St. Louis City public schools · A system dynamics model
The city's public schools lost more than 5,000 students between 2015 and 2021, and the pressures behind that loss have not eased. What might it take to turn it around?
To look at that question, we built a working simulation of the city's public school system, district and charter together. It lets us try policy options out on the simulation rather than on real children, and see which ones change the direction of travel.
It covers one part of a much bigger picture. That is enough, we hope, to be worth arguing with. Below is the short version. The second part hands you the controls.
There is no painless way to get more money to each student. In the model, consolidating about 60 of 100 buildings does roughly what doubling state revenue would. Both routes end in the same place, but they are nothing alike for the families involved.
Closing schools is not an accounting move. Closures fall hardest on neighborhoods that have already lost the most, in the places where a school may be one of the last public buildings still open, and a shuttered school can leave a large vacant site behind. The simulation cannot see any of that. It counts what a building costs to run, and nothing else.
It also cannot see the cost of leaving things as they are. The baseline is not a neutral hold; it is continued decline, year after year, landing on the same neighborhoods and on the children whose families have the fewest alternatives. Both paths carry a cost, and only one of them requires anybody to decide.
Much of the decline starts outside the schools. Fewer families live in the city. From there a chain takes over: money follows students out, so less reaches those who stay, results slip, and more families leave. In the model that chain makes the same fall about five times bigger.
Funding is tied to enrollment, so a student who leaves takes their share of the money with them. The costs that do not shrink, like the building, the heating and the front office, are then spread across fewer children, so less reaches each one. Results slip, parents notice, and those who can look elsewhere do.
Switching that chain off in the model changes the picture sharply: enrollment still falls, but by under a tenth instead of well over a third. The population change comes from outside the schools. The amplification does not, which makes it the part there is any hold over.
No single intervention looks like enough. Everything together does about twice what the strongest one alone does. And the model cannot see what closing a school costs a neighborhood.
Run on its own, the strongest intervention lifts educational attainment by about a third by 2030. Run together, the whole set lifts it by about two thirds, which is more than the individual effects add up to, because several of them feed the same loops.
That is an argument for combinations rather than for any one policy. It is not an argument that every intervention is worth doing, which is a judgement about people and places that sits outside what a model of this kind can weigh.
Families in the city describe a cycle: they look elsewhere for schooling, which leaves fewer dollars to meet the needs of the students who stay, which drives teacher turnover and burnout, which lowers confidence in the system further.
The simulation is one way of thinking about that cycle. It draws on 96 interviews with parents, caregivers, teachers and school staff in St. Louis City, on a survey of families, and on state statistics from Missouri. It is the work of Skip Designed, a social design organization that puts people with lived experience and practice first.
Share of the 78 parents and caregivers interviewed. Interviews were semi-structured and coded using a grounded theory approach.
It shows how the pieces move together over a decade, and roughly how effective one policy is next to another.
It is not a forecast. It averages every school into one, and it covers district and charter together rather than either alone. It was built on 2023/24 data.
Read the direction, not the decimal place. If the simulation says enrollment falls by 38%, do not plan around 38%. What is worth taking seriously is that it falls rather than levels off, that it falls faster over time rather than steadily, and that one policy moves it further than another. Those patterns hold even when the exact figures are off.
It stops in 2030 on purpose. Every year a simulation runs, small errors compound and the assumptions behind it age, so the further out it goes the less the numbers are worth. Six years is far enough to see a trend turn and near enough that the structure still resembles the system it was drawn from.
The next part hands you the controls. You can switch policies on and off, change how big or how fast they are, and watch what happens to enrollment, funding and results. There is also a map of how the parts of the system push on each other, which is where these numbers come from.
Change what we assumed and see whether the story still holds. This is the one place a policy can be tried without a real child living through it.
Pick a ready-made scenario or switch on your own combination of interventions. Pin a run to keep it on the charts while you change something else. Tap or hover any policy name or chart title for a definition.
Arrows show cause and effect; a sign says whether the two move together (+) or in opposite directions (−). Boxed variables are stocks, and the pipes with valves are the flows that fill and drain them. Pick a loop to trace it.
The map is wider than the screen. Drag it sideways to follow a loop round, or pinch to zoom in on a label.
Layout and naming follow the CoModel version of this diagram. The simulation behind it also carries a safety loop, where rising student need lowers perceived safety and that lowers trust. It is live in the simulator but not drawn here.