Schelling Segregation
Thomas Schelling placed pennies and dimes on a checkerboard. Each coin was “happy” if at least a third of its neighbors were the same type — a mild preference, not hostility to difference. When unhappy coins moved to random empty squares, the board segregated almost completely. Mild individual preferences produced extreme collective separation.
The model doesn’t require anyone to want segregation. It emerges from aggregated choices, each reasonable in isolation. “I’d like at least some neighbors who share my background” sounds modest. But when everyone acts on this preference, homogeneous clusters form, boundaries sharpen, and the integrated middle empties out. Nobody designed the outcome; nobody can easily undo it.
Schelling segregation generalizes beyond race and neighborhoods. Companies segregate by culture as people leave environments where they don’t fit. Online communities segregate by viewpoint as the moderate middle finds both sides uncomfortable. Markets segregate by quality tier as goods cluster with similar goods. Mild sorting preferences create starkly sorted worlds.
The mechanism is feedback loops. Early clustering makes the cluster more attractive to similar types and less attractive to different ones. Each move reinforces the pattern, which triggers more moves, which reinforces further. The equilibrium is self-sustaining even if the initial conditions were random.
The insight has uncomfortable implications. You can’t infer individual preferences from collective outcomes. A segregated city doesn’t prove its residents are hostile to integration — it’s consistent with mild preferences plus dynamics. Conversely, tolerant individuals don’t guarantee integrated outcomes. The gap between intention and result can be vast.
Interventions must address dynamics, not just preferences. Making people more tolerant helps, but if threshold dynamics remain, segregation can persist. Breaking up clusters, creating mixing opportunities, and changing the choice architecture may matter more than changing hearts.
Related: emergence, feedback loops, threshold models, second order effects, path dependence