Embed Size (px)
Citation preview
Zegeer et al. 1
DEVELOPMENT OF CRASH MODIFICATION FACTORS FOR 1
UNCONTROLLED PEDESTRIAN CROSSING TREATMENTS 2
3 Charlie Zegeer1; Phone: 919-962-2202; Email: [email protected] 4 5 Craig Lyon3; Phone: 613-422-2542; Email: [email protected] 6 7 Raghavan Srinivasan1; Phone: 919-962-7418; Email: [email protected] 8 9 Bhagwant Persaud2; Phone: 416-979-5000 X6464; Email: [email protected] 10 11 Bo Lan1; Phone: 919-962-0465; Email: [email protected] 12 13 Sarah Smith1; Phone: 919-962-2202; Email: [email protected] 14 15 Daniel Carter1; Phone: 919-962-8720; Email: [email protected] 16 17 Nathan J. Thirsk1; Phone: 919-962-2202 18 19 John Zegeer4; Phone: 954-828-1730; Email: [email protected] 20 21 Erin Ferguson4; Phone: 510.433.8066; Email: [email protected] 22 23 Ron Van Houten5; Phone: 269 387 4471; Email: [email protected] 24 25 Carl Sundstrom1; Phone: 919-962-2202; Email: [email protected] 26
27 1UNC Highway Safety Research Center (UNC HSRC) 28 730 Martin Luther King Jr, Blvd, CB #3430 29 Chapel Hill, NC 27599 30 31 2Ryerson University, Civil Engineering Department, Toronto, Canada 32 33 3Persaud and Lyon, Inc., Canada 34 35 4Kittelson and Associates, Inc. (KAI) 36 37 5Center for Education and Research in Safety (CERS), Western Michigan University, Kalamazoo, MI 38 39 Number of Words: 5745 40 Number of Tables: 5 41 Number of Figures: 1 42 43 Total words: 5745 + 5 x 250 + 1 x 250 = 7245 words 44 45
Submitted for Presentation at the 2017 Annual Meeting of the Transportation Research Board 46 47
November 9, 2016 48 49
Zegeer et al. 2
ABSTRACT 1 2 The objective of this study was to develop crash modification factors (CMFs) for four treatment types: 3 Rectangular Rapid Flashing Beacons (RRFBs), Pedestrian Hybrid Beacons (PHBs), Pedestrian refuge 4 islands, and Advance Yield or STOP markings and signs. A total of 975 treatment and comparison sites 5 were selected from 14 different cities throughout the U.S. Most of the treatment sites were selected at 6 intersections on urban, multi-lane streets, since these are the situations which have a high risk for 7 pedestrian crashes and where countermeasures are typically most needed. For each treatment site, relevant 8 data were collected regarding the treatment characteristics, traffic, geometric, and roadway variables, and 9 the pedestrian crashes and other crash types that occurred at each site. 10
Cross-sectional regression models and before/after empirical Bayesian analysis techniques were 11 used to determine the crash effects of each treatment type. All four of the treatment types were found to 12 be associated with reductions in pedestrian crash risk, compared to the untreated sites. Pedestrian hybrid 13 beacons were associated with the greatest benefit to pedestrian crash risk (55% reduction), followed by 14 RRFBs (47% reduction), refuge islands (32% reduction), and advance yield/stop lines and signs (25% 15 reduction). The results for RRFBs are based on a very limited sample, and must be used with caution. 16 17 18
Zegeer et al. 3
STUDY BACKGROUND AND OBJECTIVES 1 2 There continues to be a problem in the U.S. related to the safety for pedestrians who attempt to cross 3 streets, particularly on high-speed, high-volume multi-lane roads. Furthermore, there is a need to better 4 understand the safety effects of some of the more promising treatments on pedestrian crashes. Numerous 5 studies have been conducted in the U.S. and abroad in recent years on the effects of various geometric and 6 traffic control treatments at unsignalized crossings. However, most of those evaluations have relied on 7 behavioral and operational measures of effectiveness, (e.g., pedestrian/vehicle conflicts, vehicle speeds, 8 driver yielding behavior) instead of crashes as the measure of effectiveness (1). The lack of crash-based 9 evaluations for such pedestrian treatments largely results from sample size issues; that is, pedestrian 10 crashes usually do not cluster to the same degree at specific sites, so a larger number of treatment and 11 comparison sites are needed to obtain sufficient statistical power to detect a change in crashes due to the 12 treatment, compared to the evaluation of countermeasures for vehicle/vehicle crashes which occur at a 13 higher frequency. 14
Thus, there is a need to conduct evaluations over a wider region and over a longer time period in 15 order to obtain an adequate sample size to allow for the development of Crash Modification Factors 16 (CMFs) or Crash Modification Functions (CMFunctions) that can provide guidance as to the most-17 effective pedestrian crossing treatments to use. The NCHRP 17-56 (14) project sought to develop crash 18 modification factors (CMFs) for selected pedestrian crossing treatments for various traffic and roadway 19 conditions, to the extent possible. Initially, eight treatments were considered, but it was not possible to 20 find a sufficient sample of sites for all these treatments. Also, consideration was given to selecting 21 treatments for evaluation which were considered to be of most particular interest to local and state traffic 22 and safety officials. Based on these considerations, the following four treatments selected for crash-based 23 evaluation in this study: 24
Rectangular rapid flashing beacons (RRFBs) 25 Pedestrian Hybrid Beacons (PHBs) 26 Pedestrian refuge islands (RI) 27 Advance Yield or STOP markings and signs (AS) 28 The next section is a description of the known research literature on these treatments. This is 29
followed by the details of the data, analysis, results, recommendations, and limitations. Further details 30 about the study are available in Zegeer et al. (14). 31 32 LITERATURE REVIEW 33 34 Advanced yield or stop markings are a type of pavement marking placed before a crosswalk to increase 35 the distance at which drivers stop or yield to allow pedestrians to cross. Increasing the distance between 36 yielding vehicles and pedestrians increases the ability of motorists in other lanes to see the pedestrian as 37 the pedestrian crosses and to yield accordingly. Pedestrian visibility of oncoming traffic is, improved. Ten 38 previous studies indicated that advance yield or stop markings and signs reduced pedestrian-vehicle 39 conflicts and increased motorist yielding (1). 40
Pedestrian refuge islands, sometimes referred to as center islands, refuge islands, or pedestrian 41 islands, are raised areas that help protect pedestrians who are crossing the road at intersections and mid-42 block locations. The presence of a median refuge island in the middle of a street or intersection allows 43 pedestrians to focus on one direction of traffic at a time as they cross and gives them a place to wait for an 44 adequate gap between vehicles. Islands are appropriate for use at both uncontrolled and signalized 45 crosswalk locations. Where the road is wide enough and on-street parking exists, center islands can be 46 combined with curb extensions to further enhance pedestrian safety (7). 47
Three studies reviewed reported crash reduction factors for constructing raised medians or 48 pedestrian islands. One study found that replacing a 6 foot painted median with a wide raised median 49 reduced pedestrian crashes by 23 percent (9) which was consistent with findings from another study that 50
Zegeer et al. 4
pedestrian crash rates for roads with 10 foot medians were 33 percent lower than for roads with 4 foot 1 painted medians (8). Finally, a before-after study was conducted to evaluate the safety effectiveness of 2 raised pedestrian refuge islands. Researchers found that there was a 73 percent reduction in mid-block 3 pedestrian crashes but a 136 percent increase in total crashes. It was noted that the decrease in safety 4 related to vehicle-island crashes might be helped by better island design and lane alignment (10). 5
Pedestrian hybrid beacons (PHBs), also known as HAWK beacons (short for High-Intensity 6 Activated crosswalk beacon), were developed by a Tucson traffic engineer, Dr. Richard Nassi, in the late 7 1990s as a means for providing safe pedestrian crossings where minor streets intersected with major 8 arterials (3,4). A PHB is a special type of beacon used to warn and control traffic at an unsignalized 9 location to assist pedestrians crossing a street or highway at a marked crosswalk. The PHB signal head 10 consists of two red lenses over a single yellow lens. It displays a red indication to drivers when activated, 11 which creates a gap for pedestrians to use to cross a major roadway. The PHB signal indication for 12 motorists is not illuminated until it is activated by a pedestrian, triggering the warning flashing yellow 13 lens on the major street. After a set amount of time, the traffic signal indication changes to a solid yellow 14 light to inform drivers to prepare to stop. The beacon then displays a dual solid red light to drivers on the 15 major street and a walking person symbol to pedestrians. At the conclusion of the walk phase, the beacon 16 displays an alternating flashing red light to drivers, and pedestrians are shown an upraised hand symbol 17 with a countdown display informing them the time left to cross. The first PHB was installed in Tucson in 18 2000. The PHB was considered an experimental treatment until 2009, when it was included in the 19 MUTCD. PHBs are now widely used in Tucson and have since been installed in Georgia, Minnesota, 20 Florida, Michigan, Virginia, Arizona, Alaska, and Delaware (5). 21
One study reviewed reported crash reduction factors for installing PHBs in Tucson, AZ. Results 22 of the analysis showed a statistically significant reduction in total crashes of 29 percent as well as a 23 statistically significant reduction in pedestrian crashes of 69 percent. There was a 15 percent reduction in 24 severe crashes; however, this result was not statistically significant (3). 25
A rectangular rapid flashing beacon (RRFB) is a type of amber LED installed to enhance 26 pedestrian crossing signs at midblock crossings or unsignalized intersections. RRFBs can be automated or 27 pedestrian-actuated, and feature an irregular, eye-catching flash pattern to call attention to the presence of 28 pedestrians. The RRFB was given interim approval as a crossing sign enhancement by FHWA in 2008 29 (6). Nine studies were reviewed. All of them indicated that RRFBs increased motorist yielding helping 30 thereby reducing pedestrian-vehicle conflicts (1). Figure 1 shows photographs for the four treatments 31 evaluated in this study. 32
33 DATA COLLECTION 34 35 The research team used multiple methods for identifying cities and states that had installed pedestrian 36 crossing treatments at unsignalized intersections. The team conducted exploratory field work to identify 37 specific treatment and comparison sites in multiple cities across the U.S. Based on detailed information 38 obtained from each city, in terms of available treatments of interest for this study, U.S. geographic 39 distribution of cities, and other factors, the 14 cities selected for data collection for this study were: 40 Alexandria and Arlington, VA, Cambridge, MA, Chicago, IL, New York, NY, Miami and St. Petersburg, 41 FL, Tucson, Scottsdale, and Phoenix AZ, Portland and Eugene, OR, Charlotte, NC and Milwaukee, WI. 42
A total of 499 treatment sites and 476 comparison sites were included in the final database. 43 Comparison sites were selected in such a way to be as similar as possible to the treatment sites, except 44 that no treatment was present. Efforts were made to select comparison sites which were similar to the 45 treatment sites in terms of number of lanes, traffic volume, area type, nearby location, and other 46 characteristics (e.g., at a school, bus stop, shopping area). TABLE 1 is a summary of the number of 47 treatment and comparison sites that were found and coded into the database. As shown in TABLE 1, 48 treatment sites often had combinations of two or more of the studied treatments. 49
50
Zegeer et al. 5
1
FIGURE 1 Example of advance stop line and sign (A), raised median island (B), pedestrian hybrid 2 beacon (C), and rectangular rapid flashing beacon (D); Sources: PBIC, FHWA, www.wichita.org 3 4
Most of the sample sites were on roads with four or more lanes. Specifically, of the 975 total 5 sites, 136 (27%) treatment and 87 (18%) comparison sites were on two-lane roads. The project team 6 decided to focus more on sites with four or more lanes and pedestrian crossings on arterial streets and at 7 transit stops, since these are the types of sites where pedestrian crashes are more likely to occur 8 (compared to local, two lane streets) and where pedestrian safety treatments are more likely needed. 9
The most common crosswalks in the database were those that had no marked crosswalk (e.g., at 10 untreated comparison sites), continental-style crosswalks, or ladder crosswalks. The transverse lines 11 crosswalks were found at about 100 crossing sites. Multiple crosswalk types (80 sites) are those that have 12 two or more descriptive codes (combined); for example, a yellow crosswalk with transverse line 13 markings, or a staggered crosswalk with continental markings, or a diagonal crosswalk with ladder 14 markings. Most of the treatment (75%) and comparison (76%) sites were at intersections. Furthermore a 15 great majority of treatment and comparison sites were in suburban areas. In terms of transit use, 16 approximately half the comparison sites were at transit stops and 41% of treatment sites are at transit 17 stops. 18
Data on pedestrian crashes and other crashes for each treatment and comparison study site were 19 provided electronically by the agencies as were AADTs. Very few agencies were able to provide 20 pedestrian counts so pedestrian volume was collected in the field for 915 of the 975 sites. In the field, 21 pedestrian counts were collected for 1 to 2 hours. The team adjusted these short counts to daily volumes 22 (i.e., 24 hour volumes) based on an analysis of citywide full-day pedestrian counts from Charlotte and 23 previous research in Seattle. 24
25 26
Zegeer et al. 6
1 DATA ANALYSIS AND RESULTS 2 3 The aim of the analytical work was to use the most appropriate techniques for estimating crash 4 modification factors or functions for pedestrian involved and all crashes, by type, for various treatments 5 and treatment combinations. 6
It was quickly realized that the preferred before after study would be limited in scope and 7 robustness because of the relatively small sample sizes, especially for pedestrian crashes. Thus, 8 considerable effort was made in assembling a database for the alternative methodology, cross-sectional 9 regression analysis, recognizing the well-known limitations of this approach. In view of those limitations, 10 it was necessary to still conduct a before-after study, however limited, to complement the cross-sectional 11 analysis by seeking to corroborate the results. 12
While the primary target crash of the four pedestrian treatments is the small subset of vehicle-13 pedestrian crashes, vehicle-vehicle crashes were added to the analysis to investigate if they were affected 14 as well. If the treatments alter driver behavior this is a logical possibility. Rear-end and sideswipe crashes 15 were combined as it was hypothesized that these types of vehicle-vehicle crashes may be the crash types 16 most likely to be affected by the pedestrian treatments and would be similar in this regard. The combined 17 category was considered the target vehicle-vehicle crash category. 18 19 CMF Estimation with the Before-After Study Methodology 20 The empirical Bayes (EB) methodology for observational before-after studies was used to develop crash 21 modification factors (CMFs). In the EB approach, the CMF is estimated based on a comparison the 22 expected number of crashes that would have occurred in the after period without the treatment, and the 23 number of reported crashes in the after period. The first value is estimated in essence as a weighted 24 average of the crashes occurring before treatment and the prediction from a safety performance function 25 developed from reference group of similar but untreated sites that is adjusted for changes in traffic 26 volume and time trends between the periods before and after treatment. This methodology is rigorous in 27 that it properly accounts for regression-to-the-mean (RTM) and these other changes and has been 28 included in the 1st edition of the Highway Safety Manual (15) as the state-of-the-art for conducting 29 observational before-after studies. It is well documented in Hauer (13). Both sources can be consulted for 30 details on the methodology, which are not presented here in the interests of brevity. 31
The evaluation was conducted for 3 treatments with sufficient sites: refuge island (RI) (68 sites); 32 advanced stop (AS) (69 sites) and pedestrian hybrid beacon in combination with advanced stop (PHB & 33 AS) (27 sites). Four crash types were investigated in the before-after evaluation where possible: total, 34 rear-end, sideswipe, and pedestrian. TABLE 2 shows the estimated CMFs for RI, AS, and PHB & AS 35 combination. For each treatment and crash type, the table shows the actual crashes that occurred during 36 the after period, the expected crashes had the treatment not been implemented (estimated based on the EB 37 procedure), variance of the expected crashes, the CMF, the standard error of the CMF, and the p-value. 38
For the Refuge Island treatment, the EB estimate of crashes in the before period of the 39 treatment group is higher than the actual crashes, particularly so for vehicle-involved crashes, leading to 40 an inexplicable increase in total crashes. This was unexpected, and hence, the results for the RI treatment 41 for the vehicle-involved crash types were not considered reliable enough for inclusion. The results for 42 pedestrian crashes indicate a reduction of approximately 33%, which was significant at about the 13% 43 significance level.44
Zegeer et al. 7
TABLE 1 Treatment and Comparison Site Totals by City 1
City AS RI RRFB PHB AS+
RI
AS+
RRFB
AS+
PHB
RI+
RRFB
AS+
RI+
RRFB
AS+
RI+
PHB
Total
Treatment
Sites
Total
Comparison
Sites
Arlington & Alexandria, VA
1 21 0 0 1 0 0 0 0 0 23 22
Cambridge, MA 2 9 0 0 8 0 0 0 0 0 19 25
Charlotte, NC 0 34 0 0 0 0 2 0 0 0 36 112
Chicago, IL 0 33 0 0 0 3 0 0 0 0 36 37
Eugene, OR 0 22 1 0 1 0 0 3 2 0 29 27
Miami, FL 0 24 3 0 1 0 0 0 2 0 30 36
Milwaukee, WI 0 11 0 0 0 0 0 1 0 0 12 18
New York, NY 0 17 0 0 0 0 0 0 0 0 17 24
Phoenix, AZ 4 2 0 0 6 0 3 0 1 2 18 16
Portland, OR 18 6 1 2 33 0 0 0 1 0 61 33
Scottsdale, AZ 0 14 0 0 2 0 1 0 0 1 18 16
St. Petersburg, FL
72 1 0 0 7 22 3 0 10 0 115 45
Tucson, AZ 0 2 0 0 1 0 48 0 0 34 85 65
TOTAL 97 196 5 2 60 25 57 4 16 37 499 476
2 3 4
Zegeer et al. 8
For the Advanced Stop treatment, the results indicate, approximately, an 11 percent reduction 1 in total crashes, a 20 percent reduction in rear end and sideswipe crashes, and a 36 percent reduction in 2 pedestrian crashes. The reduction in total crashes is statistically significant at about the 8 percent level, 3 and the reduction in the other two crash types is statistically significant at the 5 percent level or lower. 4
For the combination Pedestrian Hybrid Beacon and Advance Stop treatment, the results 5 indicate, approximately, an 18 percent reduction in total crashes and a 76 percent reduction in pedestrian 6 crashes. Both these results are statistically significant at the 5 percent level or lower. The changes in rear 7 end and sideswipe crashes are not statistically significant at any reasonable significance level. The result 8 for pedestrian crashes is quite consistent with the 69 percent reduction reported by Fitzpatrick and Park 9 (3) for the same treatment at intersections in Tucson. Fitzpatrick and Park (3) also used the EB before-10 after evaluation method. 11 12 13
TABLE 2 Crash Modification Factors from the Before-After Evaluation 14
Treatment
and Number
of Sites
Crash
type
Crashes
after
EB estimates
of crashes
expected after
Variance
(EB after)
CMF* S.E. of
CMF
p-
value
RI: 68 sites Pedestrian 13 18.8 11.2 0.671 0.215 0.126
AS: 69 sites Total 671 754.7 2254.5 0.886 0.065 0.079
Rear end & Sideswipe 335 416.2 1068.8 0.800 0.076 0.008
Pedestrian 21 32.2 27.4 0.636 0.169 0.031
PHB & AS:
27 sites
Total 341 413.2 1078.5 0.820 0.078 0.021
Rear end & Sideswipe 182 205.4 460.9 0.876 0.111 0.264
Pedestrian 4 15.6 13.3 0.244 0.128 0.000
* Statistically significant results at the 10% level are indicated in bold 15 16 CMF Estimation from Cross-Sectional Regression Analysis 17 The purpose of the cross-sectional regression analysis, as it was for the before-after study, was to estimate 18 Crash Modification Factors (CMFs) or Crash Modification Functions (CMFunctions) for each of the four 19 pedestrian treatments. It was also of interest to estimate CMFs for combinations of treatments. In fact, at 20 many of the sites one treatment was installed followed by one or more additional treatments installed in 21 later years. For most combination treatments however the sample size was too limited to provide robust 22 results. The exceptions were PHB and RRFBs where most sites having these treatments also have 23 Advance Yield or Stop signs present. Several crash types were analyzed separately for each treatment, 24 including: pedestrian crashes, rear-end plus sideswipe crashes, rear-end plus sideswipe crashes, (KABC) 25 injury crashes, and total crashes. 26
The cross-sectional analysis applied Generalized Linear Modeling (GLM). Because each site-year 27 of data was used as an observation, the issue of repeated measures of the data needed to be resolved in 28 specifying the models. The approach taken within the GLM framework to account for the repeated 29 measures and nested nature of the data was Generalized Estimation Equations (GEE). This ensures that 30 the correlation between within-site observations is accounted for as well as allowing for unobserved 31 heterogeneity between cities. 32
The dataset for developing CMFs for refuge islands did not include any site that had an advance 33 stop/yield sign, PHB, or RRFB installed at any time during the study period. For this and all treatments, 34 the reference group sites do not have any of the four main treatments installed. Likewise, the dataset used 35 for developing CMFs for advance stop/yield signs did not include any site that had a refuge island, PHB, 36 or RRFB installed at any time during the study period. The data for developing the CMFs for refuge 37
Zegeer et al. 9
islands and advance stop/yield signs came from Alexandria, Arlington, Cambridge, Charlotte, Chicago, 1 Eugene, Miami, Milwaukee, New York, Phoenix, Portland, Scottsdale, St. Petersburg, and Tucson. 2
The dataset for developing CMFs for pedestrian hybrid beacons did not include any site that had a 3 refuge island or RRFB installed at any time during the study period. Because many sites with PHB also 4 have an advance stop/yield sign, sites with these signs were included in the dataset. The sites were also 5 limited to the following cities in which a PHB existed: Charlotte, Portland, Phoenix, Scottsdale, Tucson 6 and St. Petersburg. 7
For pedestrian crash models, the dataset for developing CMFs for RRFB included all locations. 8 This was needed to allow for a sufficient sample of pedestrian crashes. For the other crash types the 9 dataset for RRFBs did not include any site that had a refuge island or PHB installed at any time during the 10 study period. Because many sites with RRFB also have an advance stop/yield sign sites with these signs 11 were included in the dataset. For the non-pedestrian crash models the sites were also limited to cities in 12 which a RRFB existed: Chicago, Eugene, Miami, Phoenix, Portland and St. Petersburg. 13 14 Cross-Sectional Model Results 15 This section presents the CMFs implied by the relevant parameter estimates of the cross-section models. 16 For brevity, only one set of models – for refuge islands – is presented by way of an example. The 17 following considerations were applied in making decisions on variables to include in the models. 18
The typical multiplicative crash prediction model form was eventually applied with 19 pedestrian and total AADT as separate power model terms and the categorical variables for 20 city in which a crossing is located, presence of a treatment, area type (urban or suburban) and 21 crossing location (midblock or intersection) as exponential terms. 22
Interaction terms were attempted, for example, the effect of presence of treatment and 23 crossing location, but these did not improve the model’s ability to explain the variation in 24 crashes between sites. 25
Multi-level models were attempted to develop CMFunctions by investigating if the implied 26 CMFs for presence of treatment varied by vehicle and pedestrian AADT and other variables 27 but these too did not improve the results. 28
A “City” factor variable was included in the models to account for differences in expected 29 crashes between jurisdictions that are not related to the treatment, which could include crash 30 reporting practices, weather, cultural issues, etc. Inclusion of this variables estimates a unique 31 intercept term for each city in a model. The same logic applies to the inclusion of the area 32 type variable. Models with and without the city variable and the area type were attempted 33 before deciding to include these variables in the final models. (For the most part these 34 variables were included.) 35
36 Model Results Example – Refuge Islands 37
The following model forms pertain to each crash type for the refuge island models to predict the 38 expected number of crashes per year. 39
40
𝑃𝐸𝐷𝐶𝑅𝐴𝑆𝐻 = 𝑒𝑥𝑝(𝑎+𝐶𝑖𝑡𝑦+𝑏∗𝑅𝑒𝑓𝑢𝑔𝑒 𝐼𝑠𝑙𝑎𝑛𝑑 𝑃𝑟𝑒𝑠𝑒𝑛𝑐𝑒+𝑐∗𝐴𝑟𝑒𝑎𝑇𝑦𝑝𝑒)𝐴𝐴𝐷𝑇𝑒𝑃𝐸𝐷𝐴𝐴𝐷𝑇𝑓 1) 41
𝑇𝑂𝑇𝐶𝑅𝐴𝑆𝐻 = 𝑒𝑥𝑝(𝑎+𝐶𝑖𝑡𝑦+𝑏∗𝑅𝑒𝑓𝑢𝑔𝑒 𝐼𝑠𝑙𝑎𝑛𝑑 𝑃𝑟𝑒𝑠𝑒𝑛𝑐𝑒+𝑑∗𝑀𝑖𝑑𝑏𝑙𝑜𝑐𝑘_𝑖𝑛𝑡)𝐴𝐴𝐷𝑇𝑒𝑃𝐸𝐷𝐴𝐴𝐷𝑇𝑓 2) 42
𝐼𝑁𝐽𝐶𝑅𝐴𝑆𝐻 = 𝑒𝑥𝑝(𝑎+𝐶𝑖𝑡𝑦+𝑏∗𝑅𝑒𝑓𝑢𝑔𝑒 𝐼𝑠𝑙𝑎𝑛𝑑 𝑃𝑟𝑒𝑠𝑒𝑛𝑐𝑒+𝑑∗𝑀𝑖𝑑𝑏𝑙𝑜𝑐𝑘_𝑖𝑛𝑡)𝐴𝐴𝐷𝑇𝑒𝑃𝐸𝐷𝐴𝐴𝐷𝑇𝑓 3) 43
𝑇𝐴𝑅𝐺𝐸𝑇𝐶𝑅𝐴𝑆𝐻 = 𝑒𝑥𝑝(𝑎+𝐶𝑖𝑡𝑦+𝑏∗𝑅𝑒𝑓𝑢𝑔𝑒 𝐼𝑠𝑙𝑎𝑛𝑑 𝑃𝑟𝑒𝑠𝑒𝑛𝑐𝑒+𝑑∗𝑀𝑖𝑑𝑏𝑙𝑜𝑐𝑘_𝑖𝑛𝑡)𝐴𝐴𝐷𝑇𝑒𝑃𝐸𝐷𝐴𝐴𝐷𝑇𝑓 4) 44
𝐼𝑁𝐽𝑇𝐴𝑅𝐺𝐸𝑇𝐶𝑅𝐴𝑆𝐻 = 𝑒𝑥𝑝(𝑎+𝐶𝑖𝑡𝑦+𝑏∗𝑅𝑒𝑓𝑢𝑔𝑒 𝐼𝑠𝑙𝑎𝑛𝑑 𝑃𝑟𝑒𝑠𝑒𝑛𝑐𝑒+𝑑∗𝑀𝑖𝑑𝑏𝑙𝑜𝑐𝑘_𝑖𝑛𝑡)𝐴𝐴𝐷𝑇𝑒𝑃𝐸𝐷𝐴𝐴𝐷𝑇𝑓 5) 45
Target crashes include rear-end and sideswipe crashes. 46
Zegeer et al. 10
1 where, 2 AADT = total AADT on roadway being crossed 3 AreaType = 1 if Suburban; 0 if Urban 4 City = represents an intercept term specific for each city 5 Midblock_Int = 1 if intersection; 0 if midblock 6 PEDAADT = total pedestrian AADT for midblock or intersection 7 Refuge Island Presence = 1 if present; 0 if not present 8 9 Parameter estimates for all models are shown in TABLE 3. As noted above, the estimates for the 10
City variable are not provided in the interest of brevity, and since in any case they do not impact the 11 estimated CMFs. 12
13 14
TABLE 3 Parameter Estimates for Refuge Island Regression Models 15
Parameter Estimate (standard error)
Parameter Pedestrian Total Injury RE+SS Injury RE+SS
a -10.4246
(1.6409)
-5.5953
(0.7761)
-6.4572
(0.8886)
-8.3157
(0.9754)
-9.7130
(1.2009)
b -0.3578
(0.2153)
-0.2981
(0.0956)
-0.3369
(0.1148)
-0.2999
(0.1258)
-0.3254
(0.1460)
c -0.5715
(0.3127)
n/a n/a n/a n/a
d n/a 0.4730
(0.1083)
0.4312
(0.1183)
0.4140
(0.1224)
0.3728
(0.1395)
e 0.6977
(0.1694)
0.5192
(0.0756)
0.5375
(0.0846)
0.7235
(0.0947)
0.7780
(0.1157)
f 0.3295
(0.0486)
0.1224
(0.0247)
0.1141
(0.0260)
0.1041
(0.0237)
0.0986
(0.0263)
Overdispersion n/a 0.7608
(0.0305)
0.7075
(0.0482)
0.9149
(0.0521)
0.9837
(0.1075)
16 With the exception of the parameter estimates for refuge island presence and area type in the 17
pedestrian crash model, all parameter estimates are statistically significant at the 5 percent level (i.e., a 18 value of no effect is not within 1.96 standard errors). Those two estimates are, however, statistically 19 significant at the 10th percent level (a value of zero is not within 1.64 standard errors). 20
The estimated parameters indicate that fewer crashes of all types are expected when a refuge 21 island is present, and for pedestrian crashes, in suburban areas versus urban areas. Greater numbers of 22 crashes of all types are expected at higher levels of vehicle and pedestrian AADT. With the exception of 23 pedestrian crashes, the models also indicate that more crashes are expected at intersections than at 24 midblock crossings. 25 26 CMFs Implied from Cross-Sectional Models 27 TABLE 4 summarizes the results from the cross-sectional analyses for treatments and crash type cases 28 where the analyses support a CMF recommendation. The CMFs implied by the parameter estimates for a 29 treatment is estimated as (eb), where b is the estimated coefficient for the presence of treatment. Standard 30 errors are also presented for assessing statistically significance. Effects that are statistically significant at 31 the 10% level are indicated in boldface. Also presented for comparison are the before-after study results 32
Zegeer et al. 11
presented earlier. As is evident, a comparison of results from the two studies is only possible for 1 pedestrian crashes, the main target crash type. 2 3 TABLE 4 CMFs Implied from Cross-Sectional Analyses - Compared to Before-After Study CMFs 4
Treatment Crash Type Cross-section study Before-after study
Estimate Standard error Estimate Standard error
Refuge Islands
Pedestrian 0.699 0.152 0.671 0.215
Total 0.742 0.071
All Injury 0.714 0.082
Rear End/Side Swipe Total 0.741 0.093
Rear End/Side Swipe Injury 0.722 0.106
Advance Stop (AS)
Pedestrian 0.863 0.290 0.636 0.169
Total 0.886 0.065
Rear End/Side Swipe Total 0.800 0.076
PHB Pedestrian 0.526 0.206 0.38* n/a
PHB+AS
Pedestrian 0.62** 0.140 0.244 0.128
Total 0.820 0.078
Rear End/Side Swipe Total 0.876 0.111
RRFB Pedestrian 0.526 0.377
*As mentioned earlier, the sample size for the PHB alone treatment was very limited for conducting a before after 5 evaluation. However, the before-after evaluation estimated CMFs for AS and PHB+AS combinations. Using these 6 results and assuming that CMFs are multiplicative for treatment combinations, the CMF for PHB alone was 7 calculated as the CMF for PHB + AS divided by the CMF for AS, i.e., 0.244/0.636 = 0.38. 8 **Estimated for comparison purposes 9 10
For refuge islands, the cross-sectional CMFs are all statistically significant (p<0.10) and 11 consistently of the order of 0.7 (crash reductions of 30%). Note that the refuge island may be provided by 12 both a continuous median or a smaller island provided at the crossing. Additional models were attempted 13 allowing the safety effects to differ by continuous median versus a short refuge island but the results were 14 inconclusive. 15
For the presence of advance stop or yield signs, the estimated cross-sectional model parameters 16 indicated that fewer pedestrian, total and injury crashes are expected when a sign is present and more 17 crashes are expected for target and target injury crashes. For total crashes the estimate is extremely close 18 to 0, which would indicate no effect. For the non-pedestrian crash types, the sample sizes are sufficiently 19 large that the large standard errors of these estimates as well as the inconsistent direction of effect can be 20 used to question their reliability in estimating a CMF. For this reason it was concluded that the cross-21 sectional models do not support a CMF for vehicle-vehicle crashes. For pedestrian crashes, which have 22 much smaller sample sizes, regardless of the large standard error, the results appear logical in direction of 23 effect and magnitude. 24
For the presence of PHB, the p-value for pedestrian crash CMF estimated from the cross-25 sectional model is on the high side but, as for advance stop/yield signs, because the direction of the effect 26 is intuitive (CMF <1), the point estimate of the CMF for pedestrian crashes is still recommended for 27 application. No CMFs are recommended for the other crash types due to the statistical significance and/or 28 illogical effects implied. 29
For RRFBs, as was the case for advance stop/yield signs, because the direction of the effect is 30 intuitive (CMF <1), the point estimate of the CMF for pedestrian crashes from the cross-sectional is still 31 recommended for application since the high p-value is likely mainly a result of the scarcity of these 32
Zegeer et al. 12
crashes. No CMFs are recommended for the other crash types due to the statistical significance and/or 1 illogical effects implied. 2
In general, the before-after results, where they could be obtained, corroborate the direction of 3 effect for pedestrian crashes from the cross-sectional analysis and in the case of refuge islands, the order 4 of magnitude. For the other treatments, AS, PHB and PHB+AS, the CMF point estimates are larger (the 5 benefits smaller) for the cross-sectional study, which might be expected on theoretical grounds as a result 6 of the issues with cross-sectional studies. However, given the large standard errors, in particular for the 7 CMFs for the cross-section analysis, it cannot be concluded that the estimates are different statistically. 8 9 Recommended CMFs 10 As presented in TABLE 4, credible CMFs for pedestrian crashes and vehicle-involved crashes were 11 estimated by either cross-sectional or before-after analysis, or both, for refuge islands, advanced 12 stop/yield (AS), PHB, and PHB+AS. The CMF for RRFB was based on a very limited sample, and hence 13 should be used with caution. 14
The two methods have strengths and limitations. Although the before-after analysis is preferred, 15 especially the empirical Bayes approach used for this research, the limited sample size of sites and 16 crashes limits the robustness of the results. By contrast, the well-known limitations of deriving CMFs 17 from cross-sectional regression analysis suggest that results obtained from that analysis, albeit from a 18 substantial sample, should be interpreted and used with caution. It is encouraging, however, that the 19 results of the limited before-after study corroborate those from the cross-section study for pedestrian 20 crashes, the main crash type of interest. 21
On balance, it seems difficult to choose one approach over another where two sets of results are 22 obtained, as was the case for pedestrian crashes, which logically suggests that equal weighting of the 23 CMFs is not unreasonable for recommending CMFs for practical application. With this in mind, where 24 there are two CMFs in TABLE 4, the median point values are presented in TABLE 5 as a final 25 recommendation for practical application. Where there is a CMF value from only one study, that value is 26 recommended. In each case, there is an appropriate annotation in indicating the basis of the 27 recommendation. For practical applications where a conservative benefit estimate is desired for a 28 contemplated treatment, the higher of two CMF values from TABLE 4 could be considered. As noted 29 earlier, results from the before-after and cross-sectional analyses are only reported for treatments and 30 crash type cases where the analyses support a CMF recommendation. 31
As mentioned earlier, CMFunctions were explored in order to determine the effectiveness of 32 these treatments corresponding to different levels of AADT, posted speed limit, area type, number of 33 lanes, and other factors. However, these CMFunctions did not provide useful results. Future research 34 could investigate whether these treatments are more or less effective under different conditions. 35
36 LIMITATIONS AND RECOMMENDATIONS 37 38 Although several CMFs were estimated, it is important to note that expected safety effects of applying the 39 treatments may vary across specific applications. A great majority of treatment sites included in this study 40 were intentionally selected at urban (and suburban) areas on multi-lane roads. While most of the 41 treatments were at intersections, there was also a substantial number of midblock locations represented in 42 the database. For example, it is likely that factors such as operating speed, AADT, roadway width, lane 43 width, and other factors can be expected to influence the efficacy of any of these interventions. The 44 limited data available did not allow for such a disaggregation of expected effects. Future work is needed 45 to make more precise predictions about how much of a crash reduction should be expected at locations 46 with specific characteristics. 47 48 49
Zegeer et al. 13
TABLE 5 Recommended CMFs 1
Treatment Crash Type Recommended CMF Study Basis
Estimate Standard error
Refuge Islands
Pedestrian 0.685 0.183 Median from two studies
Total 0.742 0.071 Cross-section
All Injury 0.714 0.082 Cross-section
Rear End/Side Swipe Total 0.741 0.093 Cross-section
Rear End/Side Swipe Injury 0.722 0.106 Cross-section
Advance Stop (AS)
Pedestrian 0.750 0.230 Median from two studies
Total 0.886 0.065 Before-after
Rear End/Side Swipe Total 0.800 0.076 Before-after
PHB Pedestrian 0.453 0.167 Median from two studies
PHB+AS
Pedestrian 0.432 0.134 Median from two studies
Total 0.820 0.078 Before-after
Rear End/Side Swipe Total 0.876 0.111 Before-after
RRFB Pedestrian 0.526 0.377 Cross-section
2 3
It is also likely that these treatments may have different effects on number of crashes and the 4 severity of crashes. For example, interventions that slow drivers and reduce the probability of a crash but 5 may also have an effect on the severity of the crash when one occurs. Hence it may be possible to have a 6 CMF for total crashes and a different CMF for incapacitating and fatal crashes. Another factor that could 7 not be fully assessed is how the use of several of these treatments used together would influence crashes. 8 It should also be remembered that the CMF for the RRFB was based on a very limited sample (i.e., 50 9 treatment sites), and hence should be used with caution. 10
One of the data limitations for this study, and other pedestrian and bicycle-related studies, is the 11 lack of available exposure data related to walking (and bicycling). As a result of the lack of such data 12 from most of the 14 agencies selected for this study, it was necessary to conduct short-term (i.e., 1 or 2 13 hour) counts, and extrapolate to obtain an estimate of the average annual daily pedestrian traffic (i.e., 14 pedestrian AADT’s). This required a substantial effort to develop pedestrian AADT’s, based on the actual 15 hourly count, the time of day of the count (e.g., 4:00 to 5:00 p.m. was the usual count time), and the type 16 of area within the city. It was recognized that while estimating pedestrian daily volume was necessary in 17 the analysis, the accuracy of such estimates (i.e., extrapolating short-term counts) could have been greatly 18 improved if a larger sample of data (e.g., 8-or-10-hour counts) was available from the city transportation 19 agencies. Unfortunately, we found that only two of the cities (Charlotte and St. Petersburg) had existing 20 pedestrian counts at any of our selected sites. Certainly, if more city and State DOT’s were routinely 21 collect pedestrian (and bicycle) counts, such count data would have value not just for research studies, but 22 also to help determine the need and justification for all types pedestrian improvements on a routine basis. 23
It is also important to remember that most of the four treatment types evaluated in this study did 24 not have samples from all (or even most) of the 14 test cities. For example, approximately 90 percent of 25 the pedestrian hybrid beacons (PHBs) used in this study were from Tucson, Arizona, and that city was 26 found to have installed most of this treatment type that was found to exist nationwide, when study cities 27 were being selected. Likewise, approximately 75 percent of the rectangular rapid flashing beacon (RRFB) 28 sites in this study were from St. Petersburg, Florida due the frequent use of these devices in that city. 29 Therefore, readers should not necessarily assume that the crash modification factors (CMFs) found in this 30 study for a given treatment will necessarily be the same in all cities. Individual cities and state 31 jurisdictions have differences in driver and pedestrian behaviors, terrain, laws, weather patterns, and 32
Zegeer et al. 14
many other factors that can affect the way that any countermeasure will perform. Therefore, it is 1 recommended that agencies select among these or any other countermeasure with caution, and to try to 2 identify and address the specific safety problem at a location and then the countermeasure (s) considered 3 most likely to address the specific problem(s). After implementation of any countermeasures, it is also 4 advisable that monitoring of crashes and behaviors be a routine practice at treatment sites, to insure that 5 the countermeasure is operating effectively. 6 7 ACKNOWLEDGEMENTS 8 9 This work was sponsored by the National Cooperative Highway Research Program through Project 17-10 56. The opinions and conclusions expressed or implied herein are those of the Contractor. They are not 11 necessarily those of the Transportation Research Boards, the Academies, or the program sponsors. The 12 study authors greatly appreciate the input and guidance provided by the project panel throughout the 13 project period and to the NCHRP Technical Manager, Lori Sundstrom for her project oversight and 14 advice. We also appreciate the assistance from all of the city officials throughout the U.S. who were 15 contacted regarding information on the status of pedestrian treatments which have been implemented, and 16 particularly for the support and cooperation of officials from the 14 cities where treatments were selected. 17 Also, officials from the respective state DOTs were very helpful in providing crash data for the sites 18 within their cities. This study could not have been conducted without assistance from those city and state 19 DOT officials. 20 21
Zegeer et al. 15
REFERENCES 1 2 1. Mead, J., C. Zegeer, and M. Bushell. Evaluation of Pedestrian-Related Roadway Measures: A 3
Summary of Available Research. Chapel Hill, NC: Federal Highway Administration, 2013 4
(http://www.pedbikeinfo.org/data/library/details.cfm?id=4872). 5
2. City of Madison Traffic Engineering Division. Year 2 Field Evaluation of Experimental ‘In-Street’ 6
Yield to Pedestrian Signs. City of Madison Department of Transportation, Madison, 7
Wisconsin, 1999. 8
3. Fitzpatrick, K. and E. S. Park. Safety Effectiveness of the HAWK Pedestrian Crossing Treatment. 9
Publication FHWA-HRT-10-042. FHWA, U.S. Department of Transportation, 2010. 10
4. Nassi, R.B., and M. J. Barton. New Traffic Control for an Old Pedestrian Crossing Safety Problem. 11
APWA Reporter, Vol. 75, No. 6, 2008, pp. 44-49. 12
5. Chalmers, M. New Traffic Signals Make It Safer for Pedestrians. USA Today, 8/9/2010 13
6. Federal Highway Administration. Rectangular Rapid Flash Beacon (RRFB). Publication FHWA-14
SA-09-009, FHWA, U.S. Department of Transportation, 2009. 15
7. Zegeer, C. V., J. Stutts, H. Huang, M. J. Cynecki, R. Van Houten, B. Alberson, R. Pfefer, T. R. 16
Neuman, K. L. Slack, and K. K. Hardy. NCHRP Report 500 Volume 10: A Guide for 17
Reducing Collisions Involving Pedestrians. Washington, D.C., Transportation Research 18
Board, National Research Council, 2004. 19
8. Cairney, P. Pedestrian Safety in Australia. Publication FHWA-RD-99-093, FHWA, U.S. 20
Department of Transportation, 1999. 21
9. Claessen, J. and D. Jones. The road safety effectiveness of wide raised medians. Proc. 17th ARRB 22
Conference, 17(5), 1994, pp. 269-284. 23
10. Bacquie, R., D. Egan, and L. Ing, L. Pedestrian Refuge Island Safety Audit. Presented at 2001 ITE 24
Spring Conference and Exhibit, Monterey, CA, 2001. 25
11. Manual on Uniform Traffic Control Devices for Streets and Highways (MUTCD), USDOT Federal 26
Highway Administration, 2009. 27
12. Gross, F., B. Persaud, and C. Lyon. Guide to Developing Quality Crash Modification Factors. 28
Report FHWA-SA-10-032, 2010. 29
13. Hauer, E., Observational Before-After Studies in Road Safety, Pergamon Press, 1997. 30
14. Zegeer, C., C. Lyon, R. Srinivasan, B. Persaud, B. Lan, S. Smith, D. Carter, N. Thirsk, J. Zegeer, E. 31
Ferguson, R. Van Houten, and C. Sundstrom. Development of Crash Modification Factors for 32
Uncontrolled Pedestrian Crossing Treatments, Final Report from NCHRP Project 17-56, 33
Washington, D.C., Transportation Research Board, 2016. (in press). 34
15. Highway Safety Manual, 1st edition, AASHTO, Washington, DC, 2010. 35
