A key to democratizing urban solutions is building better models
By relying on sharper data, new approaches to modeling can lead to faster policies and greater consensus.
A/B testing — the ability to compare two different versions of the same product or service — is a key tool in the technologist’s toolbox, but one that’s more difficult for urbanists to deploy. For example, if you wanted to study solutions to transit problems on the east side of Manhattan, ideally you could build and compare two different options: bus lanes on First and Second avenues, and the Second Avenue subway. Of course, in the case of the multi-billion-dollar Second Avenue subway project, carrying out this type of A/B test is wildly impractical. And even when such testing could be more easily implemented, such as re-routing fixed-route bus service, doing so can be unnecessarily disruptive to riders.
So how do urbanists seek information on the potential efficacy of urban planning solutions? We build models. As Sidewalk reimagines cities for the digital age, Model Lab is exploring new ways to approach modeling as a means toward addressing big urban challenges. Models can be used to shape, test, get stakeholder feedback on, and adjust ideas related to land use, transportation, government processes, and many other areas of city life.
When cities tackle transportation problems, for instance, they create simulation models in which travelers move about cities: going to work, dropping children off at school, running errands. These simulations are based on theories of traveler behavior developed and tested by academics and practitioners. For example, one theory posits that travelers consider every minute waiting for a bus about twice as annoying as every minute riding on a bus. These theories are tested and calibrated against survey data collected by the Census Bureau and local governments. Once the simulations do an adequate job of replicating what’s happening today, we modify model inputs to simulate what might happen five, 10, or 20 years from now. These inputs may include a new transit service or wider roadways or higher bridge tolls or myriad other policy and planning ideas. The goal is to learn how people may benefit from, or be burdened by, these changes.
In the best case, these models deliver insights to planners which they have not previously considered. Transportation systems, like so many systems that impact cities, are dynamic and complex. For example, when individuals decide to drive rather than take the bus, desirable roads become congested; travelers may then respond to congestion by leaving for work earlier in the morning. The conceit of transportation models (and those looking at other areas of urban life) is that by representing individual decision-making we can learn something about how these decisions lead to a collective impact on a system, and in turn, about how the impacted system affects individual decision-making.
Thinking with models
The greatest value of building models is the disciplined way they require us to think about the impact of city policies and infrastructure on residents and visitors. Consider the increasingly common speculation of a future with shared autonomous vehicles. The potential benefits of shared autonomous vehicles are compelling: greater mobility, less congestion, less pollution. But how can we predict how many of us will want to share rides in autonomous vehicles with strangers?
Let’s consider all the questions a modeler would pose about sharing rides in autonomous vehicles.
What would make a traveler more or less willing to share a ride? What is the traveler trying to optimize? Cost? Time? Comfort? Convenience? All of the above? A first pass of model variables may include: time and cost, relative to other alternatives; a willingness to share a small space with other people; the physical difficulty of entering or exiting a shared vehicle; and, the perceived burden of out-of-my-way travel necessary to serve other passengers. Does your model have any variables we missed?
A modeler’s disciplined approach to asking these questions helps translate speculation into a systematic procedure for predicting a traveler’s response to sharing rides in autonomous vehicles. Now let’s consider actions that a local government might take to encourage sharing.
How much more expensive must traveling in a car by yourself be to motivate sharing? How can curb space be allocated more efficiently to reduce the entering/exiting burden? What type of regulations are needed to encourage sharing but protect consumers (from unfair prices or out-of-the-way travel)? What other strategies can cities use to incentivize sharing?
Thinking about how individual preferences intersect with policies — or infrastructure and services — is only the beginning of the fun of building models. Our next step is to translate these preferences and policies into mathematical expressions and then turn these mathematical expressions into computer simulations. Then we compare, over and over, the performance of these simulations against observed outcomes. When we’re done we have a model — one that we hope is useful.
Building consensus with models
Models provide the opportunity to create a dispassionate venue in which ideas can be explored and tested by anyone interested in the topic at hand. To achieve this goal we must first build models that resonate with decision-makers and the public as credible, legible, realistic, and compelling. We must then allow anyone the ability to create their own solutions, and investigate and explore solutions created by others.
One area in which Model Lab hopes to improve current practice is by leveraging location data generated from mobile phones. Historically, to pick a representative example, transportation models have been informed by very sparse and expensive datasets. That’s starting to change. Mobile phones send location signals and high rates of mobile phone adoption means signals are sent every second all across urban areas. These signals can be anonymized to protect consumer privacy and then made useful to urban planners. The resulting very large data, combined with the maturity of machine learning techniques developed in recent years, can lead to models that are informed with fresher, cheaper, and more precise data than ever before.
Efforts to improve models using these advances are already underway. Firms such as AirSage, Teralytics, and Streetlight Data have introduced mobile location data to the transportation planning field. Researchers and practitioners are now using location-based data to test the usefulness of machine learning techniques to urban planning problems. Their work inspires us, as does the decades of progress made in the transportation planning and travel modeling communities.
Mobile location data, paired with machine learning techniques, provide the opportunity to train and deploy models much, much faster and inform policy conversations. As the pace of transportation change accelerates, the lag time between observing a behavior and having it inform planning decisions via modeling tools can make policymakers less nimble and less effective. Right now, planners must wait years to simulate the efficacy of, say, expanding a bike-share system. Our goal is to reduce this “latency” towards zero: observing, learning, and deploying can happen together. And when it does, virtual A/B testing in the urban planning space may be as effective as A/B testing in the technology space.
The leveraging of mobile location data is only valuable if we can clearly and cogently communicate what we are doing with that data. Model Lab will strive to build transparent and accessible planning tools. Transparent in that anyone can look under the hood and see what’s going on — from interested residents to modeling experts. Accessible in that anyone can engage with the tool and participate in the conversation.
Solving problems faster
To think about how advanced models can help residents and decision-makers more efficiently improve quality of life in cities, imagine the following scenario.
A city is considering whether to dedicate a travel lane on a major roadway to transit vehicles. Officials hold a public workshop to brainstorm potential ideas: transit lanes on the curbside of the road or on the median side; transit lanes for half the corridor or for the full corridor or none at all; transit lanes all day or only during rush hour; transit plus high-occupancy vehicle lanes or transit plus right turn lanes. Local residents at the meeting break into small groups and explore how these alternatives perform with a model.
Many of the people have engaged in this type of exercise before and are eager to get started. The model provides immediate feedback on how each alternative performs in terms of traffic congestion, transit travel times, transit mode share, greenhouse gas emissions, and other measures. The small groups tinker with their ideas, seeking to optimize the measures they feel are most important — informing their group discussion. The modeling tool used in the workshop is the same one that will be used throughout the project’s life. This consistency allows people to engage at the same level as professional planners do.
Having simulated their alternatives, the small groups reconvene and discuss whether the model results met their expectations. The city planners encourage the participants to be skeptical of the model results and question the assumptions driving the simulation. Models are only as good as the information we put into them, and important assumptions are often contentious. The modeling software allows the groups to adjust certain assumptions about behavior or future land developments. After leaving the meeting, residents are encouraged to try out other ideas and alternatives with the model, which are accessible online.
Today, the planning work described above can take months, if not years. If cities can reduce this time to weeks and extract good ideas from the community, we can create a future in which residents are more engaged and governments are more nimble, responsive, and effective.
If you’re an urban planner excited about how models may improve your work, we’d love to hear from you. And if you’re a technologist interested in helping us advance our vision of better models, please reach out to learn more about joining our team.