Schelling's Segregation Model

The Setup

Place two groups on a grid. Each cell is occupied by a member of group A, a member of group B, or left empty. Every occupied cell looks at its eight surrounding neighbors and asks a single question: "Are enough of my neighbors like me?"

"Enough" is defined by a threshold, a fraction of similar neighbors that satisfies the agent. If the answer is yes, the agent stays. If the answer is no, the agent relocates to a random empty cell.

That is the entire model. There is no hostility, no strategic calculation, no awareness of the global pattern. Each agent acts on a local, modest preference.

What Happens

Set the threshold at 33%: each agent is content so long as roughly one in three neighbors shares its type. That sounds tolerant. Run the simulation and watch.

Within a few dozen steps the grid reorganizes itself into large, homogeneous clusters. Segregation emerges not from anyone demanding it but from the cumulative effect of mild, local dissatisfaction. Agents on the boundary of a cluster are the most likely to be unhappy, so they move, which shifts the boundary, which makes new agents unhappy, which triggers further movement. The process feeds itself until it reaches an equilibrium that looks nothing like the preferences that produced it.

Raise the threshold to 50% and the segregation is sharper, faster, more complete. Lower it to 20% and there is still visible clustering, though neighborhoods stay more mixed. The striking result is the nonlinear relationship: a small increase in individual preference produces a disproportionately large increase in collective segregation.

Why It Matters

Schelling's model is one of the earliest and clearest demonstrations of emergence in the social sciences. Three lessons survive fifty years of subsequent research.

Intentions do not predict outcomes. Every agent in the model is individually tolerant. The segregated outcome is not what anyone wanted. Policy that focuses solely on individual attitudes will miss the structural dynamics that produce collective results.

Tipping points are real. The relationship between preference and segregation is not proportional. There are threshold values beyond which the system flips rapidly from mixed to segregated. This makes the problem both harder (small changes in attitudes can trigger large shifts) and more hopeful (small structural interventions at the right point can prevent the cascade).

Simple rules, complex outcomes. The model uses no sophisticated decision-making, no communication between agents, no memory. The complexity lives in the interaction, not in the individuals. This insight, that macro phenomena can arise from micro rules without any central direction, is foundational to agent-based modeling and complexity science.

Limitations and Extensions

The original model is a grid with random relocation. Real cities have geography, housing costs, social networks, and institutional forces. Researchers have extended the model with heterogeneous preferences, network topologies, income constraints, and historical path dependence. These extensions generally reinforce the core finding: mild preferences plus decentralized movement produce clustering. They also reveal that the details of the relocation mechanism (where agents can move to, and in what order) matter as much as the preference threshold.

Clark and Fossett (2008) showed that when agents can choose from a menu of neighborhoods ranked by desirability, segregation outcomes depend heavily on whether desirability correlates with group composition. Pancs and Vriend (2007) demonstrated that even agents who strictly prefer 50/50 integration can produce full segregation, because the dynamics of sequential moves do not converge to the individually preferred state.

The Deeper Question

Schelling's model asks: when a pattern exists in the world, is it because someone chose it? The answer, disturbingly often, is no. Segregation, inequality, polarization, and market bubbles can all emerge from choices that look rational and moderate at the individual level. Understanding the gap between micro-motives and macro-behavior is not optional for anyone who wants to understand how societies organize themselves.