Researchers Designed an Algorithm to Save Schools Money and Improve Equity. The District Loved it. Then Things Got…
Skip to content
Social Impact Policy Oct 1, 2020

Researchers Designed an Algorithm to Save Schools Money and Improve Equity. The District Loved it. Then Things Got Messy.

A tale of bus routes in Boston shows the promises and pitfalls of using new technology to change entrenched systems.

School children exit bus

Lisa Röper

Based on the research of

Dimitris Bertsimas

Arthur Delarue

Sébastien Martin

As a researcher, Sébastien Martin develops algorithms to tackle extremely complex problems in the world of transportation logistics. It’s something he feels passionately about.

He knows not everyone matches his level of enthusiasm. “I’m this young professor, and the only thing I know how to talk about well is transportation optimization,” he jokes, “which is maybe not the subject to talk about with random people.”

Yet, two years ago, one of Martin’s algorithms was the talk of the town in Boston, generating a huge public controversy and protests that eventually led the mayor to reverse course on a major school-policy change.

At issue was whether the Boston Public Schools system should change the start times of the majority of its schools. The goal was threefold: save money because new start times would allow school buses to more easily travel to multiple schools each morning and afternoon; give high schoolers later start times, which research shows is beneficial to their health and learning; and make sure that start times were equitably distributed across neighborhoods and socioeconomic groups.

“In many districts, especially in cities, transportation is an important tool we have to guarantee some level of equity.”

— Sébastien Martin

The district’s proposal was based on an algorithm developed by Martin and colleagues. An initial algorithm looked only at the question of transportation costs and was able to show the district how to remove 50 busses from its fleet. After that successful project, the researchers tackled the question of start times.

In 2017, the Boston School Committee unanimously approved their recommended plan. And that’s when things got crazy.

For Martin, this experience demonstrates the complexity of using algorithms to help make policy decisions. Algorithms can suggest potential answers to difficult problems, but ultimately these decisions are made by humans. This is crucial, since the real world is always messier than the well-defined problems algorithms can solve.

Yet, he’s no less bullish on algorithms’ power.

“I’m convinced algorithms can help a lot,” says Martin, who is an assistant professor of operations at the Kellogg School. “But you don’t have to use an algorithm that decides the school start time. That’s not even what we were doing in Boston. The algorithm tries billions and billions of different combinations and gives examples of the best ones. Then it’s the humans who decide. But the algorithm will have shown them solutions they would have never imagined.”

While the Boston plan failed, Martin sees an opportunity to change start times, and thus improve student health, equity, and transportation efficiency in many districts, for an unlikely reason: COVID-19. Given that school schedules have been turned upside down, he says, it may be easier to make additional changes to the school day, so that, when we emerge from the pandemic, start times have been improved.

“We had to break down the status quo because teaching the same way is not feasible right now,” he says. “So we have an amazing opportunity to start anew.”

A Massively Complex Problem

On the list of school issues that parents worry about—safety, teacher quality, extracurricular opportunities—the logistics of bus routes likely ranks low. Yet the impact of transportation plans is immense.

Buses gobble up a large chunk of district budgets. In Boston, transportation costs make up about 11 percent of the district’s billion dollar annual budget. This is money that schools would happily spend on teachers or nurses or social workers instead. Furthermore, district buses are often how low-income students are able to travel to high-performing schools.

“In many districts, especially in cities, transportation is an important tool we have to guarantee some level of equity,” Martin says.

So the stakes are high. And the logistics are mind-numbingly complex.

How far is it reasonable to expect students to walk to a bus stop? At the same time, if you have lots of stops to create short walking times, then your busses stop constantly. What’s the maximum amount of time students can spend on a bus before it becomes onerous? Yet if routes are too short, then costs balloon. How do you ensure that wheelchair-accessible buses get to the right kids? And, for efficiency’s sake, how do you arrange for the same bus to do multiple routes each morning and afternoon? Then, of course, once you’ve found the best possible solution that takes into account all these factors, you have to start over next fall, as students leave and join the system, or move to a different home or new school.

Across the country, many districts use specialized software to set their bus routes. But the problem is so complicated that it can require months of additional manual work. And in the end, it is hard to know if the new routes are the best option. So the Boston School District decided to change its process.

It launched its Transportation Challenge in 2017. Martin and Arthur Delarue, both PhD students at MIT at the time, were intrigued. So, along with their advisor, Dimitris Bertsimas, they accepted the challenge and a new research project was born.

“These problems are extremely complicated,” Martin says. “Finding the perfect solution is completely impossible, even with the best possible computers. But, still, by being clever, we can find very good solutions.”

Tackling the Complex Problem

The team’s algorithm ended up winning the competition.

The researchers’ solution involved breaking from the prevailing system for setting bus routes. In many districts, Boston included, each school determines its optimal routes, and then those best options are combined into the district-wide plan. Instead, the algorithm looked at each school in isolation and generated a few good options for each one; it then sought to knit different permutations of those routes together into the most cost-effective, district-wide solution.

“Sometimes the solution that’s best for one school is not necessarily best for the overall transportation system,” Martin says. For example, using three buses instead of two may feel costlier to the individual school, but if those three routes can be more easily combined with other schools’ routes, there’s a greater overall savings.

“Even if you use more buses for one specific school, in the end your costs are actually much lower,” he says.

The researchers’ solution let the Boston School District remove 50 buses from its fleet of about 650 in 2017 without requiring the students to spend more time on the bus. This meant an annual savings of about $5 million. Boston continues to use the algorithm for its school bus routes each year.

Looking at School Start Times

Yet, the district knew it could achieve more of its objectives by adjusting something that the first algorithm didn’t touch: school start times. With this additional factor on the table, bus routes could be more easily linked. Which is why, as a second part of the Transportation Challenge, the district asked researchers to see what they could do if start times were in play.

“We showed clear correlations between the level of wealth in a neighborhood and how good their start times are.”

— Sébastien Martin

The idea of optimizing transportation costs by changing start times is not new. In fact, the reason so many districts, like Boston, have staggered start times in the first place is so that buses can do multiple routes each morning. But, over time, this led to two problems.

The first is a growing inequity among start times at schools in Boston’s higher- and lower-income neighborhoods.

Most parents prefer start times between 7:45 and 8:45 a.m. Yet, Martin says, “we showed clear correlations between the level of wealth in a neighborhood and how good their start times are.”

He doesn’t think these inequalities are there on purpose. Changing a school start time significantly affects the lives of the families, so these changes don’t happen often. And when they do, they stay that way for decades. Without an algorithm that can understand the complex ramifications of start times on transportation and equity, negative consequences can slowly build up over the years.

Secondly, medical research shows that high school students’ brains are changing in ways that make it difficult for them to get enough sleep if school starts before 8:30 a.m. And less sleep can lead to decreased academic achievement and increased rates of obesity and depression. Yet a CDC report found that the vast majority of American high schools start before 8:30 a.m.

The district reasoned that by changing start times, it could save money and address equity and public health issues at the same time.

Martin, Delarue, and Bertsimas’s solution combined new start times for 85 percent of schools with the ability to cut 120 buses from the district fleet. The Boston School Committee unanimously approved the proposal in December 2017, with the intention of changing school start times for the following academic year.

That never happened.

Many parents were outraged at the proposal. Some schools were slated to change start times by nearly two hours. Parents who packed a school-district meeting that January complained that the new times would be disruptive to their families, and that children would need to wake up far too early, and thus go to bed too early to have family time in the evenings.

While, according to the Boston Globe, most of the families that packed this meeting appeared to be white, there was also push back among Black and brown parents, as well as from advocacy groups like the NAACP. They argued that parents who work lower-wage jobs often cannot easily accommodate changes to their families’ schedules, and that very early or late start times would necessitate more before- or after-school childcare.

Ultimately, an hour before a planned protest in front of City Hall, the mayor called off the change.

Martin doesn’t fault any of the parents for advocating for their families.

“In the end, you fight for your kids,” he said. “Obviously, you need to make your voice heard. That’s the basis of democracy.”

Continued Optimization

While Martin has sympathy for the parents who voiced their concerns—and great admiration for the school officials caught in the crossfire—he wishes the entire project hadn’t been scuttled.

“There were many things that were not perfect,” he says. “But the question is whether the current system is better than this.” And, he adds, while changes like this are always hard, they are necessary. “High schoolers’ health is at stake, and the equity of start-time assignments could easily be improved. At the same time, tens of millions of dollars could be saved from transportation and reinvested in the children’s education.”

Martin still has high hopes for his team’s algorithm. Indeed, after the Boston Globe covered the start-time controversy, a number of other districts reached out to the researchers for help. They ended up forming a startup to help districts with their optimization needs. Additionally, much of the bus-route algorithm is open source, so Martin assumes there are districts using it without the researchers necessarily knowing.

Today, as so many school districts adjust to education in the time of COVID-19, bus routes and schools’ start times are likely being rethought, too. Martin sees this as an opportunity for districts to revisit the bigger questions around start times. Given that the status quo has already been disrupted for most schools and families, why not try to land on a healthier and more equitable system of start times for schools post-pandemic?

“If we are already paying the price of change, why not try to make things a little better for our children,” he says.

Beyond COVID, Martin is gratified to see that this research continues to be used for more long-term projects. School transportation and especially school start times are multifaceted issues, which means they create an opacity that can hide inequities and inefficiencies. It’s easy for one school to say that it wants to push back its start time by 30 minutes or to go from three busses to two. But it’s nearly impossible, without the help of an algorithm, to see what those changes will do to neighboring schools or overall budgets.

“Anyone can argue for something, and no one really knows what the bigger consequences are,” he says. “That’s something that algorithms can do. They can cast light on some very opaque matter and help us make more informed decisions. They can be very useful tools, but we are still responsible for making the right decision.”

About the Writer
Emily Stone is the senior editor at Kellogg Insight.
About the Research
Bertsimas, Dimitris, Arthur Delarue, and Sebastien Martin. 2019. “Optimizing Schools’ Start Time and Bus Routes.” PNAS. 116(13): 5943–5948.

Read the original

More in Social Impact