In 1971, Thomas Schelling showed that a city of individually tolerant people can still end up deeply divided. The model requires no bigotry — just a mild preference for familiar neighbors.
scroll to begin
Imagine a city laid out on a grid. Each cell is a household — either orange or blue, with some cells left vacant. The two types are randomly placed, roughly 50-50.
Right now, the city is well-integrated.
Each household has a single rule: at least one-third of my neighbors should be like me.
This is an extremely tolerant threshold. It means accepting being in the minority two-thirds of the time. It does not demand a majority — not even close.
White outlines mark the households whose one-third condition isn't met. In the random starting arrangement, a small fraction of households are unhappy — surrounded by too many of the other type.
Unhappy households relocate to a random vacant cell. A single round of moves. The city looks nearly the same, but the dynamics have already started.
Some moves satisfied one household but made a neighbor unhappy.
Each round of moves creates new unhappiness at the margins. Departures make source neighborhoods less diverse, triggering further departures. Schelling called this a tipping process.
Within a few dozen rounds, everyone is satisfied. No household wants to move. But look at the map — the city has sorted itself into large, nearly homogeneous clusters.
Each household only asked for 33% similar neighbors. What they got is a neighborhood that's typically 70–80% similar. The collective outcome vastly overshoots the individual preference.
Nobody wanted this. The system produced it anyway.
Raise the threshold to 50% — each household wants to be in the majority, not even a strong majority — and the segregation becomes nearly total. Entire quadrants go single-color.
Individual preferences do not aggregate linearly. A city of tolerant individuals can produce an intolerant outcome. The micro motive and the macro behavior are qualitatively different.
Desegregation cannot be achieved by changing attitudes alone. It requires changing the structure of the system.
Schelling's model contains no racists. No household demands homogeneity. A preference for one-third same-type neighbors sounds not just tolerant but generous. Yet the collective result is a city divided into blocks that are overwhelmingly uniform.
The mechanism is self-reinforcing. When one household leaves, the neighborhood becomes less diverse, which may push another household past its threshold. Each departure tips the balance further.
Each line tracks the average same-type neighbor fraction over time. All start near 50% (random) and converge to 70–80% (segregated).
25 independent simulations. 30×30 grid, 10% vacant, threshold 33%.
The relationship between individual preference and collective segregation is not linear. Below a critical threshold (around 30%), the city stays well-mixed. Above it, segregation jumps sharply. This is a phase transition — the same kind of abrupt shift found in magnetism, epidemics, and percolation.
Each dot is one simulation. The mean (line) reveals the sharp transition: mild preferences produce radical outcomes.
12 simulations at each threshold. 20×20 grid, 10% vacant. Segregation measured as average same-type neighbor fraction.
Schelling published his model in 1971, originally working it out with coins on a checkerboard. It has since been used to study residential segregation, political polarization, language clustering, and the formation of social cliques in schools and workplaces.
The core lesson persists: in any system with even mild sorting preferences, small individual biases amplify into large collective patterns. Integration is not the natural equilibrium — it has to be actively maintained.
Adjust the tolerance threshold and watch the city evolve. Even small changes in individual preference produce dramatically different collective outcomes.