Escalation

Why do disputes so often grow far past the point either side originally wanted?

Look at the end of almost any spiralled dispute — a border incident, a strike, a family feud over an estate — and ask each party whether, at the outset, they would have chosen this. Nobody says yes. That is a peculiar fact about a process made entirely of deliberate choices. Nobody chose the destination, yet somebody chose every step. What kind of mechanism has that property?

Take the first move seriously. It is usually modest and meant to end the dispute cheaply — a show of resolve intended to make the other side conclude that continuing is not worth it. That is not aggression for its own sake; it is an attempt at a short path. So notice something immediately: the escalatory step is chosen because it looks like the way to avoid a long fight.

Now think about how it lands on the other side. They did not see the reasoning, only the act. To them it reads as an unprovoked move that has to be answered, and their answer is likewise modest and likewise meant to settle the matter. Each side is now responding to the other's last move, and each experiences itself as reacting while the other initiates. Run that loop a few turns and both are further up than either intended, with each move locally sensible.

But reciprocity alone would let a dispute come back down as easily as it went up. Why is coming down so much harder? Because of what the steps do to the cost of stopping. This is Thomas Schelling's insight, and it inverts the intuition. In a contest of resolve, the way to prevail is to become visibly unable to back down — to burn the bridge behind you, so the other side sees that yielding is the only remaining move on the board. Commitment is not stubbornness. It is the deliberate destruction of your own options because that is what makes a threat believable.

Ask what happens when both sides use that technique. Each step removes one side's exit, and the other side must then either yield or take a step that removes its own. The ladder is built by two parties who are each rationally trying to end the climb. What began as a dispute over the original stake becomes a dispute over which side has to be seen to lose, and that second stake grows with every rung.

Three things do the ratcheting. Audience costs are the sharpest: a leader who says publicly that they will not yield has recruited an audience that will punish them for yielding, which is exactly why the statement was made — and it means the exit is now barred by one's own supporters. Prior investment adds a second pull, familiar from ordinary sunk cost, though the two are worth keeping apart: with sunk cost the past is genuinely irrelevant and treating it as relevant is an error, whereas here the past really has changed the present situation, because reputation and audience persist. The third is mechanical — mobilisation timetables, standing orders, delegated authority, alert states that trigger the other side's alert states. Schelling called the residue of this the threat that leaves something to chance: nobody has to decide on the war for the risk of it to rise.

So how does anything ever come down? Not by one side being persuaded it was wrong. It comes down when the cost of stopping is lowered, which is a different task from persuasion. Hence the standard instruments: a private channel where a concession costs no audience; a third party who can be blamed for the compromise; a formula that lets each side describe the outcome as a success on its own terms; small unilateral steps that invite reciprocation without conceding the dispute. Salient limits help too — a line both sides recognise without negotiating it, like the one between conventional and chemical weapons, gives a place to stop that neither has to propose.

I would not overstate how well this is understood. That crises spiral is not seriously disputed, but how much weight to put on misperception, on deliberate commitment tactics, and on domestic politics is argued case by case, and the same episode is routinely read all three ways.

The strongest correction to make is to the word "irrational". Escalation does not require anyone to lose their head. It is generated by ordinary, defensible steps whose combined effect nobody chose, which is why it survives sober people, professional staffs, and full awareness of the pattern.

A second refinement: the mechanism is symmetrical in structure but need not be symmetrical in cause. One side may be escalating deliberately, as a bargaining technique, while the other escalates reflexively because it reads the move as aggression — and from inside, those look identical.

And a third: de-escalation is a cost problem, not a belief problem. Asking whether the other side has been convinced is asking the wrong question. The question is whether they have somewhere to step that does not humiliate them — and that is usually something you can build for them rather than something they can find alone.

• How tacit limits are recognised in a crisis without either side proposing them.
• Why third-party mediators can settle disputes the parties could have settled themselves.

Every term the collection defines is gathered in the glossary.