Schelling segregation

Schelling’s model (1971)

Thomas Schelling was an economist and game theorist who asked a deceptively simple question: what happens if people have only a mild preference for living near others like themselves? Not hostility toward difference — just a gentle pull toward similarity. He placed two types of agents on a grid, gave each a tolerance threshold (the minimum fraction of neighbors that must be the same type), and let unhappy agents relocate to vacant cells. The result was striking. Even when agents were happy with 66% of their neighbors being different — a threshold of just 33% — the grid sorted itself into large, homogeneous clusters. Near-complete segregation emerged from near-complete tolerance. The model was published in the Journal of Mathematical Sociology in 1971, and it has been unsettling comfortable assumptions ever since.

The gap between intention and outcome

This is the core insight, and it runs deeper than housing patterns. Each individual agent in the model is perfectly content in a mixed neighborhood. No agent wants segregation. No agent is even mildly bigoted in any meaningful sense — they simply have a weak preference for not being entirely surrounded by the other type. But when every agent acts on that preference simultaneously, the collective result is a pattern that none of them intended and most of them would find troubling if they could see it. The macro-level pattern is qualitatively different from the micro-level preference. This is the fundamental challenge of collective behavior: individual rationality does not aggregate into collective rationality. The whole is not just more than the sum of its parts — it can be something that no part wanted at all.

Tipping dynamics

Try moving the tolerance slider slowly upward and watching the segregation index. The relationship is not linear. At low thresholds (0–20%), the grid stays well-mixed. At moderate thresholds (30–40%), segregation appears but the system still has some mixing. Then, around 50%, something qualitative happens: the segregation index jumps sharply upward. This is a tipping point — a small change in individual behavior produces a discontinuous change in collective outcomes. The system exhibits hysteresis, too: once segregated, reducing the threshold doesn’t immediately restore mixing. The segregated state is an attractor. These dynamics are closely related to phase transitions in physics and critical phenomena in complex systems.

Real-world resonance

Schelling developed his model in the context of American housing segregation, and its relevance there is obvious. But the mechanism it reveals — mild individual preferences amplified into stark collective patterns — appears everywhere. School sorting, where families with slight preferences for academic homogeneity produce completely stratified schools. Online echo chambers, where a gentle preference for agreeable content produces hermetic ideological bubbles. Language death, where bilingual speakers’ slight preference for the dominant language erodes minority languages to extinction. In each case, the question is the same: how much of the pattern we observe is driven by strong individual preferences, and how much is an emergent artifact of weak ones interacting at scale?

Nobel recognition

In 2005, Schelling shared the Nobel Prize in Economics with Robert Aumann “for having enhanced our understanding of conflict and cooperation through game-theory analysis.” The segregation model was a central part of the citation. It demonstrated something the Nobel committee called “the scope for collective action” — the idea that understanding aggregate outcomes requires understanding the feedback loops between individual decisions and the social environment those decisions create. Schelling’s work stands at the intersection of economics, sociology, and what we would now call complexity science.

Connection to collective intelligence research

This model connects directly to Joshua Becker’s research on collective behavior. The central theme is the same: collective outcomes emerge from individual decisions in ways that no individual intended or desired. In Becker’s work on social influence and collective intelligence, the question is whether group interactions amplify individual wisdom or individual bias. In Schelling’s model, the answer is unambiguous: mild individual preferences get amplified into extreme collective patterns. Understanding when and how this amplification occurs — and when social structure can prevent it — is one of the defining questions of collective behavior research. The network dynamics experiment in this lab explores a related question: how the topology of social connections shapes whether groups converge on truth or drift toward error. And the evolutionary game theory simulation shows spatial structure enabling cooperation — the mirror image of spatial structure enabling segregation here.