Safety at Midblock Pedestrian Signals (2023)

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   Literature Review A literature search was conducted using several search techniques and resources, among which the Transportation Research Information Services (TRIS) databases. TRIS can search several databases, including the Highway Research Information Service database for domestic literature, the Highway Research in Progress database for ongoing research studies, and the International Road Research Database for relevant foreign literature. Based on that search, as expected, only a few studies have included the MPS. This chapter summarizes the literature on the safety effectiveness of pedestrian treatments that include a signal controller. Also provided is a summary of the literature that discusses how pedestrian volume at a signalized intersection can be estimated, because pedestrian volume is a key element in a safety analysis of pedestrian treatments. Midblock Pedestrian Signal Effectiveness TCRP Project D-08/NCHRP Project 3-71 (3) included several MPSs as part of its driver yielding study (see Table 1). That study introduced the concept of “red” and “yellow” devices to emphasize that if an agency wants to ensure high driver yielding at pedestrian crossing treatments, using a device that shows a red indication to the motorists is needed. Red devices include a high-intensity activated crosswalk (HAWK) beacon (officially called a pedestrian hybrid beacon, PHB), a half signal, or an MPS. Yellow devices include the rectangular rapid flashing beacon (RRFB), in-road warning lights, the LED-embedded pedestrian or school crossing sign, and others. These devices do not include a traffic signal controller. As shown in Table 1 and Figure 2, the MPS—similar to the other red devices studied—has very high driver yielding rates. An unpublished analysis of 108 MPS installations in Los Angeles included 5 years of crash data from October 1, 2013, to September 30, 2018. The study found a similar crash rate distribution between the MPS and the PHB. A 2020 study in Minnesota (4) investigated the impact on driver yield rates for pedestrian- activated crossing (PAC) systems at 34 locations. The midblock crossings in the study included RRFBs, PHBs, and one signal; the driver-yield rates and sample sizes are provided in Table 2. The author suggests the large difference between driver yield rates when the signal was and was not activated was because “the system in place was a standard signal which is a well-recognized traffic control device that clearly gives one party or the other the right-of-way; when pedestrians attempted to cross when they did not have the right-of-way, drivers were much less likely to yield.” As part of the 2020 Minnesota study, Hourdos (4) attempted to estimate the effects of PACs on pedestrian crash rates using simulation, but was not successful due to limitations with the study method. C H A P T E R 2

4 Safety at Midblock Pedestrian Signals Crossing Treatment TCRP Project D-08/NCHRP Project 3-71 Other Studies Compliance—Staged Pedestrian Crossing Compliance—General Population Pedestrian Crossing Compliance—Literature Review # of Sites Range (%) Average (%) # of Sites Range (%) Average (%) # of Sites Range (%) Average (%) MPS 2 97 to 100 99% 4 91 to 98 95% na na na Half signal 6 94 to 100 97% 6 96 to 100 98% 1 99 99% HAWK (PHB) 5 94 to 100 97% 5 98 to 100 99% 1 93 93% na = not applicable; there were no sites. Source: Fitzpatrick et al. (3) Site Driver Yield Rate When Activated Sample Size Driver Yield Rate When Not Activated Sample Size Treatment Lanes Crossed 7-Maple Plain 88.5% NP 56.5% NP PHB 2 9-Red Wing 66.2% NP 42.8% NP PHB 2 9-Red Wing from Island 93.0% NP 100% NP PHB 2 25-Wayzata 73.1% 130 55.1% 49 RRFB 1 4-Lewiston 81.8% 22 50.0% 6 RRFB 2 U2a-Wayzata 72.1% 172 66.7% 30 RRFB 2 U2b-Wayzata 68.2% 198 35.5% 107 RRFB 2 11-Anoka 98.0% 151 14.7% 61 Signal 2 NP = not provided. Table 1. Findings from data collected in 2003 for a TCRP/NCHRP study. Figure 2. Site average and range for driver yielding by crossing treatment. Table 2. Driver yield rates at midblock crossings by treatment.

Literature Review 5 Half-Signal Effectiveness Johnson (5) investigated the safety of half signals in Portland, OR, based on data collected from the city in a thesis for Portland State University. Half signals are located at four-way intersections and include a typical green-yellow-red traffic signal for automobiles on the major road, a stop sign for motorists on the minor road, and a pedestrian signal with actua- tion for pedestrians and/or bicyclists on the minor road. The treatment is prevalent in Canada, but the MUTCD (2) discourages its use in the United States. The treatment can be found in Portland, OR, and Seattle, WA, with the most recent installation of 47 half-signal intersections in Portland in 1986. Johnson reported on the crash history results from published half-signal studies as shown in Table 3. For the 2015 study, Johnson considered 442 crashes over a 10-year period for the 47 half-signal intersections in Portland. Of the 442 crashes, 16 involved pedestrians. The calcu- lated crash rates for the half signals (0.158 and 0.178 crashes per million entering vehicles for three-leg and four-leg half signals, respectively) did not differ significantly from comparison groups (e.g., minor-street stop-controlled intersections and signalized intersections). The matched comparison only showed rear-end (RE) crashes as being statistically significantly different, with half signals having more RE crashes than the minor stop-controlled intersections comparison group. The lack of pedestrian volume limited the ability to determine if the greater number of pedestrian crashes at half-signal intersections versus signalized intersections was statistically significant. PHB Safety Effectiveness Table 4 summarizes the findings from the three studies that found a statistically significant relationship between the PHB and crashes. In a 2010 FHWA study, researchers conducted a before-and-after evaluation of the safety performance of the device (10). Using an empirical Bayes (EB) method, their evaluations compared the observed crash frequency after installation of the PHB to the EB estimate of the expected crash frequency for the same after period without the PHB. NCHRP Research Report 841 (11) investigated the safety effectiveness of the PHB and developed several crash modification factors. The 2019 Arizona Department of Transportation (ADOT) study (12) also used an EB before-and-after analysis. The 2019 ADOT study included a cross-sectional evaluation that considered a larger sample size of PHB installations. Two relevant findings are: • Midblock (i.e., two legs) versus intersection (i.e., three or four legs) does not make a differ- ence with respect to safety at PHBs since no statistical difference in crashes between midblock Location (Study) Year Number of Sites Time Period Change in Pedestrian Crashes Before to After Change in Auto Crashes Before to After Half-Signal Crash Rate per Million Entering Vehicles Seattle, WA (6) 1974 18 0.6–2.5 yr −100%a, b +8% 0.525 Seattle, WA (7) 1988 22 14.5 yr −65% −10% na Seattle, WA (8) 2001 19 4–16 yr na −20.4% na Canada (9) 2003 25 3.25–5 yr −66% −23% 0.230 Portland, OR (5) 2015 47 10 yr Not a before/after study Not a before/after study 0.158 for 3-leg 0.178 for 4-leg a Total number of pedestrian crashes went from 4 in the before period to 0 in the after period. b Significant at the 85% confidence interval level. na = not available. Table 3. Johnson crash history findings.

6 Safety at Midblock Pedestrian Signals locations and those PHBs at three- or four-leg intersections was found in the cross-sectional evaluation. • The cross-sectional evaluation showed no statistically significant difference between the lower-speed and higher-speed PHB sites (posted speeds at 35 mph or lower versus 40 mph or higher) for all crash types except RE crashes. For RE crashes, fewer RE crashes were present when the posted speed limit was 35 mph or lower. Coordinated Signals Hauer (13) in 2020 discussed the benefits of retiming coordinated signalized intersections to improve the opportunities for pedestrians to cross midblock. Sources for Pedestrian Volume Studies on the effectiveness of pedestrian treatments can be limited when pedestrian volumes are not available, since pedestrian volume has been found to be the most influential factor in explaining the variation in counts of pedestrian crashes (14). The research team reviewed the literature on the methods available to estimate pedestrian volumes. On-Site Counts Several techniques are available for counting the number of pedestrians and bicyclists at a site, including manual counts, inductive loops, thermal cameras, infrared counters, and pedestrian signal actuation data. Kothuri (15) tested several technologies in a parking lot and at an inter- section. The author’s conclusions were that inductive loops and a thermal camera counted bicycles accurately in a controlled environment but failed to do so at an intersection. Passive infrared counters were found to count pedestrians accurately at the intersection sidewalk, and pedestrian signal actuation data could be a cost-effective surrogate for pedestrian demand at signalized intersections. Because this study was done about 10 years ago, the technology may have improved since. Crash Type/Site Type 2010 FHWA Study (10) 2017 NCHRP Study (11) 2019 ADOT Study (12) Total 29** 18.0** 18.2** Fatal and injury 15 NG 25.2** Pedestrian related 69** 56.8** 45.7** Fatal and injury pedestrian related NG NG 45.0** Rear end NG NG 20.5** Fatal and injury rear end NG NG 28.6** Angle NG NG 22.6** Fatal and injury angle NG NG 24.5* Rear end and sideswipe NG 12.4 NG Treated sites 21 PHB sites 27 PHB with advance stop markings and signs sites 52 PHB Sites Reference group 102 unsignalized intersections 3,129 sites in Charlotte, NC, Portland, OR, Phoenix, AZ, Scottsdale, AZ, Tucson, AZ, and St. Petersburg, FL, that did not have the following treatments: PHB, RRFB, refuge island 101 unsignalized intersections and 56 signalized intersections NG = crash reduction not generated for this crash type. Statistical level indications: * Statistically significant results with 90% confidence level. ** Statistically significant results with 95% confidence level. Table 4. Percentage of crash reduction at PHBs for several studies.

Literature Review 7 Direct-Demand Models While vehicle volume is frequently available for a street, pedestrian volume is seldom available. Turner et al. (16) provides a summary of techniques used to estimate and evaluate exposure to risk in pedestrian and bicyclist safety analyses. The authors note that geographic scale is a critical element, and provide four scale categories: • Regional (e.g., city, county, or state) • Network (e.g., traffic analysis zone, census tract, or census block group) • Road segment • Point (e.g., midblock or intersection street crossing) With respect to crash evaluations for MPSs, exposure estimates at the point scale group are needed. Direct-demand models are widely used for pedestrian and bicyclist volume estima- tion, but they require local data—transportation system variables, built environment vari- ables, socioeconomic characteristics, weather, typology—and are probably not transferable to different areas. Schneider et al. (17) in 2012 developed and applied a pedestrian intersection volume model for San Francisco, CA. A sample of counts at 50 intersections was collected and adjusted to produce annual pedestrian crossing estimates at each sampled intersection. Next, the authors developed a log-linear regression model to identify the relationship between annual pedestrian volume estimate and various explanatory variables including land use, transportation system, local environment, and socioeconomic characteristics near each sampled intersection (see Table 5). Griswold et al. (18) developed a statewide pedestrian exposure model using log-linear regres- sion. Their database included more than 1,200 count locations in California. The model included the following variables: • Employment density • Population density • Number of schools • Number of street segments • Intersections with principal arterial and minor arterial roadways • Four-way intersections • U.S. Census Bureau’s American Community Survey journey-to-work walk mode share Table 6 provides the variables and coefficient estimates for the recommended model. The model was used to estimate pedestrian volume at state highway intersections within 656 ft Model Variablea Recommended ModelCoefficient t-Value p-Value Total households within ¼ mi (ten thousands) 2.12 0.040 Total employment within ¼ mi (hundred thousands) 2.22 0.032 Intersection is in a high-activity zone 3.79 0.000 Maximum slope on any intersection approach leg (hundreds) −3.07 0.004 Intersection is within ¼ mi of a university campus 1.45 0.154 Intersection is controlled by a traffic signal 4.03 0.000 Constant 1.81 2.43 1.27 −9.40 0.635 1.16 12.9 33.29 0.000 Note: The dependent variable is the natural logarithm of annual pedestrian intersection crossing volume at each of 50 study intersections. This represents the sum of all crossings on each approach leg within 50 ft of intersections. Annual volume estimate is extrapolated from a 2-hour manual count taken in September 2009 or July–August 2010. The extrapolation method accounts for variations in pedestrian activity by time of day, day of week, weather, and land use. N = 50; adjusted R2 = 0.804; F (test value) = 34.4 (p < 0.001). a All distances used to calculate the model variables are straight-line distances rather than roadway network distances. Source: Schneider et al. (17). Table 5. Preferred model of pedestrian volume for San Francisco.

8 Safety at Midblock Pedestrian Signals (200 m) of a census block with a minimum population density of 500 people per square mile. The estimates are available at https://dot.ca.gov/programs/safety-programs/ped-bike/exposure. Schneider et al. (19) in 2021 developed pedestrian intersection crossing volume models for the Milwaukee, WI, metropolitan region. Negative binomial regression was used to relate annual pedestrian volumes at 260 intersections to roadway and surrounding neighborhood socio- economic and land-use variables. The range of annual pedestrian volumes in the model dataset was 1,000 to 650,000. The following variables describing the area surrounding an intersection had statistically significant, positive associations with annual pedestrian volumes: • Population density within 1,312 ft (400 m) • Employment density within 1,312 ft (400 m) • Number of bus stops within 328 ft (100 m) • Number of retail businesses within 328 ft (100 m) • Number of restaurant and bar businesses within 328 ft (100 m) • Presence of a school within 1,312 ft (400 m) • Proportion of households without a motor vehicle within 1,312 ft (400 m) While three models (see Table 7) had a good overall statistical fit, the authors recommended model C. They noted that the presence of a traffic control signal or a park within 400 m tended to have consistent positive associations with annual pedestrian volumes, but they were not significant at the 95% confidence level when included with the other seven variables in the final models. The authors also expected being within 400 m of a university campus would be signifi- cant; however, only six intersections had that characteristic. Le et al. (20) conducted a study to explore options for collecting or estimating pedestrian volume data in Dallas, TX, particularly at intersections with high pedestrian activity. The authors successfully developed a direct-demand model that estimates pedestrian volumes at signalized and stop-controlled intersections (see Table 8). The final model showed that pedestrian volume increases 4 times within downtown; increases 12% per school within 1 mi of an intersection; increases 4.8 times per 1% increase in commercial/multifamily residential land uses within 300 ft of an intersection; increases 4.7 times with the presence of higher education, hospitals, or malls; and decreases 36% per 5-mph increase in the intersections’ maximum posted speed limit. Geedipally (21) developed a regression model to estimate the pedestrian crossing volumes at midblock locations. The variables found to influence the crossing volumes were posted speed limit, number of bus stops, sidewalk width, and area type (see Table 9). Variable Scale Manipulation Transformation Estimate Pr (> ItI) Intercept na na na 5.58 < 2e-16 *** Number of employees ¼ mi Weighted log 0.390 < 2e-16 *** Population ½ mi Truncated 0.000142 < 2e-16 *** Number of street segments ½ mi Weighted log 0.302 2.08e-05 *** Walk commute mode share ½ mi na na 2.84 6.25e-08 *** Number of schools ½ mi na log 0.0444 1.38e-05 *** Principal arterial Intersection na na 0.457 4.17e-16 *** Minor arterial Intersection na na 0.384 6.23e-10 *** Four-way intersection Intersection na na 0.413 7.38e-09 *** Dependent variable: log (annual value estimate). Adj. R2 = 0.714. *** p < 0.001. na = not applicable. Source: Griswold et al. (18). Table 6. Final California pedestrian exposure model.

Literature Review 9 Variable A. Base Model B. Square Root Model C. Cube Root Model Beta p-Value Beta p-Value Beta p-Value Constant 8.334 0.000 7.629 0.000 7.071 0.000 PopDen400 0.000140 0.001 na na na na SRPopDen400 na na 0.019 0.000 na na CRPopDen400 na na na na 0.100 0.000 EmpDen400 0.000021 0.046 na na na na SREmpDen400 na na 0.00581 0.005 na na CREmpDen400 na na na na 0.036 0.003 BusStp100 0.336 0.000 na na na na SRBusStp100 na na 0.434 0.000 na na CRBusStp100 na na na na 0.477 0.001 Retail100 0.108 0.026 na na na na SRRet100 na na 0.208 0.000 na na CRReBa100 na na na na 0.471 0.000 RestBar100 0.116 0.062 na na na na SRReBa100 na na 0.208 0.050 na na CRReBA100 na na na na 0.244 0.044 SchDum400 0.515 0.001 0.478 0.003 0.499 0.002 Pct0Veh400 5.307 0.000 4.184 0.001 4.330 0.000 Sample Size (n) 260 260 260 Log-likelihooda −2,792 −2,774 −2,772 Akaike information criterion (AIC)a 5,601 5,565 5,560 Bayesian information criterion (BIC)a 5,629 5,593 5,588 a Lower absolute values of log-likelihood; AIC and BIC indicate better overall model fit. na = not applicable. Source: Schneider et al. (19). Table 7. Final annual pedestrian crossing volume models for Milwaukee. Parameter Estimate Std. Error p-value Intercept (b0) 5.3048 0.5157 < 0.0001 Indicator variable for the signalized intersection (lsig) 0.9630 0.1787 < 0.0001 Number of schools within 1 mi (bsch) 0.1566 0.0272 < 0.0001 Commercial and multifamily proportion (bco+mf) 1.4305 0.2250 < 0.0001 Posted speed limit (bpsi) −0.0578 0.0148 0.0001 Central business district indicator (bcbd) 0.9682 0.3178 0.0026 Special generator indicator (bspl) 1.2568 0.3458 0.0004 Number of bus stops (bbus) 0.0487 0.0565 0.3895 Dispersion parameter (δ) 0.7693 0.0717 < 0.0001 Log likelihood −1,213.04 na na AIC 2,444.07 na na Note: Italicized value means the variable is not significant at 5% level. Est. = estimate; std. error = standard error; na = not applicable. Source: Le et al. (20). Coefficient Variable Value Std. Dev. t-statistic p-value ΒO Intercept 7.9409 1.0544 7.53 < 0.0001 βpsl Posted speed limit −0.09465 0.02704 −3.50 0.0009 βbus Bus stops 0.2314 0.1187 1.95 0.0558 βsww Sidewalk width −0.06762 0.04771 −1.42 0.1614 βarea Area type (1.0 if commercial, 0.0 otherwise) 0.5777 0.2314 2.50 0.0152 K Inverse dispersion parameter 0.6459 0.109 2 5.92 < 0.0001 Observations: 64 midblock locations. Source: Geedipally et al. (21). Table 8. Estimated parameters for signalized and stop-controlled intersections in Dallas, TX. Table 9. Calibrated coefficients for pedestrian crossing volumes at midblock locations.

10 Safety at Midblock Pedestrian Signals Pedestrian Volume Estimates Using Traffic Signal Data Two recent papers explore the use of pedestrian push-button data as the source for pedestrian exposure. One of the papers uses Arizona data (22) and one uses Utah data (23). The researchers collected multiple hours of pedestrian crossing data at a few locations, developed adjustment factors, and then applied those factors to the pedestrian push-button counts from signalized intersections. The study using Utah data provided graphs showing the relationships between the unique pedestrian detections and pedestrian crossing volume. Pedestrian Volume Estimates Using Local Engineering Judgment Another approach is to use local engineering judgment to assign the site to a general level of pedestrian activity category developed from historical pedestrian count data. A recent ADOT PHB study (12) established typical pedestrian volumes by general level of pedestrian activity (see Table 10). The pedestrian volume values are based on the data from the 2010 FHWA study (10). The Highway Safety Manual (HSM; 24) data are included in the table as a comparison. Key Findings from Literature Key findings from the literature review that influence the MPS safety analysis include the following: • The effectiveness of the MPS has been evaluated using driver yielding as a safety surrogate in a previous study. • Previous safety analyses for similar pedestrian traffic control devices demonstrated that pedestrian volume should be considered in the analysis along with the number of legs at the crossing and the posted speed limit on the major street. • While the availability of pedestrian counts is limited, researchers have successfully developed models to estimate pedestrian volume. These models are specific to the area providing the counts used in model development. Creation of these models also requires a notable amount of staff resources. General Level of Pedestrian Activitya PH Bb P ed . M aj . 2 4 hr PH B Pe d. C ro ss 2 4 hr PH B Pe d. A ll 24 h r U ns ig .c Pe d. M aj . 2 4 hr U ns ig . P ed . C ro ss 2 4 hr U ns ig . P ed . A ll 24 h r Si g. d P ed . M aj . 24 h r Si g. P ed . C ro ss 24 h r Si g. P ed . A ll 24 h r H SM e S ig . 3 Le g H SM S ig . 4 Le g High 950 1,180 2,130 320 290 610 820 700 1,520 1,700 3,200 Medium–high 490 480 970 190 180 370 410 530 940 750 1,500 Medium 170 220 390 90 90 180 210 290 500 400 700 Medium–low 90 40 130 40 40 80 110 170 280 120 240 Low 40 20 60 10 20 30 60 60 120 20 50 a The team assumed the general level of high pedestrian activity to be the 90th percentile value (rounded to the nearest 10) for the group of sites. The medium–high was the 75th percentile, the medium was the 50th percentile, the medium–low was the 25th percentile, and the low was the 10th percentile value (rounded to the nearest 10). Other assumptions include that the PHB is controlling the vehicles on the major street and that the pedestrian count for “all” is the sum of the pedestrians crossing the major legs and the pedestrians crossing the cross-street legs (if any). b PHB values are based on 52 PHB (HAWK) intersections in Arizona. c Unsig. values are based on 98 unsignalized intersections in Arizona. d Sig. values are based on 33 signalized intersections in Arizona. e HSM values are from HSM (24) Tables 12–15, pp. 12–37. Table 10. Assumed pedestrian volume by general level of pedestrian activity.