Pedestrian Safety and Transit Corridors

This research examines the relationship between pedestrian accident locations on state-owned facilities (highways and urban arterials) and the presence of riders loading and alighting from bus transit. Many state facilities are important metropolitan transit corridors with large numbers of bus stops users, resulting in increased exposure of pedestrians to traffic and in increased numbers of collisions. The research also examines the association between pedestrian collisions and other travel generators (concentrations of retail activity and housing) as well as environmental conditions (wide roadways, high traffic volumes, and high speed limits). Based on a retrospective sampling approach and logistic regression models, the study shows that bus stop usage is associated with pedestrian collisions along state facilities. Less strong, but significant associations exist between retail location and size, traffic volume, and number of traffic lanes, and locations with high levels of pedestrian-vehicle collisions. The findings suggest that facilities with high numbers of bus riders need to accommodate people walking safely along and across the roadway. They support the development of state DOT programs for multimodal facilities, which integrate travel modes in major regional facilities within local suburban communities and pay specific attention to the role of transit in shaping the demand for nonmotorized travel on the facilities. Journal of Public Transportation, Vol. 7, No. 2, 2004 74 Problem Statement Collisions between motor vehicles and pedestrians along state highways with transit routes are associated with high rates of injury and death of pedestrians and constitute a significant societal problem. In Washington State more than 30 percent of vehicle-pedestrian collisions are not on city streets where travel on foot may be expected, but on large state roads that are typically considered regional or transregional facilities designed for moving traffic (Washington State Department of Transportation 1997). Between January 1995 and December 2000 state facilities accounted for over 1,795 collisions involving more than 1,995 pedestrians (Table 1). Of these, 175 pedestrians were killed and 376 disabled. Using federal and state cost formulas, average yearly societal costs were more than $100,000,000. Collisions are especially concentrated in metropolitan areas. King County, Washington’s most urbanized county with 20 percent of the state’s population, has a disproportionate number of pedestrian-vehicle collisions. With 56 pedestrian fatalities and 144 disabling injuries, the county accounts for 36 percent of state societal costs associated with pedestrian collisions over this same six-year period. Within King County, collisions are concentrated on State Route 99 (SR 99), which accounts for 43 percent of pedestrian vehicle collisions in the county and 16 percent for the state as a whole. Originally part of US 99, first commissioned in 1926 and stretching from Canada to Mexico, SR 99 became the urbanized region’s second most important north-south thruway after the construction of Interstate 5 in the 1960s. Much of the corridor presents difficult and dangerous conditions for pedestrians. Development along the highway is strip commercial, the facility is wide with four to six travel lanes, traffic volumes are high, ranging from 20,000 to 40,0000 ADT, and large segments have no curbs and no sidewalks. It is also an important transit corridor. Washington King SR99 in State County King Co. 1995-2000 Avg. Yearly 1995-2000 Avg. Yearly 1995-2000 Avg. Yearly Collisions 1795 299 670 112 289 48 Pedestrians 1895 316 714 119 303 51 Fatal 175 29 56 9 23 4 Injuries Disabling 376 63 144 24 65 11 Injuries Societal $610,208,000 $101,701,333 $222,015,000 $37,002,500 $97,414,000 $16,235,667 Cost Table 1. Reported Pedestrian Collisions on State Routes, 1995–2000 Pedestrian Safety and Transit Corridors 75 State highways like SR 99 are common in many metropolitan areas. Designed for transregional traffic, these facilities have been lined with and surrounded by suburban development. As an alternative arterial street network was never developed (Untermann 1984; Southworth and Owens 1993), such regional facilities now also carry substantial local traffic. They even act as main streets, containing most of a community’s retail, commercial, and institutional uses. The mismatch of facility design and current patterns of use may be an important reason why there are high collision rates in these places. The financial and political costs of converting highways back to their former, narrower purposes would be enormous, and it is thus important to understand the relationship between new use patterns and collisions. Figure 1: Development along SR 99 north of Seattle. Regional highways have become urbanized roads with a variety of activities and uses. Journal of Public Transportation, Vol. 7, No. 2, 2004 76 National, state, and local road design programs are being developed and implemented to address the growing demand for multimodal transportation on facilities of regional significance within metropolitan areas (U.S. Department of Transportation and Federal Highway Administration 2003; Florida Department of Transportation 2001; Huang, Stewart et al. 2001). Transit plays a significant role in generating pedestrian traffic on these highways, making it essential that facility design insure the safe integration of transit users with the driving public. This research supports the development of these programs and policies, and specifically the need for safety investments in regional traffic facilities that act as de facto transit and pedestrian zones. Research Objective The main purpose of this research is to examine the relationship between pedestrian accident locations on state facilities and the presence of riders loading and alighting from bus transit, controlling for other factors. Transit riders are pedestrians exposed to potential vehicle collisions. Transit commuters, for example, depart and return on opposite sides of roadways, necessitating at least one daily crossing. Large numbers of bus stops users are therefore likely to be associated with increased collisions. The research also examines other pedestrian travel generators, such as concentrations of retail activity and housing, as well as physical environmental conditions that affect pedestrian safety, such as wide roadways, high traffic volumes, high speed limits, traffic signalization, and crosswalk markings (Zegeer, Stewart et al. 2002; Koepsell et al. 2001). The approach taken in this article differs from most previous safety research, which focuses on the increased risk of collision associated with facility characteristics while controlling for exposure. In other words, researchers have so far been interested in identifying unsafe conditions independent of the location and magnitude of pedestrian activity (Zegeer, Seiderman et al. 2002). This focus makes sense in areas where pedestrian volumes are high or evenly distributed along facilities, but it is less appropriate along state highways in suburbanized areas, where the presence of pedestrians tends to be sporadic. In these latter cases, the risk of collision is likely related not only to an interaction of pedestrian behavior and environmental factors, but also to actual pedestrian activity at certain locations. Unfortunately, data on the location and volume of pedestrian activity along suburban highways are incomplete or missing. In response, this study uses data on Pedestrian Safety and Transit Corridors 77 locations with high numbers of pedestrian-vehicle collisions and examines whether these locations are associated with potential pedestrian generators. The argument is that limited resources for improving pedestrian safety make it essential to target public spending at the most dangerous locations. To do so requires an understanding of the links between pedestrian generators, most specifically bus stops, and pedestrian-vehicle collisions. Variables and Data Sources The study area for the project is the urbanized area of King County. Highway segments with large numbers of pedestrian-vehicle collisions are treated as the dependent variable. Indicators of pedestrian activity including bus stop usage and land uses that likely generate pedestrian traffic are treated as one category of independent variables. Roadway and facility conditions are treated as a second category of independent variables. Pedestrian-Vehicle Collision Locations Primary data are based on Pedestrian Accident Locations (PALs) identified by the Washington State Department of Transportation (WSDOT). WSDOT defines a PAL as four or more collisions over a six-year period along a 0.10-mile section of roadway (528 feet). The concept of PAL was developed in transportation planning to identify highway segments that have large numbers pedestrian collisions. This research is first in using PAL to analyze underlying factors. PALs are used as the dependent variable because data on the precise locations of individual pedestrian collisions have not been available. Data on individual collisions would yield more analytical power, facilitating the use of nondichotomous variables and allowing for testing the effect of different spatial aggregations of collisions (beyond the 0.10 mile segments used). For the 1995–2000 data period, 47 percent of the State’s 120 PALs were located in King County (Table 2, Figure 2). King County PALs contained 55 percent of the total pedestrian collisions, 60 percent of fatalities, and 56 percent of disabling injuries located within all the PALs in Washington State. Because of large concentrations of PALs and continuously urbanized environmental conditions on SR 99, a separate analysis was done for this facility. SR99 contains 57 percent of PALS in King County and 27 percent of PALS in the State as a whole. Calculated societal costs for SR99 PALs average more than $10,000,000 a year. Journal of Public Transportation, Vol. 7, No. 2, 2004 78 Table 2. PALs, Constituent Injuries, and Costs in Washington State, King County, and SR99 in King County Indicators of Pedestrian Activity Indicators of pedestrian activity include bus stop usage, location of retail uses, concentrations of residences, and the locations of supermarkets, fast food restaurants, and school sites. It is hypothesized that these suburban pedestrian activity generators are positively associated with PAL sites. Data for bus stop usage are from METRO (the county transit agency) Automatic Passenger Counts (APC). Total daily boardings and alightings for each stop were averaged for two counting periods, fall 2000 and fall 2001. Land use data are from King County Assessor’s data for each tax parcel attached to a geospatial database of approximately 500,000 parcels. Indicators of Roadways Conditions Data for roadways include traffic volumes, roadway width and number of lanes, traffic speed, and speed limits. As volumes, speeds, and roadway size increase, it is hypothesized that pedestrian risk, especially for street crossing, also increases. WSDOT geospatial (GIS) data on state highways were used for geocoding and mapping PALs. Data on traffic volumes as well as roadway attributes, such as travel lanes and posted speed limits, were obtained from the Puget Sound Regional Council (PSRC). All data were spatially overlaid and combined using GIS. The number of intersections per one-half mile of linear roadway, or intersection density, was calculated from King County Network data using GIS and also used as an independent variable. Because the relationship between intersection density and environmental conditions is not well understood, the direction of the relaWashington King SR99 in State County King Co. 1995-2000 Avg. Yearly 1995-2000 Avg. Yearly 1995-2000 Avg. Yearly # of Pals 120 NA 57 NA 33 NA Collisions 554 92 305 51 186 31 Fatal 30 5 18 3 13 2 Injuries Disabling 123 21 69 12 45 8 Injuries Societal $173,919,000 $28,986,500 $98,327,000 $16,387,833 $64,795,000 $10,799,167 Cost Pedestrian Safety and Transit Corridors 79 tionship to PALs was not hypothesized. High intersection density may increase pedestrian risk because of frequent vehicle turning movements. Yet very low intersection density may also increase risk: because signalized intersections are typically placed one-half mile apart or more in suburban environments, pedestrians may engage in risky mid-block highway crossings rather than choose to make the long walk to a protected crossing and back. Figure 2. Pedestrian Accident Locations (PALs) on State Routes in King County Journal of Public Transportation, Vol. 7, No. 2, 2004 80 Research Design This research uses a retrospective sampling approach (Ramsey et al. 1994 ) to test variables for their power to distinguish between a set of predefined locations, in this case all PALs, and a set of other, randomly selected locations, in this case nonPAL sample points (hereafter referred to as “sample points”). Problems of spatial correlation precluded treating highways as a continuous series of points, and a limited number of sample points were drawn representing 0.10-mile segments. Sample points were not drawn along controlled or limited access facilities including large portions of SR 99 in the City of Seattle and Interstate 90. Approximately 50 sample points were drawn along SR 99 and 75 on other state facilities within the urbanized area (see Table 3). Data for pedestrian activity generators and roadway characteristics were attached to each PAL and sample point. Bus boardings and alightings were aggregated for each 500-foot highway segment, approximately overlapping the length of 528foot PAL segments. Land-use generators, such as the total floor area of commercial uses, were measured based on walking sheds of one-quarter mile from the center of PALs and sample points. Housing unit densities were measured using a one-half mile figure as a proxy for the potential for generalized pedestrian activity. Table 4 describes the variables and the data sources used. The principal modeling technique was binary logistic regression. Analysis was performed on three sets of data: (1) all facilities, (2) SR 99, and (3) facilities other than SR 99. Separating SR 99 was justified because it is more substantially developed and used than other facilities, and contains a disproportionate number of PALs. Table 3. PAL and Sample Points on SR99 and Other State Routes in King County Other SR99 State Routes Total PALs 33 23 56 Sample Points 49 76 125 Total 82 99 181 Pedestrian Safety and Transit Corridors 81 Ta bl e 4. P ri nc ip al V ar ia bl es Journal of Public Transportation, Vol. 7, No. 2, 2004 82 Analyses and Findings Variables were examined in terms of their means and standard deviations. Correlation analysis was used to explore basic relationships between variables and test for multicolinearity. Descriptive Statistics Basic descriptive statistics PALs and sample points on all state facilities in King County and for SR99 are presented in Tables 5 and 6. Mean bus stop use within all PAL and sample points is 54 persons day. On SR 99 only, this about doubles to a mean of 101 persons. The areas around PALs and sample points are clearly urbanized, although substantial variation is found in the mean of 100,000 square feet of retail space within one-quarter mile. Compared to the entire data set, points along SR 99 have more housing units, with a mean of almost 2,000 within the buffer zones. Also, a higher percentage of SR 99 points are located near groceries, fast food restaurants, and schools. State facilities are heavily trafficked, with a mean of 40,000 daily vehicles for all PALs and sample points, and a mean of 57,000 on SR 99. SR 99 also has more travel lanes. Finally, both sets show mean off-peak speeds modeled at just over 30 miles an hour. They have similar numbers of intersections per one-quarter mile of highway, with a mean of about 4.6, or about one every 300 feet, but there is a fair degree of variation in this figure. Table 5. Descriptive Statistics for PAL and Sample Points on All State Facilities in King County N Minimum Maximum Mean Std. Deviation BUS250 181 0 93 5.4 12.88 RETQRTMI 181 0 8.95 0.96 1.29 DUHLFMI 181 0 5578 1536 1031 HWYGRCRY 181 0 1 0.13 0.33 HWYFSTFD 181 0 1 0.38 0.49 SCHOOL 181 0 1 0.29 0.46 24HR_VOL 176 0.4 109.6 40.3 26.1 LAN_OP 176 2 8 3.9 1.2 CSPD_OP 176 12.1 44.7 31.5 5.9 INTSECT 181 1 13 4.57 3 Pedestrian Safety and Transit Corridors 83 Correlations Pearson correlation coefficients show that variables have only weak to moderate relationships with each other. Only two variables have Pearson coefficients above 0.5: 24-hour traffic volumes (24HR_VOL) with the number of dwelling units located within one-half mile of points (DUHLFMI) where the correlation is 0.51; and (24HR_VOL) with the number of travel lanes (LAN_OP) where the correlation is 0.69. These and all other correlations make basic sense and do not present statistical problems. Logistic Regression Logistic regression was used due to the dichotomous nature of the dependent variable (whether a point is a PAL or not). The technique assesses other variables in terms of their power to predict the value of the dependent variable. In this case, the probability that a site is a PAL divided by the probability it is a non-PAL sample site (an odd ratio) is linearly regressed against the vector of the predictor variables. The exponential function of variable coefficients, Ex(B), can be interpreted as a multiplicative effect on the odd ratio of a one-unit change in the variable. The intercept cannot be interpreted. All variables were entered into the regressions. Table 6. Descriptive Statistics for PAL and Sample Points on SR99 in King County N Minimum Maximum Mean Std. Deviation BUS250 82 0 93 10.1 17.6 RETQRTMI 82 0 8.95 1.17 1.47 DUHLFMI 82 18.64 5578.104 1984.49511 1181.436518 HWYGRCRY 82 0 1 0.16 0.37 HWYFSTFD 82 0 1 0.46 0.5 SCHOOL 82 0 1 0.38 0.49 24HR_VOL 82 12.9 109.6 57 24.6 LAN_OP 82 2 8 4.5 1 CSPD_OP 82 24 43.8 33.4 5.1 INTSECT 82 1 13 4.9 3.3 Journal of Public Transportation, Vol. 7, No. 2, 2004 84 Model 1: Results for ALL PAL and Sample Points in King County Model 1 is statistically significant at the 0.01 level (Table 7). The Cox and Snell R Square suggests that about a third of the variance in the dependent variable (whether a point is a PAL or not) is explained by the independent variables. Overall, 80 percent of total points are correctly predicted, with about 91 percent of non-PALs correctly predicted and only 57 percent of PAL correctly predicted. Two variables are statistically significant: BUS250, the number of people boarding and alighting from bus within 250 feet of the center of a PAL or sample point expressed in tens of bus users; and RETQRTMI, the amount of building area in retail uses within one-quarter mile of the center of a PAL or sample point, expressed in 100,000s of square feet. The value of Exp(B) for BUS250 suggests increasing bus stop usage by 10 people increases the odds that a point will be a PAL by 1.17 times. This supports the principal hypothesis of the study that increased bus stop usage is positively related to Pedestrian Accident Locations. Likewise, with RETQRTMI the value of Exp(B) suggests that adding 100,000 square feet of retail uses (about the size of two grocery stores) increases the odds that a point will be a PAL by about 1.5. In addition to increased pedestrian activity, increased levels of retail activity may also be associated with environmental factors that increase risk such as large numbers of active driveways along highways. The research cannot separate these possible effects. Model 2: Results for SR99 PAL and Sample Points in King County Similar to the first model, Model 2 is statistically significant at the 0.01 level with about a third of the variation in the dependent variable explained (Table 8). Slightly better than the first model, Model 2 classifies about 75 percent of all points correctly, with 86 percent of non-PAL sample points and about 61 percent of PAL points correctly classified. Bus stop usage is statistically significant, but unlike Model 1, retail activity (RETQRTMI) is not. The value of Exp(B) value suggests that an increase of 10 bus stop users increases the odds a site is a PAL by 1.16, similar to Model 1. The descriptive statistics for PAL and sample points helps explain these results (Table 9). On average, PALs have six times the bus stop usage of sample points, and about 50 percent more retail footage, but for all other variables, differences between PALs and sample points are slight. In other words, most points along SR 99 have similar conditions in terms of nearby land uses, traffic volumes and speeds, number of lanes, and intersection densities. The lack of variation in these variables urges cauPedestrian Safety and Transit Corridors 85 Table 7. Variables in Model 1 for All PAL and Non-PAL Sample Points in King County B S.E. Wald Sig. Exp(B) BUS250 0.158 0.033 23.029 0 1.171 RETQRTMI 0.398 0.195 4.165 0.041 1.489 DU1000 0.128 0.264 0.234 0.628 1.137 HWYGRCRY -0.889 0.706 1.584 0.208 0.411 HWYFSTFD 0.382 0.451 0.72 0.396 1.465 SCHOOL 0.011 0.485 0.001 0.981 1.011 24hr_VOL -0.118 0.118 1.007 0.316 0.888 LAN_OP 0.561 0.297 3.565 0.059 1.752 CSPD_OP -0.03 0.041 0.541 0.462 0.97 INTSECT 0.04 0.083 0.231 0.63 1.041 Constant -3.22 1.486 4.698 0.03 0.04 Summary Stats. Chi-Sq. DF Sig. -2 log Cox and likelihood Sneel R Sq. 71.5 10 0 146.7 0.33 Table 8. Variables in Model 2 for SR99 PAL and Non-PAL Sample Points in King County B S.E. Wald Sig. Exp(B) BUS250 0.151 0.039 15.409 0 1.163 RETQRTMI 0.332 0.246 1.828 0.176 1.394 DU1000 0.05 0.364 0.019 0.891 1.051 HWYGRCRY -0.231 0.896 0.067 0.796 0.794 HWYFSTFD -0.685 0.682 1.008 0.315 0.504 SCHOOL 0.065 0.633 0.011 0.918 1.067 24hr_VOL -0.266 0.18 2.17 0.141 0.767 LAN_OP 0.111 0.416 0.072 0.789 1.118 CSPD_OP 0.018 0.071 0.063 0.801 1.018 INTSECT 0.18 0.134 1.807 0.179 1.198 Constant -2.243 2.306 0.946 0.331 0.106 Summary Stats . Chi-Sq. DF Sig. -2 log Cox and likelihood Sneel R Sq.