Tens of thousands of people are killed or injured in traffic collisions each year. To improve road safety and combat life-threatening crashes, over 20 U.S. cities have adopted Vision Zero, an initiative born in Sweden in the 1990’s that aims to reduce traffic-related deaths and serious injuries to zero. Vision Zero is built upon the belief that crashes are predictable and preventable, though determining what kind of engineering, enforcement and educational interventions are effective can be difficult and costly for cities with limited resources.
While many cities have access to data about where and why serious crashes occur to help pinpoint streets and intersections that are trouble spots, the use of predictive algorithms and advanced statistical methods to determine the effectiveness of different safety initiatives is less widespread. Seeing the potential for data and technology to advance the Vision Zero movement here in the U.S., DataKind and Microsoft came together to help cities take advantage of data science to potentially predict and identify key factors and patterns of crashes to inform the work of city officials and aid in their efforts to reduce traffic fatalities and injuries to zero.
Three U.S. cities - New York, Seattle and New Orleans - partnered with DataKind to see how data science could further their Vision Zero efforts. Each city had specific questions that they wished to address related to better understanding the factors contributing to crashes and what types of engineering treatments or enforcement interventions may be most effective in helping each of their local efforts and increase traffic safety for all.
To execute this project, the first and largest multi-city, data-driven collaboration of its kind to support Vision Zero efforts within the U.S., DataKind launched its first ever Labs project. Labs projects differ from other DataKind programs in that they aim to make sector-wide impact and are primarily executed by DataKind staff. In this case, DataKind staff data scientists Erin Akred, Michael Dowd and Jackie Weiser did the bulk of the work with volunteers brought on as support, including Sean Chang and dozens of volunteers from Microsoft, University of Washington’s E-Science Institute and other Seattle data scientists that contributed at a weekend DataDive.
The cities provided information about their priority issues, expertise on their local environments, access to their data, and feedback on the models and analytic insights. The overall collaboration was enabled by Microsoft providing resources and expertise in support of the collaborative model and project goals.
Map showing locations of crashes that occurred at street improvement projects locations in New York.
According to the City of New York, on average, vehicles seriously injure or kill a New Yorker every two hours, with vehicle collisions being the leading cause of injury-related death for children under 14 and the second leading cause for seniors. Looking to improve traffic safety on its streets, the city wanted to understand what existing interventions are working and where there is potential for improvement to help inform how the city can better allocate its resources to protect its residents.
The team leveraged datasets from the New York City’s Department of Transportation, NYC OpenData, New York State as well as other proprietary data to examine the effectiveness of various street treatments to help inform the city’s future planning and investment.
Before they could answer these questions though, they first needed to answer a more basic one: How many cars are on the road? Knowing total traffic volume or “exposure” is useful for many calculations, but most cities don’t have a citywide measure of it. To overcome this, the Labs team designed an innovative exposure model that can accurately estimate traffic volume in streets throughout the city. The model has two main components. The first is an algorithm that propagates traffic counts on a single street segment to adjacent street segments. It assumes that traffic on one city block is very similar to traffic on adjacent blocks. This process can be run many times and allows one to widely propagate traffic count values along neighboring streets. However, some streets may not have any traffic counts available, so the second component of the model is a machine learning model, with high predictive accuracy, that predicts traffic volumes on streets based on their characteristics.
The team also created a crash model for New York, allowing the city to examine individual locations and test how different street characteristics impacts the number of injuries. For example, the city can look at a particular street and determine whether it is safer for the street to be a one- or two-way road.
The exposure model will prove to be invaluable to the City of New York, filling a crucial void in vehicle volume data that many cities face. With it, the city can now perform initial safety project feasibility studies very quickly and provide context for a variety of other safety research work that requires an “exposure” rate. The model can also be altered to estimate other defined traffic volume measures, like peak hour traffic volumes. This can be of great benefit to the city, as it means they can estimate, based on the number of cars on the road, how many traffic-related injuries may occur. It can also help inform future work related to traffic congestion and additional safety projects.
New York can also use the crash models to test the potential impact different engineering, land use and traffic scenarios would have on total injuries and fatalities in the city. They will continue to build upon the work started by DataKind, as the models developed set the stage for future research in crash prediction, congestion relief and city safety projects.
The team was able to leverage the work started in New York City to help develop and refine the approaches for both Seattle and New Orleans.
This “exposure” model developed for New York and Seattle showing estimates of citywide traffic volume, a key piece of information needed for advanced analyses that most cities don’t have.
While Seattle has seen a 30 percent decline in traffic fatalities over the last decade, traffic collisions are still a leading cause of death for Seattle residents age 5-24. Older adults are also disproportionately affected so this trend could grow as the population ages. To supplement the findings of the City’s Bicycle and Pedestrian Safety Analysis project and provide policy makers and engineers with actionable information for developing and implementing interventions, Seattle sought to find out what mid-block street designs are most prone to collisions and the probability of a crash at such locations. Seattle also wanted to understand what factors, or combination of factors (such as traffic volume, street characteristics and environmental variables), may contribute to crashes at intersections.
Using Seattle’s collision, roadway traffic, exposure data and environment characteristics, the DataKind team developed models to uncover collision patterns between vehicles and pedestrians or bicyclists and determined the extent to which driver behavior and/or street design is a contributing factor to crashes and the severity level of types of crashes. The team also applied the methodology developed for their work with New York to calculate exposure or total traffic volume citywide for Seattle.
By incorporating incident-specific information such as time of day, weather, lighting conditions and behavioral aspects, the team was also able to further develop a crash model to evaluate elements that may contribute to crashes at intersections and to what extent driver behavior, road conditions and street design played a role.
With the collision pattern models, the DataKind team was able to determine several variables that were the greatest predictors of street segment collisions - traffic volume, whether the land use was commercial or not, the number of traffic lanes, street width and pedestrian concentration.
With the crash factor model, the team was able to identify several environmental and behavioral patterns and trends that resulted in higher levels of injury severity. For instance, whether a driver is making a right turn or left turn at a given intersection will influence the severity of the collision. They were also able to identify which months of the year were better or worse for crash incidents. Interestingly, the number of crosswalks was found to be significant, with more crosswalks at an intersection resulting in reduced crash severity.
With this additional knowledge, Seattle will be able to pinpoint high risk areas and the factors that can be addressed to help reduce future crashes. Seattle recently passed a levy to fund multi-modal transportation improvements city-wide. The results from this project, along with additional safety studies, will help guide over $1 million in spending for future safety projects and further Vision Zero efforts in local communities.
While New Orleans hasn’t officially adopted Vision Zero, the city government and community are working together to make roads safer. New Orleans was named a “silver” level bicycle friendly community by the League of American Bicyclists and had the 8th highest share of bicycle commuters among major U.S. cities. New Orleans also leads Southern U.S. cities in bicycle commuting. Yet, a disproportionately high number of the state’s pedestrian crashes occur in New Orleans and the number of bicycle crashes doubled from 2010 to 2014.
To help the city protect its growing number of roadways users, New Orleans wanted to understand the impact that future street treatments, such as bike lanes and traffic signage, could have on preventing traffic injuries and fatalities.
The DataKind team created an Impact Assessment tool that could be used to test the effectiveness of installed treatments, which would then be used to better inform the placement of future street treatments, both individual interventions and groups of interventions applied simultaneously.
Specifically, the tool takes a set of treatment locations and uses different statistical methods to create sets of comparison locations. These comparison locations are used as a point of reference to gauge the impact of the treatment on traffic safety by comparing crash rates before and after the installation of interventions to similar intersections that did not receive interventions. The tool includes visualizations to examine generated comparison groups, as well as methods for using manually selected comparison groups.
As an example, New Orleans could select a treatment, such as a bike lane, and compare the crash rates before and after the bike lane was installed. The city can then compare these crash rates to other comparison sites. The comparison sites are especially important because they allow the city to control for outside factors, such as overall growth in population or traffic. Comparing the site with the treatment to similar untreated sites, we can see if the crash rate went up more slowly at the treated site, thus confirming that the treatment provided an improvement in safety even if it did not reduce the absolute crash rate.
New Orleans has integrated the Impact Assessment tool into their systems and will be collecting more data to maximize the tool’s potential and evaluate the effectiveness of additional street features. These findings will help inform the placement of future street treatments.
Overall, the work accomplished by the Vision Zero Labs team proved to be invaluable for the cities of New York, Seattle and New Orleans, equipping them with powerful insights, models and tools that can help inform future planning to prevent severe traffic collisions and keep all road users safe. With this knowledge, the cities can better determine how to best allocate resources and investments towards improvements in infrastructure and policy changes.
In addition to aiding the participating cities in their efforts to make streets safer, the project showed how data science can be effectively used to address complex civic issues like transportation safety. A particular example is the technique developed in this project around estimating road use volume even when complete relevant data is lacking. This technique is relevant both for safety analyses and broader transportation planning activities. These are the kinds of cutting edge and scalable solutions DataKind’s Labs projects aim to deliver to achieve sector wide impact.
The project also showed how collaboration between the public and private sector and amongst partner organizations can help benefit the greater good and result in innovative and scalable solutions to address complex and critical issues like traffic safety. Cities around the world will be able to benefit from the results of the Vision Zero Labs project and can adopt the methodologies and learnings from the work to reduce traffic-related injuries and fatalities in their own communities.
This Vision Zero Labs project provides a model for how collaboration between the public and private sector and among partner organizations can help benefit the greater good and result in innovative and scalable solutions to address complex and critical issues.