, or the greater one of posted speed and 85th percentile speed are used instead.

NCHRP 17-59 26 Five of the reviewed policies identified countermeasures that should be considered for deficient sight distance. The treatments include: (1) remove the obstruction; (2) reduce speed; (3) install warning signs; and (4) install extra traffic control devices. Several additional notable items are found in Highway Safety Investigation Manual of the Oregon DOT. This policy is the only one that states that the observer should see the entire sighting rod/tape. It also describes a method to measure the grade when evaluating the ISD by using a SmartLevel placed on the ground 250 feet to the left/right of the intersection on the major road to measure the grade. This project team adopted a similar procedure in the data collection plan. Review of Uncontrolled Intersection Sight Distance Methodologies Introduction Stop and yield signs provide an economical solution for traffic right-of-way assignment at locations that require some level of regulatory control, but do not possess the traffic volumes to warrant a signal or other higher-capacity form of control. Consequently, stop signs and yield signs are the most widely utilized traffic control devices for intersection right-of-way assignment in the United States. However, because the use of these signs – particularly stop signs – leads to inconveniences and inefficiencies for motorists, in addition to installation and maintenance costs for road agencies, they should be used with discretion. At intersection approaches with very low volumes and adequate intersection sight distances (ISD), it may be most economical to utilize no traffic control due to the relatively low likelihood of vehicles arriving at the intersection at the same time. However, although substantial evidence suggests that the absence of traffic control does not impact crashes for very low volume roadways (Mounce 1981; Souleyrette et al. 2006; Stockton et al. 1981), uncontrolled public roadway intersections are used sparingly in most areas of the United States, typically only at very low volume, low speed local streets. The recommended daily traffic volumes thresholds for no traffic control at an intersection are 150 (Souleyrette et al. 2005) and 2,000 (Mounce 1981) vehicles per day on the minor and major street, respectively. Uncontrolled intersections present unique behavioral challenges for approaching drivers. The absence of traffic control for right-of-way assignment forces drivers to revert to the basic right- of-way rule found in Section 11-401 of the Uniform Vehicle Code, which states “When two vehicles approach or enter an intersection from different highways at approximately the same time, the driver of the vehicle on the left shall yield the right-of-way to the vehicle on the right” (National Committee on Uniform Traffic Laws and Ordinances 2000). This places a substantial amount of vigilance on drivers along both the major and minor approaches to 1) detect the intersecting roadway in the absence of a regulatory traffic control device; 2) detect an oncoming vehicle on an intersecting approach; 3) assess the approach speed and distance of the oncoming vehicle; 4) possess an understanding of the right-of-way requirement; and 5) take appropriate action. As uncontrolled intersections are uncommon in several areas of the United States, the

NCHRP 17-59 27 level of driver understanding of this basic rule is likely correlated with driver familiarity/experience with uncontrolled intersections. Furthermore, engineers must also consider that uncontrolled intersections lack visual cues from traffic control devices indicating the need to stop, which suggests utilizing conservative designs. For approaching drivers to decide whether to stop or proceed, uncontrolled intersections must be provided with clear sight triangles along both the major and minor roadway approaches that extend to the decision point, which, as previously described, is the location at which the approaching driver must begin to decide whether to stop or proceed. The need for approach sight distance is similar to that for yield controlled intersections, but is vastly different from stop- controlled intersections, which only require departure sight triangles, contributing to substantially lower setback requirements for obstructions at approaches where stop signs exist. It is also important to consider that the approach sight triangles must be clear at all times, which is often a seasonal issue at intersections adjacent to farmland or other vegetated areas, where crops or vegetation limit the sight distance during portions of the year. In many of these seasonal sight obstruction cases, it is prudent for agencies to utilize two-way or four-way stop control rather than uncontrolled or yield controlled intersections unless the obstructing vegetation can be permanently removed. The need for stop-controlled is amplified as the approach speed increases, as the sight triangles are more likely to pass over private property or may otherwise become impractically large for the roadway agency to maintain. Uncontrolled ISD Methodologies The 2011 Green Book (the current AASHTO procedure at the time of this writing) method for determination of the minimum ISD needed on the approach to an uncontrolled intersection (Case A) is based on allowance for vehicles approaching on either or both roadways to stop prior to reaching the intersection (AASHTO 2011). This method, which was first introduced by AASHTO in the 2001 Green Book (AASHTO 2001), is based on research performed in the mid- 1990s by Harwood et al. (1996), and detailed in NCHRP Report 383. The design ISD values provided in the 2011 Green Book (2011) are based on a 2.5 second perception reaction time along with the assumption that approaching drivers will have already slowed to 50 percent of their midblock running speed (85th percentile speed or design speed for design purposes) upon reaching the intersection, regardless of whether a vehicle is present on the intersecting approach. This preliminary deceleration is assumed to occur during all or part of the perception reaction time at a modest rate up to 5 ft/s2, while braking to a stop occurs at braking rates assumed for stop sight distance (SSD) (11.2 ft/s2) (AASHTO 2011; Harwood et al. 1996). NCHRP 383 also established the eye height and object height requirements of 3.5 feet that were ultimately incorporated into the 2001 and subsequent Green Books (Harwood et al. 1996). Additionally, the 2011 Green Book introduced a provision for intersections with skew angles, which recommended against the use of no traffic control at intersections with oblique angles (AASHTO 2011). Adjustments to the sight triangle legs are provided for grades exceeding ±3 percent.

NCHRP 17-59 28 Prior to the 2001 Green Book, the AASHTO method for determining the required approach ISD for uncontrolled intersections (termed “Case 1” in the 1994 and prior Green Books) was determined based on providing adequate sight distance for drivers on both approaches to adjust speed – but not stop (AASHTO 1994). This procedure assumed that upon detection of the conflicting vehicle, one driver would accelerate while the other would decelerate, thus avoiding any conflict. The ISD calculation was based solely on the distance traversed by a vehicle approaching at the assumed speed (typically the design speed) and a 2.0 second perception reaction time plus 1.0 second speed adjustment time. This method was essentially the same as those presented in the 1990 and 1984 Green Books (AASHTO 1984; AASHTO 1990). Although this method did not provide adequate sight distance for stopping, it was deemed rational due to the relatively low likelihood of vehicular conflicts associated with such low traffic volumes. However, the 1994 Green Book acknowledged that intersections with sight triangles designed based solely on the premise of speed adjustment are not necessarily safe and suggests that the preferred design is to allow sight triangles that are large enough to provide drivers with sufficient time to stop the vehicle prior to reaching the intersection (AASHTO 1994). This statement provided clear impetus for development of the stopping-based ISD design method recommended in NCHRP 383, which served as the basis for the current AASHTO policy. The decision to move toward the full stopping model for determination of design ISD in the 2001 Green Book was to provide a more conservative ISD estimate than that provided by speed adjustment model alone. Specifically, the pre-2001 AASHTO method relied on the drivers of both vehicles taking the correct action; namely one driver decelerates, while the other driver either maintains a constant speed or accelerates (AASHTO 1984; AASHTO 1990; AASHTO 1994). A potential conflict or collision may result if both drivers chose to decelerate. Several researchers had found fault with the speed-reduction ISD model found in pre-2001 Green Books, noting specifically that the sight triangles provide inadequate time for evasive action to be taken in the event of a pending conflict. Easa (2000) suggested that the pre-2001 method resulted in sight triangles that were inadequate for 1 in 1000 intersection approach events (1 in 10,000 is the minimum desirable). McGee and Hooper (1983) suggested that peripheral perception that is typically required for intersection sight distance would necessitate a raising of the perception reaction time to 3.4 seconds. McGee et al. (1984) also recommended a deceleration-based model that would allow time for evasive maneuvers. Similarly, Easa (1998) reasoned that the speed- reduction model caused significant underestimations of ISD requirements, especially when the speed differential between the two intersecting roads was substantial. He claimed that while utilization of the 85th percentile speed or design speed was adequate for the non-yielding approach, a small percentile (e.g., between 1 and 5 percent) approach speed should be selected for the yielding approach. This better accounted for drivers traveling under the design speed and provided safer and more accurate sight distances when designing a roadway. Because the current AASHTO method for determining ISD assumes that drivers decelerate to 50 percent of the midblock running speed prior to reaching an uncontrolled intersection, SSD is not

NCHRP 17-59 29 necessarily accommodated by the sight triangle. However, sight triangles of such magnitude are neither practical nor necessary for typical low volume roadways where uncontrolled intersections are most common. Larger approach sight triangles are more likely to require clearance of obstructions across private property. This is a difficult proposition that particularly affects intersections in rural areas where speeds are typically higher and crop farming is more prevalent. Consequently, using SSD to determine the approach sight triangles at higher speed rural locations would generally prompt the use of two-way or all-way stop control to eliminate the need for such impractically large sight triangles on the intersection approaches. Furthermore, the traffic volumes typically experienced at uncontrolled intersections would generally not be large enough to provide a substantial collision risk. For locations with existing non-moveable obstructions, the critical approach speed for the yielding approach may be determined based on the design speed (or speed limit) along the non- yielding approach and the lateral distance to the obstruction. If the critical speed on the yielding approach is found to be less than the typical approach speed, the speed limit should be lowered or an advisory speed sign applied to encourage approach speeds at or below the critical speed (Pignataro 1973). Unfortunately, such speed zones are relatively ineffective at lowering vehicular approach speeds. Figure 1 provides a comparison of the uncontrolled intersection sight distance design values provided by 1) the 1984, 1990, and 1994 AASHTO Green Books; 2) the 2001, 2004, and 2011 Green Books, along with NCHRP 383; 3) current SSD method; and 4) other alternative methods found in the literature. Compared to the ISD design values for uncontrolled intersections found in the pre-2001 Green Book, those recommended by NCHRP 383 and the current AASHTO Green Book are nearly identical at speeds of 40 mph or less, but begin to increase considerably above the pre-2001 values at speeds of approximately 40 mph and above. Figure 1 also displays that stopping sight distance is nearly always greater than intersection sight distance for the no control condition, regardless of the ISD method utilized.

NCHRP F Discussio The ISD Green Bo than the s underesti or less. R and rema importan follows:  In m (A sp R m  In 5 un ut re 17-59 igure 1. C n of Param model propo oks (AASH peed-reduct mation of IS egardless, it ins as the un t parameters itial approa ethod sugge ASHTO 20 eed for stop ecommenda odel as it m itial slowin 0 percent of controlled ilized in the stricted sigh omparison eters for Cu sed by Har TO 2001; A ion ISD mo D with the has remain controlled I pertaining ch speeds: N st using the 11). Fambr ping sight d tion: Utilize ore accurate g on the app the midbloc urban inters current AA t distance, L of ISD Desi rrent AASH wood et al. ( ASHTO 20 del. Awadal current AAS ed as the pre SD method to ISD were CHRP 383 design spee o et al. (199 istances. the 85th per ly represent roach: NCH k running sp ections in A SHTO meth ovegrove ( 30 gn Values f TO Methodo 1996) and u 04; AASHT lah (2009) d HTO mode ferred unco recommend investigated (Harwood e d or the 85th 7) suggested centile midb s the speed RP 383 sug eed based o rizona (Harw od (AASHT 1978) found or Uncontr logy tilized in th O 2011) has id suggest s l at midbloc ntrolled ISD ed herein. N in greater d t al. 1996) a percentile m using the 8 lock runnin of the traffic gests that dr n data colle ood et al. 1 O 2011). A that approa olled Inters e 2001, 200 received fa light (i.e., 3 k running sp method for evertheless etail with th nd the curre idblock ru 5th percentil g speed in t stream on t ivers initiall cted at seve 996). This a t intersectio ch speeds o ections. 4, and 2011 r less critici 0 feet or les eeds of 50 over a deca , several e findings a nt AASHTO nning speed e operating he stopping he approach y decelerate n low speed ssumption i ns with n the major sm s) mph de s ISD . to s road

NCHRP 17-59 31 are correlated with the frequency with which vehicles on the intersecting roadway are present. Specifically, drivers typically show less of a speed reduction at locations when low volumes are present on the intersecting roadway, in many cases exceeding the safe approach speed. Recommendation: While 50 percent of the midblock running speed represents a reasonable baseline assumption, evidence suggests that drivers do not reduce speeds to such levels. Furthermore, NCHRP 383 only utilized seven low speed intersections in Phoenix, Arizona. Prior research suggests that ISD is most sensitive to approach speed. Therefore, additional research on approach speed reductions is needed, particularly at locations with midblock running speeds of 35 mph and greater.  Perception reaction times: Current AASHTO design value is 2.5 s (AASHTO 2011), which corresponds to the value recommended by McGee and Hooper (1983) and is supported by NCHRP 383 (Harwood et al. 1996). Recommendation: 2.5 seconds is a reasonable and consistent value for perception reaction time. However, detection of a vehicle approaching on the intersecting roadway often requires detection from the peripheral vision, potentially increasing the perception reaction time. Additional research may be necessary for this subject.  Deceleration rates used during initial slowing while approaching from upstream: Current AASHTO method recommends values up to 5 ft/s2 (AASHTO 2011), which are based on the NCHRP 383 field studies (Harwood et al. 1996). Recommendation: Additional research on this subject should be performed, particularly for higher speed approaches and to determine where the initial departure from the midblock running speed begins.  Deceleration rates used during stopping: NCHRP 383 and current AASHTO method recommend the default SSD value of 11.2 ft/s2, which is based on a controlled braking maneuver (Fambro et al. 1997). Previous AASHTO braking assumptions were based on locked-wheel braking maneuver, resulting in much higher values. Controlled braking at signalized intersection in response to the yellow change interval is typically assumed as 10 ft/s2 Recommendation: 11.2 ft/s2 is appropriate for use for ISD design.  Driver eye height and object height: Current AASHTO method recommends 3.5 feet for both parameters based on NCHRP 383, which represents the near maximum value for the portion of a passenger car that should be visible to be detected (AASHTO 2011; Harwood et al. 1996). The 1994 Green Book utilized 3.5 feet for driver eye height and 4.25 feet for object height (AASHTO 1994). Recommendation: 3.5 feet for both parameters is recommended.

NCHRP 17-59 32  Skew angle: Current AASHTO method recommends against using no traffic control at intersections with oblique angles. Recommendation: Intersections with oblique angles should not be left uncontrolled. FIELD DATA COLLECTION This section presents a summary of the field data collection effort in North Carolina, Ohio, and Washington. Site Selection Data were collected in North Carolina (NC), Ohio (OH), and Washington (WA), for the reason that they could individually and collectively offer the following:  Completeness, quality, and availability of data. FHWA’s Highway Safety Information System (HSIS) is a multistate database that contains quality crash, roadway inventory, and traffic volume data for a select group of States. NC, OH, and WA have all participated in the HSIS program for many years.  Regional and terrain diversity. WA, OH, and NC represent the Pacific Northwest, Midwest, and Southeast, respectively. In addition, in NC and WA, data were collected in three distinct regions in each state to better account for the terrain diversity.  Proximity of project team members. This allowed efficient use of project resources for making the field-intensive ISD measurements. Criteria for Inclusion Initially, the project team collected data for 250 sites in each state based on the preliminary sample size estimate. However, the results of the preliminary analysis of the NC data indicated that there were fewer crashes than expected at the study sites. As a result, approximately 20 sites were added in each state. Therefore, a total of 832 sites were ultimately included in the field study. A Geographic Information Systems (GIS) database was created from State Department of Transportation (DOT) data and served as a starting point for the site selection in each state. The availability of traffic volume data—particularly on the minor roadways was a key consideration to the overall site selection approach, as this data element can become the most expensive one to obtain. The term “site” herein refers to each minor road leg of an intersection; that is, a three-leg intersection (with one minor road approach) would constitute one site, while a four-leg intersection (having two minor road approaches) would constitute two sites. With all the GIS layers displayed on one map, the project team looked for the corridors that have volumes available not only for the mainline but also for many of its cross streets. In this step, the site selection in urban areas of NC was different since very little traffic data were available for urban areas. As a result, the project team randomly selected sites regardless of traffic data for urban

NCHRP 17-59 33 sites. Once the corridor was identified, the project team checked the traffic control type and number of lanes on the minor road for each intersection along the corridor using Google Earth. Only stop-controlled intersections with two-lane minor roads were included in the preliminary potential site database. Once the potential sites were identified, a series of factors were checked to see if the sites were eligible for the analysis and modeling. The following precluded an intersection from being used in the analysis:  A site had undergone recent and significant physical change (e.g., the addition of a turn lane), as determined by a review of historical aerial imagery in Google Earth.  An intersection was relatively new—having opened after 2005. (Because the crash analysis period included data from 2008-2010, any road opened on January 1, 2006 or later would not have had a two-year acclamation period prior to the crash years.)  The minor road approach was characterized by a significantly steep (i.e., more than three percent) positive grade. This factor was verified in the field.  For rural sites - the site had a signalized intersection (not a flashing beacon or pedestrian hybrid beacon) within the intersection sight distance. This factor did not impact locations in urban areas.  A site was not in the HSIS roadway data file. Since the crashes were taken from the HSIS database, it was essential to ensure the intersections existed in the HSIS database. Additionally, the team attempted to avoid sites with extreme skews or unusual alignments. Generally, the team avoided sites with skews greater than 30 degrees based on the AASHTO Green Book guidance that deviation from 90 degrees is permissible, and that an angle of at least 60 degrees provides most of the same benefits as a right-angle intersection (AASHTO 2011). However, approximately 7 percent of sites include a skew greater than 30 degrees, which is representative of the populations of intersections along the corridors included in the study. An unpublished report by Harkey et al. (in-progress) revealed that any intersection angle that is less than 90 degrees would affect the safety of the intersection. Therefore, actual skew was measured at all sites in order to assess the effect on crashes. Figure 2, Figure 3, and Figure 4 show the data collection sites in each state.

NCHRP 17-59 Figure 2. Figu Data Collec re 3. Data C 34 tion Site in ollection S North Car ites in Ohio olina. .

NCHRP Data Co Table 5 p were iden required Required their inco elements by allowi are defin each stud office). It for the re significan Access to form is sh 17-59 llection Me rovides the tified by th or desired f elements re rporation in are conside ng for a mo ed as desired y location. T was deeme ason that 1) t physical c streamline own in App Figure 4 thodology specific var e project tea or a crash-ba present the to the mode red more an re detailed i , the projec able 5 also d necessary the availabl hanges had and standar endix A. . Data Coll iables for cr m, and whet sed cross-s minimum in l is consider cillary in na nvestigation t team colle displays the to field veri e data were been made. dize the data 35 ection Sites ash, roadwa her data for ectional ana formation n ed essential ture and, if a . Note that w cted all vari data source fy the data e not always u A data entry entry proce in Washing y, and traffi these variab lysis through ecessary to to the study vailable, se hile severa ables (both r s of each da lements tha p-to-date a form was d ss across th ton. c/operations les were co this resear conduct the . Meanwhil rve to enhan l of the vari equired and ta element ( t were colle nd 2) to ensu eveloped w e entire proj categories t nsidered ch effort. analysis, an e, desired ce the analy ables in Tab desired) fo i.e., field or cted in the o re that no ithin Micro ect team. Th hat d sis le 5 r ffice soft is

NCHRP 17-59 36 Table 5. Crash, Roadway, and Traffic/Operations Data Elements of Interest. Data Type Data Element Require/ Desire Description/Comment Data Sources Crash Location Req. Typically consists of county, route, and milepost HSIS Database Type Req. Often recorded as first harmful event HSIS Database Severity Req. Worst injury sustained during crash HSIS Database Initial direction of vehicles Req. Direction of travel immediately before crash HSIS Database Sequence of events Des. Events of crash listed in chronological order HSIS Database Light condition Des. Defined as light or dark based on police report HSIS Database Weather condition Des. Defined as dry or wet based on police report HSIS Database Vehicle type Des. Identifies the type of vehicle based on police report HSIS Database Road- way Major road functional class Req. E.g., principal/minor arterial, major/minor collector, local HSIS Database Minor road functional class Req. E.g., principal/minor arterial, major/minor collector, local HSIS Database Area type Req. E.g., rural, urban HSIS Database Number of through lanes Req. Both directions of the major road Google Earth Divided/undivided major road Req. Undivided, divided, or two-way left turn lane Google Earth Available ISD Req. Measured looking left and right Field Confirm visibility of traffic control Req. Simple yes/no confirmation that minor road driver has clear vantage of the traffic control device(s) at major road Field Horizontal curvature Req. Indicating presence and direction looking left and right Google Earth Lane width Des. Width of each lane in feet Field(NC), Google Earth (WA, OH)

NCHRP 17-59 37 Data Type Data Element Require/ Desire Description/Comment Data Sources Shoulder width Des. Width of paved shoulder in feet Field(NC), Google Earth (WA, OH) Median width Des. Width of median in feet Field(NC), Google Earth (WA, OH) Access density Des. Number of stop-controlled intersections or driveways along major road within 0.25 mi of subject intersection (count a 4- leg intersection as 2) Google Earth Lighting presence Des. Field check for luminaires at the intersection Field Vertical grade Des. Measured with smart level at 3 locations left and right along major road Field Quality of sight distance Des. Appraise ISD quality in a subjective way (scores from 1-3 with 1 being least objects and 3 being many objects) Webinar Traffic / Ops Major road AADT Req. Including year that volume was determined State DOT, HSIS Database Minor road AADT Req. Including year that volume was determined State DOT, local agencies, field counts Traffic control Req. Stop, yield, etc. Field Posted speed limit Req. Nearest upstream posted speed limit on approach of interest Field Presence of left turn lanes Req. Along major road Field Presence of right turn lanes Req. Along major road Field Presence of left turn lane Req. Along minor road Field Presence of right turn lane Req. Along minor road Field

NCHRP 17-59 38 Data Type Data Element Require/ Desire Description/Comment Data Sources Intersection angle Des. Approximate angle between major road and approach of interest Google Earth Presence of on-street parking Des. Along major road Field

NCHRP 17-59 39 Crash Data Four years of crash data were compiled from HSIS accident and vehicle files (i.e., 2008-2011 in OH and WA, 2009-2012 in NC). The project team used SQL code to build queries in MS Access, used county/route/milepost to link the site data to crash files, and used case numbers to link the accident file to the vehicle file. Only target crashes were considered for the analysis, which were defined by the following criteria:  The crash has a milepost that is within a 250-foot radius of the intersection milepost.  The crash involves a vehicle on the mainline and a vehicle on the minor road. In addition to target crashes, several subsets of the target crashes were investigated, including the following:  Injury crashes: A subset of target crashes which result in one or more fatalities or injuries. In all regions, injury crashes were defined as KABC crashes on the KABCO scale.  Angle crashes: A subset of target crashes where the crash type is defined as an angle crash. Angle crashes include all right-angle crashes as well as angle crashes with turning vehicles. o Right-angle crashes: A subset of target crashes where the crash involves vehicles traveling straight from major and minor roadways. o Left-turn crashes: A subset of angle crashes where the crash involves a vehicle on the major road and a vehicle turning left from the minor road. o Right turn crashes: A subset of angle crashes where the crash involves a vehicle on the major road and a vehicle turning right from the minor road.  Daytime crashes: A subset of target crashes where the crash occurs during lighted conditions (i.e., daylight). Table 6 presents the combination of maneuvers that were used to identify the crashes of interest across the three states. Due to limitations in crash data, it was not possible to identify left-turn and right-turn crashes from NC data. Only OH and WA data were used to explore these specific crash types. Angle crashes were identified based on the crash type, which is also described first harmful event, as shown in Table 7. Although several crash types and conditions were analyzed, the final results are based on target crashes and target fatal and injury crashes due to the lack of sample size and the similarity between target, angle, and daytime crash results.

NCHRP 17-59 40 Table 6. Combination of Maneuvers. Crash Type Maneuvers of Vehicle on Major Road Maneuvers of Vehicle on Minor Road Target crashes Traveling straight, turning right, or turning left or parked Traveling straight, turning right, turning left Right-angle crashes Traveling straight Traveling straight Right turn crashes Traveling straight Turning right Turning right Turning left Left-turn crashes Traveling straight Turning left Turning right Turning left Table 7. Crash Types Used in Identifying Angle Crashes. State Crash Type(s) NC Angle; left-turn, different roadway; right turn, different roadway OH Angle WA Angle To more effectively quantify the effect of ISD on expected crash frequency, the project team coded an additional field that describes the directional approach for each crash. Each site was divided into two analysis units (i.e., left direction and right direction) by major road direction, as shown in Figure 5. The term “analysis unit” here refers to each major road approach of one site; that is, the major road at the right side when standing at the minor road belongs to the right- directional analysis unit, and the major road at the left side when standing at the minor road belongs to left-directional analysis unit.

NCHRP Using the was trave informati  If ca ro  If ca ro In the en values of represent Assessme An explo results of sight for driver in number o 17-59 vehicle tra ling on the on, the cras the major r tegorized a adway and the major r tegorized a adway and d, this appro an approach s the target nt of Interse ratory comp the literatur the minor ro appropriate f point obje Figure 5. L vel direction major road a h, roadway, oad vehicle s a right-dire traffic data a oad vehicle s a left-direc traffic data f ach resulted level ISD crash freque ction Sight onent of thi e review, th ad driver co gap selectio cts on the ne eft- and Rig data, the pr nd which v and traffic d was travelin ctional targ ssociated w was travelin tional target or the left-d in target cr (i.e., one ob ncy for one Distance Qu s study is th e project tea uld play eith n. In the pil ar and far s 41 ht-Directio oject team i ehicle was tr ata were co g from the r et crash and ith the right g from the l crash and t irectional an ash counts b servation of minor road ality e notion of s m hypothes er a benefic ot study, the ide of the m nal Analys dentified wh aveling on t mbined as f ight directio the crash da -directional eft direction he crash dat alysis unit. eing associa target crash and major ro ight distanc ized that roa ial or a detr team devel inor approac is Unit. ich crash-in he minor ro ollows: n, the target ta were me analysis un , the target c a were merg ted with spe frequency i ad approac e quality. B dside objec imental role oped a meth h leg. Upon volved veh ad. Using th crash was rged with it. rash was ed with cific, measu n the databa h combinati ased on the ts in the line in assisting od to count review of t icle is red se on). of the the hese

NCHRP 17-59 42 data, the NCHRP project panel suggested eliminating collecting this data element. Instead the panel directed the project team took a more subjective approach to appraising ISD quality—one that was more qualitative in nature—to explore the hypothesis that this has an impact on safety To develop this subjective approach, the team hypothesized that discrete roadside objects (e.g. poles, posts, trees, signs, guardrail, buildings, etc.) may assist drivers with estimating the distance and speed of an approaching vehicle on the major roadway. The team also suggested that roadside objects may clutter the sight triangle of a driver in some situations, potentially causing momentary obstruction of an oncoming vehicle, further confounding the gap estimating process. The project team held a web conference call to discuss possible subjective scoring criteria. A sample training set of intersection photos were viewed from all three data collection states representing a range of roadside objects. After discussion, it was decided that a more quantitative approach would yield repeatability and consistency in the scoring. Thus, ISD quality of each site was assessed based the photos taken from the best available DP looking both left and right. Each photo was scored as possessing one of three conditions based on the number of objects on the far side of the major road approach or on the amount of clutter within the sight distance triangle, taking into account the height and girth of the objects. A condition score of “1” represented the least amount of visual clutter or objects, while a score of “3” represented the most clutter. Separate scores were provided for a leftward and rightward view at each site. Three senior project team members (one from each state) provided independent scoring of each individual site and the most common score between the three represented the final score for the particular direction of view at that site. For example, if two of the three members considered the ISD quality as condition 1, the ISD quality would be considered as condition 1 regardless of the score given by the third member. A set of figures are presented to illustrate this scoring process. Figure 6 represents ISD quality condition 1 because very few objects including signs, mailboxes, utility poles are in the field of view. Figure 7 represents ISD quality condition 2 and Figure 8 represents ISD quality condition 3. Compared to Figure 6 and Figure 7, Figure 8 has numerous objects including signs, light posts, and trees in the field of view.

NCHRP 17-59 Fig Fig ure 6. ISD ure 7. ISD 43 Quality Co Quality Co ndition 1. ndition 2.

NCHRP Field Co The field foliage is sighting p measurem  T th  T gr The proje Final App Selection at each si Green Bo was base ISD mea The Gree through l the mino minor roa entering 17-59 llection of A measureme robust). Th erson and o ents, the pr he traffic co e minor roa he grade me ade. ct team pro roach to C of Decision te. The ISD ok and the D d on the loc surements w n Book stan ane (regardl r road center d vehicles a the major ro Fig vailable ISD nts were tak e data collec ne or two ru oject team v ntrol device d driver. asured at th ceeded with ollecting Av Point: For measureme P location ation of min ere taken at dard DP is l ess of wheth line. The dr nd their sto ad. Upon co ure 8. ISD en during la tion teams c nners). Upo erified the f (s) at the int e minor road further field ailable ISD the majority nts were bas representati or approach each site— ocated 14.5 er a bicycle iver selected pping points mpleting th 44 Quality Co te spring, su onsisted of n arrival to ollowing ite ersection wi was less th measureme of sites, tw ed on the st ve of typica drivers upo two for each feet back fr lane or park DP was lo to replicate e data collec ndition 3. mmer, and two or three the site and ms to ensur th the majo an or equal nts if the si o sets of av andard DP l l drivers at e n stopping a DP in each om the edge ing lane wa cated based where mos tion, the pro early fall (i. members ( prior to per e site eligib r road was c to three perc te met both ailable ISD ocation spec ach study ap nd looking. direction (l of the neare s present) a on observat t drivers act ject team d e., when the including on forming fiel ility: learly visibl ent positive requirement were measu ified by the proach, wh In total, fou eft and right st major ro nd 4 feet fro ion of sever ually stop b etermined th e d e to s. red ich r ). ad m al efore at

NCHRP the drive reason, th Equipme road DPs used to d  T an sc li si  T sl m ab h fr in A it th Procedur away fro the neare this unsa when the the range between between horizonta measurin Based ini project te available than or e “1,321 fe 17-59 r selected D e project te nt: A sighti looking in etermine the he rangefind d was moun ope was 42 mits the imp ghting perso he target con ow moving etal pole su ove the gro as sufficient om the rang creases its c dditionally, em at nation e target wil e: The runn m the DP, id st travel lan fe and then t full metal t finder’s sco the CP and D the CP and t l curves or g wheel to d tial field me am used 1,3 ISD. There qual to 1,320 et” to indica P was the sa am determin ng height of each directio limits of av er includes ted on top o in. above th acts of the v n. sisted of a vehicle sign ch that the b und, as show density to p efinder, whi onspicuity t because thi al home imp l be straightf ers moved t eally positio e, with the e hey stood c arget just be pe. This loc P lacked cu he DP (i.e., significant v etermine th asurements 20 feet (or o fore, the exa feet. Avail te an exact me location ed that the s 42 in. was u n. The com ailable ISD a scope with f a monopo e ground. T ision charac standard me mounted to ottom of the n in Figure rovide adeq le its bright o the sightin s sign is typi rovement s orward. he target alo ning thems xception tha loser to the o gan to disap ation becam rvature, the the availabl ertical curva e distance fr combined w ne quarter m ct value wa able ISDs o value was n 45 as the stand tandard DP sed to meas bination of a looking left 4x magnifi d such that he magnific teristics of tal 14-in. by a piece of l sign was 4 9. The meta uate bounce orange colo g person. cally an in-s tores, replic ng the majo elves in the t conditions utside edge pear from th e the CP for rangefinde e ISD). If th ture betwee om the CP t ith input fro ile) as the s recorded if bserved to b ot determine ard DP 83 p was approp ure the avai laser range and right fr cation the ation the 14-in. umber or 2 in. l sign -back r tock ation of r road center of rendered of the trave e view of th that directio r was used t e major road n the CP an o the DP. m the NCH threshold di the availab e greater tha d. Figure 9. Use ercent of th riate for fur lable ISD fr finder and m om each DP l lane. The r e sighting p n of view. o determine included o d DP, the ru RP project stance for m le ISD obse n 1,320 fee Sighting Ta d in Ohio. e time. For t ther analysis om two min etal target . unners stop erson throu If the major the distance ne or more nner used th panel, the easuring the rved was les t were label rget his . or was ped gh road e s ed as

NCHRP 17-59 46 Summary of Selected Sites Data were ultimately collected at a total of 832 stop-controlled approaches. Table 8 displays general characteristics of the sites that were ultimately selected for data collection, stratified by state, area type, and terrain. Table 8. Sites Stratified by State, Area Type, and Terrain. Terrain NC (N=286) OH (N=281) WA (N=265) Total (N = 832) Rural Urban Rural Urban Rural Urban Rural Urban Level 34 1 128 62 71 49 233 112 Mountainous 48 36 0 0 1 0 49 36 Rolling 74 93 48 43 55 89 177 225 Grand Total 156 130 176 105 127 138 459 373 Three types of major roads were considered in this study (i.e., two-lane, four-lane divided, and four-lane undivided). All minor approaches were on two-lane, two-way roadways. Table 9 shows the distribution of major route facility types by area type in the three states. Table 9. Sites Stratified by State, Major Route Type, and Area Type. Major Route Facility Type NC OH WA Total Rural Urban Rural Urban Rural Urban Rural Urban 2-lane Major 65 43 112 53 72 63 249 159 4-lane Major Undivided 40 46 23 33 7 47 70 126 4-lane Major Divided 51 41 41 19 48 28 140 88 Grand Total 156 130 176 105 127 138 459 373 Initially, three years of crash data were obtained for each site. Since the average crash frequency for three years was relatively low, the project team decided to extend the analysis period by one year, to include 2011. Additionally, the 2008 HSIS data for NC has known quality issues, so 2008 was excluded and years 2009-2012 were used instead. For both WA and OH, 2012 HSIS data were not available; therefore 2008-2011 HSIS data were utilized. Table 10 shows the number of target crashes for each state stratified by year.

NCHRP 17-59 47 Table 10. Target Stratified by State and Year. State 2008 2009 2010 2011 2012 Total Average Per Site WA 36 38 46 37 -- 157 0.592 OH 64 75 80 84 -- 303 1.078 NC -- 44 72 75 75 266 0.930 Total 100 157 198 196 75 726 0.873 The following provides an overview of the data collected in each of the three states with the corresponding summary statistics for all variables. North Carolina Data were collected for 286 sites in NC, consisting of 572 approach directions (each site had two directions) for analysis. There were a grand total of 2,288 direction-years of data available (i.e., 572 directions * 4 years). Table 11 provides the summary statistics, including the minimum, maximum, and average value for continuous variables stratified by major route type and area type. In preparation for data analysis, variables were transformed into a format appropriate for modeling and new names were assigned. Table 11 presents the variable names, brief descriptions, and summary statistics, including the number of observations (i.e., direction-years), minimum, maximum, mean, and standard deviations for all independent continuous variables. Table 12 presents the variable names, brief descriptions, and proportion of occurrence for categorical or indicator variables

NCHRP 17-59 48 Table 11. Summary Statistics for NC Continuous Independent Variables. Variable Name Variable Description Obs. Minimum Maximum Mean St. Dev Accdens Number of access points within 0.25 miles 2,288 0 47 12.95 9.75 Angle Intersection angle (deg) 2,288 41 90 80.48 10.65 Avaiisd Standard available ISD (ft) 2,288 99 1,321 934.79 362.91 Bestisd Best available ISD (ft) 2,288 208 1,321 952.30 347.29 Grd100 Grade 100 ft from the intersection (percent) 2,288 -7.2 7.1 -0.14 2.48 Grd250 Grade 250 ft from the intersection (percent) 2,288 -9.5 7.7 -0.13 2.55 Grd500 Grade 500 ft from the intersection (percent) 2,288 -8.1 7.9 -0.06 2.58 LW Lane width (ft) 2,288 9 14 11.51 0.79 Majaadt Major route average annual daily traffic (veh/day) 2,288 1,600 45,000 13,365 8,061 Minaadt Minor route average annual daily traffic (veh/day) 2,288 100 7,790 1,140 931 MW Median width (ft) 2,288 0 93 11.34 17.94 Spdlmt Posted speed limit (mph) 2,288 20 60 47.31 7.93 SW Shoulder width (ft) 2,288 0 12 1.39 1.93 Target Target crashes 2,288 0 7 0.12 0.40 Injury Target fatal and injury crashes 2,288 0 2 0.05 0.23

NCHRP 17-59 49 Table 12. Summary Statistics for NC Categorical Independent Variables. Variable Name Variable Description Proportion Areatype Urban area type 45.5 % Rural area type 54.5 % Legs Three legs 55.2 % Four legs 44.8 % Direction Left direction 50.0 % Right direction 50.0 % Lttrnmaj Left turn lane on the major route 44.8 % No left turn lane on the major route 55.2 % Rttrnmaj Right turn lane on the major route 33.6 % No right turn lane on the major route 66.4 % Lttrnmin Left turn lane on the minor route 7.0 % No left turn lane on the minor route 93.0 % Rttrnmin Right turn lane on the minor route 7.3 % No right turn lane on the minor route 92.7 % Majrtepark Parking on the major route 1.4 % No parking on the major route 98.6 % Lighting Lighting is present 41.6 % No lighting is present 58.4 % Median Divided median is present 32.5 % No divided median is present 67.5 % Majclass Major route is classified as principal arterial (Majparterial) 35.3 % Major route is classified as major arterial (Majmajarterial) 12.6 % Major route is classified as minor arterial (Majminarterial) 44.4 % Major route is classified as major collector (Majmajcollector) 0.0 % Major route is classified as collector (Majcollector) 4.9 % Major route is classified as minor collector (Majmincollector) 0.0 % Major route is classified as local (Majlocal) 2.8 %

NCHRP 17-59 50 Variable Name Variable Description Proportion Minclass Minor route is classified as principal arterial (Minparterial) 0.3 % Minor route is classified as major arterial (Minmajarterial) 2.8 % Minor route is classified as minor arterial (Minminarterial) 0.7 % Minor route is classified as major collector (Minmajcollector) 0.0 % Minor route is classified as collector (Mincollector) 2.8 % Minor route is classified as minor collector (Minmincollector) 9.1 % Minor route is classified as local (Minlocal) 84.3 % Minor route is unclassified (Minunclassified) 0.0 % Crve A horizontal curve is present 61.2 % No horizontal curve is present 38.8 % Terrain Level 12.2 % Rolling 58.4 % Mountainous 29.4 % Isdquality Quality level 1 – few objects 10.7 % Quality level 2 – medium objects 83.7 % Quality level 3 – many objects 5.6 % Ohio Data were collected for 281 sites in OH, consisting of 562 approach directions for analysis. There were a grand total of 2,248 direction-years of data available (i.e., 562 * 4 years). Table 13 provides the summary statistics, including the minimum, maximum, and average value for continuous variables stratified by major route type and area type. In preparation for data analysis, variables were transformed into a format appropriate for modeling and new names were assigned. Table 13 presents the variable names, brief descriptions, and summary statistics, including the number of observations (i.e., direction-years), minimum, maximum, mean, and standard deviations for all independent continuous variables. Table 14 presents the variable names, brief descriptions, and proportion of occurrence for categorical or indicator variables.

NCHRP 17-59 51 Table 13. Summary Statistics for OH Continuous Independent Variables. Variable Name Variable Description Obs. Minimum Maximum Mean St. Dev Accdens Number of access points within 0.25 miles 2,248 0 47 12.79 12.02 Angle Intersection angle (deg) 2,248 17 90 78.89 15.11 Avaiisd Standard available ISD (ft) 2,228 100 1,321 1,089 333.55 Bestisd Best available ISD (ft) 2,248 273 1,321 1,144 281.93 Grd100 Grade 100 ft from the intersection (percent) 2,248 -10.2 7.7 -0.07 1.58 Grd250 Grade 250 ft from the intersection (percent) 2,248 -9.3 5.3 -0.16 1.49 Grd500 Grade 500 ft from the intersection (percent) 2,248 -8.7 6.9 -0.13 1.55 LW Lane width (ft) 2,248 10 18 11.41 1.37 Majaadt Major route average annual daily traffic (veh/day) 2,248 267 32,341 7,257 6,423 Minaadt Minor route average annual daily traffic(veh/day) 2,248 41 9,212 1,386 1,600 MW Median width (ft) 2,248 0 96 7.12 17.24 Spdlmt Posted speed limit (mph) 2,248 25 65 49.20 10.39 SW Shoulder width (ft) 2,248 0 10 2.72 2.67 Target Target crashes 2,248 0 8 0.13 0.50 Injury Target fatal and injury crashes 2,248 0 5 0.07 0.33

NCHRP 17-59 52 Table 14. Summary Statistics for OH Categorical Independent Variables. Variable Name Variable Description Proportion Areatype Urban area type 37.4 % Rural area type 62.6 % Legs Three legs 27.8 % Four legs 72.2 % Direction Left direction 50.0 % Right direction 50.0 % Lttrnmaj Left turn lane on the major route 28.6 % No left turn lane on the major route 71.4 % Rttrnmaj Right turn lane on the major route 1.2 % No right turn lane on the major route 98.8 % Lttrnmin Left turn lane on the minor route 3.6 % No left turn lane on the minor route 96.4 % Rttrnmin Right turn lane on the minor route 2.8 % No right turn lane on the minor route 97.2 % Majrtepark Parking on the major route 6.0 % No parking on the major route 94.0 % Lighting Lighting is present 30.6 % No lighting is present 66.4 % Median Divided median is present 22.4 % No divided median is present 77.6 % Majclass Major route is classified as principal arterial (Majparterial) 33.1 % Major route is classified as major arterial (Majmajarterial) 0.0 % Major route is classified as minor arterial (Majminarterial) 27.8 % Major route is classified as major collector (Majmajcollector) 31.3 % Major route is classified as collector (Majcollector) 4.6 % Major route is classified as minor collector (Majmincollector) 3.2 % Major route is classified as local (Majlocal) 0.0 %

NCHRP 17-59 53 Variable Name Variable Description Proportion Minclass Minor route is classified as principal arterial (Minparterial) 0.0 % Minor route is classified as major arterial (Minmajarterial) 0.0 % Minor route is classified as minor arterial (Minminarterial) 0.3 % Minor route is classified as major collector (Minmajcollector) 10.7 % Minor route is classified as collector (Mincollector) 0.4 % Minor route is classified as minor collector (Minmincollector) 0.7 % Minor route is classified as local (Minlocal) 87.9 % Minor route is unclassified (Minunclassified) 0.0 % Crve A horizontal curve is present 27.2 % No horizontal curve is present 72.8 % Terrain Level 67.6 % Rolling 32.4 % Mountainous 0.0 % Isdquality Quality level 1 – few objects 12.8 % Quality level 2 – medium objects 76.3 % Quality level 3 – many objects 10.9 % Washington Data were collected for 265 sites in WA, consisting of 524 approach directions for analysis. In WA, data were only considered in one direction for some sites due to inconsistencies with the number of approach lanes (i.e., one site had three lanes on the major road, with one lane on one approach and two-lane on the other approach). There were a grand total of 2,096 direction-years of data available (i.e., 524 directions * 4 years). Table 15 shows the summary statistics, including the minimum, maximum, and average value for continuous variables stratified by major route type and area type. In preparation for data analysis, variables were transformed into a format appropriate for modeling and new names were assigned. Table 15 presents the variable names, brief descriptions, and summary statistics, including the number of observations (i.e., direction- years), minimum, maximum, mean, and standard deviations for all independent continuous variables. Table 16 presents the variable names, brief descriptions, and proportion of occurrence for categorical or indicator variables.

NCHRP 17-59 54 Table 15. Summary Statistics for WA Continuous Independent Variables. Variable Name Variable Description Obs. Minimum Maximum Mean St. Dev Accdens Number of access points within 0.25 miles 2,096 1 50 10.35 9.34 Angle Intersection angle (deg) 2,096 8 90 80.91 12.85 Avaiisd Standard available ISD (ft) 2,096 136 1,321 979.56 358.62 Bestisd Best available ISD (ft) 2,096 228 1,321 1,001 343.83 Grd100 Grade 100 ft from the intersection (percent) 2,096 -9.1 8.2 -0.02 2.10 Grd250 Grade 250 ft from the intersection (percent) 2,096 -8.6 8.6 -0.01 2.22 Grd500 Grade 500 ft from the intersection (percent) 2,096 -8.3 7.0 -0.11 2.24 LW Lane width (ft) 2,096 10 15 11.53 0.74 Majaadt Major route average annual daily traffic (veh/day) 2,096 537 36,009 13,619 7,971 Minaadt Minor route average annual daily traffic (veh/day) 2,096 88 15,759 1,432 1,698 MW Median width (ft) 2,096 0 80 12.37 23.23 Spdlmt Posted speed limit (mph) 2,096 25 70 48.07 10.45 SW Shoulder width (ft) 2,096 0 14 4.12 3.14 Target Target crashes 2,096 0 4 0.07 0.33 Injury Target fatal and injury crashes 2,096 0 3 0.03 0.20

NCHRP 17-59 55 Table 16. Summary Statistics for WA Categorical Independent Variables. Variable Name Variable Description Proportion Areatype Urban area type 52.5 % Rural area type 47.5 % Legs Three legs 60.9 % Four legs 39.1 % Direction Left direction 50.6 % Right direction 49.4 % Lttrnmaj Left turn lane on the major route 51.1 % No left turn lane on the major route 48.9 % Rttrnmaj Right turn lane on the major route 13.9 % No right turn lane on the major route 86.1 % Lttrnmin Left turn lane on the minor route 6.9 % No left turn lane on the minor route 93.1 % Rttrnmin Right turn lane on the minor route 8.8 % No right turn lane on the minor route 91.2 % Majrtepark Parking on the major route 0.8 % No parking on the major route 99.2 % Lighting Lighting is present 76.1 % No lighting is present 23.9 % Median Divided median is present 29.0 % No divided median is present 71.0 % Majclass Major route is classified as principal arterial (Majparterial) 53.8 % Major route is classified as major arterial (Majmajarterial) 0.8 % Major route is classified as minor arterial (Majminarterial) 34.4 % Major route is classified as major collector (Majmajcollector) 11.0 % Major route is classified as collector (Majcollector) 0.0 % Major route is classified as minor collector (Majmincollector) 0.0 % Major route is classified as local (Majlocal) 0.0 %

NCHRP 17-59 56 Variable Name Variable Description Proportion Minclass Minor route is classified as principal arterial (Minparterial) 1.9 % Minor route is classified as major arterial (Minmajarterial) 0.0 % Minor route is classified as minor arterial (Minminarterial) 9.2 % Minor route is classified as major collector (Minmajcollector) 21.2 % Minor route is classified as collector (Mincollector) 7.6 % Minor route is classified as minor collector (Minmincollector) 7.6 % Minor route is classified as local (Minlocal) 0.0 % Minor route is unclassified (Minunclassified) 52.5 % Crve A horizontal curve is present 53.4 % No horizontal curve is present 46.6 % Terrain Level 45.6 % Rolling 54.0 % Mountainous 0.4 % Isdquality Quality level 1 – few objects 3.1 % Quality level 2 – medium objects 56.3 % Quality level 3 – many objects 40.6 % ANALYTICAL METHOD A cross-sectional study design with count regression modeling was used to quantify the relationship between safety and ISD. A cross-sectional study design is a type of observational study used to analyze a representative sample of observations at a specific point in time. The safety effect is estimated by taking the ratio of the estimated expected crash frequency for two groups, one with the feature of interest and the other without the feature of interest. In this case, the feature of interest is the ISD. For this method to work, the two groups should be similar in all regards except for the feature of interest. In practice, this is difficult to accomplish and multivariable regression models were used to estimate the safety effects of one feature while controlling for other characteristics that vary among sites. In this case, expected crash frequency was the dependent variable of interest and explanatory variables listed in Tables 11 through 16 were considered for the right-hand-side of the model, including traffic volumes and other roadway and operational characteristics. Regression coefficients were estimated during the modeling process for each of the explanatory variables. The coefficients represent the expected change in the dependent variable (expected crash frequency) due to a unit change in the explanatory variable, all else being equal.

NCHRP 17-59 57 The current state-of-the-practice for developing statistical road safety models of this type is to assume a log-linear relationship between expected crash frequency and site characteristics. Generalized Linear Modeling (GLM) techniques were applied to develop the models, and a log- linear relationship was specified using a negative binomial error structure. The negative binomial error structure also has advantages over the Poisson distribution in that it allows for over- dispersion that is often present in crash data (i.e., the variance is greater than the mean). It is typical in statistical road safety modeling for this “additional dispersion” to be specified as a dispersion parameter multiplied by the expected number of crashes squared (NB-2 model). Therefore, the models of the expected numbers of target crashes for this analysis were structured as follows: λi = EXP(Xiβ+εi) where: λi = expected number of target crashes for the ith observation, with an observation in this analysis being one minor road and major road approach combination; Xi = a matrix of explanatory variables associated with λi, including available intersection sight distance; β = a vector of parameters to be estimated that quantify the relationships between the explanatory variables and λi; εi = a disturbance term, where EXP(εi) is gamma-distributed with a mean equal to one and variance equal to αi. In the NB-2 model, the variance in the expected number of crashes (quantifying the unexplained diversity in the same of observations) is then written as λi + αiλi2. The appropriate model was determined after exploratory data analysis, according to the procedure outlined in Hauer (2015). In addition, the model specifications considered several key aspects of this research context:  ISD is very likely associated with the expected number of target crashes in a non-linear fashion. The sensitivity of the expected number of target crashes to changes in ISD is expected to be highest when ISD is shorter, and decrease as ISD increases (e.g., the safety benefit of increasing ISD from 300 to 600 feet is expected to be substantially larger than the safety benefit of increasing ISD from 1000 to 1300 feet). If this is true, an “inverse ISD” specification is likely appropriate.  The impact of ISD on target crashes and target fatal and injury crashes is expected to vary with speed (e.g., the safety benefit of increasing ISD from 300 to 600 feet is expected to be substantially larger on high speed roads compared to low speed roads). A variable that captures speed and ISD interactions is needed to capture this effect. With operating speeds not available, speed limit could serve as a surrogate for operating speed.

NCHRP 17-59 58  The impact of ISD on target crashes and target fatal and injury crashes is also expected to vary with major and minor road traffic volumes. Variables that capture volume and ISD interactions are needed to capture these effects. The main effects of ISD as well as ISD, speed, and traffic volume interactions were the primary variables of interest in the matrix of explanatory variables, Xi. However, a number of other traffic and geometric variables were included to decrease unexplained variation in expected crash frequency and to try and minimize omitted variable bias. Omitted variable bias would involve over- or under- estimating the safety effects of ISD due to other variables that influence crash frequency and are correlated with ISD, but are excluded from the model. However, the breadth of the model specifications had to be balanced with available data. Statistical significance guided model specification, but sound underlying theory and parameter stability throughout model testing informed specifications more than achieving a p-value of 0.05 or less for all parameter estimates.