Source VHB NCHRP 17 74 Application of CMFs

Source: VHB NCHRP 17 -74: Application of CMFs for Access Management 1

Background 2

What is the difference in safety performance for these two multilane, median-divided facilities? A B 3

Overview Do current crash prediction models adequately capture the differences in access management features? • Highway Safety Manual (HSM) provides crash prediction models for assessing and quantifying safety consequences of planning, designing, and operating highway facilities • Access management is effective component in operation and safety of roadways but few opportunities to quantify safety effects • Validity of applying crash modification factors (CMFs) for access management strategies to existing safety performance functions (SPFs) o HSM established SPFs without consideration of base conditions for many access management features 4

Objectives 5

Objective of Project Verify reliability of existing SPFs from HSM to quantify safety performance of urban and suburban arterials • Verify application of existing SPFs Quantify safety performance of access features • Refine existing CMFs • Develop new CMFs Develop guidance for applying SPFs and CMFs to quantify safety performance of access features • Do SPFs perform well across sites with different access features? • If not, what adjustment factors (or CMFs) are appropriate? Identify opportunities for future research • Key gaps in data and/or methods that would enhance crash prediction 6

Approach 7

Phase 1 Focus on information gathering and refinement of work plan • Literature review • Survey of practitioners • Data reconnaissance • Gap analysis Source: VHB 8

Literature Review • Summarized safety effects of access management strategies at three levels: site, intersection, and corridor • Identified existing high-quality CMFs and SPFs for access management strategies (which were relatively limited) • Informed gap analysis, identifying areas where high-quality CMFs and SPFs are not available 9

Survey Distribution and representation of respondents 10

Survey of Practitioners • Most indicated their agencies do not quantify safety effects of access management strategies to support decision-making process • 80+% indicated their agencies do not have a policy or procedure for assessing safety effects of access management strategies • Most indicated a priority need for estimating effects of combination strategies, while less than half indicated a priority need for estimating effects of individual strategies • Corridor level analysis is highest priority as opposed to segment/intersection- or site-level analysis 11

Survey Question 1: Do you quantify the safety effects of access management strategies to support related decisions? 60% 50% 40% 30% 20% 10% 0% Yes (please describe in the Comments section below) No 12

Survey Question 1 Survey results show that 15 of 28 do not quantify the safety effects of access management strategies to support policy decisions. “In the past we have used an FHWA chart …This chart is now quite old, so an updated analysis would be helpful. ” “Only because we have not advanced to this level of analysis. ” “Only for projects that request Safety Program funding (state or federal). ” Marked No “…But we want to. ” 13

Survey Question 2: Do you have a policy or procedure for assessing the safety effects of access management strategies? 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Yes (please provide a link or brief description in the comment box below) No 14

Survey Question 2 Survey results show that 23 of 28 respondents indicated their agencies do not have a policy or procedure for assessing safety effects of access management strategies “We probably do, related to the typical section choices …, but I don't have the link. ” “Traffic safety does a good job detailing the benefits of RCI's [restricted crossing intersections]. I am not aware of any other procedure. ” “We have general policy, but not specifically to the safety effects of access management. ” “Occasionally a researcher will gather data for some projects, but nothing that is routine. ” 15

Survey Question 5: What are your priority needs in terms of the level of analysis? Respondents ranked priority needs based on a spatial scale. How many respondents ranked each level as priority… 1 2 3 Site Level 5 9 15 Intersection Level 8 15 6 Corridor Level 16 5 8 16

Survey Question 7: Are there other access management strategies or combination(s) of strategies that should be prioritized in this project? Of the respondents, 21 of 27 indicated “No”. “How do access management techniques impact pedestrians and bicyclists? Both from a crash perspective, and from how people feel using these facilities? ” “The effect of access management on non-motorized. E. g. , increased exposure. ” 17

Data Reconnaissance Determined feasibility of evaluating priority strategies and identified potential data / methodological issues to consider Review Questions • What data are readily available in existing databases? • How many miles of road and counts of intersections (by facility type) are available? • What is the availability of suitable reference locations? • What is the quality of existing data? • Which agencies provided the data? • Who is the point of contact for acquiring the data? 18

Gap Analysis • Summarized current knowledge and identified gaps to establish research priorities and appropriate study design(s) • Met with panel to discuss priorities and determine focus of research, considering schedule, budget, and data availability 19

Primary Research Question How does existing HSM Predictive Method perform for sites with similar geometry, but different access management features (e. g. , access density and spacing, corner clearance, and turning restrictions) not captured in the crash prediction model? 20

Strategies Investigated Intersection Strategies Sub-strategy Site Type Manage spacing of access on crossroads near freeway interchanges Establish interchange ramp terminal spacing Right-turn treatment Channelize right-turn lane 3 -legged stop-control 3 -legged signal 4 -legged stop-control 4 -legged signal Segment Strategies Sub-strategy Site Type Manage location, spacing, and design of median openings Create directional median opening Regulate median opening density 4 -lane divided Regulate median opening spacing Manage location and spacing of unsignalized access Establish corner clearance criteria Manage spacing of signals Establish unsignalized access spacing Manage driveway corner clearance at signalized and unsignalized intersections Establish traffic signal density criteria 2 -lane undivided 3 -lane with TWLTL 4 -lane undivided 4 -lane divided 5 -lane with TWLTL Establish traffic signal spacing criteria 21

Phase 2 Implemented work plan including: • Data collection • Analysis • Documentation Source: VHB 22

Data Collection • Leveraged existing datasets to minimize duplication of efforts and maximize amount of data for analysis in study • Used existing crash, roadway, and traffic characteristic data from NCHRP Project 17 -62 and supplemented with additional access management variables 23

Manual Data collected for: Intersections • Ohio and North Carolina 3, 500 intersections 240 intersections Segments • Ohio and Minnesota 4, 000 segments 500 segments 24

Focus: Urban/Suburban Arterials Ohio Intersection Types Sites Multi-Vehicle Crashes Single-Vehicle Crashes Three-Legged Stop-Controlled (3 ST) 1, 792 3, 319 803 Four-Legged Stop-Controlled (4 ST) 439 1, 325 208 Three-Legged Signalized (3 SG) 218 2, 217 158 Four-Legged Signalized (4 SG) 488 5, 630 335 North Carolina Intersection Types Sites Multi-Vehicle Crashes Single-Vehicle Crashes Three-Legged Stop-Controlled (3 ST) 52 259 45 Four-Legged Stop-Controlled (4 ST) 36 38 Three-Legged Signalized (3 SG) 19 516 37 Four-Legged Signalized (4 SG) 102 5, 793 256 25

Focus: Urban/Suburban Arterials Ohio Segment Types Sites Miles Total Crashes Two-Lane Undivided (2 U) 337 219 1, 948 Two-Lane with TWLTL (3 T) 113 48 618 Four-Lane Divided (4 D) 387 218 3, 219 Four-Lane Undivided (4 U) 216 97 1, 395 Four-Lane with TWLTL (5 T) 157 91 2, 759 Sites Miles Total Crashes Two-Lane Undivided (2 U) 139 29 287 Two-Lane with TWLTL (3 T) 47 8 102 Four-Lane Divided (4 D) 133 24 624 Four-Lane Undivided (4 U) 95 15 219 Four-Lane with TWLTL (5 T) 28 5 145 Minnesota Segment Types 26

Data Analysis Focus on assessing, validating, and enhancing HSM Part C Predictive Method to investigate effect of access management variables on crash predictions 1. Evaluate if Part C Predictive Method shows prediction bias for sites with different access management strategies 2. If substantial prediction bias exists, determine if there is a discernable and explainable pattern 3. Attempt to create CMFs to remove prediction bias for access management features and determine if relationships makes sense 4. Using a validation dataset, determine if findings are consistent among datasets 27

Analysis Methodology Approach 1: Calibration of Existing SPFs • SPFs developed using data from select jurisdiction(s) • Calibration needed when applied to another jurisdiction • Calibration factor: Level of Variable Calibration Factor CMFs 1 0. 9 1. 0 2 1. 1 1. 2 3 1. 2 1. 3 28

Cumulative Residual (CURE) Plots Good Performance • Cumulative residuals largely inside 2 standard deviation boundaries • Minimal sudden jumps or ranges of declines/increases 250 Cumulative Residuals 200 150 100 50 0 -50 0 10 20 30 40 50 60 70 80 -100 -150 -200 -250 Cumulative Residuals HSM Prediction-[Median Opening Density] Upper 95% Conf. Limit Lower 95% Conf. Limit 29

Cumulative Residual (CURE) Plots Poor Performance • Many data points outside 2 standard deviation boundaries • Underpredicts up to ~25 intersections/mile then overpredicts 250 Cumulative Residuals 200 150 100 50 0 -50 0 50 100 150 200 250 300 350 -100 -150 HSM Prediction-[Unsignalized Intersection Density] Cumulative Residuals Upper 95% Conf. Limit Lower 95% Conf. Limit 30

Analysis Methodology Approach 2: Develop New SPFs • Use generalized linear regression modeling (GLM) • Add access management variables to existing HSM models • Estimate new intercept and parameters for additional variables Where: • • SPF = prediction from HSM SPF CMF 1, CMF 2, …, CMFN = values of HSM CMFs intercept = constant term to account for differences between data used for this project and original data used to develop HSM SPFs f(AM) = function representing relationship between crashes and access management related variable(s) 31

Documentation • Developed Practitioner’s Guide to assist transportation planners, designers, and traffic engineers in quantifying safety impacts of access management strategies and making more informed access-related decisions on urban and suburban arterials • Vetted guide with focus group of 40 practitioners representing diverse perspectives from highway safety and access management "Like the material. This is important to this type of audience since the target audience is transportation professionals with or without prior experience in access management and highway safety. " "Like the additional guidance on the selection of CMFs. The selection process is an important component to properly assessing safety performance. " 32

Practitioner’s Guide Outline Chapter 1: Introduction • Background on purpose and need Chapter 2: Definitions for Quantitative Safety Analysis • Terms, general approaches, selecting an approach Chapter 3: Safety Effects of Access Management • Discussion of individual strategies and isolated safety impacts Chapter 4: Predictive Method for Segment- and Intersection. Level Analysis • HSM Predictive Method and adjustment factors Chapter 5: Predictive Method for Corridor-Level Analysis • 2 methods: combine results from Chapter 4 or use corridor models Chapter 6: Communicating Results • Target audiences, measures, formats 33

Results 34

Segment Results For most strategies • Minimal bias, direction of effect was illogical and/or not statistically significant • • Median opening spacing Number of median openings by type Unsignalized intersection spacing Number of unsignalized access points and density Corner clearance Signalized intersection spacing Number of signalized intersections and density 35

Intersection Results Channelized right-turn lane • CMF for channelizing a RT lane (not RT lane presence) • 3 ST • Fewer crashes when channelized • Combined datasets to estimate CMF (0. 72) • Contradictory results for other facility types • Low statistical significance Site Type Ohio CMF (standard error) North Carolina CMF (standard error) Combined CMF (standard error) 4 ST 0. 96 (0. 36) -- -- 4 SG 1. 27 (0. 51) 0. 89 (0. 18) -- 3 ST 0. 69 (0. 28) 0. 75 (0. 32) 0. 72 (0. 25) 3 SG 0. 42 (0. 19) 1. 27 (0. 64) -- 36

Intersection Results Distance to ramp terminal • 3 ST and 4 ST • More crashes expected when ramp within 1, 500 ft. • CMFs from 1. 32 to 2. 12 • 3 SG and 4 SG • No observed bias for signalized intersections Site Type Calibration Factor <= 1, 500’ > 1, 500’ CMF for Ramp Terminal <= 1, 500’ Using < 1, 500’ Using Calibration Factors GLM (standard error) 3 ST, 4 ST 1. 14 0. 60 1. 90 (1. 14) 2. 12 (0. 91) 3 SG and 4 SG 0. 97 1. 04 0. 93 (0. 46) 0. 82 (0. 24) 37

Impact of HSM Segmentation Impact of HSM data structure • Roads segmented into intersections and segments • Crashes are assigned to one of the two site types A – all crashes assigned to intersection B – if identified as intersection-related assigned to intersection 38

Impact of HSM Segmentation Segment-related crashes decrease as intersection density increases (counterintuitive) • Likely a result of how crashes are assigned to intersections Corridor analysis is more appropriate for variables such as intersection density or spacing 39

Conclusion 40

Conclusions For segment-level predictions Current method accounts for number and type of driveways along segment For intersection-level predictions Current method accounts for presence of left- and right-turn lanes, left-turn signal phasing (at signalized intersections), and right-turnon-red restrictions (at signalized intersections) 41

Conclusions Scenarios where the existing models do not perform well across sites with different access management features. • Distance to ramp terminal: • Bias not observed in predictions for signalized intersections. • CURE plots did not show substantial bias for stop-controlled intersections but did indicate potential underprediction for distances under 1, 500 feet. • Recommended CMF (adjustment factor) for 3 - and 4 -legged stop-controlled intersections is 2. 12 if there is a ramp terminal within 1, 500 feet. • CMF is intuitive: more crashes predicted as distance to ramp terminal is smaller. • Channelizing right-turn lanes: • Results inconsistent and not statistically significant at 95% confidence level for 3 - and 4 legged signalized intersections and 4 -legged stop-controlled intersections. • 3 -legged stop-controlled intersections, recommended CMF (adjustment factor) is 0. 72 for presence of a channelized right-turn lane (compared to base condition of right-turn lane with no channelization). 42

Conclusions Part C Predictive Method performs relatively well across a range of several other access management features not accounted for in existing method Sites with similar geometry, but different: • Median opening spacing • Number of median openings by type • Corner clearance 43

Conclusions Part C Predictive Method should not be used to estimate safety effect of variables related to access spacing and density • Only applicable to estimating safety performance of individual segments and intersections, assuming independence among each unit of analysis • Method does not consider potential interactions among adjacent or nearby sites (e. g. , access spacing and density) • May even produce counterintuitive results (e. g. , fewer estimated segment crashes with an increase in number of intersections along a corridor) 44

Conclusions As number of intersections in a segment increased, expected number of segment crashes often decreased • Contrary to expectations… more intersections = less conflicts • Data structure may contribute to finding (e. g. , if more intersections are present, then likely that a crash is assigned to intersection and not segment) • Given this finding, data structure for HSM predictive method is not conducive to segment-level analysis for some variables (e. g. , intersection density, spacing, etc. ) • Variables related to spacing and density better assessed at corridor level Source: VHB 45