Psychology
Agent-Based Modeling of Urban Segregation Dynamics
Quick fact
In the classic Schelling model, even when residents are perfectly content to live in neighborhoods where up to 70% of their neighbors are of a different group, the outcome is still a city with striking segregation.
Why this is interesting
Have you ever wondered why so many cities are deeply divided by race or income, even when most people say they'd be happy to live in diverse neighborhoods? What if the culprit isn't hatred, but a surprising side effect of individual preferences?
Read the full explanation
Understanding Agent-Based Modeling of Urban Segregation Dynamics
Imagine a city as a checkerboard of cells, each holding one household. Each household has a preference: they want at least a certain number of their neighbors to be similar to them, say 30%. If they find themselves with too few similar neighbors, they move to a random empty cell. This is an agent-based model (ABM): a simulation where many individual 'agents' follow simple rules, and we watch what pattern emerges. At first, the city might be randomly mixed. But after a few rounds of moves, clusters of similar households begin to appear. The surprising part: the final segregation is far more extreme than the individual tolerance would suggest. The mechanism is a feedback loop. When a few households move because they are uncomfortable, their old neighborhoods become less diverse for those left behind, which triggers more moves, and the cascade continues. This is a classic example of emergence, where the macro pattern is not designed by anyone but arises from micro-interactions.
A deeper explanation
The core of the Schelling model lies in the concept of a 'tolerance threshold'. Each agent has a threshold: the minimum acceptable fraction of similar neighbors. The rule is simple: if the current fraction of similar neighbors falls below the threshold, the agent moves to an empty spot that satisfies the threshold. The critical insight is that this process produces a tipping point. Because the city is a spatial grid, moving reduces similarity in the neighborhood abandoned and increases it in the destination, effectively 'poisoning' one neighborhood and 'enriching' another. This creates a self-reinforcing dynamic that amplifies small initial imbalances. The model works with arbitrary group labels, so it applies to race, class, or even language. It matters because it explains why integration is so hard to achieve even with tolerant people. It also demonstrates why agent-based modeling is powerful: it lets us explore non-linear, path-dependent processes that are hard to capture with traditional equations. It shows how macro patterns can be unintended consequences of local behavior, a principle that extends to traffic jams, crowd dynamics, and political polarization.