AN EVALUATION OF THE RELATIVE SAFETY OF PEDESTRIAN INFRASTRUCTURE USING DRIVER BEHAVIOR AND CONFLICT AS SURROGATES FOR CRASHES By Steven York Stapleton A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Civil Engineering – Master of Science 2017 ABSTRACT AN EVALUATION OF THE RELATIVE SAFETY OF PEDESTRIAN INFRASTRUCTURE USING DRIVER BEHAVIOR AND CONFLICT AS SURROGATES FOR CRASHES By Steven York Stapleton A field study was performed at 40 uncontrolled midblock crosswalks and 26 signalized intersections on low-speed roadways selected from the areas surrounding three major urban college campuses across lower Michigan. An array of existing traffic control devices existed at the study sites, including various crosswalk marking strategies, along with additional treatments, such as pedestrian hybrid beacons (PHBs), rectangular rapid flashing beacons (RRFBs) and single in-street signs (R1-6). The sites also collectively included a diverse set of roadway and traffic characteristics, including crossing widths, number of lanes, and median presence, along with vehicular, pedestrian, and bicyclist volumes. Three initial evaluations were performed for the midblock segments and signalized intersection study sites, including: driver yielding compliance, vehicle-pedestrian conflicts, and non-motorized traffic crash data. Ultimately, only crash data and driver yielding compliance to pedestrians were included in the final analysis. The yielding compliance study found that the type of crosswalk treatment has a strong influence over driver yielding compliance. While yielding compliance improves substantially when crosswalk markings are utilized, the highest compliance rates are achieved when an additional enhancement device (i.e., RRFB, PHB, or R1-6 sign), is also provided. The primary limitation towards prediction of pedestrian crashes is the lack of a reliable exposure data to represent the amount of pedestrian activity on a given segment or intersection. This thesis is dedicated to my late brother, Andrew Mark Stapleton, who always encouraged me to do my best in all my endeavors. iii ACKNOWLEDGEMENTS I would like to acknowledge my graduate advisor, Dr. Timothy Gates, who has been a constant source of support throughout my academic career, and whose guidance was instrumental in writing this thesis. In addition, I would like to acknowledge Dr. Peter Savolainen of Iowa State University whose expertise played a major role in this thesis. I would also like to acknowledge my student collaborators in this research: Trevor Kirsch, Santosh Miraskar, Daniel Srbinovski, and Gentjan Heqimi. Lastly, this research was funded, in part, through a grant provided by the U.S. Department of Transportation’s University Transportation Centers program through the Transportation Research Center for Livable Communities at Western Michigan University. iv PREFACE This thesis presents three methods of evaluating the relative safety of pedestrian infrastructure: crash analysis, conflict analysis, and yielding compliance. Crash and conflict analyses are presented to demonstrate the shortcomings of traditional crash analysis for pedestrian safety, as well as conflict, which is a common surrogate for crashes. While the utility of yielding compliance as a measure of effectiveness stands alone, crash and conflict analyses are included to establish the need for yielding compliance as a surrogate measure of effectiveness for crashes. v TABLE OF CONTENTS LIST OF TABLES ....................................................................................................................... viii LIST OF FIGURES ....................................................................................................................... ix KEY TO ABBREVIATIONS ..........................................................................................................x CHAPTER 1: INTRODUCTION ....................................................................................................1 1.1 Problem Statement ................................................................................................................3 1.2 Research Approach ...............................................................................................................4 1.3 Objectives .............................................................................................................................5 1.4 Study Constraints ..................................................................................................................5 CHAPTER 2: LITERATURE REVIEW .........................................................................................7 2.1 Highway Safety Manual and the Predictive Method ............................................................7 2.2 Using Conflict as a Surrogate Measure for Crashes ...........................................................12 2.3 Using Staged Crossing to Evaluate Pedestrian Crossing Treatments .................................13 2.4 Predicting Crashes Using Behavioral Information .............................................................14 2.5 Safety Performance of Midblock Pedestrian Crosswalk Treatments .................................16 2.5.1 Marked and Unmarked Crosswalks ............................................................................16 2.5.2 In-Street Pedestrian Crossing Signs ............................................................................17 2.5.3 Rectangular Rapid Flashing Beacon ...........................................................................18 2.5.4 Pedestrian Hybrid Beacon...........................................................................................20 CHAPTER 3: DATA COLLECTION ...........................................................................................22 3.1 Site Selection ......................................................................................................................22 3.2 Field Data Collection ..........................................................................................................27 3.2.1 Staged Pedestrian Crossing Events .............................................................................27 3.2.2 Naturalistic Pedestrian Crossing Events .....................................................................31 3.2.3 Pedestrian-Vehicle Conflicts ......................................................................................31 3.2.4 Road User Volumes ....................................................................................................31 3.3 Pedestrian Crash Data Collection .......................................................................................32 CHAPTER 4: CRASH ANALYSIS ..............................................................................................35 4.1 Data Summary ....................................................................................................................36 4.2 Analytical Procedures .........................................................................................................39 4.3 Results and Discussion .......................................................................................................40 CHAPTER 5: DRIVER YIELDING BEHAVIOR AT SIGNALIZED INTERSECTIONS.........45 5.1 Data Summary ....................................................................................................................45 5.2 Results and Discussion .......................................................................................................48 vi CHAPTER 6: DRIVER YIELDING COMPLIANCE AT MIDBLOCK CROSSINGS ..............52 6.1 Data Summary ....................................................................................................................53 6.2 Data Analysis ......................................................................................................................55 6.3 Results and Discussion .......................................................................................................56 6.4 Conclusions and Recommendations ...................................................................................62 CHAPTER 7: CONCLUSIONS ....................................................................................................65 7.1 Driver Behavior During Pedestrian Crossing Attempts .....................................................65 7.2 Recommendations ...............................................................................................................67 APPENDIX ....................................................................................................................................69 REFERENCES ..............................................................................................................................86 vii LIST OF TABLES TABLE 1. Michigan Pedestrian Crashes by Location Type and Traffic Control, 2012-2016 ........2 TABLE 2. Safety Performance of Crossing Treatments from Literature ......................................21 TABLE 3. Number of Study Sites by Crossing Type and City .....................................................23 TABLE 4. Characteristics of Midblock Crosswalk Study Sites ....................................................24 TABLE 5. Characteristics of Signalized Intersection Study Sites .................................................26 TABLE 6. Characteristics of Midblock Study Segments ..............................................................36 TABLE 7. Descriptive Statistics for Analysis of Pedestrian Crashes on Midblock Segments .....38 TABLE 8. Negative Binomial Results for Vehicle-Pedestrian Crashes on Uncontrolled Midblock Segments ........................................................................................................................................41 TABLE 9. Summary of Naturalistic Driver Yielding Behavior Data at Signalized Intersections 47 TABLE 10. Naturalistic Driver Yielding Compliance Rates by Site Characteristics ...................48 TABLE 11. Negative Binomial Results for Naturalistic Driver Yielding Compliance at Signalized Intersections .................................................................................................................49 TABLE 12. Summary of Site Characteristics for Midblock Yielding Compliance Assessment ..54 TABLE 13. Logistic Regression Results for Driver Yielding Compliance..................................57 TABLE 14. Driver Yielding Compliance by Crosswalk Treatment..............................................58 TABLE 15. Interaction of Lane Position with Roadway Cross-Section and Crosswalk Treatment .......................................................................................................................................60 TABLE 16. Aerial Imagery for Study Sites...................................................................................70 viii LIST OF FIGURES Figure 1. Typical Pedestrian Crosswalk Enhancements in Michigan ..............................................3 Figure 2. Typical Video Camera Setup for Recording Motorist Yielding Behavior .....................27 Figure 3. Location of Dilemma Zone Relative to Crosswalk ........................................................30 Figure 4. Screenshot of Staged Crossing Attempt .........................................................................30 Figure 5. Distinction between Pedestrian Crashes at a Minor Street Intersection: (1) Stop Controlled Leg Crash vs. (2) Uncontrolled Midblock Crosswalk Crash .......................................33 Figure 6. Pedestrian Crashes per Marked Crosswalk with Hourly Vehicular Traffic Volume and Hourly Pedestrian Crossings by Site .............................................................................................37 Figure 7. Pedestrian Crashes per Mile with Hourly Vehicular Traffic Volume and Hourly Pedestrian Crossings by Site ..........................................................................................................37 Figure 8. Relationship Between Driveway and Crosswalk Density ..............................................42 Figure 9. Screenshot of Noncompliant Turning Vehicle at Signalized Intersection .....................46 Figure 10. Yielding Rates vs. Pedestrian-Turning Vehicle Interactions per 15 Minute Interval ..50 Figure 11. Yielding Compliance by Lane Position and Treatment ...............................................61 ix KEY TO ABBREVIATIONS HSM Highway Safety Manual AADT Annual average daily traffic ADT Average daily traffic SPF Safety performance function CMF Crash modification factor AASHTO American Association of State Highway and Transportation Officials FHWA Federal Highway Administration PHB Pedestrian hybrid beacon RRFB Rectangular rapid flashing beacon MOE Measure of effectiveness x CHAPTER 1: INTRODUCTION The safety of pedestrians continues to be a critical transportation issue, both nationally and throughout Michigan. Approximately 65,000 pedestrians are injured in traffic crashes in the United States annually, including approximately 5,000 fatalities [1]. A query of the Michigan Traffic Crash Database via the Michigan Traffic Crash Facts website [michigantrafficcrashfacts.org] showed that between 2012 and 2016, 11,395 pedestrian crashes occurred on roadways in Michigan, representing a 3.7 percent increase over the previous 5-year period of 2007 to 2011. Such crashes resulted in 761 fatal crashes involving pedestrians, representing an 18.9 percent increase over 2007 to 2011. During the same period, crashes not involving pedestrians decreased by 1.81 percent while fatal crashes not involving pedestrians decreased by 2.68 percent. While pedestrian-involved crashes comprised only a small portion (0.8 percent) of all crashes that occurred between 2012 and 2016, consider that fatal crashes involving pedestrians accounted for 17.2 percent of all fatal crashes in Michigan during that period. When considering the vulnerability and relative risk, pedestrians were 27 times more likely to be fatally injured when involved in a traffic crash compared to occupants of motor vehicles. Crashes involving pedestrians occur most frequently within urban and suburban areas, particularly on or near college campuses, since these areas experience the highest levels of pedestrian activity and traffic volumes. Further, there is considerably greater distraction present for both motorists and pedestrians in such areas, and the focus of motorists is often drawn away from the roadway. As a result, pedestrians are often put into situations where approaching motorists do not see them or are surprised by their presence, which may lead to conflicts and traffic 1 crashes. Unfamiliar drivers, which are particularly common on college campuses, further exacerbate these safety issues. Various efforts have been implemented to address pedestrian safety issues throughout the United States, including “Complete Streets” policies, “Safe Routes to School” programs, and other initiatives. However, while these efforts have improved safety and connectivity for non-motorized road users, they have also facilitated increases in pedestrian and bicyclist travel, thereby leading to an increased exposure and subsequent crash risk. Such risks may be mitigated by the application of appropriate engineering treatments to enhance motorists’ awareness of crossing pedestrians, while also encouraging pedestrians to cross at these engineering crossing areas. However, given limited financial resources, adequate guidance is necessary to assist agencies in determining when and where to implement pedestrian safety treatments in the most cost effective manner possible. As can be observed in Table 1, the need for effective pedestrian safety countermeasures is particularly important at non-intersection (i.e., midblock) locations, especially at such locations where no signal exists (i.e., uncontrolled). Also problematic for pedestrian safety are intersections with no traffic control, including uncontrolled legs of stop controlled intersections, as vehicular operations are similar to that experienced at midblock areas but with the additional risk of turning traffic. TABLE 1. Michigan Pedestrian Crashes by Location Type and Traffic Control, 2012-2016 Road User Type Pedestrian Type of Location Non Intersection – No Signal Non Intersection - Signal Intersection – No Control Intersection – Stop or Yield Intersection – Signal Crash Statistics, 2012 - 2016 Number Number of Fatal Crashes as of Fatal Percent of All Crashes Crashes Crashes 5,794 548 9.5% 538 1,443 955 2,387 2 35 82 16 68 6.5% 5.7% 1.7% 2.8% 1.1 Problem Statement A variety of pedestrian safety treatments are available for implementation at such locations, including pedestrian hybrid beacons (PHBs), rectangular rapid flashing beacons (RRFBs), and instreet pedestrian signs (R1-6), examples of which are displayed in Figure 1. Resource constraints make it imperative that agencies are able to identify those locations that are at the highest risk for pedestrian-involved crashes so that appropriate countermeasures may be implemented. As such, there is a clear need for well-supported guidelines to assist in determining appropriate locations for specific pedestrian safety treatments. Single R1-6 R1-6 Gateway Configuration Pedestrian Hybrid Beacon Rectangular Rapid Flashing Beacon Figure 1. Typical Pedestrian Crosswalk Enhancements in Michigan Typically, these types of network screening activities have been done on the basis of historical crash data. More recently, development of safety performance functions (SPFs) has provided a promising approach for quantifying the level for pedestrian crashes at specific intersections or road segments. The Highway Safety Manual (HSM) currently provides an 3 aggregate pedestrian SPF, which is based upon land use characteristics [2]. However, since pedestrian crashes are particularly rare, such an approach limits the ability to proactively identify sites with the potential for crashes that are not reflected by recent crash data. As a result, research is limited in terms of disaggregate-level studies considering the effects of motor vehicle/bicycle/pedestrian volumes, roadway geometry, and other factors on pedestrian crashes. Furthermore, research has also been limited with respect to how these factors influence the underlying behaviors of both motorized and non-motorized road users. Therefore, alternative surrogate measures for the identification of roadway locations which possess comparatively high safety risks should be investigated. 1.2 Research Approach To address these issues, a field study was performed on low-speed roadways within three Michigan cities to determine factors related to pedestrian safety risk. A variety of existing traffic control devices were considered, including various crosswalk marking strategies, along with additional treatments, including PHBs, RRFBs and single in-street R1-6 signs. A diverse set of roadway and traffic characteristics were also considered, including crossing width, number of lanes, and median presence, along with vehicular, pedestrian, and bicyclist volumes collected during the study period. Three primary evaluations were performed for both segments and signalized intersections, which included: driver yielding compliance, vehicle-pedestrian conflicts, and non-motorized traffic crash data, and attempts were made to examine the relationships between the behavioral measures and the crash data. 4 1.3 Objectives This study sought to identify factors which contribute to pedestrian safety. Traditional crash-based modeling for evaluating pedestrian safety is challenging due to the small number of pedestrian crashes as well as the lack of reliable exposure data for pedestrians. This is reflected in the lack of pedestrian-specific safety performance functions (SPFs) at midblock areas in the Highway Safety Manual (HSM) as well as the lack of statistically significant crash modification factors (CMFs) for the RRFB and in-street R1-6 sign. To address these challenges, the following objectives were set:  To evaluate the safety of pedestrian crossing treatments using a measure of effectiveness other than crashes  To determine the safety impact of cross-sectional and site characteristics other than the pedestrian crossing treatment  To use statistical analysis to directly compare different crossing treatments  To provide for a methodology which states and local agencies can use to evaluate their own pedestrian infrastructure 1.4 Study Constraints In order to meet study objectives, there were several constraints in site selection. Sites were limited to low speed locations on or near large public university campuses in Michigan during daytime hours (i.e., 9:00 AM to 4:00 PM) in the fall. This was done for several reasons:  Pedestrian volume: This study sought to evaluate driver yielding compliance in areas where pedestrian activity already exists, and therefore, selecting sites with moderate-to-high pedestrian volume is imperative. College campuses and surrounding areas are reliable sources of pedestrian traffic. Times were selected during the mid-day to align with 5 pedestrian travel behavior on university campuses. Lastly, observations took place in the fall to allow for observations when class is in session and pedestrian volumes are high, and to avoid the winter months where pedestrian activity would be expected to decrease  Land-use and demographic characteristics: Utilizing sites located on university campuses allows for similar characteristics of land use, zoning, population density, as well as demographic characteristics of drivers and pedestrians, among sites  Driver speed: It is expected that speed is a factor in pedestrian safety. However, the previously mentioned constraints limit the type of road which can be evaluated. On and surrounding the three university campuses studied, speed limits greater than 25 mph are rare, and have low pedestrian activity. Therefore, consistency was provided by only including sites with speed limits of 25 mph. The following chapters describe the data collection and analytical methods along with results, conclusions and recommendations. 6 CHAPTER 2: LITERATURE REVIEW To address pedestrian safety, a variety of pedestrian crossing safety treatments have been devised. The Manual of Uniform Traffic Control Devices (MUTCD) is a manual published by the Federal Highway Administration (FHWA) which lists and provides detail and guidance regarding the use of traffic control devices in the United States [3], and state design guides must be in substantial compliance with the MUTCD in order to provide nationwide consistency. To this end, the MUTCD includes several pedestrian crossing treatments, including crosswalk markings, instreet signs (R1-6), and the pedestrian hybrid beacon (PHB). In addition, FHWA has offered interim approval to the rectangular rapid-flashing beacon (RRFB). In evaluating these treatments, a variety of methods have been explored. The Highway Safety Manual (HSM) provides a method for estimating the mean number of crashes at a site. However, various surrogates for crashes also exist in evaluating safety, such as conflicts (near crashes) as well as driver yielding compliance to pedestrians. These methods for evaluating pedestrian safety, as well as the safety performance of various pedestrian crossing treatments, are presented in the sections below. 2.1 Highway Safety Manual and the Predictive Method The HSM was first published by the American Association of State Highway and Transportation Officials (AASHTO) in 2009. The HSM was created to provide tools to quantitatively measure safety performance with regards to frequency, severity, and type of crashes. The HSM uses two tools, safety performance functions (SPFs), which predict crashes at an intersection or segment under base conditions, and crash modification factors (CMFs), which describe the reduction in crashes when a countermeasure is installed. Although the use of 7 regression modeling to predict crashes and assess safety existed prior to the HSM’s publication, the HSM provided a standard, national reference for highway safety. Part C of the HSM contains the predictive method, which estimates annual average crash frequency as a function of geometric design, the presence and type of traffic control devices, and traffic volume. There are three components to the predictive method: SPFs, CMFs, and calibration factors. SPFs provide an estimate of crashes for a roadway segment or intersection under base conditions as a function of average annual daily traffic (AADT) and segment length, while CMFs are used to take into account the impact of geometric design, traffic control, and any other factor present at an intersection or segment that has an impact on the total number of crashes. Calibration factors account for regional variation. The predictive method allows practitioners to estimate the total number of crashes annually at an intersection or road segment under current conditions, hypothetical or forecasted future conditions, and the total number of annual crashes using an alternative plan. Crashes are countable events which are never negative, which therefore lends itself to using a Poisson distribution to estimate crashes. However, crash data are typically overdispersed, and therefore, a negative binomial distribution is used when developing SPFs. While SPFs in their most basic form take into account only segment length and AADT, CMFs adjust the total number of crashes based on the specific conditions at an intersection. There are CMFs for geometric design features, such as lane and shoulder width, as well as traffic control features such as a protected left turn lane. The CMF is multiplied by the SPF to determine the number of crashes: a CMF >1 indicates that crashes will be higher than base conditions, while a CMF <1 indicates the opposite. The predictive method in the HSM covers rural two-lane two-way roads (Chapter 10), rural multilane highways (Chapter 11), and urban and suburban arterials (Chapter 12). The HSM only 8 considers pedestrian crashes for urban and suburban arterials. Chapter 12 has separate methodologies for predicting vehicle-pedestrian crashes at segments, intersections with signals, and stop-controlled intersections (3-leg, where minor leg is stop controlled, and 4-leg, where two minor legs are stop controlled). The HSM does not provide SPFs for other intersection For predicting crashes for pedestrians at segments, the HSM method takes the base SPF for road segments and multiplies it by an adjustment factor which takes into account the posted speed limit in relation to 30 mph and the type of road (2-lane undivided, 3-lane with two-way leftturn lane, 4-lane undivided, 4-lane divided, and 5-lane with two-way left-turn lane). For pedestrians at stop controlled intersections, the base SPF for intersections is multiplied by an adjustment factor which takes into account intersection type. For predicting vehicle-pedestrian crashes at signalized intersections, the HSM provides a more sophisticated method than adjustment factors, and in fact has developed an SPF and CMFs unique to pedestrians. The formula for total crashes is as follows: 𝑁𝑝𝑒𝑑𝑖 = 𝑁𝑝𝑒𝑑𝑏𝑎𝑠𝑒 ∗ 𝐶𝑀𝐹1𝑝 ∗ 𝐶𝑀𝐹2𝑝 ∗ 𝐶𝑀𝐹3𝑝 , (1) Where,  Npedi = predicted average crash frequency of vehicle-pedestrian collisions  Npedbase = predicted number of vehicle-pedestrian collisions per year for base conditions at signalized intersections  CMF1p, CMF2p, CMF3p, etc. = crash modification factors for vehicle-pedestrian collisions at signalized intersections The base SPF is as follows: 𝐴𝐴𝐷𝑇 𝑁𝑝𝑒𝑑𝑏𝑎𝑠𝑒 = exp⁡(𝑎 + 𝑏 ∗ ln⁡(𝐴𝐴𝐷𝑇𝑡𝑜𝑡 ) + 𝑐 ∗ ln⁡(𝐴𝐴𝐷𝑇𝑚𝑖𝑛 ) + 𝑑 ∗ ln⁡(𝑃𝑒𝑑𝑉𝑜𝑙) + 𝑒 ∗ 𝑚𝑎𝑗 𝑛𝑙𝑎𝑛𝑒𝑠𝑥) (2) 9 Where,  a = intercept term  b=coefficient for AADTtot  AADTtot=AADT for major and minor legs  c= coefficient for 𝐴𝐴𝐷𝑇𝑚𝑖𝑛  AADTmin=AADT for minor leg of intersection  AADTmaj=AADT for major leg of intersection  d=coefficient for PedVol  PedVol=pedestrian volume at intersection  e=coefficient for nlanesx  nlanesx=number of lanes at intersection x 𝐴𝐴𝐷𝑇 𝑚𝑎𝑗 The HSM provides estimates for PedVol based on general levels of activity. The CMFs provided take into account the number of bus stops within 1000 ft of the intersection, the presence of schools within 1000 ft of the intersection, and the number of alcohol sales establishments within 1000 ft of the intersection [2]. The HSM provides additional CMFs in Part D. Those pertaining to vehicle-pedestrian crashes include:  Permit right-turn-on-red [2]  Convert minor-road stop control to all-way stop control [4]  Remove unwarranted signal [5]  Provide intersection illumination [6] The HSM provides no predictive method for pedestrian crashes at midblock crossing locations or anywhere in rural locations, and no provides no CMFs beyond those already 10 mentioned. In addition, the HSM does not provide a distinction between injury and fatal crashes, and assumes no vehicle-pedestrian crashes are property damage only [2]. There are additional CMFs describing vehicle-pedestrian crashes which have not been included in the HSM, but which can be found in the CMF Clearinghouse [cmfclearinghouse.org]. For vehicle-pedestrian crashes, these include:  Install raised pedestrian crosswalks [6]  Provide a raised median [7]  Implement automated speed enforcement cameras [8]  Install bicycle lanes [9]  Install a traffic signal [10]  Provide split phases [10]  Increase cycle length for pedestrian crossing [10]  Install high-visibility crosswalk [10]  Convert from yield signal control to signalized control [11]  Install lighting [12]  Install flashing yellow arrow [13]  Install pedestrian countdown timer [13]  Implement a leading pedestrian interval [14]  Installation of a High intensity Activated crosswalk (HAWK) pedestrian-activated beacon at an intersection [15]  Raised median with marked or unmarked crosswalk at an uncontrolled intersection [16] 11 2.2 Using Conflict as a Surrogate Measure for Crashes Vehicle conflict has been used as a surrogate for crashes as crashes are rare and unexpected events. In presenting the concept of surrogacy, a TRB white paper compares the use of surrogate measures within the field of traffic safety engineering with the use of the same in the medical field. They conclude that acceptable surrogates must be “fully correlated with the clinically meaningful outcome” of reducing or eliminating crashes, and they must “fully capture the effect of the treatment,” meaning the surrogate measure must be physically related to crashes [17]. Surrogates should go beyond mere near crashes, and instead be measures which take into account the mechanisms of crashes. In doing so, a hierarchical Bayesian approach can be adopted to take into account some surrogate measures being more strongly correlated with crashes than others [18]. In doing so, these surrogate measures will capture some, but not all of the factors that lead to crashes [17]. Additionally, just as crashes are rare events, severe conflicts are also rare, which may lead to an under-prediction of crashes when relying on conflict as a surrogate measure [19]. Researchers rarely observe crashes directly during their study period. One problem with validating the relationship between near crashes or conflict and crashes is the study’s duration: typically, the time period in which behavior and vehicle interaction is being studied is a much shorter time period than that in which crash reports are analyzed [19]. However, some studies have been able to collect sufficient data to correlate crashes with conflicts and near crashes. A 2006 study performed by researchers at the Virginia Tech Transportation Institute outfitted 100 vehicles with monitoring equipment which continuously recorded driving data for a period of 1 year, logging approximately 2 million vehicle miles, 43,000 hours of data, and utilizing 241 drivers in order to determine causal relationships between a host of safety related factors. The data collected included crashes and near-crashes, which were defined as rapid evasive maneuvers by 12 the study vehicle. [20]. An evaluation of this data in 2010 looked at the use of conflict as a surrogate measure for motor vehicle crashes, and using Poisson regression found that there was a significant (p-value < 0.001) positive relationship between crashes and near crashes. In particular, the authors endorsed using near-crashes as a surrogate measure for crashes in small-scale studies with a low number of crashes [21], although the focus of these studies was not specific to vehiclepedestrian crashes. 2.3 Using Staged Crossing to Evaluate Pedestrian Crossing Treatments In evaluating the safety performance of pedestrian crosswalk treatments, driver yielding compliance to pedestrians been a primary performance measure [22]. Yielding performance can be measured by utilizing trained staged pedestrians who make street-crossing attempts using standardized procedures while recording driver yielding behavior associated with each attempt. A study published in 2006 by researchers from the Texas A&M Transportation Institute (TTI) used trained staged pedestrians to evaluate yielding compliance of motorists to pedestrians at unsignalized intersections. The authors chose to measure yielding instead of crashes as the measure of effectiveness (MOE) as their review of the existing literature found that the most common MOE for pedestrian crossing treatments was yielding. Staged pedestrians were used to provide a consistent crossing procedure without regional variability, as well as to provide a sufficient sample size. The trained pedestrians approached the pedestrian crossing and indicated their intent to cross by facing the oncoming traffic without stepping into the crosswalk. The pedestrian would only enter into the crosswalk once motorists had yielded [22]. A 2013 study at Western Michigan University used a similar procedure to measure yielding compliance. In order to determine whether the motorists ought to be scored as “yielding” or “not yielding,” study authors applied the concept of the dilemma zone, which is commonly applied 13 when determining the timing of the amber interval at a traffic signal. In order to provide enough time for a driver to react and comfortably decelerate to a stop, the common formula for amber interval timing [23], shown below, was utilized in the formula. 1.47𝑣 𝑌 = 𝑡 + 2(𝑎+𝐺𝑔) (3) The distance required was taken by multiplying the amber indication time by the posted speed limit. As such, vehicles were only scored as not yielding if the staged pedestrian approached the crosswalk prior to the motorist entering the dilemma zone and did not yield. Another deviation from the TTI procedure is that the staged pedestrians indicated their intent to cross by placing one foot in the crosswalk and the other on the curb [24], which more closely follows the typical crosswalk right-of-way laws followed by municipalities in Michigan and elsewhere [25]. The study also compared the yielding results for staged and unstaged pedestrian crossings, and found no significant difference in results, supporting the use of staged pedestrians to measure driver yielding compliance to pedestrians [24]. Later research by the previously mentioned TTI team incorporated these improvements in their staged pedestrian crossing procedures by using the AASHTO stopping sight distance formula to determine dilemma zone, and having staged pedestrians place one foot into the crosswalk to indicate intent to cross. The reason for placing one foot in the crossing to indicate intent is that in Texas, where the study took place, motorists are only compelled by law to yield to pedestrians already in the crosswalk. Additionally, all staged pedestrians wore similar clothing [26]. 2.4 Predicting Pedestrian Crashes Using Behavioral Information A recent study published in 2014 sought to use behavioral information to predict pedestrian crashes at signalized and midblock crossing locations. The research combined observed pedestrian conflicts with crossing distance and building setback. The authors studied 100 pedestrian crossing 14 locations in Connecticut. Sites included signalized and unsignalized mid-block crossings, 3-leg intersections, and 4-leg intersections. The research considered crossing type, traffic control, speed limit, presence of median or pedestrian refuge island, crossing distance, number of lanes, on-street parking, and building setback. Conflicts were observed using a variation of the Swedish Traffic Conflict Technique. Pedestrian crossings were categorized as undisturbed passages, potential conflicts, minor conflicts, or serious conflicts. Vehicular volume was calculated using Department of Transportation (DOT) volume counts, which were adjusted for the time and day of the week that the observations occurred (traffic volume counts were not taken at the time of observation). Pedestrian counts taken during observations were converted to Annual Average Daily Pedestrian Volume (AADPV) using the following formula: 𝐴𝐴𝐷𝑃𝑉 = 𝐴𝐴𝐷𝑇 ∗ 𝑃/𝑉𝑜 ⁡ (4) Where, P = Pedestrian volume during observation period, and Vo = Calculated vehicular volume during observation period. The researchers used negative binomial and ordered proportional odds to estimate crashes. The research found that minor conflicts have a p-value of 0.1628 for predicting KAB crashes, and serious conflicts have a p-value of 0.1318 for predicting KABCO crashes (significance level is 0.10). Greater crossing distance and small building setbacks were associated with larger numbers of pedestrian-vehicle crashes, while pedestrian volume was not significant [27]. 15 2.5 Safety Performance of Midblock Pedestrian Crosswalk Treatments Various forms of pedestrian crossing treatments have been devised, including crosswalk markings, as well as enhancement devices such as the in-street sign, RRFB, and PHB. The following subsections explore the known safety performance of these treatments in detail, which is summarized in Table 2. 2.5.1 Marked and Unmarked Crosswalks Using pavement markings to indicate pedestrian crossing areas is the most basic pedestrian safety treatment. One of the first studies evaluating the safety of marked crosswalks, published in 1972, found that installing pavement markings at crosswalks resulted in an increase in pedestrian crashes at these locations, although the analysis did not consider pedestrian exposure but rather evaluated total crash numbers alone [28]. Since that was published, other studies have come to similar conclusions, including studies in Sweden [29] and Ontario [30], as well as a Swedish study which considered pedestrian exposure and found crash rates increase [31]. These studies evaluated before and after numbers without further statistical analysis to determine whether the change in crashes was significant, nor did they control for regression-toward-the-mean bias or take into account trends in crashes. The Ontario study, for instance, found that crashes were increasing year-over-year in the before period as well as the after period [30]. More recently, studies have analyzed the effects of marking crosswalks in a more comprehensive manner. An evaluation of 2,000 marked and unmarked crosswalks in 30 cities, representing all regions of the United States, found that among locations with marked crosswalks, two-lane roads and locations with raised medians are less crash-prone than marked crosswalks on multilane and undivided roads. However, the authors found that on two-lane roads and multilane roads at traffic volumes with average daily traffic (ADT) of less than 12,000 vehicles per day there 16 was no significant difference in crashes between marked and unmarked crosswalks, even with pedestrian volume included as a factor. When traffic volumes were greater than 12,000 ADT marked crosswalks were associated with an increase in crashes relative to unmarked crosswalks [16]. While most studies evaluate the change in crashes when pavement markings are installed, a study in Israel looked pedestrian behavior when crosswalk markings were removed. They found that pedestrians are more likely to stop and look for traffic at sites where markings were removed, which led to fewer conflicts, but also led to longer waiting times at crossing locations and fewer vehicles yielding to pedestrians [32], indicating that pedestrians will likely choose a marked crosswalk over unmarked when given the choice. When included as a factor in studies of marked crosswalks, pedestrians express a clear preference for marked crosswalks [16]. Due to this preference, it may be safer at a network level to choose safe locations for marked crosswalks rather than eliminate them altogether in order to encourage users to use safe crossing locations. 2.5.2 In-Street Pedestrian Crossing Signs Several studies have shown that treatments can be added to a marked crosswalk to improve pedestrian safety, such the addition of an in-street sign along a roadway centerline advising drivers to yield to pedestrians. This treatment is included in the MUTCD as the R1-6 sign [3]. A 2007 study evaluating four crosswalks in San Francisco found that driver yielding rates at crosswalks treated with the in-street pedestrian crossing signs (R1-6) ranged from 60 percent to 74 percent, while sites without any treatment had yielding rates ranging from 20 percent to 60 percent [33]. Another study conducted in Pennsylvania found this type of treatment to be effective in increasing driver yielding behavior with driver yielding increasing from 17 percent to 24 percent at midblock crossings [34]. A compendium of research on midblock crosswalk treatments found that in-street 17 pedestrian crossing signs may be most effective on two lane roads, although this type of treatment has still been found to increase driver yielding in additional lane configuration situations [35]. Adding two additional R1-6 signs to both edgelines in addition to the centerline in the Gateway configuration (Figure 1) has been shown to increase yielding rates more than a single R1-6 sign. In one Michigan study, driver yielding compliance rates went from 25 percent with markings alone to 57 percent with a single R1-6, but increased to 82 percent when signs were installed in the Gateway configuration. Other sites in Michigan showed similarly dramatic increases in driver yielding compliance in spite of this treatment costing as low as $450 per sign [13]. Guidance from the Michigan Department of Transportation recommends the Gateway treatment when traffic volumes are less than 12,000 ADT in most cases, or 25,000 ADT on threelane roads with pedestrian refuge islands based on prior research of the effectiveness of the Gateway treatment [36]. The R1-6 treatment also has a channelizing effect for pedestrians, with pedestrians choosing treated crosswalks over those which are unmarked or with markings alone [34] [37] [38]. 2.5.3 Rectangular Rapid Flashing Beacon The RRFB has been shown to increase driver yielding rates. A before and after analysis by Brewer found that yielding rates at sites with the RRFB treatment increased by a range of 35 percent to 79 percent. Pedestrian compliance with RRFB treatment was also strong, with 94 percent of non-staged pedestrians activating the treatment [39]. Similarly, at a high-volume shareduse trail crossing location in Florida, yielding by drivers increased from 2 percent to 35 percent after the treatment was installed. Looking exclusively at when the beacon was activated, driver yielding increased to 54 percent. However, unlike the study previously mentioned, user activation of the beacon was much lower, with 32 percent of users activating the beacon, and only 51 percent 18 of users crossing when the beacon was activated [40]. However, it is important to note that this location is a shared-use path with large bicycle volumes [40], unlike the previous study which observed a pedestrian only facility [39]. In subsequent studies, a sign was added near the push button saying, “push button to activate beacons” to improve pedestrian compliance, but the low beacon activation rate persisted [40]. A 2010 paper published by FHWA observed a more geographically diverse set of sites, with locations in St. Petersburg, Florida, Washington, D.C., and Mundelein, Illinois, to note how the RRFB impacted driver yielding behavior. The research found that in St. Petersburg, using 4 RRFBs (89 percent average driver yielding rate) was more effective for driver yielding than the typical 2 RRFB (82 percent average driver yielding rate) setup, and both setups were more effective for driver yielding than no treatment at all (18 percent average driver yielding rate). Additionally, yielding drivers left more distance between their front bumper and the crosswalk when compared with the baseline treatment. However, the research did not find this type of treatment had a significant impact on evasive behavior by either pedestrians or motorists. Similarly, the research found that the RRFB treatment was also associated with increased driver yielding in Washington, D.C. (yielding was 1.7 percent with no treatment, and 85 percent with the RRFB treatment), and that yielding drivers left more room between their front bumper and the crosswalk when compared with the baseline treatment, as previous. The research also considered modifying the RRFB to flash its LED in the drivers’ eyes. FHWA found that yielding increased from 0 percent with no treatment to 80 percent with the typical RRFB treatment and to 89 percent with the beacon’s LED flashing in the drivers’ eyes. Lastly, the research also looked at combining the RRFB with advanced warning devices, which resulted in no change in driver yielding 19 compared to a RRFB treatment with no advanced warning signs [41]. The RRFB has been shown to be more effective when mounted overhead than when mounted on either side of the road [42]. A study in Bend, Oregon found that in addition to increasing driver yielding from an average rate of 17.8 percent before treatment to 79.9 percent after treatment, the RRFB also significantly reduced pedestrian motorist conflicts from 4.4 per 100 crossings to 1.4 per 100 crossings. Motorist speeds were also reduced [43]. More recently, a CMF was developed for the RRFB, which found a 47 percent reduction in crashes [44]. However, the result was not statistically significant, and therefore questions remain as to the effect that this treatment has on crashes. 2.5.4 Pedestrian Hybrid Beacon The PHB, also known as the High-intensity Activated crossWalK beacon (HAWK), is a crosswalk treatment that is effective in improving pedestrian safety. Shurbutt found a 69 percent reduction in pedestrian crashes, a 15 percent reduction in severe crashes, and a 29 percent reduction in total crashes when the PHB treatment was applied [41]. Similarly, another study found that the PHB was associated with a 28 percent reduction in total crashes and a 58 percent reduction in pedestrian crashes [42]. Additionally, research by Fitzpatrick (2016) found that pedestrian and motorist compliance with the PHB was strong: a study of 20 locations in Austin, Texas and Tucson, Arizona found that only 6 percent of pedestrians crossed during the beacon’s dark indication and that driver yielding when the PHB was activated was 96 percent. The same study found that about half of vehiclepedestrian conflicts, defined as events when a vehicle or pedestrian takes evasive action to avoid a collision, occurred when the beacon was dark. Furthermore, in one study, PHB installation was correlated significantly with an increase in pedestrian volume at the treatment location. Out-of- 20 crosswalk pedestrian volume also increased, with anecdotal evidence showing that out-ofcrosswalk pedestrians were typically following the PHB indications [45]. A summary of the safety performance of these treatments is presented in Table 2 below. TABLE 2. Safety Performance of Crossing Treatments from Literature Treatment Typical Measure of Effectiveness Safety Performance  Crosswalk markings Crashes   In-street signage Driver yielding compliance RRFB Driver yielding compliance PHB Crashes    Crosswalk markings associated with an increase in both crashes [28] [30] and crash rate [29] compared to unmarked crosswalks Crosswalk marking associated with no significant change in crashes at ADT<12,000 compared to unmarked crosswalks [16] Single in-street R1-6 sign associated with increases in driver yielding compliance to pedestrians [3] [33] [34] [35] Multiple R1-6 signs in the “gateway” configuration have generated yielding compliance comparable to PHB [13] Associated with yielding compliance improvements, including at high volume sites [40] Associated with reductions in pedestrian crashes [41] [42] 21 CHAPTER 3: DATA COLLECTION In order to assess the safety performance of various pedestrian crossing treatments, it was initially necessary to collect data specific to existing locations in the field where such treatments have been implemented. First, this involved the identification of sites which possess varying geometric, operational, and other highway characteristics in addition to the pedestrian crossing treatment of interest. After the selection of appropriate field locations, behavioral data was collected in the field at each site, including data for both staged and naturalistic crossing events, in order to assess driver compliance to traffic control as well as quantify the occurrence of conflicts. Historical traffic crash data were also collected for each site from the annual databases maintained by the Michigan State Police. The data collection activities for this study are detailed in the subsections that follow. 3.1 Site Selection The study locations were selected to provide diversity among existing crosswalk treatments and roadway characteristics, along with a range of vehicular and pedestrian volumes. This included the identification of both midblock crossings (including uncontrolled legs at two-way stopcontrolled intersections) as well as signalized intersections. To ensure adequate pedestrian activity, the locations were selected on or near college campuses or commercial business districts. A total of 66 sites were selected, including 40 uncontrolled midblock locations and 26 signalized intersections. Sites were selected to provide a broad range of site and cross-sectional characteristics (i.e., lane width, presence of auxiliary lanes, type of crosswalk marking) with moderate to high pedestrian volume. 22 The sites were selected from three Michigan cities and all sites were on or near major university campuses, which provided for a degree of consistency in terms of land use characteristics in addition to having similar driver and pedestrian demographics among all sites. This included 35 sites from the midtown area of Detroit (Wayne State University), 20 sites from East Lansing (Michigan State University), and 11 sites from Kalamazoo (Western Michigan University). Relevant site characteristics, including crosswalk treatment, crossing distance, median presence, pedestrian signage, lighting, speed limit, and access point density, as well as other highway features, were initially collected using Google Earth satellite imagery and were later validated in the field. Table 3 shows the distribution of the study sites by crossing type and city for both the midblock crossing locations and signalized intersections. Tables 4 and 5 display the basic site characteristics for the 40 midblock crossing locations and 26 signalized intersections included in the study, respectively. Aerial photos of site locations are provided in the Appendix. As it was not possible to obtain speed data during the field data collection, in order to control for operating speeds, only sites with posted speed limits of 25 mph were selected. Furthermore, few sites with speed limits greater than 25 mph met site selection criteria (i.e., on or near a university campus, high pedestrian activity) and sites that did meet this criteria had distinguishing features which made direct comparison with other sites difficult. Thus, the results of this study are limited to low speed locations. TABLE 3. Number of Study Sites by Crossing Type and City Type of Crossing Detroit East Lansing Kalamazoo TOTAL Uncontrolled Midblock 14 18 8 40 Signal Controlled 21 2 3 26 23 TABLE 4. Characteristics of Midblock Crosswalk Study Sites Cross Street or Landmark Total Street Crossing Dist. (ft.) Crosswalk Type Median Presence Atchison Hall 61 Continental Yes W. Palmer Ave. 102 Continental Yes PS 5 94 Continental Yes W. Hancock St. 65 Unmarked Yes W. Ferry Ave. 94 Continental Yes PS 1 58 Continental Yes Cass Ave. W. Kirby St. 50 Unmarked No Detroit Cass Ave. Kohn Building 48 Continental No 9 Detroit Cass Ave. Prentis St. 50 Unmarked No 10 Detroit Cass Ave. W. Ferry Ave. 46 Unmarked No 11 Detroit Matthaei Center 40 Continental No 12 Detroit Lodge Service Dr. W. Palmer Ave. Shapero Hall 69 Continental Yes 13 Detroit John R St. Garfield St. 52 Continental No 14 Detroit Cass Ave. W. Willis St. 46 Unmarked No Bogue St. Snyder Hall 51 Continental Yes Chestnut Rd. Wilson Hall 30 Continental No E. Circle Dr. Olin Health Center 30 Continental No Site No City 1 Detroit 2 Detroit 3 Detroit 4 Detroit 5 Detroit 6 Detroit 7 Detroit 8 15 16 17 18 19 20 21 22 23 24 E. Lansing E. Lansing E. Lansing E. Lansing E. Lansing E. Lansing E. Lansing E. Lansing E. Lansing E. Lansing Primary Street Anthony Wayne Dr. Anthony Wayne Dr. Anthony Wayne Dr. Anthony Wayne Dr. Anthony Wayne Dr. W. Palmer Ave. E. Grand River Ave. Red Cedar Rd. Red Cedar Rd. Charles St. 53 Standard Yes Eng. Building 54 Continental No Spartan Stadium 26 Continental No S. Shaw Ln. Anthony Hall 24 Continental Yes N. Shaw Ln. Erickson Hall 24 Continental Yes N. Shaw Ln. Intl Center 24 Continental Yes N. Shaw Ln. Planetarium 22 Continental Yes 24 GPS Coordinates (latitude, longitude) 42.355525, 83.071787 42.359443, 83.073668 42.358867, 83.073625 42.353380, 83.070549 42.359004, 83.073511 42.361002, 83.071057 42.358933, 83.068318 42.360288 83.069138 42.352980, 83.064925 42.360655, 83.069312 42.355802, 83.075208 42.360128, 83.072945 42.354613, 83.060026 42.350500, 83.063508 42.730708, 84.471992 42.723477, 84.487475 42.732702, 84.479157 42.734166, 84.479936 42.724922 84.482371 42.730383, 84.485680 42.72522 84.478992 42.725996, 84.478987 42.726024, 84.48091 42.726009, 84.476540 TABLE 4. (cont’d) Site No 25* 26 27* 28* 29 30 31 32 33* 34* 35* 36 37* 38 39 40* City Primary Street Cross Street or Landmark Total Street Crossing Dist. (ft.) Crosswalk Type Median Presence GPS Coordinates (latitude, longitude) 42.726029, 84.475715 42.725352, E. Lansing N. Shaw Ln. Holmes Hall 47 Continental Yes 84.463949 42.725444, E. Lansing N. Shaw Ln. Holmes Hall 47 Continental No 84.464775 42.725350, E. Lansing N. Shaw Ln. Holmes Hall 29 Continental No 84.464765 Gd River 42.732966, E. Lansing W. Circle Dr. 25 Continental No Ramp 84.481141 Wharton 42.723256, E. Lansing Wilson Rd. 50 Continental Yes Center 84.469874 E. Wilson 42.721974, E. Lansing Wilson Rd. 28 Continental No Hall 84.488592 W. Wilson 42.72195, E. Lansing Wilson Rd. 28 Continental No Hall 84.489091 W. Michigan 42.284981 Kalamazoo Student Rec 40 Standard No Ave. 85.609556 Univ Prog 42.285626 Kalamazoo Dormitory Rd. 22 Standard No Bldg 85.610532 42.286131, Kalamazoo W. Walnut St. Health Plaza 73 Standard No 85.580328 Knollwood Western 42.279918, Kalamazoo 26 Continental No Ave. View 85.620105 42.28278, Kalamazoo Rankin Ave. Welborn Hall 40 Standard No 85.61954 Western 42.286838 Kalamazoo Gilkison Ave. 32 Standard No Heights 85.614961 Goldsworth 42.288534 Kalamazoo Valley Pond 38 Standard No Dr. 85.615437 42.286639 Kalamazoo Dormitory Rd. PS 1 41 Continental No 85.610737 Note: an asterisk indicates that staged crossing data were not collected at this location E. Lansing N. Shaw Ln. Shaw Hall 24 25 Continental Yes TABLE 5. Characteristics of Signalized Intersection Study Sites City Primary Street Cross Street Average Street Crossing Dist (ft) Crosswalk Type Right-Turnon-Red Permitted 41 Detroit 2nd Warren 71 Continental Yes 42 Detroit Lodge Service Dr Warren 59 Continental No 43 Detroit Randolph Jefferson 98.5 Continental Yes 44 Detroit Cass Palmer 61.5 Continental No 45 Detroit Cass Putnam 44 Continental Yes 46 Detroit Cass Library 49 Continental No 47 Detroit 2nd Forest 53.5 Continental No 48 Detroit Trumbull Warren 54.5 Standard No 49 Detroit Anthony Wayne Dr Forest 57 Continental No 50 Detroit Cass Forest 45 Continental No 51 Detroit Cass Antoinette 43 Standard No 52 Detroit Cass Milwaukee 43.5 Standard No 53 Detroit Shelby Lafayette 38.5 Continental No 54 Detroit Shelby Fort 49.5 Continental Yes 55 Detroit Cass Fort 60 Continental No 56 Detroit Washington Congress 46 Continental No 57 Detroit Washington Larned 47 Continental Yes 58 Detroit John R Warren 69 Standard No 59 Detroit Cass Michigan 79 Continental No 60 Detroit 3rd Michigan 86.5 Continental No 61 Detroit Woodward Jefferson 91 Continental No 62 E. Lansing Farm Lane River Trail 40 Continental No 63 E. Lansing Red Cedar South Shaw 40 Continental Yes 64 Kalamazoo Dormitory Michigan 49 Brick No 65 Kalamazoo Howard Michigan 83.5 Standard No 66 Kalamazoo Howard Valley 57 Standard No Site No 26 GPS Coordinates (latitude, longitude) 42.35523, -83.06869 42.353966, 83.073030 42.329755, 83.041970 42.361390, 83.069739 42.356984, 83.067176 42.358282, 83.067948 42.353082, 83.067608 42.352160, 83.078716 42.352304, 83.069939 42.353745, 83.065361 42.363285, 83.070801 42.368564, 83.074049 42.33101, -83.04943 42.330134, 83.048774 42.329286, 83.050711 42.328742, 83.049213 42.327990, 83.048613 42.357710, 83.062145 42.331683, 83.052623 42.331653, 83.057522 42.328695, 83.044569 42.727240, 84.477869 42.725437, 84.482282 42.284827, 85.610292 42.281626, 85.621621 42.286532, 85.623164 3.2 Field Data Collection After the selection of sites was completed, observational field data related to the behavior of motorists and pedestrians during crossing events were collected during August, September, and October of 2015. The data were collected during daytime periods and under fair weather conditions for two to four hours per site, which was chosen to provide for high pedestrian volume. Covertly positioned elevated high-definition video cameras were temporarily installed at each location to record the staged pedestrian crossing attempts along with vehicle and pedestrian volumes. The videos were later reviewed to extract volume and behavioral information. Using video recordings provided two primary advantages over using on-site human observers: 1) the number of necessary field personnel at each site was reduced and 2) permanent record of the interactions was provided, which improved training and quality assurance procedures. Figure 2 displays an example of the video camera setup and field-of-view. Figure 2. Typical Video Camera Setup for Recording Motorist Yielding Behavior 3.2.1 Staged Pedestrian Crossing Events Staged pedestrian crossing events were utilized for the assessment of driver yielding compliance, and took place at 31 midblock crossing locations. The staged crossing events utilized observers trained to follow a uniform crossing protocol for each approaching driver, thereby reducing external bias. Consistency was provided among the positioning, stance, gesture, eye contact, and aggressiveness used by the pedestrian while entering the crosswalk, in addition to 27 control over external features such as the style and conspicuity of clothing. The staged crossing events also ensured a sufficient sample size at each location, which improved data collection efficiency at locations with low pedestrian crossing volumes. The staged crossing events followed protocols established in prior research [13] [35]:  The staged pedestrian approached the crossing at any time when approaching vehicles were within sight of the crossing. Where present, active devices (PHB, RRFB) were activated at this time. Staged crossing attempts were avoided while other pedestrians were attempting to cross the same crosswalk.  The staged pedestrian indicated an intention to cross by standing at the curb or roadway edge with one foot in the crosswalk and facing oncoming traffic. This action occurred when the vehicle approached a predetermined location upstream of the crosswalk, which was determined using the standard kinematic equation for the timing of an amber interval at a traffic signal based on the default reaction time (1.0 s) and deceleration rate (10 ft/s 2) parameters, provided earlier in equation 1. For 25 mph, this distance was calculated to be 104 ft, which was rounded to 110 ft in order to provide additional buffer space for the driver to make the yielding decision, reflecting the minimum value of 3.0 s for yellow light timing [23]. This distance was measured from the near edge of either the crosswalk, stop line, or pedestrian landing and was marked with a roadside object (Figure 3). In this manner, motorists were afforded ample distance to comfortably stop for the staged pedestrian. Vehicles already beyond this boundary point when the crossing was initiated were considered too close to comfortably stop and were not considered.  The staged pedestrian began to cross when the motorist in the nearest lane had begun to yield and maintained eye contact with the motorist at all times. 28  If additional vehicles were approaching from other lanes, the staged pedestrian crossed halfway into the lane where a motorist had already stopped or yielded and waited until the intention of the approaching motorist was determined. This process was completed as many times as necessary to cross the entire roadway or reach a median.  After concluding the midblock crossing, the procedure was then repeated from the opposite direction at the same crosswalk. An event was classified as a yielding event when a motorist that was initially positioned upstream of the 110 ft boundary point at the start of the staged crossing attempt slowed or stopped to allow the pedestrian to safely cross. For motorists in the nearest lane to the pedestrian, the yielding assessment was made on the basis of the initial intention to cross the roadway. For motorists in the additional lanes, in either the same or the opposite direction, this assessment was made once the pedestrian had crossed to within a half-lane distance of their position. Vehicles in lanes other than the near lane were evaluated irrespective of whether a vehicle in the near lane was present or yielded. These procedures are consistent with the crosswalk right-of-way requirements included within the Uniform Traffic Code for Cities, Townships, and Villages that has been adopted as a local ordinance by many Michigan municipalities [25], including all three cities studied. Staged crossing events, used to evaluate driver yielding compliance at uncontrolled midblock crosswalks, were recorded on a per-event basis. An example of a staged pedestrian indicating intent to cross is shown in Figure 4. 29 Figure 3. Location of Dilemma Zone Relative to Crosswalk Figure 4. Screenshot of Staged Crossing Attempt 30 3.2.2 Naturalistic Pedestrian Crossing Events Naturalistic driver yielding compliance for vehicles turning on permissive signal indications was also recorded during naturalistic pedestrian crossing events at signalized intersections. According to state law, during a permissive signal indication, the driver would must yield to pedestrians in this scenario [25]. Thus, driver yielding compliance was scored accordingly for each permissive turning event where pedestrians were present either at or within the crosswalk. 3.2.3 Pedestrian-Vehicle Conflicts In addition to the staged crossing events, the data related to pedestrian-vehicle conflicts were also collected. The pedestrian conflict data were collected from the aforementioned highdefinition videos. Each video was manually reviewed to classify the types and frequency of evasive maneuvers taken by either party at each of the midblock and signalized intersection locations. The purpose of recording the naturalistic (i.e., not staged) events was to gather ancillary data on evasive maneuvers taken by motorists or pedestrians when the driver (or pedestrian in some cases) did not properly yield the right-of-way. Conflicts were defined as cases where the driver or pedestrian took evasive action to avoid a collision. A vehicular evasive maneuver was recorded if the driver had to take evasive action such as swerving or extreme braking to avoid striking a crossing pedestrian. Alternatively, a pedestrian evasive maneuver was recorded if the pedestrian had to take evasive action such as hurried walking or stepping back to the curb to avoid a collision with a motorist. 3.2.4 Road User Volumes Volumes of vehicles, bicycles, and naturalistic (i.e., non-staged) pedestrian crossings were collected from the videos at each study location during the study period. Pedestrians that crossed within 10 ft of the crosswalk were included in the pedestrian crossing volume for the particular 31 crosswalk. Bicyclists were only counted if using the bike lane or traffic lane. Bicyclists utilizing the sidewalk were not counted as a part of this study, but were included as pedestrians if crossing at the crosswalk. All volume data were tallied in 15-minute intervals and were subsequently converted to hourly volumes. Where multiple crosswalks existed at a single location, the pedestrian volumes for each crosswalk were averaged and converted to an hourly volume. 3.3 Pedestrian Crash Data Collection In addition to evaluating driver yielding compliance, traffic crash data were obtained from queries of the annual traffic crash databases maintained by the Michigan State Police for the period of 2005 – 2014 for each study location. This period was utilized due the relative infrequency of vehicle-pedestrian crashes, although it is acknowledged that uncontrolled changes will have occurred at each site during this time period. Historical traffic crashes were selected from each of the ten annual databases by comparing the location associated with each crash to the particular study location. After the initial query of crashes from the annual statewide databases was completed, a secondary screening was performed in order to ensure crashes were selected which were truly occurring at the specified locations. This involved obtaining the Michigan UD-10 crash report form associated with each crash from the Michigan Traffic Crash Report System also maintained by the Michigan State Police. After each crash report form was collected, the responding officer’s narrative and description of the crash was reviewed in order to determine the precise location of the crash. A key component of this manual review was to identify pedestrian and bicycle crashes which truly occurred along the segment or specific crossing location of interest. Figure 5 shows the diagram included in a typical UD-10 crash report form for two different crash events occurring at the same site. Science Road (running North-South) is stop controlled, 32 while Shaw Lane (running East-West) is uncontrolled. Crash 1, shown on the left in Figure 5, which occurred in the crosswalk crossing Science Road would be categorized as having occurred at the stop-controlled leg of the, and therefore would not be included as a crash for the midblock crosswalk analysis. Crash 2, on the other hand, occurred on the crosswalk crossing Shaw Lane, which is uncontrolled, and therefore was included in the midblock crosswalk crash analysis. Figure 5. Distinction between Pedestrian Crashes at a Minor Street Intersection: (1) Stop Controlled Leg Crash vs. (2) Uncontrolled Midblock Crosswalk Crash The pedestrian crashes were initially investigated on a per-crosswalk basis. In order to reduce the impact of crash coding inaccuracies and to capture a slightly broader area of influence of the subject crosswalk, rather than simply within the crosswalk itself, a 150 ft buffer distance on either side of the crosswalk along the subject roadway was utilized for the crash query. This distance was truncated to exclude the influence area of any nearby traffic signals or stop controlled intersections. Upon completion of the crash data review for each crosswalk, it was determined that only 14 pedestrian crashes occurred within 150 feet of the 40 midblock crosswalks during the entire 10- 33 year period of investigation. These 14 crashes occurred at 11 crosswalks, while 29 of the crosswalks did not experience a single pedestrian crash during the 10-year period. The maximum number of pedestrian crashes at any given crosswalk during the 10-year period was two. Due to the lack of crashes over a 10 year period, crash analysis performed at the site level (i.e. at a particular midblock crossing location or crossings at signal or stop controlled intersections) was not included in the final analysis. To expand the sample of crashes for analysis, it was decided to expand the query to include crashes that occurred along the entire homogeneous uncontrolled segment of roadway adjacent to the subject crosswalk. A segment was considered homogeneous if it maintained the same crosssectional features (i.e., laneage, roadway width, and median presence/absence) and no stop, yield, or signal control for vehicles along the subject roadway. Segment endpoints were thus defined by the first stop sign, yield sign, traffic signal, or change in cross-section encountered along the subject roadway. This process yielded a total of 25 unique uncontrolled midblock segments, as several segments included two or more of the individual study crosswalks. In such cases, the site data collected at the individual crosswalks were aggregated across the entire segment. 34 CHAPTER 4: CRASH ANALYSIS Pedestrian crash data for 25 homogeneous uncontrolled segments were utilized for the crash data analysis, as initial screening of the pedestrian crash data at individual crosswalk level yielded impractically small samples for analysis. It is again noted that the segments were defined as homogenous roadway sections which maintain the same cross-sectional features (e.g., roadway width, laneage, median presence, etc.) with no stop signs, yield signs, or traffic signals along the subject roadway (stop or yield signs may have existed on the cross-streets or driveways). The segment start and end points were defined by a traffic control signal, stop sign, yield sign, or change in primary cross-sectional characteristics. For segments which contained multiple crosswalks from which volume and behavioral information were extracted, values were averaged to in order to conduct the analysis of historical crash data. Segment endpoints, as well as the number of crosswalks and driveways within each segment, are shown in Table 6. The crash data included the most recent 10 years of data (2005 – 2014). 35 TABLE 6. Characteristics of Midblock Study Segments Cluster 5* NodeToNode Distance (ft) 555 Cluster 6* 550 2 3 Anthony Wayne Kirby Palmer 17* 675 1 6 Palmer 2nd Cass 22* 2870 1 6 Lodge Svc Dr Trumbull Warren 27* 775 2 3 John R Forest Canfield 102 1760 4 1 Bogue Gd River Lansing River Tr 103 1150 3 5 Chestnut Shaw Wilson 104 1350 7 4 E Circle W Circle Farm Ln 107 765 1 2 Gd River M.A.C. Division 109 705 1 5 Red Cedar S Shaw Wilson 110 2375 6 8 Red Cedar N Shaw Chestnut 114 1090 3 2 S Shaw Red Cedar Farm Ln Cluster 3 1085 3 5 N Shaw Red Cedar Farm Ln Cluster 4 1530 5 8 N Shaw Farm Ln Bogue Cluster 1 1990 7 7 Shaw Owen Entrance Hagadorn 125 4305 16 7 W Circle Beal Kalamazoo 126 1975 2 6 Wilson Bogue Shaw Cluster 2 795 3 6 Wilson Birch Chestnut 213^ 690 3 9 Knollwood Michigan Auditorium 215^ 845 1 1 Gilkison Parking Dormitory Rd 217^ 2065 2 2 Rankin Valley Dormitory Rd 201^ 605 1 2 Michigan Western Ave Dormitory Rd Cluster 7^ 885 2 2 Dormitory Rd Michigan Tennis Courts 208^ 1020 2 7 Walnut Burdick Jasper 214^ 580 2 3 Rankin Michigan Business Ct Site ID Number of Crosswalks Number of Driveways Primary Street 1 1 Palmer Anthony Wayne 2nd Endpoints Note: Detroit segments denoted by a (*), and Kalamazoo sites by a (^) 4.1 Data Summary After compiling the crash data by segment, a series of basic graphical displays were generated and data screening measures were performed. Figures 6 and 7 depict the 10-year pedestrian crashes normalized per crosswalk (Figure 6) and per mile (Figure 7) for each observed segment along with hourly vehicular and pedestrian crossing volumes. From these figures it 36 appears that very little, if any, trends can be observed between pedestrian crashes and vehicular volumes and especially between pedestrian crashes and pedestrian crossing volumes. The relationship between pedestrian crashes and volumes was further investigated using negative 2500 2000 1500 1000 500 214 208 201 Cluster 7 217 215 213 126 Cluster 2 125 Cluster 1 Cluster 4 Cluster 3 114 110 109 107 104 103 27 102 22 17 Cluster 6 0 Site Number Pedestrian Crash per Marked Crosswalk Vehicle Hourly Volume Vehicle Hourly Volume and Pedestrian Crossings per Hour 4.00 3.50 3.00 2.50 2.00 1.50 1.00 0.50 0.00 Cluster 5 Pedestrian Crashes per Marked Crosswalk binomial modeling techniques, as described in the following subsection. Pedestrian Crossings per Hour Figure 6. Pedestrian Crashes per Marked Crosswalk with Hourly Vehicular Traffic 2500 25 2000 20 1500 15 1000 10 5 500 0 0 Site Number Pedestrian Crashes per Mile Vehicle Hourly Volume Vehicle Hourly Volume and Pedestrian Crossings per Hour 30 Cluster 5 Cluster 6 17 22 27 102 103 104 107 109 110 114 Cluster 3 Cluster 4 Cluster 1 125 126 Cluster 2 213 215 217 201 Cluster 7 208 214 Pedestrian Crashes per Mile Volume and Hourly Pedestrian Crossings by Site Pedestrian Crossings per Hour Figure 7. Pedestrian Crashes per Mile with Hourly Vehicular Traffic Volume and Hourly Pedestrian Crossings by Site 37 A summary of the traffic crash data and relevant site characteristics for the 25 midblock segments analyzed is provided in Table 7. TABLE 7. Descriptive Statistics for Analysis of Pedestrian Crashes on Midblock Segments Factor Level or Unit Mean Std. Dev. Pedestrian Crashes Segment Length Hourly Pedestrian Vol. Hourly Bicycle Vol. Hourly Vehicular Vol. Uncontrolled Marked Crosswalk Density Driveway Density Min Max Ten year total Miles Pedestrians/hour Bicycles/hour Vehicles/hour 1.2 0.25 85.82 6.73 459.8 1.98 0.17 72.03 8.25 441.81 0 0.1 10.5 0 74.8 8 0.82 282.14 30.67 2,329.20 Per mile 13.05 6.27 1.84 27.38 Per mile 24.21 Two-Way Two0.64 Lane (Baseline) Multilane Cross-section 0.08 Undivided Multilane Divided 0.28 No Additional 0.56 Lanes (Baseline) Bicycle Lane* 0.32 Auxiliary Laneage Shoulder 0.04 Parking Lane* 0.12 Standard Crosswalk 0.28 Crosswalk treatment (Baseline) Continental 0.72 Crosswalk *Certain segments had both a bike lane and a parking lane 15.18 6.25 68.87 - 0 1 - 0 1 - 0 1 - 0 1 - 0 0 0 1 1 1 - 0 1 - 0 1 Overall, the segments evaluated as a part of this study averaged approximately one quarter mile in length, with the shortest segment measuring a tenth of mile and the longest homogenous segment measuring more than four-fifths of a mile. Additionally, the study segments experienced 1.2 pedestrian crashes on average over the 10-year analysis period, with several segments experiencing zero pedestrian crashes and one segment experiencing eight crashes. With respect to the number of marked crosswalks, on average the study segments contained approximately 13 38 crosswalks per mile, with a minimum crosswalk density of 1.84 per mile and a maximum of 27.4 per mile. The number of access points averaged 24.2 per mile across all study segments with a minimum density of 6.25 per mile and a maximum of 68.9 per mile. Approximately 28 percent of the study segments were multilane divided highways, eight percent multilane undivided highways, and 64 percent two-lane two-way highways. Approximately 12 percent of the study sample included segments which included parking lanes. 4.2 Analytical Procedures For estimating a number of expected events given random data, the Poisson distribution is usually the most appropriate model. However, one of the underlying assumptions of the Poisson distribution is that the variance is equal to the mean, which is oftentimes not the case in the analysis of traffic safety data. In this case, the negative binomial distribution was used to address the dispersion of the pedestrian crash data between the segments. In fact, the HSM encourages using the negative binomial distribution for estimating or predicting crashes [2]. The negative binomial is a generalized form of the Poisson model. In the Poisson regression model, the probability of road segment i experiencing yi events during a specific period is given by: y P(yi ) = EXP(−λi )λi i yi ! (5) where P(yi) is probability of segment i experiencing yi events during the period and λi is equal to the expected number of events for the segment, E[yi]. Poisson regression models are estimated by specifying this Poisson parameter λi as a function of explanatory variables. The most common functional form of this equation is λi = EXP(βXi), where Xi is a vector of explanatory variables (e.g., AADT, segment length, etc.) and β is a vector of estimable parameters. The negative binomial model is derived by rewriting the Poisson parameter for each segment i as λi = EXP(βXi 39 + εi), where EXP(εi) is a gamma-distributed error term with mean 1 and variance α. The addition of this term allows the variance to differ from the mean as VAR[yi] = E[yi] + αE[yi]2. The α term is also known as the over-dispersion parameter, which is reflective of the additional variation in event counts beyond the Poisson model (where α is assumed to equal zero, i.e., the mean and variance are assumed to be equal). One concern with using crash data from three different cities in Michigan is unobserved heterogeneity between cities, as each city and college campus has different characteristics which cannot be accounted for, such as driver and pedestrian demographics, land use characteristics, and policies for maintaining pavement markings and signs. In order to account for these differences, a city-specific random effect was incorporated into the model. 4.3 Results and Discussion Several versions of the pedestrian crash model were estimated. Variables were removed (and in some cases re-added) in a stepwise manner. Most significantly, it was found that neither hourly vehicular traffic volumes, nor yielding compliance, nor vehicle-pedestrian conflicts were significant predictors for pedestrian crash occurrence. The final negative binomial model results for estimating pedestrian-vehicle crashes at midblock segments are shown in Table 8, which includes the parameter estimate, standard error, and the exponential of the parameter estimate (for cases where the natural logarithm of the factor was not taken), and p-value for each. It should be noted that the natural logarithms were taken of segment length, crosswalk length, and the hourly pedestrian volume. This conversion allows for the associated parameter estimates (β) to be more easily interpreted when determining the elasticity of the parameter with respect to traffic crash occurrence. Specifically, the parameter estimates for the log transformed variables represent the percent increase in crashes associated with a one-percent increase in the 40 specific variable. For the binary variables, the pseudo-elasticity (shown as follows) represents the percent change in crashes when the binary variable is changed from zero to one: 𝜆 𝐸𝑥𝑖𝑗𝑖 = 𝐸𝑋𝑃(𝛽𝑗 )−1 𝐸𝑋𝑃(𝛽𝑗 ) , (6) TABLE 8. Negative Binomial Results for Vehicle-Pedestrian Crashes on Uncontrolled Midblock Segments Parameter Estimate Std. Error z value Pr(>|z|) Intercept -22.5 5.01 Segment Length (ln ft) 1.64 0.426 Hourly pedestrian volume (ln) 0.774 0.255 Average crosswalk length (ln ft) 2.06 0.737 <13 Crosswalks per mile baseline 13-18 Crosswalks per mile 2.72 0.926 >18 Crosswalks per mile 1.99 0.662 Standard crosswalk baseline Continental crosswalk -1.58 0.717 No auxiliary lane present baseline Auxiliary lane present -0.787 0.499 Overdispersion parameter 9.42E-05 Note: response variable is 10-year pedestrian crash frequency Odds Ratio -4.49 3.85 3.04 2.80 <0.001 <0.001 0.00240 0.00507 2.94 3.01 0.00331 0.00263 15.2 7.34 -2.21 0.0273 0.206 -1.58 0.115 0.455 Not surprisingly, the results show that an increase in segment length is associated with a corresponding increase in vehicle-pedestrian crashes. This is consistent with prior research, for which the primary factors in predicting crashes at segments are segment length and vehicular volume [2], although a relationship between crashes and vehicular volumes was not found here, likely due to the small crash sample size. The number of vehicle-pedestrian crashes also increased as hourly pedestrian volumes increased, which is in general agreement with the models presented in the HSM [2]. Initial models showed a positive correlation between driveway density and pedestrian crashes. Although no existing studies linking driveway density with pedestrian crashes in 41 particular could be found, the result is consistent with existing research showing a positive relationship between driveway density and total crashes [2]. However, further analysis found a correlation between driveway density and crosswalk density (Figure 8), with one or the other being found significant but not both. Due to the manner in which pedestrian related crashes were collected (recall Chapter 3) whereby crashes were excluded if the pedestrian was hit while crossing a driveway, driveway density was removed from the model while marked crosswalk density remained. Figure 8. Relationship Between Driveway and Crosswalk Density Greater crosswalk density along the segment was also associated with increased crash frequency. Segments with crosswalk densities of 13 crosswalks per mile or greater were found to 42 predict significantly more crashes than those with fewer than 13 crosswalks per mile. This is consistent with prior research indicating that marked crosswalks are associated with higher crash rates than unmarked crosswalks [16] [46] due to the generally greater midblock pedestrian crossing activity along the segment. The tendency for pedestrians to select marked crosswalks over unmarked was reflected in pedestrian volumes: among the 40 midblock sites evaluated in this study, unmarked crosswalks had an average crossing volume of 14 pedestrians per hour, while marked crosswalks (with and without enhancement devices) had average crossing volumes of 102 pedestrians per hour. The channelizing effect which marked crosswalks have, in addition to the increased crash rates associated with marked crosswalks indicate that engineers should be conservative in their placement of midblock crosswalks, choosing locations with narrow crossing widths and low motorist speeds. Crosswalk marking pattern also had an effect on crashes, with segments utilizing continental crosswalks showing fewer pedestrian-vehicle crashes along the segment compared to those segments with standard crosswalks. This indicates that choosing a more visible crosswalk marking strategy may mitigate the increase in crashes associated with marked crosswalks. Special treatments like the R1-6, RRFB, and PHB were not specifically analyzed due to the treatment not being in effect for the entire 10 year study period, as well as crashes being analyzed at a segment rather than node-level. Furthermore, crosswalk length was positively correlated with pedestrian crashes. This is supported by previous research which found fewer pedestrian-involved crashes on narrower roads [47]. It should be noted that on median divided segments crosswalk length was the crossing distance for one direction of traffic. While the presence of an auxiliary lane (i.e. parking lane, bike lane, or shoulder) adds to crosswalk length, these lanes were found to reduce pedestrian crash occurrence. While the factor was not significant (p-value=0.115), the reduction in crashes could 43 be due to the traffic calming effects and subsequent lower speeds associated with on-street parking [48]. The crash analysis performed had several limitations. Enhancement devices (i.e., R1-6, RRFB, and PHB) were not analyzed due crashes being analyzed at a segment rather than nodelevel, as enhancement devices treat a particular crosswalk rather than an entire segment. Likewise, segment analysis of crashes does not allow for analysis of treatments present at signalized intersections. Most importantly, it must be noted that the small total 10-year sample size of 30 pedestrian crashes across the 25 segments is relatively small and clearly a limitation of this study, and therefore caution should be taken with interpretation of crash analysis results. Furthermore, no association between driver yielding compliance and pedestrian crash occurrence was found. 44 CHAPTER 5: DRIVER YIELDING BEHAVIOR AT SIGNALIZED INTERSECTIONS Pedestrian crossings at signalized intersections are an important safety consideration for roadway agencies, and such crossings will continue to become more important as non-motorized safety programs further encourage travel via walking in the future. Twenty-six signalized intersections were identified across the three Michigan cities in order to further evaluate pedestrian crossing safety. Field observational data for driver yielding compliance as well as historical traffic crash data were collected and analyzed along with historical traffic crash data at each location in order to assess the selected safety performance measures. Due to the small sample size of pedestrian crashes at the intersections studied, only naturalistic driver yielding compliance is included in the final analysis. 5.1 Data Summary Vehicle-pedestrian naturalistic yielding compliance data were collected at each of the 26 signalized intersections considered as a part of this study. Yielding in the context of this study was only assessed for cases where turning vehicles (right and left) encountered one or more pedestrians in the crosswalk. According to state law, during a permissive signal indication, the driver would must yield to pedestrians within the crosswalk in this scenario [25]. Figure 9 shows an example of a right-turning vehicle not yielding to pedestrians within the crosswalk who have the right-ofway. Thus, driver yielding compliance was scored accordingly for each crossing pedestrian’s encounter with a turning vehicle. Videos at signalized intersections were positioned such that signal indication was visible, which allowed for consistent and accurate naturalistic observations of driver and pedestrian yielding behavior at these intersections. As vehicles must slow down to turn, dilemma zone was not a concern. These data were aggregated into 15-minute intervals for 45 subsequent analysis to simplify the data collection process. Data from 104 unique 15-minute intervals were collected. However, only 84 unique 15-minute intervals had any turning vehiclepedestrian interactions, and therefore 84 intervals were included in the final analysis. A summary of the naturalistic yielding compliance behavior collected at the 26 signalized intersections is presented in Table 9. Figure 9. Screenshot of Noncompliant Turning Vehicle at Signalized Intersection 46 TABLE 9. Summary of Naturalistic Driver Yielding Behavior Data at Signalized Intersections Continuous Factors Factor Driver yielding Pedestrian-turning vehicle interactions Vehicle volume Bicycle volume Pedestrian volume Right-turn Left-turn Categorical Facotrs Factor Geometry Laneage Crosswalk Treatment Directionality Pedestrian signal Right-turn-on-red Median Level or Unit number of events in a 15min period number of events in a 15min period veh/15-min interval bicycles/15-min interval peds/15-min interval percent of total vehicles percent of total vehicles Level or Unit signalized crosswalk 4-leg intersection 3-leg intersection bike lanes present parking lanes present no additional lanes standard crosswalk continental crosswalk brick paver one-way two-way no countdown timer countdown timer Permitted Prohibited Not present Present 47 Mean SD Min Max 5.23 8.18 0 70 5.93 8.64 0 73 259.58 1.48 58.2 0.17 0.14 144.3 1.99 66.29 0.1 0.09 56 0 2 0 0 679 12 415 0.46 0.45 Proportion of Periods 0.08 0.73 0.19 0.31 0.77 0.08 0.25 0.72 0.04 0.44 0.56 0.24 0.76 0.72 0.28 0.72 0.28 Number of Sites 2 19 5 7 20 3 6 19 1 11 15 5 21 20 6 19 7 5.2 Results and Discussion The yielding compliance rates were disaggregated by intersection characteristics of interest and are presented in Table 10. Additionally, a statistical model was estimated based upon the negative binomial regression techniques outlined in Chapter 4. The final model results are presented in Table 11, which estimates driver yielding compliance at signalized intersections based upon several explanatory variables. It should be noted that Table 11 includes the coefficient estimate, standard error, odds ratio (for cases where binary indicator variables were utilized), and p-value for each variable. TABLE 10. Naturalistic Driver Yielding Compliance Rates by Site Characteristics Category Parameter Intersection geometry Three-leg Four-leg Signalized crosswalk One-way Two-way Standard Continental Brick paver No countdown timer Countdown timer Permitted Prohibited Not present Present Directionality Crosswalk treatment Pedestrian signal Right-turn-onred Median Percent Number of Number of Interactions of Turning Interactions per location Locations Vehicles Yielding 4 178 44.5 93.26% 20 418 20.9 86.36% 1 21 21.0 80.95% 11 253 23.0 90.91% 14 364 26.0 86.26% 6 101 16.8 84.16% 18 475 26.4 89.05% 1 41 41.0 87.80% 6 110 18.3 88.18% 19 507 26.7 88.17% 18 370 20.6 85.14% 7 247 35.3 92.71% 18 428 23.8 88.08% 7 189 27.0 88.36% 48 TABLE 11. Negative Binomial Results for Naturalistic Driver Yielding Compliance at Signalized Intersections Category Parameter Intercept Volume 15-min pedestrianturning vehicle interactions 15-min vehicle volume (ln) 15-min pedestrian volume (ln) Approach Signalized crosswalk configuration Three-leg Four-leg Crosswalk Brick paver type Standard Continental Estimate Std. Error z value Pr(>|z|) -3.818 0.029 0.859 0.003 -4.443 8.860E-06 8.716 < 2e-16 0.568 0.115 4.950 7.420E-07 0.458 0.092 4.972 6.610E-07 0.290 0.303 4.064 4.820E-05 3.030 0.002 baseline 1.179 0.919 baseline -0.586 -0.636 0.270 0.225 -2.173 -2.821 0.030 0.005 Odds Ratio 3.25 2.51 0.56 0.53 A four-leg intersection is shown in the negative binomial model to result in fewer yielding events compared to a signalized pedestrian crosswalk with an adjacent driveway, while a three-leg intersection is more likely to result in yielding behavior. The relationship of yielding behavior between three- and four-leg intersections is also shown in raw yielding rates, for which a three leg intersection has a yielding rate almost 7 percentage points higher than a four-leg intersection. Previous research has shown three-leg intersections to be associated with reduced numbers of pedestrian crashes [47]. It can also be seen in Table 10 that the three-leg intersection has more than double the observed number of pedestrian-turning vehicle interactions per location compared with four-leg intersections due to the necessity of vehicles turning at the dead-end leg. The regression modeling shows that increasing volumes of pedestrians and vehicles was associated with increased yielding compliance for turning vehicles. More importantly, an increasing number of pedestrian-vehicle interactions (i.e., yielding opportunities), was also associated with improved 49 driver yielding compliance, which is shown in Figure 10. The improved yielding performance associated with increasing numbers of pedestrian-turning vehicle interactions could be due to driver familiarity. On intersections with high pedestrian and turning vehicle volume, the driver is more likely to expect pedestrians, and consequently yield to them. Figure 10. Yielding Rates vs. Pedestrian-Turning Vehicle Interactions per 15 Minute Interval Looking at crosswalk type, the decorative brick paver crosswalk performed better than the more conventional standard and continental crosswalks. Caution should be taken in interpreting this result due to the small sample size (one site with 41 interactions). In spite of strong demand by local communities for these types of crosswalks due to aesthetics, research is currently limited in evaluating the safety impact of this crosswalk treatment. Guidance from FHWA indicates that 50 textured crossings, such as non-slip brick pavers, can increase driver attention by means of noise, vibration, and contrasting colors [50]. While brick pavers performed the best out of the three crosswalk treatments evaluated, at first glance, the standard crosswalk markings performed better than the continental pattern. However, when looking at the standard error for these two crosswalk marking patterns, there is no significant difference between them in turning vehicle yielding compliance. This is a surprising result, as the continental crosswalk is more visually conspicuous and performed better than the standard crosswalk at mid-block intersections. This could be due to the nature of turning vehicles. Turning is a complex task, particularly turning right on red. As most pavement markings are white, the crosswalk patterns could be in the driver’s periphery in contrast with the brick paver crosswalks which have both a contrasting color and texture, and therefore warn the driver in multiple ways. In addition, turning vehicles must slow down, and at lower speeds, they may be more aware of the pedestrian attempting to cross than the pavement markings themselves. 51 CHAPTER 6: DRIVER YIELDING COMPLIANCE AT MIDBLOCK CROSSINGS In selecting an alternative measure of effectiveness for crashes at midblock crossings, two options were considered: vehicle-pedestrian near-crashes (conflicts) and driver yielding compliance to staged pedestrians. As previously described in Chapter 3, naturalistic driver and pedestrian behavior was evaluated, including vehicle-pedestrian conflicts, which were defined as evasive maneuvers by drivers (i.e., swerving or extreme braking to avoid collision) or pedestrians (i.e., hurried walking or stepping back to the curb to avoid collision). However, this method of evaluating safety had several challenges. Primarily, determining driver and pedestrian intent was difficult or impossible to determine from video review, particularly related to the pedestrian evasive actions, which made up a majority of the conflict data sample. Simply put, it was often impossible to discern whether the pedestrian was forced to make an evasive action, such as walking faster, running, or making a path change, or did so voluntarily. As driver and pedestrian intent could not be ascertained from video review, all potential evasive actions by pedestrians or drivers were scored as conflicts. Ultimately, this resulted in 151 conflict events over a time period of 99.25 h, or more than 1.5 conflicts/h. Previous research correlating conflicts with motor vehicle crashes using specially equipped vehicles indicate that true conflicts, similar to crashes, are rare and random events [20]. Given the rarity of true conflicts, this large number of conflicts raised concerns, which were confirmed by the unusual results found in the preliminary negative binomial regression analysis. Ultimately, after further investigation of the data collection methods, evasive action event scoring, and modeling results, the vehiclepedestrian conflict data collected as a part of this study were deemed invalid for further evaluation. Driver yielding compliance to pedestrians, on the other hand, allowed for a consistent 52 methodology from observation-to-observation, as all pedestrian crossing attempts which were evaluated were attempted by trained researchers, providing for consistency. Furthermore, the staged crossing procedure allowed for sample sizes sufficient for meaningful analysis. Lastly, the methodology reflects a performance measure which is meaningful to pedestrians and engineers alike (driver yielding compliance to pedestrians) as a safety as well as an operational metric. 6.1 Data Summary Driver yielding compliance data were extracted from the 31 sites where staged pedestrians were utilized, resulting in a total of 1,281 observations, which were either scored as “yielded” or “did not yield.” These data are summarized in Table 12. However, although 1,281 data points were extracted for this study, data for the site with the RRFB could not be included in the model, as that site showed a 100 percent yielding compliance rate, which is a result incompatible with the statistical method utilized here. Thus, only 1,245 yielding compliance observations were included in the final analysis, although the RRFB compliance rate was included in subsequent discussions. Note that the summary statistics in Table 12 exclude the RRFB site, unless noted otherwise. Sites with in-street R1-6 signs, PHB, and RRFB treatments all utilized the continental style (i.e., markings parallel to the traffic direction) crosswalk. Utilization of R1-6 signs were limited to a single sign placed on the centerline within the crosswalk, and the three sign “gateway” application of this sign was not used in this study. 53 TABLE 12. Summary of Site Characteristics for Midblock Yielding Compliance Assessment Categorical Factors Factor Level or Unit Proportion of Observations Number of Sites Driver Actiona Yield Did not yield Near (curb) lane Center or far lanes Unqueued vehicle Queue leader Queue follower Unmarked Standard only Continental only In-street R1-6 sign PHB RRFB (excl. from model) <30 ft 31-40 ft 41-50 ft >50 ft One-Way Two-Way 2 lanes 3 lanes 4 lanes Two-lane Undivided multilane Divided multilane None Bike, parking or shoulder <50 pedestrians/h >50 pedestrians/h 0.61 0.39 0.70 0.30 0.66 0.21 0.13 0.20 0.07 0.58 0.08 0.04 0.03 0.54 0.11 0.31 0.04 0.55 0.45 0.85 0.10 0.04 0.45 0.05 0.50 0.37 0.63 0.54 0.46 5 3 17 3 2 1 15 4 9 2 15 15 24 4 2 14 3 13 12 18 15 15 Level or Unit Mean SD Min Max 11.13 0.49 200.20 101.36 7.93 22 2 218 5 0 54 4 1,204 662 31 Vehicle Lane Position Position of Vehicle in Queue Crosswalk Treatment Crossing Width (excludes median) Traffic Direction at Crosswalk Through Lanes at Crosswalk Roadway Cross-Section Auxiliary Lane Pedestrian Crossing Volume Continuous Factors Factor Crossing Width (excludes median) ft 34.91 Through Lanes at Crosswalk count 2.19 Vehicle Volume at Crosswalk vehicles/h 439.30 Pedestrian Crossing Volume pedestrians/h 85.95 Bicycle Volume bicycles/h 9.16 Note: The RRFB site was excluded from the summary statistics, except where noted a Dependent variable 54 6.2 Data Analysis As driver yielding compliance is a binary (yes/no) outcome, logistic regression provides an appropriate framework for determining those vehicle, pedestrian, and roadway factors associated with driver yielding behavior. Within the context of this study, the logistic regression model takes the general form: 𝑝 ln [1−𝑝𝑖 ] = 𝛼 + 𝛽′𝑋𝑖 , 𝑖 (7) where pi is the response probability of driver i yielding to a pedestrian, α is an intercept term, β' is a vector of estimable parameters, and Xi is a vector of predictor variables (e.g., crosswalk treatment, pedestrian/vehicular volumes). One concern that arises within the context of this study is the potential correlation in compliance rates within individual locations due to common, unobserved factors (i.e., unobserved heterogeneity). Failure to account for such correlation may lead to biased or inefficient parameter estimates. To account for this concern, a site-specific random effect is added for each location j, resulting in: 𝑝 ln [1−𝑝𝑖 ] = 𝛼𝑗 + 𝛽𝑋𝑖 , 𝑖 (8) This approach allows for the constant term to vary across locations, but maintain the same value for all crossing events observed at an individual location. In addition to impacting the constant term, unobserved heterogeneity can also lead to explanatory parameters varying across locations. For example, various site characteristics may occur in combination with other factors (e.g., land use, local design practices) that are not directly accounted for as a part of the analysis. To address this issue, a series of site-specific random parameters can be similarly introduced as follows: 55 𝛽𝑗 = 𝛽 + 𝑢𝑗 , (9) where 𝛽 is the vector of estimable parameters and 𝑢𝑗 is a randomly distributed term for each location j with mean zero and variance 𝜎 2 . Parameters that are found to vary across study locations take this random parameter form while those parameters that are shown to have homogeneous impacts across locations are treated as traditional fixed parameters (i.e., 𝑢𝑗 is equal to zero). Model estimation was done through simulated maximum likelihood using 10,000 Halton draws. The variables from Table 12 were considered as potential predictors when estimating this mixed effects logistic regression model. Several preliminary versions of the models were estimated, and in many cases, categorical factors were utilized over the continuous analogs in order to improve model fit. The variables found to be statistically significant in the preliminary model were then each considered as normally distributed random parameters. Those parameters shown to vary across locations were retained as random parameters, with the remaining variables included as fixed parameters. 6.3 Results and Discussion The final model results for driver yielding compliance are displayed in Table 13, which includes the coefficient estimate, standard error, t-statistic, and odds ratio for each variable included in the mixed effects logistic regression model. The base conditions for the model were included as follows: unmarked crosswalk, undivided roadway cross-section, subject vehicle in the lane nearest to the curb, and subject vehicle not queued. 56 TABLE 13. Logistic Regression Results for Driver Yielding Compliance Variable Fixed Parameters Constant Crosswalk Treatment Crossing Width Pedestrian Volume Vehicle Lane Position Vehicle Position in Queue Random Parameters Crosswalk Treatment Cross-Section Coefficient Standard Estimate Error Level or Unit Unmarked In-Street R1-6 Sign PHB ln ft ln ped/hr Near (curb) lane Other lane Unqueued vehicle Queue leader Queue follower -3.416 baseline 2.83458 2.93714 0.54656 0.18541 baseline 0.83107 baseline 0.45534 -0.35329 Unmarked Standard only mean Standard only st. dev. Continental only mean Continental only st. dev. Undivided Divided mean Divided st. dev. baseline 1.01445 2.01124 1.24722 0.30001 baseline -0.34902 0.38509 t-stat Odds Ratio 1.217 0.005 N/A 0.64039 0.66567 0.31341 0.07108 <0.0001 <0.0001 0.0822 0.0099 17 18.9 1.7 1.2 0.12889 <0.0001 2.3 0.13549 0.17239 0.0011 0.0411 1.6 0.7 0.3484 0.5202 0.2022 0.0873 0.0044 <0.0001 <0.0001 0.0011 2.8 N/A 3.5 N/A 0.1184 0.10342 0.0032 0.001 0.7 1.5 N=1,245 Initial log-likelihood (constant only) = -833.24 Log-likelihood at convergence = -629.73 McFadden Pseudo R2=0.244 The results of the mixed effects logistic regression model revealed several interesting findings. The type of crosswalk treatment had the strongest association with driver yielding compliance of any variables included in the model. Compared to unmarked crossing areas, each of the crosswalk treatments provided significant improvements in driver yielding compliance during the staged pedestrian crossing attempts. Both the standard and continental crosswalks were 57 shown to increase compliance over unmarked crosswalks. On average, compliance rates were 2.8 times higher for standard crosswalks and 3.5 times higher for continental crosswalks. These effects were shown to vary across sites and this variability was particularly pronounced for the standard crosswalks, which may be reflective of the settings under which either type of crosswalk was installed. The inclusion of an R1-6 in-street sign, PHB, or RRFB provided substantial improvements in yielding compliance over the standard and continental crosswalks. To further enhance discussion of the crosswalk treatment results, the raw yielding compliance summary statistics are displayed for each treatment type in Table 14. TABLE 14. Driver Yielding Compliance by Crosswalk Treatment Crosswalk Treatment Number of Locations Number of Observations Percent of Drivers Yielding Unmarked Standard only Continental only In-Street Sign (R1-6) PHB RRFB ALL 5 3 17 3 2 1 31 261 88 744 101 51 36 1,281 28.70% 50.00% 66.30% 95.00% 98.00% 100.00% 62.00% The raw yielding compliance rates for each type of treatment revealed several interesting findings that generally followed the results of the mixed effect model. First, the PHB yielding compliance rate of 98 percent was in general agreement with PHB yielding compliance (85 to 97 percent) observed in other states [51]. The single RRFB location showed 100 percent yielding compliance, which was substantially higher than the 22 to 94 percent rates observed in other states [26] [41] [51] [52]. The PHB and RRFB locations also displayed higher yielding rates compared to rates observed at several Michigan PHB and RRFB locations in 2012 (77 percent, on average, for both devices) [16]. Considering that the current PHB and RRFB study sites were also included 58 in the 2012 study suggests that yielding compliance may improve over time as drivers become more familiar with these devices. However, although prior studies have also shown improvements in driver compliance rates over time [13] [41], these results should be viewed with caution due to the small number of PHB and RRFB locations observed in the current study. The sites with an R1-6 sign positioned within the crosswalk showed a yielding compliance rate of 95 percent, which was similar to rates observed at the PHB and RRFB locations and substantially higher than crosswalks with no additional treatment. Although crosswalks with R1-6 signs have shown compliance rates of up to 87 percent in prior studies [51], such a high level of compliance was a surprising result given the substantially lower cost of the R1-6 sign compared to RRFBs and PHBs. Turning to other variables of interest, there was significant variability in compliance based upon the lane where the subject vehicle encountered the pedestrian. Drivers traveling in the near (curb) lane were 2.3 times less likely to yield for a pedestrian compared to drivers traveling in any other lane. This effect may be reflective of differences in driver expectancy based upon pedestrian location and behavior. When crossing attempts were initiated at the near (curb) lane, approaching drivers may have either not been observed by the approaching driver or the driver may not have realized their intention to cross. In contrast, the pedestrians’ intensions were likely clearer while attempting to cross the other lanes where the individual was completely within the roadway as the driver approached. The pedestrians were also likely more conspicuous to approaching drivers overall. Regarding the roadway cross-section variables, drivers’ likelihood to yield increased as the crossing distance increased. Interpretation of the parameter estimate suggests a crossing width of 48 ft (i.e., a four-lane street) would result in a 46 percent greater likelihood of driver yielding compared to a width of 24 ft (i.e., a two-lane street). This again may be due to the increased 59 conspicuity of the pedestrian to approaching drivers. In contrast, drivers were, on average, 30 percent less likely to yield on divided roadways compared to undivided roadways. This effect was shown to vary across locations due to unobserved heterogeneity between sites, which suggests the presence of additional factors affecting yielding compliance. Further investigation of the interaction effects of lane position and roadway cross-section on yielding compliance was performed, with the raw yielding compliance rates displayed in Table 15. Near-lane yielding compliance was lower across all roadway cross-section types. Near-lane compliance rates were substantially lower for multilane divided roadways, suggesting potential issues with visual occlusion of the pedestrian in the median. Similarly, compliance in lanes other than the near lane was considerably higher on multilane undivided roadways than for two-lane or divided roadways, further confirming that drivers were more aware of crossing pedestrians as the exposure time was increased. TABLE 15. Interaction of Lane Position with Roadway Cross-Section and Crosswalk Treatment Variable 2-Lane Multilane - Undivided Multilane - Divided Unmarked Standard only Continental only In-Street Sign (R1-6) PHB RRFB TOTAL Number of Observations Near Lane Other Lane 390 170 36 23 464 198 166 95 66 22 575 169 40 61 25 26 18 18 890 391 Yielding Compliance Near Lane Other Lane 55.60% 74.70% 80.60% 91.30% 51.90% 79.80% 19.90% 44.20% 34.80% 95.50% 61.40% 82.80% 92.50% 96.70% 96.00% 100.00% 100.00% 100.00% 54.80% 78.20% Turning to the interaction between lane position and crosswalk treatment, the results for which are also displayed in Table 15, yielding compliance was again lower in the near lane across 60 all crosswalk treatments. Near-lane yielding compliance was especially poor for unmarked crosswalks (19.9 percent), improving to 34.8 percent and 61.4 percent where standard crosswalks and continental crosswalks were used, respectively. Yielding compliance at standard crosswalks was particularly sensitive to lane position, increasing from 34.8 percent for drivers in the near lane to 95.5 percent for drivers in any other lane. Yielding compliance was far less sensitive to driver lane position at locations where additional treatments (i.e., R1-6 sign, PHB, RRFB) were utilized, further emphasizing the effectiveness of these treatments (Figure 11). Far lane yielding compliance was higher than near lane yielding compliance for sites without enhancement devices, Percentage of Drivers Yielding likely due to the increased conspicuity of the pedestrian approaching lanes other than the curb lane. 100% 75% 50% 25% 0% Unmarked Standard Continental In-Street only only Sign (R1-6) Treatment Type Near Lane PHB RRFB Far Lane Figure 11. Yielding Compliance by Lane Position and Treatment The vehicle’s position within the queue also affected the likelihood of driver yielding. The logistic regression results displayed in Table 13 suggest that queue leaders were 1.6 times more likely to yield compared to unqueued drivers and were 2.3 times more likely to yield compared to queued drivers that were not in the lead position. These results are not surprising, as queued drivers in many cases are simply following the leading vehicle, who obviously also did not yield for the 61 pedestrian. Past research on the PHB has shown that queued drivers will tend to follow the queue leader without first checking for pedestrians attempting to cross [42] [45]. Finally, greater pedestrian volumes were associated with an increase in yielding compliance. This was not a surprising result, as greater pedestrian activity would serve to raise driver awareness at the particular crosswalk. However, although preliminary analyses showed yielding rates to decrease with hourly traffic volume, this effect was not statistically significant in the final analysis when considering other pertinent factors. 6.4 Conclusions and Recommendations The results of this study provide several important insights to inform subsequent decisions by road agencies as to the installation of pedestrian crosswalk treatments. A mixed effects logistic regression model was estimated to account for intra-site correlation in yielding rates, as well as for the effects of unobserved heterogeneity across study locations. The results demonstrate the importance of applying robust analytical methods to examine driver-pedestrian interactions. Ultimately, the findings provide a clear indication that the type of crosswalk treatment has a strong influence over driver yielding compliance. While yielding compliance improves substantially when crosswalk markings are utilized, much greater compliance is obtained when additional enhancement devices, such as RRFBs, PHBs, or in-street R1-6 signs, are also provided. Yielding compliance rates for the various crosswalk treatments were shown to be in agreement with previous research performed outside of Michigan, and also showed improvements across all treatment types compared to prior studies performed within Michigan. This is an important finding, which suggests that compliance improves as drivers become more familiar with a particular treatment. 62 It was also found that yielding compliance is highly sensitive to both the roadway crosssection and lane position of the vehicle relative to the location of the crossing pedestrian. Drivers were much less likely to yield when the driver encountered the staged pedestrian at the nearside curb lane compared to any other lane. This is not a surprising result, as the pedestrian is in a less conspicuous and less vulnerable position when waiting near the curb, compared to encounters that occurred while the pedestrian was approaching any other lane. While this result is reflective of the interaction between motorists and pedestrians attempting to cross, it does indicate the necessity for yielding compliance studies to control for the driver lane position. And while low curb-lane compliance persisted across each of the observed types of roadway cross sections (two-lane, multilane undivided, and multilane divided), it was particularly low on median divided roadways. This may be indicative of potential obstructions within the median that reduce the visibility of pedestrians waiting to cross. Perhaps most importantly, however, yielding compliance showed little sensitivity to the particular travel lane of the subject vehicle at locations where additional treatments (i.e., in-street R1-6 sign, PHB, RRFB) were utilized, further validating the effectiveness of these devices. Road agencies are advised to place crosswalks in otherwise unmarked locations where pedestrians frequently cross and, when necessary, install additional treatment. Providing marked crosswalks in locations with light to moderate vehicle volumes will result in higher yielding compliance and will typically not require additional treatment unless special circumstances (i.e., school, hospital, etc.) exist. For midblock crosswalks in locations with high vehicle and/or high pedestrian volumes, particularly at multilane locations, additional low-cost treatments such as instreet pedestrian crossing signs may further increase compliance and provide subsequent safety benefits. Due to high costs, RRFBs and especially PHBs, should only be installed at select 63 locations displaying high pedestrian and vehicular volumes, particularly where other treatments have proven to be ineffective. While the results of this study provide important insights to guide subsequent investment strategies for mid-block crossings, there are some important limitations that must be stated. First, the results are limited to low-speed locations only. Yielding compliance is likely different on higher speed roadways, where pedestrian activity is typically less frequent. Furthermore, all sites selected in this study were on or near public universities in the Midwest during the early fall when school was in session. Therefore, the samples of pedestrians and drivers included in this study are a non-random sample and it is unclear how these trends would extrapolate to a broader population. 64 CHAPTER 7: CONCLUSIONS Ultimately, this thesis found driver yielding compliance to pedestrians to be an ideal surrogate for crashes in analyzing pedestrian safety. At midblock crossing areas, statistical analysis of yielding compliance found that enhancement devices were associated with increased propensity to yield on the part of drivers. Using yielding compliance as a measure of effectiveness in and of itself is not novel. Rather, the primary contribution this thesis makes is the analysis technique. Binary logistic regression with mixed effects was used to determine the probability of driver yielding based on not only crosswalk treatment, but also on site and cross-sectional characteristics. A cross-sectional study was ideal for these purposes, as driver familiarity with new devices was not a concern. Ultimately, the study design combined with the analysis technique found yielding performance for the PHB and in-street R1-6 sign to be similar to each other on lowspeed roads based on odds ratio, which is an important finding considering the high cost of installing the PHB. This was an unexpected result, and direct comparison of these treatments was only possible because of the analysis method. 7.1 Driver Behavior During Pedestrian Crossing Attempts The driver yielding compliance results at midblock crosswalks indicated that the type of crosswalk treatment has a strong influence over driver behavior when encountering a pedestrian in the crosswalk. While yielding compliance improves substantially when crosswalk markings are utilized, much greater compliance is obtained when additional enhancement devices, such as RRFBs, PHBs, or in-street R1-6 signs, are also provided. Yielding compliance rates for the various crosswalk treatments were shown to be in agreement with previous research performed outside of Michigan, and also showed improvements across all treatment types compared to prior studies 65 performed within Michigan. This is an important finding, which suggests that compliance may improve as drivers become more familiar with a particular treatment. Driver yielding compliance at midblock crosswalks was shown to increase as the pedestrian crossing volumes increased, but decrease as the vehicular volume increased. It was also found that yielding compliance is highly sensitive to both the roadway cross-section and lane position of the vehicle relative to the location of the crossing pedestrian. Drivers were much less likely to yield when the driver encountered the staged pedestrian at the nearside curb lane compared to any other lane. This is not a surprising result, as the pedestrian is in a less conspicuous and less vulnerable position when waiting near the curb, compared to encounters that occurred while the pedestrian was approaching a driver in any other lane. While this result is reflective of the interaction between motorists and pedestrians attempting to cross, it does indicate the necessity for yielding compliance studies to control for the driver lane position. And while low curb-lane compliance persisted across each of the observed types of roadway cross sections (two-lane, multilane undivided, and multilane divided), it was particularly low on median divided roadways. This may be indicative of potential obstructions within the median that reduce the visibility of pedestrians waiting to cross. Interestingly, vehicle-pedestrian conflicts were found to be lower at midblock crosswalks on divided roadways compared to undivided roadways. Perhaps most importantly, however, yielding compliance showed little sensitivity to driver lane position at locations where additional treatments (i.e., in-street sign, PHB, RRFB) were utilized, providing further evidence of the effectiveness of these treatments. Considering signalized intersections, yielding compliance was greater at 3-leg intersections compared to 4-leg intersections. Additionally, yielding compliance for turning vehicles at signalized intersections actually improved as the turning vehicle and pedestrian crossing volumes 66 increased (and subsequent number of pedestrian-vehicle interactions increased). This effect was particularly strong when considering only right-turning vehicles. Readers should also be aware of the limitations of the field study. First, the results are limited to low speed locations only. Driver and pedestrian behavior is likely different on higher speed roadways and pedestrian activity is typically less frequent. Furthermore, all sites selected in this study were on or near public universities in the Midwest during the early fall when school was in session. Therefore, both the pedestrians and drivers on which this model is based on may be more likely to fit a younger demographic than the pedestrian population at large. Finally, and most importantly, although the investigation of pedestrian crashes at the study sites provided some indication of relationships between the various site, traffic, and behavioral factors, the small sample size of crashes across the study sites did not provide definitive results nor did it allow for formal SPF development. 7.2 Recommendations In evaluating the safety of existing pedestrian crossing sites, road agencies are advised to use yielding compliance as their performance measure. Crash analysis is typically infeasible due to the low number of pedestrian crashes at crossing locations, in addition to the lack of meaningful pedestrian exposure data. Meaningful pedestrian-vehicle conflicts (i.e. near crashes) are also extremely rare, and therefore the labor costs to collect enough data would also be infeasible. While there is no objective measure for what constitutes satisfactory yielding compliance at any given location, engineering judgement and public feedback can be useful in determining yielding compliance targets in locations with high pedestrian activity. When additional treatments are 67 installed to improve pedestrian safety, agencies should conduct a yielding compliance study before and after the treatment is installed to determine its effectiveness. Road agencies are advised to place crosswalks in otherwise unmarked locations where pedestrians frequently cross and, when necessary, install additional treatment. Providing marked crosswalks at midblock locations on low speed roadways with light to moderate vehicle volumes will result in higher yielding compliance and will typically not require additional treatment unless special circumstances (i.e., school, hospital, etc.) exist. For midblock crosswalks on low speed roadways with high vehicle and/or high pedestrian volumes, particularly at multilane locations, additional low-cost treatments such as in-street pedestrian crossing signs (R1-6) may further increase compliance and provide subsequent safety benefits, whether used in a single installation on the centerline (studied here) or in a gateway configuration on both the centerline and at the edges of the roadway. Due to high costs, RRFBs and especially PHBs, should only be installed at select locations displaying high pedestrian and vehicular volumes, particularly where other treatments have proven to be ineffective. 68 APPENDIX 69 APPENDIX This appendix contains aerial imagery for study sites (Table 16). Refer to Tables 4 and 5 for site descriptions. TABLE 16. Aerial Imagery for Study Sites Site 1 Site 2 Image Source: Google Earth Image Source: Google Earth Site 3 Site 4 Image Source: Google Earth Image Source: Google Earth 70 TABLE 16. (cont’d) Site 5 Site 6 Image Source: Google Earth Site 7 Image Source: Google Earth Site 8 Image Source: Google Earth Image Source: Google Earth 71 TABLE 16. (cont’d) Site 9 Site 10 Image Source: Google Earth Image Source: Google Earth Site 11 Site 12 Image Source: Google Earth Image Source: Google Earth 72 TABLE 16. (cont’d) Site 13 Site 14 Image Source: Google Earth Site 15 Image Source: Google Earth Site 16 Image Source: Google Earth Image Source: Google Earth 73 TABLE 16. (cont’d) Site 17 Site 18 Image Source: Google Earth Site 19 Image Source: Google Earth Site 20 Image Source: Google Earth Image Source: Google Earth 74 TABLE 16. (cont’d) Site 21 Site 22 Image Source: Google Earth Image Source: Google Earth Site 23 Site 24 Image Source: Google Earth Image Source: Google Earth 75 TABLE 16. (cont’d) Site 25 Site 26 Image Source: Google Earth Site 27 Image Source: Google Earth Site 28 Image Source: Google Earth Image Source: Google Earth 76 TABLE 16. (cont’d) Site 29 Site 30 Image Source: Google Earth Site 31 Image Source: Google Earth Site 32 Image Source: Google Earth Image Source: Google Earth 77 TABLE 16. (cont’d) Site 33 Site 34 Image Source: Google Earth Image Source: Google Earth Site 35 Site 36 Image Source: Google Earth Image Source: Google Earth 78 TABLE 16. (cont’d) Site 37 Site 38 Image Source: Google Earth Site 39 Image Source: Google Earth Site 40 Image Source: Google Earth Image Source: Google Earth Site 41 Site 42 Image Source: Google Earth Image Source: Google Earth 79 TABLE 16. (cont’d) Site 43 Site 44 Image Source: Google Earth Image Source: Google Earth Site 45 Image Source: Google Earth Site 46 Image Source: Google Earth 80 TABLE 16. (cont’d) Site 47 Site 48 Image Source: Google Earth Site 49 Image Source: Google Earth Site 50 Image Source: Google Earth Image Source: Google Earth 81 TABLE 16. (cont’d) Site 51 Site 52 Image Source: Google Earth Image Source: Google Earth Site 53 Site 54 Image Source: Google Earth Image Source: Google Earth 82 TABLE 16. (cont’d) Site 55 Site 56 Image Source: Google Earth Image Source: Google Earth Site 58 Site 57 Image Source: Google Earth Image Source: Google Earth 83 TABLE 16. (cont’d) Site 59 Site 60 Image Source: Google Earth Image Source: Google Earth Site 61 Site 62 Image Source: Google Earth Image Source: Google Earth Site 63 Image Source: Google Earth Site 64 Image Source: Google Earth 84 TABLE 16. (cont’d) Site 65 Site 66 Image Source: Google Earth Image Source: Google Earth 85 REFERENCES 86 REFERENCES [1] Federal Highway Administration, "Pedestrian and Bicycle Safety," [Online]. Available: http://safety.fhwa.dot.gov/ped_bike/. [Accessed September 2016]. [2] American Association of Highway and Transportation Officials, Highway Safety Manual, 2010. [3] Federal Highway Administration, "Manual on Uniform Traffic Control Devices," December 2009. [Online]. Available: https://mutcd.fhwa.dot.gov/pdfs/2009r1r2/mutcd2009r1r2edition.pdf. [Accessed 19 July 2017]. [4] J. Lovell and E. Hauer, "The Safety Effect of Conversion to All-Way Stop Control," Transportation Research Record, no. 1068, pp. 103-107, 1986. [5] B. Persaud, E. Hauer, R. Retting, R. Vallurupalli and K. Mucsi, "Crash Reductions Related to Traffic Signal Removal in Philadelphia," Accident Analysis & Prevention, vol. 29, no. 6, pp. 803-810, 1997. [6] R. Elvik and T. Vaa, Handbook of Road Safety Measures, Oxford, United Kingdom: Elsevier, 2004. [7] P. Alluri, A. Gan, K. Haleem, S. Miranda, E. Echezabal, A. Diaz and S. Ding, "Beforeand-After Safety Study of Roadways Where New Medians Have Been Added," Report BDK80 977-18, State of Florida Department of Transportation, 2012. [8] K. Shin, S. Washington and I. Van Schalkwyk, "Evaluation of the Scottsdale Loop 101 Automated Speed Enforcement Demonstration Program," Accident Analysis & Prevention, vol. 41, no. 3, pp. 393-403, 2009. [9] L. Chen, C. Chen, R. Srinivasan, C. McKnight, R. Ewing and M. Roe, "Evaluating the Safety Effects of Bicycle Lanes in New York City," American Journal of Public Health, vol. 102, pp. 1120-1127, 2012. [10] L. Chen, C. Chen and R. Ewing, "The Relative Effectiveness of Signal Related Pedestrian Countermeasures at Urban Intersections - Lessons from a New York City Case Study," Transport Policy, vol. 32, pp. 69-78, 2014. [11] S. Jensen, "Safety Effects of Intersection Signalization: A Before-After Study," in TRB 89th Annual Meeting Compendium of Papers, Washington, D.C., 2010. [12] P. Wanvik, "Effects of Road Lighting: An Analysis Based on Dutch Accident Statistics 1987-2006," Accident Analysis & Prevention, vol. 41, no. 1, pp. 123-128, 2009. 87 [13] R. Van Houten, J. LaPlante and T. Gustafson, "Evaluating Pedestrian Safety Improvements," Report RC-1585. Michigan Department of Transportation, 2012. [14] A. Fayish and F. Gross, "Safety Effectiveness of Leading Pedestrian Intervals Using the Empirical Bayes Method," in TRB 88th Annual Meeting Compendium of Papers, Washington, D.C., 2009. [15] K. Fitzpatrick and E. Park, "Safety Effectiveness of the HAWK Pedestrian Crossing Treatment," Report FHWA-HRT-10-042, Federal Highway Administration, U.S. Department of Transportation, 2010. [16] C. Zegeer, R. Stewart, H. Huang, P. Lagerwey, J. Feaganes and B. Campbell, "Safety Effects of Marked versus Unmarked Crosswalks at Uncontrolled Locations: Final Report and Recommended Guidelines," Report FHWA-HRT-04-100, Federal Highway Administration, U.S. Department of Transportation, 2005. [17] A. Tarko, G. Davis, N. Saunier and T. Sayed, "White Paper: Surrogate Measures of Safety," in TRB 88th Annual Meeting Compendium of Papers, Washington, D.C., 2009. [18] V. Shankar, P. Jovanis, J. Aguero-Valverde and F. Gross, "Analysis of Naturalistic Driving Data: Prospective View on Methodological Paradigms," Transportation Research Record, no. 2061, pp. 1-8, 2008. [19] A. Tarko, "Use of Crash Surrogates and Exceedance Statistics to Estimate Road Safety," Accident Analysis & Prevention, vol. 45, pp. 230-240, 2012. [20] T. Dingus, S. Kauer, V. Neale, A. Petersen, S. Lee, J. Sudweeks, M. Perez, J. Hankey, D. Ramsey, S. Gupta, C. Bucher, Z. Doerzaph, J. Jermeland and R. Knipling, "The 100-Car Naturalistic Driving Study, Phase II - Results of the 100-Car Field Experiment," Report DOT-HS-810-593, National Highway Traffic Safety Administration, U.S. Department of Transportation, 2006. [21] F. Guo, S. Klauer, J. Hankey and T. Dingus, "Near Crashes as Crash Surrogate for Naturalistic Driving Studies," Transportation Research Record, no. 2147, pp. 66-74, 2010. [22] S. Turner, K. Fitzpatrick, M. Brewer and E. Park, "Motorist Yielding to Pedestrians at Unsignalized Intersections: Findings from a National Study of Improving Pedestrian Safety," Transportation Research Record, no. 1982, pp. 1-12, 2006. [23] B. Wolshon and A. Pande, "Traffic Engineering Handbook, 7th Edition," John Wiley and Sons Inc., 2016. [24] R. Van Houten, L. Malenfant, B. Huitema and R. Blomberg, "Effects of High-Visibility Enforcement on Driver Compliance with Pedestian Yield Right-of-Way Laws," Transportation Research Record, no. 2393, pp. 41-49, 2013. 88 [25] Michigan State Police, Uniform Traffic Code for Cities, Townships, and Villages, 2003. [26] K. Fitzpatrick, M. Brewer and R. Avelar, "Driver Yielding at Traffic Control Signals, Pedestrian Hybrid Beacons, and Rectangular Rapid-Flashing Beacons in Texas," Transportation Research Record, no. 2463, pp. 46-54, 2014. [27] S. Islam, V. Serhiuenko, J. Ivan, N. Ravishanker and P. Garder, "Explaining Pedestrian Safety Experience at Urban and Suburban Crossings Considering Observed Conflicts and Pedestrian Counts," Journal of Transportation Safety & Security, vol. 6, no. 4, 2014. [28] B. Herms, "Pedestrian Crosswalk Study: Crashes in Painted and Unpainted Crosswalks," Transportation Research Record, no. 406, 1972. [29] L. Ekman, "On the Treatment of Flow in Traffic Safety Analysis," University of Lund, Lund, Sweden, 1996. [30] S. Yagar, "Safety Inpacts of Installing Pedestrian Crosswalks," in Proceedings of the Effectiveness of Highway Safety Improvements Conference,, New York, NY, 1985. [31] L. Ekman and C. Hyden, "Pedeestrian Safety in Sweden," Report FHWA-RD-99-091, Federal Highway Administration, Washington, D.C., 1999. [32] V. Gitelman, R. Carmel, F. Pesahov and S. Hakkert, "An examination of the influence of crosswalk marking removal on pedestrian safety as reflected in road user behaviours," Transportation Research Part F, no. 46, pp. 342-355, 2017. [33] I. Banerjee and D. Ragland, "Evaluation of Countermeasures: A Study on the Effect of Impactable Yield Signs Installed at Four Intersections in San Francisco," in TRB 86th Annual Meeting Compendium of Papers, Washington, D.C., 2007. [34] C. Strong and M. Kumar, "Safety Evaluation of Yield-to-Pedestrian Channelizing Devices: Final Report," Report FHWA-PA-2006-010-050114, Pennsylvania Department of Transportation, 2006. [35] K. Fitzpatrick, S. Turner, M. Brewer, P. Carlson, B. Ullman, N. Trout, E. Park, J. Whitacre, N. Lalani and D. Lord, "Improving Pedestrian Safety at Unsignalized Crossings," Report TCRP 112/NCHRP 562, Transportation Research Board of the National Academies, Washington, D.C., 2006. [36] Western Michigan University, T.Y. Lin International, "User Guide for R1-6 Gateway Treatment for Pedestrian Crossings," December 2016. [Online]. Available: http://mdotcf.state.mi.us/public/tands/Details_Web/mdot_user_guide_gateway_treatment. pdf. [Accessed 19 July 2017]. 89 [37] R. Benekohal, J. Medina and M. Wang, "Crosswalk Signing and Marking Effects on Conflicts and Pedestrian Safety in UIUC Campus," Report FHWA-IL/UI-TOL-22, Federal Highway Administration, U.S. Department of Transportation, 2007. [38] S. Pulugurtha, S. Nambisan, M. Dangeti and V. Vasudevan, "Evaluation of Effectiveness of Traffic Signs to Enhance Pedestrian Safety," in TRB 89th Annual Meeting Compendium of Papers, Wshington, D.C., 2010. [39] M. Brewer, K. Fitzpatrick and R. Avelar, "Rectangular Rapid Flashing Beacons and Pedestrian Hybrid Beacons: Pedestrian and Driver Behavior Before and After Installation," Transportation Research Record, no. 2519, pp. 1-9, 2016. [40] W. Hunter, S. Raghavan and C. Martell, "Evaluation of the Rectangular Rapid Flash Beacon at a Pinellas Trail Crossing in St. Petersburg, Florida," Report BA784, Florida Department of Transportation, 2009. [41] J. Shurbutt and R. Van Houten, "Effects of Yellow Rectangular Rapid-Flashing Beacons on Yielding at Multilane Uncontrolled Crosswalks," Report FHWA-HRT-10-043, Federal Highway Administration, U.S. Department of Transportation, 2010. [42] K. Hunter-Zaworski and J. Mueller, "Evaluation of Alternative Pedestrian Traffic Control Devices," Report FHWA-OR-RD-12-09, Federal Highway Adminsitration, U.S. Department of Transportation, 2012. [43] J. Ross, D. Serpico and R. Lewis, "Assessment of Driver Yielding Rates Pre- and PostRRFB Installation, Bend, Oregon," Report FHWA-OR-RD-12-05, Federal Highway Administration, U.S. Department of Transportation, 2011. [44] C. Zegeer, C. Lyon, R. Srinivasan, B. Persaud, B. Lan, S. Smith, D. Carter, N. J. Thirsk, J. Zegeer, R. Ferguson, R. Van Houten and C. Sundstron, "Development of Crash Modification Factors for Uncontrolled Pedestrian Crossing Treatments," Transportation Research Record: Journal of the Transportation Research Board, no. 2636, pp. 1-8, 2017. [45] K. Fitzpatrick and M. Pratt, "Road User Behaviors at Pedestrian Hybrid Beacons," Transportation Research Record, no. 2586, pp. 9-16, 2016. [46] Z. Sawalha and T. Sayed, "Evaluating Safety of Urban Arterial Roadways," Journal of Transportation Engineering, vol. 127, no. 2, pp. 151-158, 2001. [47] S. Ukkusuri, S. Hasan and H. M. A. Aziz, "Random Parameter Model Used to Explain Effects of Built-Environment Characteristics on Pedestrian Crash Frequency," Transportation Research Record, no. 2237, pp. 98-106, 2011. [48] W. E. Marshall, N. W. Garrick and G. Hansen, "Reassessing On-Street Parking," Transportation Research Record, no. 2046, pp. 45-52, 2008. 90 [49] C. Lee and M. Abdel-Aty, "Comprehensive Analysis of Vehicle-Pedestrian Crashes at Intersections in Florida," Accident Analysis & Prevention, no. 37, pp. 775-786, 2005. [50] Federal Highway Administration, "Pedestrian Accommodations at Intersections," [Online]. Available: https://safety.fhwa.dot.gov/ped_bike/univcourse/pdf/swless15.pdf. [Accessed 18 July 2017]. [51] K. Fitzpatrick, S. Chysler, R. Van Houten, W. Hunter and S. Turner, "Evaluation of Pedestrian and Bicycle Engineering Countermeasures: Recancular Rapid-Flashing Beacons, HAWKs, Sharrows, Crosswalk Markings, and the Development of an Evaluation Methods Report," Report FHWA-HRT-11-039. Federal Highway Administration, 2011. [52] K. Fitzpatrick, R. Avelar, I. Potts, M. Brewer, J. Robertson, C. Fees, J. Hutton, L. Lucas and K. Bauer, "Investigating Improvements to Pedestrian Crossings with an Emphasis on the Recangular Rapid-Flashing Beacon," Report FHWA-HRT-15-044. Federal Highway Administration, 2015. [53] L. Miranda-Moreno, J. Strauss and P. Morency, "Exposure Measures and Injury Frequency Models for Analysis of Cyclist Safety at Signalized Intersections," in TRB 90th Annual Meeting Compendium of Papers, Washington, D.C., 2011. [54] J. Oh, J. Jun, K. Eungcheol and K. Myungseob, "Assessing Critical Factors Associated with Bicycle Collisions at Urban Signalized Intersections," in TRB 87th Annual Meeting Compendium of Papers, Washington, D.C., 2008. [55] S. Turner, G. Wood, T. Hughes and R. Singh, "Safety Performance Functions for Bicycle Crashes in New Zealand and Australia," Transportation Research Record, no. 2236, pp. 66-73, 2011. [56] J. Schepers, P. Kroeze and J. Wust, "Road Factors and Bicycle-Motor Vehicle Crashes at Unsignalized Priority Intersections," Accident Analysis & Prevention, vol. 43, pp. 853861, 2011. [57] J. Shurbutt and R. Van Houten, "Effects of Yellow Rectangular Rapid-Flashing Beacons on Yielding at Multilane Uncontrolled Crosswalks," Report FHWA-HRT-10-043, Federal Highway Administration, U.S. Department of Transportation, 2010. [58] D.-G. Kim, S. Washington and J. Oh, "Modeling Crash Types: New Insights into the Effects of Covariates on Crashes at Rural Intersections," Journal of Transportation Engineering, vol. 132, no. 4, pp. 282-292, 2006. 91