An area-level model of vehicle-pedestrian injury collisions with implications for land use and transportation planning

https://doi.org/10.1016/j.aap.2008.10.001Get rights and content

Abstract

There is growing awareness among urban planning, public health, and transportation professionals that design decisions and investments that promote walking can be beneficial for human and ecological health. Planners need practical tools to consider the impact of development on pedestrian safety, a key requirement for the promotion of walking. Simple bivariate models have been used to predict changes in vehicle-pedestrian injury collisions based on changes in traffic volume. We describe the development of a multivariate, area-level regression model of vehicle-pedestrian injury collisions based on environmental and population data in 176 San Francisco, California census tracts. Predictor variables examined included street, land use, and population characteristics, including commute behaviors. The final model explained approximately 72% of the systematic variation in census-tract vehicle-pedestrian injury collisions and included measures of traffic volume, arterial streets without transit, land area, proportion of land area zoned for neighborhood commercial and residential-neighborhood commercial uses, employee and resident populations, proportion of people living in poverty and proportion aged 65 and older. We have begun to apply this model to predict area-level change in vehicle-pedestrian injury collisions associated with land use development and transportation planning decisions.

Introduction

In the 20th century, pedestrian needs were rare priorities in urban and transportation planning (Frumkin et al., 2004). Yet, environments that support walking can benefit human health by reducing motor vehicle collisions, motor vehicle-related noise and air pollution, and increasing physical activity and social cohesion (Cavill, 2001, Ewing, 2006, Leyden, 2003, Lavizzo-Mourney and McGinnis, 2003). To achieve walkable communities, planning professionals need practical tools to assess and mitigate the impact of development on pedestrian safety, including vehicle-pedestrian collisions.

Traffic collisions are a major cause of mortality in the United States (Mokdad et al., 2004), and the leading cause of death for persons aged 4–34 (Subramanian, 2006). Nationwide, pedestrians account for 11% of motor vehicle collision fatalities, with approximately 4700 pedestrian deaths in 2006 (NHTSA, 2006a). 15% of those people killed while walking died in California (NHTSA, 2006b).

Among California cities, San Francisco has historically had the highest per capita vehicle-pedestrian injury collision rate (STPP and California Walks, 2002). In stark contrast with the national figure of 11%, pedestrians account for half of San Francisco traffic deaths, with 13 fatalities and 726 non-fatal vehicle-pedestrian collisions in 2006. Pedestrian injuries and fatalities in San Francisco have declined over the last decade, attributed to intersection and mid-block pedestrian safety countermeasures, traffic calming, law enforcement, and improved planning efforts. Still, San Francisco's injury rate remains approximately 100/year/100,000 population (CCSF MTA, 2007, U.S. Census Bureau, 2000) or over five times the Healthy People 2010 national target of no greater than 19 pedestrian injuries/year/100,000 people; San Francisco's fatal injury rate of 2/year/100,000 is twice the national target (US DHHS, 2000).

Motor vehicles and pedestrians are two necessary component causes of vehicle-pedestrian injury collisions. San Francisco is a relatively dense, urban city, with approximately 776,000 residents and over 250,000 additional non-resident employees. By 2025, residential and job growth are expected to increase trips to, from, and within San Francisco by 12% (SFCTA, 2004). Of the projected 5 million trips in 2025, 3.3 million will be within San Francisco and over 50% of those are estimated to be auto trips. Both the relatively high frequency of pedestrian injuries and fatalities and the projected growth in San Francisco's traffic and population underscore the need to prioritize pedestrian safety needs in land use and transportation planning processes.

Currently, limited planning tools are available to evaluate the impacts of land use planning on pedestrian safety conditions. The Pedestrian and Bicycle Crash Analysis Tool software identifies pre-crash actions that lead to collisions, and links them to potential mitigation strategies (PBCAT, 2007). Crossroads software (Crossroads, 2007) and zonal analysis (USDOT, 1998) identify collision patterns and areas with high densities of pedestrian injuries.

Tools for prospectively forecasting the impacts of transportation and land use development on future vehicle-pedestrian collisions would complement the above methods for assessing existing collision patterns. To be useful in a planning context, a vehicle-pedestrian injury collision forecasting model needs to be based on available or routinely produced data, provide meaningful, easily interpreted, robust estimates, and be applicable in diverse areas to routine land use and transportation planning decisions. We are not aware of any vehicle-pedestrian injury collision forecasting tools in general use by planners for environmental or health impact assessments.

Empirically, increases in road facility vehicle volume increase the probability of vehicle-pedestrian conflicts on that facility (Lee and Abdel-Aty, 2005). A simple way to forecast change in vehicle-pedestrian collisions associated with change in vehicle volume is by applying a road safety function—which describes the relationship between traffic volume and collisions. The following power function (1.1) is an empirically supported parametric form of a road safety function, where AADT = Average Annual Daily Traffic:Δ(%),vehicle-pedestriancollisions=FutureAADTBaselineAADTβ1×100Typically β < 1, and empirical evidence suggests that 0.5 is a reasonable parameter (Lee and Abdel-Aty, 2005). At β = 0.5, vehicle-pedestrian collisions are forecasted to increase proportional to the square root of AADT, with a 50% increase in AADT predicting a 22% increase in collisions. Fig. 1 graphically illustrates the relationship between change in vehicle volume and change in the number of collisions as β varies. Applying this power function (1.1) to estimate collision increases associated with traffic volume changes due to area-level development is more challenging and requires simplifying assumptions, including: (1) development does not affect pedestrian flow and behavior; (2) development does not implement pedestrian safety countermeasures; and (3) AADT changes at intersections or street segments selected for evaluation are reasonable surrogates for changes at adjacent area roadways. (We included an example application in the Appendix A.)

As vehicle volume is not the only variable mediating the impacts of development on vehicle-pedestrian injury collisions, a multivariate area-level model might more robustly predict related change in collisions. In this paper, we describe our development of a context-specific regression model for forecasting vehicle-pedestrian injury collisions that includes local traffic volume and environmental and area-level population determinants associated with vehicle-pedestrian injury collisions.

Fig. 2 describes the conceptual framework that informed our model development. Specifically, we sought to understand how an area's built environmental context – street and land use characteristics – as well as compositional factors, including resident and employee population size, population characteristics and travel behaviors, predict the area-level distribution of vehicle-pedestrian injury collisions. Vehicle-pedestrian injury collisions are also associated with a number of individual-level factors including age, alcohol consumption, and other driver or pedestrian behaviors (Laflamme and Diderichsen, 2000, Ryb et al., 2007, Wazana et al., 1997). In the typical study of individual-level determinants, the environmental context of the injury is viewed as a “given” (Christoffel and Gallagher, 1999); however, individual behaviors occur in and are influenced by the environment, which is the focus of our research.

Previous research on environmental correlates of vehicle-pedestrian collisions shows that traffic volume is a significant predictor (Brugge et al., 2002, LaScala et al., 2000, Lee and Abdel-Aty, 2005, Loukaitou-Sideris et al., 2007, Roberts et al., 1995), while injury severity is largely determined by vehicle speed (Ewing, 2006, NHTSA, 1999). Other roadway characteristics associated with pedestrian injuries include street type (e.g., residential, freeway, arterial) and intersection and street design features (e.g., traffic and pedestrians signals, signage, lighting) (Ewing, 2006, Retting et al., 2003). Similarly, the land use type in an area has been associated with vehicle-pedestrian collisions (overall and fatal)—with increases predicted by increasing proportions of land used for commercial, mixed use, park, retail, or community uses (Geyer et al., 2005, Kim et al., 2006, Loukaitou-Sideris et al., 2007, Wedagama et al., 2006).

Pedestrian volumes, at the intersection-level as well as larger geographic regions, are also associated with increased pedestrian injury risk, though individual risk may be attenuated as pedestrian volumes increase (Geyer et al., 2005, Jacobsen, 2003). Actual pedestrian count data is not routinely collected in the United States; however, U.S. Census data on population or commute travel mode data can serve as a surrogate for pedestrian volume (Jacobsen, 2003).

Aside from pedestrian volumes, specific population characteristics can affect vehicle-pedestrian collision risk. Vehicle-pedestrian collisions are a leading cause of injury and death for youth (Walton-Haynes, 2002). Nationally, youth aged 10–20 have the highest population rates of pedestrian (non-fatal) injury at 35 injuries/100,000, well above the overall population rate of 20/100,000 (NHTSA, 2006b). Seniors aged 65 and over actually have non-fatal injury rates slightly lower than the overall population rate (some have speculated due to less pedestrian activity); however, seniors are more likely to die when hit by a vehicle based on national and local data (NHTSA, 2006b, Sciortino and Chiapello, 2005a). The elderly and children take longer to cross a street, increasing their exposure for injury (Demetriades et al., 2004), and children also have less developed cognitive, perceptual, motor and traffic safety skills (Johnson et al., 2004). Lower income children have a higher rate of pedestrian injury than higher income children, though the mechanisms contributing to this disparity – including the physical and social environment – are not well understood (Laflamme and Diderichsen, 2000, Johnson et al., 2004, LaScala et al., 2004).

Findings from many of the above studies may be specific to local contexts, and the resulting findings and risk estimates therefore may not be generalizable. In addition, some of the above studies did not adjust for confounding by important covariates, while others standardized outcome variables by factors we would like to understand as predictors—such as street length or land area.

Vehicle-pedestrian collisions tend to be dispersed throughout urban areas, and these dispersion patterns are missed by intersection or other micro-level analyses that focus on “black spots” with pre-existing high crash rates (Campbell et al., 2004, Morency and Cloutier, 2006). For example, from 2001 to 2005, eliminating all vehicle-pedestrian injury collisions at the five San Francisco intersections with 10 or more collisions during that period would leave over 98% of the city's vehicle-pedestrian injury collisions unaddressed (CCSF MTA, 2006). However, based on our data review, almost 10% of San Francisco's vehicle-pedestrian injury collisions were concentrated in two of 176 census tracts. A macro-level approach focused on census tracts could inform area-wide community transportation safety planning, and complement micro-level traffic safety mitigation measures such as intersection signalization (Lovegrove and Sayed, 2006).

Transportation researchers have modeled motor vehicle collisions at an area-level using multivariate regression methods, aggregate variables and linked datasets (Hadayeghi et al., 2003, Ladron de Guevara et al., 2004, Lovegrove and Sayed, 2006). Positive associations between collisions and traffic volume or vehicle miles travelled, population density, road network, and area-level socio-demographic characteristics are consistently significant in these macro-level models, which include pedestrian collisions with all motor vehicle collisions. Given potentially different determinants and risk estimates, separate macro-level vehicle-pedestrian collision models are warranted. For example, Loukaitou-Sideris et al. (2007) analyzed the spatial distribution of vehicle-pedestrian collisions in Los Angeles, and found pedestrian exposure, traffic, socioeconomic and land use variables were predictive of census-tract collision density.

To evaluate and model census-level predictors of vehicle-pedestrian injury collisions in San Francisco, we used cross-sectional, aggregated data, to (1) describe the distribution of vehicle-pedestrian injury collisions and select environmental and population characteristics in San Francisco census tracts; and (2) estimate the nature and strength of the independent effect of census-tract traffic volume on census-tract vehicle-pedestrian injury collisions, adjusting for covariates. We then discuss the strengths and limitations of this approach and its potential for practical application to predict change in vehicle-pedestrian injury collisions associated with land use development and transportation planning decisions.

Section snippets

Methods

This area-level model is based on cross-sectional data for San Francisco, California County, aggregated at the level of the census tract (outlined in Fig. 3). We selected our analytic variables based on the previous literature and our interest in environmental predictors of vehicle-pedestrian injury collisions as detailed in Fig. 2.

Results

There were 4039 recorded vehicle-pedestrian injury collisions in San Francisco's 176 census tracts from 2001 to 2005, with a median 14 and mean 23, ranging from 0 to 191 vehicle-pedestrian injury collisions in a tract (Table 1). As illustrated in Fig. 3, vehicle-pedestrian injury collisions were dispersed throughout the city, with evident concentrations in areas near freeways and highways that carry high traffic volumes from bridges and highways, as noted in previous literature (UCSF SFDPH, 2004

Discussion

In San Francisco, California, in a multivariate regression model at the census-tract level, statistically significant predictors of vehicle-pedestrian injury collisions include traffic volume, arterial streets without public transit, proportions of land area zoned for neighborhood commercial use and residential-neighborhood commercial use, land area, employee population, resident population, proportion of people living in poverty, and proportion of people aged 65 and over. All model variables

Conclusion

Consistent with previous national and international findings (Roberts et al., 1995, Lee and Abdel-Aty, 2005, Brugge et al., 2002, LaScala et al., 2000), our study provides additional evidence that traffic volume is a primary environmental cause of vehicle-pedestrian injury collisions at the area level. In addition to traffic volume, employee and resident populations, arterial streets without public transit, proportions of land area zoned for neighborhood commercial use and

Acknowledgements

The authors acknowledge the helpful contributions of Cynthia Comerford Scully, MA, Environmental Planner, San Francisco Department of Public Health, for ArcGIS data management; and Tom Rivard, Senior Environmental Health Specialist, San Francisco Department of Public Health, for guidance and support in utilizing the traffic count database.

References (57)

  • N. Cavill

    Walking and health: making the links

    World Transport Policy and Practice

    (2001)
  • CCSF MTA (City and County of San Francisco Municipal Transportation Agency Traffic Engineering Division), 2007. San...
  • CCSF MTA (City and County of San Francisco Municipal Transportation Agency Traffic Engineering Division), 2006. San...
  • Census Transportation Planning Package, 2000. http://www.mtc.ca.gov/maps_and_data/datamart/census/ctpp2000, accessed on...
  • CHP (California Highway Patrol), 2008. Accident Investigation Unit. Statewide Integrated Traffic Records System....
  • T. Christoffel et al.

    Injury Prevention and Public Health: Practical Knowledge, Skills, and Strategies

    (1999)
  • Crossroads Software Inc., 2007. http://www.crossroadssoftware.com/. Accessed on 1 March...
  • Ewing, R., 2006. Fatal and Non-fatal Injuries. Understanding the Relationship Between Public Health and the Built...
  • H. Frumkin et al.

    Urban Sprawl and Public Health: Designing, Planning, and Building for Healthy Communities

    (2004)
  • Geolytics Inc., 2004. Neighborhood Change Database (NCDB)...
  • Geyer, J., Raford, N., Ragland, D., Pham, T., 2005. The Continuing Debate about Safety in Numbers—Data from Oakland,...
  • A. Hadayeghi et al.

    Macrolevel accident prediction models for evaluating safety of urban transportation systems

    Transportation Research Record

    (2003)
  • P.L. Jacobsen

    Safety in numbers: more walkers and bicyclists, safer walking and bicycling

    Injury Prevention

    (2003)
  • Johnson, E., Geyer, J., Rai, N., Ragland, D.R., 2004. Low Income Childhood Pedestrian Injury: Understanding the...
  • K. Kim et al.

    Influence of land use, population, employment, and economic activity on accidents

    Transportation Research Record

    (2006)
  • F. Ladron de Guevara et al.

    Forecasting crashes at the planning level: simultaneous negative binomial crash model applied in Tucson, Arizona

    Transportation Research Record

    (2004)
  • L. Laflamme et al.

    Social differences in traffic injury risks in childhood and youth—a literature review and a research agenda

    Injury Prevention

    (2000)
  • R. Lavizzo-Mourney et al.

    Making the case for active living communities

    American Journal of Public Health

    (2003)
  • Cited by (288)

    • Evaluating interventions for phone distracted pedestrians in a virtual reality environment

      2023, Transportation Research Part F: Traffic Psychology and Behaviour
    View all citing articles on Scopus
    1

    Present address: Marin Community Clinic, 250 Bon Air Road, Greenbrae, CA 94904, United States.

    View full text