← back to essays/

Schelling's Segregation Model

2006-10-02

In 1971, economist Thomas Schelling asked a simple question: what happens if people have a mild preference for living near others who are similar to them? Not a strong preference, not hostility toward anyone different, just a slight comfort with familiarity.

His answer was surprising and, decades later, remains one of the most important insights in social science: even very mild individual preferences can produce extreme collective segregation.

The Rules

The model is disarmingly simple:

That's it. No agent dislikes the other group. No agent is trying to create a segregated neighborhood. Each one simply wants some of their neighbors to look like them.

Watch It Happen

The slider below controls the preference threshold, the minimum percentage of similar neighbors each agent wants. Start at 30% (a very mild preference) and watch what emerges:

33%
40×40
15%
0
Steps
--
Unhappy agents
--
Avg % similar neighbors
--
Segregation index
Click "Initialize" to place agents randomly, then "Run" to watch them sort themselves.

The Surprise

At a 33% threshold, each agent is happy as long as one-third of their neighbors are like them. They're fine being in the minority. And yet the grid quickly sorts itself into distinct clusters. At 50%, the segregation is stark. At 70%, it's nearly total.

No agent wanted this outcome. No agent intended to create homogeneous neighborhoods. Each one just wanted to not be completely surrounded by the other group. But as unhappy agents at the borders move away, they shift the composition for those left behind, triggering more moves, in a cascade that produces far more segregation than any individual preferred.

Compare Thresholds

Click below to run the model at several thresholds simultaneously and compare the results after 100 steps:

0% preference
--
25% preference
--
33% preference
--
50% preference
--
75% preference
--

Why This Matters Beyond Housing

Schelling's model isn't just about neighborhoods. It reveals a general principle: individual preferences that seem harmless and tolerant can produce collective outcomes that look extreme and intentional.

The Transferable Insight

Schelling's model teaches something that's easy to say but hard to internalize: you cannot infer individual intentions from collective outcomes. A segregated neighborhood doesn't necessarily mean residents are hostile. A homogeneous team doesn't necessarily mean the hiring manager is biased. An echo chamber doesn't necessarily mean its members are closed-minded.

The system-level pattern can be far more extreme than any individual's preference. This matters because it changes where you look for solutions. If the problem is structural, emerging from the dynamics of many mild preferences interacting, then addressing it requires changing the structure (incentives, defaults, rules), not just individual attitudes.

The habit worth building: when you see an extreme outcome, resist the urge to assume it required extreme intentions. Ask instead what mild preferences, interacting at scale, could have produced it.