Bystander effect

Why is a person in trouble less likely to be helped when more people are watching?

Common sense says a crowd is a resource: more people, more hands, more chance that someone competent is among them. The laboratory finding runs the other way — add bystanders and the odds that any given one acts go down. Before asking why, it is worth asking what exactly is being claimed, because the story most people know about this effect is largely false, and the correction changes what needs explaining.

Take the famous case first. The 1964 murder of Kitty Genovese is remembered through a New York Times report that thirty-eight witnesses watched the attack and did nothing. That account does not survive scrutiny. Later work — including a 2007 review by Manning, Levine and Collins — found the number unsupported, most neighbours heard fragments rather than watching an attack, the layout meant few had a view at all, at least one shouted at the attacker, the police were called, and a neighbour came down and held her as she died. So the founding anecdote is not evidence of anything. What follows rests on experiments, not on it.

Those experiments are real. Darley and Latané staged a seizure over an intercom in 1968: participants who believed they alone could hear it reported it 85% of the time, dropping to about 31% when they believed four others could hear it too. In a companion study, smoke seeped into a waiting room; three-quarters of lone participants reported it, but with two calm confederates present, only about one in ten did. Same event, same person, different number of witnesses.

Now, why? The temptation is to reach for one cause. Resist it — these two studies already point at different things. In the seizure, each participant was alone in a cubicle and could not see the others; whatever suppressed action was not about being watched or about reading faces. In the smoke room, participants could see the others were unbothered.

Pull the mechanisms apart. The first is diffusion of responsibility: if five people can act, the felt obligation on each is roughly a fifth, and the shame of not acting is shared out too. This needs only the belief that others are present, which is why it works through an intercom.

The second is pluralistic ignorance: nobody knows if this is an emergency, so each looks at the others for a reading. But everyone else is also looking, and looking calm while they do — so each takes everyone else's composure as evidence there is nothing wrong. A room of privately worried people collectively concludes that all is fine. This requires visibility, not numbers as such.

The third is audience inhibition: acting means risking looking foolish in front of witnesses if you have misread the situation. This one scales with how observable you are, and it is why the effect weakens when the emergency is unambiguous — nobody looks foolish helping someone who is obviously bleeding.

Does the effect hold outside a laboratory? Here the picture complicates in a useful way. Philpot and colleagues in 2019 analysed CCTV footage of real public conflicts in three cities and found intervention in around nine cases in ten — and, contrary to the simple prediction, more bystanders slightly raised the chance that someone stepped in. That is not a refutation of the lab work; it measures a different quantity. The lab asks whether this particular bystander acts, and the answer is still less often in a group. The street asks whether anyone acts, and with enough people, someone usually does.

Latané and Darley framed helping as a chain of decisions rather than a single choice: notice the event, interpret it as an emergency, take responsibility, know what to do, and act. That framing is what lets the three mechanisms sit in different places instead of competing — pluralistic ignorance blocks the interpreting step, diffusion of responsibility the taking-responsibility step, audience inhibition the final acting step. A crowd can break the chain in any of three places, and which one it breaks changes what would fix it.

Two honest qualifications. The effect is real in the laboratory and has been confirmed by meta-analysis, but a 2011 meta-analysis by Fischer and colleagues found it shrinks, and can reverse, when the situation is dangerous or unambiguous — plausibly because clarity removes both pluralistic ignorance and the fear of embarrassment, while danger makes a lone bystander seek allies. And the popular framing, that crowds make people callous, is not what any of this shows. The participants in these studies were not indifferent; those who did not report the seizure were visibly agitated. They were stuck, not cold.

• Whether pluralistic ignorance can be broken simply by one person saying out loud that they are unsure.
• How the CCTV evidence should change what bystander-intervention training actually teaches.

Every term the collection defines is gathered in the glossary.