Building an O.R. framework to drive school district policies

A nationwide bus driver shortage is currently stifling U.S. school systems. Facilitating access to school is an important mission of school districts, and most manage an expensive fleet of buses (the “yellow buses”), either owned or outsourced, to transport their students. School transportation is already extremely costly for school districts. In the city of Boston alone, more than $120 million is spent annually on the school transportation system, which is 10% of the public education budget. Schools are now struggling to find bus drivers, which significantly impacts school operations. According to a recent survey by a school transportation company, 78% of respondents from public school districts flagged bus driver shortage as a restriction, and 81% reported that COVID-19 has exacerbated this issue [1]. Pandemic-related safety protocols have reduced effective bus capacity, requiring increases in the resources needed to transport the same number of students to school. Limited budgets hinder the hiring process for bus drivers. Some districts have even asked teachers to drive buses before school or paid families to opt out of bus services [2].

John Hanlon, CEO of the school transportation company AlphaRoute and former COO of Boston Public Schools, is currently implementing optimization solutions for dozens of school districts. He highlights the high potential for operations research (O.R.) to help school districts in these challenging times:

School districts and public agencies in general are known for continually rolling out status quo approaches to old and new problems alike. When I think of the value that sophisticated operations research can provide, I doubt that there is a sector in the world that is more ripe for this innovation than the public sector. The pandemic is causing a staffing shortage in so many different areas of public school administration that novel approaches are required to manage the lack of resources. We are seeing this very clearly with the driver shortage. Schools are trying to hire more drivers, and they are hiring them, but they’re losing them just as fast. Using new techniques to optimize fleets and use fewer routes will help even more. School systems could become more efficient if they leveraged developments in operations research, and they absolutely shoulds.

The O.R. community agrees with Hanlon: the school bus routing and scheduling problem (SBSP) is one of the oldest and most studied routing problems, with more than 100 papers published over the last 60 years (see reviews in [3, 4]). The goal of the SBSP is generally to design models and algorithms that can improve the bus routes and the driver schedules to limit the number of buses and drivers needed, as well as limit the amount of time children have to spend on the bus. This can be formulated as an optimization problem. Students in each school district are first assigned to bus stops; these bus stops are then connected into bus routes going to each school. Schools can vary their start and end times to allow each bus driver to serve several routes every morning and afternoon, and their schedules are optimized to maximize their utilization. Most of the work in the optimization literature has therefore focused on improving these tactical and operational decisions to reduce transportation costs given districts’ resources and constraints. Each school system has its unique transportation characteristics, and a general trend in the SBSP literature is the growing complexity that is needed to capture these idiosyncrasies. Yet, as Hanlon argues, many school systems are far from implementing state-of-the-art solutions.

The Recipe for Successful Implementation of Academic Work

Partnerships between researchers and school districts are a great way for the O.R. community to have a positive impact on society while doing innovative research. The authors have partnered with Tyler Maybee, program manager on the transportation team for Denver Public School (DPS). Following an INFORMS Annual Meeting, he mentions how his attendance – and our subsequent partnership – were beneficial to the school district:

The INFORMS Annual Meeting is a great opportunity to learn about current research in the field of operations. Many of the presentations provide valuable insight, which I was able to take back to Denver and work on applying creative solutions to long-existing problems within our operations groups. Additionally, it was great to meet Dr. Smilowitz and create a partnership between Northwestern and Denver Public Schools around operations research, especially pertaining to our Transportation Department. Creating the partnership with Northwestern has been invaluable in helping DPS create deeper analysis and better approaches around operational issues such as bell times. Collaborating with academics allows the ability for DPS to continually challenge the mentality of ‘this is how it has always been done.’

Maybee’s story is not unusual, and many school districts are partnering with academics, although implementations are still rare. A recent example of a success story is the work of Bertsimas et al. [5], which resulted in the successful implementation of a novel large-scale bus scheduling algorithm that enabled Boston Public Schools to save $5 million annually. During the pandemic, Arthur Delarue, Zhen Lian and Sebastien Martin also created an optimization framework for the City of San Francisco to reduce transportation costs and help high school students get much-needed rest by changing the start times of schools. This change was implemented in fall 2021 and is saving millions of dollars in transportation costs. Smaller school districts near universities have also drawn on resources from researchers, such as the Northwestern University and Evanston/Skokie School District 65 partnership.

If these projects can be successful, why are such partnerships still the exception rather than the rule? First, researchers need to put in extra work toward implementation and navigate the complex political landscape of school systems. Researchers must also remember that optimization is not the answer to every problem, but just a piece of the puzzle. Maybee says, “The optimal solution often does not take into consideration the individual needs of the pieces involved; in our scenario the pieces are schools. Schools have leaders with their own agendas and priorities that might conflict with the most optimal solution. Once one piece is changed, the solution has the potential to unravel quickly if not managed well.”

He adds, “The largest complexity is certainly the reaction of parents to any policy decision, which dictates the school reaction, and the school reaction for the large part can dictate how many voices in central administration react. Administrators, school board members and parents will accept known inefficiencies and higher costs to ensure students have the best education opportunities available to them. Finding that balance between fiscal responsibility and putting students first is the tightrope public school systems must try and walk every day.”

As researchers, we find that doing the extra step and learning the powerful interplay with optimization practitioners is completely worth it. It leads to new research ideas and gives a lot of meaning to our work. Most school districts are eager to collaborate and could indeed benefit from the knowledge of the O.R. community. As noted in Smilowitz and Keppler [6], a best practice for maintaining a good relationship is defining a tangible deliverable and focusing on solving interesting technical problems. But what is needed to foster more partnerships between operations research and schools? Hanlon says that school systems should take the first step:

I don’t know if it’s fair to say that the research community could do more on its own. I think school systems need to work more closely with the research community to develop new solutions in support of district operations. They already do this quite well on the academic side, partnering with universities on curricular, attendance and assessment projects. They just need to direct some of this same energy to the operations side, bringing in researchers to try to rethink the status quo. When I worked at Boston Public Schools with a team from MIT, they transformed how we thought of transportation. There must be other opportunities for work like that to take place in public systems.

A potential avenue to increase awareness is for operations researchers to tackle the problems that are the most important to school districts. Next, we provide an example of such a problem, on which we think our community could have an enormous impact.

School Transportation and Policy: Beyond Traditional Optimization

Much of the literature on school transportation is focused on optimization. Indeed, an efficient way to limit the growing transportation costs of school districts is to carefully optimize the transportation operations. At the same time, school districts constantly face policy decisions that inevitably affect the transportation system. Should they open a new school or close an existing one? Should they allow district-wide schools or favor neighborhood schools? Should they update the school choice policy where applicable? Given the weight of the transportation budget, it is not possible to move forward with any policy without understanding its consequences on transportation. While implementing techniques found in the school transportation literature has been particularly successful at reducing transportation costs, it is still not well understood how these optimized transportation costs are affected by policy changes. In short, transportation optimization is often treated as a “black box” by school districts, which makes it difficult to make appropriate policy decisions. Hanlon explains how school districts struggle with this issue and how his company is trying to help:        

Most people I talk to at school transportation departments across the country feel like they are not part of the decision-making process. The policy decisions are just passed on to transportation, and it’s on them to figure out the aftermath. Essentially, policy decisions are made in a vacuum without the ability to understand the consequences on transportation. This guides our work with AlphaRoute, as we find that school districts make better decisions when they can evaluate the impact on transportation and various trade-offs, before those policies are established. An example is the matter of school choice – how students are assigned to school. Too many school boards do not fully appreciate the consequences of these policies on transportation, but they are important. We can build models to allow school boards to make sure that their policies will not over-burden transportation, while still allowing for important political concerns to be met as well. This is vital, especially amidst a massive driver shortage.

It is clear that optimizing transportation systems is not enough. Opening the “black box” of transportation costs and enabling districts to make transportation-aware policy decisions would surely bridge the gap between operations research and school operations.

Building on our work with school districts and the individuals supporting these districts, such as Maybee and Hanlon, we are designing a framework to understand the complex interactions between school policies and transportation costs. Using a general model of school districts and their enrollment patterns, we simulate a variety of districts over time to explore the impact of various policy changes. We then build on the broad base of O.R. literature to implement state-of-the-art school transportation algorithms, tuned to the key characteristics of the school districts, and use them to estimate the transportation costs. This ultimately enables us to provide policymakers with insights into the relationship between the education policies under study and transportation costs. We believe that such a framework can allow the work of the O.R. community to reach more districts. This can then support the nation’s public school districts as they face the many challenges ahead.

Acknowledgments

Financial support from the National Science Foundation [Grant CMMI-1727744] is gratefully acknowledged.

References

1. HopSkipDrive, “State of School Transportation 2021 Report,” https://www.hopskipdrive.com/school-transportation-2021/.
2. Crockett, 2021, “Why America has a school bus driver shortage,” https://thehustle.co/why-america-has-a-school-bus-driver-shortage/.
3. A. Ellegood, S. Solomon, J. North and J. F. Campbell, 2020, “School bus routing problem: Contemporary trends and research directions,” Omega, Vol. 95, p. 102056.
4. Park and B.-I. Kim, 2010, “The school bus routing problem: A review,” European Journal of Operational Research, Vol. 202, No. 2, pp. 311-319.
5. Bertsimas, A. Delarue, W. Eger, J. Hanlon and S. Martin, 2020, “Bus routing optimization helps Boston Public Schools design better policies,” INFORMS Journal on Applied Analytics, Vol. 50, No. 1, https://doi.org/10.1287/inte.2019.1015. 
6. Smilowitz and S. Keppler, 2020, “On the use of operations research and management in public education systems,” TutORials in Operations Research, pp. 84-105.