Crowd crush

Why does a crowd in which everyone pushes to leave faster end up leaving more slowly?

The familiar story is panic: something frightens a crowd, everyone bolts for the exit, and in the stampede people are trampled. It has an obvious moral — keep your head, and you will get out. But before accepting the story, notice how much of it is assumption. Were the people who died actually running? Were they trying to leave at all? And does pushing harder really move you out faster?

Each of those turns out to be doubtful, and correcting them is most of the explanation.

Start with the deaths. In the crowd disasters that have been investigated in detail, the cause of death is overwhelmingly compressive asphyxia — the chest cannot expand against the surrounding pressure — rather than trampling. That single forensic fact rules out the stampede picture, because compressive asphyxia does not require anyone to be running or to fall. People have died standing upright, held vertical by the bodies around them.

So what produces the pressure? Density. Below about two people per square metre you walk where you like. As density climbs you lose first your preferred speed, then your preferred direction, and somewhere above roughly five per square metre you lose the ability to move your own body at all: you go where the mass goes. British safety guidance for standing spectator areas caps density at 47 people per ten square metres for exactly this reason — it is an administrative limit chosen with a margin, not a physical threshold, and the transition it guards against is gradual.

Once individual locomotion has stopped, the crowd stops being a collection of people and starts behaving as a compressible medium. Push at the back and the push does not stay at the back; it travels. Analysis of video from the 2006 Jamarat Bridge disaster during the Hajj showed the crowd moving first in stop-and-go waves and then into a regime the researchers called crowd turbulence, in which people are thrown several metres sideways by forces they did not generate and cannot resist. Nobody in that footage is panicking. Most of them cannot see anything wrong.

Now the question as posed. Why should pushing harder empty a space more slowly? The cleanest answer comes from the physics of a bottleneck. When many bodies converge on a gap wider than one person but narrower than the flow feeding it, they can lock into a stable arch across the opening — the same arching that jams grain in a hopper — and each additional unit of force presses the arch tighter rather than through. Simulations of escape behaviour published in 2000 found precisely this: as the desired speed of the agents rose past a point, the flow through the exit fell. It has been called the faster-is-slower effect, and it has been reproduced in granular experiments and in controlled pedestrian trials.

Two honesties about that result. It is a model finding first and a field observation second; the original paper framed it as "escape panic", and crowd scientists have criticised that framing hard, since the psychological premise of irrational selfish flight is largely absent from real disasters even where the clogging physics is present. And arching at a doorway is only one contributor — in the worst events the fatal density is not at the exit at all.

Which brings us to the mechanism that actually kills. The people generating the force are usually nowhere near the people being crushed, and cannot see them. A crowd is opaque: your information horizon is a few metres of shoulders. So the front compresses against an obstruction while the back, experiencing an ordinary pleasant press, keeps supplying pressure — and the pressure accumulates, because a crowd transmits force but does not transmit the news that force is doing harm. Shouting does not carry. The feedback loop that would normally stop you hurting someone is severed.

Notice that this makes the question's premise partly wrong. The crowd need not be trying to leave: Hillsborough in 1989 was a crush at an entrance, and the Itaewon crush of 2022 happened in a sloped alley where two flows met. This is a story about density and force transmission, not about motive.

The useful correction is that "crowd panic" inverts cause and effect. Disorder is not the input that produces the crush; the crush is a physical outcome of density and geometry, and any disorder tends to be a response to it. Investigations repeatedly find helping behaviour, not selfishness, right up to the point where people can no longer control their own bodies.

That reframing changes what counts as a fix. If panic were the cause, the remedy would be crowd discipline — announcements, exhortation, calm. If density is the cause, the remedy is arithmetic done in advance: limit how many people enter a space, count them, design against convergent flows and against gaps that narrow in the direction of travel, and give the back of the crowd a way to learn what the front knows.

One caveat. The relative weight of arching, of upstream pressure waves, and of local geometry differs from disaster to disaster, and reconstructing any particular event is contested work done from partial video. The general mechanism is solid; the apportionment in any given case is argued over for years.

• Why counterflow and merging flows are so much more dangerous than a single dense stream.
• How granular physics came to be the working model for pedestrian bottlenecks, and where the analogy fails.

Every term the collection defines is gathered in the glossary.