bystander effect experiment purpose

HOLIDAY SAVINGS: Save 30% Sitewide with HOL30 + free shipping on orders over $75.

On The Site

“Laboratory of Experimental Psychology” by Popular Science Monthly Volume, licensed for use on Wikimedia Commons .

Exploring the Bystander Effect

May 1, 2017 | yalepress | Current Affairs , Psychology

Joel E. Dimsdale—

The very public murder of young Kitty Genovese in New York City motivated the next social psychology exploration on the nature of malice. On the night of March 13, 1964, Genovese left work and was walking on a street in Kew Gardens, Queens, when she was chased down and stabbed. The murder was doubly horrific. She didn’t die suddenly; rather, her assailant kept chasing her, stabbing and slashing her over and over again in an attack that lasted more than thirty minutes. She called out in terror: “Please help me! Please help me!” But no one did.

Two investigators, John Darley and Bibb Latané, designed a series of experiments to find out why people do not intervene even in life-threatening situations. If Milgram conducted his studies on the infl uence of authority, looking over his shoulder at Nazi Germany, Darley and Latané wanted to learn what accounts for bystander apathy. Why did no one intervene to save Genovese’s life? Why did so many people stand by and do nothing to save the lives of Nazi victims? It was a topic Hannah Arendt had wrestled with in another context: “Under conditions of terror most people will comply but some people will not. . . . No more is required, and no more can reasonably be asked, for this planet to remain a place fit for human habitation.” Sadly, the number of people who resist complying is small.

To be fair, emergency situations are often sudden, ambiguous, and out of the typical life experience of the beholder. But, as Darley and Latané would learn, there was something about the social environment itself that influences how a person would respond to emergencies. Painstakingly, first at Columbia University and New York University and then at Princeton University, the investigators grappled with this problem in multiple experiments.

Darley and Latané confronted their subjects with a threatening situation and assessed how the subjects would respond. Then, in an ingenious twist, Darley and Latané studied the subject alone or else in small groups. How would a subject respond to an emergency in the presence of other people who appeared to ignore what was happening?

In one experiment, subjects reported to a room to talk about the problems of urban life while smoke poured from the room’s heating register—not just a few puffs, but so much smoke that by the end of the session “vision was obscured in the room by the amount of smoke present.” When subjects waited alone, 75 percent of them quickly reported the smoke, but when they entered a room with two other seated people who studiously ignored the smoke, the subjects’ behavior was strikingly different. Only 10 percent of these subjects reported the smoke even though they “coughed, rubbed their eyes, and opened the window.”

In another experiment, subjects arrived at the lab and were greeted by a receptionist who stood up, drew a curtain, and noisily clambered up on a chair to grab some folders. The receptionist surreptitiously turned on a tape recorder, which played the sound of a loud crash, a scream, and the following script: “Oh, my God, my foot . . . I . . . can’t move . . . it. Oh . . . my ankle, I can’t get this . . . thing . . . off me.” The tape played sounds of weeping and moaning. Here again, the question was simple: Would anyone check on the receptionist, and how long would it take? It may be comforting that 70 percent of the subjects who waited alone checked on the receptionist. It will not be comforting to learn that when the subjects were waiting with a stranger who ignored the receptionist’s plight, only 7 percent of the subjects intervened.

A third experiment revealed even more indifference. In this setting, subjects were ushered into individual cubicles and asked to talk via intercom with other subjects about the problems they experienced in college. One subject, secretly a confederate of the investigators, revealed that in addition to all the usual stresses of college, he was embarrassed because he had a seizure disorder. He then grew increasingly incoherent, saying: “I-er-umI think I-I need-er-if-if could-er-er-somebody er-er-er-er-er-er-er give me a little-er-give me a little help here because-er-I-er-I’m-er-er h-h-having a-a-a real problem-er-right now and I-er-if somebody could help me out it wouldit would-er-er s-s-sure be-sure be good . . . because-er-there-er-er-a cause I-er-I-uh-I’ve got a-a one of the-er sei—-er-er-things coming on and-and-and I could really-er-use some help so if somebody would-er give me a little hhelp-er-uh-uh-uh (choking sounds). . . . I’m gonna die-er-er I’m . . . gonna die-er-help-er-er-seizure-er (chokes, then quiet).”

When subjects were alone in the cubicle, 85 percent got up within a minute to check on the subject who was presumably having a seizure. Subjects were then tested with other people in the same cubicle. If the subject was paired with one person who had been secretly instructed to ignore the situation, 62 percent of the subjects got up to check on the presumably ill student. If however, the subject was tested with four confederates who ignored the apparent seizure, then the chance of the subject responding fell to only 31 percent, and it took such people, on average, three minutes to check on their fellow student.

There have been countless variations on the design, but the inherent message remains the same: in social situations, there is a diffusion of responsibility. If a bystander sees that other witnesses are doing nothing, then he or she will also do nothing—“bystander apathy,” as Darley and Latané put it so pungently.

From Anatomy of Malice by Joel E. Dimsdale , published by Yale University Press in 2016. Reproduced by permission.

Joel E. Dimsdale is distinguished professor emeritus and research professor in the department of psychiatry at the University of California, San Diego. He lives in San Diego, CA.

Further Reading

Recent Posts

• Italian Futurism and Avant-Garde Sculpture
• The Great Transformation: A Conversation with Odd Arne Westad and Jian Chen
• A Conversation with Susan Owens about Drawing and the History of Art
• Trump and His Allies Could Corrupt Future Elections
• Where We Stand: A Conversation with Djamila Ribeiro and Padma Viswanathan
• Formation of a New Language: Chicago, 1943-56

Sign up for updates on new releases and special offers

Newsletter signup, shipping location.

Our website offers shipping to the United States and Canada only. For customers in other countries:

Mexico and South America: Contact W.W. Norton to place your order. All Others: Visit our Yale University Press London website to place your order.

Shipping Updated

Learn more about Schreiben lernen, 2nd Edition, available now. 

Bystander Effect In Psychology

Udochi Emeghara

Research Assistant at Harvard University

B.A., Neuroscience, Harvard University

Udochi Emeghara is a research assistant at the Harvard University Stress and Development Lab.

Learn about our Editorial Process

Saul McLeod, PhD

Editor-in-Chief for Simply Psychology

BSc (Hons) Psychology, MRes, PhD, University of Manchester

Saul McLeod, PhD., is a qualified psychology teacher with over 18 years of experience in further and higher education. He has been published in peer-reviewed journals, including the Journal of Clinical Psychology.

Olivia Guy-Evans, MSc

Associate Editor for Simply Psychology

BSc (Hons) Psychology, MSc Psychology of Education

Olivia Guy-Evans is a writer and associate editor for Simply Psychology. She has previously worked in healthcare and educational sectors.

On This Page:

Take-home Messages

• The bystander effect is a social psychological phenomenon where individuals are less likely to help a victim when others are present. The greater the number of bystanders, the less likely any one of them is to help.
• Factors include diffusion of responsibility and the need to behave in correct and socially acceptable ways.
• The most frequently cited real-life example of the bystander effect regards a young woman called Kitty Genovese , who was murdered in Queens, New York, in 1964 while several of her neighbors looked on. No one intervened until it was too late.
• Notice the event (or in a hurry and not notice).
• Interpret the situation as an emergency (or assume that as others are not acting, it is not an emergency).
• Assume responsibility (or assume that others will do this).
• Know what to do (or not have the skills necessary to help).
• Decide to help (or worry about danger, legislation, embarrassment, etc.).
• Latané and Darley (1970) identified three different psychological processes that might prevent a bystander from helping a person in distress: (i) diffusion of responsibility; (ii) evaluation apprehension (fear of being publically judged); and (iii) pluralistic ignorance (the tendency to rely on the overt reactions of others when defining an ambiguous situation).
• Diffusion of responsibility refers to the tendency to subjectively divide personal responsibility to help by the number of bystanders present. Bystanders are less likely to intervene in emergency situations as the size of the group increases, and they feel less personal responsibility.

What is the bystander effect?

The term bystander effect refers to the tendency for people to be inactive in high-danger situations due to the presence of other bystanders (Darley & Latané, 1968; Latané & Darley, 1968, 1970; Latané & Nida, 1981).

Thus, people tend to help more when alone than in a group.

The implications of this theory have been widely studied by a variety of researchers, but initial interest in this phenomenon arose after the brutal murder of Catherine “Kitty” Genovese in 1964.

Through a series of experiments beginning in the 1960s and 1970s, the bystander effect phenomenon has become more widely understood.

Kitty Genovese

On the morning of March 13, 1964, Kitty Genovese returned to her apartment complex, at 3 am, after finishing her shift at a local bar.

After parking her car in a lot adjacent to her apartment building, she began walking a short distance to the entrance, which was located at the back of the building.

As she walked, she noticed a figure at the far end of the lot. She shifted directions and headed towards a different street, but the man followed and seized her.

As she yelled, neighbors from the apartment building went to the window and watched as he stabbed her. A man from the apartment building yelled down, “Let that girl alone!” (New York Times, 1964).

Following this, the assailant appeared to have left, but once the lights from the apartments turned off, the perpetrator returned and stabbed Kitty Genovese again. Once again, the lights came on, and the windows opened, driving the assaulter away from the scene.

Unfortunately, the assailant returned and stabbed Catherine Genovese for the final time. The first call to the police came in at 3:50 am, and the police arrived in two minutes.

When the neighbors were asked why they did not intervene or call the police earlier, some answers were “I didn”t want to get involved”; “Frankly, we were afraid”; “I was tired. I went back to bed.” (New York Times, 1964).

After this initial report, the case was launched to nationwide attention, with various leaders commenting on the apparent “moral decay” of the country.

In response to these claims, Darley and Latané set out to find an alternative explanation.

Decision Model of Helping

Latané & Darley (1970) formulated a five-stage model to explain why bystanders in emergencies sometimes do and sometimes do not offer help.

At each stage in the model, the answer ‘No’ results in no help being given, while the answer ‘yes’ leads the individual closer to offering help.

However, they argued that helping responses may be inhibited at any stage of the process. For example, the bystander may not notice the situation or the situation may be ambiguous and not readily interpretable as an emergency.

The five stages are:

• The bystander must notice that something is amiss.
• The bystander must define that situation as an emergency.
• The bystander must assess how personally responsible they feel.
• The bystander must decide how best to offer assistance.
• The bystander must act on that decision.

Figure 1. Decision Model of Helping by Latané and Darley (1970).

Why does the bystander effect occur?

Latane´ and Darley (1970) identified three different psychological processes that might interfere with the completion of this sequence.

Diffusion of Responsibility

The first process is a diffusion of responsibility, which refers to the tendency to subjectively divide the personal responsibility to help by the number of bystanders.

Diffusion of responsibility occurs when a duty or task is shared between a group of people instead of only one person.

Whenever there is an emergency situation in which more than one person is present, there is a diffusion of responsibility. There are three ideas that categorize this phenomenon:

• The moral obligation to help does not fall only on one person but the whole group that is witnessing the emergency.
• The blame for not helping can be shared instead of resting on only one person.
• The belief that another bystander in the group will offer help.

Darley and Latané (1968) tested this hypothesis by engineering an emergency situation and measuring how long it took for participants to get help.

College students were ushered into a solitary room under the impression that a conversation centered around learning in a “high-stress, high urban environment” would ensue.

This discussion occurred with “other participants” that were in their own room as well (the other participants were just records playing). Each participant would speak one at a time into a microphone.

After a round of discussion, one of the participants would have a “seizure” in the middle of the discussion; the amount of time that it took the college student to obtain help from the research assistant that was outside of the room was measured. If the student did not get help after six minutes, the experiment was cut off.

Darley and Latané (1968) believed that the more “people” there were in the discussion, the longer it would take subjects to get help.

The results were in line with that hypothesis. The smaller the group, the more likely the “victim” was to receive timely help.

Still, those who did not get help showed signs of nervousness and concern for the victim. The researchers believed that the signs of nervousness highlight that the college student participants were most likely still deciding the best course of action; this contrasts with the leaders of the time who believed inaction was due to indifference.

This experiment showcased the effect of diffusion of responsibility on the bystander effect.

Evaluation Apprehension

The second process is evaluation apprehension, which refers to the fear of being judged by others when acting publicly.

People may also experience evaluation apprehension and fear of losing face in front of other bystanders.

Individuals may feel afraid of being superseded by a superior helper, offering unwanted assistance, or facing the legal consequences of offering inferior and possibly dangerous assistance.

Individuals may decide not to intervene in critical situations if they are afraid of being superseded by a superior helper, offering unwanted assistance, or facing the legal consequences of offering inferior and possibly dangerous assistance.

Pluralistic Ignorance

The third process is pluralistic ignorance, which results from the tendency to rely on the overt reactions of others when defining an ambiguous situation.

Pluralistic ignorance occurs when a person disagrees with a certain type of thinking but believes that everyone else adheres to it and, as a result, follows that line of thinking even though no one believes it.

Deborah A. Prentice cites an example of this. Despite being in a difficult class, students may not raise their hands in response to the lecturer asking for questions.

This is often due to the belief that everyone else understands the material, so for fear of looking inadequate, no one asks clarifying questions.

It is this type of thinking that explains the effect of pluralistic ignorance on the bystander effect. The overarching idea is uncertainty and perception. What separates pluralistic ignorance is the ambiguousness that can define a situation.

If the situation is clear (for the classroom example: someone stating they do not understand), pluralistic ignorance would not apply (since the person knows that someone else agrees with their thinking).

It is the ambiguity and uncertainty which leads to incorrect perceptions that categorize pluralistic ignorance.

Rendsvig (2014) proposes an eleven-step process to explain this phenomenon.

These steps follow the perspective of a bystander (who will be called Bystander A) amidst a group of other bystanders in an emergency situation.

• Bystander A is present in a specific place. Nothing has happened.
• A situation occurs that is ambiguous in nature (it is not certain what has occurred or what the ramifications of the event are), and Bystander A notices it.
• Bystander A believes that this is an emergency situation but is unaware of how the rest of the bystanders perceive the situation.
• A course of action is taken. This could be a few things like charging into the situation or calling the police, but in pluralistic ignorance, Bystander A chooses to understand more about the situation by looking around and taking in the reactions of others.
• As observation takes place, Bystander A is not aware that the other bystanders may be doing the same thing. Thus, when surveying others’ reactions, Bystander A “misperceives” the other bystanders” observation of the situation as purposeful inaction.
• As Bystander A notes the reaction of the others, Bystander A puts the reaction of the other bystanders in context.
• Bystander A then believes that the inaction of others is due to their belief that an emergency situation is not occurring.
• Thus, Bystander A believes that there is an accident but also believes that others do not perceive the situation as an emergency. Bystander A then changes their initial belief.
• Bystander A now believes that there is no emergency.
• Bystander A has another opportunity to help.
• Bystander A chooses not to help because of the belief that there is no emergency.

Pluralistic ignorance operates under the assumption that all the other bystanders are also going through these eleven steps.

Thus, they all choose not to help due to the misperception of others’ reactions to the same situation.

Other Explanations

While these three are the most widely known explanations, there are other theories that could also play a role. One example is a confusion of responsibility.

Confusion of responsibility occurs when a bystander fears that helping could lead others to believe that they are the perpetrator. This fear can cause people to not act in dire situations.

Another example is priming. Priming occurs when a person is given cues that will influence future actions. For example, if a person is given a list of words that are associated with home decor and furniture and then is asked to give a five-letter word, answers like chair or table would be more likely than pasta.

In social situations, Garcia et al. found that simply thinking of being in a group could lead to lower rates of helping in emergency situations. This occurs because groups are often associated with “being lost in a crowd, being deindividuated, and having a lowered sense of personal accountability” (Garcia et al., 2002, p. 845).

Thus, the authors argue that the way a person was primed could also influence their ability to help. These alternate theories highlight the fact that the bystander effect is a complex phenomenon that encompasses a variety of ideologies.

Bystander Experiments

In one of the first experiments of this type, Latané & Darley (1968) asked participants to sit on their own in a room and complete a questionnaire on the pressures of urban life.

Smoke (actually steam) began pouring into the room through a small wall vent. Within two minutes, 50 percent had taken action, and 75 percent had acted within six minutes when the experiment ended.

In groups of three participants, 62 percent carried on working for the entire duration of the experiment.

In interviews afterward, participants reported feeling hesitant about showing anxiety, so they looked to others for signs of anxiety. But since everyone was trying to appear calm, these signs were not evident, and therefore they believed that they must have misinterpreted the situation and redefined it as ‘safe.’

This is a clear example of pluralistic ignorance, which can affect the answer at step 2 of the Latané and Darley decision model above.

Genuine ambiguity can also affect the decision-making process. Shotland and Straw (1976) conducted an interesting experiment that illustrated this.

They hypothesized that people would be less willing to intervene in a situation of domestic violence (where a relationship exists between the two people) than in a situation involving violence involving two strangers. Male participants were shown a staged fight between a man and a woman.

In one condition, the woman screamed, ‘I don’t even know you,’ while in another, she screamed, ‘I don’t even know why I married you.’

Three times as many men intervened in the first condition as in the second condition. Such findings again provide support for the decision model in terms of the decisions made at step 3 in the process.

People are less likely to intervene if they believe that the incident does not require their personal responsibility.

Critical Evaluation

While the bystander effect has become a cemented theory in social psychology, the original account of the murder of Catherine Genovese has been called into question. By casting doubt on the original case, the implications of the Darley and Latané research are also questioned.

Manning et al. (2007) did this through their article “The Kitty Genovese murder and the social psychology of helping, The parable of the 38 witnesses”. By examining the court documents and legal proceedings from the case, the authors found three points that deviate from the traditional story told.

While it was originally claimed that thirty-eight people witnessed this crime, in actuality, only a few people physically saw Kitty Genovese and her attacker; the others just heard the screams from Kitty Genovese.

In addition, of those who could see, none actually witnessed the stabbing take place (although one of the people who testified did see a violent action on behalf of the attacker.)

This contrasts with the widely held notion that all 38 people witnessed the initial stabbing.

Lastly, the second stabbing that resulted in the death of Catherine Genovese occurred in a stairwell which was not in the view of most of the initial witnesses; this deviates from the original article that stated that the murder took place on Austin Street in New York City in full view of at least 38 people.

This means that they would not have been able to physically see the murder take place. The potential inaccurate reporting of the initial case has not negated the bystander effect completely, but it has called into question its applicability and the incomplete nature of research concerning it.

Limitations of the Decision-Helping Model

Schroeder et al. (1995) believe that the decision-helping model provides a valuable framework for understanding bystander intervention.

Although primarily developed to explain emergency situations, it has been applied to other situations, such as preventing someone from drinking and driving, to deciding to donate a kidney to a relative.

However, the decision model does not provide a complete picture. It fails to explain why ‘no’ decisions are made at each stage of the decision tree. This is particularly true after people have originally interpreted the event as an emergency.

The decision model doesn’t take into account emotional factors such as anxiety or fear, nor does it focus on why people do help; it mainly concentrates on why people don’t help.

Piliavin et al. (1969, 1981) put forward the cost–reward arousal model as a major alternative to the decision model and involves evaluating the consequences of helping or not helping.

Whether one helps or not depends on the outcome of weighing up both the costs and rewards of helping. The costs of helping include effort, time, loss of resources, risk of harm, and negative emotional response.

The rewards of helping include fame, gratitude from the victim and relatives, and self-satisfaction derived from the act of helping. It is recognized that costs may be different for different people and may even differ from one occasion to another for the same person.

Accountability Cues

According to Bommel et al. (2012), the negative account of the consequences of the bystander effect undermines the potential positives. The article “Be aware to care: Public self-awareness leads to a reversal of the bystander effect” details how crowds can actually increase the amount of aid given to a victim under certain circumstances.

One of the problems with bystanders in emergency situations is the ability to split the responsibility (diffusion of responsibility).

Yet, when there are “accountability cues,” people tend to help more. Accountability cues are specific markers that let the bystander know that their actions are being watched or highlighted, like a camera. In a series of experiments, the researchers tested if the bystander effect could be reversed using these cues.

An online forum that was centered around aiding those with “severe emotional distress” (Bommel et al., 2012) was created.

The participants in the study responded to specific messages from visitors of the forum and then rated how visible they felt on the forum.

The researchers postulated that when there were no accountability cues, people would not give as much help and would not rate themselves as being very visible on the forum; when there are accountability cues (using a webcam and highlighting the name of the forum visitor), not only would more people help but they would also rate themselves as having a higher presence on the forum.

As expected, the results fell in line with these theories. Thus, targeting one’s reputation through accountability cues could increase the likelihood of helping. This shows that there are potential positives to the bystander effect.

Neuroimaging Evidence

Researchers looked at the regions of the brain that were active when a participant witnessed emergencies. They noticed that less activity occurred in the regions that facilitate helping: the pre- and postcentral gyrus and the medial prefrontal cortex (Hortensius et al., 2018).

Thus, one’s initial biological response to an emergency situation is inaction due to personal fear. After that initial fear, sympathy arises, which prompts someone to go to the aid of the victim. These two systems work in opposition; whichever overrides the other determines the action that will be taken.

If there is more sympathy than personal distress, the participant will help. Thus, these researchers argue that the decision to help is not “reflective” but “reflexive” (Hortensius et al., 2018).

With this in mind, the researchers argue for a more personalized view that takes into account one’s personality and disposition to be more sympathetic rather than utilize a one-size-fits-all overgeneralization.

Darley, J. M., & Latané´, B. (1968). Bystander intervention in emergencies: Diffusion of responsibility . Journal of Personality and Social Psychology, 8 , 377–383.

Garcia, Stephen M, Weaver, Kim, Moskowitz, Gordon B, & Darley, John M. (2002). Crowded Minds. Journal of Personality and Social Psychology, 83 (4), 843-853.

Hortensius, Ruud, & De Gelder, Beatrice. (2018). From Empathy to Apathy: The Bystander Effect Revisited. Current Directions in Psychological Science, 27 (4), 249-256.

Latané´, B., & Darley, J. M. (1968). Group inhibition of bystander intervention in emergencies . Journal of Personality and Social Psychology, 10 , 215–221.

Latané´, B., & Darley, J. M. (1970). The unresponsive bystander: Why doesn’t he help? New York, NY: Appleton-Century-Croft.

Latané´, B., & Darley, J. M. (1976). Help in a crisis: Bystander response to an emergency . Morristown, NJ: General Learning Press.

Latané´, B., & Nida, S. (1981). Ten years of research on group size and helping . Psychological Bulletin, 89 , 308 –324.

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 , 555-562.

Prentice, D. (2007). Pluralistic ignorance. In R. F. Baumeister & K. D. Vohs (Eds.), Encyclopedia of social psychology (Vol. 1, pp. 674-674) . Thousand Oaks, CA: SAGE Publications, Inc.

Rendsvig, R. K. (2014). Pluralistic ignorance in the bystander effect: Informational dynamics of unresponsive witnesses in situations calling for intervention. Synthese (Dordrecht), 191 (11), 2471-2498.

Shotland, R. L., & Straw, M. K. (1976). Bystander response to an assault: When a man attacks a woman. Journal of Personality and Social Psychology, 34 (5), 990.

Siegal, H. A. (1972). The Unresponsive Bystander: Why Doesn’t He Help? 1(3) , 226-227.

Van Bommel, Marco, Van Prooijen, Jan-Willem, Elffers, Henk, & Van Lange, Paul A.M. (2012). Be aware to care: Public self-awareness leads to a reversal of the bystander effect. Journal of Experimental Social Psychology, 48 (4), 926-930.

Further information

Latané, B., & Nida, S. (1981). Ten years of research on group size and helping. Psychological Bulletin , 89, 308 –324.

BBC Radio 4 Case Study: Kitty Genovese

Piliavin Subway Study

• Anxiety Disorder
• Bipolar Disorder
• Schizophrenia
• Adjustment Disorder
• Agoraphobia
• Antisocial Personality Disorder
• Borderline Personality Disorder
• Childhood ADHD
• Dissociative Identity Disorder
• Narcissistic Personality Disorder
• Oppositional Defiant Disorder
• Panic Attack
• Postpartum Depression
• Schizoaffective Disorder
• Seasonal Affective Disorder
• Sex Addiction
• Social Anxiety
• Specific Phobias
• Teenage Depression

Wellness Topics

• Black Mental Health
• Emotional Health
• Sex & Relationships
• Understanding Therapy
• Workplace Mental Health

Original Series

• My Life with OCD
• Caregivers Chronicles
• Empathy at Work
• Sex, Love & All of the Above
• Parent Central
• Mindful Moment

News & Events

• Mental Health News
• Live Town Hall: Mental Health in Focus
• Inside Mental Health
• Inside Schizophrenia
• Inside Bipolar
• ADHD Symptoms Quiz
• Anxiety Symptoms Quiz
• Autism Quiz: Family & Friends
• Autism Symptoms Quiz
• Bipolar Disorder Quiz
• Borderline Personality Test
• Childhood ADHD Quiz
• Depression Symptoms Quiz
• Eating Disorder Quiz
• Narcissim Symptoms Test
• OCD Symptoms Quiz
• Psychopathy Test
• PTSD Symptoms Quiz
• Schizophrenia Quiz
• Attachment Style Quiz
• Career Test
• Do I Need Therapy Quiz?
• Domestic Violence Screening Quiz
• Emotional Type Quiz
• Loneliness Quiz
• Parenting Style Quiz
• Personality Test
• Relationship Quiz
• Stress Test
• What's Your Sleep Like?

Treatment & Support

• Find Support
• Suicide Prevention
• Drugs & Medications
• Find a Therapist

Understanding the Bystander Effect

Picture this: You’re walking down the street when you hear someone call for help. They’re being assaulted. What do you do? Most of us would like to think that we’d intervene — or at the very least, call 911. But the truth is, this isn’t what always happens.

On October 17, 2021, a woman was sexually assaulted on a train near Philadelphia, in full view of several passengers. Yet not a single one of them helped her or called the authorities, even though it was very obvious what was happening.

Onlookers time and again have withheld help to someone in need. Psychologists call this the bystander effect.

What is the bystander effect?

In short, the bystander effect is the name given to the phenomenon where people in a group fail to offer help to someone during an emergency, even though they are witnesses to the event.

In fact, research from 2014 suggests that the bigger the group, the less likely it is that anyone will come to help.

What started the research

In 1964 , a woman named Catherine “Kitty” Genovese was attacked and repeatedly stabbed by a serial killer named Winston Moseley, despite calling out for help in her apartment courtyard. As many as 38 people were said to have witnessed her being murdered.

The press coverage at the time alleged that none of these witnesses came to her aid. These sensationalized early accounts have since been disproven — but nevertheless, the case jump-started psychological research into the bystander effect.

This includes some groundbreaking research by John Darley and Bibb Latané.

How psychology explains the bystander effect

In a series of experiments, Darley and Latané found that people tend to feel a moral responsibility to help someone in distress if they believe they are the only witnesses. But if they’re surrounded by others, they’re significantly less likely to feel like they have to intervene.

In fact, in 1969, Latané found that while 70% of people would help a woman in distress if they were the only bystander, only 40% would come to her aid if other people were present.

More recently, studies have found that people are less likely to speak up if they witness cyberbullying that takes place in larger online group forums, according to a 2015 review .

Examples of the bystander effect

• ignoring bullying or cyberbullying
• filming an assault instead of calling 911
• assuming someone else will help
• walking past a person lying on the street (Content warning: graphic)
• apathy toward climate change

Why does it happen?

Researchers think that there are two group dynamics at work in the bystander effect, which is why we’re less likely to act when we’re surrounded by others.

Diffusion of responsibility

In a group, we can feel less individual responsibility to help others. This has been observed in children as young as 5 and adults alike, per 2015 research and 2017 research respectively.

It happens for a simple reason: When we’re in a group, it’s easier to assume that someone else will step up and do something, so we don’t do anything ourselves. This leads to the bystander effect. The problem is, when everyone assumes that someone else will act, no one actually does.

Social referencing

When we’re in a group, we can look to others to decide what is appropriate behavior and what’s not.

So if there is a crisis — and it’s not clear what we should do because of the confusion — we often look at what everyone else is doing to get social cues.

If we don’t see anyone doing anything, we might assume there’s a reason for the inaction and draw a false conclusion that no action is needed, according to older research by Latané and Darley.

For example, if two people are arguing but no one else seems to care, we might figure it’s just a quarrel and keep walking — even if that argument turns physical.

What makes bystanders more likely to intervene?

Some research from 2011 indicated that increased levels of danger could push bystanders to intervene.

Why? Well first of all, if the situation is dangerous to you and the victim, you’re more likely to pay attention to what’s going on. And second, in a dangerous situation, being in a group might help you feel more empowered and like you can actually help, per 2013 research .

For example, if you see someone attacking another person with a knife and you’re alone, you might be tempted to run away rather than help. But if you’re with a group, you might be more confident that you — and the group as a whole — can stop the knife-wielder if you work together.

How to counteract bystander apathy

Here are some theories on how to wake from the bystander effect:

Know what to do

If you’re an onlooker, you’re more likely to feel empowered to intervene if you know strategies to do so safely . Depending on the situation you encounter, you can choose to:

• intervene directly
• distract the attacker, if there is one
• delegate (bring in others to help)
• delay (offer follow-up resources and emotional support to the affected)

You can also take steps to be more prepared for emergencies, like taking first aid classes or getting CPR training.

Use the bystander effect positively

Some psychologists believe that simply being aware of the bystander effect could make us all more likely to react.

After all — if we know it can happen, we might be more determined to make sure we aren’t the ones that stood by and let something horrible occur.

Shame and guilt can be powerful motivators, so being observed in a group might make you feel like you have to help.

Other research from 2018 has suggested that the more we see or hear about people helping others (like donating blood), the more likely we are to do good ourselves.

Call out to a specific person

According to a 2018 research review , studies have found that people are more likely to help people they already know. This is why people are less likely to come to the aid of a stranger.

So, psychologist and professor Ken Brown said in a 2015 TEDx talk , if you’re in a crisis and you need help from others, try to focus on getting just one person to help. You’re more likely to get help, 2015 research says, if you’re direct and personal.

For example, you can call out to a specific bystander in the crowd, identifying them by the color of their shirt or hair, and ask them to call 911.

In fact, Brown recently commented to Psych Central about his 2015 TED Talks, noting the power of “asking for help in ways that reduce uncertainty.” You might’ve noticed that the audience member whom he called out of the crowd to ask for help stayed and assisted him for an unusually long period of time.

Brown reflects, “She continued to help me, despite the discomfort of standing, because I had made a direct request of her and once she started, it was clear that I wanted her to keep doing it,” until he eventually thanked her in front of the audience (but after the recording stops).

Let’s recap

The bystander effect happens more than we’d like, but there are some things we can all do to overcome it and help others — and it all starts with knowing that this phenomenon even exists.

18 sources collapsed

• Aakvaag HF, et al. (2014). Shame and guilt in the aftermath of terror: The Utøya island study. https://onlinelibrary.wiley.com/doi/10.1002/jts.21957
• A new look at the killing of Kitty Genovese: The science of false confessions. (2017). https://www.psychologicalscience.org/publications/observer/obsonline/a-new-look-at-the-killing-of-kitty-genovese-the-science-of-false-confessions.html
• Beyer F, et al. (2017). Beyond self-serving bias: Diffusion of responsibility reduces sense of agency and outcome monitoring. https://academic.oup.com/scan/article/12/1/138/2628052
• Brown K. (2021). Personal interview.
• Bystander effect. (n.d.) https://dictionary.apa.org/bystander-effect
• Bystander intervention. (n.d.) https://www.rockefeller.edu/education-and-training/bystander-intervention/
• Cieciura J. (2016). A summary of the bystander effect: Historical development and relevance in the digital age. http://www.inquiriesjournal.com/articles/1493/a-summary-of-the-bystander-effect-historical-development-and-relevance-in-the-digital-age
• Darley JM, et al. (1968). Bystander intervention in emergencies: Diffusion of responsibility. https://doi.apa.org/record/1968-08862-001?doi=1
• Fischer P, et al. (2011). The bystander-effect: A meta-analytic review on bystander intervention in dangerous and non-dangerous emergencies. https://pubmed.ncbi.nlm.nih.gov/21534650/
• Fischer P, et al. (2013). The positive bystander effect: Passive bystanders increase helping in situations with high expected negative consequences for the helper. https://pubmed.ncbi.nlm.nih.gov/23421000/
• Hortensius R, et al. (2018). From empathy to apathy: The bystander effect revisited. https://journals.sagepub.com/doi/10.1177/0963721417749653
• Kassin SM. (2017). The killing of Kitty Genovese: What else does this case tell us? https://web.williams.edu/Psychology/Faculty/Kassin/files/Kassin%20(2017)%20-%20Kitty%20Genovese.pdf
• Latané B, et al. (1969). A lady in distress: Inhibiting effects of friends and strangers on bystander intervention. https://www.omerdank-strategy.com/wp-content/uploads/2020/03/A_lady_in_distress_Inhibiting_effects_of.pdf
• Machackova H, et al. (2015). Brief report: The bystander effect in cyberbullying incidents. https://www.sciencedirect.com/science/article/abs/pii/S0140197115001049
• Obermaier M, et al. (2014). Bystanding or standing by? How the number of bystanders affects the intention to intervene in cyberbullying. https://journals.sagepub.com/doi/abs/10.1177/1461444814563519
• Plötner M, et al. (2015). Young children show the bystander effect in helping situations. https://journals.sagepub.com/doi/abs/10.1177/0956797615569579
• Raihani NJ, et al. (2015). Why humans might help strangers. https://www.frontiersin.org/articles/10.3389/fnbeh.2015.00039/full
• Studte S, et al. (2018). Blood donors and their changing engagement in other prosocial behaviors. https://onlinelibrary.wiley.com/doi/10.1111/trf.15085

Read this next

We're unpacking the exchange theory and breaking down what you're really attracted to in your friendships or romantic relationships.

Albert Bandura's social learning theory talks about modeling and positive reinforcement. How can it explain behaviors and mental health conditions?

You're hungry, you eat. You're thirsty, you drink. The drive reduction theory has an equation that explains these behaviors. But, what about the rest?

Feeling love for your parents is natural, but according to Freud's theory, it may be a sign of Oedipus complex if those feelings stray into…

The psychology behind conspiracy theories offers explanations of why some people are more likely to believe conspiracy theories, even those that feel…

Take the first step in feeling better. You can get psychological help by finding a mental health counselor. Browse our online resources and find a…

If you’re experiencing anxiety, these 15 essential oils may help ease your symptoms.

Here's how to encourage leadership to create a more empathetic workplace if employees feel their needs aren't met.

Research indicates that some vitamin deficiencies may put you at a greater risk of depression. Here are nine deficiencies linked to depression.

• Bipolar Disorder
• Therapy Center
• When To See a Therapist
• Types of Therapy
• Best Online Therapy
• Best Couples Therapy
• Managing Stress
• Sleep and Dreaming
• Understanding Emotions
• Self-Improvement
• Healthy Relationships
• Student Resources
• Personality Types
• Guided Meditations
• Verywell Mind Insights
• 2024 Verywell Mind 25
• Mental Health in the Classroom
• Editorial Process
• Meet Our Review Board
• Crisis Support

How Psychology Explains the Bystander Effect

Kendra Cherry, MS, is a psychosocial rehabilitation specialist, psychology educator, and author of the "Everything Psychology Book."

Amy Morin, LCSW, is a psychotherapist and international bestselling author. Her books, including "13 Things Mentally Strong People Don't Do," have been translated into more than 40 languages. Her TEDx talk,  "The Secret of Becoming Mentally Strong," is one of the most viewed talks of all time.

How the Bystander Effect Works

• Real-Life Example
• Explanations

What Is the Meaning of Bystander Effect?

The bystander effect, also known as bystander apathy, refers to a phenomenon in which the greater the number of people there are present, the less likely people are to help a person in distress.

If you witnessed an emergency happening right before your eyes, you would certainly take some sort of action to help the person in trouble, right? While we might all like to believe that this is true, psychologists suggest that whether or not you intervene might depend upon the number of other witnesses present.

When an emergency situation occurs, the bystander effects holds that observers are more likely to take action if there are few or no other witnesses.

Being part of a large crowd makes it so no single person has to take responsibility for an action (or inaction).

In a series of classic studies, researchers Bibb Latané and John Darley found that the amount of time it takes the participant to take action and seek help varies depending on how many other observers are in the room. In one experiment , subjects were placed in one of three treatment conditions: alone in a room, with two other participants, or with two confederates who pretended to be normal participants.

As the participants sat filling out questionnaires, smoke began to fill the room. When participants were alone, 75% reported the smoke to the experimenters. In contrast, just 38% of participants in a room with two other people reported the smoke. In the final group, the two confederates in the experiment noted the smoke and then ignored it, which resulted in only 10% of the participants reporting the smoke.

Additional experiments by Latané and Rodin (1969) found that 70% of people would help a woman in distress when they were the only witness. But only about 40% offered assistance when other people were also present.

What Is a Real-Life Example of the Bystander Effect?

The most frequently cited example of the bystander effect in introductory psychology textbooks is the brutal murder of a young woman named Catherine "Kitty" ​Genovese. On Friday, March 13, 1964, 28-year-old Genovese was returning home from work. As she approached her apartment entrance, she was attacked and stabbed by a man later identified as Winston Moseley.

Despite Genovese’s repeated calls for help, none of the dozen or so people in the nearby apartment building who heard her cries called the police to report the incident. The attack first began at 3:20 AM, but it was not until 3:50 AM that someone first contacted police.

An initial article in the New York Times sensationalized the case and reported a number of factual inaccuracies. An article in the September 2007 issue of American Psychologist concluded that the story is largely misrepresented mostly due to the inaccuracies repeatedly published in newspaper articles and psychology textbooks.  

While Genovese's case has been subject to numerous misrepresentations and inaccuracies, there have been numerous other cases reported in recent years. The bystander effect can clearly have a powerful impact on social behavior, but why exactly does it happen? Why don't we help when we are part of a crowd?

Why Does It Happen?

There are two major factors that contribute to the bystander effect. First, the presence of other people creates a diffusion of responsibility .

Because there are other observers, individuals do not feel as much pressure to take action. The responsibility to act is thought to be shared among all of those present.

The second reason is the need to behave in correct and socially acceptable ways. When other observers fail to react, individuals often take this as a signal that a response is not needed or not appropriate.

Researchers have found that onlookers are less likely to intervene if the situation is ambiguous. In the case of Kitty Genovese, many of the 38 witnesses reported that they believed that they were witnessing a "lover's quarrel," and did not realize that the young woman was actually being murdered.

A crisis is often chaotic and the situation is not always crystal clear. Onlookers might wonder exactly what is happening. During such moments, people often look to others in the group to determine what is appropriate. When they see that no one else is reacting, it sends a signal that perhaps no action is needed.

Preventing the Bystander Effect

What can you do to overcome the bystander effect ? Some psychologists suggest that simply being aware of this tendency is perhaps the greatest way to break the cycle. When faced with a situation that requires action, understand how the bystander effect might be holding you back and consciously take steps to overcome it. However, this does not mean you should place yourself in danger.

But what if you are the person in need of assistance? How can you inspire people to lend a hand? One often recommended tactic is to single out one person from the crowd. Make eye contact and ask that individual specifically for help. By personalizing and individualizing your request, it becomes much harder for people to turn you down.

Manning R, Levine M, Collins A. The Kitty Genovese murder and the social psychology of helping: the parable of the 38 witnesses . Am Psychol. 2007;62(6):555-62. doi:10.1037/0003-066X.62.6.555

Darley JM, Latané B. Bystander “apathy.” American Scientist. 1969;57:244-268.

Latané B, Darley JM. The Unresponsive Bystander: Why Doesn’t He Help?  Prentice Hall, 1970.

Solomon LZ, Solomon H, Stone R. Helping as a function of number of bystanders and ambiguity of emergency . Pers Soc Psychol Bull . 1978;4(2):318-321. doi:10.1177/014616727800400231

By Kendra Cherry, MSEd Kendra Cherry, MS, is a psychosocial rehabilitation specialist, psychology educator, and author of the "Everything Psychology Book."

An official website of the United States government

Official websites use .gov A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

• Publications
• Account settings
• Advanced Search
• Journal List

Bystander Responses to a Violent Incident in an Immersive Virtual Environment

Aitor rovira, richard southern, david swapp, jian j zhang, claire campbell, mark levine.

• Author information
• Article notes
• Copyright and License information

* E-mail: [email protected]

Competing Interests: The authors have declared that no competing interests exist.

Conceived and designed the experiments: MS ML. Performed the experiments: AR DS RS. Analyzed the data: MS. Wrote the paper: MS ML AR RS DS JJZ CC. Computer animation: RS JJZ.

Received 2012 Mar 1; Accepted 2012 Nov 22; Collection date 2013.

This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are properly credited.

Under what conditions will a bystander intervene to try to stop a violent attack by one person on another? It is generally believed that the greater the size of the crowd of bystanders, the less the chance that any of them will intervene. A complementary model is that social identity is critical as an explanatory variable. For example, when the bystander shares common social identity with the victim the probability of intervention is enhanced, other things being equal. However, it is generally not possible to study such hypotheses experimentally for practical and ethical reasons. Here we show that an experiment that depicts a violent incident at life-size in immersive virtual reality lends support to the social identity explanation. 40 male supporters of Arsenal Football Club in England were recruited for a two-factor between-groups experiment: the victim was either an Arsenal supporter or not (in-group/out-group), and looked towards the participant for help or not during the confrontation. The response variables were the numbers of verbal and physical interventions by the participant during the violent argument. The number of physical interventions had a significantly greater mean in the in-group condition compared to the out-group. The more that participants perceived that the Victim was looking to them for help the greater the number of interventions in the in-group but not in the out-group. These results are supported by standard statistical analysis of variance, with more detailed findings obtained by a symbolic regression procedure based on genetic programming. Verbal interventions made during their experience, and analysis of post-experiment interview data suggest that in-group members were more prone to confrontational intervention compared to the out-group who were more prone to make statements to try to diffuse the situation.

Introduction

A violent and unprovoked attack by one person on another unfolds in close view of an unrelated bystander: under what conditions will the bystander be likely to intervene to help the victim? In this paper we address the hypothesis that group affiliation between the bystander and the victim provides a powerful incentive for the bystander to try to intervene to stop the attack, or prevent harm to the victim, and in particular that this operates even though the perpetrator and victim are virtual human characters. Our experiment involved fans of an English football team, Arsenal. In one experimental condition (in-group) the fan conversed with a virtual character that was clearly an Arsenal supporter and in another condition the character was just a general football enthusiast but not an Arsenal fan (out-group). The virtual character was later threatened by a perpetrator that, in the in-group condition, specifically attacked his Arsenal affiliation. Our expectation was that based on group affiliation, those in the in-group would intervene more than those in the out-group. First we place this in the general context of studies of bystander intervention, and then describe the detailed design of the experiment and the results.

Research on the behaviour of bystanders in emergencies began with the response to the rape and murder of Kitty Genovese in New York in 1964. Social psychologists Bibb Latane and John Darley read a report on the murder in the New York Times suggesting that 38 witnesses had watched the murder unfold over 30 minutes from their apartment windows– and yet failed to intervene. In order to understand why this might have happened, they set out to create laboratory based experimental analogies of the event. They set up carefully choreographed situations in which bystanders were faced with a non-violent emergency situation while on their own or in the presence of others [1] , [2] . The research led to the discovery of the ‘bystander effect’ – the idea that people are more likely to intervene on their own than in the presence of others [1] . This is one of the most reliable and robust findings in social psychology [3] , [4] .

However, as Cherry pointed out [5] , through translating the events surrounding the Genovese murder into laboratory settings, Latane and Darley neglected some of the key features of the event. Despite the fact that the original murder involved violence by a man against a woman, subsequent experimental analogies tended to remove both the gendered nature of the attack and the violence. Although there are thousands of studies using non-violent emergency settings, it is possible to find only a few experiments that did retain violence as the emergency variable [6] , [7] , [8] , [9] , [10] . These found results that were at odds with the traditional bystander paradigm. In violent emergencies, what seemed to be most important about the likelihood of bystander intervention was not the presence of others, but rather the bystanders’ beliefs about the nature of the relationship between perpetrator and victim [9] , [10] . In an experiment that did vary the number of bystanders to a violent emergency Harari et al. [6] showed that the presence of others actually enhanced the likelihood of bystander intervention in a simulated rape situation. This finding has been supported by contemporary work which presents violence to participants by means of a CCTV video link, where the presence of others is not found to inhibit helping [11] and can sometimes enhance it [12] . A recent meta-analysis by Fisher and colleagues [4] confirms that intervention behaviour in violent emergencies does not fit the traditional bystander effect explanation.

If violent emergencies are different in some way, it is important to understand the processes at work. Almost all violence research shares a similar limitation. In order to circumvent the practical and ethical problems of presenting violence in experimental settings, these experiments tend to avoid placing participants in direct contact with the violence itself. The only exception is the work described in [10] in which a role-play setting was used, and confederates actually staged a violent confrontation in front of naive participants who were also taking part in the role-play game. However, it is highly unlikely that contemporary ethics boards would allow this kind of design. The other studies either have the violence happening at a distance where it is possible to avoid the event [6] or present the violence as happening contemporaneously but where it can only be heard [7] , [8] , [9] , or happening in another room where it can be seen on CCTV link [11] . This distancing of participants from the violence is required to satisfy the ethical and practical difficulties of experimental design, but may itself introduce psychological effects that interfere with the veridical nature of the situation. Imagining the violence, or having it happen in another room, is not the same as being physically where the violence erupts.

In [13] we argued that the use of immersive virtual environments (IVE) goes some way towards solving this problem, since there is mounting evidence that when people are faced with events and situations in an IVE they tend to behave and respond as if these were real [14] . IVEs portray a simulated computer generated reality at life size that is sensorially surrounding. Participants perceive this world through wide field-of-view stereo vision and sound. The form of perception involves more or less natural sensorimotor contingencies - meaning that the whole body is used for perception much as in physical reality, based at least on head-gaze direction and orientation achieved through head-tracking. This gives rise to the sensation of being in the virtual place that is depicted, a place-illusion. Additionally when there are dynamically unfolding events in the environment that personally refer to the participant, and where actions of the participant apparently cause responses in the virtual environment, this gives rise to a plausibility-illusion, meaning that events have the illusory quality of being real. When the participant has the double illusion - of being in the virtual place and where events that are happening are apparently really happening, this can give rise to behaviour and responses that are appropriate to the situation as if it were playing out in reality [15] .

IVEs provide therefore a powerful tool for experimental studies in social psychology [16] and classic effects such as proxemics [17] where distances that people maintain between themselves are governed by social norms, have been reproduced several times in IVEs with respect to virtual humanoid characters [18] , [19] , [20] . Moreover, IVEs have been useful for experiments that would otherwise be difficult to carry out in any other way, such as the study of male risk taking in the presence of observers, specifically the differential effects of the observers being male or female [21] .

Closer to the present study which focuses on responses to violence, the Stanley Milgram obedience paradigm [22] has been reproduced with IVE avoiding the ethical difficulties of deception [23] , [24] . IVEs provide environments completely under control of a computer program but where people respond realistically. Every experimental condition can be exactly reproduced across trials as needed, and hence can be used for laboratory based experiments.

It has been argued before that IVEs provide an excellent tool for the study of prosocial behaviour [25] . The experiment described in the present study is specifically concerned with the likelihood of prosocial behaviour when participants are placed in direct proximity to violent behaviour. We explore the hypothesis that the psychological relationships between bystanders and the others involved are important in bystander behaviour, in this case specifically the relationship between the bystander and the victim [12] , [26] , [27] , [28] . The experimental conditions provide a context where it is certain that the violence between perpetrator and victim is of the same magnitude and intensity for each experimental trial. Participants (n = 40) all supporters of the Arsenal Football Club , entered into a virtual reality that represents a bar. A male virtual human (V) approached and conversed with them about football for a few minutes. In one condition V wore an Arsenal football shirt and spoke enthusiastically about the club (in-group condition). In a second condition V wore an unaffiliated red sports shirt, and asked questions about Arsenal without special enthusiasm, using neutral responses and displaying ambivalence about Arsenal’s prospects (out-group condition). After a few minutes of this conversation another male virtual human (P, perpetrator) who had been sitting by the bar walked over to V (victim) and started an argument that he continually escalated until it became a physical attack ( Figure 1 ).

Figure 1. The Victim and Perpetrator.

The Victim (V) is in the red shirt, with an Arsenal emblem in the in-group condition, and with a plain football shirt of the same colour in the out-group condition. The perpetrator (P) had been sitting by the bar. (a) P stood up to approach V and (b) started an argument. (c) As the argument progressed V made conciliatory statements and postures while (d) P became ever more aggressive finally pushing V violently against a wall.

The main response variable was the extent to which the participant attempted to intervene during this confrontation. Interventions were verbal utterances or physical moves towards the two virtual characters and were coded from video recordings by two independent researchers (Methods). There were two binary factors group and LookAt . Group was whether V was in-group (Arsenal supporter) or out-group with respect to the participant. LookAt was whether or not occasionally during the confrontation V would look towards the participant or not ( LookAt  = ‘on’ or ‘off’). The experiment used a between-groups design, with n = 40, 10 participants allocated arbitrarily to one of the four cells of the 2×2 design. The degree of support for the Arsenal club was similar between the 4 experimental conditions ( Text S1 ). At the end of their session they answered a questionnaire, and this was followed by an interview and debriefing. The data from two participants could not be used due to video recording failures.

Numbers of Interventions

Table 1 shows the means and standard errors of the numbers of interventions indicating that the mean number of interventions was higher for the in-group than the out-group, but that the LookAt factor had no effect. Two-way analysis of variance was carried out on the response variables, the number of physical ( nPhys ) and number of verbal ( nVerbal ) interventions. ANOVA for nPhys indicates that the mean is greater for the in-group than for the out-group condition (P = 0.02) but with no significant differences for the LookAt factor and no interaction effect. However, the residual errors of the fit were strongly non-normal (Shapiro-Wilk test P = 0.0008). To overcome this problem a square root transformation was applied to nPhys . This resulted in the same conclusions for group (P = 0.016, partial η 2  = 0.15) and no significance for LookAt (P = 0.297, partial η 2  = 0.03). The normality of the residuals is improved although not ideal (Shapiro-Wilk P = 0.034). For the response variable nVerbal the results were similar: ANOVA of nVerbal on group and LookAt shows no significant interaction term, group has significance level P = 0.095, and for LookAt P = 0.228. However, again the residual errors are far from normal (SW P = 0.0008). The square root transformation gives P = 0.060, partial η 2  = 0.10 for group and P = 0.112, partial η 2  = 0.07 for LookAt . The residual errors are compatible with normality (SW P = 0.24).

Table 1. Means and Standard Errors of Numbers of Interventions.

n = 9 for each of the two Out-group cells, n = 10 for each of the two In-group cells, n = 38 in total.

The factor LookAt represents whether the V avatar was programmed to occasionally look toward the participants. Additionally, the post experience questionnaire included the statement ( VictimLooked ) “After the argument started, the victim looked at me wanting help” which was scored on a scale from 1 (least agreement) to 7 (most agreement). VictimLooked therefore represents the belief of the participants as to whether the victim looked towards them for help . There is no significant difference between the mean VictimLooked score of those who were in the group LookAt  = ‘on’ (mean 3.3, SD = 1.8, n = 20) and those in the group LookAt  = ‘off’ (mean 4.0, SD = 1.5, n = 20) (P = 0.12, Mann-Whitney U). Hence the response to this question was not based on the number of actual looks of the victim towards the participant, and therefore was a belief. It turns out that VictimLooked plays a significant role in the number of interventions.

Figure 2 shows the scatter plots of nPhys and nVerbal on the questionnaire response VictimLooked for the out-group and in-group. These reveal a quite different relationship in the two cases. In the case of the in-group there is a positive association between the number of interventions (verbal or physical) and the perception that the victim was looking towards the participant for help. In the case of the out-group there appears to be no relationship in the nPhys case and a possible negative relationship in the nVerbal case. Using the same strategy as above in order to obtain residual errors compatible with normality, ANCOVA of nPhys 0.5 on group with VictimLooked as a covariate shows that the slopes of the regression line are different between the in-group and out-group (P = 0.004, partial η 2  = 0.22 for the slopes, SW P = 0.18). For the number of verbal interventions, using nVerbal 0.5 the difference in slopes between in-group and out-group is significant at P = 0.004 (partial η 2  = 0.22 for the slope, SW P = 0.12).

Figure 2. Number of interventions by VictimLooked and Group .

(a) For the verbal interventions and (b) for the physical interventions.

These results indicate that the response to the belief that the victim was looking towards the bystander for help was different between the in-group and out-group. For those in the in-group condition the greater their belief that the victim was looking to them for help the greater the number of verbal and physical interventions. For those in the out-group condition there is no such association. These results are further corroborated using multivariate analysis of variance on the response vector ( nPhys 0.5 , nVerbal 0.5 ) ( Text S2 ).

Numbers of Interventions - Symbolic Regression

The previous section provided standard analyses for these types of data. Even though this revealed positive results consistent with our initial hypothesis, in this section we also employ a quite different method using symbolic regression, to throw further light on the experimental results. The purpose is to consider the relationship between the number of interventions, and the experimental factors, but now also including any possible influence of the subjective variables as elicited through the post-experience questionnaire ( Table 2 ). Standard statistical analysis is based, amongst other things, on the assumption of linearity in the parameters. But in such a complex situation as the one under consideration, on what grounds is such an assumption valid when considering the multivariate influence of a number of factors potentially influencing bystander intervention? Symbolic regression does not rely on such linearity, being a method for discovering relationships between variables using the technique of genetic programming [29] ( Text S3 ). It has recently been shown to be able to discover complex physical laws automatically [30] , using a program called Eureqa, which was used in the analysis presented below. In the context that we apply this technique here, we consider it as a data reduction method. It allows us to succinctly represent the original data but with quite simple equations while preserving the variance in the original data. It is not a technique that can be compared with statistical significance testing, it is rather a data exploration method, that can lead to understanding of complex data, where models generated by this technique can be used for hypothesis formation in later experimental study.

Table 2. The Post-Questionnaire and Corresponding Variable Names.

All items were presented as statements on a 1–7 Likert scale where 1 meant least agreement and 7 most agreement with the corresponding statement.

The operators that were used for the symbolic regression were: Constant, +, −, ×, /, sqrt, exp, log. The program was run for both nPhys and nVerbal . The population size (number of formulae per generation) was chosen by Eureqa as 2560. For each analysis the program was run on a 40 core cluster (see Methods) and left running for many hours until the solution set of equations stabilized. The fitness metric used was mean absolute error.

We consider first nPhys . The Eureqa program was left to run for more than 2000 core hours. It reported 28 equations. Each has an associated size parameter that represents the complexity of the equation (ranging from 1, least, to 53, most complex), a fitness value, the square of the correlation coefficient between the response variable and the fitted values from the equation, and the Akaike Information Criterion (AIC). The AIC is an information theoretic measure of the relative goodness of fit of a model to the data. Smaller AIC values represent better goodness of fit, taking also into account the complexity of the model. The AIC is often used in model selection procedures, as discussed extensively in [31] .

The model with the smallest AICs is shown in Eq (1). Here group is 0 for out-group and 1 for in-group. Similarly LookAt is 0 for ‘off’, and 1 for ‘on’. The other variables are from the questionnaire ( Table 2 ).

(R 2  = 0.85, AIC = 108, Size = 26).

Figure 3a shows the relationship between the observed and fitted number of interventions based on Eq (1) (the diagram is very similar for all the top fitting equations generated). The high fitting equations all, of course, give similar results and Eq (1) is marginally preferred since it has high explanatory power (in terms of correlation) and the smallest AIC, and on the range of complexity of the models produced is about half way along the scale amongst all generated equations.

Figure 3. The fitted number of interventions by VictimLooked from Eqs.

(1) and (2). (a) The fitted against observed values for nPhys . (b) The fitted against observed for nVerbal .

The equation shows a clear distinction between in-group and out-group. For the out-group (group = 0) the entire first term, on the left-hand side of the plus sign, vanishes (20 of the 28 equations generated have this exponential term). For the in-group (group = 1), it can be seen that LookAt has a very small but positive influence on the number of interventions but VictimLooked has a greater influence. As it ranges from 1 to 7 the number of interventions increases by 0.015*exp( VictimLooked ), which is, for example, 2 for VictimLooked  = 5, and 16 for VictimLooked  = 7, other things being equal.

The second term only includes a few of the questionnaire variables. Examining this term, the number of interventions is proportional to concern about the safety of others, and the feeling that the fight should be stopped. It is inversely proportional to the feeling of wanting to get out, and the fear that other people might turn up to make things worse.

Now we turn to the number of verbal interventions nVerbal , and follow the same analysis. Here the genetic program ran for 1930 core hours. 28 equations were produced with size complexity ranging from 1 to 71. The equation with the lowest AIC is shown in Eq (2).

(R 2  = 0.93, AIC = 83, Size = 29).

As before all the high fitting equations give very similar results and we take Eq. (2) as representative. Figure 3b shows the plot of fitted by observed values over the data set for Eq. (2). Examining the equation we see this time there is no effect of group . The number of verbal interventions is proportional to the feeling of the need to stop the fight, and inversely proportional to the fear that other people might arrive and make things worse. Also there is a positive association with participant fears for their own safety. The most interesting variable again is VictimLooked , the belief that the V avatar was looking towards the participant for help. The variable MoveAway is strongly related with VictimLooked which must be taken into account otherwise the equations explode into huge values as VictimLooked increases. Figure 4 shows that there is a very strong positive correlation between these two variables (apart from 1 outlier) (r = 0.71, P = 3.3×10 −7 ), with regression line MoveAway  = −0.38+0.82 VictimLooked . Moreover 22 out of the 28 equations include the exponential term involving these two variables. We maintain this relationship when examining the effect of VictimLooked on nVerbal rather than fixing MoveAway at a constant value, and taking this into account high values of VictimLooked are associated with a larger number of interventions.

Figure 4. Scatter diagram of MoveAway against VictimLooked .

Note the one outlying point when VictimLooked  = 1 and MoveAway  = 7.

The Interviews

After the experimental trial there was a short interview with the participants, followed by their debriefing where the purposes of the experiment were explained. The interviews concentrated on several main questions: their feelings and responses during their experience, the extent to which they judged their responses to be realistic, factors that might have increased their intervention, and factors that drew them out of the experience. Summaries of the interviews were coded into key codes and frequency tables constructed, using the HyperResearch software [32] .

We consider first the responses and feelings of participants during their experience. Table 3 shows the codes and two example sentences of each code and Table 4 the code frequencies.

Table 3. Codes for the Interview Questions: What feelings/responses did you have while this was happening?

Table 4. frequencies of the codes in table 3 ..

The impression from the interviews as shown in Table 4 is that those in the out-group tended to sympathize with or feel sorry for V. Also many of them wanted to just leave the situation, felt uninvolved, or a few found the situation silly. For those in the in-group it seems to be more anger and frustration that could be the driving force of their intervention, and their response was more likely to be a confrontational one. None of them felt uninvolved, found the situation funny or silly, felt sorry for V or wanted to leave. Some of the in-group expressed surprise at their own responses even though they were aware that it was virtual reality, whereas none of the out-group expressed such surprise. This fits with the fact that many of the out-group felt uninvolved and none of the in-group felt so.

Tables 5 and 6 give the results for the interview question regarding the authenticity of response in comparison with reality. We do not show the separate tables for in-group and out-group since there is no difference between them in this regard, although there is some suggestion of a difference between the LookAt groups. It seems that those in the LookAt ‘off’ group were more likely to remark on the lack of interaction, and to contrast their behaviour in virtual reality and reality. They were less likely to report their responses as being realistic. In the combined sample just over half found that their responses were realistic.

Table 5. Codes for the Interview Question: Were your responses realistic?

Table 6. frequencies of the codes in table 5 ..

Participants were asked what might have increased or decreased their degree of intervention. The results are shown in Tables 7 and 8 . Most frequently they said that if the setup had been more interactive (i.e., the characters responding to their actions after the argument had started) then they would have been more likely to intervene. There were two other aspects that are opposed. On the one side a number of participants said that they would have been more likely to intervene if the perpetrator had become more aggressive. On the other side some participants said that they might have intervened had the perpetrator been less aggressive. Others emphasized that had the victim explicitly called for help they would have been more likely to have intervened. Another important contributory factor could have been greater rapport - for example, the victim having been a friend - or someone in need such as a child.

Table 7. Codes for the Interview Question: What would have made it more likely for you to intervene?

Table 8. frequencies of the codes in table 7 ..

Finally participants were asked to talk about technical factors that drew them out of the experience. It will be seen from the video (Video S1) that, for example, there is no lip sync when the characters talk. This is very obvious when looking at the video, but barely noticeable when immersed in the environment with the life-sized characters. The combination of gesture and natural turn taking in conversation, amongst other things, are probably factors in making this glaring defect not noticeable. Only 5 out of 40 people mentioned the lack of lip sync and it was the fifth most mentioned aspect in this question. Table 9 shows the list of topics raised by the participants and the number of times they were mentioned. By far the greatest number of issues were concerned with ‘plausibility’ of the situation itself, and the technical factors tend to come down lower in the list.

Table 9. Frequencies of Statements in Response to the Interview Question: What factors tended to draw you out of the experience?

The principal finding of this research with respect to the bystander issue is that participants in the in-group condition made more attempts at physical and verbal intervention than those in the out-group condition. Second, for those in the in-group the number of physical interventions was associated with the belief that the victim was looking towards them for help.

This second finding relies on the important distinction between the experimentally manipulated LookAt factor, and the questionnaire report after the experiment about how much the subjects thought that the victim was looking towards them for help ( VictimLooked ). To be clear, LookAt refers to whether or not in fact the program was making the victim sometimes look towards the participant. The second refers to the reported belief of the participant that the victim was looking towards him for help . The analysis of covariance (and Figure 2 ) showed that the belief that the victim was looking towards the participant for help had a differential effect depending on group. For those in the in-group condition, if they believed that the victim was looking towards them for help their number of interventions tended to be greater. For those in the out-group condition this relationship did not occur. This would not be surprising if it occurred in reality. If you consider you have group affiliation with someone and that person is looking to you for help surely this would be a more important event, more likely to move you to action, than if someone with whom you have no affiliation looks towards you for help. It is especially striking then that this also occurs also in virtual reality (where the only real people were the participants themselves): the more that the participants believed that the victim was looking towards them for help the more often did they intervene - but only those in the in-group condition.

Third, the use of symbolic regression as a data exploration method complemented and supported the results found from the classical analysis. Specifically, it provided a further demonstration that those in the in-group and out-group conditions responded quite differently to the influence of the LookAt factor and the VictimLooked response. Additionally, for those in both in-group and out-group the feeling that they should stop the argument was positively associated with an increased number of physical interventions, as was concern for the safety of the victim. However, the fear that other people might turn up to make the situation worse was inversely related to the number of physical interventions as was the feeling of wanting to get out.

The picture looks different for the number of verbal interventions. Here the group did not seem to play much role. Important factors contributing positively to the number of such interventions were the feelings by participants that they ‘should stop it’, concern for their own safety, and a strong perception that the victim was looking towards them for help. The factors that contributed negatively were the feeling of wanting to move away from the protagonists, and also the fear that other people might turn up to make the situation worse. However, in the vast majority of equations generated by the symbolic regression the belief that the victim was looking towards them for help is always together with the feeling of wanting to move away from the protagonists. These two variables have opposite effects, but in these data they are very strongly positively correlated. When VictimLooked is high, and MoveAway is held at its correlated value according to the regression relationship between them, then the number of verbal interventions becomes very high.

The out-group and in-group participants had about the same reported desire to stop the argument, the same level of feeling of being torn about intervening or not, and the same level of anxiety or fear. However, those in the in-group condition expressed greater anger and frustration, whereas those in the out-group condition were more likely to feel sorry for the victim, feel uninvolved or find the situation silly. Those in the in-group condition were more likely to react in a confrontational way compared with those in the out-group, who were looking more to defuse the situation. When we classify the verbal interventions as to whether they were more aimed at defusing the situation or more confrontational, amongst the out-group 17% were confrontational compared to 40% for the in-group, and 73% were defusing utterances compared to 60% for the in-group. These data suggest that the in-group were more likely to respond to the situation through anger and confrontation compared to the out-group, who were either less likely to become involved at all, or more likely to make verbal interventions to defuse the situation. This is not too surprising since by insulting the Arsenal affiliation of the victim in the in-group situation, the perpetrator was also of course indirectly insulting the participants who were all Arsenal supporters.

These data also suggest that physical interventions were more related to the safety of the victim, whereas verbal interventions were more related to safety of the self. The equations for the verbal interventions are more likely to include the ‘own safety’ than those for the physical interventions.

A final point regarding the ‘out-group’ is that in a sense it is not really an ‘out-group’ condition. Rather it is simply not ‘in-group’. Recalling the fact that all the participants were Arsenal supporters, for the ‘out-group’ the victim was portrayed as a football supporter of unknown affiliation (though highly unlikely to be Arsenal). The fact that there are clearly different results between the in-group and out-group condition is therefore a quite strong one: it is ‘in-group’ versus simply not ‘in-group’.

An important issue is the extent to which these findings are generalizable. We have shown an example where the group affiliation was a real one: strong supporters of a particular football team. This is unlike many laboratory based experiments where an abstract group affiliation is created for the purposes of the experiment. Our experimental manipulation involved activating the Arsenal affiliation through the virtual character V wearing an Arsenal football shirt, and talking enthusiastically about the club (in-group). The affiliation was not activated for those in the out-group condition, since V was not wearing an Arsenal shirt, and did not engage in enthusiastic conversation about the club. Our interest focused on the extent to which this activated (or not) psychological group affiliation impacted intervention behaviour. Our procedure was therefore designed to generate meaningful psychological group membership - the Arsenal fans were representative of a particular group. Our claim is that it is the perception that the victim belongs to the same group as the participant (in this context he was ‘one of us’) that leads people to be more likely to intervene. Hence our general hypothesis is that had the group identification been through some other means (social class, race, members of a tennis club, or even arbitrary groups conjured for an experiment) the results would have been similar.

It could be argued that the group of participants might have been too diverse in order to draw these types of conclusions. However, we argue that diversity of the sample is not relevant to this study. Arsenal fans are clearly made up of men, women, Londoners, working class, middle class, and people of different ethnic origins. However, the point is that under some circumstances they come to define themselves as members of the same group (in this case Arsenal fans) - and when this aspect of identity is important to them they are more likely to intervene to help a victim of violence when they think that person shares group membership with themselves in this context. Such group membership can be so powerful that it has been shown to at least temporarily cut across even racial bias in a context where group affiliation was created in a laboratory setting [33] , [34] . It has further been argued with respect to the famous Milgram obedience and Zimbardo Stanford prison experiments [35] that group identification is an excellent predictor of conformity [36] , [37] . For example, it was demonstrated, on the basis of the complete set of Milgram’s experiments, that the more that subjects identified with the experimenter and his causes (science, answering an important scientific problem) the more likely that they would administer the shocks. On the other hand they would be more likely to disobey the more that they identified with the Learner (representing the general community). Milgram’s original set of experiments provided a range of circumstances that led to varying degrees of identification with one of these groups (science or the community), and the degree of obedience varied accordingly.

Now we consider how our experiment could be improved. In [15] the concept of ‘plausibility’ of experiences in IVEs was introduced, referring to the illusion of participants that the virtual events are really happening (even though they know that this is not the case). It was argued that plausibility depends at least on three factors: (i) the extent to which there are events that refer personally to the participant, (ii) the extent to which the environment responds to actions of the participant, (iii) and the credibility of the scenario in terms of how much they fit expectations from a similar situation in reality. With respect to the technical setup there were no differences between in-group and out-group, and this is reflected in the fact that there are no differences in reported responses and feelings elicited through the interviews. However, the evidence does suggest ( Table 6 ) a greater tendency for the group with LookAt ‘on’ to say that their responses were realistic, and for those with LookAt ‘off’ to mention the lack of interaction. This is consistent with (i) above.

However, an overwhelming conclusion from these data is that the plausibility of the experience would be greatly improved through more interactivity (i.e., (ii) above). Recall that there was an interactive episode at the start of the experiment, where in order to establish the in-group and out-group conditions, the eventual victim did have a conversation with the participant. However, once the argument started there was no further interaction in the sense that the virtual characters did not respond to anything that the participant said or did, except for the pre-programmed LookAt factor. Another aspect of plausibility that would need to be improved based on the results of this experiment is the credibility of the scenario itself (iii). As seen from Table 9 the types of factors that drew people out of the scenario were to do with the setting rather than the technical aspects of the display: no other people around in the pub, it did not look like a real English pub, and the dialogue with the victim itself not being realistic. More than 50% of the statements made in Table 9 refer to these types of general credibility, and the remainder are specific technical issues such as ‘Illumination not realistic’ or ‘Lack of facial animation’, none of which were commonly stated. By technical issues we refer to aspects of the scenario that require only programming to solve (such as the provision of lip sync). By more general credibility issues we refer to the simulation itself - aspects that require a better understanding of what needs to be there for this to be believable as a fight in a London pub.

Apart from the introduction of interactivity and other issues relating to credibility, there are several improvements for later versions of this experiment. For example, we have not said anything about the role of the social identity of the perpetrator with respect to the participant. Moreover there are clearly other issues involved - such as participant fear of being harmed by the perpetrator. This has not been considered at all, but could also be incorporated into an experiment through manipulation of the appearance of the perpetrator (for example, to look more or less menacing). Finally, future experiments will also manipulate the number of bystanders, and thus directly tackle the question of the role of the number of bystanders in intervention.

In this paper we have shown that immersive virtual reality can be usefully exploited to study the likelihood of bystander intervention in interpersonal violent incidents. The paradigm allows the investigation of what participants did do and think during an actual experience involving violence rather than their opinion of what they might do or what they think others might do - whether based on watching a video or on a verbal description of a situation [38] . Moreover we have exploited the powerful tool of genetic programming to explore these data in a deeper way than is possible with normal statistical methods, highlighted by the elegant distinction between the in-group and out-group conditions shown in Eq. (1).

Of course, there is still no proof that what participants would do in a physically real situation would match that which we find in virtual reality. However, as reported in the introduction to this paper there is evidence to suggest that people do respond realistically in IVEs. In fact since these experiments can never be carried out in reality, ultimately the question of the validity of people’s responses to the virtual situation can never be known through laboratory based experiments of any kind. However, our approach can be used in the process of constructing theories, that can then be further tested with the use of experiments in virtual reality, and moreover ultimately examine how well these theories fit what might be found in actual experiences in the field.

To conclude, we note that the findings for this type of research can also have implications for policy. For example, by creating an atmosphere where it is thought that not running away from a violent scene is the right thing to do, and by encouraging people to ask for help when they are victims of such a situation, it may be possible to engineer pro-social behaviour in specific circumstances where this is thought desirable by policy makers, and actually to manipulate the same variables to avoid it in other situations (e.g., “do not approach this man since he is considered armed and dangerous”). Here it would be a question of using the group to enforce social norms for the prevention of violent behaviour. The key to tackling the so called ‘walk–on-by’ society lies in using the power of group identification to promote social solidarity – and to persuade and empower bystanders to intervene, in situations where this is considered by the authorities to be appropriate.

Ethics Statement

The experiment was approved by the UCL Research Ethics Committee, and was carried out under written informed consent from each participant.

The Virtual Reality System

A four screen projection system driven by a 5 PC cluster was used. We refer to this by generic name ‘Cave’ being the type of system described in [39] . The Cave has three 3 m×2.2 m back-projected screens: front, left, and right, and a 3 m×3 m front projection surface on the floor. The computers in the cluster contain Intel Pentium 3.2 GHz processors with 1 gigabyte of RAM and Nvidia Quadro FX 5600 graphics cards. The display resolution is 1024×768 pixels for each screen.

The participants were fitted with Crystal Eyes shutter glasses that were synchronized with the projectors, delivering active stereo at 45 Hz each eye. Head-tracking was performed with an InterSense IS-900 tracking device.

The program was written using the XVR programming platform as described in [40] . The virtual characters were animated using the Hardware Accelerated Library for Character Animation, HALCA [41] .

The Scenario

Two professional actors were hired to act the scene for the character animation motion capture. A Vicon motion capture system with 6 infrared cameras was used to capture their motions simultaneously. Sound was also recorded at the same time using Audacity software (audacity.sourceforge.net) with two wireless microphones attached to each actor. This raw data was then cleaned up, synchronized and split into pieces so that each one could be later assigned to a button on the interface to be played when needed during the study.

During the experiment the free-flowing conversation between the participant and V was achieved by operator control. A number of utterances had been recorded for V, each one making a statement or asking a question of the participant. Each such utterance was selected interactively by a hidden operator who could hear the responses of the participant. The operator sat by a computer screen, and all the phrases were represented visually as selectable buttons on the screen. When a button was selected (by point-and-click with the mouse) then V would say the phrase with a corresponding animation.

There was a defined script that the operator followed, but when the participant said something that fell outside of the script, then a number of general phrases could be selected by the operator in order to keep the conversation going in a natural way. For example, if the participant said something out of line, the operator could select a phrase such as “Totally agree with you” which would then be said by V. The overall effect for most of the time for most participants sounded as if it were a normal conversation between two people.

Procedures and Scenario Details

40 male participants were recruited by advertisements around the UCL campus, where we specified that we needed football supporters (‘soccer’ in American usage). They were required to complete a questionnaire that asked about their favourite team in the English Premier League and how much they supported this team. We only recruited those who supported Arsenal Football Club to the level of at least 4 on a scale from 1 (not at all) to 7 (very much so). They were paid £7 ($10–12) for their participation. The experiment was approved by the UCL Research Ethics Committee.

Upon arrival at the laboratory participants were given a short questionnaire to complete that obtained information as to their English proficiency, medication, recent alcohol intake, degree of computer game playing, and past familiarity with virtual reality. Their age was obtained at the recruitment stage, in order to ensure that no one under 18 would be recruited.

After this they were given an information sheet to read, and the same information was again told to them verbally. This described the equipment that would be used. It also warned them that some people experience a degree of nausea in virtual reality systems, and that they were free to withdraw at any time without giving reasons. They were told that they would virtually visit a bar where something was to take place and that they should feel free to interact with other people there. They were warned that experience was going to involve discussion about football and the language and situation depicted was realistic. It included the statement “If you are someone who would be put off by witnessing realistic scenes that might include bad language or aggressive behaviour, then you should not take part in this experience.”

After participants had agreed to take part they were given a consent form to read and sign, and again told that they were free to leave the experiment without having to give reasons. They were then invited to take off their shoes to enter the Cave, and put on the eyeglasses.

The participants entered the virtual reality and were asked to look around and observe the scene, which was a bar of size 4.5 meters deep by 18 meters wide. The participant was then left alone in the bar having been instructed to look around for items related to football for 2 minutes. This allowed them to become familiar with the bar and accommodate to the virtual reality display including the shutter glasses and the overall brightness of the scene. After this time, a virtual character entered the scene and started a conversation with the participant by saying “You alright mate?” in the in-group version, and “Hi, how is it going?” in the out-group one.

Not every conversation was the same across all participants in each detail due to different responses by the participants. Table S1 shows two such conversations, one when V is an Arsenal football club fan (in-group) and the other when just a general football fan (out-group).

After about 2 minutes of this conversation they were interrupted by another character (P) that had been sitting by the bar, who stood up and approached V and said to him “Oy! Have you got a problem?” and then accused V of “staring” at him ( Figure 1 ). This quickly became a strong verbal attack on V, with V remaining submissive throughout. Eventually after 140 s the P avatar started to violently push V, at which point the program ended and the participant took off the glasses and left the Cave.

The participants then were asked to complete a questionnaire about their various responses to the situation ( Table 2 ) followed by an interview and debriefing where the purposes of the experiment were explained to them, and they were asked not to discuss it with others for 3 months in case they spoke with a future participant. They were then paid the £7 ($10) and the experimental trial was complete.

Experimental Design

The experimental design was 2×2 between groups, with 10 participants arbitrarily assigned to each of the 4 cells. The two factors were Group (in-group/out-group) and LookAt (off/on). In-group was signified by V wearing an Arsenal football shirt, and maintaining an initial enthusiastic conversation about Arsenal. Out-group was signified V wearing a football shirt the same as for the in-group except without the Arsenal insignia, and during the conversation his responses were neutral and did not show much interest in the team ( Table S1 ).

LookAt referred to whether V had been programmed to occasionally look at the participant during the confrontation or not. If ‘yes’ then 5 times during the confrontation V looked toward the participant for 3 seconds. This was possible since the head-tracking streamed continual real-time data to the computer program about the position and orientation of the participant’s head. If ‘no’ there was no particular programmed action that would lead V to look towards the participant, but this may occasionally have occurred by chance (depending on where the participant was standing at the time).

Response Variables

Our major response variable of interest concerned the extent to which participants intervened during the confrontation. A video camera was mounted above the Cave looking down at the scenario and recorded each entire experimental trial. The video for two participants could not be analyzed, one for a participant in condition ‘out-group’ and LookAt ‘off’ and the other in condition ‘out-group’ and LookAt ‘on’. These participants were eliminated from all analysis involving counts of the number of interventions. The videos were analyzed independently by two different people, covering the time from when P first accosted V to the end. They had been instructed to count the number of verbal ( Verbal ) and physical ( Physical ) interventions.

Video Analysis

Figure S1 shows two stills from one of the recordings - with first P at the start of the confrontation and then a moment while the participant was intervening by placing himself between V and P and raising his hand.

First one of the experimenters carried out a review of all the videos noting the number of times that the participant said something to the virtual characters (variable nVerbalApprox ) and the number of physical interventions - meaning the number of times that the participant moved closer to the characters or reached out towards them ( nPhysApprox ). Second and independently someone not associated with the research, and not knowing its purposes was hired to carry out a complete video analysis using the ELAN system and as paid work ( www.lat-mpi.eu/tools/elan ).

The instructions were to record the total number of utterances during the relevant period ( nVerbalElan ) and also the number of physical interventions ( nPhySElan ). The instructions for the physical interventions were to regard as an intervention or an attempt at intervention an action accompanied with verbal intervention or reaching out to either of the avatars. When not accompanied by these it was considered intervention if the participant was walking with purpose towards the avatars or was followed by another form of physical or verbal intervention. Walking or stepping towards the avatars was not considered to be intervention if the participant took a step forwards, backwards, or to the left or right when far from the avatars, if they walked or stepped forwards and this was followed by them standing passively watching the avatars or when they were walking around the environment and the avatars and appeared to be simply investigating the surroundings.

Although the procedures used for the approximate and ELAN based intervention recordings were not the same the results are strongly correlated. Table S2 shows the correlation coefficients between the various measures of intervention. The approximate and ELAN based methods were consistent, with highly significant positive correlations.

Since the ELAN based method was carried out by someone not involved in the research team and more thorough, with the notion of ‘intervention’ more rigorously and conservatively defined, we base all analysis on this, so that nPhys  =  nPhysElan , and nVerbal  =  nVerbalElan .

Supporting Information

A still from a video recording from above. (a) The participant can be seen near the centre with the victim to his left, and the perpetrator to his right. (b) The participant has stepped between the victim and perpetrator standing in front of the latter and raising his hand.

Examples of Conversations between the Virtual Character V, and participant S.

Pearson Correlation Coefficients Between the Intervention Variables.

Degree of Support for the Arsenal Football Club.

MANOVA for the Number of Physical and Verbal Interactions.

Symbolic Regression.

The experimental scenario for the in-group condition.

Acknowledgments

We thank Victoria Stafford for the ELAN video analysis and the actor Morgan Roberts who performed in the video.

Funding Statement

This work was funded by the UK EPSRC project “Visual and Behavioural Fidelity of Virtual Humans with Applications to Bystander Intervention in Violent Emergencies” (EP/F032420/1; EP/F030215/1; EP/F030355/1). http://www.epsrc.ac.uk . European Senior Research Grant TRAVERSE grant number 227985. http://erc.europa.eu/ . The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

• 1. Darley JM, Latané B (1968) Bystander Intervention in Emergencies - Diffusion of Responsibility. Journal of Personality and Social Psychology 8: 377–383. [ DOI ] [ PubMed ] [ Google Scholar ]
• 2. Latané B, Rodin J (1969) A lady in distress: Inhibiting effects of friends and strangers on bystander intervention. Journal of Experimental Social Psychology 5: 189–202. [ Google Scholar ]
• 3. Latané B, Nida S (1981) 10 Years of Research on Group-Size and Helping. Psychological Bulletin 89: 308–324. [ Google Scholar ]
• 4. Fischer P, Krueger JI, Greitemeyer T, Vogrincic C, Kastenmüller A, et al. (2011) The bystander-effect: A meta-analytic review on bystander Intervention in dangerous and non-dangerous emergencies. Psychological Bulletin 137: 517–537. [ DOI ] [ PubMed ] [ Google Scholar ]
• 5. Cherry F (1995) The “stubborn particulars” of social psychology: essays on the research process: Routledge.
• 6. Harari H, Harari O, White RV (1985) The reaction to rape by American male bystanders. The Journal of social psychology 125: 633–658. [ DOI ] [ PubMed ] [ Google Scholar ]
• 7. Schwartz SH, Gottlieb A (1976) Bystander reactions to a violent theft: Crime in Jerusalem. Journal of Personality and Social Psychology 34: 1188–1199. [ DOI ] [ PubMed ] [ Google Scholar ]
• 8. Schwartz SH, Gottlieb A (1980) Bystander anonymity and reactions to emergencies. Journal of Personality and Social Psychology 39: 418–430. [ DOI ] [ PubMed ] [ Google Scholar ]
• 9. Shotland RL, Straw MK (1976) Bystander response to an assault: When a man attacks a woman. Journal of Personality and Social Psychology 34: 990–999. [ Google Scholar ]
• 10. Borofsky GL, Stollak GE, Messe LA (1971) Sex differences in bystander reactions to physical assault. Journal of Experimental Social Psychology 7: 313–318. [ Google Scholar ]
• 11. Fischer P, Greitemeyer T, Pollozek F, Frey D (2006) The unresponsive bystander: Are bystanders more responsive in dangerous emergencies? European Journal of Social Psychology 36: 267–278. [ Google Scholar ]
• 12. Levine M, Crowther S (2008) The responsive bystander: How social group membership and group size can encourage as well as inhibit bystander intervention. Journal of Personality and Social Psychology 95: 1429–1439. [ DOI ] [ PubMed ] [ Google Scholar ]
• 13. Rovira A, Swapp D, Spanlang B, Slater M (2009) The use of virtual reality in the study of people's responses to violent incidents. Frontiers in Behavioral Neuroscience 3: 59 doi: 10.3389/neuro.08.059.2009 . [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ]
• 14. Sanchez-Vives MV, Slater M (2005) From Presence to Consciousness through Virtual Reality. Nature Reviews Neuroscience 6: 332–339. [ DOI ] [ PubMed ] [ Google Scholar ]
• 15. Slater M (2009) Place Illusion and Plausibility can lead to realistic behaviour in immersive virtual environments. Philos Trans R Soc Lond 364: 3549–3557. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ]
• 16. Blascovich J, Loomis J, Beall A, Swinth K, Hoyt C, et al. (2002) Immersive virtual environment technology as a methodological tool for social psychology. Psychological Inquiry 13: 103–124. [ Google Scholar ]
• 17. Hall ET (1973) The hidden dimension. Leonardo: 94–94.
• 18. Wilcox LM, Allison RS, Elfassy S, Grelik C (2006) Personal space in virtual reality. ACM Transactions on Applied Perception (TAP) 3: 412–428. [ Google Scholar ]
• 19. Llobera J, Spanlang B, Ruffini G, Slater M (2010) Proxemics with Multiple Dynamic Characters in an Immersive Virtual Environment ACM Transactions on Applied Perception (TAP) 8: Article 3.
• 20. Bailenson J, Blascovich J, Beall A, Loomis J (2003) Interpersonal Distance in Immersive Virtual Environments. Personality and Social Psychology Bulletin 29: 819–833. [ DOI ] [ PubMed ] [ Google Scholar ]
• 21. Frankenhuis WE, Dotsch R, Karremans JC, Wigboldus DHJ (2010) Male physical risk taking in a virtual environment. Journal of Evolutionary Psychology 8: 75–86. [ Google Scholar ]
• 22. Milgram S (1974) Obedience to Authority: McGraw-Hill.
• 23. Slater M, Antley A, Davison A, Swapp D, Guger C, et al. (2006) A virtual reprise of the Stanley milgram obedience experiments. PLoS ONE 1: e39 doi: 10.1371/journal.pone.0000039 . [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ]
• 24. Cheetham M, Pedroni AF, Antley A, Slater M, Jáncke L (2009) Virtual milgram: empathic concern or personal distress? Evidence from functional MRI and dispositional measures. Frontiers in Human Neuroscience 3. [ DOI ] [ PMC free article ] [ PubMed ]
• 25. Gillath O, McCall C, Shaver PR, Blascovich J (2008) What can virtual reality teach us about prosocial tendencies in real and virtual environments? Media Psychology 11: 259–282. [ Google Scholar ]
• 26. Levine M, Cassidy C, Brazier G, Reicher S (2002) Self-categorization and bystander non-intervention: Two experimental studies. Journal of Applied Social Psychology 32: 1452–1463. [ Google Scholar ]
• 27. Levine M, Thompson K (2004) Identity, place, and bystander intervention: Social categories and helping after natural disasters. Journal of Social Psychology 144: 229–245. [ DOI ] [ PubMed ] [ Google Scholar ]
• 28. Levine M, Prosser A, Evans D, Reicher S (2005) Identity and emergency intervention: How social group membership and inclusiveness of group boundaries shape helping behavior. Personality and Social Psychology Bulletin 31: 443–453. [ DOI ] [ PubMed ] [ Google Scholar ]
• 29. Koza JR (1992) Genetic programming: on the programming of computers by means of natural selection: The MIT press.
• 30. Schmidt M, Lipson H (2009) Distilling free-form natural laws from experimental data. Science 324: 81–85. [ DOI ] [ PubMed ] [ Google Scholar ]
• 31. Burnham KP, Anderson DR (2002) Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach: Springer.
• 32. Hesse-Biber S, Dupuis P, Kinder TS (1991) HyperRESEARCH: A computer program for the analysis of qualitative data with an emphasis on hypothesis testing and multimedia analysis. Qualitative Sociology 14: 289–306. [ Google Scholar ]
• 33. Kurzban R, Tooby J, Cosmides L (2001) Can race be erased? Coalitional computation and social categorization. PNAS 98: 15387–15392. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ]
• 34. Van Bavel JJ, Cunningham WA (2009) Self-categorization with a novel mixed-race group moderates automatic social and racial biases. Personality and Social Psychology Bulletin 35: 321–335. [ DOI ] [ PubMed ] [ Google Scholar ]
• 35. Zimbardo P (2007) The Lucifer effect: Understanding how good people turn evil. Random House Inc, New York.
• 36. Reicher SD, Haslam SA, Smith JR (2012) Working Toward the Experimenter Reconceptualizing Obedience Within the Milgram Paradigm as Identification-Based Followership. Perspectives on Psychological Science 7: 315–324. [ DOI ] [ PubMed ] [ Google Scholar ]
• 37. Haslam SA, Reicher SD (2012) Contesting the ‘‘Nature’’ Of Conformity: What Milgram and Zimbardo’s Studies Really Show. PLoS Biol 10: e1001426. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ]
• 38. Banyard V (2008) Measurement and Correlates of Prosocial Bystander Behavior: The Case of Interpersonal Violence. Violence and Victims 23: 83–97. [ DOI ] [ PubMed ] [ Google Scholar ]
• 39. Cruz-Neira C, Sandin DJ, DeFanti TA, Kenyon RV, Hart JC (1992) The CAVE: audio visual experience automatic virtual environment. Communications of the ACM 35: 64–72. [ Google Scholar ]
• 40. Tecchia F, Carrozzino M, Bacinelli S, Rossi F, Vercelli D, et al. (2010) A flexible framework for wide-spectrum vr development. Presence: Teleoperators and Virtual Environments 19: 302–312. [ Google Scholar ]
• 41. Gillies M, Spanlang B (2010) Comparing and evaluating real-time character engines for virtual environments. PRESENCE: Teleoperators and Virtual Environments 19: 95–117. [ Google Scholar ]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

• View on publisher site
• PDF (501.1 KB)
• Collections

Similar articles

Cited by other articles, links to ncbi databases.

• Download .nbib .nbib
• Format: AMA APA MLA NLM

Add to Collections

Bystander Effects

• Open Access
• First Online: 31 May 2023

Cite this chapter

You have full access to this open access chapter

• Øyvind Kvalnes 2  

3519 Accesses

The lesson from research on the bystander effect is that the more people who are present when a verbal intervention is required to change the course of events in a positive way, the less likely it is that anyone will speak up. Although exceptions occur, where a high number of bystanders can increase the likelihood of intervention, the main pattern relevant for building a communication climate is that people tend to hesitate to break out of a passive group. The two main reasons for the bystander effect are diffusion of responsibility and pluralistic ignorance. Knowledge about these psychological phenomena can inform efforts to establish and maintain a well-functioning communication climate.

You have full access to this open access chapter,  Download chapter PDF

• Bystander effect
• Pluralistic ignorance
• Diffusion of responsibility
• Communication climate
• Devil’s advocate

A guest lecturer steps onto the podium in an auditorium filled with about 100 students. She is introduced as a professor from Copenhagen Business School, and the students are encouraged to raise their hands to comment and ask questions during the lecture. What they do not know is that the lecturer in front of them is an actor who has been instructed to talk nonsense for an hour. She can use academic terminology and hint at knowledge of the core concept in the course, but in an unstructured and nonsensical manner. This is a social psychology experiment conducted for pedagogical purposes to see whether the students will intervene or remain silent during the lecture. The arguments they will hear from the podium will make little sense, both in the course context and otherwise. How will the students respond to the situation?

The false professor begins to speak. She uses course concepts, but haphazardly and inconsistently. Unease spreads among the students, but none of them raises their hand to ask for clarification or question the professor’s claims. For a full lecture hour, the students silently listen to the nonsense from the podium. Then the course coordinator tells them that they have taken part in an experiment and that the professor is an actor. The coordinator opens the discussion for reflection and comments. Students shyly admit that they did not understand what the professor was talking about but assumed that others did because they had also remained silent. Others say that they thought it would be impolite to challenge the guest lecturer, a visitor from another country who might have misunderstood the context of the course.

The purpose of the experiment was to introduce the students to the bystander effect, a psychological phenomenon where the number of people present in a situation tends to affect decision-making (Darley and Latané 1968 ; Manning et al. 2007 ). Research on this phenomenon has indicated that the likelihood that someone will provide help to a victim decreases with an increase in the number of people who are present. Recent studies have explored bystander effects in the context of how likely it is that people will intervene when they witness cyberbullying (You and Lee 2019 ), sexual assault (Kettrey and Marx 2021 ), and violence (Levine et al. 2020 ). The general pattern is that the more people who are at the scene, the less likely it is that anyone will act and provide help. Some studies indicate that bystanders in a group are more likely to intervene in the presence of danger. Being part of a larger group can provide protection and enhance initiatives rather than lead to passivity (Liebst et al. 2021 ).

Bystander effects can occur in situations at work where a number of people spot something that should be verbally addressed and can choose to take an initiative or not (Kvalnes 2017 ). In a study involving employees and managers working in a Fortune 500 organisation, Hussain et al. ( 2019 ) introduced the concept of voice bystander effect. They found that the more some information is shared among employees, the less any particular employee feels individually responsible for bringing up that information with their managers. Countering bystander effects can be a core concern for building and maintaining a constructive communication climate in an organisation or project. Such effects can occur at critical quality moments in a project process. Several project members can be aware that the person in charge of operations has made a crucial mistake, a weakness exists in the execution plan, or one particular team member has not had sufficient training and is not qualified for the task.

From the outset, one would assume that the more project members who know about these issues and who are in position to address them, the more likely it would be that one of them would do so. Research on bystander effects points in the opposite direction. It seems that the higher the number of project members who know, the more likely it is that all will remain passive and silent bystanders. In a debriefing of the situation, the project members’ silence may appear to be mysterious. Why does none of them speak up? Knowledge about bystander effects can help one to understand the lack of initiative and demystify the silence.

The chapter on critical quality moments included examples where silence or a lack of initiative from those present can be explained in terms of the bystander effect and its causes. Research points to diffusion of responsibility as one main reason why people are passive in numbers (Darley and Latané 1968 ; Barron and Yechiam 2002 ). We tend to make the mistake in moral mathematic of thinking that responsibility is a unit we share evenly and fairly among those present (Parfit 1984 ). In the fake lecturer example, the 100 students in the auditorium each thought they only had 1/100 of the responsibility for speaking up about the professor who talked nonsense on the podium. With so little responsibility, it is easy to justify to oneself that one has remained silent despite having misgivings. If 15 engineers look critically at the drawings and specifications of an installation, each of them can mistakenly think they have 1/15 of the responsibility for actually applying their expertise to identify possible weaknesses. If five people are supposed to control the documentation ahead of a money transfer, each of them can mistakenly assume that they only have one-fifth the responsibility to look closely at the details. In all three examples, people can also remain passive due to an assumption that the others are shaper and more alert than they are at present.

A nurse told me that diffusion of responsibility could set in even when only two persons are involved. She worked at a hospital where from time to time they would treat prisoners, some of whom were considered dangerous. It was important to make sure that after treatment they did not leave the hospital with sharp objects that they could use to harm others with later. To make sure this did not happen, both a trained nurse and a police officer would search each prisoner thoroughly upon departure. First, the nurse would conduct a search, and then hand over the prisoner to the police, who would conduct a second search. On one such occasion, the system failed, and a prisoner was able to smuggle a surgical knife out of the hospital. Both the nurse and the police officer had thought that the other person would do the job properly. They appeared to have split the responsibility for searching the prisoner in half, and each considered themselves to have only half the responsibility for doing a thorough search of the prisoner.

The second main reason for bystander effects goes under the label of pluralistic ignorance , a tendency to adjust our initial interpretation in light of what we take to be other people’s interpretation of it (Miller and McFarland 1987 ; Rendsvig 2014 ). Each of the students in the auditorium may initially have thought that the professor was talking nonsense but suppressed that thought when seeing the other students remain passive and seem to understand what she was saying. They interpreted the other students’ passivity as social proof (Cialdini et al. 1999 ) that everything was as it should be. In the engineering example, each of the 15 engineers may have had doubts about the drawings in front of them but kept silent because the other 14 seemed satisfied with what they saw. Similarly, each of the five members of the team who were responsible for quality control of the payment documents could have had misgivings about small details, but still kept silent due to the lack of protest from any of the others.

One common countermeasure against bystander effects is to give one person the role of being devil’s advocate (Nemeth et al. 2001 ; Brohinsky et al. 2021 ). This person is responsible for being extra critical and looking for weaknesses in the proposals on the table. The strategy aims to avoid diffusion of responsibility because it places the task of speaking up about critical issues firmly in the hands of one person. It also attempts to counter pluralistic ignorance, because the devil’s advocate is not supposed to adjust personal judgement to fit in with what other people take to be the case.

In the fake professor example, the course instructor could have elevated one student to be devil’s advocate and adopt a stance that is critical of what the guest professor had to say, instead of asking the whole group what they thought about the guest lecturer. If all the individuals in a group know in advance that at some point they may be chosen to give a response, then they are likely to be more alert and prepared than they would be if the responsibility were spread thinly out to everyone in the room and would remain so for the rest of the proceedings.

Informal devil’s advocates can operate in a range of organisational contexts. Some individuals always tend to speak up and take responsibility for being critical. They have not been formally assigned to that role, but their dissent emerges as a recurring pattern of the group process (Brohinsky et al. 2021 ). Colleagues will look to those individuals to voice a concern or point to weaknesses and mistakes in a proposal, because that is what they normally do. On one occasion, a group of Norwegian bank executives were gathered to make a major decision about the way forward. The CEO presented his preferred alternative and the arguments for it. Unease spread among the other executives, much like when a conductor gives the wrong tone to the singers in a choir. Most of them sensed that the CEO built his argument on a faulty assumption.

In this group, one person usually took on the role of being devil’s advocate. She would never hesitate to address flawed assumptions or weaknesses in an argument. On this occasion, she was completely silent. The people around the table glanced at her, waiting for an initiative, but it never came. No one else voiced their concerns or pointed to the weakness in the basic assumption. A decision was made in accordance with the CEO’s suggestion. Eventually, it led to exactly the sort of negative outcome that the other executives had feared.

In a debriefing, the group met to discuss why no one had opposed the CEO’s proposal. Many explained that they had expected the devil’s advocate to speak up, and that because she was silent, they began to have doubts about their judgement of the case. Several of them had reasoned that because she was passive, they had probably misjudged the proposal. Why had she not spoken? She explained that on the day she had been distracted by an ongoing dramatic event in her family and was checking her cell phone for news. She had not been able to focus on work-related issues. The lesson the group learned from this process was that the role of being devil’s advocate should circulate among them and not lie with the same person every time.

We have seen that bystander effects can occur in situations where an initiative is needed to stop a negative turn of events. Passivity in a group can also be a challenge in situations where an opportunity exists to provide acknowledgement and praise to individuals or groups in the organisation. A group of colleagues has done an exceptional job and they deserve vocal appreciation. Here is a critical quality moment. People are about to leave the meeting, and now is the chance to express praise in front of the whole unit. Whose responsibility is it to take this initiative?

The number of people present in such situations can create passivity even here. These colleagues really deserve a show of appreciation for what they have done. Many are present and in a position to raise their voice and provide it. Who will take responsibility for doing it in a group of 50? It could be the leader of the unit, but if that person for some reason is incapacitated, someone else needs to take charge. Considering what we know about bystander effects and passivity in numbers, the unit could identify someone to take on the role of being what we can call God’s advocate, with special responsibility to identify excellence and effort in the workplace and speak up about them.

Recent empirical research on the bystander effect provides a more nuanced outlook than the initial idea that a higher number of people present makes it less likely that anyone would intervene. The bystander effect weakens when social bonds exist between the person needing help and the bystanders (Levine and Manning 2013 ). Studies of surveillance camera footage of violent episodes indicate that danger sharply increases the likelihood of intervention from bystanders (Liebst et al. 2021 ; Lindegaard et al. 2021 ). These findings have in common that a sense of social connectedness and importance can weaken and nullify bystander effects. When we care about those affected by the events in front of us and see that what is happening will have a significant effect on them, we may be more likely to intervene and speak up than if we were witnessing strangers in trouble but not in real danger.

The purpose of this chapter has been to show that knowledge about bystander effects—what causes them as well as what strengthens and weakens them—is highly relevant for efforts to build and maintain a constructive communication climate. Diffusion of responsibility can occur even among the best of colleagues. Pluralistic ignorance can also emerge in settings where colleagues glance at each other, without saying a word. We have seen that it can make sense to appoint a devil’s advocate and a God’s advocate in work settings to place responsibility for making interventions firmly with specific individuals. We have also observed that these roles need to circulate and not stay with the same individuals. Long-serving advocates may have bad days where they are distracted from seeing events clearly or lack energy to speak up. On those days, their silence can be interpreted wrongly as a sign that everything is fine, and therefore no one calls for an intervention. An alternative is to give more people experience in keeping a critical and appreciative eye on proceedings at work and speaking up about what they see, which may serve to mobilise dissent and appreciation that is more authentic in work contexts.

Barron, G. and E. Yechiam (2002). “Private e-mail requests and the diffusion of responsibility.” Computers in Human Behavior 18(5): 507–520.

Article   Google Scholar  

Brohinsky, J., et al. (2021). “The Devil’s Advocate.” Science & Education: 1–22.

Google Scholar  

Cialdini, R. B., et al. (1999). “Compliance with a request in two cultures: The differential influence of social proof and commitment/consistency on collectivists and individualists.” Personality and Social Psychology Bulletin 25(10): 1242–1253.

Darley, J. M. and B. Latané (1968). “Bystander intervention in emergencies: Diffusion of responsibility.” Journal of personality and social psychology 8(4): 377–383.

Hussain, I., et al. (2019). “The voice bystander effect: How information redundancy inhibits employee voice.” Academy of Management Journal 62(3): 828–849.

Kettrey, H. H. and R. A. Marx (2021). “Effects of bystander sexual assault prevention programs on promoting intervention skills and combatting the bystander effect: A systematic review and meta-analysis.” Journal of Experimental Criminology 17(3): 343–367.

Kvalnes, Ø. (2017). Fallibility at work: Rethinking excellence and error in organizations. London, Palgrave Macmillan.

Book   Google Scholar  

Levine, M. and R. Manning (2013). “Social identity, group processes, and helping in emergencies.” European Review of Social Psychology 24(1): 225–251.

Levine, M., et al. (2020). “Rethinking the bystander effect in violence reduction training programs.” Social Issues and Policy Review.

Liebst, L. S., et al. (2021). “Cross-national CCTV footage shows low victimization risk for bystander interveners in public conflicts.” Psychology of violence 11(1): 11.

Lindegaard, M. R., et al. (2021). “Does Danger Level Affect Bystander Intervention in Real-Life Conflicts? Evidence From CCTV Footage.” Social Psychological and Personality Science: 19485506211042683.

Manning, R., et al. (2007). “The Kitty Genovese murder and the social psychology of helping: The parable of the 38 witnesses.” American Psychologist 62(6): 555.

Miller, D. T. and C. McFarland (1987). “Pluralistic ignorance: When similarity is interpreted as dissimilarity.” Journal of personality and social psychology 53(2): 298.

Nemeth, C., et al. (2001). “Devil's advocate versus authentic dissent: Stimulating quantity and quality.” European Journal of Social Psychology 31(6): 707–720.

Parfit, D. (1984). Reasons and persons. Oxford, Oxford University Press.

Rendsvig, R. K. (2014). “Pluralistic ignorance in the bystander effect: informational dynamics of unresponsive witnesses in situations calling for intervention.” Synthese 191(11): 2471–2498.

You, L. and Y.-H. Lee (2019). “The bystander effect in cyberbullying on social network sites: Anonymity, group size, and intervention intentions.” Telematics and Informatics 45: 101284.

Download references

Author information

Authors and affiliations.

Dept Leadership & Organizational Devt, BI Norwegian Business School, Oslo, Norway

Øyvind Kvalnes

You can also search for this author in PubMed   Google Scholar

Rights and permissions

Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License ( http://creativecommons.org/licenses/by/4.0/ ), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.

Reprints and permissions

Copyright information

© 2023 The Author(s)

About this chapter

Kvalnes, Ø. (2023). Bystander Effects. In: Communication Climate at Work. Palgrave Macmillan, Cham. https://doi.org/10.1007/978-3-031-28971-2_3

Download citation

DOI : https://doi.org/10.1007/978-3-031-28971-2_3

Published : 31 May 2023

Publisher Name : Palgrave Macmillan, Cham

Print ISBN : 978-3-031-28970-5

Online ISBN : 978-3-031-28971-2

eBook Packages : Business and Management Business and Management (R0)

Share this chapter

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Publish with us

Policies and ethics

• Find a journal
• Track your research