Determining Guidelines for Ramp and Interchange Spacing (2011)

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.

Chapter 3 Research Activities and Findings

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-1 Chapter 3 RESEARCH ACTIVITIES AND FINDINGS This chapter presents the findings of the Operations and Safety Work Plans. The panel approved these work plans during the February 2009 Panel Meeting in Washington DC. The NCHRP Project 3-88 research team (3-88 project team) spent the balance of the project executing these plans. These work plans identified and evaluated critical factors associated with varying ramp and interchange spacing dimensions. The work plan execution has also assisted the 3-88 project team with developing the Guidelines that are summarized in Chapter 4. Traffic operations have long been considered when planning and designing freeway facilities and associated ramps and interchanges. Interchanges and ramps should be located with sufficient space to allow exiting and entering maneuvers to occur with few impacts to speed and capacity of the mainline freeway. The 2004 AASHTO Green Book provides ramp and interchange spacing guidance that is based on limited, decades-old operational and design considerations. The 2000 HCM provides procedures for calculating the level of service at ramp junctions, weaving segments, and basic freeway segments. The basic freeway segment procedure identifies a reduction in mainline freeway speed as the density of interchanges on a freeway increases. However, the HCM provides no guidance regarding the maximum or desirable spacing of ramps and interchanges; it simply acknowledges that spacing has an impact on traffic operations. The 3-88 project team analyzed interchange spacing considerations by reviewing data from NCHRP Project 3-92: Production of the 2010 Highway Capacity Manual. NCHRP Project 3-92 proposes to update the density factors that are used to predict speed on a basic freeway segment. The NCHRP Project 3-92 researchers found ramp density to have a greater impact on freeway speeds than interchange density. Thus, ramp spacing became the focus of NCHRP Project 3-88, with interchange spacing remaining a key element. The 3-88 project team analyzed the impact of ramp spacing on mainline freeway speed with microscopic simulation models of selected ramp combinations, a crash dataset created for the project, sensitivity tests of existing HCM analysis procedures, and new analysis of data sets from two other NCHRP projects. The 3-88 project team calibrated simulation models with field data to assess freeway operational performance with various freeway and ramp volumes while considering a range of ramp spacing values. Under a variety of traffic volumes, the 3-88 project team modeled different spacings of entry-entry and entry-exit ramp combinations and measured the impact on mainline freeway speed. The 3-88 project team created a crash database specifically for this project to assess the impact of ramp and interchange spacing on crash frequency. The high-quality dataset allowed a large range of potential safety-influencing features to be considered and included in the model.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-2 3.1 ANALYSIS OF OTHER NCHRP DATA SETS The 3-88 project team reviewed and analyzed data from three other NCHRP projects. Field data collection was a relatively limited part of NCHRP Project 3-88, and field-calibrated simulation models played a large role in the traffic operations work plan. Using data from other projects created much broader and diverse datasets without incurring the significant costs associated with field data collection. These datasets provided the 3-88 project team with additional means of analyzing the effects ramp and interchange spacing have on freeway operation, specifically speed. Reviewing these datasets also increased the 3-88 project team’s understanding of the outcomes, findings, and limitations of previous spacing-related operational research. The three datasets utilized were from NCHRP Project 3-37 Capacity of Ramp-Freeway Junctions, NCHRP Project 3-75 Analysis of Freeway Weaving Sections, and NCHRP Project 3-92 Production of 2010 Highway Capacity Manual. 3.1.1 NCHRP Project 3-37 Capacity of Ramp-Freeway Junctions NCHRP Project 3-37 was conducted in the early 1990s by Polytechnic University (Brooklyn, NY), with the final report issued in March 1994 (76) The purpose of NCHRP Project 3-37 was to create new analytical models for analyzing ramp-freeway junctions for the 1994 HCM. At the time, the existing models for ramp-freeway junctions were based on limited datasets collected in the early 1960s. 3.1.1.1 Project 3-37 Background and Key Findings NCHRP Project 3-37 collected data at 42 single-lane onramps (“merge locations”), 16 single-lane offramps (“diverge locations”) and 10 special sites with features such as double-lane ramps, system interchanges, or metered ramps. These sites were in 15 cities in 10 states. At each site most data, including speed and volume, was collected by the NCHRP Project 3-37 researchers with video cameras. Five cameras were used at each site and spaced 500 feet apart to provide 2,000 feet of coverage. At merge sites, coverage began 500 feet upstream of the ramp-freeway junction and ended 1,500 feet downstream of the junction. At diverge sites, coverage began 1,500 feet upstream of the ramp-freeway junction and ended 500 feet downstream of the junction. For both merge and diverge sites, the ramp-freeway junction was defined as “the point at which the edges of the ramp lane and the freeway lane meet.” This is often referred to as the painted gore, although NCHRP Project 3-37 did not use that term. Data at each site was collected in 15-minute bins, resulting in a total of 341 15-minute bins of data. Speeds reported in the datasets were from the station downstream of a merge or upstream of a diverge that reported the lowest average speed throughout the data collection period. NCHRP Project 3-37 defined the ramp influence area, or the portion of the freeway in which turbulence due to a ramp exists. The ramp influence area

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-3 includes only the right two lanes plus the acceleration lane or deceleration lane associated with the ramp, and extends 1,500 feet upstream of an offramp and 1,500 feet downstream of an onramp. The choice of 1,500 feet is somewhat arbitrary, as the NCHRP Project 3-37 researchers did not collect data further than 1,500 feet away from a ramp. Because the influence area only contains two lanes, an important step in the methodology developed in NCHRP Project 3-37 is the prediction of the volumes in lanes 1 and 2 (V12). The NCHRP Project 3-37 researchers developed several equations to predict V12 based on different geometric configurations. On six-lane freeways (three lanes each direction), some of these equations, summarized in Exhibit 3-1 below, consider distance to and volumes on adjacent ramps. When adjacent ramps were found to impact V12, there was generally a direct relationship with ramp volume and an inverse relationship with distance to the ramp. Adjacent ramps were not found to have an effect on V12 on eight-lane freeways, and on four-lane freeways there is no need to predict V12 as it is simply the total volume of the freeway. Onramp Offramp Upstream on No impact Increases V Upstream off 12 Increases V No impact 12 Downstream on No impact No impact Downstream off Increases V Increases V12 Exhibit 3-1 Impacts of Adjacent Ramps on V 12 12 on Six-Lane Freeways (76) Once V12 3.1.1.2 NCHRP Project 3-37 Data Analysis for NCHRP Project 3-88 is determined, the methodology no longer considers adjacent ramps in the remaining steps to determine ramp-freeway junction level of service. The primary objective of the traffic operations research for NCHRP Project 3-88 was to examine the impacts of ramp and interchange spacing on mainline freeway speed. The 3-88 project team reviewed the NCHRP Project 3-37 datasets and determined that they included sufficient data to analyze the impacts of ramp spacing on speed, but not to analyze the impacts of interchange spacing on speed. The 3-88 project team analyzed two sets of data – one from the Project NCHRP 3-37 merge sites and one from the NCHRP Project 3-37 diverge sites. NCHRP Project 3-37 “special” sites (multilane ramps, ramp meters, etc.) were excluded. The 3-88 project team chose to keep the merge and diverge datasets separate because of the way in which the data had originally been collected. Since NCHRP Project 3-37 was focused on single ramps, speeds were collected immediately downstream of merging ramps and immediately upstream of diverging ramps. Combining the datasets and

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-4 analyzing speeds between ramps would have created an inconsistency with where the speeds were collected along the freeway. For each site, the NCHRP Project 3-37 datasets provided the distance from the study ramp to the nearest downstream ramp and upstream ramp. For this project, the 3-88 project team used the distance to the downstream ramp as the “distance between ramps” for merge sites and the distance to the upstream ramp as the “distance between ramps” for the diverge sites. This choice was made by the 3-88 project team for two reasons. First, it places the speed collection points between the ramps (or within the distance being measured). Second, if full, conventional interchange forms are at the sites, it places the speed collection points between interchanges rather than within them. Generally, the pairing of downstream ramps with merge sites and upstream ramps with diverge sites created EN-EX combinations. When it did not, the 3-88 project team excluded the site from further analysis. There were not enough EN-EN, EX-EX, or EX-EN combinations from which meaningful conclusions could have been drawn to apply to NCHRP Project 3-88. For all analysis, the 3-88 project team chose distance between ramps to be the independent variable and mainline freeway speed to be the dependent variable. Many variables are known to affect freeway speed, and the 3-88 project team made an effort to identify impacts on speed that may have been a reflection of these variables rather than ramp spacing. First, for both the merge and diverge datasets, data was segregated by freeway level of service (LOS). LOS was determined by computing density from the volume and speed of the freeway as if it were a basic segment. For each LOS, data in 15- minute bins was plotted onto a graph. Three versions of each graph (with the same data points) were then created. One indicated whether each point represented stable or unstable flow, a second indicated the number of lanes on the freeway, and a third indicated different ranges of ramp volumes. In many cases, volume was only available from one ramp (the one that had been studied in NCHRP Project 3-37), so the volume of the second ramp was not considered. All of these plots are included in Appendix B The 3-88 project team observed few trends in the data that are applicable to NCHRP Project 3-88. Freeways operating at LOS E experienced a wide range of speeds between ramps regardless of spacing, and freeways operating at LOS F experienced lower speeds than other levels of service. Most of the eight-lane freeways experienced slower speeds between the ramps than the four- or six-lane freeways, but the small sample of eight-lane freeways makes it difficult to draw conclusions for use in NCHRP Project 3-88. Ramp volumes (entering volumes from the NCHRP Project 3-37 merge sites and exiting volumes from the NCHRP Project 3-37 diverge sites) had no apparent impact on speeds. Most notable was the lack of any apparent relationship between ramp spacing and freeway speeds between the ramps. This may be due to the range

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-5 of ramp spacings in the dataset. NCHRP Project 3-37 does not seem to have prioritized close ramp spacing when selected data collection sites, which is not surprising since the purpose of the project was to analyze single ramps. Of the 40 sites analyzed by the 3-88 project team, 27 had ramp spacings between 2,000 and 5,280 feet, 11 had ramp spacings of over 5,280 feet, and only two had spacings under 2,000 feet. The 3-88 project team considered including lane width, shoulder width, and speed limit in the analysis but was unable to do so. Lane width and shoulder width were not in the dataset. An undefined variable in the dataset believed to be speed limit or free-flow speed had the same value at nearly every site. 3.1.1.3 Summary of Analysis of NCHRP Project 3-37 Data The 3-88 project team considered data from NCHRP Project 3-37, which studied operations in ramp-freeway junction areas, to investigate what impact, if any, ramp spacing has on mainline freeway speed. The 3-88 project team only analyzed EN-EX combinations with single-lane ramps, as these comprised a majority of the NCHRP Project 3-37 dataset. Most of the sites had ramp spacings that were several times larger than the current AASHTO minimum of 1,600 feet, with one site having a ramp spacing of over 17,000 feet. Within this range of spacings, there was no apparent impact of ramp spacing on freeway speeds. These findings suggest that if ramp spacing has an impact on freeway speeds, it is only when ramps are spaced closer than most of data from NCHRP Project 3-37. Future analysis should focus on this closer range of spacing, perhaps with spacings of 3,000 feet or less. 3.1.2 NCHRP Project 3-75: Analysis of Freeway Weaving Sections NCHRP Project 3-75 was conducted by Polytechnic University in partnership with KAI, with the final report issued in January 2008 (77). The purpose of NCHRP Project 3-75 was to “calibrate new and/or updated models for prediction of performance in freeway weaving sections” for use in the upcoming 2010 HCM. Draft chapters for the 2010 HCM were produced as part of this effort as well (78, 79) 3.1.2.1 NCHRP Project 3-75 Background and Key Findings NCHRP Project 3-75 collected data at 14 locations in six different states. Most sites were not between the loop ramps at a cloverleaf interchange, but rather between diagonal ramps at different interchanges. The sites were a mixture of Type A, B, and C weaving. One site was on a collector-distributer roadway, and another was at a two-sided weaving section (one ramp was on the left side of the freeway). The 3-88 project team did not analyze any of the data from these two sites, as they both contain conditions that are outside the scope of NCHRP Project 3-88. The NCHRP Project 3-75 researchers collected data at most sites using aerial photographs taken from a fixed-wing aircraft. Data was collected at these

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-6 sites for two hours. The NCHRP 3-75 researchers collected data differently at several other sites. Researchers from the Next Generation Simulation (NGSIM) project provided reduced data for two sites, a state DOT provided video data for one site, and the NCHRP Project 3-75 researchers experimented with a ground-based data collection system at one site. Weaving sections, by definition, have an auxiliary lane between the entrance ramp and the exit ramp; all of the data collection sites for NCHRP Project 3- 75 have an auxiliary lane. In this respect, NCHRP Project 3-75 fundamentally differs from NCHRP Project 3-88. The weaving procedures developed by the NCHRP Project 3-75 researchers are largely new and replaced models that are decades old. The weaving methodology of the 2000 HCM is based upon the weaving methodology of the 1985 HCM with only minor changes. The 1985 methodology itself is based on data collected from 1963 and 1983. The NCHRP Project 3-75 researchers felt that the age of this data and the methodologies developed from it may limit its effectiveness in analyzing weaving sections today, and elected to concentrate their efforts on new data collected and new methodologies developed as part of NCHRP Project 3-75. Major differences between the weaving procedures of the 2010 HCM and those of the 2000 HCM include eliminating different types of weaving (A, B, and C) and eliminating the 2,500 foot maximum weaving length. In place of the different weaving types, several calculations involving lane changing are performed. Interchange density, which has never previously been used in HCM weaving procedures, is a term in these lane changing calculations. In place of the 2,500-foot limit on the length of a weaving section, an equation is used. The equation requires two variables as inputs: the ratio of the weaving volume to total volume at the site, and the number of lanes from which a weaving maneuver may be made by changing only one or zero lanes. In most situations, this equation allows for weaving analysis to be applied to sections longer than 2,500 feet. The 2010 HCM also defines the length of a weaving section differently than previous editions. The 2000 HCM defines the length of a weaving section as the distance from a point on the entry gore where the right-most edge of the freeway pavement is two feet from the left-most edge of the ramp pavement to a point on the exit gore where these edges are 12 feet apart. The NCHRP 3-75 researchers believed this definition was based on typical loop ramp designs from decades ago, and wanted to replace it because most weaving sections on modern freeways do not involve loop ramps. The NCHRP Project 3-75 researchers considered four definitions of weaving length:

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-7 • Short length (LS) – the distance between the end points of barrier markings (such as solid stripes) that prohibit or discourage lane- changing; • Base length (LB) – the distance between the points where the left edge of the ramp travel lane(s) and the right edge of the freeway travel lanes meet; • Long length (LL) – the distance between physical barriers in the merge and diverge gore areas; and, • Average length – the average of the short length and base length. The first three of these definitions are depicted below in Exhibit 3-2. Exhibit 3-2 Possible Definitions of Weaving Segment Length. NCHRP Project 3-75 and the 2010 HCM chose LS and NCHRP Project 3-88 chose L B Statistical analysis performed by the NCHRP Project 3-75 researchers ultimately found that the short length (LS 3.1.2.2 NCHRP Project 3-75 Data Analysis for NCHRP Project 3-88 ) best fit the data that was collected. This was somewhat of a surprise to the NCHRP Project 3-75 researchers; videos recorded for NCHRP Project 3-75 showed barrier markings to not be well-observed in the field. However, the short length was ultimately used in all procedures developed in NCHRP Project 3-75. In cases where there is no barrier marking beyond the painted gore, the short length is simply equal to the base length. The 3-88 team examined the dataset from NCHRP Project 3-75 and determined that it was suitable for investigating the impact of ramp spacing on mainline freeway speed. The 3-88 project team analyzed the NCHRP Project 3-75 data in a manner similar to the NCHRP Project 3-37 data. The 3-88 project team aggregated the 5-minute bins of NCHRP Project 3-75 data into 15-minute bins, and then computed the LOS of the basic freeway segment upstream of the weaving section. The 3-88 project team planned on segregating the data by freeway LOS, but this was ultimately not done because nearly all time periods had LOS C conditions on the freeway. The 3-88 project team then plotted all of the NCHRP Project 3-75 data onto a single graph, with ramp spacing as the independent variable and speed of all vehicles in the weaving section as the dependent variable. The base length

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-8 was used to define ramp spacing when creating these graphs for consistency with other elements of NCHRP Project 3-88. The initial plot of ramp spacing versus the speed of all vehicles in the weaving section, shown in Exhibit 3-3, did not reveal any relationship between the two variables. A subsequent graph that only considered the speed of vehicles remaining on the freeway through the entire segment (i.e., not entering or exiting) also did not reveal any trends. Different versions of these graphs were then created to explore what impact other variables might have on the results. One graph created by the 3-88 project team indicated the number of upstream lanes at each site, and other graphs indicated ranges of onramp volumes, ranges of offramp volumes, ranges of total ramp volumes, interchange densities, 2000 HCM weaving type, and free-flow speed (FFS) at the sites. These graphs can be seen in Appendix A. None of these variables appear to have an impact on speed. 0 10 20 30 40 50 60 70 0 500 1000 1500 2000 2500 3000 Ramp Spacing (Base Length), feet Sp ee d B et w ee n R am ps fo r A ll Ve hi cl es , m ph Exhibit 3-3 Ramp Spacing (weaving section length) Versus Speed at NCHRP 3-75 Data Collection Sites (15-minute bins). The presence of FFS in the dataset also allowed the 3-88 project team to investigate the reduction in speed in the weaving section. In theory, such an analysis minimizes the impact of freeway conditions themselves, and instead highlights the impact of closely spaced ramps on freeway conditions. The 3- 88 project team prepared a graph of ramp spacing versus the reduction from FFS, shown in Exhibit 3-4. Once again, multiple versions of this graph were prepared, with indications for the number of upstream lanes, the range of ramp volumes, etc. These graphs can be seen in Appendix B. Ultimately, none of these graphs revealed any relationship between ramp spacing and a reduction in freeway speed.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-9 0 10 20 30 40 50 0 500 1000 1500 2000 2500 3000 Ramp Spacing (Base Length), feet Sp ee d R ed uc tio n B et w ee n R am ps fo r A ll Ve hi cl es , m ph Exhibit 3-4 Ramp Spacing (weaving section length) Versus Reduction in Speed at NCHRP Project 3-75 Data Collection Sites (15-minute bins). 3.1.2.3 Summary of Analysis of Project 3-75 Data Data from NCHRP Project 3-75, which studied operations in weaving segments, was used to investigate what impact, if any, ramp spacing has on mainline freeway speed. Most data was collected between diagonal ramps serving different interchanges, not between loop ramps at the same interchange. Although the NCHRP Project 3-75 researchers ultimately incorporated ramp spacing (i.e., the length of the weaving section) into weaving analysis procedures they developed, the 3-88 project team did not feel that the relationship between ramp spacing and speed was strong enough to use in design-based guidance and recommendations. 3.1.3 NCHRP Project 3-92: Production of the 2010 Highway Capacity Manual NCHRP Project 3-92 is being led by KAI. The scope of NCHRP Project 3- 92 includes producing the next edition of the HCM and conducting miscellaneous research projects that are needed to supplement analysis procedures in some chapters. One research project conducted by Polytechnic University was to recalibrate the freeway speed-flow curves that are in the 2000 HCM (5). Interchange density is one factor that affects FFS in the methodology of the 2000 HCM, and the NCHRP Project 3-92 researchers reevaluated its impact. The 3-88 project team speculated it may be able to incorporate any new findings related to the impact of interchange density on FFS into NCHRP Project 3-88.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-10 Ultimately, the NCHRP Project 3-92 researchers chose to remove interchange density from the FFS prediction model and replace it with ramp density. Within the NCHRP Project 3-92 dataset, the relationship between FFS and interchange density was counterintuitive: as interchange density increased, FFS also increased. The NCHRP Project 3-92 researchers found that the relationship between FFS and ramp density, while weak, was intuitive: as ramp density increased, FFS decreased. The 3-88 project team noted that NCHRP Project 3-92’s focus on ramp spacing versus interchange spacing may be consistent with the 3-88 project team’s perspective that emphasizes ramp spacing values over interchange spacing. 3.1.4 Summary of Analysis of Previous NCHRP Datasets The 3-88 project team reviewed datasets from three previous NCHRP projects that were related to ramp and interchange spacing: NCHRP Project 3-37 Capacity of Ramp-Freeway Junctions, NCHRP Project 3-75 Analysis of Freeway Weaving Sections, and NCHRP Project 3-92 Production of 2010 Highway Capacity Manual. Overall, the 3-88 project team felt that, in these datasets, the relationship between ramp spacing and freeway speed was not strong enough to be the basis of any design guidance. 3.2 HCM METHODOLOGIES Federal and state design guidance for ramp spacing values generally provide a single minimum recommended ramp spacing dimension for each possible ramp combination (EN-EX, EN-EN, etc). Little guidance is available to designers with regard to how minimum ramp spacing needs may vary under different traffic-volume scenarios. The 3-88 project team explored applying HCM principles that incorporate traffic volumes to consider their influence on ramp spacing values. The HCM provides two sets of methodologies for analyzing the impacts of ramps on a freeway’s operation. The impact of an exit or entrance ramp is analyzed with the HCM’s ramp-freeway junction procedure. The impact of an entrance ramp followed closely by an exit ramp, with an auxiliary lane between the ramps, is analyzed with the HCM’s weaving procedures. When no auxiliary lane exists, or when a different combination of ramps is present (such as an entry followed by an entry), the HCM’s ramp-freeway junction procedures are used to analyze each ramp-freeway junction separately, with the distance between the ramps being an input to the procedure in some cases. The 3-88 project team investigated weaving and ramp-freeway junction procedures to see if, with known traffic volumes and a desired LOS, planning-level spacing guidance can be developed. The 3-88 project team found that, for most cases, such guidelines could not be developed due to the

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-11 complexity of the methodologies or lack of spacing terms in the procedures. However, some findings, with limited applicability to these guidelines, are presented below. The 3-88 project team investigated ramp-freeway junction guidelines with the methodology of the 2000 HCM, which will be virtually unchanged in the 2010 HCM. Weaving guidelines were investigated using the draft procedures of the 2010 HCM, which has an entirely new weaving procedure. The weaving procedures were found to be too complex to form the basis of any simple, conceptual ramp spacing dimension guidelines. 3.2.1 Ramp-Freeway Junctions The HCM provides a procedure for analyzing ramp-freeway junctions on two-, three-, and four-lane freeways. The procedure determines the LOS for the right two lanes of the freeway at a single merging or diverging ramp. On three- and four-lane freeways, the procedure includes a step that calculates the volume in the right two lanes given the freeway’s directional flow. When analyzing an entry ramp on a three-lane freeway, the calculation of the volume in the freeway’s right two lanes (and ultimately the LOS of the ramp- freeway junction) takes into account the distance to the next exit ramp downstream. The four-lane, ramp-freeway junction procedure does not take adjacent ramps into account, primarily due to the small amount of data collected from four-lane freeways at the time the HCM methodology was developed. Thus, ramp spacing plays no role in the analysis for ramp-freeway junctions on two- or four-lane freeways, but it does on three-lane freeways. This is true for both merging ramp and diverging ramp procedures. Given the inconsistency in the two procedures and the limitation of having no ramp spacing value correlation, the 3-88 project team concluded there was limited value in pursuing HCM applications for two- or four-lane freeways. Using the HCM procedures for a merge ramp-freeway junction on a three- lane freeway, the 3-88 project team solved the equations for the term representing distance to an adjacent downstream exit ramp. 3.2.1.1 Calculations The calculation of minimum ramp spacing to achieve a desired LOS began with Equation 25-5 of the 2000 HCM (Equation 3-1 below), which determines the density of a merge influence area. R R 12 AD =5.475+0.00734V +0.0078V -0.00627L (Equation 3-1) where DR = density of merge influence area (pc/mi/ln) VR = on-ramp peak 15 minute flow rate (pc/h) V12 L = flow rate entering ramp influence area (pc/h) A = length of acceleration/deceleration lane (ft)

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-12 The flow rate in Lanes 1 and 2 of a freeway immediately upstream of a merge (V12 12 F FMV =V P× ) is given by: (Equation 3-2) where VF P = freeway peak 15 minute flow rate (pc/h) FM remaining in lanes 1 and 2 = proportion of approaching freeway flow For a six-lane freeway with three lanes in each direction, PFM FM D downP =0.5487+0.2628V /L is calculated using the following equation: (Equation 3-3) where VD immediately upstream of merge = demand flow rate on adjacent downstream Ldown = distance to adjacent downstream ramp (ft) From Equation 1, 12 5.475 0.00734 0.00627 0.0078 R R AD V LV − − += (Equation 3-4) From Equations 3-2 and 3-4, 12 R R FM F F V D -5.475-0.00734V +0.00627P = = V 0.0078V FM D downP =0.5487+0.2628V /L D down FM 0.2628VL = P -0.5487 D F down R R A F 0.00205V VL = (D -5.475-0.00734V +0.00627L )-0.00428V (Equation 3-5) At this point it is necessary to choose a desired density to determine minimum “acceptable” ramp spacing. The 3-88 project team selected three densities for analysis – the maximum densities under LOS C, D, and E. For example, LOS C corresponds to a density of 28 to 35 pc/mi/ln. To calculate

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-13 the minimum distance to the downstream ramp, DR Thus, = 35 pc/mi/ln is substituted into equation 3-5. The 3-88 project team applied the upper value of the density range for each LOS grade. D F down min, C R A F 0.00205V VL = (29.525-0.00734V +0.00627L )-0.00428V Equation 3-6) where Ldown min,C ramp to maintain a LOS “C”(ft) = minimum distance to adjacent downstream Equation 3-6 contains a term for each relevant volume (both ramps and the freeway) as well as the length of the acceleration lane for the merging ramp. Assuming an acceleration lane length, the 3-88 project team created a chart that can be used to see if approximate entry-exit ramp spacing values will create a design that achieves LOS C. Similar charts were also developed for LOS D and E. For all charts the 3-88 project team assumed the following inputs: • Peak-hour factor of 0.92 • Passenger-car equivalent for trucks of 1.5 • Driver population factor of 1.0 • Acceleration lane length of 600 feet • For the freeway: • 60 mph free-flow speed • 10% trucks • 0% RVs • For the ramps: • 5% trucks • 0% RVs 3.2.1.2 Outcome Charts developed by the 3-88 project team are shown in Exhibits 3-5 through 3-7.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-14 Exhibit 3-5 Minimum Ramp Spacing to Achieve LOS C on a Three-Lane Freeway Exhibit 3-6 Minimum Ramp Spacing to Achieve LOS D on a Three-Lane Freeway

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-15 Exhibit 3-7 Minimum Ramp Spacing to Achieve LOS E on a Three-Lane Freeway 3.2.1.3 Applying the Charts The charts shown above may be used in the initial planning stages of a project to see if conceptual designs are feasible from a traffic operations perspective. For example, if an agency desires LOS D for its facility, the user would apply the LOS D chart. A user of the chart should begin by finding the freeway volume being studied on the x axis. The user should then find the set of curves associated with the volume on the entry ramp. In the charts, curves are only provided for entry-ramp volumes of 500 vehicles per hour (vph) and 1,750 vph for ease of presentation. For example: Using the LOS D chart (Exhibit 3-6), with a (three-lane) freeway volume of 3,000 vph and an entrance-ramp volume of 1,750 vph, proposed ramp spacing of 3,500 feet should result in LOS D or better operation on the freeway regardless of the volume on the downstream exit ramp. If the entrance ramp volumes were 1,250 or 800 vph, ramp spacings of 2,500 feet and 1,500 feet, respectively, would provide LOS D operation. For entrance- and exit-ramp volumes not shown in any of the exhibits, interpolation may be used. 3.2.2 Weaving The 2010 HCM will include an entirely new weaving analysis procedure developed from a new dataset. Noteworthy differences in the 2010 weaving procedure in comparison to the 2000 weaving procedure include:

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-16 • A new way of measuring the length of a weaving section. • Replacement of Type A, B, and C weaving with “ramp weaves” and “major weaves.” • A variable maximum weaving length based upon traffic volumes and configuration of the section, instead of a 2,500-foot threshold for all cases. The 3-88 project team attempted to develop simple charts or guidelines (like those developed from the ramp-freeway junction procedures) to approximate the performance of weaving sections. However, the 3-88 project team was unable to do so due to the complexity of the methodologies and the number of variables involved. The methodology has several decision points where the results of one calculation determine which calculation should be performed next. The 3-88 project team concluded that, given the wide range of variables, there would be limited value in exploring the weaving application for planning-level ramp spacing guidance. The 3-88 project team suggests designers should simply conduct a complete HCM weaving analysis early in a project’s development as the ramp configurations are being investigated. 3.3 SIMULATION MODELING The 3-88 project team conducted simulation modeling to assess the impact of ramp spacing on freeway speed. Two ramp combinations were investigated: an entry ramp followed by another entry ramp (EN-EN), and an entry ramp followed by an exit ramp (EN-EX). A base model of each of these ramp combinations was constructed and calibrated with the same ramp spacing as the field data collection sites. The 3-88 project team varied spacing, and then collected point travel speeds at various locations within the model as a means of comparing different ramp spacings. Traffic volumes for mainline through, entering, and exiting movements were varied to test the impact of ramp spacing under differing demands. For the EN-EX model, an auxiliary lane between the entry and exit ramps was later added so that the 3-88 project team could assess potential operational benefits an auxiliary lane. The 3-88 project team’s simulation modeling was conducted using VISSIM because of its ability to realistically reflect freeway merge, diverge, and operational characteristics. VISSIM’s ability to model detailed unique driver behavior characteristics, vehicle fleet types and character allow the model to be calibrated to the site-specific conditions to be used in this evaluation. Additionally, project team members had prior experience with VISSIM and its widespread use would allow the study to be replicated by others. 3.3.1 Data Collection The 3-88 project team collected data at one EN-EX and one EN-EN site in the Phoenix area. The project panel had previously concluded these ramp

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-17 combinations are the highest priority for study as they have the greatest vehicle interaction. EX-EN and EX-EX ramp combinations have fewer merging and weaving maneuvers. At each site, speed and volume data were collected for a 24-hour period with side-mounted digital wave radar on the freeway and with tubes on the ramps. This data is included in Appendix C. Video footage of each site was recorded to provide a visual record of traffic operations at each site, but was not used to capture speeds or volumes. Congestion did not occur at the EN-EX site. The EN-EN site experienced a greater range of volumes and congestion during peak periods. The data collected at both sites was used to calibrate the respective simulation models. The EN-EX combination studied for this project was not a weaving segment, as there was not an auxiliary lane between the ramps. The 3-88 project team and project panel felt that the wealth of research on weaving segments made them less of a research priority than EN-EX combinations without auxiliary lanes. 3.3.2 Site Selection The site-selection process began by establishing the characteristics of what might be considered “ideal criteria and features” of a potential data collection site, and then reviewing aerial photographs of selected freeways in 10 states. From this effort the 3-88 project team identified 16 EN-EX sites and two EN-EN sites as preliminary candidates for collecting data. A list of these sites and the site-selection considerations were provided in Interim Report 1. From this list the 3-88 project team recommended and the panel concurred on the two sites located in the Phoenix area: an entry ramp followed by an exit ramp on southbound SR 51 between Union Hills Road and Bell Road in Phoenix, and an entry ramp followed by an entry ramp on eastbound Loop 202 between Priest Drive and North Center Parkway in Tempe. 3.3.2.1 Selection Criteria The sites were selected for several reasons: • A thorough examination of the sites did not reveal any unusual geometrics or other features that would make them unsuitable for use in research. The sites are of modern design, and do not have low speed ramps, short merge areas, or other constrained features that are common on some older freeways. • Overpasses between the ramps provided a vantage point for data collection at both sites. • Data from the Arizona Department of Transportation’s Freeway Management System (ADOT FMS) was available as a supplement to data collected by the 3-88 project team.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-18 • Both sites are in the same metropolitan area, which minimized data collection costs. Kittelson & Associates, Inc. (KAI) staff visited both data collection sites on March 30, 2009. The 3-88 project team conducted field visits to verify the site’s basic geometric characteristics and identified suitable vantage points for video cameras. The 3-88 project team did not identify any fatal flaws with either site. 3.3.2.2 Crash History The 3-88 project team obtained crash data and average annual daily traffic volumes (AADT) from ADOT as part of the review of the two traffic operations, data-collection sites. Crash data covered the three-year period beginning January 1, 2006, and ending December 31, 2008. The 3-88 project team obtained average crash rates and average injury crash rates for the freeway on which each site is located from the Maricopa Association of Governments (MAG). This data was used to evaluate the safety performance of the two Phoenix sites. At both sites, the study segment was defined as beginning 0.3 miles upstream of the first ramp’s painted gore and ending 0.3 miles downstream of the second ramp’s painted gore. The 0.3 mile distance is consistent with the methodology of FHWA’s Interchange Safety Analysis Tool (ISAT). At both sites, this segment definition resulted in a segment of 1.0 miles in length. The southbound AADT at the EN-EX site is approximately 42,000 vehicles. This is the lowest-volume segment of SR 51, as it is near the northern end of the freeway and on the edge of the urbanized area. For comparison, the AADT at the EN-EN site serves approximately 101,000 vehicles in the eastbound direction. This is one of the highest-volume segments of Loop 202, as it is near downtown Phoenix, downtown Tempe, and Sky Harbor Airport. The 3-88 project team calculated crash rates, in crashes per million vehicle miles of travel, for the study segments using the raw crash data and AADT provided by ADOT. Rates were calculated for all crashes and for injury crashes only. The injury crashes include possible injuries, non-incapacitating injuries, and incapacitating injuries. The rates are presented in Exhibit 3-8: EN-EX (SR 51) EN-EN (Loop 202) Crash Rate for Segment 0.54 1.71 Average Crash Rate for Entire Facility from MAG 1.58 1.50 Injury Crash Rate for Segment 0.15 0.50 Average Injury Crash Rate of Entire Facility from MAG 0.67 0.62 Exhibit 3-8 Crash Rates on Traffic Operations Study Segments (crashes/million vehicle miles of travel)

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-19 The crash rate at the EN-EX site is well below the facility average, while the crash rate at the EN-EN site is slightly higher than the facility average. The crash rate at the EN-EN site is higher than the facility average, which may correspond to the high traffic volumes at the site. This is not necessarily indicative of a crash problem at the study site. In general, the relationship between crashes and traffic volume is not linear. The distribution of crashes by type at each site is provided in Exhibit 3-9. Forty-eight percent of crashes at the EN-EX site (SR 51) were single-vehicle crashes. Most of these single-vehicle crashes were collisions with the median barrier. These crashes were distributed throughout the segment. For at least the first 18 months of the crash data period, SR 51 had a dirt median with a cable barrier. By April 2009, (four months after the end of the three-year crash data period), ADOT had added a concrete barrier and HOV lanes to the median. Seventy-one percent of collisions at the EN-EN site (Loop 202) were rear- end collisions. Many of the rear-end collisions occurred during the late afternoon, when volumes become high and stop-and-go traffic frequently occurs. EN-EN (SR 51) EN-EN (Loop 202) Single Vehicle 48% 11% Sideswipe 28% 15% Rear End 20% 71% Backing 0% 1% Other 4% 3% Exhibit 3-9 Crash Types on Traffic Operations Study Segments 3.3.3 Data collection process KAI staff engaged teaming partner Traffic Research & Analysis, Inc. (TRA) to determine data collection logistics and develop a detailed scope of work and budget. The 3-88 project team initially assumed that pneumatic tubes and video cameras would be used for all data collection, with tubes used to collect speed and volume on the ramps and video cameras used to collect speed and volume on the freeway. TRA suggested collecting data with side- mounted digital wave radar because of their previous success in obtaining accurate and precise data for ADOT. KAI independently investigated other applications of these tools and ultimately concurred with the TRA recommendation to use this equipment to collect the data. 3.3.3.1 Permitting TRA frequently collects freeway data in the Phoenix area, and has permits from ADOT that allow them to mount their radar units along freeways. At the EN-EN site, the video cameras on the overpass had to be set up on the roadway shoulder because there was no sidewalk. TRA obtained permission

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-20 from the City of Tempe, which maintains this roadway, to set up the cameras on the shoulder. 3.3.3.2 Equipment Side-mounted digital wave radar measures speed and traffic volume by lane. The radar units are mounted on roadside objects such as signs, and measure freeway conditions at the mounting location. Exhibit 3-10 shows a close-up view of a radar unit and a roadside-mounted unit collecting data for this project. Exhibit 3-10 Side-Mounted Digital Wave Radar Used by TRA for Data Collection. Left photo: Close-up view of the radar unit. Right photo: Radar unit collecting data at EN-EN site. KAI staff contacted the Texas Transportation Institute (TTI), which had recently evaluated TRA’s side-mounted digital wave radar units and found them to be satisfactory for counting vehicles on a freeway. (81) In comparison to a video counted by multiple people, the radar units had average volume errors of less than 5 percent. Lanes furthest from the mounting location (generally closest to the median, since the units are typically mounted on the right side of the road) had the highest errors – up to 10 percent at one test location. TTI found little difference in error count between high- and low-volume periods. Based on this favorable review of the side-mounted radar, the 3-88 project team elected to use it for measuring speeds and volumes on the mainline freeway. Using side-mounted radar significantly reduced data collection costs because it eliminated the need for a person to manually count vehicles and compute speeds by watching a video recording. Video was recorded and used to qualitatively observe operational characteristics, such as lane changing, that could not be measured by the radar. 3.3.3.3 Scope of Data Collection The 3-88 project team collected four hours of data, capturing both off-peak and peak conditions, at each site. On the freeway itself, speed and volume data was collected prior to the first ramp, between the ramps, and after the second ramp, with a video camera at each collection point.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-21 Radar units eliminated the need for cameras at each of these three locations. With radar, the primary purpose of the video changed from collecting data at a specific location to providing a visual record of vehicular activity on the entire study segment. To best accomplish this, two cameras were placed on the overpass in the middle of each segment. One camera looked upstream towards the first ramp, and the other looked downstream towards the second ramp. Radar also enabled speed and volume data to be collected for longer than four hours at each site. Radar units were set up the day before data collection was scheduled to occur, and taken down the day after. Since the units were continuously collecting data, an entire 24-hour day of data was downloaded instead of only four hours. Video was collected for four hours as initially planned, during both off-peak and peak conditions. Data was collected in 15- minute bins. Exhibits 3-11 and 3-12 show the locations of data collection equipment (radar, tubes, and cameras) at the EN-EX and EN-EN sites, respectively. At the EN-EX site on SR 51, three radar units, two tubes, and two video cameras were used. At the EN-EN site on Loop 202, two radar units, two tubes, and two video cameras were used. The data collection plan for the EN-EN site initially called for three radar units: one upstream of the first ramp, one between the ramps, and one downstream from the second ramp. However, the section of Loop 202 downstream from the second ramp is elevated, and there are no signs or other roadside objects on which a radar unit could have been mounted. There is also no vantage point from which a camera could have been set up to record traffic operations. Fortunately, an ADOT FMS detector is located there and the 3-88 project team was able to obtain speed and volume data from the detector for the same time period that all other data at the EN-EN site was collected. 3.3.3.4 Arizona Department of Transportation Freeway Management System (ADOT FMS) The Arizona Department of Transportation operates an extensive Freeway Management System (ADOT FMS) (82). This system collects data, including speeds and volumes per lane, on most segments of most freeways in the Phoenix metropolitan area through the use of loop detector stations. This data is primarily used to monitor traffic conditions in real time, but is also archived and made available to the public on an ftp site. The EN-EX site on SR 51 lies outside of the ADOT FMS coverage area, although a station is located at an interchange approximately one mile downstream of the segment’s end. The 3-88 project team compared average daily volumes from 2008, presented in 15-minute bins, at the downstream station with volumes collected for this project. This comparison confirmed

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-24 that freeway volumes on the day of data collection were typical. The EN-EN site on Loop 202 is well covered by ADOT FMS, with detectors located upstream of the first ramp, between the ramps, and downstream of the second ramp. Data from the ramps themselves is also available. The 3-88 project team used this data to verify that volumes on the day of data collection were typical. 3.3.3.5 As-Built Plans As-built plans for both data collection sites provided the 3-88 project team with data that could not easily or accurately be measured in the field or from aerial photographs. 3.3.4 Data collection plan execution TRA collected data at the EN-EX site on Tuesday, April 14, 2009, and at the EN-EN site on Thursday, April 16, 2009. A KAI team member was present during video data collection (scheduled from 2-6 p.m.) at both sites to observe the process and monitor traffic conditions. At the EN-EX site, data collection was conducted smoothly as scheduled. 3.3.4.1 Issues At the EN-EN site, video data collection was conducted from 3:30-7:30 p.m. instead of 2-6 p.m. due to a crash on the freeway. Just before 2 p.m., a bicyclist riding on the freeway was struck by an automobile near the end of the taper of the first on-ramp. Numerous emergency vehicles responded to this crash, which forced drivers on the first onramp to cross the painted gore to merge, closed the freeway’s right lane for a brief time, and distracted drivers. The last of these vehicles cleared the site around 3:30 p.m., and video recording was started then so that four hours could be captured before nightfall. No data collected prior to 5 p.m. was used for quantitative analysis or modeling due to the potential for lingering operational impacts related to the crash. Low traffic volumes at the end of the video data period allowed for off-peak conditions to be captured, and the 3-88 project team deemed this data adequate for its use. Although several hours of video data from a crash-free period and nearly a full day of speed and volume data were captured at the EN-EN site on the first data collection day, a second day of data collection was scheduled at the site from 2-6 p.m. on Tuesday, May 5, 2009. The benefits of such data would have been limited, as crash-free data was collected during both peak and off- peak periods on the first day. Unfortunately, ADOT had begun a long-term construction project that included closing a lane on Loop 202 by the time the second day of data collection occurred. Data collected on this day is not suitable for analysis or modeling, and was not used.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-25 The long-term nature of the construction project made it infeasible to collect additional data at the Loop 202 site. The 3-88 project team is confident that the data collected on April 16, 2009 is suitable for analysis and modeling purposes. 3.3.4.2 Data collection results At the EN-EX site on SR 51, traffic volumes exceeded 1,150 vehicles per hour per lane during the peak on the segment between the ramps. The highest volume hour was during the AM peak. On-ramp volumes peaked at over 1,200 vehicles per hour, and off-ramp volumes peaked at over 500 vehicles per hour. No congestion was observed at this site. At the EN-EN site on Loop 202, traffic volumes exceeded 1,950 vehicles per hour per lane during the peak on the segment between the ramps. The 3-88 project team observed congestion and decreases in speed. Volumes on the first onramp peaked at over 600 vehicles per hour, and volumes on the second onramp peaked at over 800 vehicles per hour. 3.3.5 Existing Conditions Model Construction The 3-88 project team constructed VISSIM simulation models for the following two test sites: • Entry-entry (EN-EN) site on Route 202 near North Priest Drive and North Center Parkway • Entry-exit (EN-EX) site on Route 51 between East Union Hills Drive and East Bell Road These locations both include four basic highway lanes, single-lane on- and off-ramps and no auxiliary lanes between ramps. At the EN-EN site ramp spacing is 2,180 feet, and at the EN-EX site ramp spacing is 2,100 feet, measured from painted gore to painted gore. The 3-88 project team constructed the VISSIM model using as-built drawings, on-the-ground photography, and video data collection to ensure study-area models accurately reflect roadway plan, profile, and cross section, as shown in Exhibit 3-13. Only study-area ramps and freeway mainlines were modeled. Ramp-terminal operations and metering of traffic flows were beyond the scope of this project and were not included in the models. This represented a worst-case scenario for freeway operations, as ramp meters are intended to improve freeway operation.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-26 Exhibit 3-13 VISSIM Models Built Using Available Data Sources. Left photo: VISSIM model layered over existing high- resolution aerial photography. Right photo: VISSIM model layered over most recent striping plan. The two base models built using existing geometric data and traffic volumes served as a validation tool to ensure that calibration could be conducted based on the field collected data. 3.3.6 Simulation Calibration Good calibration of the simulation model is a key step to developing results in which one can be confident. The 3-88 project team collected a wealth of field data in the form of spot or point speeds from the two study sites in Phoenix. Field data collection is discussed in detail in section 3.1.1. Exhibit 3- 14 shows the data collection points used to report speed for the two VISSIM simulation models. Data collected at these points match the field data collection point, and were used to calibrate the model to the existing point where speed data was collected. Exhibit 3-14 VISSIM data collection points were placed at both gore points (painted tips) and three additional intermediate collection points were evenly spaced between the gore points. EN-EX model pictured above

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-27 The 3-88 project team added operating characteristics to the simulation model to reflect test-site conditions. The following characteristics were field measured and included in the simulation model for calibration: • Weekday AM and PM peak hour traffic volumes (mainline, entering ramp, and exiting ramp). Measured peak-hour traffic volumes were directly entered into VISSIM. For calibration, these exact volumes are generated within the peak hour. In later evaluation of ramp spacing alternatives, traffic volumes were allowed to vary to reflect day-to-day fluctuation in travel demand based upon random seeding. • Operating speed ranges (mainline and ramps) during weekday AM and PM peak hours. Posted speeds along the site study areas were noted, but field-measured, free-flow, average speed ranges were collected and those speed ranges were input into VISSIM to develop reasonable desired speed profiles. • Vehicle fleet mix (percent trucks and buses versus passenger cars). Field-measured fleet mix from each study area was input into VISSIM. • Driver behavior characteristics (such as lane changing, gap acceptance/merging, car following distances) were generally compared between observed field video recordings and VISSIM simulation. The resulting VISSIM speeds were compared to the field measured speeds as an indicator of calibrated driver behavior in addition to animation visual inspection. Thirty different random seed, traffic model runs were averaged to conclude in a single “VISSIM result” to compare to field-measured data to determine model calibration. Field measured data was collected on April 14 and 16 2009, as documented in the previous section of this report. The same approach of averaging 30 random seeds in VISSIM was used for alternative evaluation results presented in the following section. For the purposes of this research effort, point speed on the freeway mainline was the primary calibration and performance measurement. The 3-88 project team set a desired calibration target of +/- 1 mph (1-2%) for three distinct, point-speed measurements when comparing field-measured averages and VISSIM-measured averages across all lanes. Both models were calibrated to these levels. This allowed for sufficient confidence to begin varying ramp spacing and traffic volumes, as described in the following section. 3.3.7 Alternative Ramp Spacing Evaluations The 3-88 project team selected ramp spacings to model based on a combination of design literature including the AASHTO Green Book, established practices, and field ramp spacing. Ramp spacing is defined as painted tip of gore point to painted tip of gore point. This is consistent with

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-28 the HCM 2010 and the of ramp spacing definition documented in the Guidelines.by the 3-88 project team The 3-88 project team created four alternative ramp spacing VISSIM models to test the impacts of ramp spacing, as follows: 1. EN-EN site with ramp spacing of 700 feet 2. EN-EN site with ramp spacing of 2,500 feet 3. EN-EX site with ramp spacing of 1,000 feet 4. EN-EX site with ramp spacing of 2,500 feet These ramp spacings represent reasonable ranges over which the 3-88 project team could investigate the sensitivity of speed to ramp spacing. The 3-88 project team assumed that ramps longer than 2,500 feet would have diminishing influence from upstream/downstream ramps, and ramps closer than 700 or 1,000 feet are closer than typically found on freeway systems. For each of these four ramp spacing models, three mainline volumes entering the segment upstream of the first ramp were analyzed: • 1,250 mainline entering vehicles per hour per lane (vphpl) (LOS B upstream of ramps and LOS C at highest-volume point within the segment studied); • 1,500 mainline entering vehicles per hour per lane (LOS C upstream of ramps and LOS D at highest-volume point within the segment studied); and, • 1,750 mainline entering vehicles per hour per lane (LOS D upstream of ramps and LOS F at highest-volume point within the segment studied). The 3-88 project team selected these traffic volumes with the aid of Exhibit 13-6 of the 2000 HCM to represent different levels of service. For each ramp spacing model and mainline volume variant, a total of nine ramp-volume combinations were evaluated, as shown in Exhibit 3-15. These ramp volumes represent combinations of three distinct types of ramp operation, up to capacity as defined in Exhibit 13-20 of the 2000 HCM (14). By looking at the nine combinations illustrated in Exhibit 3-15, a wide variety of intermediate ramp-volume combinations can then be inferred.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-29 Ramp 1 Volume (vph) 750 1250 1750 R am p 2 V ol u m e (v p h ) 750 1 2 3 1250 4 5 6 1750 7 8 9 Exhibit 3-15 Ramp-Volume Combinations With three various mainline-volume scenarios and nine different ramp- volume scenarios for each mainline-volume scenario, this results in a total of 27 different simulation models per ramp spacing for each of the EN-EN and EN-EX geometries. The EN-EX site in Phoenix where data was collected, as well as the models described above, do not have an auxiliary lane between the ramps. However, the 3-88 project team chose to create a second set of EN-EX models with an auxiliary lane. This allowed the 3-88 project team to investigate the operational influence of adding an auxiliary lane and better compare findings to the wealth of research that has been conducted on weaving sections. Including the auxiliary lane models, a total of 156 models were created and run for this project. 3.3.8 VISSIM Modeling Results Using the calibration settings in the VISSIM model and the alternatives described in the previous section, the 3-88 project team adjusted the VISSIM models to meet the prescribed traffic volumes and ramp spacing for evaluation. For each alternative, the 3-88 project team conducted 30 simulation iterations or runs using a consistent set of random seed variables. These 30 results were then summarized and averaged for each point-speed location within VISSIM. The 3-88 project team collected specific speed information at each of the five data collection points (see Exhibit 3-14), and then averaged the speeds across 30 runs. This resulted in a total data set of approximately 200,000 entries. Once the data collection period of the modeling was complete, the 3-88 project team compared reported speeds for each spacing alternative and volume scenario.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-30 3.3.8.1 Measures of Effectiveness The distinct measure of effectiveness for this evaluation is point speed. As shown in Exhibit 3-14, the point speed is measured at five distinct locations between ramp gore points for the EN-EN and EN-EX models, and under all ramp spacing alternatives. This means the measurement locations are closer together when ramp spacing is closer and further apart when ramp spacing is longer. This provides a relative comparison of speeds between each ramp spacing alternative. The 3-88 project team used two evaluation approaches to convey the relative comparison of point speeds under various ramp volumes,: 1) A comparison of the lowest speeds. This evaluation simply compares the lowest speeds occurring within each mainline segment, regardless of the location within the segment. Exhibit 3-16 shows an example of this lowest- speed-reported comparison. The lowest speeds were chosen by the 3-88 project team for comparison in lieu of comparing the average speed within each mainline segment since averaging the speeds would have dampened the actual operational difference between each ramp spacing alternative. For example, in the top left cell, the lowest point speed for a 1000-foot ramp spacing of 64 mph may be measured at the midpoint between gore points, while the lowest point speed for the 2,500-foot spacing (65 mph) was measured at the downstream ramp gore point. This comparison highlights the effect of ramp spacing on the lowest speed between gore points, but also shows absolute speeds that can be used in comparison to free-flow speed. Exhibit 3-16 Example of Comparison of Lowest Speeds Reported 2) The maximum corresponding point-speed difference. This evaluation considers the speed difference between the two ramp spacing alternatives at

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-31 each measurement location. The 3-88 project team determined the speed differential at each of the five measurement points and the maximum speed differential is considered for each ramp-volume combination. The speed measurement points are equivalent in this comparison, regardless of ramp spacing,. Comparing lowest reported speed does not necessarily compare measurements at the same point. For example, if the speed occurring at the downstream gore point in the 1,000-foot model is 35 miles per hour, and in the 2,500-foot model is 50 miles per hour, then the corresponding point- speed difference is 15 miles per hour. Exhibit 3-17 shows an example of the maximum corresponding point-speed differentials for each volume scenario. From the nine data points shown in the exhibit, the 3-88 project team inferred expected trend zones showing the anticipated maximum speed differential at corresponding points under different ramp-loading conditions. Exhibit 3-17 Example of Maximum Corresponding Point Speed Difference 3.3.8.2 Entry – Exit Analysis Results Exhibits 3-18, 3-19 and 3-20 summarize the lowest-reported-speed comparisons and maximum corresponding point-speed differentials for the two EN-EX ramp spacing alternatives, 1,000 feet and 2,500 feet. The lowest reported speeds did not necessarily occur at the same measurement point for the two spacing alternatives. For the EN-EX model the average free-flow speed is 66 mph.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-32 A) B) Exhibit 3-18 EN-EX Mainline Entering Volume 1,250 vphpl, 1,000- ft spacing and 2,500-ft spacing: A) Comparison of Lowest Speed Reported, B) Maximum Corresponding Point-Speed Difference Simulation model results indicate the following: • Ramp spacing does not significantly impact speed (i.e., >2 mph) at low to moderate exit volumes and uncongested conditions. • The greatest ramp spacing effect (speed differential) occurs under congested conditions (i.e., speeds <50 mph). • At the highest exiting (ramp 2) volume (1,750 vph), the lowest speed for the 2,500-foot spacing was 12 mph higher than the lowest speed for the 700-foot spacing. • The largest observed point-speed differential at corresponding points is 15 mph. • Speed differentials at corresponding points are slightly higher then the differential observed for the lowest speeds; however, they do show a consistency relative to ramp volumes.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-33 A) B) Exhibit 3-19 EN-EX Mainline Entering Volume 1,500 vphpl, 1,000- ft spacing and 2,500-ft spacing: A) Comparison of Lowest Speed Reported, B) Maximum Corresponding Point-Speed Difference Simulation model results indicate the following: • Comparing the lowest speeds, ramp spacing does not significantly impact speed (i.e., >2 mph) at low to moderate exit volumes and uncongested conditions; however, the maximum speed differences indicate a significant impact under moderate exiting volumes. • The greatest ramp spacing effect (speed differential) occurs under congested conditions (i.e., speeds <50 mph). • The greatest speed difference between the two spacing alternatives (16 mph) occurred at a low-entry volume and high-exit volume, not when both ramp volumes are high as in the previous data for a mainline entering volume of 1,250 vphpl. This suggests that at this mainline volume, along with high ramp volumes, the system’s level of congestion is controlling speed and the effect of ramp spacing diminishes. • Comparing the data between the 1,500 vphpl and 1,250 vphpl entering mainline-volume scenarios shows an increasing influence of ramp spacing across ramp volumes.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-34 A) B) Exhibit 3-20 EN-EX Mainline Entering Volume 1,750 vphpl, 1,000- ft spacing and 2,500-ft spacing: A) Comparison of Lowest Speed Reported, B) Maximum Corresponding Point-Speed Difference Simulation model results indicate the following: • Comparing the lowest speeds, ramp spacing does not significantly impact speed (i.e., >2 mph) at low exit volumes and uncongested conditions. • The greatest ramp spacing effect (speed differential) occurs under congested conditions (i.e., speeds <50 mph). • As shown in Exhibit 3-20-A, the greatest difference in the speeds between the two spacing alternatives (15 mph) occurred at a low- entry volume and high-exit volume, not when both ramp volumes are high as in data for a mainline entering volume of 1,250 vphpl. This indicates that the system’s level of congestion is controlling speed and the effect of ramp spacing diminishes as both entering and exiting volume increase to their highest levels. • The largest observed point-speed differential at corresponding points is 15 mph. • Speed differentials at corresponding points are higher then the differential observed for the lowest speeds; however, they do show a consistency relative to ramp volumes. • Comparing the data between the 1,750 vphpl, 1,500 vphpl and 1,250 vphpl entering mainline-volume scenarios shows an increasing influence of ramp spacing across ramp volumes.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-35 3.3.8.3 Entry – Entry Analysis Results The 3-88 project team observed several general trends during the modeling and analysis of the EN-EN site. • The lowest observed speeds generally occur at the gore point of the second entry ramp. • The highest corresponding point-speed differentials generally occur at the first gore point or an intermediate measurement point between gore points. • Longer ramp spacing can increase corresponding point speeds by up to 11 mph, but has diminishing impacts as mainline and ramp- traffic volumes increase towards capacity. Exhibits 3-21, 3-22 and 3-23 document the lowest-reported-speed comparisons and maximum corresponding point-speed differentials for the two EN-EN ramp spacing alternatives, 700 feet and 2,500 feet. For the EN- EN model the average free-flow speed is 64 mph. Exhibit 3-21 EN-EN Mainline Entering Volume 1,250 vphpl, 700-ft spacing and 2,500-ft spacing: Comparison of Lowest Speed Reported Simulation model results indicate the following: • The highest observed speed differentials were less then 2 mph for all ramp-volume combinations. As such, ramp spacing does not significantly impact mainline operations for the EN-EN ramp configuration when the mainline entering volume is low (1,250 vphpl). The graph for corresponding speed differentials was omitted for this mainline volume due to it showing no speed differentials greater then 2 mph.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-36 A) B) Exhibit 3-22 EN-EN Mainline Entering Volume 1,500 vphpl, 700-ft spacing and 2,500-ft spacing: A) Comparison of Lowest Speed Reported, B) Maximum Corresponding Point-Speed Difference Simulation model results indicate the following: • In comparing lowest speeds, the speed difference between the two ramp spacing alternatives was less then 2 mph for all ramp- volume combinations, • The largest observed point-speed differential at corresponding points is 3 mph, but only for the highest ramp volumes. • Similar to the previous data for an entering mainline volume of 1,250, ramp spacing does not significantly impact mainline operations for the EN-EN ramp configuration when the mainline entering volume is moderate (1,500 vphpl).

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-37 A) B) Exhibit 3-23 EN-EN Mainline Entering Volume 1,750 vphpl, 700-ft spacing and 2,500-ft spacing: A) Comparison of Lowest Speed Reported, B) Maximum Corresponding Point-Speed Difference Simulation model results indicate the following: • With an entering mainline volume of 1,750 vphpl, ramp spacing has a significant (>2 mph) effect on mainline speed at moderate and high ramp-volume scenarios. • The greatest ramp spacing effect (speed differential) occurs under congested conditions (i.e., speeds <50 mph). • The maximum corresponding point-speed differentials (Exhibit 3- 23-B) indicate a larger ramp spacing affect than the speed differences between the lowest observed speeds (Exhibit 3-23-A). The uncharacteristic trend shown in Exhibit 3-23-B is primarily due to the varying speed profiles of the two ramp spacing alternatives under each ramp-volume scenario. 3.3.8.4 Auxiliary Lane Analysis Results Comparisons of the EN-EX ramp configuration with and without an auxiliary lane between the ramps are shown in Exhibits 3-24 and 3-25. The 3- 88 project team created models for both spacing dimensions previously analyzed: 1,000 feet and 2,500 feet. The data presented in these exhibits represent the increase in mainline speed resulting from adding an auxiliary lane.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-38 A) B) C) Exhibit 3-24 Effect of Auxiliary Lane on Mainline Speed (1,000-ft ramp spacing) A) Mainline Entering Volume = 1,250 vphpl B) Mainline Entering Volume = 1,500 vphpl C) Mainline Entering Volume = 1,750 vphpl Simulation model results for the 1,000-foot ramp spacing indicate the following: • Adding an auxiliary lane results in higher mainline speeds for all mainline and ramp-volume combinations.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-39 • In general, at low exit ramp volume (750 vph), the benefit of adding an auxiliary lane is relatively minor (< 2mph), regardless of the entry ramp volume. • At moderate to high mainline entering and ramp volumes, the increased speeds resulting from an auxiliary lane are significant, reaching as high as 18 mph. A) B) C) Exhibit 3-25 Effect of Auxiliary Lane on Mainline Speed (2,500-ft ramp spacing) A) Mainline Entering Volume = 1,250 vphpl B) Mainline Entering Volume = 1,500 vphpl C) Mainline Entering Volume = 1,750 vphpl

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-40 Simulation model results for the 2,500-foot ramp spacing indicate the following: • Adding an auxiliary lane results in higher mainline speeds for all mainline and ramp-volume combinations. • At low (750 vph) to moderate (1,250) exit-ramp volumes and low mainline entering volume (1,250 vphpl), the benefit of adding an auxiliary lane is relatively minor (< 2mph), regardless of the entry- ramp volume (Exhibit 3-25A). Similarly, at moderate mainline entering volume (1,500 vphpl) and low exit volume, the benefit is minor (Exhibit 3-25-B). • At high mainline and ramp volumes, the increased speeds resulting from an auxiliary lane are significant, reaching as high as 20 mph. • In general, the increased speed provided by an auxiliary lane for the 2,500-foot spacing is less than provided for the 1,000-foot spacing at low to moderate ramp volumes. This makes sense since the longer spacing provides greater length for lane changing without an auxiliary lane present. 3.3.8.5 Summary of results • EN-EN Models • The lowest mainline speeds within this ramp configuration occur at the second onramp. • At low to moderate mainline entering volumes (<1,500 vphpl), ramp spacing generally has little effect on speed within the mainline segment regardless of ramp-volume levels. • At high mainline entering volumes (>1,750 vphpl), ramp spacing has a significant impact on mainline segment speeds across moderate to high ramp volumes. • EN-EX Models Without Auxiliary Lane • In general, the level of exiting volume has the greatest influence on mainline segment speeds. • At low mainline entering volumes (<1,250 vphpl), ramp spacing significantly affects mainline segment speed at high (>1,750 vphpl) exit-ramp volumes. • At moderate and high mainline entering volumes (>1,500 vphpl) ramp spacing significantly affects mainline segment speed at moderate and high exit-ramp volumes. • Auxiliary Lane Models • Adding an auxiliary lane results in increased point speeds when compared to equivalent non-auxiliary lane conditions. The

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-41 benefit of an auxiliary lane is minor at low mainline and exit volumes; however, it becomes significant as traffic volumes increase. • Adding an auxiliary lane on a longer ramp spacing (2,500 ft) generally has less benefit than adding an auxiliary lane to a shorter ramp spacing (1,000 ft). 3.4 SAFETY WORK PLAN Freeway interchanges, by their nature, coincide with increased lane changing, acceleration and deceleration on and near the mainline. Traffic operations are adversely affected and decline with higher interchange ramp densities (i.e., shorter ramp spacing). The effects are captured at both the interchange level (e.g., free-flow speed and capacity decrease as interchange density increases) and at the ramp level (e.g., speed decreases as weaving length decreases) by algorithms in the 2000 HCM. Analogous safety relationships are not as well established. The first addition of the Highway Safety Manual does not include a quantitative safety effect of ramp or interchange spacing. The literature review, summarized in Chapter 2, demonstrated that only two studies directly explored relationships between ramp spacing and safety. The first, by Cirillo, was dated by almost 40 years and did not consider key variables that likely influence the spacing-safety relationship (e.g., ramp volumes, number of though lanes) (70). The second, by Bared et al., used modern analysis techniques, but several limitations were identified (72). Therefore, the NCHRP 3-88 research approach for the safety work plan called for focused research effort to explore the safety effects of ramp and interchange spacing. A study by the Texas Transportation Institute that included a safety assessment of weaving length was published while the NCHRP 3-88 work plan was under way. The 3-88 project team compared the results of the Texas study to the results of this study in Section 3.4.6. The remainder of this chapter describes the safety-related research effort of NCHRP 3-88. 3.4.1 Key Issues Interchange spacing, defined from cross-street centerline to cross-street centerline, is not as meaningful as ramp spacing, defined from painted gore to painted gore, from a safety modeling and analysis standpoint. For a given interchange spacing, freeway segments between the cross streets may have different numbers, types, combinations and spacings of interchange ramps. In addition, cross streets associated with some ramps are difficult to identify for atypical interchange types, and may not be centered between exit and entrance ramps. As a result, the 3-88 project team focused on developing relationships between ramp spacing and safety. The relationships can be aggregated to determine interchange spacing effects for different interchange forms.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-42 Much of the following safety discussion is related to the scenario of an entrance ramp from one cross street followed by an exit ramp to a downstream cross street (EN-EX). This is a common ramp-sequence scenario, and one in which operational analyses are frequently conducted and safety information is frequently needed. The 3-88 project team explored the scenario of two consecutive entrance ramps (EN-EN) from a safety perspective, but with lesser detail than the EN-EX. The following describe some of the challenges of other ramp scenarios: • The scenario of an exit ramp followed by an exit ramp (EX-EX) was uncommon; in the the initial scan of 650 directional miles of freeway the 3-88 project team identified less than 20 such locations. • An exit ramp followed by an entrance ramp (EX-EN) is common within a single interchange; observed spacing dimensions between these ramps did not significantly vary and generally ranged from 2,400-4,400 feet, far above the minimum recommended value of 500 feet identified in the AASHTO Green Book. Data that included interchange ramp traffic and coded ramp locations referenced to a mainline milepoint, required elements for a safety analysis, were most readily available in data files from Washington State and California. The 3-88 project team explored data sets and supplemental data sources from both states in detail. The 3-88 project team discovered discrepancies between electronically coded data in California ramp files obtained through FHWA’s Highway Safety Information System (HSIS) and video data observed through the University of California, Berkeley’s Performance Measurement System (PeMS). Therefore, the 3-88 project team focused its data collection efforts on Washington State only. The 3-88 project team compared the safety findings using the Washington data to recently published ramp spacing/safety findings using Texas data to address potential concerns regarding the transferability of findings using data from Washington alone. The Texas research objectives and strategy was consistent with the goal of the NCHRP 3-88 safety effort: To understand and quantify general accident trends associated with ramp and interchange spacing. The 3-88 project team felt that such an in-depth, comprehensive effort using data from one state was much more likely to provide greater insights into the safety phenomenon of interest than less comprehensive information from a greater number of spatially dispersed states. 3.4.2 Variable Notation and Definitions The following variable notations and definitions are used throughout the remainder of the safety section:

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-43 a. L = segment length defined from physical entrance gore to physical exit gore (miles); b. ln(L) = natural logarithm of the segment length; c. ADT = two-way average daily traffic upstream of the entrance ramp in an EN-EX or EN-EN ramp sequence (veh/day); d. DADT = one-way (directional) average daily traffic upstream of the entrance ramp in an EN-EX or EN-EN ramp sequence (veh/day); e. ln(ADT) = natural logarithm of the ADT; f. ADTEN g. ADT = average daily traffic on the entrance ramp of an EN-EX ramp sequence at the entrance ramp-freeway terminal (veh/day); EN-1 h. ADT = average daily traffic on the first (upstream) entrance ramp of an EN-EN ramp sequence at the entrance ramp-freeway terminal (veh/day); EN-2 i. ln(ADT = average daily traffic on the second (downstream) entrance ramp of an EN-EN ramp sequence at the entrance ramp-freeway terminal (veh/day); EN) = natural logarithm of the ADTEN j. ln(ADT ; EN-1) = natural logarithm of the ADTEN-1 k. ln(ADT ; EN-2) = natural logarithm of the ADTEN-2 l. ADT ; EX m. ln(ADT = average daily traffic on the exit ramp of an EN-EX ramp sequence at the exit ramp-freeway terminal (veh/day); EX n. S = ramp spacing defined from painted entrance gore to painted exit gore (feet); ) = natural logarithm of the ADTEX; o. S-1 p. AuxLn = indicator variable for the presence of an auxiliary lane between an entrance ramp and exit ramp (1 = auxiliary lane present; 0 = no auxiliary lane); = inverse of ramp spacing (1/feet); q. %BarrL = the length of a barrier adjacent to the median shoulder divided by the total length of the segment (unitless decimal); r. %BarrR = the length of a barrier adjacent to the right shoulder divided by the total length of the segment (unitless decimal); s. MainEn = indicator variable for the vertical relationship between the cross street for the entrance ramp and the freeway mainline (1 = mainline over cross street; 0 = mainline under cross street); t. MainEx = indicator variable for the vertical relationship between the cross street for the exit ramp and the freeway mainline (1 = mainline over cross street; 0 = mainline under cross street);

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-44 u. NoLn2 = indicator variable for the number of directional through lanes on the freeway segment (1 = two lanes; 0 = three or four lanes); v. NoLn3 = indicator variable for the number of directional through lanes on the freeway segment (1 = three lanes; 0 = three or four lanes); w. NoLn4 = indicator variable for the number of directional through lanes on the freeway segment (1 = four lanes; 0 = three or four lanes); x. Total = expected number of crashes of all severities and types; y. FplusI = expected number of crashes involving at least one occupant fatality or injury; z. SingV = expected number of crashes involving only one motor vehicle; aa. MultV = expected number of crashes involving more than one motor vehicle; bb. Truck = expected number of crashes involving at least one large truck; cc. Peak = expected number of crashes occurring during defined peak- hour time periods; and, dd. α = overdispersion parameter of the negative binomial regression model. 3.4.3 Data Collection The 3-88 project team collected data from Washington State using several different information sources: interchange diagrams available through Washington Department of Transportation’s (WSDOT’s) Interchange Web Viewer; freeway network maps and aerial photographs available through Google Maps and Google Earth; video logs that are part of WSDOT’s State Route Web; and electronic crash, roadway, ramp and vehicle files provided by FHWA’s HSIS. Google Maps Street View was also used to supplement or verify information collected from WSDOT’s video logs. The NCHRP 3-88 project team used four primary steps in the data collection process: 1. Gather ramp locations and ramp-related features in both directions of freeway travel; 2. Define freeway segments for safety analysis; 3. Collect traffic and geometric data for defined freeway segments; and, 4. Determine crash frequencies and severities on each defined freeway segment. Each of the four steps is summarized in the remainder of this section.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-45 3.4.3.1 Ramp Locations and ramp-related features The 3-88 project team used Google Maps to scan Washington State’s freeway network and locate potential corridors of interest. The 3-88 project team observed and recorded ramp locations and ramp-related features for approximately 550 directional miles of Interstate 5 (I-5); 600 directional miles of I-90; 260 directional miles of I-82; 50 directional miles of I-405; 50 directional miles of State Route (SR) 167; 40 directional miles of SR 18; 20 directional miles of SR 512; 20 directional miles of SR 14; and 10 directional miles of SR 101. The 3-88 project team used some of these segments only for analysis of less common ramp combinations to increase sample size. The 3-88 project team collected the general ramp type (characterized for this study as either diagonal, direct, semi-direct, turning roadway, or loop) as well as the interchange number and general interchange type (system or service) associated with each ramp using interchange diagrams and verified these elements with aerial photography from Google Earth. An example interchange diagram and corresponding satellite photograph is illustrated in Exhibit 3-26. The 3-88 project team marked each interchange number and name in Google Earth for quick future referencing if needed.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-46 Exhibit 3-26 Example of Interchange Diagram and Google Earth Aerial Photography to Collect Interchange Number, General Interchange Type and Ramp Type The 3-88 project team collected physical ramp-gore locations, associated cross-street locations, and the beginning or end of the acceleration or deceleration lane tapers from interchange diagrams and verified using WSDOT’s State Route Web video logs. Painted ramp-gore locations were collected from the video logs and verified using a measurement tool in Google Earth. The 3-88 project team also obtained the vertical relationship between the freeway and cross street (i.e., freeway mainline under or over cross street) associated with each ramp from the interchange diagrams and verified the relationships with video logs and Google Maps Street View. Finally, the 3-88 project team recorded each ramp identification (ramp ID) number.. The

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-47 ramp ID, along with the ramp milepost, were used to collect the daily exiting or entering traffic volumes from the HSIS Washington ramp files. Examples of the aforementioned ramp features, as they appear in the interchange diagrams and video logs, are illustrated in Exhibit 3-27. Exhibit 3-27 Example of Ramp Features Collected from Interchange Diagrams and Video Logs Milepoints shown on WsDOT’s video logs and interchange diagrams (see Exhibit 3-23) are based on a State Route Milepost (SRM) system, which includes sequential numbers, increasing from south to north or west to east, in 1/100th mile increments. The SRM system does not include adjustments that are effective over the entire route following realignments and accompanying route lengthening or shortening. Instead, the SRM includes “back” and “ahead” indicators, which are notations that distinguish “new Begin Lane Taper Physical Gore Ramp ID Cross Street Location Mainline under Cross Street Begin Lane Taper Physical Gore Cross-Street Location Mainline under Cross Street Painted Gore

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-48 milepoints” from “old milepoints” of the same value that are either upstream or downstream of the realignment locations (see Exhibit 3-28). WSDOT’s road data also include an Accumulated Route Mileage (ARM) for each route, also measured to 1/100th mile. The ARM starts at the beginning of each state route, increases from south to north or west to east, and is adjusted to account for previous route realignments. The ARM numbers are provided in the Washington roadway and crash files obtained by the 3-88 project team through HSIS and were used for segment length and spacing calculations and to link road and crash files. WSDOT’s State Route Web includes both the SRM and ARM milepoints, which allowed the 3-88 project team to determine locations of previous route realignments and identify and correct for any potential crash miscounts over a three year period for a defined road segment. An example of the SRM and ARM numbers in State Route Web at a location where a route was lengthened during a previous reconstruction project is shown in Exhibit 3-28. 1) Realignment and lengthening of the route by 0.07 mile 2) Beginning of an exit ramp taper and its ARM milepoint (Accum Miles) and SRM milepoint (Mile Post) 3) Physical gore of the exit ramp and its ARM milepoint (Accum Miles) and SRM milepoint (Mile Post) Exhibit 3-28 Example of mile post adjustment information 3.4.3.2 Freeway segments for safety analysis The 3-88 project team then used the ramp data to define freeway segments; the freeway segments were the base observation units for the safety analysis. Each row in the ramp database, created through execution of steps described in the previous section, represented one ramp. Each row in the subsequent freeway segment database represented one ramp combination. The 3-88 project team created two segment databases - one for each ramp sequence studied. The 3-88 project team identified consecutive rows in the ramp

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-49 database where an entrance ramp was followed by an exit ramp and combined them into one row (i.e., one EN-EX freeway segment). The beginning of the segment was defined by the milepost of the physical entrance ramp gore; the end of the segment was defined by the milepost of the physical exit ramp gore. Ramp spacing was defined as the distance from the painted entrance gore (i.e., the merging tip) to the painted exit gore (the diverging tip) for the EN-EX scenario. The EN-EX freeway segment boundaries and the defined ramp spacing dimension are shown in Exhibit 3- 29. Exhibit 3-29 Illustration of Defined Segment Boundaries and Ramp Spacing for EN-EX The 3-88 project team combined consecutive rows in the ramp database where an entrance ramp was followed by another entrance ramp to form an EN-EN segment. The freeway segment was defined from the physical gore of the first (upstream) entrance ramp to the end of the acceleration lane taper of the second (downstream) entrance ramp. The 3-88 project team defined the segment this way to capture crashes associated with merging activities at both ramp locations. Ramp spacing was defined as the distance between merging tips for the EN-EN scenario. The EN-EN freeway segment boundaries and the defined ramp spacing dimension are shown in Exhibit 3- 30. Exhibit 3-30 Illustration of Defined Segment Boundaries and Ramp Spacing for EN-EN The 3-88 project team excluded segments from the dataset if the team identified construction activity on or near the segment from 2005 through 2008. The 3-88 project team identified temporary traffic control devices on the video logs or construction areas present on current and archived Google

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-50 Earth photographs for these segments. Missing traffic volume counts (discussed in section 3.4.3.3) for a segment also indicated possible construction activity; segments with missing volume counts were excluded. The 3-88 project team did not spend any additional resources (e.g., time to personally interview Washington DOT personnel) to identify work zones from 2005-2008. Work zone presence is likely not correlated with traffic and geometric variables included in the safety models; higher levels of unexplained variability in expected crash counts, a less serious flaw than omitted variable bias (discussed in section 3.4.4), is expected if some work zones were missed during the screening process. The 3-88 project team also excluded several other types of ramps from the dataset. The project team excluded rest-area ramps between entrance and exit ramps associated with two consecutive cross streets. The 3-88 project team also excluded segments with any type of ramp metering and high occupancy vehicle (HOV) facilities. Finally, the 3-88 project team excluded EN-EX weaving segments between cloverleaf ramps. These areas could have different safety performance characteristics than weaving areas between two consecutive interchanges, and the 3-88 project team did not have sufficient resources to develop a separate model for such areas. The final datasets used by the 3-88 project team to estimate the safety models described in section 3.4.5 consist of 155 EN-EX segments and 30 EN-EN segments. 3.4.3.3 Traffic and geometric data for defined freeway segments The 3-88 project team collected traffic and geometric data for each defined freeway segment. Freeway mainline traffic volumes were collected from HSIS roadway files using route number and mainline milepost variables to identify the correct volume measurement. The mainline traffic volume assigned to each defined freeway segment represented the directional average daily traffic just upstream of the physical entrance ramp gore of the EN-EX and upstream of the first (upstream) physical entrance ramp gore of the EN- EN. The HSIS files included bidirectional traffic volumes. The 3-88 project team used the process described below to estimate directional traffic. Each defined freeway segment was linked to its nearest Automated Data Collection (ADS) station. These traffic data collection stations are located throughout Washington State’s highway system. Data collected by this automated system are summarized in WSDOT’s annual traffic reports and include directional mainline traffic volumes. The 3-88 project team used the directional volume information to estimate a directional traffic volume ratio (D). The 3-88 project team then assumed that the directional traffic volume ratio for each defined freeway segment was the same or very close to the volume ratio at the nearest ADS station. All defined freeway segments had an estimated directional traffic volume ratio falling between 0.49 and 0.51.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-51 The 3-88 project team determined entering and exiting traffic volumes using the ramp ID number and ramp milepost variables, and represented the average daily traffic on the entrance and exit ramp-freeway terminals, respectively. The number of through lanes was determined using HSIS roadway files and confirmed with video logs, Google Earth aerial photography, and Google Maps Street View. The presence of an auxiliary lane between an entrance and exit ramp was determined from the interchange diagrams and also confirmed with video logs. The number of lanes on the entrance and exit ramps (at the ramp-freeway terminal) was determined from video logs alone. Finally, the presence of a barrier (concrete or steel guardrail) adjacent to the right and median shoulders, as well as the respective barrier length, was collected using video logs. Descriptive statistics of the traffic and geometric variables for the 155 EN-EX and EN-EN segments are provided in Exhibit 3-31 and Exhibit 3-32, respectively. Safety models for EN-EN segments included only traffic volumes, segment length and spacing as variables because of the small sample size. Only descriptive statistics for those variables are included in Exhibit 3-32. Variable Mean Standard Deviation Minimum Maximum L 1.59 1.12 0.24 5.29 DADT 41,644 21,281 15,928 104,079 ADT 5,799 EN 4,935 113 31,395 ADT 5,727 EX 5,051 84 31,495 S 7,369 5,903 686 26,770 AuxLn 0.1871 0.3912 a 0 1 %BarrL 0.5499 0.4296 0 1 %BarrR 0.3559 0.2897 0 1 MainEn 0.3613 0.4819 0 1 MainEx 0.3548 0.4800 0 1 NoLn2 0.3612 0.4819 0 1 NoLn3 0.4645 0.5004 0 1 NoLn4 0.1742 0.3805 0 1 a Exhibit 3-31 Descriptive Statistics of Traffic and Geometric Data from 155 EN-EX Segments The mean of an indicator variable is interpreted as the proportion of segments with the indicator value equal to 1 (e.g., 18.59% of the 156 segments have an auxiliary lane present). Variable Mean Standard Deviation Minimum Maximum L 0.64 0.34 0.23 1.56 DADT 42,254 23,449 3,459 99,030 ADT 8,774 EN-1 6,220 208 21,649 ADT 6,354 EN-2 5,271 707 27,570 S 2,821 3,146 686 17,160 Exhibit 3-32 Descriptive Statistics of Traffic and Geometric Data from 30 EN-EN Segments

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-52 3.4.3.4 Crash frequencies and severities on each defined freeway segment The 3-88 project team counted the number of crashes occurring on each freeway segment (i.e., between the physical entrance gore and exit gore for the EN-EX and the upstream physical exit gore and downstream end of acceleration lane taper for EN-EN) in the years 2005, 2006, and 2007 using the route number and milepost variables. The following crash counts included: • Number of crashes of all severities and types; • Number of crashes resulting in at least one occupant fatality or injury; • Number of crashes involving only one vehicle (i.e., single-vehicle crashes); • Number of crashes involving more than one vehicle (i.e., multiple- vehicle crashes); • Number of crashes involving a large truck; and, • Number of crashes occurring during predefined peak hours. The 3-88 project team classified a vehicle as a large truck if it was coded in HSIS vehicle files as 1) truck (over 10,000), 2) truck tractor, 3) truck tractor & semi-trailer, or 4) other truck combinations. The 3-88 project team defined peak hours as 7-10 a.m. and 3-6 p.m. The 3-88 project team summarized the large truck and peak hour safety models in the Second Interim Report for NCHRP 3-88. The proportion of explained variation by the truck safety model was smallest (i.e., was the weakest of all safety models estimated), indicating the model specification for truck crashes was missing key variables. Truck-volume data available through HSIS were tested by the 3-88 project team, but the data quality appeared suspect and were ultimately excluded from the model. Any future comprehensive model of truck crashes should include truck volumes. The peak hour safety models were not any different that the safety models for total crashes. Only the safety models for the expected number of total crashes, fatal plus injury crashes, multiple vehicle crashes, and single vehicle crashes are included in this report as a result of these earlier findings. The 3-88 project team included crashes in the analysis only if they were coded as occurring in the roadway or roadside of the freeway mainline and in the same direction of travel served by the interchange ramps. Crashes coded as having occurred on the ramp proper or in the opposing direction of freeway travel were not assigned to the segment of interest. The 3-88 project team used the HSIS impact location and travel direction variables to identify these appropriate crashes. One limitation of this approach is that it may not capture the complex interactions between cross-median, head on collisions. Descriptive statistics of the observed crash frequencies for the 155 EN-EX

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-53 segments and 30 EN-EN segments are provided in Exhibit 3-33 and Exhibit 3-34, respectively. Crash Type Mean Standard Deviation Minimum Maximum Total 33.4 31.0 2 189 FplusI 11.5 11.4 0 65 SingV 11.7 10.1 0 60 MultV 21.8 25.4 0 141 Exhibit 3-33 Descriptive Statistics of Observed Crash Frequencies from 2005 through 2007 (inclusive) on 155 EN-EX Segments Crash Type Mean Standard Deviation Minimum Maximum Total 28.8 29.3 3 131 FplusI 10.1 11.8 1 54 SingV 6.2 4.1 0 17 MultV 22.5 26.9 0 120 Exhibit 3-34 Descriptive Statistics of Observed Crash Frequencies from 2005 through 2007 (inclusive) on 30 EN-EN Segments 3.4.4 Modeling Approach The 3-88 project team explored the relationship between ramp spacing and safety using a negative binomial regression modeling approach. In 1986, Jovanis and Chang introduced the use of Poisson regression to model the relationships between crash frequency, traffic volumes, and weather conditions (83). Miaou later used the negative binomial regression model, a more general form of the Poisson regression model, to explore the relationship between crash frequencies, daily traffic, and highway geometric design variables (59). Negative binomial regression has become the most widely used technique to model crash frequency-geometric design relationships since that time. In the negative binomial model, the expected number of crashes of type i on segment j is expressed as: μ ij = E(Yij) = exp(Xjβ + ln Lj where: μ ) ij = E(Yij X ) = the expected number of crashes of type i on segment j; j β = regression coefficients estimated with maximum likelihood that quantify the relationship between E(Y = a set of traffic and geometric variables characterizing segment j (including ramp spacing); ij L ) and variables in X; j ln L = length of segment j; and, j = the natural logarithm of segment length.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-54 The following crash frequencies were modeled by the NCHRP 3-88 project team: • Expected number of crashes of all severities and types (Total); • Expected number of crashes involving at least one occupant fatality or injury (FplusI); • Expected number of crashes involving only one motor vehicle (SingV); • Expected number of crashes involving more than one motor vehicle (MultV); Ramp spacing was the primary variable of interest in the matrix of explanatory variables, Xj • Segment length . However, the 3-88 project team included a number of other traffic and geometric variables to decrease unexplained variation in expected crash frequency and to try and minimize omitted variable bias. Omitted variable bias involves over- or under- estimating the safety effect of ramp spacing due to other variables that influence crash frequency and are correlated with ramp spacing, but are excluded from the model. Measures of the following explanatory variables were included in the NCHRP 3-88 EN- EX safety models: • Freeway traffic • Ramp traffic • Ramp spacing • Presence of an auxiliary weaving lane • Barrier presence and length • Vertical relationship between the freeway mainline and cross streets • Number of freeway through lanes The number of EN-EN segments included in the safety analysis by the 3-88 project team was much smaller than for the EN-EX sample. Model specifications for the EN-EN were less robust as a result. Measures of the following explanatory variables were included in the NCHRP 3-88 EN-EN safety models: • Segment length • Freeway traffic • Ramp traffic • Ramp spacing

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-55 The 3-88 project team included segment length, L, in the models as an offset variable (i.e., the regression coefficient for the natural logarithm of segment length is constrained to 1.0), and captures the linear increase in expected crash frequency with an increase in segment length due to increased exposure. The 3-88 project team evaluated model fit using the McFadden Pseudo R- Squared. The McFadden Pseudo R-Squared (ρ2 ρ ) is analogous to the R- squared value used to express the goodness of fit of a standard, ordinary least squares regression model. It is expressed as: 2 )0( )( L fullL = where: ρ2 L(full) = log-likelihood of the model with explanatory variables; and, = McFadden Pseudo R-Squared; L(0) = log-likelihood of the intercept-only model. The McFadden Pseudo R-Squared may take a value between 0 and 1; the value moves closer to 1 as model fit improves. 3.4.5 Model Results Model estimation results are summarized in Exhibit 3-35 (EN-EX) and Exhibit 3-36 (EN-EN). Models for EN-EX segments have the following form, which is consistent with the general modeling discussion in the section 3.4.4: E(Yi) = exp (constant + 1.0*ln(L) + b2*ln(DADT) + b3*ln(ADTEN) + b4*ln(ADTEX) + b5*S-1 + b6*AuxLn + b7*%BarrL + b8*%BarrR + b9*MainEn + b10*MainEx + b11 with all variables defined above and b *NoLn2) i E(Y equal to estimated regression coefficients listed in Exhibit 3-35. The EN-EX model can also be expressed as: i) = L1.0 DADTb2 ADTENb3 ADTEXb4 exp(constant + (b5/S) + b6*AuxLn + b7*%BarrL + b8*%BarrR + b9*MainEn + b10*MainEx + b11 For example, the model for the expected number of crashes of all types and severities (i.e., E(Y *NoLn2) i E(Y ) = Total) is expressed as: i) = Total = L1.0 DADT1.122 ADTEN0.1766 ADTEX0.0174 exp(-10.75 + (448.6/S) – 0.2283*AuxLn + 0.1026*%BarrL + 0.4243*%BarrR + 0.1184*MainEn + 0.0221*MainEx + 0.1184*NoLn2)

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-56 Variable Total FplusI SingV MultV Constant -10.75* -12.85* -2.092** -18.51* Ln(L) 1.000 1.000 1.000 1.000 Ln(DADT) 1.122* 1.204* 0.3461* 1.677* Ln(ADTEN 0.1766* ) 0.1706* 0.0271 0.2809* Ln(ADTEX 0.0174 ) 0.0366 0.0198 0.0486 S 448.6* -1 171.1 -39.44 691.2* AuxLn -0.2283** -0.2622** -0.2241 -0.2788** %BarrL 0.1026 0.2976* -0.1072 0.2265** %BarrR 0.4243* 0.3336* 0.3858* 0.4121* MainEn 0.1184 0.1029 0.0861 0.1879** MainEx 0.0221 0.0512 0.0727 0.0016 NoLn2 0.1184 0.1934 -0.0694 0.3657* α 0.1643* 0.1643* 0.1262* 0.2293* ρ 0.1528 2 0.1528 0.0354 0.2965 *parameter statistically significant with probability of type I error ≤ 0.05 ** parameter statistically significant with probability of type I error ≤ 0.1 Exhibit 3-35 Summary of EN-EX Model Estimation Results; Dependent Variable is Expected Number of Crashes in One Direction of Freeway Travel between 2005 and 2007 (expected crashes per 3 years) Models for EN-EN segments have the following form: E(Yi) = exp (constant + 1.0*ln(L) + b2*ln(DADT) + b3*ln(ADTEN-1) + b4*ln(ADTEN-2) + b5*S-1 with all variables defined above and b ) i E(Y equal to estimated regression coefficients listed in Exhibit 3-36. The EN-EN model can also be expressed as: i) = L1.0 DADTb2 ADTEN-1b3 ADTEN-2b4 exp(constant + (b5 For example, the model for the expected number of crashes of all types and severities (i.e., E(Y /S)) i E(Y ) = Total) is expressed as: i) = Total = L1.0 DADT1.140 ADTEN-10.1730 ADTEN-20.0222 exp(-11.73 + (434.3/S))

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-57 Variable Total FplusI SingV MultV Constant -8.812* -11.08* -3.277** -14.55 Ln(L) 1.000 1.000 1.000 1.000 Ln(DADT) 0.8095* 0.8116* 0.3949** 1.096 Ln(ADTEN-1 0.3387* ) 0.4530* 0.0805 0.5873 Ln(ADTEN-2 0.0931 ) 0.1332 0.0762 0.1101 S 418.1 -1 54.74 234.2 346.9 α 0.2770* 0.2924 0.2382* 0.2952* ρ 0.1262 2 0.1451 0.0507 0.1669 *parameter statistically significant with probability of type I error ≤ 0.05 ** parameter statistically significant with probability of type I error ≤ 0.1 Exhibit 3-36 Summary of EN-EN Model Estimation Results; Dependent Variable is Expected Number of Crashes in One Direction of Freeway Travel between 2005 and 2007 (expected crashes per 3 years) 3.4.5.1 EN-EX Model Interpretation The signs (positive or negative) of the estimated NCHRP 3-88 model parameters, the statistical significance of the parameters, the relative parameter magnitudes across the different crash frequency models, and the relative levels of model fit were consistent with what one would expect at locations of merging, diverging, acceleration, and deceleration maneuvers associated with the EN-EX segments. The model for expected number of multiple-vehicle crashes demonstrated a combination of the largest ramp spacing effect and the best model fit. The parameter for ramp spacing associated with the expected number of total crashes (i.e., all severities and types) was also statistically significant, a result that is expected given that a majority of these crashes involve multiple vehicles. The parameter for ramp spacing associated with the expected number of crashes resulting in a fatality or injury was positive, but not statistically significant and much smaller than the ramp parameter for crashes of all severities and types. While the results suggest an increase in the frequency of severe crashes with decreasing ramp spacing, the expected proportion of crashes resulting in a fatality or injury appears to decrease as ramp spacing decreases. The result of the NCHRP 3-88 research is consistent with published findings reported by Milton et al. (74) (see Chapter 2). The parameter for ramp spacing associated with expected single-vehicle crashes was negative, indicating a decrease in the frequency of single-vehicle crashes as ramp spacing decreases. The finding is expected; the opportunity for single-vehicle crashes decreases as lane-change intensity increases. Single- vehicle crashes are associated with lower-volume, higher-speed conditions. Each of the findings related to the crash frequency-ramp spacing relationship are discussed in greater detail in section 3.4.6. The 3-88 project team concluded that findings associated with other traffic and geometric variables also were generally consistent with previous safety

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-58 modeling research and with expectations. The parameter for ln(ADT) was highly significant for all models, greater than 1.0 for multiple-vehicle crashes and less than 1.0 for single-vehicle crashes. The parameters indicate a smaller likelihood of a crash involving one vehicle compared to the likelihood of a crash involving more than one vehicle as traffic volumes increase. The finding is consistent with freeway safety research conducted by Bonneson and Pratt (2008) (84). The parameter for ln(ADTEN) was positive and statistically significant for all crash types except single vehicle, indicating entrance volumes are associated with multiple-vehicle crashes resulting from increased merging maneuvers. The parameter value for ln(ADTEN The amount of exiting traffic, represented by ln(ADT ) is less than 1.0 for all crash types; this finding is consistent with previous operational research indicating that the merge maneuver becomes easier as entrance volumes increase because the entering vehicles become a more dominant movement. EX The presence of an auxiliary lane between the entrance and exit ramps was modeled as an indicator variable (i.e., 1 = auxiliary lane present; 0 = auxiliary lane not present). Model parameters indicate that, for a given ramp spacing, fewer crashes are expected when an auxiliary lane is present. The auxiliary lane effect was largest for multiple-vehicle crashes and fatal-plus-injury crashes. Approximately 24 to 23 percent fewer multiple-vehicle crashes and fatal-plus-injury crashes, respectively are expected when an auxiliary lane is present compared to when there is not an auxiliary lane for the same ramp spacing dimension. Approximately 20 percent fewer total crashes (i.e., of all severities and types) are expected when an auxiliary lane is present; the respective regression parameter is also statistically significant. The difference was similar in magnitude for also single-vehicle crashes, but is not statistically significant. ), did not appear to influence expected crash frequencies. The results indicate that exiting traffic directly causes a minimal, if any, safety disturbance. The result is consistent with its operational counterpart; free-flow speed adjustments are made only for interchanges with entrance ramps in the 2000 HCM freeway segment methodology. The presence and length of a barrier adjacent to the right-side and median shoulders generally was associated with an increase in crashes, as expected (i.e., there is less room for roadside recovery before striking an object). However, the expected frequency of single-vehicle crashes decreased when a barrier was adjacent to the median shoulder (the parameter was not statistically significant). The result may indicate that crashes with a median barrier close to the traveled way may often be reported as multiple-vehicle crashes due to a redirection into the traveled way and into other vehicles after an initial barrier strike.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-59 Results also show an increase in the expected number of fatal and injury crashes with barrier presence and length. For a barrier adjacent to the right shoulder, the increase is slightly smaller in magnitude and less statistically significant than for crashes of all severities and types, indicating a reduction in the proportion of crashes resulting in fatalities and injuries. However, a barrier adjacent to the median shoulder is associated with an increase in the proportion of crashes resulting in fatalities and injuries. Median width appears to be a key missing variable at this stage; median barriers are expected to increase crash frequency, but decrease crash severity, for narrower median widths. In addition, some detail is lost by the fact that all injury levels are currently combined into one crash outcome category. Parameters for variables representing the vertical relationship between the freeway mainline and cross street indicate that an increased number of crashes are expected when an entrance ramp joins a freeway from a cross street that passes below the freeway. The effect was largest and statistically significant for multiple-vehicle crashes. The result is expected; vehicles are not as likely to reach freeway speeds before merging when traveling on an entrance ramp with a positive grade. Available sight distance may also be limited when joining a freeway from a lower elevation. The same phenomena were not found for exit ramps; parameters were generally positive, but small and not statistically significant. This was also expected; the vertical relationship between mainline and cross streets for exit ramps is more likely to influence crash frequency on the ramp proper and ramp-cross street terminal. These crash types were excluded from the NCHRP 3-88 analysis. However, the positive parameters for all crashes do indicate a small safety benefit of having the cross street associated with the exit ramp pass over the freeway mainline. Finally, the expected frequencies of all crash categories except single-vehicle were larger when only two through lanes per direction were present (compared to when three or four were present). The effect was largest and most statistically significant for multiple-vehicle crashes. For a given traffic volume, three or four lanes (compared to two) provide additional room for through moving traffic to move away from merging traffic, decreasing the probability of merging conflicts. The safety finding has an operational counterpart; the weaving intensity factor, and resulting travel speeds, decrease as the number of lanes increase in the HCM weaving analysis methodology. 3.4.5.2 EN-EN Model Interpretation The signs (positive or negative) of the estimated NCHRP 3-88 model parameters, the statistical significance of the parameters, the relative parameter magnitudes across the different crash frequency models, and the relative levels of model fit were also consistent with what one would expect for the EN-EN models. The EN-EN model specifications were much more

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-60 limited than the EN-EX models because of the small sample sizes associated with this scenario. Therefore, model interpretations must be much more general with less attention to statistical significance The model for expected number of total crashes demonstrated the largest ramp spacing effect and the model for expected number of multiple vehicle crashes demonstrated the best model fit. None of regression parameters associated with ramp spacing were statistically significant. Again, the parameter for ramp spacing associated with the expected number of crashes resulting in a fatality or injury was positive, but much smaller than the ramp parameter for crashes of all severities and types indicating the expected proportion of crashes resulting in a fatality or injury decreases as ramp spacing decreases. The parameter for ln(ADT) was significant for all models, greater than 1.0 for multiple-vehicle crashes and less than 1.0 for single-vehicle crashes as expected (see discussion in 3.4.5.1). The parameter for ln(ADTEN-1) was positive and statistically significant for all crash types except single vehicle, indicating higher entrance volumes on the first (upstream) ramp of an EN- EN sequence are associated with multiple-vehicle crashes resulting from increased merging maneuvers. The parameter value for ln(ADTEN-1 Increasing amounts of entering traffic on the second (downstream) entrance ramp, represented by ln(ADT ) is less than 1.0 for all crash types; this finding is consistent with previous operational research (see discussion in 3.4.5.1). EN-2), were associated with higher crash frequencies, but the effect was not as large as for ln(ADTEN-1 3.4.6 Model Validation through Comparisons of NCHRP 3-88 EN- EX Findings to a Recent Texas Study ). The result may indicate that crashes associated with entrance ramps occur primarily at or downstream of the merge location. The crashes associated with the volumes on the second entrance ramp are not captured by within the segment boundaries shown in Exhibit 3-30. The 3-88 project team conducted a focused safety research effort with the philosophy that a smaller sample of complete data would provide more information about ramp spacing-safety relationships than a larger, multistate sample of data with less quality control efforts and missing variables. The 3- 88 project team considered data from Washington most relevant to the research objectives. Data collection efforts were focused in Washington because of the availability of key data sources (e.g., online video logs and interchange diagrams) and key data elements (e.g., ramp gore mileposts, ramp ID numbers, ramp volumes, and freeway crash-location variables). The transferability of results to other states and geographic regions is generally the primary concern with focused, single-state efforts such as the one undertaken to date. A study by the Bonneson and Pratt of Texas Transportation Institute (TTI) that included safety modeling of freeway segments and other freeway features was published after the safety work plan

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-61 of this project began (84). Where possible, the 3-88 project team compared the safety findings from analyses of Washington data to related safety findings from Texas. Similar underlying safety trends in two states as different in location, climate, topography, and freeway design as Washington and Texas would build confidence in the general transferability of the 3-88 results. The comparison efforts and conclusions are described in the remainder of this section. A weaving accident modification factor (AMF) for a Texas freeway was reported by Bonneson and Pratt (84) and took the form: */9.152 ,, wevL TXFIwev eAMF = for 800* ≥wevL feet where: AMFwev,FI,TX L* = accident modification factor for fatal and injury crashes using Texas freeway data; and, wev The AMF takes the value of 1.0 as the weaving length approaches infinity (i.e., a basic freeway segment). The TTI AMF was modified by the 3-88 project team to have a base condition of 2,000 feet in order to compare the results to the applicable NCHRP 3-88 AMF. The modified TTI AMF is expressed as: = weaving section length (feet). )/9.152(07645.0 2000,,, * wevL TXFIwev eAMF +−= for 800* ≥wevL feet where: AMFwev,FI,TX,2000 L* = accident modification factor for fatal and injury crashes using Texas freeway data with base condition of weaving length equals 2,000 feet; and, wev The relationship between AMF = weaving section length (feet). wev,TX,2000 and ramp spacing is plotted in Exhibit 3-37.

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-62 Exhibit 3-37 Comparison of Accident Modification Factors from NCHRP 3-88 and TxDOT Study 4703 for Fatal and Injury Crashes (AMF set to 1.0 for Ramp Spacing of 2,000 Feet) The NCHRP 3-88 AMF developed with Washington data most applicable for comparison to the Texas AMF is given by: )/1.171(05995.0 2000,,, S WAFIEXEN eAMF +− − = where: AMFEN-EX,FI,WA,2000 S = ramp spacing, defined from painted gore of entrance ramp to painted gore of exit ramp (feet). = accident modification factor for fatal and injury crashes using Washington freeway data with base condition of ramp spacing equals 2,000 feet; and The relationship between AMFEN-EX,WA,2000 Differences between the Washington and Texas studies include: and ramp spacing is also plotted in Exhibit 3-37. The findings are very similar, and support the conclusion in one is likely to learn more about a transferable, underlying safety phenomenon associated with ramp spacing using a small sample of carefully collected data than with a larger sample of incomplete data. • The Texas AMF is for weaving segments only; the Washington AMF is currently applicable to both weaving segments and entrance-exit segments without an auxiliary lane.

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-63 • The Washington segments were defined from entrance gore to exit gore; the Texas freeway segments included one, more than one, or sometimes only part of a weaving segment. • Ramp spacing in the Texas data ranged from 800 to 4,000 feet; spacing in the Washington data ranged from 700 to 27,000 feet. 3.4.7 Safety Conclusions The primary focus of the NCHRP 3-88 research was the scenario of an entrance ramp followed by an exit ramp, a commonly occurring ramp sequence and one in which safety information is frequently needed. The 3-88 project team also explored the safety/ramp spacing relationship at locations where an entrance ramp is followed by another entrance ramp, but at less level of detail than the EN-EX scenario. The 3-88 project team used data from freeway segments in Washington State to estimate a series of negative binomial regression models. While ramp spacing was the key variable of interest, a number of other traffic and geometric variables were included in the model specification to avoid over- or under-estimating the ramp spacing- safety effect. The 3-88 project team found the signs (positive or negative) of the estimated model parameters, the statistical significance of the parameters, the relative parameter patterns across the different crash frequency models, and the relative levels of model fit were intuitive. The 3-88 project team reduced modeling results to a set of accident modification factors (AMFs) and safety performance functions (SPFs). The AMFs can be used to estimate the expected incremental safety effect of different ramp spacing dimensions under a set of fixed traffic and roadway conditions (e.g., alternatives to upgrade existing ramps with no difference in ramp volumes between different alternatives). The SPFs can be used to estimate the expected number of crashes for scenarios where ramp spacing, ramp traffic and mainline traffic are all changing (e.g., the addition of a new interchange between two existing interchanges with expected travel pattern impacts). NCHRP 3-88 research results show that the expected number of total crashes increases as ramp spacing decreases. The sensitivity of ramp spacing to total crashes appears highest for ramp spacing values less than 2,000 feet and becomes nearly negligible for spacings beyond 3,000 feet. The crash increase is largely a result of an increase in multiple-vehicle crashes; the expected number of single-vehicle crashes decreased as ramp spacing decreased. The finding is consistent with expectations that ramp presence and spacing is most directly associated with multiple-vehicle crashes resulting from increased numbers and densities of merging, diverging, acceleration, and deceleration maneuvers. The expected number of severe crashes, those resulting in at least one occupant fatality or injury, also increased as ramp spacing decreased. The increase was at a much lower rate than for total crashes, indicating that much

Final Report NCHRP 3-88 Chapter 3: Research Activities and Findings Guidelines for Ramp and Interchange Spacing 3-64 of the total crash increase is a result of an increase in less severe collisions (likely lower speed sideswipe and rear-end collisions). The increase in severe crashes appears to become relatively negligible beyond 1,000 to 1,500 feet. The 3-88 project team established the presence of an auxiliary lane between the entrance and exit ramps as an indicator variable (i.e., 1 = auxiliary lane present; 0 = auxiliary lane not present). Model parameters indicate that, for a given ramp spacing, fewer crashes are expected when an auxiliary lane is present. The auxiliary lane effect is largest and most statistically significant for multiple-vehicle crashes and fatal-plus-injury crashes. Approximately 24 percent fewer multiple-vehicle crashes and 23 percent fewer fatal-plus-injury crashes are expected when an auxiliary lane is present compared to when there is not an auxiliary lane for the same ramp spacing dimension. At the time of this research, no interaction between ramp spacing and weaving lane presence had been captured; however, comparisons to a Texas study indicate this interaction may be small or nonexistent. Model parameters indicate fewer crashes with an auxiliary lane present for a given ramp spacing. Ramp spacing was the primary variable of interest; however, the 3-88 project team included a number of other traffic and geometric variables to decrease unexplained variation in expected crash frequency and to try and minimize omitted variable bias. Variables representing the vertical relationship between the freeway mainline and cross street indicate that an increased number of crashes are expected when an entrance ramp joins a freeway from a cross street that passes below the freeway. The effect is largest and statistically significant for multiple-vehicle crashes and truck crashes. The presence and length of a barrier adjacent to the right-side and median shoulders was generally associated with an increase in crashes as expected (i.e., there is less room for roadside recovery before striking an object). Capturing this variable was key to estimating the safety effect of ramp spacing. Barriers normally are present near urban areas where spacing is shorter. The effect of ramp spacing on the expected number of crashes would have been overestimated without variables for the presence and length of barriers included in the safety model. NCHRP 3-88 research found the parameter for freeway mainline traffic was highly significant for all models, greater than 1.0 for multiple-vehicle crashes and less than 1.0 for single-vehicle crashes. The parameters indicate a smaller likelihood of a crash involving one vehicle compared to the likelihood of a crash involving more than one vehicle as traffic volumes increase and is consistent with previous research (Bonneson and Pratt, 2008) (13). The parameter for entrance ramp traffic was positive and statistically significant in the EN-EX models for all crash types except single-vehicle, indicating entrance volumes are associated with multiple-vehicle crashes resulting from increased merging maneuvers. The parameter for ln(ADTEN-1) was positive and statistically significant for all crash types in the EN-EN models except single vehicle, indicating higher entrance volumes on the first (upstream)

NCHRP 3-88 Final Report Guidelines for Ramp and Interchange Spacing Chapter 3: Research Activities and Findings 3-65 ramp of an EN-EN sequence are associated with multiple-vehicle crashes resulting from increased merging maneuvers. Increasing amounts of entering traffic on the second (downstream) entrance ramp, represented by ln(ADTEN-2), were associated with higher crash frequencies, but the effect was not as large as for ln(ADTEN-1 Expected frequencies of all crash categories except single-vehicle were larger when only two through lanes per direction were present (compared to when three or four were present). The effect was largest and most statistically significant for multiple-vehicle crashes. The NCHRP 3-88 findings indicate that for a given traffic volume, three or four lanes (compared to two) provide additional room for through moving traffic to move away from merging traffic, decreasing the probability of merging conflicts. ). The result may indicate that crashes associated with entrance ramps occur primarily at or downstream of the merge location. The transferability of results to other states and geographic regions is generally the primary concern with focused, single-state efforts such as the one undertaken to date. Where possible, the 3-88 research team compared the safety findings from analyses of Washington data to related safety findings from a recent Texas study (84). The findings of AMF and SPF comparisons showed very similar safety relationships in Washington and Texas, and built confidence that the general safety trends uncovered in NCHRP 3-88 are transferable to areas with other geographic characteristics.