Nine times out of ten, someone helps

Cameras caught 219 real street conflicts in three cities. Someone stepped in 91 percent of the time, and more bystanders made help likelier, not rarer.

Researchers collected surveillance footage of 219 real fights and assaults from the streets of Amsterdam, Cape Town and Lancaster. Not staged emergencies. Not undergraduates in a psychology building being told that smoke is filling the room. Actual conflicts, recorded by cameras that had no idea they were producing data.

In roughly 91 percent of them, at least one bystander stepped in. On average, 3.76 people did.

That number is worth sitting with, because most of us carry the opposite belief around. Ask someone whether a stranger would help them in the street and you tend to get a shrug, a remark about cities, something about how nobody wants to get involved anymore.

The footage disagrees.

The finding that inverts the folklore

There is a second result in the same study that is harder to swallow than the first. The more bystanders were present, the more likely it was that somebody acted.

That runs backwards to what most of us were taught. The bystander effect — more witnesses, less help, everyone assuming someone else will handle it — is one of the few pieces of social psychology that escaped the seminar room and became general knowledge. It turns up in textbooks, in talks, in the safety briefing at work.

And it is not wrong. Darley and Latané's experiments have held up: an individual is measurably less likely to act when others are around. Diffusion of responsibility is real.

But that answers a question about individuals, and it is not the question a person on the ground is asking. Nobody lying on the pavement wants the per-capita probability of assistance. They want to know whether help is coming. Those are different questions, and for fifty years the field mostly measured the first one.

Add more people and each one becomes individually less likely to act. Add more people and the odds that at least one of them acts go up anyway. Both are true. The second is the one that matters if it is your face on the pavement.

The story that was never true

The whole field grew out of one night in Queens in 1964, and the received version of that night does not survive contact with the record.

In 2007 Rachel Manning, Mark Levine and Alan Collins went back to the trial transcripts and the archive material, and checked the three claims the story rests on: that there were 38 witnesses, that they watched the murder, and that they did nothing. Their conclusion was that there is no evidence for any of the three.

Some of what they found is almost mundane. There were two attacks, not the three the newspaper described. One witness, Robert Mozer, shouted at the attacker and scared him off — that part is in the trial record, corroborated by a second witness. After that first attack Kitty Genovese got to her feet and walked, slowly and unsteadily, around the corner of the building toward her apartment entrance, which is precisely when she went out of sight of the people at their windows. The second, fatal attack happened inside part of a building where almost nobody could have seen it. None of the witnesses reported seeing the stabbing. A neighbour named Sophie Farrar, telephoned about what was happening, called the police and then ran to Genovese. She was still alive when the police arrived.

Even the number is unstable. The famous headline in the New York Times read "37 Who Saw Murder Didn't Call the Police." The article underneath it said thirty-eight.

The three of them call the story a parable — retold for its moral rather than its accuracy, and taught uncritically for decades to students who went on to write the textbooks.

Mark Levine's name is on that 2007 paper. It is also on the 2020 CCTV study. Thirteen years apart, the same researcher took the founding anecdote apart and then went out and measured what actually happens.

What it does not prove

I would rather not oversell this. The cameras were mounted in city-centre entertainment districts, and the incidents were largely loud, visible, night-time conflicts — the kind of emergency that is obvious to everybody at once. That is a specific situation, and it is the one most favourable to intervention.

It tells you nothing about someone having a stroke alone on a quiet street at four in the morning. Nothing about who gets helped and who does not, which is its own uncomfortable question. Nothing about the internet, where the audience is enormous and the cost of scrolling past is zero.

So: not a law of human nature. A measurement, taken in three cities, of one kind of trouble. It is still the largest such measurement anyone has made from real footage, and it points the other way from what we assumed.

The better question

The researchers suggest dropping "why don't people help?" and asking "what makes intervention work?" instead. That strikes me as the more useful reframing, and not only for psychologists.

If nine times in ten somebody already steps forward, the problem was never a shortage of willing people. It is that willingness stays invisible until the moment it is needed. Nobody walking down a street knows that three of the people around them would help.

Which is more or less the premise this whole site is built on.

Sources: Philpot, R., Liebst, L. S., Levine, M., Bernasco, W., & Lindegaard, M. R. (2020). "Would I be helped? Cross-national CCTV footage shows that intervention is the norm in public conflicts." American Psychologist, 75(1), 66–75. Manning, R., Levine, M., & Collins, A. (2007). "The Kitty Genovese murder and the social psychology of helping: the parable of the 38 witnesses." American Psychologist, 62(6), 555–562. Checked 8 September 2026.