The bystander effect: why crowds freeze, and when they actually help

The bystander effect is the finding that a person is less likely to help in an emergency when other people are present than when they are alone. The reason is not apathy. Responsibility spreads across the group, and each person reads the calm of the others as a sign that nothing is wrong.

The effect is real, but two parts of the popular story are wrong. The famous "38 witnesses" of the Kitty Genovese murder did not exist in the form the newspapers described. And in real, dangerous conflicts caught on camera, at least one bystander usually does step in.

• Darley and Latané (1968): 85% of people who thought they alone heard a staged seizure reported it in time. Only 31% did when they thought four others were listening.
• Latané and Darley (1968): 75% of students alone in a room reported smoke coming through a vent, against 10% of those who sat with two people who ignored it.
• Latané and Darley's five-step model says helping can fail at any of five points: notice, interpret, take responsibility, decide how to help, act.
• Three processes drive the effect: diffusion of responsibility, evaluation apprehension and pluralistic ignorance.
• A 2011 meta-analysis found the effect is weaker in dangerous emergencies. A 2020 CCTV study found that someone intervened in 90.9% of real public conflicts.

What is the bystander effect in psychology?

In social psychology, the bystander effect (also called bystander apathy, although apathy is not the cause) describes an individual's behaviour. The more bystanders a person believes are present, the lower the chance that this person will help, and the longer they take to do it.

That is a claim about each individual, not about the crowd as a whole. This difference explains most of the confusion about the effect, and the section on when bystanders do help returns to it.

The Kitty Genovese story, and what the archives show

The research began with a murder. In the early hours of 13 March 1964, Catherine "Kitty" Genovese was attacked and killed near her home in Kew Gardens, Queens. On 27 March, The New York Times reported that 38 neighbours had watched the attack. According to the report, none of them had called the police. The story became a national symbol of urban indifference, and many textbook chapters on helping still open with it.

In 2007, Rachel Manning, Mark Levine and Alan Collins went back to the archive material. In their American Psychologist article, they found no evidence that 38 witnesses were present, no evidence that the witnesses saw the murder, and no evidence that the witnesses did nothing. Few people saw any part of the attack, and most neighbours could not see the place of the fatal assault. At least two people called the police.

The correction matters, but it has limits. The attack and the killing were real, and the help that came was too late. What the archives remove is the parable of 38 silent watchers. They do not remove the laboratory findings, which never rested on the newspaper account.

The two experiments that defined the effect

John Darley and Bibb Latané did not accept the explanation that New Yorkers had stopped caring. They suspected that the number of witnesses was part of the cause, and they designed experiments to test that idea.

The seizure study (1968)

Darley and Latané tested 72 students at New York University, 59 women and 13 men. Each student sat in a separate booth for an intercom discussion about the personal problems of college students in an urban university. Each participant believed that one, two or five other students were on the line. In fact, every other voice was a tape recording. One voice described a tendency to seizures, then later appeared to have a seizure and called for help before going quiet.

Because the participants could not see one another, nobody could copy anyone else's calm. Only the number of other listeners changed. The results, published in the Journal of Personality and Social Psychology, were clear:

Every participant who thought they were alone with the victim reported the seizure in the end. In the six-person condition, only 62% ever did. The participants who did not report were not calm. Many of them had trembling hands and were sweating, and they often asked afterwards whether the victim was all right. They had not decided that the victim did not matter. They had not decided anything.

The smoke study (1968)

The second study tested a different question: what if the other people are in the room? Latané and Darley invited male students at Columbia University to fill in a questionnaire in a waiting room. When a student had completed two pages, white smoke started to come in through a wall vent. It kept coming in puffs until it was difficult to see.

The results depended on who else was in the room:

• Alone: 18 of 24 students (75%) reported the smoke, and the median student did it within 2 minutes.
• With two confederates who ignored the smoke: 1 of 10 students (10%) reported it. The others coughed, rubbed their eyes and waved the smoke away, but kept on with their questionnaires.
• With two other real participants: in only 3 of 8 groups (38%) did anyone report the smoke. If the three students had acted independently, more than 98% of groups should have had at least one reporter.

In the interviews afterwards, the students who did not act gave harmless explanations, such as steam or air-conditioning vapour. Two of them suggested that it was "truth gas" to make them answer honestly. The inaction of the others had changed what the students thought they saw.

The five steps from noticing to helping

Latané and Darley's book The Unresponsive Bystander (1970) put the results into a five-step decision model. Each step is a point where helping can stop, and the presence of other people can make each step more difficult.

The model explains why one cure does not fix every case. A person who never noticed the event needs a different prompt from a person who noticed it and thought that someone else had already called.

Why crowds freeze: three processes

The five steps describe where helping fails. Three processes describe why the presence of others causes the failure.

Diffusion of responsibility

When one person witnesses an emergency, the whole responsibility to act falls on them. When ten people witness it, each person feels a share of it, and a share is easier to ignore. The seizure study isolated this process. The participants could not see or hear each other. The other bystanders changed only the number of people who could have helped. Milgram described a related shift in his studies of authority. Obedience was easier when the responsibility seemed to belong to someone else (see our article on the Milgram obedience experiment).

Evaluation apprehension

People do not want to look foolish. If you run to help and the situation turns out to be harmless, an audience sees your mistake. This fear holds back the last step, the act itself, and it is strongest when the situation is ambiguous.

Pluralistic ignorance

In an ambiguous situation, each person looks at the others to decide what is happening. Everyone keeps a calm face while they decide, so everyone sees a group of calm people and concludes that there is no emergency. The group as a whole holds a belief that no one in it holds privately. The smoke study shows this process. The confederates' indifference made the smoke look harmless, and among three real participants, each student's hesitation became evidence for the other two. Copying the reactions of others is a mental shortcut. Like the shortcuts in our overview of cognitive biases, it is usually useful. Here it fails at the worst time.

When bystanders do help

The laboratory results are often summarised as "people in crowds do not help". Later research shows where that summary breaks down.

The effect shrinks when the danger is clear

In 2011, Peter Fischer and colleagues combined 105 effect sizes from bystander studies published from the 1960s to 2010. The studies had more than 7,700 participants in total. Their meta-analysis in Psychological Bulletin found an overall effect of g = -0.35, a small-to-moderate reduction in helping when others are present.

The size of the effect depended on the situation. It was weaker when the emergency was clearly dangerous, when a perpetrator was present, and when the cost of helping was physical. A clear danger removes the ambiguity that pluralistic ignorance needs. Other bystanders can also give physical support. The effect was weaker when the other bystanders were men or knew each other. It was also weaker when they were real participants instead of passive confederates.

On camera, intervention is the norm

In 2020, Richard Philpot and colleagues coded 219 real public conflicts recorded by surveillance cameras in Amsterdam, Cape Town and Lancaster. At least one bystander intervened in 90.9% of the conflicts, with an average of 3.76 interveners for each incident. The rates were similar in the three cities, although perceived public safety differs a lot between them. Each additional bystander made it slightly more likely, not less, that the victim got help.

This does not disprove the laboratory findings. Fischer's meta-analysis and Philpot's study measure different things. The bystander effect says that each person is less likely to act when others are present. The CCTV study asks whether anyone in the group acts. Both can be true at the same time. Each person in a crowd holds back a little, but a crowd has many people who can step in.

Bystander effect vs related terms

Several terms in this topic are close in meaning, and exam questions often test the difference.

Social loafing is about effort on a shared task, not about helping. The two share a cause: in both cases, one person's contribution feels less necessary when others are present.

How to overcome the bystander effect

The five-step model gives the most useful advice for both sides.

The best-tested advice is to ask one person directly. In a study of 400 chat groups, Patrick Markey found that naming a specific person almost removed the effect of group size and made help come much faster. That study was online. In a physical crowd, the five-step model suggests the same approach: pick one person and give them one task, such as "Call an ambulance." This answers step 2 (it is an emergency), step 3 (it is their job) and step 4 (what to do).

If you are a bystander, treat an ambiguous situation as a question, not an answer. The calm faces around you may belong to people who are also waiting for someone else to move. The model suggests a simple first step: ask the person if they are all right, or tell the group what you see.

For revision, learn the experiments, the five steps and the three processes against each other, as in our guide to memorising psychology terms. Save "bystander effect", "diffusion of responsibility" and "pluralistic ignorance" to a Social psychology list in Psychology World and drill them.