The city’s Republican mayor was elected on a platform to govern by data. Now, he's deploying volunteer analysts to execute on his vision.
Politicians have an agenda; data doesn’t. That’s the ethos behind Tulsa Mayor G. T. Bynum’s effort to shape the Oklahoma city’s policies using statistics collected by a team of his own citizens.
Bynum, a Republican, ran on this platform under the shadow of the 2016 presidential election—against an incumbent whose campaign characterized Bynum as liberal in contrast to the standing mayor’s strong conservative vision. “What if, instead of responding with partisanship, we responded with a focus on results?” Bynum recalled thinking, in a TED Talk a year later. “And so we decided to respond not with a negative ad but with something people find even sexier—data points.”
It would be data, he vowed, that would guide his decision-making process, not partisan beliefs. But after he was elected, Bynum realized he didn’t have the budget to hire the data team to execute on his campaign promises. So while some other cities employ full-time data staffs, Bynum sought out volunteers.
Called the Urban Data Pioneers, this group of government workers and community data buffs are counting the city’s potholes and abandoned buildings; timing the city’s traffic lights; and measuring the population’s stability and growth. Then, they’re mapping their findings. After more than a year of executing 10-week projects, they’re starting to identify city trends, and getting results: After analyzing data on the relationship between education and income, for example, they increased federal financial aid sign-ups in the city by 10 percent.
“My principal interest in it was from the standpoint of improving efficiency and performance,” said Bynum. “What I only appreciated about it as our campaign went on was that it was a platform by which you're able to pull people out of their normal kind of philosophical differences and predispositions, and focus on the tangible reality in front of them.”
When Bynum initially announced the group’s birth on social media and sent out a City Hall-wide email, he expected 15 people to show up. But the first meeting attracted 60 participants, and since, the program’s ranks have grown to 120. Of the approximately 50 active members, about two-thirds of them are government employees, and the rest are community members.
“I know there are a lot of employees who maybe in that traditional structure of things got kind of lost in the bureaucracy—kind of just a cog in this giant structure that we have here,” said Bynum. “And all of a sudden they are part of a small team that's focused on fixing a very specific problem. They have agency: They are empowered in a way that they weren't before.”
Mike Dougherty, who recently started a job in water utilities for the city, is a self-proclaimed data nerd. He spends three to five hours a week working with the Urban Data Pioneers, he says, developing his skills, and forging a greater connection to the city. “It’s been a lot of fun,” he said. “It’s helped me grow as an employee, but it’s also kind of cool to see something you do every day can help you answer these big nebulous questions.”
The approach is novel because it seeks volunteers outside of the formal structure of a data team or full-time staff. But it comes with many of the same concerns as other cities’ efforts to focus more on data: Citizen data collectors, like staff ones, bring their unconscious and conscious biases to the table. And, though Bynum sees a numbers-driven policy approach as less divisive than a politically motivated one, data is hardly non-partisan: Someone has to decide what information is collected or what hypothesis to test, and once data is collected or analyzed, someone has to decide how it’s used.
Those top-level goals were initially determined by Bynum with James Wagner, chief of performance strategy and innovation for the city, who then oversees each cohort’s 10-week process. Later iterations let teams pitch projects based on available data, and this summer, they’re planning on announcing an open call for project suggestions from the general public. While many of the team members are experienced in the field of data analysis or statistics, each team is partnered with a subject matter expert, and hooked up to a group chat to share notes. At the end of each 10-week project, the teams give a presentation, complete with data visualizations and top-line findings.
What’s clear from the first three project cycles is that even cases that tested largely intuitive correlations have yielded more precise information. By measuring what was driving differences in per-capita incomes between city tracts, for example, one cohort found that the determining factor was the level of higher education attained in each block. Common knowledge tells us that more education leads to better jobs which leads to higher incomes, but the Urban Data Pioneers, working in collaboration with a community-based data collection non-profit, Tulsa Data Science, Inc., were able to back that up with regression models, and break the designation of “higher ed” down into levels. They found that higher levels of education were the most important factor leading to a higher per-capita income in a Tulsa city tract, beating out other factors analyzed in a vacuum like racial makeup and family structure.
That doesn’t mean those other factors should be ignored. But for Bynum there was one clear takeaway: Get more people to and through college. Nudged by the data, the city found that very few Tulsa high schoolers were filling out federal financial aid forms in their senior year of high school, which meant they were missing out on an Oklahoma program that pays for four years of college—but only for eligible seniors who have all their documents submitted. Millions of dollars in education support was being left on the table, to the long-term detriment of the city. Along with the Tulsa Chamber of Commerce, Bynum led a drive to get more people signed up. They’ve since increased the percentage of high schoolers applying by more than 10 percent.
“I don't know if the sense of urgency on that would have been as great if we hadn't seen this data proven out,” said Bynum. “That college is an area we need to be focused on every bit as much as high school graduation rates.”
Equally illuminating were the findings in reverse, however: The higher the proportion of students that had dropped out of high school in each city tract, the lower the per capita income. Steve Green, who worked on the project with Tulsa Data Science and the Urban Data Pioneers, has since moved out of Tulsa, but at the time was working for Tulsa Public Schools. Research he had done there on the associations between truancy, missing school, and cutting school with eventual drop-outs dovetailed nicely with the UDP’s findings, and he urged the mayor to implement his suggestions. “That, to me, was the coolest part of all of it: that I was able to connect these two things that I had been working on,” said Green.
This kind of synergy is another benefit to the pioneer program, in Bynam’s eyes. Forging partnerships between different sectors of the government—the police force, and the code enforcement folks, and the meter men—isn’t something that happens every day. “The data pioneers really helped crack the shell on this,” Bynum said. “We’re moving away from the historically bureaucratic, siloed structure that is very formal and not necessarily very efficient.”
Identifying ways to grow Tulsa’s population and make it a more “stable place to live” was another city priority, so volunteers were tasked with finding a way to predict moves in and out of the city, and eventually, cut down on them. Using more than 15 years of water utility billing records—which show, for example, how many water main turn-ons and turn-offs there were during each billing cycle—they made a predictive regression model, and found that the biggest factor leading to move-outs were collection actions, like late fees. Other factors that discouraged “neighborhood stability” were property abandonment, foreclosure, and crime; and those that promoted neighborhood stability were high home ownership rate, low crime rate, and sometimes, type of housing.
Weaving these findings into a policy agenda is more complicated. Dougherty, who worked on the stability project, sees potential for using the models to encourage the city to invest in certain neighborhoods. “Maybe you see that property values are declining, while the neighborhood itself is fairly stable,” he said. “That would be a good area for a new development.” That impulse, shaped by data, is certainly not devoid of partisan implications—targeting specific neighborhoods for economic revitalization and leaving others behind; driving gentrification in some while leaving others intact.
But the data collection isn’t over yet, and they’re hoping to drill down on a more granular level before taking policy action. As it stands now, a lot of the information lacks nuance. The word “move-outs,” for example, obscures the fact that often, the moves they’re recording are forced: In many cases, unpaid water bills lead to forced evictions, or preemptive, informal evictions to avoid them. Developing city-sponsored water assistance, or more progressive foreclosure policies, could be another way to address the perceived instability driven by poverty.
Measuring the relationship between blight and violent crime could be another sticky case if left to subjectivity. Using data collected by the Tulsa Police Department and Love Your Block, a grant-funded neighborhood revitalization program, Urban Data Pioneers is testing questions like: Does blight lead to violent crime?
As any data scientist will tell you, correlation definitely doesn’t equal causation. And as research has proven, when community members are tasked with reporting “unsafe areas” via 311 calls or apps like Findit, Fixit, people of color and the homeless are often policed for simply living their lives. For this reason, data-driven algorithms intended to predict crime has been criticized for perpetuating cycles of racist policing.
“We’ve been able to show at least between reported blight and reported crime—of course we know there's unreported crime, too—but that there are some spatial correlations there,” said Wagner. “In terms of causation, though, we never really made that case.” Instead of using the data to justify increased policing, the action items outlined are more focused on predicting and avoiding blight in Tulsa, and “reducing the likelihood that a property will become so deteriorated that the only option is demolition.”
Most of these analyses have been executed using existing city data, but introducing tools for the public to get involved in collecting it is next. “They didn't have a full picture of where blight was happening in the city because it's only collected if it's reported by somebody,” said Wagner. “So they’re now working on the strategy to engage the public—people can go out and collect blight data by answering 12 questions on an app.”
Tulsa still “isn’t exactly a data science hot spot,” laughed Green. But the city is making strides in not only harnessing the community’s existing skills, but developing them.“This is about creating agency and empowering people with skills,” said Wagner. ”We really want that skillset to be infused throughout all the departments in the city.”
After winning the Engaged Cities Award from Bloomberg Philanthropies’ Cities of Service program, Tulsa is starting a new data science training module called Level Up, which Wagner hopes will lure more people into civilian data collection; and Tulsa Data Science is working outside city bureaucracy to hold trainings and internships for local data enthusiasts. The city is also planning to use the grant to educate other cities about the program. Already, Fort Collins, Colorado is hoping to pilot the Tulsa model, Wagner said.
Funding from Esri was provided to support our project, “Data City."