communicoupling · concept 11 of 26
Two crowds, identical in their average sentiment — one erupts into a riot, the other goes quietly home. Granovetter's threshold model shows why: collective behaviour is decided not by the average feeling but by the exact distribution of individual tipping points, and a single stubborn holdout can flip the outcome.
We explain riots, fads, bank runs, and standing ovations by how people felt on average — the crowd was angry, the mood was ripe. It is the natural story: enough fuel, and the fire catches. But Granovetter built two crowds with the same average radicalism that behave completely differently, because what matters is not how many are willing to act but at what point each one will: how many others they need to see go first.
Each person carries a private number — a threshold, the size of the crowd already acting that would be enough to pull them in too. The zealot has a threshold of zero; they throw the first stone alone. The cautious need to see half the square move before they will. Line these numbers up and the whole outcome can hinge on a single person's — the one missing link in a chain of dominoes that would otherwise have run to the end.
The distribution, not the average
A crowd's fate is not read off its mean. It is computed, step by step, as a chain of threshold-crossings. Start with the instigators — the threshold-zero few who will act with no one beside them. Their action raises the count of people acting from zero to a handful. Now anyone whose threshold that handful satisfies joins in, which raises the count again, which satisfies the next tier of thresholds, and so on. The cascade runs until it reaches a step where no one new is willing — the equilibrium, where it stops.
The mean tells you nothing about whether this chain completes, because completion depends on there being no gaps in the sequence of tipping points. The two headline presets prove it: they have byte-for-byte the same average threshold, the same number of people, the same spread of radicalism. In one, every rung of the ladder is occupied, so the count climbs one step at a time all the way to the top. In the other, one rung is empty — the person who would have joined at that exact moment is instead standing one notch higher — and the climb stops dead the instant it reaches the missing rung. Same fuel, opposite fire. The structure that decides the outcome is invisible in any average, any headcount of the willing, any poll of the mood.
What to try
Load The crowd that ignites, then …that goes home. Watch the mean tile: it does not move. Watch the equilibrium: it collapses from everyone to almost no one. The average predicted nothing.
On the igniting crowd, drag the bar at threshold 1 down to zero and drop that person one bar to the right. You have added no radicalism — you have opened a domino gap, and the whole cascade stalls.
On a stalled crowd, read the detector: it names the single empty rung. Hit Lower one holdout into the gap and one changed tipping point carries the count from a handful to the whole square.
The fragility of ignition
Because the cascade advances only as fast as each rung is filled, the difference between a spark that catches and one that dies is often a single threshold. Remove the instigators entirely — the All talk, no spark preset — and a crowd of people who are each nearly ready to riot does absolutely nothing, because no one is willing to be first. Restore one threshold-zero person and the same crowd can run away completely. The count that matters is not how many are angry; it is whether the exact sequence of tipping points forms an unbroken staircase from the first mover to the last.
This is why ignition is fragile in a way that averages hide. A gap of one missing threshold does not slow the cascade down — it halts it, permanently, one step short of where the momentum would have carried it. And the person who fills that gap need not be a radical; they need only carry the tipping point that happens to sit at the frontier. The critical individual is not the most committed person in the crowd. They are whoever occupies the empty rung — ordinary, interchangeable in their feelings, decisive only in their timing. Which is exactly why no one can point to them in advance.
Reading collective action
The model reaches wherever joining depends on how many have already joined. A riot is the pure case: each person's willingness to smash a window rises with the number of windows already smashed. A bank run is a threshold cascade with money at stake — you withdraw when enough others have, and the enough-others is self-fulfilling. Product adoption, viral fads, a standing ovation that either sweeps the hall or dies in the third row, the decision to strike or to whistleblow when it is safe only in numbers — all of them turn on distributions of tipping points, not on how the room feels on average.
So "they were all just angry" is a category error. It names the fuel and ignores the structure that decides whether the fuel ever catches — and it is wrong symmetrically. It explains too much: two identically angry crowds behave differently, and anger cannot tell them apart. And it explains too little: a crowd that riots was rarely angrier than one that stayed home; it was arranged differently, with its tipping points forming a chain instead of leaving a gap. The mood is the alibi. The distribution is the mechanism.
You pull your deposit when enough others have. The threshold to run falls as the queue grows, and a solvent bank can fail on nothing but the shape of everyone's patience.
Whether a hall rises depends on the few zero-threshold enthusiasts near the front, and on whether the timid tiers behind them form an unbroken ladder up to a full house.
Speaking is safe only in numbers, so silence holds until one low-threshold voice makes the second cheaper, then the third — or until the missing first voice keeps everyone mute.
The mapping
| Mechanism | Social life |
|---|---|
| a threshold | How many others a person needs to see act before they will join. |
| the instigators | The threshold-zero few who start anything, acting with no one beside them. |
| the distribution | The hidden structure that decides the outcome — invisible in any average. |
| the cascade | The chain of threshold-crossings unrolling in time, each crossing enabling the next. |
| the domino gap | A missing threshold that breaks the chain — momentum halted one step short. |
| the equilibrium | Where the cascade stops: the count at which no one new is willing. |
Where it tears
Thresholds are almost impossible to know before the fact. The model reconstructs why a cascade ran or stalled with total clarity once the outcome is in — but the critical individual is invisible until they act, and their empty rung looks like every other quiet person in the crowd. Explaining collective behaviour is not the same as forecasting it, and this model is far better at the first.
Here a threshold is a bare number: any thousand strangers acting are as good as any other. Real thresholds shift with identity — you will join a march the moment your friends do, and never mind a stadium of people you scorn. Weighting neighbours by trust, status, or similarity (try the only neighbours mode for a first taste) changes ignition completely; a raw headcount is the crudest possible version of what moves us.
Showing that a riot turned on a distribution rather than a mood is not a verdict on the riot. Thresholds describe the mechanics of joining — who goes when — and are silent on whether the cause was just, the target deserving, or the act defensible. The mechanism explains how people are pulled in; it does not tell you whether they should have been, and reading it as an excuse mistakes an account of the how for a judgement of the whether.