National Academies Press: OpenBook

Pedestrian Safety Prediction Methodology (2008)

Chapter: Chapter 2. Literature Review

« Previous: Chapter 1. Introduction
Page 6
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 6
Page 7
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 7
Page 8
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 8
Page 9
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 9
Page 10
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 10
Page 11
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 11
Page 12
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 12
Page 13
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 13
Page 14
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 14
Page 15
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 15
Page 16
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 16
Page 17
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 17
Page 18
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 18
Page 19
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 19
Page 20
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 20
Page 21
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 21
Page 22
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 22
Page 23
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 23
Page 24
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 24
Page 25
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 25
Page 26
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 26
Page 27
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 27
Page 28
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 28
Page 29
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 29
Page 30
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 30
Page 31
Suggested Citation:"Chapter 2. Literature Review." National Academies of Sciences, Engineering, and Medicine. 2008. Pedestrian Safety Prediction Methodology. Washington, DC: The National Academies Press. doi: 10.17226/23083.
×
Page 31

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

3 CHAPTER 2. LITERATURE REVIEW A review of completed and ongoing research has been conducted to identify current knowledge or methodologies that may be appropriate for predicting pedestrian safety effects related to roadway geometrics and other factors. The review of literature and research in progress addressed the following issues: • Current knowledge based on completed research concerning pedestrian safety and its relationship to pedestrian exposure and to geometric design, traffic control, and other features that can be modified by highway agencies • Ongoing research on pedestrian safety relationships • Safety modeling approaches that have been applied to pedestrian safety in completed or ongoing work • Other safety modeling approaches, including approaches that are being used in the development of other portions of the HSM, whether or not those modeling approaches have been applied to pedestrian safety Key sources that have been consulted in the literature review include: • The AASHTO Guide for Planning, Design, and Operation of Pedestrian Facilities (1), which was written, in part, by the University of North Carolina (UNC) Highway Safety Research Center (HSRC) in NCHRP Project 15-20, and was revised and edited by Midwest Research Institute (MRI) in NCHRP Project 20-7(161) • The Guide for Reducing Collisions Involving Pedestrians (2) written by HSRC in NCHRP Report 500, Volume 10 • The evaluation of safety differences between marked and unmarked crosswalks at intersections performed by HSRC for FHWA (3) • The evaluation of the safety differences between roadways with and without sidewalks performed by HSRC (4) • The synthesis on pedestrian safety at channelized right turns (5) prepared by MRI in NCHRP Project 3-72 • The draft material on pedestrian safety prepared by iTRANS Consulting, Inc., for the interim report of NCHRP Project 17-27 (6) and for eventual use in Parts A and D of the HSM FHWA’s Pedestrian and Bicycle Crash Analysis Tools (7) (PBCAT) has been used as a source of ideas for the HSM methodology, as has a recent paper by Lyon and Persaud (8) entitled

4 Pedestrian Collision Models for Urban Intersections. Another recent paper by Lyon et al. (9) has proposed an “index of pedestrian activity” for use in safety modeling. The results of the literature review are presented below in sections that address the effect of intersection characteristics on pedestrian safety, the effect of roadway segment characteristics on pedestrian safety, predictive models for pedestrian crashes, estimating pedestrian exposure data from count periods less than 24 hours, and estimating pedestrian volumes from land use characteristics. EFFECT OF INTERSECTION CHARACTERISTICS ON PEDESTRIAN SAFETY Research included in this section of the literature review is focused on the pedestrian safety effect of various intersection characteristics. Many pedestrian safety studies use non- crash-based measures of effectiveness, such as: • Changes in pedestrian behavior (e.g., increased use of pedestrian pushbuttons, increased use of crosswalks, decreased crossings on red signal, etc.) • Changes in motorist behavior (e.g., increased number of yields, reduction in speed, etc.) • Other surrogate safety measures (e.g., conflicts, avoidance maneuvers, etc.) In contrast to these measures of safety, this literature review is limited to those studies that focused on pedestrian crashes as the unit of analysis. The following discussion combines literature reviewed in this research and literature reviewed for the Project 17-27 interim report (6). The effects of the following intersection characteristics on pedestrian safety are addressed: • Pedestrian volume • Traffic volume • Crossing width • Raised pedestrian crosswalks • Crosswalk marking • Crosswalk illumination • Median refuge islands • Raised intersections • Bus stop location • Pedestrian-related signing • Pedestrian signal type • Pedestrian signal timing • Right-turn-on-red • One-way streets

5 Pedestrian Volume Pedestrian volume, also called pedestrian exposure, has been found by several studies to be one of the most influential factors in predicting pedestrian crashes. Zegeer et al. (10) studied pedestrian crashes at 1,297 signalized intersections in 15 cities. They collected data on 2,081 crashes over a three to six year period. The analysis found that the volume of pedestrians crossing at an intersection was the most influential variable in explaining the variation in pedestrian crashes. The frequency of pedestrian crashes generally increased with increasing pedestrian volume. Intersections with less than 1,200 pedestrians per day experienced an average of 0.178 pedestrian accidents per intersection per year, while intersections with 1,200 pedestrians per day or more experienced an average of 0.553 pedestrian accidents per intersection per year, and intersections 3,500 pedestrians per day or more experienced 1.002 pedestrian accidents per intersection per year. Brude and Larsson (11) studied the effect of pedestrian and traffic volumes on pedestrian crashes at intersections in Swedish municipalities. The analysis included crash data from 285 intersections (121 signalized, 155 unsignalized, and 9 roundabouts). Each site had 100 or more pedestrian crossings per day. They found pedestrian volume to have a significant and positive relationship to pedestrian crashes in a single predictive model that covered all intersection types. Lyon and Persaud (8) developed pedestrian crash prediction models for urban 3-leg and 4-leg signalized and unsignalized intersections in Toronto. They included 684 4-legged signalized intersections averaging 7.72 pedestrian crashes over 11 years, 263 3-legged signalized intersections averaging 4.05 pedestrian crashes over 11 years, and 122 3-legged stop controlled intersections averaging 1.3 crashes over 11 years. Pedestrian volume was found to have a significant and positive relationship to pedestrian crashes for all intersection types. Additionally, models that included pedestrian volumes had a better fit than those without pedestrian volumes. Zegeer et al. (3) compared the safety aspects of marked and unmarked crosswalks at uncontrolled locations under various traffic and roadway conditions. They studied 1000 marked crosswalks and 1000 matched (comparison) unmarked crosswalks (about 80 percent of these sites were at intersections. They analyzed 229 pedestrian crashes that occurred over an average of five years. Pedestrian volume was found to have a significant and positive relationship to pedestrian crashes in predictive models developed for both marked and unmarked crosswalks. Traffic Volume Traffic volume has also been found to be a major contributing factor to pedestrian crashes. Zegeer et al. (10) found that traffic volume was the second most important factor in explaining pedestrian crashes. The analysis showed that, for a specific pedestrian volume level, the frequency of pedestrian accidents per intersection per year generally increased as the motor vehicle traffic volume increased. Studies by Brude and Larsson (11) and Zegeer et al. (3) also found that the number of incoming vehicles per day at intersections was a significant and positive variable in predicting pedestrian crashes.

6 The above studies used data on average daily traffic (ADT), which is generally collected on a regular basis for most major roads in developed areas. Some studies also included turning vehicle movements. Leden (12) compared pedestrian crashes with left-turning traffic at semi- protected left-turn schemes with pedestrian crashes with right-turning vehicles. He found that an increase in left-turn volume was associated with a larger increase in pedestrian crashes compared to a similar increase in right-turn volumes. Lyon and Persaud (8) include left-turning traffic volume in predictive models for pedestrian crashes. They found that the proportion of left- turning traffic to total traffic had a significant and positive relationship to pedestrian crashes for signalized intersections. For stop-controlled intersections, left-turning traffic volume was found to be more important than total entering volume and was included as a solitary factor (not as a proportion). Crossing Width The recent report for NCHRP Project 17-27 (6) found only a few studies that used crash data to analyze the effect of narrowing the crossing width by curb extensions or other means; many other studies have used other measures such as vehicle speeds and pedestrian/vehicle interactions to evaluate the effect of crossing width on pedestrian safety. Project 17-27 found two studies that used crashes in their evaluation of road narrowing. The first study that addressed the effect of crossing width was a summary of pedestrian research in the United Kingdom. Davies (13) reported results from Nottingham (by Thompson and Heyden, 1991) where curb extensions were extended 8.2 ft [2.5 m] into the street and included “substantial lengths of guardrail,” (assumed to be protective railing for pedestrians). The authors reported a reduction in average pedestrian crashes from 4.7 to 1 per year after the treatment. Insufficient information was available to determine an AMF from this study. The second study that addressed the relationship between crossing width and pedestrian safety was a 2004 review that examined two studies from Denmark and Norway (14). The studies do not state if the intersections studied had 3- or 4-legs. Both were simple before-after studies not controlling for any confounding factors and therefore both of these studies have been rated as low quality. Accordingly, the standard error has been adjusted by a factor of 3. The result of a meta-analysis of these two studies is provided in Table 1, and is highly uncertain. As a general note related to NCHRP Project 17-27 (6), standard errors of the results were adjusted depending on the quality of the study. For studies considered high quality, the adjustment factor was close to 1.0. For studies rated as low quality, the adjustment factor was 2 or higher. A 2004 synthesis of pedestrian safety research examined other studies on reducing crossing speed (15). However, the studies cited in that synthesis used vehicle speeds and pedestrian/vehicle conflicts as measures of effectiveness instead of crashes.

7 TABLE 1. Effects on Crashes of Road Narrowing and Curb Extensions at Intersections (6, 14). Treatment/element Setting Intersection type and volume Crash type and severity Index of effectiveness (tadjusted) Estimate of Std. Error (s) Widen sidewalks at intersections Not reported Not reported All types, injury 1.116 1.260 Due to the design of the crash-based studies that have been reviewed, it is uncertain whether crossing width has a significant impact on pedestrian safety. Raised Pedestrian Crosswalks Elvik and Vaa (14) recently performed a meta-analysis of four international studies that evaluated raised pedestrian crosswalks. Three of these studies have been rated as low quality, and one study (15) was rated as medium-low quality. Standard errors have been adjusted by a factor of 3 in the three low quality studies and by a factor of 2.2 in the medium-low quality study. Intersection types and volumes were not reported by Elvik and Vaa. The resulting indices of effectiveness are presented in Table 2. Based on these values, raised pedestrian crosswalks appear to reduce accidents. The effects may be overstated, as none of the studies have controlled for regression-to-the-mean or long-term trends in accident frequency. Standard errors are very large. TABLE 2. Effects on Crashes of Raised Pedestrian Crossings at Intersections (6, 14). Treatment/element Setting Intersection type and volume Crash type and severity Index of effectiveness (tadjusted) Estimate of Std. Error (s) Raised pedestrian crosswalks Not reported Not reported All types, injury 0.642 0.543 Raised pedestrian crosswalks Not reported Not reported Pedestrian, injury 0.545 0.937 Raised pedestrian crosswalks Not reported Not reported Vehicle, all severities 0.697 0.667 Based on these studies, raised crosswalks appear to reduce pedestrian crashes. Crosswalk Marking A 2004 review of pedestrian safety research by Campbell et. Al. (15) reviewed 13 crash- based studies on crosswalk markings, including the study by Zegeer et al. for FHWA (3) that is considered the most comprehensive crash-based study on this topic. The 13 studies had conflicting findings, partly because of the different methodologies and data limitations which existed in some of the studies. The Zegeer et al. study (3) found that: “…on two-lane roads, the presence of a marked crosswalk alone at an uncontrolled location was association with no

8 difference in pedestrian crash rate, compared to an unmarked crosswalk. Further, on multilane roads with traffic volumes above 12,000 vehicles per day, having a marked crosswalk alone (without other substantial improvements) was associated with a higher pedestrian crash rate (after controlling for other site factors) compared to an unmarked crosswalk.” Crosswalk Illumination The 2004 review (15) covered two crash-based studies on crosswalk illumination. The first study was conducted in two parts—a pilot test and a follow-on test (16). In both stages, crashes were counted before and after the installation of sodium floodlights at the pedestrian crossing. The results of their tests, showing a significant decrease in nighttime pedestrian crashes, are shown in Table 3. TABLE 3. Crash Effects of Providing Sodium Floodlights at Pedestrian Crossings (Perth, Australia) (16). Pedestrian crashes Crashes involving vehicles alone Day Night Total Day Night Total Pilot Test (6 crossings) 5 years before 19 (1) 7 (1) 26 (2) 5 1 6 5 years after 21 (1) 2 23 (1) 9 0 9 Follow-on Test (57 additional crossings) 2 years before 57 (2) 32 (1) 89 (3) 19 2 21 2 years after 58 (2) 13 (1) 71 (3) 18 (1) 1 19 (1) Fatalities are shown in parentheses. The second study by Polus and Katz (17) conducted a crash analysis on sites before and after a combination of illumination and signing was installed at 99 test sites. Thirty-nine unilluminated control sites were also used. The results of the study, showing a decrease in crashes after illumination, are shown in Table 4. TABLE 4. Effects of Crosswalk Illumination on Pedestrian Crashes (Israel) (17). Number of night crashes Before After Illuminated sites 28 16 Unilluminated sites 10 16 Since daylight crashes remained relatively unchanged, and factors such as pedestrian and vehicle volume and weather remained unchanged, the decrease in nighttime crashes was attributed to the illumination effect.

9 The consensus from these two studies indicates that crosswalk lighting serves to decrease pedestrian crashes. Median Refuge Islands The Project 17-27 interim report (6) reviewed two crash-based studies that examined the effect of median refuge islands on pedestrian safety. Zegeer et al. (3) studied 2,000 crossing sites in 30 cities; all sites were in urban or suburban areas and included primarily arterial and collector streets. Streets covered a range of speed limits, typically 25 to 40 mph [40 to 64 km/h]. Sites were selected within a variety of area types (i.e., residential, downtown, commercial, urban fringe, etc.). Zegeer et al. found that the presence of a raised median or refuge island was associated with a significantly lower rate of pedestrian crashes on multi-lane roads (compared to no median or refuge island). This was true at marked as well as unmarked crosswalks. All sample sites used in the study were uncontrolled crossings at intersection (i.e., no traffic signals or STOP-control on intersection approach of interest) or mid-block locations. The presence of painted (not raised) medians or islands and two-way-left-turn lanes provided no significant reduction in pedestrian crash rate on multi-lane roads. The study collected and controlled for pedestrian and vehicle exposure, along with other site variables in the analysis. The purpose of a study by Lalani (18) was to compare personal injury crashes before and after installation of “Double-D” shaped refuge islands at 120 sites, including intersection and mid-block, marked, and unmarked locations in London, England. These islands were installed in conjunction with other roadway improvements, such as anti-skid surfacing, illuminated bollards, bus lanes, and crosshatch markings. This study found that refuges reduced vehicle crash frequency but increased pedestrian accident frequency at intersections. Lalani also determined that: • At intersections, vehicular accident frequency was significantly reduced only when the refuge islands were reinforced with hatch markings to channelize motor traffic; • At midblock locations, vehicular accidents were only reduced where the islands had internally illuminated bollards; and • Pedestrian accidents were only reduced at sites where the refuge islands were constructed on roads next to high pedestrian generators. (It is unclear if Lalani is referring to intersection or midblock or both.) It is possible that the results of the Lalani study (i.e., the increase in pedestrian crashes after installation of refuge islands) may be a manifestation of the fact that more pedestrians are drawn to use the crossing after a refuge island is installed. A study of all of the pedestrian crashes along a road section (with corresponding pedestrian exposure) and controlling for pedestrian exposure at the crossings would allow for quantifying this effect. Based on these two crash-based studies, it is uncertain whether a median refuge island significantly impacts pedestrian safety.

10 Raised Intersections The Project 17-27 interim report (6) mentioned only one study that examined the effect of raised intersections on pedestrian crashes. This study by Schull et al. used a simple before-after design and did not control for the potential effects of confounding factors. Details on traffic control at the intersections were not provided. The study was rated as low quality, and the results were inconclusive concerning the safety effectiveness of raised intersections. Bus Stop Location In their predictive crash model, Zegeer et al. (10) found that sites nearer to bus routes had more crashes and that the difference was small but significant. Neither the 2004 review (15) nor the Project 17-27 interim report (6) included any crash-based studies concerning the effect of bus stop location on pedestrian safety. Pedestrian-Related Signing Neither the 2004 review (15) nor the Project 17-27 interim report (6) included any crash- based studies concerning the effect of pedestrian-related signs on pedestrian safety. Pedestrian Signal Type Pedestrian signal indications can be configured in several ways. The walk indications may be solid or flashing. There may also be a countdown displayed with the signal. The Project 17-27 interim report (6) did not find any studies that related the type of pedestrian signal to pedestrian crashes. Pedestrian Signal Timing The design of pedestrian signals can involve timing schemes such as concurrent (standard) timing, exclusive phase (also called scramble timing), and other special timing sequences. The Project 17-27 interim report (6) reviewed a study from Israel by Zaidel and Hocherman (19) that analyzed 1,310 pedestrian crashes at 320 signalized intersections. Higher rates of pedestrian crashes were found at intersections with higher pedestrian and vehicle volumes, as well as at more complex intersections (i.e., the most legs or potential points of conflict). The type of signal timing provided for pedestrians had only a slight effect on pedestrian crashes and no effect on vehicle injury crashes, especially where vehicle volumes were low (less than 18,000 ADT). Intersections with exclusive phases for pedestrians had fewer crashes where vehicle and pedestrian volumes were higher. These results concur with the results of Zegeer et al. (20, 21).

11 A 1984 study by Robertson and Carter (22) found pedestrian signal indications can have different effects on pedestrian crashes, depending on the type of intersection. Robertson and Carter found that pedestrian signal indications reduce pedestrian crashes at some intersections, have little or no effect at others, and may actually increase crashes at yet other sites. The presence of pedestrian signals in itself did not have a statistically significant effect on pedestrian and vehicle delay, but the signal timing scheme had a major influence on delay. The authors suggested further study to identify the types of intersections where pedestrian signals would be most effective. AMFs could not be derived from the study results. Zegeer et al. (10) studied the effect of pedestrian signal timing at 1,297 intersections. The analysis specifically focused on pedestrian signal designs, including sites with concurrent (standard) timing, sites with exclusive timing, and sites with no pedestrian signal. After adjusting for pedestrian volume, traffic volume, and street operation, Zegeer et al. found that exclusive timing resulted in significantly lower mean crashes per year than concurrent pedestrian signal or no pedestrian signal. The findings from these studies indicate that the type of pedestrian signal can have an effect on pedestrian crashes, especially at intersections with high pedestrian and vehicle volumes. Right-Turn-On-Red Operation at Signalized Intersections The Project 17-27 interim report (6) reviews two crash-based studies that examine the effect of right-turn-on-red (RTOR) on pedestrian crashes. A study by Preusser et al. (23) examined sites in four states and found that there was a small effect of increasing right turn crashes after RTOR went into effect. This study was conducted during the mid-1970s, at which time a number of states in the eastern portion of the U.S. adopted the “permissive” type of RTOR that was already common in the western U.S. The “Western” approach to RTOR allows this maneuver at all locations that are not otherwise marked by a prohibitory sign. Of course, motorists are expected to stop and yield to pedestrians, bicyclists, and oncoming vehicles prior to making a Western RTOR. Preusser et al. evaluated several of the eastern locations and found statistically significant increases in pedestrian crashes with right-turning vehicles after RTOR was introduced. Comparison of computerized accident data from the periods before and after implementation of RTOR rule showed the following increases in accident rates: • 43 percent for pedestrians in New York State • 107 percent for pedestrians in Wisconsin • 57 percent for pedestrians in Ohio • 82 percent for pedestrians in New Orleans It should be noted that these percentages are increases in very small numbers, since RTOR-pedestrian crashes are very rare. A second part of the Preusser et al. study involved analysis of actual police crash reports. From this analysis, the authors were able to identify a common crash scenario involving RTOR.

12 Often, a driver who is stopped prior to turning right focuses on traffic coming from the left in order to identify a gap adequate to permit his right turn. Consequently, the motorist does not see a pedestrian on his right and a conflict occurs when the turn is initiated. The Preusser team found that RTOR accidents account for 1 to 3 percent of all pedestrian and bicycle accidents. A study by Clark et al. (24) examined the effects of RTOR on pedestrian safety in South Carolina and Alabama. In South Carolina, crashes at signalized intersections involving right- turning vehicles for two years before and three years after the RTOR law was implemented were compared with accidents in the same period that did not involve right-turning vehicles. A similar comparison in Alabama covered three years before and five years after RTOR was instituted. Results showed a statistically significant increase during the after period in South Carolina for right-turning property damage crashes than for crashes not involving right turns. This was not true in Alabama. There was no statistically significant difference in the rate of change in fatal or injury crashes in either state when comparing right-turning vehicles to non- right-turning vehicles. Furthermore, there was no evidence of increased pedestrian crashes resulting from RTOR in either South Carolina or Alabama. Considering both the South Carolina and Alabama results, Project 17-27 recommended an AMF of 1.067 for RTOR. Previous analysis in the current research found that this AMF only marginally involved pedestrian crashes (25). The Project 17-27 interim report (6) also cited a 2003 study that examined the scope of the problem in Canada and the U.S. and found that RTOR-related pedestrian crashes made up less than 1 percent of all reported crashes and that the crashes that did occur were not usually severe. The studies on RTOR have shown mixed results with respect to the effect of RTOR on pedestrian crashes, but it is clear that RTOR-related pedestrian crashes are rare. One-Way Streets The 2004 review (15) cited a 1978 Canadian study that studied streets in the city core and found that there were fewer crashes on one-way streets than two-way streets. These crashes were estimated to comprise about 10 percent of the total crashes in the city. The 2004 review also cites a 1973 study that examined 253 pedestrian crashes over a five-year period in New York City. The results showed the benefit to pedestrian safety of the simplified operation of one-way streets. In their predictive crash model, Zegeer et al. (3) found that there is no statistically significant difference in pedestrian crash risk between one-way and two-way streets, after controlling for other significant traffic and roadway factors. The consensus of these studies indicates that one-way streets have a higher level of pedestrian safety versus two-way streets. However, the effect of pedestrian exposure on these findings has not been studied; consideration of pedestrian exposure would be needed to more accurately define the effect of one-way streets on pedestrian safety.

13 EFFECT OF ROADWAY SEGMENT CHARACTERISTICS ON PEDESTRIAN SAFETY Research included in this section of the literature review is focused on the effect of roadway segment characteristics on pedestrian safety. Many studies have used non-crash-based measures of effectiveness, such as: • Changes in pedestrian behavior (e.g., increased use of crosswalks, types of crossing behaviors, etc.) • Changes in motorist behavior (e.g., increased number of yields, reduction in speed, etc.) • Other surrogate safety measures (e.g., conflicts, avoidance maneuvers, etc.) In contrast to these measures of safety, this literature review is limited to those studies that focused on pedestrian crashes as the unit of analysis. Many of the studies cited in this literature review are quoted from the interim report of NCHRP 17-27 (6), which consisted of draft chapters for the upcoming HSM. The effects of the following roadway characteristics on pedestrian safety are addressed: • Sidewalks • Midblock raised pedestrian crossings • Illuminated crosswalk signs • Pedestrian overpasses and underpasses • Medians and pedestrian refuges • Traffic calming • Other factors without crash-based studies Sidewalks The Project 17-27 interim report (6) reviewed a 2002 study by McMahon et al. (26) that examined “walking along roadway” pedestrian crashes. The study used a case control methodology and applied conditional and binary logistic models to determine the effects of various roadway features and socioeconomic and other census data on the likelihood that a site is a pedestrian crash site. A total of 47 crash sites were found, which were matched with 94 comparison sites (i.e., one nearby and one far-away matched comparison site for each crash site) for analysis purposes. Comparison sites were selected which were similar to the crash sites in terms of number of lanes, traffic volume, roadway and shoulder width, vehicle speeds, area type, etc. Nearby comparison sites were selected within the same neighborhood and/or within approximately 1 mi [1.6 km] of the crash site. Far-away sites were matched sites that were selected in neighborhoods or areas on the other side of the county. Physical roadway features found to be associated with a significantly higher likelihood of having a “walking along roadway” pedestrian crash included lack of a walkable area and the absence of a sidewalk augmented by higher traffic volume and higher speed limits. Using “risk

14 ratio” and controlling for other roadway factors, the likelihood of a site with a sidewalk or wide shoulder (of 4 ft [1.2 m] or wider) having a “walking along roadway” pedestrian crash was 88.2 percent lower than a site without a sidewalk or wide shoulder at the sites studied. Increased pedestrian crash risk existed for higher speed limits and for higher traffic volumes. The authors stated that these results “should not be interpreted to mean that installing sidewalks would necessarily reduce the likelihood of pedestrian/motor vehicle crashes by 88.2 percent in all situations. However, the presence of a sidewalk clearly has a strong beneficial effect of reducing the risk of a “walking along roadway” pedestrian/motor vehicle crash.” AMFs were not developed from the results. A 1983 study by Tobey (27) examined roadway segment characteristics. This study investigated the relative risks of sidewalks and other traffic and roadway characteristics, using two different measures of pedestrian exposure: pedestrian volume (P) and pedestrian volume multiplied by vehicle volume (P×V). The percent of pedestrian crashes divided by the percent of pedestrian exposure was defined as the “hazard score” for each site, and “hazard scores” were compared for sites with various traffic and roadway characteristics. If the percent of pedestrian crashes was greater than the percent of pedestrian exposure, then the “hazard score” was greater than 1.0, or a pedestrian crash risk greater than average. Where the percent of crashes was less than the percent of exposure, the “hazard score” was computed as the percentage of exposure divided by the percent of crashes and assigned a negative sign (i.e., a negative sign represents a safer than average condition). Sites with no sidewalks or pathways had the highest “hazard scores,” with values of +2.6 (using P as the exposure measure) and +2.2 (using P×V as the exposure measure). This is compared to “hazard scores” of +1.2 and +1.1 (using exposure measures of P and P×V, respectively) for sites with a sidewalk on one side of the road only. Sites with sidewalks on both sides of the road had “hazard scores” of -1.2 (using exposure measures of P and P×V), which represents a safer condition than having sidewalks on one side or no sidewalks at all. These studies indicate that sidewalks and walkways can reduce pedestrian crashes, especially “walking along roadway” crashes. Midblock Raised Pedestrian Crossings The Project 17-27 interim report (6) addresses four studies related to raised pedestrian crossings that are cited in the AASHTO Guide for the Planning, Design, and Operation of Pedestrian Facilities (1). These studies contain a total of ten estimates of effect. None of the studies have controlled for regression-to-the-mean or long-term trends in accident occurrence. For raised pedestrian crosswalks, 8 estimates have been rated as low quality and 2 as medium low quality. Thus, a high quality quantification of safety is not available for this measure. The safety effects of raised pedestrian crosswalks in Table 5 refer to pedestrian crashes or crashes involving motor vehicles only. The latter category includes all crashes that involve one or more motor vehicles, but not a pedestrian. It cannot be ruled out that the summary estimates presented in Table 5 are confounded by uncontrolled regression-to-the-mean and uncontrolled

15 long-term trends in crash occurrence. Standard errors have been adjusted by a factor of 3 for each low quality estimate of effect and a factor of 2.2 for each medium low quality estimate of effect. TABLE 5. Effects on Injury Crashes of Raised Pedestrian Crosswalks (6). Data summarized Crash type and severity Index of effectiveness (tadjusted) Estimate of Std. Error (s) All studies (10 estimates) All crashes, injury 0.642 0.543 All studies (5 estimates) Pedestrian crashes, injury 0.545 0.937 All studies (5 estimates) Vehicle crashes, injury 0.697 0.667 These studies indicate that raised pedestrian crosswalks seem to reduce pedestrian crashes. However, the lack of control on potentially confounding factors and large standard errors introduces a high degree of uncertainty on quantifying the effect of raised crosswalks. Illuminated Crosswalk Signs The 2004 review (15) of pedestrian safety research cited a study that evaluated 20 locations in Tokyo, Japan (28) with illuminated crosswalk signs. An analysis was conducted on crashes occurring before and after the installation of the illuminated crosswalk signs. The results showed that within 656 ft [200 m] on either side of the crosswalk, pedestrian crashes increased by 4.8 percent and other unrelated crashes increased by 2.4 percent. Within 164 ft [50 m], both crash types increased by 11.4 percent. The researchers concluded that the illuminated crosswalk signs showed no benefit to pedestrian crashes, but it is unclear whether the study accounted for factors such as annual volume increase. Pedestrian Overpasses and Underpasses The 2004 review (15) found one before-after study that examined pedestrian crashes at 31 locations in Tokyo, Japan, where pedestrian overpasses had been installed (28). The results (see Table 6) show the crashes occurring 6 months before and 6 months after installation. There was a reduction seen in the occurrence of crashes related to pedestrian crossing events, as well as a greater reduction of daytime pedestrian crashes over nighttime pedestrian crashes. TABLE 6. Comparison of Crashes Before and After Installation of Pedestrian Overpasses (Tokyo, Japan) (28). 656-ft section 328-ft section Type of crash Before After Reduction Before After Reduction Related Crashes 2.16 0.32 85% 1.81 0.16 91% Nonrelated Crashes 2.26 2.77 23% 1.65 1.87 –14% Total 4.42 3.09 30% 3.46 2.03 41%

16 The limited number of crash-based studies on the effect of pedestrian overpasses and underpasses does not provide definitive results. The Tokyo study indicates that overpasses can reduce pedestrian crashes, but it can be assumed that the safety effect is directly related to the level of usage of the overpasses by pedestrians, which can vary widely. Medians and Pedestrian Refuges The Project 17-27 interim report (6) reviewed a 2002 study by Zegeer et al. (3) whose primary goal was to examine the safety effects of marked versus unmarked crosswalks. One thousand crosswalks of each type were included from 30 U.S. cities. Predictive models based on five years of crash data showed that raised medians were associated with a 46 percent reduction in pedestrian crashes on marked crosswalks and a 39 percent reduction on unmarked crosswalks, for multi-lane roads with ADTs greater than 15,000 vehicles per day. Two studies on the effects of pedestrian refuge islands were also reviewed in the Project 17-27 report (6). Bacquire et al. (29) conducted a before-after analysis of raised pedestrian refuge islands. The analysis considered pedestrian and vehicle crashes at 28 sites from 3 years before to 3 years after installation. Although pedestrian crashes that were related to refuge islands were reduced by 73 percent, there was an increase in total crashes by 136 percent. The possibility that treatment sites were selected on the basis of crash history introduced potential regression-to-the-mean bias. The authors concluded that overall safety was not helped by the installation of raised pedestrian refuge islands. Cairney (30) reviewed a 1994 Australian Geoplan study which found that none of the four types of refuge islands examined was found to be very effective from a safety perspective. In fact, three types caused large increases in the adjusted pedestrian crash rate (the calculation of and adjustments made to the pedestrian crash rate were not reported), while only one type resulted in a slight reduction (2 percent reduction). However, according to Cairney, “…it seems inherently unlikely that pedestrian refuges did not reduce crashes. The method used in the Geoplan study compared crashes occurring at the site of the facility, before and after. Where pedestrian refuges are provided, it would be expected that pedestrians would be attracted to cross at this point—pedestrians who would otherwise have crossed some distance along the road, so that pedestrian flow is greatly increased at the refuge. A study of the crash history of the whole street where pedestrian refuges have been installed would therefore be necessary to determine whether there had been a reduction in pedestrian crashes.” A study by Bowman and Vecellio (31) that examined the safety effects of median types involved an analysis of 32,894 vehicular crashes and 1,012 pedestrian crashes that occurred in three U.S. cities (Atlanta, Georgia; Phoenix, Arizona; and Los Angeles/Pasadena, California). The median types which were compared were: (a) raised, (b) flush or two-way-left-turn-lane (TWLTL), and (c) no existing median (undivided). A variety of statistical tests were used, including t-tests, analysis of variance, and the Scheffe multiple comparison test. The authors did not have pedestrian volume data, but used area type (CBD and suburban areas) and land use as surrogate measures for pedestrian activity and developed pedestrian crash prediction models separately for the two area types.

17 The results of this analysis provide strong evidence that having some area of refuge (either a raised median or TWLTL) on an arterial CBD or suburban street provides a considerably safer condition for pedestrians than having an undivided road (i.e., with no refuge for pedestrians in the middle of the street). Furthermore, while this study found that suburban arterial streets with raised-curb medians had lower pedestrian crash rates, as compared to TWLTL medians, this difference was not statistically significant. This may be a clear indication that some refuge area (in the middle of wide streets) is more beneficial to pedestrian safety when crossing streets than having no refuge area. However, the safety benefits for a raised median vs. a TWLTL were not quantified. Based on the study results, Bowman and Vecellio recommended that in CBD areas, whenever possible, divided cross-sections should be used due to their lower crash rates for pedestrians and motor vehicles. Cairney (30) also presented the results of the following studies related to medians: • A study by Moore and McLean reports early research in New South Wales by Johnson in 1962 and Leong in 1970. These studies found that providing narrow medians reduced vehicle-to-vehicle crashes but had no effect on pedestrian crashes. The author did not report the sample size of the studies, the type of statistical analysis, or whether data variables such as pedestrian exposure were collected and controlled for in the analyses. Crash data were not reported. • A 1986 study by Scriven in Adelaide, South Australia, found that medians were effective in reducing pedestrian crashes. On arterial roads, pedestrian crash rates were directly related to median width. Roads with the narrowest medians (4 ft [1.2 m]) had pedestrian crash rates that were four times higher than routes with the widest medians (10 ft [3.0 m]). • A 1994 study by Claessen and Jones found that replacing a 6-ft [1.8-m] painted median with a wide raised median reduced pedestrian crashes by 23 percent. According to Cairney, this conclusion was consistent with Scriven’s finding that pedestrian crash rates for roads with 10-ft [3.0-m] medians were 33 percent lower than for roads with 4-ft [1.2-m] painted medians. Overall, these studies indicate that medians and pedestrian refuges serve to reduce pedestrian crashes. The emphasis on safety effectiveness indicates that raised medians are better than nonraised, and wider medians and refuges are better than narrow ones. Traffic Calming Although past studies indicate that traffic calming may cause a decrease in total crashes, no crash-based studies were found to determine any significant effect of traffic calming on pedestrian crashes.

18 Other Factors Without Crash-Based Studies No crash-based studies were found for the following roadway segment (midblock) factors: • Pedestrian-activated flashing yellow beacon • Overhead electronic LED signs (animated eyes) • In-pavement lighting EXISTING PREDICTIVE MODELS FOR PEDESTRIAN CRASHES This section reviews existing predictive models for pedestrian crashes that have been developed in previous work. The first portion of this section presents the provisional approach for estimating the frequency of vehicle-pedestrian crashes in the HSM methodology already developed as part of this research. This provisional approach is being replaced with the pedestrian safety prediction methodology presented in this report. The second portion of this section reviews four previous pedestrian modeling efforts reported in the literature. Current Approach to Estimating Pedestrian Safety in the HSM Methodology Separate pedestrian safety estimation procedures are used for intersections and roadway segments. Each set of procedures is presented below (25). Intersections Safety predictions for a particular intersection are developed as a combination of base models, calibration factors, and accident modification factors (AMFs) using the following general approach: Nint = (Nbi + Npedi + Nbikei) Ci ( 1 ) Nbi = Nbibase (AMF1i AMF2i AMF3i AMF4i AMF5i) ( 2 ) where: Nint = predicted number of total intersection-related accidents per year after application of accident modification factors Nbi = predicted number of total intersection-related accidents per year (excluding vehicle-pedestrian and vehicle-bicycle collisions) Nbibase = predicted number of total intersection-related accidents per year for nominal or base conditions (excluding vehicle- pedestrian and vehicle-bicycle collisions) Npedi = predicted number of vehicle-pedestrian collisions per year Nbikei = predicted number of vehicle-bicycle collisions per year

19 Ci = calibration factor for at-grade intersections developed for use for a particular geographical area AMF1i … AMF5i = accident modification factors for intersections Four types of intersections on arterial roadways are considered: • Three-leg intersections with STOP control on the minor-road approach (3ST) • Three-leg signalized intersections (3SG) • Four-leg intersections with STOP control on the minor-road approaches (4ST) • Four-leg signalized intersections (4SG) Specific procedures for estimating Nbi are presented in the HSM methodology. The number of vehicle-pedestrian collisions per year for an intersection is estimated from Nbi as: Npedi = Nbi fpedi ( 3 ) where: fpedi = pedestrian safety adjustment factor Table 7 presents the values of fpedi for use in Equation (3). All vehicle-pedestrian collisions are considered to be fatal-and-injury accidents. These adjustment factors are based upon average proportions of pedestrian crashes to total crashes, for the given intersection type. TABLE 7. Pedestrian Safety Adjustment Factors for Intersections (25). Intersection type Pedestrian safety adjustment factor (fpedi) 3ST 0.008 3SG 0.005 4ST 0.016 4SG 0.017 NOTE: These factors apply to the methodology for predicting total accidents (all severity levels combined). All pedestrian collisions resulting from this adjustment factor should be treated as fatal-and-injury accidents and none as property-damage-only accidents. Roadway Segments Safety predictions for a particular roadway segment are developed as a combination of base models, calibration factors, and AMFs using the following general approach: Nrs = (Nbr + Npedr + Nbiker) Cr ( 4 ) Nbr = Nbrbase (AMF1r AMF2r AMF3r) ( 5 )

20 where: Nrs = predicted number of total roadway segment accidents per year Nbr = predicted number of roadway segment accidents per year excluding vehicle-pedestrian and vehicle-bicycle collisions Nbrbase = predicted number of total roadway segment accidents per year for nominal or base conditions excluding vehicle-pedestrian and vehicle-bicycle collisions Npedr = predicted number of vehicle-pedestrian collisions per year Nbiker = predicted number of vehicle-bicycle collisions per year AMF1r … AMF3r = accident modification factors for roadway segments Cr = calibration factor for roadway segments developed for use for a particular geographical area Five types of roadway segments are considered: • Two-lane undivided arterials (2U) • Three-lane arterials including a center TWLTL (3T) • Four-lane undivided arterials (4U) • Four-lane divided arterials (i.e., including a raised or depressed median) (4D) • Five-lane arterials including a center TWLTL (5T) Specific procedures for estimating Nbr are presented in the HSM methodology. The number of vehicle-pedestrian collisions per year for a roadway segment is estimated from Nbr as: Npedr = Nbr fpedr ( 6 ) where: fpedr = pedestrian safety adjustment factor Table 8 presents the values of fpedr for use in Equation (6). All vehicle-pedestrian collisions are considered to be fatal-and-injury accidents. These adjustment factors are based upon average proportions of pedestrian crashes to total crashes, for the given roadway type. In applying Table 8, roadways with traffic speeds or posted speed limits of 48 km (30 mph) or less should be considered low speed; roadways with traffic speeds or posted speed limits greater than 48 km (30 mph) should be considered intermediate or high speed. The original version of Table 8 in the first and second drafts of the HSM chapter used the column headings “urban” and “suburban;” these column headings have been changed in the third and fourth drafts of the HSM Chapter to “low speed” and “intermediate or high speed,” because the use of the terms “urban” and “suburban” was confusing to some reviewers.

21 TABLE 8. Pedestrian Safety Adjustment Factors for Roadway Segments in Urban and Suburban Areas (25). Pedestrian safety adjustment factor (fpedr) Road type Low speed Intermediate or high speed 2U 0.031 0.003 3T 0.030 0.002 4U 0.044 0.008 4D 0.018 0.006 5T 0.036 0.004 Note: These factors apply to the methodology for predicting total accidents (all severity levels combined). All pedestrian collisions resulting from this adjustment factor should be treated as fatal-and-injury accidents and none as property- damage-only accidents. Predictive Models From the Literature Four studies have previously developed predictive models of pedestrian crashes (3,8,11,12). Three of these studies (3,8,12) estimated negative binomial regression models that used the following functional form: )X..βXβPedVolβADTβexp(β nn3x210ped ++++=N ( 7 ) where n0 ...ββ are coefficients to be estimated. Nped is the expected number of pedestrian crashes, ADT is the annual average daily traffic (i.e., vehicular volume), PedVol is the annual average daily pedestrian volume, and n3 XX ... represent other site characteristics such as proportion of left-turn volume, number of lanes, speed limit, presence/absence of a crosswalk, and presence/absence of a median. Collectively, the following conclusions can be made from these four studies: 1. Most of the studies that have tried to predict pedestrian crashes have been conducted at intersections. Even the sample in the Zegeer et al. (3) study that looked at both intersections and nonintersection locations consisted mostly of intersections (more than 80 percent). 2. As expected, increase in total traffic volumes and pedestrian volumes lead to higher pedestrian crashes. However, the coefficient for pedestrian volumes in most of these models is significantly less than 1.0. This implies that the relationship between pedestrian volumes and pedestrian crashes is nonlinear (e.g., a 50 percent increase in pedestrian volumes will lead to a less than 50 percent increase in pedestrian crashes, all other things being equal). 3. Based on the work by Lyon and Persaud (8), it is clear that the amount of turning traffic plays an important role in vehicle-pedestrian crashes. In fact, at 3-leg intersections with minor-road STOP control in Toronto, left-turn volumes were

22 found to be more important than total volumes in predicting the number of pedestrian crashes. However, it is also likely that many jurisdictions do not have resources to collect turning volume information on a regular basis. 4. Except for Zegeer et al. (3), most of the work has focused on trying to relate just pedestrian/vehicle volumes with pedestrian crashes. Zegeer et al. (3) included information on other site characteristics and found median type, number lanes, and marked/unmarked crosswalks to be associated with pedestrian crashes. 5. Due to the limited frequency of pedestrian crashes, a large sample of sites with many years of data is needed to estimate such models. For example, Lyon and Persaud (8) utilized 122 intersections in the three-leg STOP-controlled group (the smallest intersection group) and compiled 11 years of data at these locations. Zegeer et al. (3) collected data from 2,000 locations. ESTIMATING PEDESTRIAN EXPOSURE DATA FROM COUNT PERIODS OF LESS THAN 24 HOURS The availability and use of pedestrian exposure data are important in the development of pedestrian crash models in this study. Pedestrian exposure at an intersection may be represented by such measures as pedestrian crossings per hour or estimated pedestrian volume related to the number of pedestrians crossing the street within the intersection (or on the approach legs of interest). On roadway segments, the pedestrian exposure may include the number of pedestrians who cross the street along the segment, such as for use in modeling pedestrian crashes involving street crossings. The number of pedestrians walking along the roadway may be the most appropriate pedestrian exposure measure, however, for modeling crashes involving pedestrians “walking along the road” on roadway segments. The challenge in obtaining such pedestrian exposure data is that few public agencies routinely collect pedestrian exposure data on a large number of sites, and thus, most (if not all) of the pedestrian exposure data which is needed for this study must be assumed to be collected by the project team at the selected sample sites. The collection of pedestrian exposure can be quite labor-intensive for an assumed sample of hundreds or thousands of sites, as will be needed for modeling purposes in this study. Therefore, practical and cost-efficient data collection strategies must be selected. A methodology for estimating daily pedestrian volumes at intersections was developed and utilized by Zegeer et al. (3) The methodology involved using 24-hour pedestrian counts from one city (Seattle) to develop estimated hourly pedestrian percentages for various area types (i.e., CBD, urban fringe, and residential). Counts of 11 hours were collected at multiple sites in several other cities to assist in determining hourly pedestrian volume distributions. Then, one- hour pedestrian volume counts were collected at each of the 2,000 crossing sites, and hourly adjustment factors were used to compute estimated daily pedestrian volumes. It was found, for example, that the 12-hour period from 7:00 a.m. to 7:00 p.m. represented 86 percent of the daily pedestrian crossing volume. Hourly pedestrian adjustment factors were developed for intersections in each of the three area types. For example, in CBD

23 areas, the pedestrian volume between 10:00 a.m. and 11:00 a.m. represented approximately 8.2 percent of the daily pedestrian crossing volume. Thus, a pedestrian count of 100 during that 10:00 to 11:00 a.m. period was expanded to an estimated daily pedestrian volume of 1,220 (100 divided by 0.082). It is recognized that this is a rough estimate of pedestrian exposure and that pedestrian activity may vary from day-to-day, by season, weather condition, etc. Pedestrian counts, therefore, should typically be collected on week days during relatively good weather conditions for use in developing pedestrian volumes. In fact, motor-vehicle ADT volume data are often the result of short-term counts with expansion factors applied. Using this type of methodology for estimating pedestrian exposure in this study will allow for collecting a much greater sample of total sites for use in the modeling. Only a few highway agencies have extensive records of pedestrian volume data at intersections or roadway segments. This lack of pedestrian volume data may be an impediment to applying an HSM safety prediction methodology that uses pedestrian volume as an input variable. The HSM methodology is intended primarily for application in the project development process. Extensive data collection activities are typically under taken during project development activities, so it seems reasonable to expect that the data collection could include pedestrian volumes. However, because pedestrian counts may not be a routine part of current data collection protocols, it also seems desirable to provide guidelines as part of the HSM methodology for HSM users to estimate pedestrian volumes. ESTIMATING PEDESTRIAN VOLUMES FROM LAND USE CHARACTERISTICS Transportation planners have been considering methods for estimating pedestrian volumes from land use characteristics. The following discussion reviews two related promising tools for estimating pedestrian volumes from land use characteristics. Further development of these tools would require resources beyond those available in this research, but tools like those would be reviewed below might eventually be considered for incorporation in the HSM. Space Syntax Raford and Ragland (32) observed a fairly significant correlation between predicted and observed pedestrian volumes (R2 = 0.7712, p<0.001) using a collection of modeling tools and simulation techniques called Space Syntax to analyze pedestrian movement and predict pedestrian volume in Oakland, CA. Street layout and connectivity generate “movement potentials” which are compared to sampled pedestrian counts at key locations and land use indicators such as population and employment density. The resulting correlations can be extrapolated to predict pedestrian volumes on a street-by-street level for the study area, which might include an entire city. The Space Syntax method first requires an axial line map for the analysis area. Raford and Ragland incorporated a total of 7000 street segments from Oakland. An “integration” value

24 for each segment is then developed using Space Syntax, generally based on the number of street segments accessible within 3 links. Population density and other demographic characteristics, including median household income, age, and race, were then applied to the Space Syntax model. The above description admittedly oversimplifies the Space Syntax algorithm that calculates “Integration.” That algorithm is explained in more detail in two papers: “The Common Language of Space: a way of looking at the social, economic, and environmental functioning of cities on a common basis.” Hillier, Bill, no date, available online at http://www.spacesyntax.org/ publications/commonlang.html; and “Spatial Distribution of Urban Traffic: civilizing urban traffic.” Croxford, Ben, et al., May 1995, available online at http://www.bartlett.ucl.ac.uk/ web/ben/copenhag.html. Pedestrian Flow Modeling Clifton et al. (33) have adapted the origin-destination-based demand model to standardize a modeling protocol for estimating pedestrian volumes on streets and sidewalks. The description for this model is based on conversations with the lead author, a presentation by a coauthor, and references to the model by Burnier and Clifton (34). In two case studies, the difference between pedestrian counts estimated by the model and those observed by actual pedestrian counts were within 4 percent on any given link in the modeled network. The model uses 14 trip purposes to differentiate among activities. Block-face detail is used for land use, networks, and trip making. Accessibility to activities influences the number of walking trips in the model. The barrier effects of streets (i.e., width, volume, speed, signals, etc.) are also used to influence accessibility and the distribution of trips. Stochastic path finding defines a distribution of walking routes from origins to destinations. Pedestrian trip generation characteristics (origins and destinations) for the Pedestrian Flow Model were derived from the 1996 NYMTC Household Survey, which included 11,000 households and projects a total of 59.3 million daily trips from 89,605 trip records. A total of 12,274 walking trips were recorded, estimating 9 million daily walking trips. Necessary input data includes sources readily available for most communities: • Census TIGER line files for streets • Census Block Group population and housing characteristics • Census Journey to Work data • Parcel-level property data and land uses • Aerial imagery, such as ortho-photography Easy access to data sources and the ability to apply limited manual interpretation and adjustments to the data as the model is developed increase the viability of the Pedestrian Flow Model. Detailed data can be assembled from available sources with reasonable effort. The model is sensitive to real-world factors that affect pedestrian travel such as land use, sidewalk network connectivity and quality, and barrier effects of streets. In addition outputs from the model can be used to evaluate pedestrian crash exposure and risk, and can inform development of safety priorities.

25 In addition to the two models discussed above, other methods for estimating pedestrian travel have been tried and applied in various settings. These methods include sketch planning techniques that predict pedestrian travel using simple calculations and rules of thumb about travel behavior; development and application of expansion factors to sample pedestrian counts; and developing estimates based on comparisons to similar locations. Each has limitations that render them excessively inaccurate for application to pedestrian crash risk assessment— primarily the lack of reliable street segment-level estimates. The two models discussed here are more promising for pedestrian crash risk assessment because they provide individual street segment-level estimates that have been accurate to within 4 percent of actual counts. Literature Review of Land Use Features Associated With Crash Risk Table 9 summarizes factors that literature indicates to be associated with increased or decreased pedestrian crash risk (i.e., with increased or decreased pedestrian crash frequencies). The sources in the literature on which Table 9 is based are discussed below. Identification of areas with increased pedestrian crash risk has generally relied on roadway characteristics, including geometry, traffic volumes, number of lanes, etc. Land use variables, however, have usually not been included, or are highly generalized when included. In a comprehensive analysis of pedestrian crashes at intersections, for example, Lee and Abdel-Aty (35) examined pedestrian and driver characteristics (age, gender, alcohol use), vehicle characteristics (type, speed), traffic/road geometric characteristics (traffic control device, divided/undivided, and number of lanes), and environmental characteristics (lighting, weather, and location). Of these, the location variable is the only land use-related variable, but it only indicates whether the location was urban or rural. The pedestrian characteristics variables of age and alcohol use might be proxies for adjacent or nearby land use variables, but spatial analysis examining the relationship between the distribution of pedestrians with those characteristics and the presence of the characteristics in adjacent land uses would be necessary before these variables could be replaced by land use variables. A statistically significant relationship between crashes involving pedestrians with certain demographic characteristics and those same crashes occurring in areas where those demographic characteristics are significantly present might indicate that the demographic land use variable would also correlate to pedestrian crash risk. LaScala et al (36) discuss methods of spatial analysis applied to variables that correlate to pedestrian injury crashes, and Austin et al. (37) presents a novel application of spatial statistics to land uses and expected pedestrian behavior. More detailed land use variables that may have significant correlation to pedestrian crash risk are available. These variables are discussed below. Epidemiological studies have long suggested additional correlations to various land use, demographic, or geographic characteristics. In addition, a limited number of recent planning or transportation studies have examined multiple variables and developed coefficients for numerous land use, demographic, or geographic variables that correlate with increased pedestrian crash frequencies.

26 TABLE 9. Land Use Variables Associated With Pedestrian Crash Risk. Land use variable (nongeometric roadway factor) Pedestrian crash risk Source Unemployment (less than 1.75% of population) Increases McMahon (26) Employment (number of employed) Increases Burnier and Clifton (34) Employment Decreases* Greenwald and Boarnet (40) Age of housing stock Increases McMahon (26) Single-parent households Increases McMahon (26) Population density Increases LaScala (36, 38) Population density Increases Burnier and Clifton(34) Population density Decreases* Saelens et al. (39) Proportion of children (age 0-15) Decreases LaScala (36, 38) Percent children Decreases Burnier and Clifton (34) Percent children (age 0-16) Decreases* Greenwald and Boarnet (40) Proportion of population age 16-29 Increases LaScala (36, 38) Proportion of population age 55+ Decreases LaScala (36, 38) Age (scaled) Decreases* Greenwald and Boarnet (40) Gender (proportion of males) Increases LaScala (36, 38) Education (% high school or higher) Decreases LaScala (36, 38) Education (% attended some college) Decreases Burnier and Clifton (34) Proximity to alcohol availability Increases LaScala (36, 38) Density of restaurants Increases LaScala (36, 38) Number of households in census tract Decreases Burnier and Clifton (34) Income (in $1000s) Decreases Burnier and Clifton (34) Race (percentage of nonwhite) Increases Burnier and Clifton (34) Vehicle ownership Increases Burnier and Clifton (34) Car ownership Decreases* Greenwald and Boarnet (36, 38) Density of roads Increases Burnier and Clifton (34) Percent parkland Increases Burnier and Clifton (34) Commercial accessibility Decreases Burnier and Clifton (34) Transit accessibility Increases Burnier and Clifton (34) Transit usage (number of users within area) Increases Hess (41); Vernez-Moudon (42) Mixed use Increases Burnier and Clifton (34) Land use mixture Decreases* Saelens et al. (39) Grocery store proximity (1/4 mile) Decreases Hess (41); Vernez-Moudon (42) Retail land use proximity (1/4 mile; 100K sq ft) Increases Hess (41); Vernez-Moudon (42) * Saelens et al. (39) and Greenwald and Boarnet (40) examined levels of walking associated with the land use variable, so their research looks at the denominator of the pedestrian crash risk (number of pedestrians). Increasing the denominator lowers the risk, so these variables are identified as decreasing risk, although there is conflicting research about whether pedestrian safety actually increases or decreases with higher volumes. This table presents the variable as described in the source paper. In some instances, the variable description has been modified to match other research or to clarify differences in variables. A meta-analysis of the results of the various reported studies would allow more accurate evaluation and comparison by adjusting all results to a comparable scale. In an examination of geographic correlates of pedestrian injury crashes, LaScala et al. (38) showed that pedestrian injury rates were spatially related to population density, age composition of the local population, unemployment, gender, and education. Availability of alcohol through bars was also directly related to pedestrian injury crashes. McMahon et al. (26) found that nongeometric roadway factors are associated with a significantly higher likelihood of being a crash site. Factors that were associated included high levels of unemployment, older housing stock, lower proportions of families within households, and more single-parent households. Burnier and Clifton (34) showed significant effects of land use on pedestrian crashes, as well as on pedestrian exposure. Urban downtown areas with high population density and high-

27 density roadway networks were found to have lower pedestrian crash risk. The model included thirteen variables, shown in Table 9. Saelens et al. (39) identify a number of land uses that correlate to increased levels of walking. Evidence suggested that more walking occurs in neighborhoods with higher population density, increased sidewalk connectivity, and mixtures of land uses. They also note that land use variables appear to add to variance accounted for beyond demographic predictors of walking. Greenwald and Boarnet (40) examine the built environment as a determinant of walking behavior. They found increased walking occurring in census tracts with higher percentages of younger residents, and decreased levels of walking in tracts with more older residents, higher rates of car ownership, and more employed household members. In addition, there are numerous epidemiological studies that show some level of association between land use or demographic variables and an increased crash risk. These often focus on narrow population segments; a few have been summarized in Table 9 along with those identified as part of the more comprehensive studies discussed above. At least 22 variables have been associated in various studies with an increased risk of pedestrian crashes. Table 9 shows some contradictions for some variables. For example, in separate studies employment levels and unemployment levels were each shown to increase crash risk. In one study, grocery store proximity decreased crash risk, while proximity to retail land use increased risk. These contradictions may be explained by methodology between studies, differences in study areas, or different definitions of the associated variable. In general, however, Table 9 shows that there are many potential land use, demographic, or geographic variables that warrant consideration in development of predictive models for pedestrian crash risk. In addition, most comprehensive studies, including LaScala et al. (38) and Burnier and Clifton (34), provide beta values for the coefficients because the variables are in different units or on different scales. As a result, Table 9 cannot present the relative strength of each variable because the scale of each variable is different. For example, population density may be presented on a scale, while education is typically presented with only two or three divisions that encompass the entire population. As described in a footnote above, a meta-analysis would allow effective comparison of relative significance of each variable. Another limitation of much of the existing research associating land use, demographic, or geographic variables with increased risk of pedestrian crashes is that the research does not identify whether the variable affects risk by changing the numerator of risk (number of crashes) or the denominator (volume). In many instances, which element of risk is affected may be intuitive, but probably not in all cases; however, knowing which element of risk is affected by the variable may not be necessary for developing predictive models of pedestrian crash risk. In some instances, though, knowing whether risk has increased because there are likely to be more pedestrians or whether there may be more actual crashes (perhaps for behavioral reasons associated with a population using a particular land use) may affect the predictive model. Studies have also examined land use variables and correlation to pedestrian crashes among some population groups. Roberts et al. (43) showed high levels of curb parking increases

28 the risk of child pedestrian injury. Von Kries et al. (44) associated neighborhoods with more playgrounds with decreased risk of child pedestrian injury, although the studies listed in Table 9 associate parks with an increased risk of pedestrian injuries in the general population. In an even narrower study, Agran et al. (45) found household crowding in the Hispanic population, inability of a parent to read well, and level of poverty were associated with an increased risk of pedestrian injuries for Hispanic children. Agran et al. also found that single parent households, car ownership, and level of education did not increase rate of injury among Hispanic children. Overall, the research indicates a complex relationship between land use, demographic, and geographic characteristics that is not yet well understood. Inclusion of these variables may improve models developed to predict pedestrian crashes. Closure A number of recent research projects have proposed apparently similar multivariate equations to estimate pedestrian volume, with encouraging results. Using a GIS-based statistical analysis process, called “Space Syntax,” Raford and Ragland (32) have observed fairly significant correlations between predicted and observed pedestrian volumes in Oakland, CA (R2 = 0.7712) (2003) and Boston, MA (R2 = 0.81) (2005). Other applications of the same basic process have attained R2 values of 0.84 by Hillier (46) and 0.749 by Hillier (47). Using linear regression modeling, Pulugurtha et al. (48) developed pedestrian volume estimates that attained a 2.2 percent difference from observed values. Clifton et al. (33) adapted the origin-destination based demand model to develop estimates of pedestrian counts that were within 4 percent of actual counts. Other pedestrian research projects have included volume estimation models without adequate evaluation or testing of their predictive value (33, 49, 50, 51). The “Space Syntax” approach, discussed above, develops a type of measure called “Integration” that generally describes the level of pedestrian travel possible within a specified number of intersections, called “Radius-3” or “Radius-5.” The equation used by the Space Syntax software is embedded in an add-on program called “Confeego” for the MapInfo GIS software system. Raford and Ragland (32) also include a series of land use variables in their equations.

Next: Chapter 3. Pedestrian Safety Databases »
Pedestrian Safety Prediction Methodology Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB's National Cooperative Highway Research Program (NCHRP) Web-Only Document 129, Phase 3: Pedestrian Safety Prediction Methodology explores development of improved pedestrian safety prediction models for use in the Highway Safety Manual.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!