Started in Sweden in 1997 with the goal of eliminating fatalities and severe injuries from traffic collisions, the Vision Zero initiative has since spread to more than 25 U.S. cities. These cities have committed themselves to eliminating traffic fatalities and severe injuries and have put in place local Vision Zero plans that establish a strategy for doing so.
These Vision Zero plans all contain two essential elements. For one, all Vision Zero cities acknowledge that traffic deaths are preventable and set the goal of eliminating fatalities in a set timeframe with clear, measurable strategies. Second, these plans embrace a multidisciplinary approach, bringing together a diverse set of stakeholders to address the problem.
However, the means of accomplishing these goals is a matter for cities to decide based on their needs and capabilities. In the past few years, many Vision Zero strategies have leveraged the power of data and technology in order to reduce deadly collisions. Cities have analyzed historical crash data to understand where crashes happen most often, what conditions correlate with collisions, and what road users are most vulnerable. Using this information, they have been able to target interventions in the areas that need them most.
Now cities are building on these data-driven efforts with the next wave of Vision Zero initiatives that seek to fill gaps in existing data and improve results for residents. Using a new suite of models and technologies, cities have been able to improve the accuracy, scope, and fairness of their efforts to eliminate traffic fatalities.
Prioritizing Interventions and Normalizing Collision Data
Beginning in 2015, data science nonprofit DataKind and Microsoft joined forces in an effort to help cities realize their mission to eliminate traffic fatalities. The organizations first approached New York City—a city that already had a robust Vision Zero program—offering pro bono support to fill the city’s need for data science resources in transportation.According to Rob Viola, Director of Safety Policy & Research for the New York City Department of Transportation (NYC DOT), the city was initially interested in an impact assessment tool. “We wanted a tool that would allow planners to sit down, input a project into a model, estimate what the injury reduction would be, and then be able to play with it,” he explained. “For example, what would the results be if I took a lane of traffic out? We were looking for a pie in the sky, Holy Grail type tool.”
However, the city simply did not possess enough of the necessary data to enable DataKind to produce such a tool. New York had pursued hundreds of street improvement projects—interventions that change a street’s makeup in an effort to improve safety. Still, on a granular level, these interventions were too distinct, meaning the city had very limited examples of each of the interventions it could input into its model.
Though the DataKind partnership did not meet the city’s initial goal, the work produced another valuable product: an exposure model that can estimate the number of vehicles on any given street. “We think of the number of cars on a street as something that the city probably knows, but no one actually knows this,” Viola explained. Before adopting DataKind’s model, the DOT was aware of traffic volume on only about ten percent of the city’s roads. The exposure model allows New York to estimate traffic volume on all city streets while only collecting a fraction of that data.
Adopting DataKind’s exposure model has provided the city with a number of advantages in transportation planning. For one, the DOT will be able to more easily vet potential areas for interventions. “One of the biggest constraints for interventions is how many vehicles are on the road,” Viola said. The exposure model will allow the city to identify those areas in which an intervention would not prohibitively disrupt city traffic.
Moreover, using the exposure model, the city has been able to more accurately determine the effectiveness of different street designs. By normalizing injury rates based on traffic volume, New York can better understand which street designs are truly dangerous, as opposed to those that appear in high traffic areas.
Following its work with New York, DataKind reached out to other cities to share its traffic safety insights. In Seattle, DataKind worked with the Department of Transportation in order to apply its exposure model to every corridor in the city and bolster the city’s existing Vision Zero efforts. Just as New York did, Seattle used the model to normalize where crashes and injuries happen in the city.
According to Jim Curtin, Seattle’s Traffic Safety Coordinator, this partnership has filled a technology gap in the city’s department of transportation. “We’re really grateful to Microsoft and DataKind for providing resources we simply don’t have,” he said.
The city has used this information to identify areas that pose a risk to pedestrians and bicyclists in particular. “Using DataKind’s models, we’ve been able to identify collision patterns and factors that have contributed to higher levels of injury severity,” Curtin said. “For example, we’re asking what mid-block street designs correlate with collisions involving vulnerable road users.”
Going forward, Curtin envisions the exposure model as a useful tool for galvanizing change in Seattle’s infrastructure. “Using normalized collision data informed by the exposure model, we’ll be able to spot critical locations and make a better case for interventions,” Curtin explained. “We can also consult this model when the city is working on bigger capital projects and ensure that we don’t just put things back the way they were. We can see what crash types are occurring and what measures will help prevent them.”
Drawing Insights from Near Misses
Starting in the City of Bellevue, WA through a partnership with the University of Washington, Microsoft piloted a project called Video Analytics Towards Vision Zero. The project applies machine learning to feeds from the city’s existing traffic cameras in an effort to identify conditions that lead to near misses, which are not recorded in traditional traffic data, in order to prevent accidents in the future. As opposed to collisions, which are costly learning experiences in terms of expensive damages as well as human lives, near-collision events present a zero-cost learning opportunity for cities, and video analytics is one of the first means of substantively understanding and leveraging such incidents. Since the initial release in Bellevue, New York City, Los Angeles, Seattle, Calgary, and Hamilton, Ontario have all signed on to participate in the project.The underlying video analytics system is a tracker technology that detects and follows the trajectory of moving objects and then classifies them into categories like pedestrians, bicycles, or cars using a deep neural network—a system that learns over time, simulating the central nervous systems of animals. These systems produce count reports that classify vehicles by turning movement, direction of approach, mode of transportation, speed, acceleration, and jerk. Based on a collection of near-collision conditions, the system produces risk scores, displayed on performance dashboards. A city is then able to flag high-risk locations and intervene to mitigate dangerous conditions.
In order to increase the accuracy of these systems, participating cities have appealed to the power of crowdsourcing. By analyzing pre-recorded videos, identifying what vehicles appear and how they move, online volunteers “teach” the system to identify features on its own. The Institute of Transportation Engineers (ITE) and partner cities host the Video Analytics Towards Vision Zero crowdsourcing platform on their websites, to which anyone—not only residents of participating cities—can contribute.
Analyzing Driver Behavior
Recent advances in Vision Zero analysis have sought to use information on driver behavior—ranging from cellphone use to hard braking—in order to improve traffic safety. As human error is responsible for around 90 percent of traffic accidents, analyzing and preventing risky behaviors is a logical next step for improving safety.The City of Boston capitalized on this opportunity by launching the Boston Safest Driver Competition, which ran from fall of 2016 until early 2017. For the competition, the city released a free mobile application—developed via a partnership between Boston’s Vision Zero Task Force, the Mayor’s Office of New Urban Mechanics (MONUM), and data sensing company Cambridge Mobile Telematics—that provided drivers with feedback based on acceleration, braking, cornering, and phone distraction. The app scored drivers between 0 and 100 based on their performance and each week, the city gave prizes to the top drivers, the most improved, new app users, and people who took car-free trips, and then awarded a grand prize at the end of the competition. In order to participate, all residents had to do was download the app and start driving.
The competition had a significant effect on driving behavior. 1,100 competitors saw their phone distraction drop by 47 percent and speeding drop by 35 percent, changes that could significantly reduce collisions if achieved throughout a whole city.
San Francisco-based tech company Zendrive has built upon this concept, offering a tool that uses sensors in smartphones to capture and analyze information on driver habits. The technology can sense when drivers brake or accelerate abruptly, how fast they are going at any moment, whether they are driving aggressively or distractedly, and even when they have recently stopped at a bar. The tool can then coach drivers on what they can do to drive more safely.
In 2017, Zendrive conducted a study in partnership with New York University’s Tandon School of Engineering in an attempt to understand the relationship between driver behavior and collisions. The study compared 33,450 risky driving events collected by Zendrive with four years of open New York Police Department (NYPD) crash data. Researchers found a 71 percent correlation between crash sites and examples of risky driver behavior, underscoring the potential to prevent accidents by promoting behavioral change on the roads.
Correlations between risky driving behavior and collisions (Zendrive)
A Focus on Equity Crowdsourcing for Better Data
Historically, collision data has not accurately and equitably represented crashes across communities. A study by Harvard T.H. Chan School of Public Health showed that police often underreport bike or pedestrian issues and, more troublingly, Vision Zero Network executive director Leah Shahum has found evidence of underreporting incidents involving people of color. “The police are not taking reports as seriously as they should in some situations,” she explained. “In a few of the cities we’re working with, there are certain categories that are underreported especially with walking and biking and people of color. We look at the hospital data and see higher rates of people walking and biking and people of color that weren’t captured in police data.”Moreover, collision data is often skewed by a general hesitancy to report incidents. Observers often do not report near misses or other incidents for fear of making an unwarranted 911 call.
In response, a number of cities have called on residents to share their insights into road safety via crowdsourced maps. These maps seek to fill gaps in city traffic data by leveraging community knowledge via easy to use and anonymous tools.
Boston, Philadelphia, and Denver have all recently cultivated residents as an additional data source for their Vision Zero efforts. Boston’s Vision Zero map—recent winner of Harvard’s Map of the Month—allows users to report safety concerns pinpointed by street or neighborhood. On Philadelphia’s map, residents can identify both areas where traffic conditions make them feel safe, and those where conditions make them feel unsafe. Both cities’ tools also allow users to map up-to-date accident data from police departments and emergency medical services that identify those areas with the highest concentrations of crashes and injury rates. As did Philadelphia, Denver collected resident opinions on which city streets are safe and which are not. However, Denver went one step further, also asking residents why they perceive certain areas as safe versus unsafe, adding another layer of data to inform city traffic interventions.
Denver also designed its crowdsourcing efforts to intentionally engage a representative user base. In addition to distributing an online survey to residents, the city has conducted in-person interviews in order to gather information from residents who may not have smartphone or computer access. In doing so, the city has ensured that its traffic information is more equitable and therefore more accurate.
Cities have used this more complete traffic data to inform traffic safety decisions. For example, Boston has installed 34 new traffic radar signs—digital speedometers that provide real-time speed feedback to drivers—in locations chosen in part based on information from the city’s map.
Building Equity Into Models
Traffic safety inequity is not only a matter of underreporting: traffic collisions themselves disproportionately affect residents of color and low-income residents. Understanding this disparity, many cities have implemented Vision Zero initiatives that not only seek to gather more accurate and representative data, but also acknowledge existing inequalities and prioritize interventions in traditionally underserved communities.In low-income neighborhoods and communities of color, a history of inadequate public investment has resulted in unsafe infrastructure conditions and therefore more traffic collisions. Resolving these inequities requires examining how resources are distributed among neighborhoods and what land use patterns are contributing to traffic deaths. “Look at the policies, practices and norms that have allowed these disparities to arise in the first place,” said Elva Yanez, director of health equity at the Prevention Institute, a health non-profit seeking to advance the practice of primary prevention.
Often, inequities arise as neighborhoods attracting new development receive infrastructure upgrades, paid for by developers, while lower-income neighborhoods are passed over. Instead, these lower-income neighborhoods are designated for land uses like high-speed arterials and concentrations of liquor stores that increase traffic risk.
The Vision Zero Network has responded to these concerns, publishing a case study intended to ensure that cities have the tools to make their Vision Zero efforts work for all residents, especially vulnerable and traditionally underserved populations. The paper calls for cities to create demographically representative Vision Zero leadership teams and work to emphasize equity issues within their efforts.
Portland, OR is one city that has made intentional efforts to prioritize equity within its Vision Zero initiative. The Portland Bureau of Transportation (PBOT) has integrated a Communities of Concern index into the model it uses to evaluate those areas in greatest need of traffic interventions. Areas defined as Communities of Concern score in the top quartile in equity indicators including people of color, low-income households, and poor vehicle access. Residents living in these communities often have fewer choices about how, when, and where they travel, putting them at greater risk of traffic collisions and justifying greater attention for Vision Zero interventions.
In order to ensure continued emphasis on underserved communities within the city’s data-driven Vision Zero efforts, PBOT has sought to institutionalize an equitable approach. In its Vision Zero action plan, Portland starts with an explicit statement that equity should figure prominently in the city’s plans. And, to ensure the city follows through on these plans, of the 26 members of the city’s Vision Zero task force, 10 represent organizations whose mission is to advance equity.
A number of other cities including Los Angeles, Austin, Chicago, San Francisco, and Seattle have started moving towards similar projects. These efforts start with gathering data on how traditionally underserved communities are affected by traffic collisions, then using this data to prioritize city investments.
And some cities have approached equity questions as a part of their broader public health programs. Vision Zero published a case study on a public health approach to collision reduction, stressing the need to focus on the root causes of health inequities, including disparities in collisions. The report argued that if 30,000 people were killed each year in the U.S. of a curable illness, with particular concentrations in underserved communities, we would call it a public health crisis. However, this is exactly the case with traffic collisions, and cities must therefore treat them as they would an epidemic: by using data to identify root causes of fatalities and inequities. Some cities have already started down this road. For example, Chicago has found that analyzing collision data alongside other indicators like Department of Health wellness data can create opportunities for collaboration to improve public health and safety in a coordinated and holistic way. Feelings of safety from violence, for instance—a critical aspect of wellness—are also related to frequency of collisions. Making residents of a neighborhood feel safe from gun violence may therefore be a critical means of reducing fatal collisions that has little to do with street design or speed limits.
Recent Vision Zero efforts have sought to fill in data gaps in order to improve the accuracy and fairness of traffic interventions. As technology and the quantity and quality of municipal data continue to improve, so too should the effectiveness of these interventions. However, whether or not cities achieve “Vision Zero” in the near future will hinge on their willingness to continually invest in these new technologies and deploy interventions that may be burdensome in the short term. As incentive to do so, cities must keep in mind Vision Zero’s foundational principle: “No loss of life is acceptable.”
This article was originally posted by Data-Smart City Solutions.
