SPR-721 JULY 2016 Data Needs for Tree Removal Crash Modi ication Factors on Arizona State Highways Arizona Department of Transportation Research Center Data Needs for Tree Removal Crash Modification Factors on Arizona State Highways SPR‐721 July 2016 Prepared by: Elizabeth Wemple, P.E. and Daniel Wu, AICP Cambridge Systematics, Inc. 555 12th Street, Suite 1600 Oakland, CA 94607 Christopher Monsere, P.E., Ph.D. Portland State University 301B Northwest Center for Engineering, Science, and Technology 1930 SW Fourth Avenue Portland, OR 97201 Published by: Arizona Department of Transportation 206 South 17th Avenue Phoenix, AZ 85007 In cooperation with US Department of Transportation Federal Highway Administration Disclaimer This report was funded in part through grants from the Federal Highway Administration, U.S. Department of Transportation. The contents of this report reflect the views of the authors, who are responsible for the facts and the accuracy of the data, and for the use or adaptation of previously published material, presented herein. The contents do not necessarily reflect the official views or policies of the Arizona Department of Transportation or the Federal Highway Administration, U.S. Department of Transportation. This report does not constitute a standard, specification, or regulation. Trade or manufacturers’ names that may appear herein are cited only because they are considered essential to the objectives of the report. The U.S. government and the State of Arizona do not endorse products or manufacturers. Technical Report Documentation Page 1. Report No. 2. Government Accession No. 3. Recipient's Catalog No. FHWA‐AZ‐16‐721 4. Title and Subtitle 5. Report Date July 2016 Data Needs for Tree Removal Crash Modification Factors on Arizona State Highways 6. Performing Organization Code 7. Authors 8. Performing Organization Report No. Elizabeth Wemple, P.E. and Daniel Wu, AICP Christopher Monsere, P.E., Ph.D. 9. Performing Organization Name and Address 10. Work Unit No. Cambridge Systematics, Inc. 555 12th Street, Suite 1600 Oakland, CA 94607 Portland State University 301B Northwest Center for Engineering, Science, and Technology 1930 SW Fourth Avenue Portland, OR 97201 12. Sponsoring Agency Name and Address 11. Contract or Grant No. Arizona Department of Transportation 206 S. 17th Avenue Phoenix, AZ 85007 SPR 000‐1(183) 721 13. Type of Report & Period Covered 15. Supplementary Notes 14. Sponsoring Agency Code Project performed in cooperation with the Federal Highway Administration. 16. Abstract The Arizona Department of Transportation (ADOT) Roadway Departure Safety Implementation Plan (RDSIP) has identified tree removal as a feasible countermeasure to reduce roadway departure crash frequency or severity. Previous ADOT work has identified locations for tree removal, and activities to conduct this work are underway. To test the safety effectiveness of removing the trees, ADOT intends to conduct statistically rigorous before‐after safety effectiveness analysis to estimate state‐specific crash modification factors (CMFs). This report provides a seven‐step data collection process, identifies and describes data needs, and recommends analytical methods for estimating the CMFs. The data needed for the analysis relate to crash type and severity, roadway characteristics, traffic volume, and tree removal. To complete the analysis, ADOT will compile the data before and after tree removal as outlined in this report and subsequently conduct the statistical analysis. The study team estimates that compiling, organizing, and managing the before‐after period database could be completed over the course of a month before and after tree removal has been completed. The statistical analyses may take approximately two months to conduct. This level of effort is an estimate and does not assume full‐time activity. In addition, the project analysis will not be conducted until three years after the tree removal has been completed at all sites. 17. Key Words 18. Distribution Statement Estimating crash modification factor, before after study for tree removal, CMF for tree removal This document is available to the U.S. public through the National Technical Information Service, Springfield, Virginia 22161 19. Security Classification 20. Security Classification 21. No. of Pages Unclassified Unclassified 55 23. Registrant's Seal 22. Price SI* (MODERN METRIC) CONVERSION FACTORS APPROXIMATE CONVERSIONS TO SI UNITS Symbol When You Know Multiply By LENGTH in ft yd mi inches feet yards miles in 2 ft 2 yd ac 2 mi 2 square square square acres square fl oz gal 3 ft 3 yd fluid ounces gallons cubic feet cubic yards oz lb T ounces pounds short tons (2000 lb) o Fahrenheit fc fl foot-candles foot-Lamberts lbf 2 lbf/in poundforce poundforce per square inch Symbol When You Know mm m m km millimeters meters meters kilometers 25.4 0.305 0.914 1.61 To Find Symbol millimeters meters meters kilometers mm m m km square millimeters square meters square meters hectares square kilometers mm 2 m 2 m ha 2 km milliliters liters cubic meters cubic meters 3 shown in m mL L 3 m 3 m grams kilograms megagrams (or "metric ton") g kg Mg (or "t") AREA inches feet yard 645.2 0.093 0.836 0.405 2.59 miles 2 VOLUME 29.57 3.785 0.028 0.765 NOTE: volumes greater than 1000 L shall be MASS 28.35 0.454 0.907 TEMPERATURE (exact degrees) F 5 (F-32)/9 or (F-32)/1.8 Celsius o lux 2 candela/m lx 2 cd/m C ILLUMINATION 10.76 3.426 FORCE and PRESSURE or STRESS 4.45 6.89 newtons kilopascals N kPa APPROXIMATE CONVERSIONS FROM SI UNITS Multiply By LENGTH 0.039 3.28 1.09 0.621 To Find Symbol inches feet yards miles in ft yd mi AREA 2 mm 2 m 2 m ha 2 km square millimeters square meters square meters hectares square kilometers 0.0016 10.764 1.195 2.47 0.386 square square square acres square inches feet yards miles 2 in 2 ft 2 yd ac 2 mi VOLUME mL L 3 m m3 milliliters liters cubic meters cubic meters 0.034 0.264 35.314 1.307 g kg Mg (or "t") grams kilograms megagrams (or "metric ton") o Celsius fluid ounces gallons cubic feet cubic yards fl oz gal 3 ft yd3 ounces pounds short tons (2000 lb) oz lb T MASS 0.035 2.202 1.103 TEMPERATURE (exact degrees) C 1.8C+32 Fahrenheit o foot-candles foot-Lamberts fc fl F ILLUMINATION lx 2 cd/m lux 2 candela/m N kPa newtons kilopascals 0.0929 0.2919 FORCE and PRESSURE or STRESS 0.225 0.145 poundforce poundforce per square inch lbf 2 lbf/in *SI is the symbol for the International System of Units. Appropriate rounding should be made to comply with Section 4 of ASTM E380. (Revised March 2003) CONTENTS EXECUTIVE SUMMARY ......................................................................................................................... 1 CHAPTER 1. INTRODUCTION ................................................................................................................ 3 CHAPTER 2. A REVIEW OF EXISTING LITERATURE ON CMFS .................................................................. 5 LITERATURE REVIEW SUMMARY ............................................................................................................ 5 LITERATURE REVIEW .............................................................................................................................. 6 REVIEW OF RELEVANT ADOT MATERIALS ............................................................................................ 17 CHAPTER 3. PROPOSED METHOD OF DATA AND NEEDS ANALYSIS ..................................................... 19 OVERVIEW OF ANALYSIS METHODS ..................................................................................................... 19 EVALUATION TO SELECT THE APPROPRIATE ANALYSIS METHOD ........................................................ 20 CHAPTER 4. DATA NEEDS AND DATA COLLECTION FRAMEWORK ....................................................... 25 DATA COLLECTION FRAMEWORK ......................................................................................................... 26 DATA NEEDS ......................................................................................................................................... 28 CHAPTER 5. CONCLUSION .................................................................................................................. 35 REFERENCES....................................................................................................................................... 37 APPENDIX A. NEW YORK STATE DOT ACCIDENT REDUCTION FACTORS ............................................... 41 APPENDIX B. DATA USED IN ASSESSING EFFECTIVENESS OF TREE OR FIXED OBJECT REMOVAL PROJECTS ........................................................................................................................................... 43 APPENDIX C. ADOT CRASH REPORT FORM ......................................................................................... 45 i LIST OF FIGURES Figure 1. Distribution of Fatal Arizona Crashes (2008–2012) ....................................................................... 3 Figure 2. Distribution, by Functional Class, of Fatal Crashes with Trees/Shrubs as First Harmful Event (2008–2012) ......................................................................................................................... 4 Figure 3. Fixed Object CMF for Various Offsets and Spacing Along Two‐Lane Urban and Suburban Arterials ........................................................................................................................................ 11 Figure 4. Cumulative Percent Average Setback in Segment Analysis at 30 Miles per Hour ....................... 12 Figure 5. Expected Severity Distribution by Lateral Clearance ................................................................... 13 Figure 6. Flow Chart for Study Design Selection ......................................................................................... 21 Figure 7. Data Collection Flow Chart .......................................................................................................... 26 LIST OF TABLES Table 1. Total Fatal Crashes and Fatal Crashes with Trees/Shrubs as First Harmful Event (2008 to 2012) .............................................................................................................................................. 3 Table 2. Existing CMFs Related to Tree Removal .......................................................................................... 8 Table 3. Expected Reduction in Tree Crashes related to Tree Line (Crash Reduction Factors)a ................... 9 Table 4. Crash Type Considered in Tree Removal CMF Estimates .............................................................. 14 Table 5. Ranges of Traffic Volume Considered in Tree Removal CMF Estimates ....................................... 15 Table 6. General Characteristics Considered in Tree Removal CMF Estimates .......................................... 16 Table 7. Summary of Time Period and Crash Data in EB Before‐After Analysis ......................................... 23 Table 8. Rounded Lane Width for Roadway Segmentation ........................................................................ 29 Table 9. Rounded Shoulder Width for Roadway Segmentation ................................................................. 30 Table 10. Example Field Data Collection Form ........................................................................................... 33 Table 11. Data Organization Example ......................................................................................................... 34 ii LIST OF ACRONYMS AADT ............................. Annual average daily traffic AASHTO ......................... American Association of State Highway and Transportation Officials ADOT ............................. Arizona Department of Transportation ADT ................................ Average daily traffic CMF ............................... Crash modification factor CPM ............................... Crash prediction model CRF ................................ Crash reduction factor DOT ............................... Department of Transportation EB .................................. Empirical Bayes FB .................................. Full Bayes FHWA ............................ Federal Highway Administration FIS .................................. Feature Inventory System FO .................................. Fixed object FY ................................... Fiscal year GIS ................................. Geographic Information System HSIP ............................... Highway Safety Improvement Program HSM ............................... Highway Safety Manual LIDAR ............................. Light Detection and Ranging MwRSF .......................... Midwest Roadside Safety Facility NCHRP ........................... National Cooperative Highway Research Program NYSDOT ......................... New York State Department of Transportation RDG ............................... Roadside Design Guide RDSIP ............................. Roadway Departure Safety Implementation Plan SD .................................. Standard deviation SPF ................................. Safety performance function TRB ................................ Transportation Research Board TRIS ............................... Transportation Research Information Services UNC ............................... University of North Carolina VHB................................ Vanasse Hangen Brustlin iii iv EXECUTIVE SUMMARY In recent years, the number of shrub‐ and tree‐related crashes has increased in Arizona. In 2012, fatal crashes with trees and shrubs as the first harmful event made up more than 4 percent of all statewide fatal crashes. The Arizona Department of Transportation (ADOT) Roadway Departure Safety Implementation Plan (RDSIP) has identified tree removal as a feasible countermeasure to reduce roadway departure crash frequency or severity. For this reason, ADOT has started implementing projects to reduce the number of trees in the recovery area along state highways. At the same time, states are placing more emphasis on measuring the effectiveness (i.e., change in crash frequency or crash severity) of safety investments. Safety effectiveness is measured in terms of crash modification factors (CMFs). A CMF is a multiplicative factor used to estimate the change in crash frequency or severity after a specific treatment is implemented at a specific site. CMFs are calculated using safety before‐after studies. These studies use statistical methods to compare crash frequency, type, and severity three years before and three years after a project is implemented to identify: 1) whether crash frequency, type, or severity changes; and 2) how much of the change can be attributed to the project. This research project provides ADOT with a data collection plan for conducting a before‐ after study related to the tree removal and ultimately, if possible, estimating a CMF associated with tree removal in Arizona. ADOT will need to develop a database with crash, tree removal, roadway characteristics, and traffic volume data at each tree removal site in the state in order to conduct the before‐after analysis. Data will need to be collected at each site and compiled for three years before the trees are removed and again three years after the trees are removed. This report defines the data needs; outlines a seven‐step framework for the data collection, compilation, and evaluation (Figure 7); provides a database template for organizing the data (Table 11); and discusses the likely statistical method for analysis (Chapter 3). The data needed for the analysis relate to:  Crash type and severity (e.g., number of crashes, number of crashes by severity, and crash type)  Roadway characteristics and traffic volume (e.g., terrain, roadway functional classification, number and width of lanes, width and type of shoulders, presence of rumble strips/stripes, pavement edge type/condition, speed limit, and traffic volume)  Roadside characteristics (e.g., guardrail presence and type, fences, walls, signs, and call boxes)  Tree removal (e.g., date of tree removal, average tree density before and after removal, and actual recovery area distance before and after removal) To complete this analysis, ADOT will need to compile the preceding data into a spreadsheet or database comparable to the template provided in this report. The crash, roadway, and traffic volume data are largely available from existing ADOT databases or Google Earth, if needed. These data can be compiled and organized at any time by ADOT staff members who are familiar with state databases and spreadsheets. As part of the tree removal projects, ADOT will need to collect a small amount of 1 information related to recovery area distances and tree density before and after removal. Chapter 4 includes a sample data collection form. The after‐period crash, traffic volume, and roadway data will be compiled three years after the tree removal data are collected and recorded in the database. After all tree removal projects have been completed and data have been compiled in the database, ADOT will need staff or a consultant with statistical analysis expertise to conduct the before‐after analyses. The study team estimates that compiling, organizing, and managing the before‐after period database could be completed over the course of a month before and after tree removal has been completed. The statistical analyses may take approximately two months to conduct. This level of effort is an estimate and does not assume full‐time activity. In addition, the project analysis will not be conducted until three years after the tree removal has been completed at all sites. 2 CHAPTER 1. INTRODUCTION In recent years, the number of shrub‐ and tree‐related crashes has increased in Arizona. As shown in Table 1, fatal crashes with trees and shrubs as the first harmful event made up approximately 2 to 4 percent of all fatal crashes in Arizona between 2008 and 2012. Although not tabulated in this research, the frequency of the crash type increases when severe injury crashes are also considered. Table 1. Total Fatal Crashes and Fatal Crashes with Trees/Shrubs as First Harmful Event (2008 to 2012) Trees/shrubs fatal crashes All fatal crashes Trees/shrubs fatal crashes as a percentage of all fatal crashes 2008 15 843 1.8 2009 14 709 2.0 2010 22 695 3.2 2011 28 755 3.7 2012 32 742 4.3 As shown in Figure 1, rural roadways have a higher percentage of fatal crashes with trees and shrubs as the first harmful event compared to all fatal crashes on rural roadways. Between 2008 and 2012, 66 percent of fatal crashes with trees/shrubs as the first harmful event occurred on rural roadways. In contrast, 46 percent of all fatal crashes in Arizona occurred on rural roadways. Tree‐ and shrub‐related crashes are concentrated on certain roadway functional classes. As shown in Figure 2, the two urban functional classifications with the greatest share of fatal crashes with trees/shrubs as the first harmful event are local road or street, and other principal arterials. The two rural functional classifications with the greatest share of fatal crashes with trees/shrubs as the first harmful event are principal arterial: interstate and major collector. Fatal Crashes with Tree/Shrub as the First Harmful Event All Fatal Crashes Source: National Highway Traffic Safety Administration, Fatality Analysis Reporting System, Arizona Crash Data, http://www.nhtsa.gov/FARS Figure 1. Distribution of Fatal Arizona Crashes (2008–2012) 3 Source: National Highway Traffic Safety Administration, Fatality Analysis Reporting System, http://www.nhtsa.gov/FARS Figure 2. Distribution, by Functional Class, of Fatal Crashes with Trees/Shrubs as First Harmful Event (2008–2012) To address these issues, the Arizona Department of Transportation (ADOT) is executing a long‐term roadside tree removal program as part of the ADOT Highway Safety Improvement Program (HSIP). This work is being conducted as part of the state Roadway Departure Safety Implementation Plan (RDSIP), in which tree removal was identified as a feasible countermeasure to reduce roadway departure crash frequency or severity. Recognizing the importance of safety evaluation and the opportunity to estimate Arizona‐specific crash modification factors (CMFs), ADOT initiated this project to develop data needs, a data collection plan, and optional analysis methods for developing CMFs based on the tree removal program. This report documents the anticipated safety effectiveness analysis methodology, data needs, and data collection framework. The study team expects that, after this research project has been completed, ADOT will compile the site and crash data along the tree removal segments before and after tree removal, and subsequently conduct the before‐after analysis to estimate CMFs. 4 CHAPTER 2. A REVIEW OF EXISTING LITERATURE ON CMFS This chapter presents a review of recent research and ADOT‐specific documents that are relevant to estimating the safety effects of tree removal on Arizona state highways (i.e., tree removal CMFs). The results of this review inform the development of the data collection framework and safety analysis methodology for the ADOT tree removal program, as detailed in Chapters 3 and 4. This chapter has two components:  A literature review of existing CMF research, with a focus on data needs for assessing the effectiveness of tree removal and CMF development methodologies  A review of relevant ADOT materials, such as the Arizona RDSIP, documents relevant to ADOT policies on tree removal, ADOT’s crash report form, and data already collected by ADOT LITERATURE REVIEW SUMMARY The research review provided the following key insights for identifying data needs and developing a data collection plan for ADOT:  Few research efforts have been related to developing CMFs for tree removal.  Projects that have identified CMFs have considered characteristics such as: o Lateral distance/clear zone to the fixed object (e.g., tree or utility pole) o Density of the fixed objects (e.g., utility poles) along the roadway o Traffic volume o Roadway functional classification and highway access o Roadway characteristics, including functional classification; posted speed; highway access; and shoulder width, alignment, and grade  Peng et al. 2012 showed a small decrease in fatal and serious injury crashes as the lateral clearance increased. For instance, the analysis suggested that the probability of a fatal crash decreases from 6.9 percent to 4 percent with increasing lateral clearance from 10 feet to 40 feet.  The 2011 American Association of State Highway and Transportation Officials (AASHTO) Roadside Design Guide (RDG) identifies a fixed object as a tree (or a group of trees/shrubs) with an existing or mature diameter greater than 4 inches (AASHTO 2011).  Typical ADOT procurement contracts for tree removal projects include assumptions about the density of trees/shrubs being removed. The categories of density are: o Low tree density is fewer than 16 trees per acre o Medium tree density is 17 to 32 trees per acre o High tree density is more than 32 trees per acre  Arizona’s crash report form includes crash type, location description, and severity information that will be useful for the analysis.  It will likely be necessary to gather additional information, such as the characteristics (e.g., diameter and density) of the tree that was hit, and traffic volume from field data collection, other ADOT departments, or other data sources. 5 LITERATURE REVIEW Resources The project team acquired CMF literature from two sources: 1) the Federal Highway Administration (FHWA) listserv request; and 2) a list of relevant accident, clear zones, and treatment studies identified in a presentation by the Midwest Roadside Safety Facility (MwRSF) “Crash Risks, Location, and Treatment of Roadside Trees.” (Ronald K. Faller, Ph.D., P.E., unpublished data, June 17, 2013). In September 2013, the FHWA’s Arizona Division Office sent out an information request about tree removal CMFs to a listserv of state highway engineers. These engineers provided relevant state‐specific research projects, such as CMFs for tree removals to improve sight distance and CMFs associated with maintaining clear zone minimum requirements. Many CMFs from these projects also have been incorporated in the FHWA CMF Clearinghouse. The studies within the MwRSF’s presentation were identified by state highway engineers who had been contacted through the FHWA’s information request. The MwRSF presentation’s list of studies was cross‐ checked with the Transportation Research Information Services (TRIS) and online web searches to verify that it contained current, comprehensive information. These relevant reports and studies include the National Cooperative Highway Research Program (NCHRP) Report 500 series Volume 3: A Guide for Addressing Collisions with Trees in Hazardous Locations, and Volume 8: A Guide for Reducing Collisions with Utility Poles, CMFs from Australia and Europe on modifying clear zone widths, and the NCHRP 17‐ 54: Draft Interim Report Consideration of Roadside Features in the Highway Safety Manual research report. The literature review focused on the following questions:  What CMFs related to tree removal are currently available?  Is there evidence that tree removal projects change crash severity?  Which data are used to estimate CMFs for tree removal?  What methods are used to estimate CMFs? Existing CMFs The project team reviewed the published research to identify existing CMFs related to tree removal. The objective was to understand the methodologies used to estimate tree removal CMFs and document the range of CMF values and standard deviation (SD) currently estimated. The project team found very little research that specifically estimates CMFs for tree removal. The team identified only three recent studies: Hovey and Chowdhury (2005), New York State DOT (2012), and Pennsylvania Department of Transportation (DOT). The Pennsylvania DOT research is included in the NCHRP 500 Report, Volume 3. (Transportation Research Board 2003) 6 However, trees are not the only roadside object that vehicles can collide with. The literature review also identified CMFs for safety improvements comparable to removal: removing and relocating fixed objects (e.g., light poles, utility boxes, drainage structures, and sign posts); increasing clear zone width; and changing roadside fixed object density. Table 2 summarizes all of the countermeasures, CMFs, and sources identified in the project research. The research can be grouped into three main categories as follows. 7 Table 2. Existing CMFs Related to Tree Removal A B C D E Countermeasure Remove or relocate fixed object (e.g., utility poles, trees, guardrails, sign supports, and fire hydrant) Increase distance to roadside features from:  3.3 ft to 16.7 ft  16.7 ft to 30 ft Change clear zone from:  <2 m (6.6 ft) to >8 m (26.2 ft)  Between 2 m (6.6 f) and 4 m (13.1 ft) to >8 m (26.2 ft)  Between 4 m (13.1 ft) and 8 m (26.2 ft) to >8 m (26.2 ft) Increase lateral clearance from 10 to 40 ft:  On horizon curve sections  On tangent sections Clear the minimum clear zone (30 ft) of fixed objects F Increase the clear zone by removing trees G Remove or trim shrubs/ trees to improve sight distance (unspecified intersection vs. stopping) H Crash Type All Applicability Crash Severity All All All Run‐off‐road Run‐off‐road, single vehicle Roadway Description Various roadway types (not specified) CMF a (Standard Deviation) 0.62 (0.103) Two‐lane rural roadway Fatal, serious injury, minor injury   0.78 (0.01) 0.56 (0.01)  1. 0.46 (N/A)  2. 0.63 (N/A)  3. 0.79 (N/A) Rural Fatal, serious injury, minor injury Rural, two‐lane undivided Fixed object (e.g., bridge, culvert, mailbox, pole, and tree) All Tree crashes All Various (interstate, US primary, state primary, secondary, and county) Not specified All All  Average daily traffic (ADT) per lane <5000  ADT per lane >5000 Two‐lane and multilane urban and suburban arterial roads   b Star Quality Rating Three out of five stars (per the CMF Clearinghouse) Source (Hovey and Chowdhury 2005) Five out of five stars (per the CMF Clearinghouse) (Elvik 2004). This CMF is also included in the Highway Safety Manual (HSM). Three out of five stars (per the CMF Clearinghouse) (Jurewicz and Pyta 2010) Three out of five stars (per the CMF Clearinghouse) (Peng et al. 2012) Two out of five stars (per the CMF Clearinghouse) (Ogle et al. 2009) b b 0.68 (N/A) 0.49 (N/A) c 0.024 (0.1075) Varies from 0.23 to 0.89, depending on before/after tree line (see Table 3 below)  0.57 (N/A)  0.68 (N/A) d Not specified Transportation Research Board 2003 (Pennsylvania DOT Tree Crash Reduction Factors) The analysis showed these reductions as statistically significant. New York State Department of Transportation 2012 d Not specified Zegeer and Cynecki 1984 (as CMF varies by offset referred to in HSM/NCHRP of fixed objects along 17‐54 Interim Report/ roadway, fixed object spacing (density), and NCHRP 500, Volume 8) roadway type (see Equation 2 below). a Unadjusted SD from original study. The CMF Clearinghouse also has an adjusted SD derived through the Highway Safety Manual (HSM) CMF inclusion process. b Countermeasure reported in the CMF Clearinghouse as a decrease in clearance zone width. For comparison with other countermeasures, this evaluation inverted the CMF. c This is based on an odds ratio calculation that the odds of a site having fixed object crashes are 42 times higher if minimum clear zone is not met. d Countermeasure reduction translated into CMF. Roadside fixed object density CMF Utility pole collisions All 8 Increasing Clear Zone or Lateral Clearance Clear zone or lateral clearance is the distance from the roadside to nearby obstructions and fixed objects, such as trees, utility poles, mailboxes, and culverts. Table 2 (Rows A, B, C, D, E, and F) shows that various studies have estimated CMFs or crash reduction pertaining to fixed object removal in the clear zone or clear zone expansion. The NCHRP 500 Report, Volume 3 (Transportation Research Board 2003) provides a guide for addressing tree collisions in hazardous locations. This report refers to research completed by Pennsylvania Department of Transportation (PennDOT). PennDOT’s research relates crash reduction to the tree line’s distance from the traveled way as shown in Table 3. The PennDOT findings are reported in crash reduction factors (CRFs). A CRF is equal to one minus the CMF. Table 3 shows that a greater distance between the tree line and the traveled way before and after tree removal is related to greater crash reduction. Details related to study design, sample size, and years of data were not available in the NCHRP 500 report. Table 3. Expected Reduction in Tree Crashes related to Tree Line (Crash Reduction Factors)a Source: NCHRP 500 Report, Volume 3, Appendix 8 (Transportation Research Board 2003) Transportation Resource Board (2003) a Tree line means distance of tree from travelled way. 9 Other studies provide CMFs or CRFs for all fixed objects, and not just trees. Two studies identified in the FHWA CMF Clearinghouse—Hovey and Chowdhury (2005) (Table 2, Row A) and Elvik and Vaa (2004) (Table 2, Row B)—provide CMFs for all crash types that range from 0.56 to 0.78 based on increasing lateral clearance. The clear zone increase described in Elvik and Vaa (2004) has the highest research quality of all available CMFs. Although this CMF is based on European roadway safety data, it has a Clearinghouse quality rating of five out of five stars, which indicates the evaluation was statistically rigorous, with large samples, diverse sites, and a small standard error. Three of the sources reviewed focused on two crash types: run‐off‐road and fixed object crashes. Two studies from the CMF Clearinghouse—Jurewicz and Pyta (2010) (Table 2, Row C) and Peng et al. (2012) (Table 2, Row D)—found CMFs for run‐off‐road crashes between 0.46 and 0.79 based on differences in lateral clearance. Both studies generated CMFs with FHWA Clearinghouse quality ratings of three out five stars. This rating indicates moderately rigorous study design with moderate sample size and limited site diversity. Both studies are based on rural roadways. Jurewicz and Pyta (2010), based on Australian driving conditions, found that larger changes in clear zone had greater safety benefits. For example, when the clear zone changed from fewer than 6.6 feet to greater than 26.2 feet, the CMF was estimated at 0.46; when the clear zone was changed from between 13.1 and 26.2 feet to greater than 26.2 feet, the CMF was estimated as 0.79. Ogle et al. (2009) also focused on fixed object crashes and estimated that implementing a minimum clear zone of 30 feet from the roadway yields a CMF for fixed object crashes of 0.024. Although this CMF is lower than the other CMFs shown in Table 2, it should be noted that the study has a lower Clearinghouse quality rating of two stars out of five stars. Removing and Trimming Trees to Improve Sight Distance New York State DOT (NYSDOT) calculated CRFs at state highway locations where trees were removed or trimmed based on identified accident patterns (Table 2, Row G). NYSDOT did not specify whether these locations were intersections or roadway segments. NYSDOT calculated the CRFs for different crash types, roadway characteristics, and driving conditions, as well as crash severity. The CRF for all crashes is 43 percent for roadways (CMF = 0.57) with annual average daily traffic (AADT) volumes lower than 5000 per lane. The CRF for all crashes on roadways with AADT greater than 5000 per lane is 32 percent (CMF = 0.68). NYSDOT determined these two reduction factors to be statistically significant based on a 99.9 percent level of confidence threshold. NYSDOT’s significance equation used to determine the maximum expected deviations is shown in Eq. 1 where RF = reduction factor, T = the threshold value required to gain significance expressed as a percent, and B = the adjusted number of before accidents by severity category. T = (((232.6 * (SQRT(B –.16))) – 35)/B) (Eq. 1) NYSDOT also estimated CRFs for removing and trimming trees to improve sight distance for a variety of crash types and severities. This information is summarized in Appendix A. 10 Roadside Fixed Object Density Zegeer and Cynecki (1984) evaluated utility pole collisions based on roadside fixed object density (Table 2, Row H). This study derived a CMF equation for urban and suburban arterial roadway segments with different numbers of lanes. This CMF equation is also included in the HSM and the NCHRP 17‐54 draft interim report (TRB 2013). The equation has a base condition of no roadside fixed objects. The CMF is based on fixed object offset factors derived from offset distance to fixed objects (foffset), fixed object density from both sides of the road (DFO), and fixed object collisions as a proportion of total crashes (PFO), as shown in Eq. 2. CMF2r = foffset x DFO x PFO + (1.0 – PFO) (Eq. 2) According to the research, inputs for the CMF are:  Point objects that are at least 4 inches in diameter and do not have a breakaway design.  Continuous objects that are not obstructed by the point objects; these objects are counted as one point for every 70 feet of length. Note that fixed objects in the medians of divided arterials are not considered in the CMF. To illustrate, Figure 3 shows the combined effect of the fixed object offset and density on two‐lane, undivided urban and suburban arterials. The figure’s X axis represents the distance between fixed objects in feet. The figure’s legend indicates the various CMF trend lines based on offset to fixed objects (2, 10, 20, and 30 feet). Note that all trend lines assume the proportion of fixed object collisions for two‐lane, undivided urban and suburban arterials to be 0.059. In general, the CMF is higher when objects are spaced closer together and when they are closer to the road. Source: NCHRP 17‐54: Draft Interim Report (Transportation Research Board 2013) Figure 3. Fixed Object CMF for Various Offsets and Spacing Along Two‐Lane Urban and Suburban Arterials 11 Related CMFs There are other CMFs related to roadside fixed objects. For example, numerous studies have evaluated before‐after changes in CMFs pertaining to urban or suburban landscaping or road geometry improvements involving trees. Hallenbeck et al. (2013) examined before‐after total, fatal, and tree crashes for five urban and suburban roadway segments that received landscape median treatments. The study showed that the presence of small trees in the median did not statistically increase crash rates, crash severity, or injury‐crash rates. A 2006 before‐after study on 10 Texas roadway segments with landscape improvements showed that only one of 10 sites experienced a significant reduction (83 percent) in tree collisions after landscape treatments (Mok et al. 2006). However, because the particular roadway segment also had multiple modifications and landscape treatments, it was difficult to conclude the source of the crash reduction. Maze et al. (2008) also evaluated the benefits of providing a 10‐foot clear zone along 11 urban curbed streets in Des Moines and Waterloo, Iowa. The study conducted a cumulative percent analysis to assess where the majority of the fixed object crashes occurred in relation to setback measurement. Figure 4 shows an example of the cumulative percent of fixed object crashes and their average setback for a speed limit of 30 miles per hour. The analysis suggested that in an urban setting, once the clear zone exceeds 5 feet, observed fixed object crashes are minimized. The cumulative percent analysis found that as the speed limit or ADT increased, the setback distance where 90 percent of fixed object crashes occur increased by only 1 or 2 feet, on average. Source: Maze et al. (2008) Figure 4. Cumulative Percent Average Setback in Segment Analysis at 30 Miles per Hour 12 Crash Severity Modification Peng et al. (2012) performed a multinomial regression analysis to estimate the relationship between road departure crashes and lateral clearance, side slope condition, and driveway density on rural two‐ lane roads. The study used the KABCO scale for defining injuries, where K = fatal injury, A = incapacitating injury, B = nonincapacitating injury, C = possible injury, and O = no injury. The researchers used a Light Detection and Ranging (LIDAR) gun to measure distances from the roadside to nearby obstructions and fixed objects, providing a record of lateral clearance. The distance measurement included the shoulder and is consistent with the AASHTO definition of roadside clearance. As shown in Figure 5, the analysis suggests that the probability of a fatal crash decreases from 6.9 percent to 4 percent, and the probability of an incapacitating injury crash decreases from 8.8 percent to 6.4 percent as lateral clearance increases from 10 feet to 40 feet, respectively. Source: Peng et al (2012) Figure 5. Expected Severity Distribution by Lateral Clearance Geedipally et al. (2013) similarly developed severity distribution functions for freeway roadside crashes as a function of presence of inside and outside barriers, traffic volume, presence of inside and outside rumble strips, lane width, and segment length with curve. This methodology has been integrated into the freeway chapter of the AASHTO HSM (AASHTO 2010). Data Needs As part of the literature review, the study team examined the types of data used in the research to develop the CMFs or CRFs. Appendix B includes a full list and description of the data. Overall, data used in research can be grouped into five main categories: crash data, tree characteristics (e.g., diameter, density, and offset), traffic volume, highway functional classification and access type, and other roadway characteristics (e.g., roadway alignment and grade, shoulder type, and shoulder width). Not all data 13 were used as inputs or factors for CMF functions. Some studies used these data as categories for site comparisons, and then conducted evaluations of crash outcomes. Other studies used these data as dependent and independent variables in the crash estimation models. The study team also reviewed the AASHTO RDG (AASHTO 2011) to understand national guidance about clear zone guidance. Crash Data All studies in this review used crash type data, such as the total number of crashes and run‐off‐road crashes, as a dependent variable that is affected by various factors (see Table 4). For instance, NYSDOT’s CRFs calculation evaluated how trimming and removing bushes and trees would reduce the number of total crashes, overtaking, right angle, fixed object (tree/hydrant/other), and run‐off‐road crashes (NYSDOT 2012). NYSDOT found the number of all these crash types to be significantly reduced by bush and tree trimming/removing on high‐volume roads (an ADT greater than 5000 vehicles per lane). For lower‐volume roads (fewer than 5000 vehicles per day per lane), only the number of all crash types and right‐angle crashes was significantly reduced by tree trimming and removal. Table 4. Crash Type Considered in Tree Removal CMF Estimates Crash Type 1) Total crashes; and 2) run‐off‐road casualty crashes Overtaking, right‐angle, fixed object (tree/hydrant/other), all fixed object, and run‐off‐road crashes Type of fixed object crashes (trees, poles, culvert, mail box, bridge, guard rail, and other) Study Dependent variable for most studies New York State DOT 2012 Ogle et al. 2009 Some studies also specified crash severity as a dependent variable of different factors. Typically, these studies broke severity into the following categories: fatal, injuries, and property damage only. Some studies focused on one type of crash severity. For instance, Jurewicz and Pyta (2010) examined how clear zone widths affect run‐off‐road fatal and injury crashes. Other studies, such as Peng et al. (2012), organized crash data into the five KABCO categories: fatal, incapacitating injury, nonincapacitating injury, possible injury, and property damage only. Crash data precision issues, such as accuracy of crash locations or reported crash severity, were not discussed in the reviewed literature. Tree Characteristics (Diameter, Density, Number of Trees, and Offset from Roadways) The 2011 AASHTO RDG identifies trees with an existing or expected mature diameter greater than 4 inches as potential fixed objects (AASHTO 2011). A number of trees or shrubs within close proximity to each other also may be considered to have the same effect as a tree with a diameter greater than 4 inches. The RDG specifies that large trees and shrubs should be removed from the clear zone. The width of this clear zone is a function of highway speeds, traffic volume, and roadside slopes. 14 Traffic Volume Much of the literature also used ADT or AADT data for traffic volume exposure. The traffic volume ranges and studies, shown in Table 5, have different categorical breakdowns by ADT. For instance, New York State DOT (2012) compared state highway locations with AADT of greater than 5000 vehicles per lane to those with AADT of fewer than 5000 vehicles per lane. In the Hovey and Chowdhury (2005) CRF development study, the authors applied the empirical Bayes (EB) method to estimate total crashes using total ADT and truck ADT as variables, and they estimated fatal and injury crashes based on total ADT. Table 5. Ranges of Traffic Volume Considered in Tree Removal CMF Estimates Traffic Volume Range AADT: 1500 to 6000 and 6000+ One‐way AADT: <1200 and >1200 AADT >5000 vs. AADT <5000 Average daily total and truck volumes as variables in the CRF functions Study Maze et al. 2008 Jurewicz and Pyta 2010 New York State DOT 2012 Hovey and Chowdhury 2005 Roadway Functional Classification/Highway Access Type Hovey and Chowdhury (2005) also used functional class as a parameter to forecast total crashes and fatal and injury crashes. The variable coefficients differ between 12 roadway functional classes (six urban and six rural functional classes). The study also included a variable for highway access type that has different coefficients for three access types: no access control, limited access control, and full access control. Other Roadway Characteristics Studies in this category also compare crash reduction to other roadway characteristics (see Table 6). Hovey and Chowdhury (2005) included shoulder widths as a variable influencing total crashes and fatal and injury crashes. Jurewicz and Pyta (2010) also considered crash likelihood in different traffic lanes and with paved shoulder widths. The study found that road sections with fewer than 3.5 meters (11.5 feet) each way (or 7 meters [23 feet] across) have a run‐off‐road crash likelihood 1.2 times higher than roads with seal width of more than 7 meters (23 feet) across, although this relationship was found to be statistically insignificant. 15 Table 6. General Characteristics Considered in Tree Removal CMF Estimates Characteristic Shoulder widths as variables in the CRF functions    Curve radius as a part of a run‐off‐road crash prediction model: <600 meters (1969 feet), 600 to 1500 meters (1969 to 4921 feet), and >1500 meters (4921 feet) Grade (%) as part of a run‐off‐road crash prediction model: negative, positive, or zero Traffic lane plus paved shoulder width as part of a run‐off‐ road crash prediction model: <3.5 meters (11.5 feet) and >3.5 meters (11.5 feet)  Straight roadway level, grade, and hillcrest  Curve roadway level, grade, and hillcrest Speed limit: At 30, 35, and 40 miles per hour Study Hovey and Chowdhury 2005 Jurewicz and Pyta 2010 CMF Clearinghouse (Ogle et al., 2009) Maze et al. 2008 Two of the review’s studies also looked at road alignment and grades. Jurewicz and Pyta (2010) looked at curve radius as part of a run‐off‐road crash prediction model. Curve radii were divided into three categories: less than 600 meters (1969 feet), 600 to 1500 meters (1969 to 4921 feet), and greater than 1500 meters (4921 feet). The study found that crash likelihood is 2.4 times higher on curves with a radius less than or equal to 600 meters (1969 feet) than on curves with a radius of more than 1500 meters (4921 feet or relatively straight). Similarly, for straight and curved roads, the study further recorded whether the roadway is at a level, a grade, or a hillcrest. The study compared run‐off‐road predictions between negative grade roadways and positive or zero grade roadways. The study found that roadway sections with a downhill grade have a crash likelihood 1.3 times higher than road sections with an uphill grade or no grade. Maze et al. (2008) also evaluated how the setback distance, where 90 percent of fixed object crashes occur, would differ under different speed limits. However, the study only tested three speed limits (30, 35, and 40 miles per hour), and the relationship between the number of fixed‐object crashes per year to the speed limit was inconclusive. Driving conditions (i.e., environmental characteristics) could also affect the relationship between safety improvements and crash reduction. NYSDOT documented crash conditions to evaluate whether trimming and removing trees reduced crashes on dry, wet, and snow/ice/slush roads (NYSDOT 2012). Based on the study’s available data, trimming and removing trees significantly reduced the number of crashes in all three driving conditions for roads with AADT greater than 5000 vehicles per lane, while the crash reduction in all three driving conditions was not statistically significant for roads with AADT greater than 5000 vehicles per lane. 16 REVIEW OF RELEVANT ADOT MATERIALS ADOT provided background information, which was reviewed to develop a deeper understanding of:  The existing tree removal program  Typical data likely to be available from the state crash report form  Typical tree removal project procurement materials and the data likely to be developed for or during a tree removal project ADOT HSIP‐Funded Tree Removal Project H8206 and Arizona Roadway Departure Safety Implementation Plan, FHWA Office of Safety On June 15, 2010, ADOT received a determination of eligibility to use HSIP funding for systemic tree removal, with an eligibility amount of $9.7 million. In June 2012, the FHWA Office of Safety completed the Arizona RDSIP, which included tree removal as one of the countermeasures. The RDSIP identified additional roadway segments for tree removal. On October 16, 2012, the FHWA Arizona Division granted HSIP eligibility to add the RDSIP identified segments to the June 15, 2010, approved list. The revised total HSIP funding for tree removal is $13 million, which covers approximately 700 roadway miles of tree removal. ADOT has retained a contractor to prepare scoping documents and cost estimates based on recently completed field sampling of tree density (number of trees per acre) on I‐19, SR 87, and I‐8. The final steps are determination of environmental requirements and development of the implementation schedule for final plans and construction based on input from the ADOT Environmental Planning Group. Some of the tree removal projects may be executed as standalone projects; others will be executed as part of other programmed projects (e.g., pavement preservation projects). The first standalone project is programmed at $2 million for construction in fiscal year (FY) 16. It is anticipated that a second standalone project will be designed and programmed at $4 million for construction in FY17, and a third at $4 million in FY18. ADOT Crash Report Form The existing ADOT Crash Report Form (Appendix C) was reviewed to understand the typical data available in a crash record. It is desirable that limited unusual or additional data be required as part of the process to estimate a CMF. The crash report provides the following information that may be useful:  Location, including roadway name, site specific location, and urban or rural designation  Injuries per crash and severity  First harmful event, identifying collision with a tree, bush, or stump  Light, weather, road surface, and road grade condition  Manner of crash impact, including single vehicle, angle, left turn, rear end, head on, sideswipe (same and opposite direction), rear to side, and rear to rear collision 17    Traffic unit maneuver/action, which includes an item for negotiating a curve Roadway alignment, including straight, curve left, curve right, and unknown Sequence of events, such as run‐off‐road (left or right) and collision with fixed object, including a tree, brush, or stump The following information may be needed, but is not included in the crash reports:  Roadway offset distance to the fixed object crash – This information may be included in the sketch or description provided by the reporting officer.  Characteristics (diameter and density) of the tree or shrub that was hit.  Roadway characteristics, which could include number of lanes, posted speed, shoulder type and width, and side slope. This information could possibly be gathered from other ADOT data sources.  Traffic volume – These data are also likely available through other ADOT data sources. Sample Tree Removal Project Procurement Materials – Lake Mary Road ADOT provided the project team with sample ADOT tree‐cutting project procurement materials, which the team reviewed in order to understand typical project descriptions and activities and typical tree removal pricing. In the sample project, the pricing for tree removal is set at a price per acre as a function of low, medium, and high tree density:  Low tree density is fewer than 16 trees per acre.  Medium tree density is 17 to 32 trees per acre.  High tree density is more than 32 trees per acre. In a typical tree removal project, ADOT specifies the recovery area for its contractors. The recovery area on Lake Mary Road is 30 feet from the adjacent shoulder stripe (or 42 feet from the roadway centerline if no shoulder stripe exists). The Lake Mary project also included a thinning zone between 30 feet and 100 feet (in different situations) from the shoulder stripe or right‐of‐way limits. In the thinning zone, trees are specified on an item‐by‐item basis for removal by ADOT. The ADOT materials also provide information on how the remaining tree stumps are treated as a function of diameter and slope roadside. 18 CHAPTER 3. PROPOSED METHOD OF DATA AND NEEDS ANALYSIS This chapter presents an overview of various analysis methods for developing CMFs based on the tree removal program. Based on a series of criteria developed in the FHWA’s A Guide to Developing Quality Crash Modification Factors (Gross et al. 2010), it is anticipated that the most appropriate analysis method will be the EB method’s before‐after analysis. However, this will need to be confirmed once data are collected and evaluated. The anticipated analysis method informs the development of the recommended data collection framework for the ADOT tree removal program, as detailed in Chapter 4. This chapter has two components:  An overview of safety analysis methods for developing CMFs  An evaluation to select the appropriate analysis method OVERVIEW OF ANALYSIS METHODS One of the most comprehensive sources of methods for safety effectiveness evaluation is Recommended Protocols for Developing Crash Modification Factors (Carter et al. 2012). This document was produced by the University of North Carolina’s (UNC) Highway Safety Research Center and Vanasse Hangen Brustlin, Inc. (VHB) for the FHWA CMF Clearinghouse. The methodologies for estimating crash modification factors can be divided into two broad categories: before‐after studies and cross‐sectional studies. Before‐After Studies CMFs derived from before‐after studies are based on the change in safety due to treatment implementation. Examples of before‐after study designs are:  Naïve before‐after study – This simple before and after comparison of a treatment’s safety effect compares the crash frequency in the after period with the crash frequency in the before period. A variation of this methodology compares crash rates (per vehicle miles traveled) between the before and after periods.  Before‐after study with comparison group – An untreated comparison group that is similar to the treatment group is identified to account for temporal effects or changes in traffic volumes from the before to after treatment period. The comparison group is used to calculate a comparison ratio; that is, the ratio of observed frequency in the after period to that in the before period. The observed crash frequency in the before period at the treatment sites is multiplied by this comparison ratio to estimate the expected number of crashes in the after period, had the treatment not occurred. The expected number of crashes in the after period (had the treatment not occurred) is then compared with the actual number of crashes in the after period to determine the safety impact of the treatment. Another approach is to develop safety performance functions (SPFs) using data from the untreated comparison group that relate crash frequency with site characteristics, including AADT. The SPF would account for the effect of traffic volume changes from the before period to the after period. 19   EB before‐after study – This type of study identifies a reference group of sites that have not had treatment, but are similar to the treatment sites in terms of crash risk factors (i.e., volume, roadway characteristics, and surrounding land use conditions). The method then estimates SPFs that relate crashes to independent variables, such as traffic volume and other site characteristics using the reference sites’ data. The method also calculates annual SPF multipliers that account for the temporal effects. Using the SPFs, annual SPF multipliers, and traffic volume data, the method estimates the number of crashes that would be expected for the before period in each site. For each treatment site, the method then estimates the expected number of crashes that would have occurred in the after period, had there been no treatment. The expected number of crashes in the after period without the treatment, along with the variance of this parameter and the number of reported crashes after the treatment, is used to estimate the index of effectiveness, which also is referred to as the CMF (i.e., crash modification factor) and the variance of the CMF. Full or hierarchical Bayes before‐after study – Similar to the EB methods, a reference group is used to estimate the expected crash frequency from an SPF. The full or hierarchical Bayes approach uses the distribution of likely values from the reference group instead of the point estimate. The method combines these estimates with the observed crash frequency in the before period of the treatment group to estimate the long‐term expected crash frequency without the treatment. Cross‐Sectional Studies CMFs from cross‐sectional studies are derived by comparing the safety of a group of sites with a safety improvement to the safety of a group of sites without the safety improvement. Typically, this is a comparison of average crash frequency. Because finding sites with similar characteristics for comparison is difficult, the method uses SPFs or crash prediction models (CPMs) that relate crash frequency to site characteristics. The coefficients of the variables from these SPFs or CPMs are used to estimate the CMF associated with a treatment. Carter et al (2012) provides further explanation of confounding variables in cross‐sectional studies. EVALUATION TO SELECT THE APPROPRIATE ANALYSIS METHOD Figure 6 provides a framework for identifying the most appropriate safety effectiveness analysis methodology given the conditions available to the researcher. Based on this flow chart, the EB method of before‐after analysis will likely be the appropriate analysis methodology when ADOT staff undertake the safety effectiveness evaluation. 20 Source: Gross et al. (2010) Figure 6. Flow Chart for Study Design Selection As shown in Figure 6, four key decisions lead to this conclusion:  Are data available for the treatment in your jurisdiction or can you install the treatment and collect data? Yes. Assuming the data collection efforts specified in this report move forward, data for the analysis will be available.  Are there sufficient existing or planned installations for a before‐after study? Yes. This is essentially a question of sample size. The number of sites (or miles of roadway) with tree removals should be sufficient to develop a robust before‐after study evaluation plan. The determination of sample size a priori is difficult because it depends on a number of factors, including average crash frequency, the level of statistical significance desired in the model, and the expected effect of the treatment. At the time of the before‐after analysis, the researcher can explore the suitability of the identified sample in more detail. In general, quantification of statistically significant effects that are small (i.e., less than 10 percent of crash reductions) requires larger samples. Assessing the differences in safety effects based on the extent of tree 21   clearing activities (e.g., the clearing distance) and possibly changes in crash type or crash severity further complicate the sample size question. If the treated locations do not provide a sufficient sample size for evaluating effects of different recovery area lateral clearances, the data set could potentially be supplemented with sites such as those treated by routine maintenance. It should be noted that pursuing this option would reduce the clarity of the findings as they specifically relate to the benefits of removing trees for the purpose of reducing crash frequency or severity. If needed, another option for expanding the sample size would be to determine whether other states are deploying comparable projects, and if so, combine data sets. This would be possible if a state is considering a similar research question and collecting the same data as Arizona. Are there suitable locations to develop a comparison group or reference group? Yes. Because the treatment locations are not very unique (e.g., rural four‐lane freeway and rural two‐lane highway), a large suitable pool of similar locations from which to develop reference groups or comparison groups is likely. Development of SPFs specifically for evaluation of tree removal in Arizona (evaluation‐level SPFs) is the preferred approach to conducting these analyses, as it would provide results specific to the situation and available data. A second option could be applying calibrated SPFs from the HSM (project‐level SPFs). Finally, if available, calibrated SPFs from a full deployment of SafetyAnalyst (network screening‐level SPFs) could be used if these models fit reasonably well. It should be noted, however, that this approach would provide the lowest degree of reliability in the results of the analysis, and in other states SafetyAnalyst SPFs have not fit well. The sites used to develop the models should not include the treated sites and will need to use data already collected in the ADOT data warehouse. Select before‐after study method. The final element of Figure 6 provides criteria for selecting a study method. The criteria determine whether a comparison group, EB, or full Bayes (FB) evaluation is the most appropriate. In this study, the treated sites were selected primarily because of the high number of tree‐related crashes. This selection introduces the regression‐to‐the‐mean bias that cannot easily be controlled in the comparison group approach. The study criteria suggest that additional complexity of the FB approach does not have strong justification in this proposed study. Spatial correlation is possible between sites (i.e., some climates or geographies might make a certain area have larger and more dense trees), but the data collection plan recommends variables that should help properly account for these differences. Neither a complex model form nor inclusion of prior CMF knowledge is important in this study approach. Thus, the flow chart suggests an EB before‐after methodological approach. It should be noted, however, at this point in the before‐ after safety evaluation (the data collection plan), the data requirements are nearly identical for the FB and EB studies, so the decision on analysis method can and should be re‐evaluated after data collection. 22 Overview of Empirical Bayes Concepts and Methodology In an EB before‐after study, safety effectiveness is estimated by comparing the number of crashes in a given period after implementing a treatment to an estimate of the number of crashes in the same period, had the treatment not been implemented; for example, suppose enhanced signage was implemented along a group of horizontal curves in 2009. The safety effectiveness (CMF) of the treatment would be estimated by comparing the observed number of crashes at the curves from 2010 to 2013 with the treatment in place to an estimate of what would have been the number of crashes at the curves between 2010 and 2013 if the enhanced signage had not been installed. In the EB method, safety performance functions, or SPFs, and observed crash data are used to estimate crash frequency in the after period had the treatment not been implemented. It typically is not a good idea to estimate these data by using observed crashes from the period before the treatment is installed. This is because other conditions (e.g., weather, traffic volume, and other road conditions) at the site might have changed in the after period and because of issues associated with regression‐to‐the‐mean. Seven basic steps are performed to complete a safety effectiveness evaluation using the EB method. Table 7 provides a tabular summary of the observed, predicted, and expected crash information used in the analysis. The steps to conduct the EB analysis are as follows: 1. Compile data, including crash data (A and B), SPFs, and any other data needed to estimate/use the SPFs (e.g., traffic volumes, geometric elements, and operational elements). 2. Apply SPFs to predict crashes at the treatment sites before the treatment was implemented (C) and after the treatment was implemented (D). 3. Apply the EB method to estimate expected crashes in the before period (E). 4. Calculate the ratio of predicted crash frequency in the after period with treatment to predicted crashes in the before period without treatment (D/C). 5. Estimate expected crash frequency in the after period, had the treatment not been in place (i.e., F = E*D/C). 6. Calculate safety effectiveness (or CMF) of the treatment by dividing observed crashes in the after period with the treatment in place by the estimate of safety in the after period, had the treatment not been implemented (B/F). 7. Calculate variance and standard deviation of the safety effectiveness. Table 7. Summary of Time Period and Crash Data in EB Before‐After Analysis Time Period Before After without treatment – hypothetical scenario After with treatment 23 Observed Crashes A Predicted Crashes (from SPFs) C B D Expected Crashes E F Role and Source of Safety Performance Functions As part of the data collection and organization process, ADOT will categorize segments being treated with tree removal into different site types according to characteristics described in Chapter 4 of this report. SPFs and Arizona crash type distributions will be needed for each site type. SPFs are used in the EB method for safety effectiveness evaluation to predict crash frequency before the treatment was in place at each site, predict crash frequency after the treatment is in place at each site, and to estimate expected crashes. The preferred approach is to develop and calibrate SPFs specifically for this evaluation (evaluation‐level SPFs). There would be several different SPFs consistent with the site types (e.g., two‐lane rural highways, rural multilane highways), crash severities (e.g., total crashes, or fatal and serious injury‐only crashes), and crash type. SPFs from the HSM (project‐level SPFs) also could be calibrated to site conditions, or network screening‐level SPFs from a full SafetyAnalyst deployment could be used. Both of these options should be considered only if it is not possible to develop specific SPFs for this project. FHWA has prepared two valuable documents related to calibrating or developing SPFs: Safety Performance Function Development Guide: Developing Jurisdiction‐Specific SPFs (Srinivasan and Bauer 2013); and How to Choose Between Calibrating SPFs from the HSM and Developing Jurisdiction‐Specific SPFs (Srinivasan et al. 2013). Note that, if ADOT plans to develop SPFs specifically for evaluation of tree removal, reference sites will also be needed. Reference sites are comparable to the sites receiving the before and after treatment conditions with the exception of the reference sites that do not have any tree removal activities (including tree removal maintenance activities). 24 CHAPTER 4. DATA NEEDS AND DATA COLLECTION FRAMEWORK Based on the review of existing safety literature and relevant ADOT materials and the anticipated analysis method for developing CMFs (the EB method before‐after analysis), this chapter presents the recommended data collection process and data needs. A summary list of data needs includes:  Roadway segment data – Before and after tree removal o Site location (highway number, beginning and ending milepost) o Terrain (flat, rolling, mountainous) o Roadway functional classification o Posted speed o Number of lanes o Lane width o Shoulder width o Shoulder type o AADT (all years) o Side slope o Horizontal curve o Driveway density  Tree removal data o Site location (highway number, beginning and ending milepost) o Removal location specification o Date and year of removal o Average density of trees before and after tree removal o Actual recovery area distance before and after tree removal o Tree density in thinned area before and after tree removal  Crash data o Site location (highway number, beginning and ending milepost) o Total crashes o Fatal crashes o Incapacitating injury crashes o Nonincapacitating crashes o Possible injury crash o Property damage only With the exception of the tree removal data, the majority of these data exist in ADOT databases. However, ADOT staff will need to assemble the data into a specific database for the safety effectiveness evaluation. 25 DATA COLLECTION FRAMEWORK The study team recommends a seven‐step framework for data collection, compilation, and evaluation. Figure 7 shows the seven‐step flow chart. Step A: Define Tree Clearing Project Limits and Data Needs Step B: Compile Before and After Clear Zone Distances and Tree Density Step C: Subdivide Tree Removal Sites into Homogenous Segments Step E: Identify Reference Sites and Develop or Acquire Calibrated SPFs for Reference Sites Step D: Compile Before Crash and Traffic Volume Data per Homogenous Segment Before Crash, Segment Characteristics, and Before After Clear Zone Database Step F: Compile After Data per Homogenous Segment Characteristics Unchanged or Record Changes Before and After Crash, Segment Characteristics, and Clear Zone Database Step G: Apply EB Method to Estimate CMFs Figure 7. Data Collection Flow Chart 26 As outlined in Figure 7, the steps are:  Step A: Define tree clearing project limits and data needs. ADOT begins by specifying tree removal projects and field data collection requirements. As appropriate, landscape architects should be included in the process. ADOT provides field data collection materials to contractors as appropriate. The project limits should be easily referenced so that tree density and recovery area can be easily converted to milepost measurements from the field measured distances (e.g., in feet or stations from project starting limit).  Step B: ADOT compiles recovery area lateral distances and tree density before and after tree removal at each site. Tree density will be recorded as low (16 trees per acre), medium (17 to 32 trees per acre) and high (32 trees per acre). Recovery area lateral distances before and after tree removal will be recorded. If recovery area lateral distance along the road (before or after) changes by more than 5 feet, the beginning and ending milepost of these changes should be recorded. Photos before and after tree removal should also be taken. Finally, if tree density changes across the tree removal site, the beginning and ending milepost should be recorded.  Step C: ADOT begins developing the analysis database by subdividing (as appropriate) tree removal sites into homogenous segments. Homogenous segments will be defined according to roadway characteristics, features, recovery area lateral distance, and tree density (depending on degree of variance). This information is compiled in a database to facilitate SPF development (if undertaken) and application of the EB method.  Step D: ADOT compiles before‐implementation crash and traffic volume data (three years before implementation) for the homogenous segments within each site (data developed in Steps A and B). The homogenous segments will be developed by roadway characteristics, traffic volume data, recovery area lateral distance, and tree density data. As such, the segmentation process will be iterative as field data (i.e., before and after recovery area lateral distances and before and after tree density information) are compiled.  Step E: ADOT begins the CMF development process by identifying reference sites to the homogenous segments from Steps A through C. ADOT should then develop or acquire calibrated SPFs for these reference sites. One SPF will be needed for each type of homogenous segment.  Step F: Three years after the tree removal implementation, ADOT compiles crash and traffic volume data for each homogenous segment. ADOT should record any changes to the study segment characteristics. This information is combined with the segment characteristics and crash data from Step C to form a database. Step G: ADOT applies the EB method with the calibrated SPFs from Step E and the combined crash and roadway characteristics database from Step F. 27 DATA NEEDS Three categories of data are needed for this analysis:  Site data that describe the location, context, traffic volume, and roadway characteristics of the site where ADOT implemented the tree removal (Steps A and B).  Tree removal specifications that describe field data needs, such as longitudinal extent of tree removal, average horizontal distance to tree or group of trees removed, and average tree density before and after removal (Steps A and B).  The number of crashes, crash type, and severity for three years before and three years after tree removal implementation (Steps C and D). Note that if crash frequency is low, in order to have a large enough sample size, it may be necessary to delay the study and compile two more years of data for a total of five years of after data. If this is the case, two additional years of data should be added to the before data set as well. The site data specified below will support developing and conducting an EB analysis for estimating changes in crash frequency or severity that can be associated with tree removal in Arizona. If compiling all of these data is not possible, or data gaps remain, the analyst conducting the safety effectiveness evaluation will need to conduct exploratory and sensitivity analyses to understand the scope of analyses possible. Figure 7 can be a resource for these considerations. Site Data In Steps A and B of the process, ADOT will provide field contractors with field data needs per project site. Concurrently, for each tree removal site, ADOT will begin compiling data from existing ADOT databases or Google Earth as needed to segment sites and prepare for future analyses. The data needs and definitions follow. Note that some elements may vary by travel direction (indicated as “bidirectional” below).  Site: A site is an HSIP tree removal project. If the site has trees removed from both sides of the roadway, each side of the roadway is considered a separate site. A site may have one or more homogenous segments on one or both sides of the roadway (depending on the tree removal project). The location is specified by the highway number and the beginning and ending mileposts of the site.  Homogenous segment: A homogenous segment is a section of roadway with cross‐sectional features, traffic volume, roadside features, and tree density that do not change more than as specified in the following bullets. A homogenous segment is a component of the site. Depending on roadway characteristics, a site may have different homogenous segments for each direction of travel. The location is specified by the highway number and the beginning and ending mileposts of the homogenous segment.  Terrain (bidirectional): Terrain could affect driver behaviors and vehicle performance. For instance, vehicles traveling downhill will have more momentum than vehicles traveling in the opposite direction. Terrain will be recorded by the field contractor as flat, rolling, or 28     mountainous. Assuming all other characteristics remain the same, a change from one terrain type to a different terrain type creates a new homogenous segment. Roadway functional classification: Functional classification indicates roadway and roadside characteristics and operating and posted speeds. ADOT has six functional classifications (interstate, freeway, principal arterial, minor arterial, major collector, and minor collector). These classifications are further divided by urban and rural area. ADOT has created FHWA‐ approved functional classification maps by counties, cities, towns, places, and tribes (ADOT 2016a). These maps can be used as references. Assuming all other characteristics remain the same, a change from one roadway functional classification to a different functional classification creates a new homogenous segment. Posted speed limit: ADOT establishes posted speed limits to be as near as practicable to the speed at or below which 85 percent of the drivers are traveling (ADOT 2016b). Posted speed can be revised downward based on factors such as crash experience, roadway geometrics, and adjacent development. The posted speed can be recorded in the field. A desirable option is for the consultant/ADOT designers to collect average vehicle traveling speed before and after tree removal to evaluate whether tree removal increases speed. If speed data are desired, a speed study would need to be conducted by ADOT. Assuming all other characteristics remain the same, a change from one posted speed limit to a different posted speed limit creates a new homogenous segment. Number of lanes (bidirectional): Number of travel lanes (including turning lanes or passing lanes) at each site will be compiled from existing ADOT databases. Note that the number of travel lanes of a roadway segment may vary by travel direction. ADOT has this information in its annual state highway system log (ADOT 2016c). Assuming all other characteristics remain the same, a change in the number of lanes per direction creates a new homogenous segment. Lane width: Based on the AASHTO (2010) HSM, lane width is measured to 0.1 foot of precision. Table 8 recommends rounded lane widths to assist with segmentation. A change in lane width greater than 0.5 foot creates a new homogenous segment. These data will be compiled from existing ADOT databases or can be estimated from Google Earth if gaps exist. Table 8. Rounded Lane Width for Roadway Segmentation Measured Lane Width (Feet) 9.2 or less 9.3 to 9.7 9.8 to 10.2 10.3 to 10.7 10.8 to 11.2 11.3 to 11.7 11.8 or more Source: AASHTO (2010) HSM, Volume 2 29 Rounded Lane Width (Feet) 9 or less 9.5 10 10.5 11 11.5 12 or more  Shoulder width (bidirectional): Shoulder width is defined as the distance from the edge of the travel lane to the edge of the roadway. Based on the AASHTO (2010) HSM, shoulder width is measured to 0.1 foot of precision. Table 9 recommends rounded shoulder widths to assist with segmentation. A change in shoulder width greater than 1 foot creates a new homogenous segment. These data will be compiled from existing ADOT databases or can be estimated from Google Earth if gaps exist. Table 9. Rounded Shoulder Width for Roadway Segmentation Measured Shoulder Width Rounded Shoulder Width (Feet) (Feet) 0.5 or less 0 0.6 to 1.5 1 1.6 to 2.5 2 2.6 to 3.5 3 3.6 to 4.5 4 4.6 to 5.5 5 5.6 to 6.5 6 6.6 to 7.5 7 7.6 or more 8 or more Source: AASHTO (2010) HSM, Volume 2     Shoulder type (bidirectional): ADOT designers can extract shoulder type (e.g., asphalt, portland cement concrete, gravel, or dirt) from ADOT’s highway system log. Optionally, the field contractor can collect/verify this information onsite. Assuming all other characteristics remain the same, a change in shoulder type per direction creates a new homogenous segment. Side slopes (bidirectional): In some cases, a site might have a side slope, where the cross‐ section slopes down or up from the shoulder edge. Applying concepts from the HSM CMF for flattening side slopes, a change of side slope by more than either 1 foot of horizontal distance or 1 foot of vertical distance or both will create a new homogenous segment. Horizontal curvature: Horizontal curves are defined as gradual roadway transitions between two straight roadways. If the homogenous segment includes a horizontal curve, the curve should be identified as a separate homogenous segment. The length of curve and curve radius data should be recorded and can be requested from ADOT's Geographic Information System (GIS)/Multimodal Planning Division. The radius of the curve site should be available from as‐built drawings or can be estimated from Google Earth. AADT volume (bidirectional): Daily traffic volume is recorded on select mileposts on all Arizona state highways. The most recent AADT data collected by ADOT are for 2012 and are available online as AADT reports (ADOT 2016d). Note that a site may be located between two traffic volume collection mileposts. In this case, the daily traffic volume at the site should be the weighted average of the traffic volumes and distance between the recorded traffic volume sites. The AADT should be recorded each year for three years before and three years after tree removal. This information should include the truck factor or the percentage of the AADT volume generated by trucks or commercial vehicles. Truck factor is available from ADOT’s AADT reports. 30   Presence of other roadside feature: ADOT safety staff should compile and document the location and extent of other roadside features (e.g., guardrails (presence and type), fences, walls, and call boxes) at each project site. These features are available in ADOT’s highway Feature Inventory System (FIS). The milepost and offset of features within the recovery area lateral distance should be recorded. Presence of driveways and driveway density: If driveways are present, the number of driveways per mile should be recorded. The HSM CMF for driveway density is more sensitive to changes in driveway density at lower traffic volumes. A change of driveway density greater than two driveways per mile would yield a new segment. A driveway providing access to a facility where only one or two trips per day are made (e.g., a farm or private residence) should not be included in this analysis. The site elements described above may change over time. To evaluate before and after safety effectiveness as detailed in the next section, the data may need to be confirmed by ADOT safety staff during the after data collection period and prior to analyses (Step F). Tree Removal Implementation Specifications For the purposes of this analysis, a tree is defined as vegetation with trunk diameter greater than 4 inches (AASHTO and TRB 2016). Similarly, a group of trees or shrubs with a collective diameter greater than 4 inches also is considered a tree. The ADOT designer and field contractor need to record the following key specifications of the work conducted on a segment:  Date: Date and year tree removal project is begun and completed.  Average density of trees along the roadway before and after tree removal: The density of trees removed is defined as the number of trees per acre. The density before tree removal and after tree removal should be recorded. If field contractors conducted tree thinning beyond the recovery area, they should record the distance from the edge of the traveled way to the tree line, as well as the average density of trees in the area where thinning was conducted. This information will need to be converted to milepost to match other ADOT data. Based on specifications provided by ADOT, tree density should be recorded as low (16 trees per acre), medium (17 to 32 trees per acre) and high (32 trees per acre).  Recovery area lateral distance (bidirectional): In this project, recovery area lateral distance is the actual horizontal distance from the edge of the traveled way to the tree line and includes the shoulder. Field contractors will record this distance in feet. Note that the recovery area lateral distance can vary along different parts of a site. If this variation is within 5 feet, the recovery area lateral distance can be the average distance along the segment. Recovery area lateral distance variation greater than 5 feet would reflect a new segment. Again, beginning and ending milepost of the recovery area lateral distance would be recorded and integrated into the database. If appropriate, the recovery area for the median also should be recorded. 31 The field contractor also should collect before and after photos and videos, including the following:  Photos (looking along the roadway and the roadside) before tree removal implementation  Photos (looking along the roadway and the roadside) immediately after the tree removal has been completed  If possible, a video log of the site before and after tree removal implementation Crash Data To evaluate safety effectiveness, ADOT will consider questions such as:  Whether the site has seen a decrease in recorded crashes with trees, shrubs, or stumps as one of the crash events.  Whether the site has experienced a change in crash type or severity, but not frequency. For instance, a site that previously experienced many overturn crashes with tree collisions could see only overturn crashes after tree removal.  Whether the site has experienced a shift in crash severity. For example, a site might experience fewer severe crashes after tree removal because of a decrease in fixed object crashes. As such, during the period before data collection, ADOT safety staff should record all crashes on the segments, as well as their crash type and severity (i.e., fatal, incapacitating, moderate, minor, and no injuries). The ADOT Crash Report Form Box 31 indicates the sequence of crash events and can be used to record crash type. ADOT safety staff should extract all crashes with trees, shrubs, or stumps in any of the sequence of events (Form Box 31). Note again that if crash frequency is low, it may be necessary to change the study period from three to five years. If so, two additional years of before crash and traffic volume data should be collected, as well as two additional years of after period data. Data Collection Form and Organization Table 10 is an example of a form for field contractors to use at the time of tree removal. The field contractor would collect the following information: the before and after recovery area lateral distance, the before and after average tree density and whether tree thinning beyond the clear zone was conducted, the distance from the traveled way to the tree line, and the average density of trees before and after thinning. 32 Table 10. Example Field Data Collection Form Tree Removal Work Specifications Date of field work: Site location (highway number, beginning & ending mileposts): Segment location (highway number, beginning & ending mileposts): S 260, 5.2, 10.7 S 260, 7.2, 10.0 Before After Clear zone distance (ft) Clear zone – average density of trees If thinning was conducted: Distance from edge of travelled way to tree line (ft): Average density of trees (trees/acre) Table 11 shows how the information from the data collection process can be organized for the effectiveness analysis. This table provides two sample records. Note that each data record or row represents a homogenous roadway segment at a site. For the purpose of this example, only segment data variables with values that have changed after tree removal are shown in the after tree removal implementation columns. 33 S 260, 5.2, 10.7 S 260, 6.5, 7.8 Flat S 260, 5.2, 10.7 S 260, 7.8, 9.2 Flat Major collection Major collection S 260, 5.2, 10.7 S 260, 6.5, 7.8 S 260, 6.6, 7.5 7/12/2015 15 S 260, 5.2, 10.7 S 260, 7.8, 9.2 S 260, 8.2, 9.0 7/12/2015 20 Sit e l be ocat g i io nn n in (hi g M gh w P, a en y n Se di u m ng b gm M er , be ent P) gi nn loc in ati g M on ( P, hi en gh To di wa ta ng y lc M n um ra P) sh Fa be es ta r, lc r Se ash es ve re in M o d jury er cr at as M e he in in s ju or r yc in ju Pr r a r op sh yc er e ra t sh s To y d am es ta lc ag r Fa ash e on es ta ly lc cr r as Se ash he es ve s re i n M j u od r er y cr at as M e he in in s ju or ry in cr ju as r y Pr h cr op as es er he ty s da m ag e on ly cr as he s Sit e lo be ca gi tio nn n in (hi g M gh w P, a en y n di u m ng b Se M er , gm P) be ent gi nn loc in ati g o M n( P, hi en g h di wa Re ng y m M nu ov P) m nu al be m job r, be s r, p b e ec i f gin ic n i ati n g on Da te M s( P , hig an en h w d ye di ar ng ay of M P) im pl Av em er ag e nt ( tr e at ee de io s p ns n er ity ac of re tr ee ) sr em ov ed Sit e l be ocat g i io nn n in (hi g M gh w P, a en y n di u m ng b Se gm M er , P) be ent gi lo nn ca in tio g M n (h P, en ighw di Te n g ay n rra M um P) m in (c be ou a r, n t teg ain or o u i es : s Ro ) fla ad t, wa ro llin yf un g, or ct io na lc Po la ste ss ifi d ca Nu spe ti o ed m n be (m r i le La of s ne lan pe wi es rh ou Sh dth r) ou ld Sh er w ou i ld dth er (in ty pe fee t) Cl (d ea es cr rz i pt on io e Av n) di er st ag a nc vo e e( lu an m nu in e fe ‐ al Av et er yea dail ) r1 yt ag vo e r af lu an fic m nu (A e ‐ al AD Av er yea dail T) r2 yt ag vo e ra ff i lu an m nu c( e AA Sid ‐ y al d DT e S ear aily ) lo 3 t r pe af fic Ho (sl (A op riz AD eo on T) ta fa l n Dr cu g le) iv e rv e wa (ra yd di us en in sit y fe et ) Sh ou ld er w id th Cl ea (in rz fe on et e ) di s t Av an er ce a (in vo ge fe lu An et m e nu a ) ‐y l D ea a r 1 ily Tr Av af er fic a (A vo ge AD lu An m T) n ua e ‐y l D Av ea a er r 2 ily a Tr vo ge af lu An fic m n u e (A ‐ y al D AD ea a T) i r 3 ly Tr af fic (A AD T) Table 11. Data Organization Example Before Tree Removal Implementation – Segment Data 34 After Tree Removal Implementation – Segment Data 55 2 12 3 Asphalt 25 5500 6100 6000 None None No driveways 3 30 6100 6200 6200 55 2 12 2 Asphalt 20 5600 6000 5900 None None No driveways 3 30 6000 6100 6200 Tree Removal Specifications Total Crashes with One of the Sequence of Events as Collision with Tree, Shrub, or Stump Three‐Year Period Before Tree Removal* Three‐Year Period After Tree Removal* S 260, 5.2, 10.7 S 260, 6.5, 7.8 31 1 2 5 8 15 15 0 0 2 3 10 S 260, 5.2, 10.7 S 260, 7.8, 9.2 23 0 1 3 7 12 11 0 1 0 1 9 *Crash data would be recorded per year over the three‐year period. CHAPTER 5. CONCLUSION Conducting a before‐after safety effectiveness evaluation is a multiyear and multiphase process. To complete the evaluation, the before data are collected, the sites are treated, the after data are collected for three years, and the safety effectiveness analysis is made at the end of that period. The research team anticipates that an EB analysis, as described in Chapter 3, would be the most appropriate method for estimating safety effectiveness. It would be necessary to have calibrated SPFs to complete this analysis. The data needed to conduct the before‐after analysis are detailed in Chapter 4, and the steps involved in the overall process are shown in Figure 7. In summary, the data relate to crash type and severity, roadway characteristics and traffic volume, and specifics related to tree removal. The majority of the data needed for these analyses are available in existing ADOT crash, roadway, and traffic volume databases; however a small amount of field data will need to be collected and transferred to the analysis database. It is recommended that ADOT compile data for each tree removal site into a specific database or spreadsheet developed for this analysis. Note that the data for reference sites would also be included and analyzed in order to develop safety performance functions, as explained in Chapter 3. It would also be necessary for ADOT field staff or the contractor to collect before‐and‐after tree densities and recovery area lateral distances at each project site as part of the tree removal field work. Chapter 4 includes specifications for this data. After all of the tree removal projects are completed and the data are compiled, ADOT staff (or consultants with statistical analysis expertise) would conduct the analysis to estimate the CMF related to tree removal projects in Arizona. A CMF related to tree removal in Arizona could be a valuable addition to safety research and literature. The CMF could support a cost‐benefit analysis that would demonstrate the value of the tree removal project investments in Arizona. In addition, conducting state‐specific CMF research would demonstrate to other states the feasibility and value of conducting such analyses. 35 36 REFERENCES American Association of State Highway and Transportation Officials (AASHTO). 2010. Highway Safety Manual, First Edition. Washington, D.C. American Association of State Highway and Transportation Officials (AASHTO). 2011. Roadside Design Guide, Fourth Edition. Washington, D.C. American Association of State Highway and Transportation Officials (AASHTO) and Transportation Research Board (TRB). 2016. “Trees in Hazardous Locations: Appendix 12.” Website. Accessed April 2016. http://safety.transportation.org/htmlguides/trees/app12.htm Arizona Department of Transportation. 2012. Arizona Roadway Departure Safety Implementation Plan. Arizona Department of Transportation. 2016a. “Functional Classification Maps.” Website. Accessed April 2016. http://www.azdot.gov/maps/functional‐classification‐maps Arizona Department of Transportation. 2016b. “Establishing Speed Limits.” Website. Accessed April 2016. http://www.azdot.gov/business/engineering‐and‐construction/traffic/faq/establishing‐speed‐ limits Arizona Department of Transportation. 2016c. “State Highway System Log.” Website. Accessed April 2016. http://www.azdot.gov/planning/DataandAnalysis/state‐highway‐system‐log Arizona Department of Transportation. 2016d. “Average Annual Daily Traffic.” Website. Accessed April 2016. http://www.azdot.gov/planning/DataandAnalysis Carter, D., R. Srinivasan, F. Gross, and F. Council. 2012. “Recommended Protocols for Developing Crash Modification Factors.” Final Report, National Cooperative Highway Research Program (NCHRP) 20‐7. Washington, D.C.: Transportation Research Board. http://www.cmfclearinghouse.org/collateral/CMF_Protocols.pdf. Elvik, R., and T. Vaa. 2004. Handbook of Roadway Safety Measures. Oxford, United Kingdom. Online: El Sevier. Geedipally, S. R., J. A. Bonneson, M. P. Pratt, and D. Lord. 2013. "Severity Distribution Functions for Freeway Segments." Transportation Research Record: Journal of the Transportation Research Board 2398, No. 1: 19‐27. Washington, D.C.: Transportation Research Board of the National Academies. Gross, F., B. Persaud, and C. Lyon. 2010. A Guide to Developing Quality Crash Modification Factors. Report No. FHWA‐SA‐10‐032. Washington, D.C.: Federal Highway Administration. http://safety.fhwa.dot.gov/tools/crf/resources/fhwasa10032/fhwasa10032.pdf. 37 Hallenbeck, M., P. M. Briglia Jr., Z. N. Howard, and A. St. Martin. 2013. In‐Service Evaluation of Major Urban Arterials with Landscaped Medians – Phase III. Report No. WA‐RD 636.3. Seattle: Washington State Department of Transportation Office of Research and Library Service. http://www.wsdot.wa.gov/research/reports/fullreports/636.3.pdf. Hovey, P. W., and M. Chowdhury. 2005. Development of Crash Reduction Factors. Report No. FHWA/OH‐ 2005/12. Columbus: Ohio Department of Transportation. https://www.dot.state.oh.us/Divisions/Planning/SPR/Research/reportsandplans/Reports/2005/Safe ty/14801‐FR.pdf. Jurewicz, C., and V. Pyta. 2010. “Effect of Clear Zone Widths on Run‐off‐Road Crash Outcomes.” Australasian Road Safety Research, Policing and Education Conference. Queensland University of Technology. http://eprints.qut.edu.au/38683/1/c38683.pdf. Maze, T., C. Sax, and N. Hawkins. 2008. Clear Zone – A Synthesis of Practice and an Evaluation of the Benefits of Meeting the 10 Ft Clear Zone Goal on Urban Streets. IHRB Project No. TR‐560, MTC Project No. 2006‐05, CTRE Project No. 06‐261. Ames, IA: Midwest Transportation Consortium. http://www.intrans.iastate.edu/reports/clear_zone_report.pdf Mok, J., H. C. Landphair, and J. R. Naderi. 2005. “Landscape improvement impacts on roadside safety in Texas.” Landscape and Urban Planning 78, No. 3: 263‐274. Online: El Sevier. http://aggie‐ horticulture.tamu.edu/syllabi/435/articles/mok%20article.pdf. New York State Department of Transportation. 2012. Post Implementation Evaluation System – Reduction Factor Report. Albany: New York State Department of Transportation. https://www.dot.ny.gov/divisions/operating/osss/highway‐repository/2011AccRedFac.pdf. Ogle, J., W. Sarasua, J. Dillon, V. Bendigieri, S. Anekar, and P. Alluri. 2009. Support for the Elimination of Roadside Hazards: Evaluating Roadside Collision Data and Clear Zone Requirements. Report No. FHWA‐SC‐09‐01. Columbia: South Carolina Department of Transportation. http://www.clemson.edu/t3s/scdot/pdf/projects/fhwa‐sc‐09‐01spr667.pdf. Peng, Y., S. Geedipally, and D. Lord. 2012. “Effect of Roadside Features on Single‐Vehicle Roadway Departure Crashes on Rural Two‐Lane Roads.” Transportation Research Record. Washington, D.C.: Transportation Research Board. Srinivasan, R., and K. Bauer. 2013. Safety Performance Function Development Guide: Developing Jurisdiction‐Specific SPFs. Report No. FHWA‐SA‐14‐005. Washington, D.C.: Federal Highway Administration. www.pooledfund.org/Document/Download/4484. 38 Srinivasan, R., D. Carter, and K. Bauer. 2013. How to Choose Between Calibrating SPFs from the HSM and Developing Jurisdiction‐Specific SPFs. Washington, D.C.: Federal Highway Administration. www.pooledfund.org/Document/Download/4638. Transportation Research Board. 2013. “National Cooperative Highway Research Program (NCHRP) 17‐ 54: Draft Interim Report Consideration of Roadside Features in the Highway Safety Manual.” Washington, D.C. Transportation Research Board. 2003. “National Cooperative Highway Research Program (NCHRP) Report 500: Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volume 3: A Guide for Addressing Collisions with Trees in Hazardous Locations.” Washington, D.C. http://onlinepubs.trb.org/onlinepubs/nchrp/nchrp_rpt_500v3.pdf. Transportation Research Board. 2004. “National Cooperative Highway Research Program (NCHRP) Report 500: Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volume 8: A Guide for Reducing Collisions with Utility Poles.” Washington, D.C. http://onlinepubs.trb.org/onlinepubs/nchrp/nchrp_rpt_500v8.pdf Zegeer, C., and M. Cynecki. 1984. Selection of Cost‐Effectiveness Countermeasures for Utility Pole Accidents: User Manuals. Report No. FHWA‐IP‐84‐13. Washington, D.C.: Federal Highway Administration. 39 40 APPENDIX A. NEW YORK STATE DOT ACCIDENT REDUCTION FACTORS Crash Type All All All Nonreportable Dry road Wet road Snow/ice/slush road Overtaking Right angle CMF 1. AADT per lane <5000 Crash Severity 2. AADT per lane >5000 All 1. Reduced by 43% 2. Reduced by 32% Injury 1. Reduced by 41% 2. Reduced by 19% Property damage only 1. Reduced by 35% All 1. Reduced by 51% All 2. Reduced by 16% All 2. Reduced by 26% All 2. Reduced by 31% All 2. Reduced by 64% All 1. Reduced by 63% 2. Reduced by 60% All 2. Reduced by 45% Fixed object (tree, hydrant, and other) All fixed object and run‐off‐road All Day All Day and wet road Night Road Dry road Wet road Snow/ice/slush road All fixed object and run‐off‐road 2. Reduced by 28% 1. Reduced by 48% 2. Reduced by 20% 2. Reduced by 37% 2. Reduced by 22% 2. Reduced by 36% 1. Reduced by 16% 1. Reduced by 24% 1. Reduced by 32% 1. Reduced by 9% All All All All All All All 41 Statistical Significance Yes No 42 APPENDIX B. DATA USED IN ASSESSING EFFECTIVENESS OF TREE OR FIXED OBJECT REMOVAL PROJECTS Data Type Crash type Crash condition Crash severity Detailed Description 1. Total crashes 2. Run‐off‐road casualty crashes 3. Overtaking, right‐angle, https://maps.gstatic.com/mapfiles/mapcontrols3d7.png fixed object (tree/hydrant/other), all fixed object, and run‐off‐road crashes 4. Type of fixed object crash (trees, poles, culvert, mail box, bridge, guard rail, and other) 1. Dry road, wet road, snow/ice/slush road, day and night 1. Injury, fatal, and property damage only crashes 2. All accidents, injury, and property damage only Tree offset from the 1. Minimum setback, average setback, and 15th roadway/clear zone percentile setback 2. Clear zone as a part of a run‐off‐road crash prediction model: <2 m, 2 to 4 m, 4 to 8 m, and > 8 m ADT volume 1. AADT: 1500 to 6000 and 6000+ 2. One‐way AADT: <1200 vehicles and >1200 vehicles 3. AADT > 5000 vehicles per lane vs. AADT <5000 vehicles per lane 4. Average daily total and truck volumes as variables in the CRF functions Highway access 1. As a variable in the CRF function: no access control, type limited access control, and full access control Roadway functional 1. Six urban and six rural functional classes as variables classification in the CRF functions 2. Interstate, US primary, state primary, secondary, county, and other roads Other roadway 1. Shoulder width as a variable in the CRF functions characteristics 2. Curve radius as a part of a run‐off‐road crash prediction model: <600 m, 600 to 1500 m, and >1500 m 3. Grade (%) as a part of a run‐off‐road crash prediction model: negative, positive, or zero 4. Traffic lane plus sealed shoulder width as part of a run‐off‐road crash prediction model: <3.5 m and >3.5 m 5. Straight roadway level, grade, and hillcrest 6. Curve roadway level, grade, and hillcrest 7. Speed limit at 30, 35, and 40 miles per hour 43 Study 1. and 2. Dependent variable for most studies 3. New York State DOT 2012 4. Ogle et al. 2009 1. New York State DOT 2012 1. Dependent variable for most studies 2. New York State DOT 2012 1. Maze et al. 2008 2. Jurewicz and Pyta 2010 1. Maze et al. 2008 2. Jurewicz and Pyta 2010 3. New York State DOT 2012 4. Hovey and Chowdhury 2005 1. Hovey and Chowdhury 2005 1. Hovey and Chowdhury 2005 2. Ogle et al., 2009 1. Hovey and Chowdhury 2005 2., 3., 4. Jurewicz and Pyta 2010 5., 6. Ogle et al. 2009 7. Maze et al. 2008 44 APPENDIX C. ADOT CRASH REPORT FORM 45 46 47 48 49 50