The Bystander Effect · The CCTV Record · The Capable Guardian · Collective Efficacy · The Duty to Rescue · Bystander CPR

Bystander Intervention: What the Record Shows About Who Steps In, What Stops Them, and Why the Answer Changes by Place

For half a century the standard account of bystanders came from a 1964 New York murder and a set of laboratory experiments: witnesses in a group, the account held, are less likely to help than a witness alone. The archival record has since dismantled the founding story—there is no evidence for the 38 witnesses of the original newspaper account—and camera footage of 219 real public conflicts in Lancaster, Amsterdam, and Cape Town shows at least one bystander intervening in 90.9 percent of cases, with each additional bystander raising the odds that a victim received help. What still suppresses action is narrower and more specific: ambiguity about whether an emergency is occurring, a telephone call nobody has been assigned to make, a camera raised in place of a hand, an assault read as a domestic dispute, and a weapon. Beneath the individual decision sit local structures—the capable guardian of routine activity theory, the collective efficacy measured block by block in Chicago, the informal codes that govern where police protection is not trusted—and beneath those, the law, which differs by jurisdiction: most American states impose no general duty to rescue, while France and Germany criminalize the failure to assist a person in danger. The record’s conclusion is geographic. Protection is a property of a place and its people, and a traveler carries none of it with him.

By Eunice, the desk · Collected and analyzed at 1045 ET, Sunday, November 15, 2026.

The question this report answers is the one a person in trouble is actually asking: if something happens to me here, in this street, in this city, will anyone step in—and what decides it? The research record, assembled over six decades, gives an answer that is more favorable than the famous story and more conditional than the reassurance. The famous story is the bystander effect, taught for fifty years as the finding that the presence of other witnesses makes each witness less likely to help. That finding is real in the laboratory conditions that produced it. But the murder that founded the field did not happen the way the newspapers reported it, and when researchers finally obtained camera footage of real public conflicts—219 of them, in three cities on three continents—they found intervention in nine cases out of ten, more bystanders producing more intervention, and interveners seldom harmed. The suppressors that survive scrutiny are specific: an event nobody can classify, a call nobody has been assigned, a fight read as a private matter, a weapon, and, in the medical case that has been measured most precisely, a rescuer’s fear of a lawsuit that has never once succeeded. Around those individual decisions stands the structure that decides most cases before they begin: whether the place has guardians, whether its residents trust one another enough to act for the common good, which informal code governs its public spaces, and what its law requires of a witness. Every one of those terms is local. That is the finding the rest of this report documents, study by study and statute by statute.

The founding story, and what the record did to it

The story as printed. On March 13, 1964, Catherine “Kitty” Genovese was murdered outside her apartment building in Kew Gardens, Queens. Two weeks later the New York Times reported that 38 respectable, law-abiding citizens had watched the killer stalk and stab her in three separate attacks over more than half an hour, and that none had called the police until after she was dead. The story became the founding parable of a research field and a fixture of textbooks: a crowd of witnesses, each waiting for another to act, a woman dead of their collective indifference.

The story as documented. In 2007, Rachel Manning, Mark Levine, and Alan Collins re-examined the case in the American Psychologist under the title “The Kitty Genovese murder and the social psychology of helping: The parable of the 38 witnesses.” Their archival review found no evidence for the 38 witnesses, no evidence that witnesses observed the murder as reported, and no evidence that witnesses remained inactive. The two attacks occurred in different locations, the second in an enclosed stairwell visible to almost no one; most of those who heard anything heard noises they could not classify. The Assistant District Attorney who prosecuted the case found only about six eyewitnesses usable at trial. One witness shouted from a window during the first attack and drove the attacker off. The Times itself later acknowledged that its 1964 account had grossly exaggerated the number of witnesses and what they had perceived. The parable, the reviewers concluded, survives as a parable—a teaching story whose evidentiary base does not support it.

The laboratory finding stands, in its limits. The experiments the murder prompted were real and their result has been replicated. In John Darley and Bibb Latané’s 1968 seizure study, participants who believed they were the only witness to a fellow participant’s seizure sought help 85 percent of the time; those who believed one other person was present helped 62 percent of the time; those who believed four others were present helped 31 percent of the time. The mechanism they named—diffusion of responsibility, each witness’s obligation diluted across the group—is a genuine property of the situations they built: ambiguous events, strangers who cannot see one another, and no way to know whether anyone else has acted. A 2011 meta-analysis by Peter Fischer and colleagues, covering 105 effect sizes and more than 7,700 participants in the Psychological Bulletin, put the overall effect at g = −0.35 and found it attenuated precisely where the laboratory’s conditions break down: in dangerous emergencies, when a perpetrator is present, and when the costs of intervening are physical. The clearer and graver the emergency, the weaker the effect.

What the cameras show

The CCTV record. In 2020, Richard Philpot, Lasse Liebst, Mark Levine, Wim Bernasco, and Marie Lindegaard published in the American Psychologist the first cross-national analysis of bystander behavior in real public conflicts, coded from police and municipal camera footage: 219 conflicts—95 in Lancaster, England, 63 in Amsterdam, and 61 in Cape Town. At least one bystander intervened in 90.9 percent of the conflicts. Intervention was usually plural: the average conflict drew 3.76 interveners. Each additional bystander present raised the odds that the victim received help by roughly 10 percent—the reverse of the laboratory prediction at the level of the event, which is the level at which a victim experiences it. Intervention was defined broadly, as the record of a camera allows: stepping between the parties, pulling an aggressor away, a calming gesture or touch, consoling the victim. The footage carried no audio, so purely verbal interventions went uncounted; the 90.9 percent is, on that account, a floor. The three cities differ sharply in their levels of street violence, and the researchers found no significant difference in intervention rates among them.

The risk to the intervener, measured. A companion study by Liebst and colleagues, published in Psychology of Violence in 2021, examined 93 of the recordings and the 417 interveners they contained, and coded what happened to the people who stepped in. Interveners were physically harmed in 3.6 percent of cases—12 of the 337 who intervened purely to de-escalate. Non-interveners were virtually never harmed: 0.2 percent, one of 602. The severity of the conflict at the moment of intervention did not increase the intervener’s risk. The measured picture is neither of the two popular ones: stepping in is not free of risk, and it is not the near-certain injury of the cautionary tale. It is a small, quantifiable risk, run mostly by people who de-escalate rather than fight.

The model the cameras revised. The popular training literature of the same decades organized the public into three kinds of people. The frame is Lt. Col. Dave Grossman’s, set out in the essay “On Sheep, Wolves, and Sheepdogs” in On Combat (2004), with roots in his On Killing (1995): the sheep, ordinary people not capable of hurting one another except by accident or under extreme provocation; the wolves, who prey on them; and the sheepdogs—soldiers, police, and armed citizens who make the moral choice to protect—who stand between. The essay circulated widely through American police and civilian firearms training; businesses in at least nine states trade under the sheepdog name, and the frame reached its largest audience in the 2014 film American Sniper, whose father delivers the three-types speech at the family dinner table. As a description of what cameras record, the typology does not survive the footage: in the 219 conflicts, the interveners were not a distinct protective caste but the ordinary bystanders themselves, several per incident, most of them unarmed and most of them calming rather than fighting. The desk files the frame as what the record shows it to be—the most successful piece of bystander folk theory of its era, and a publishing phenomenon the CCTV data has overtaken.

What still stops people

Ambiguity is the first gate. Before responsibility can diffuse, an event must be classified as an emergency, and classification is where the record shows most failures begin. A scream may be an attack or an argument; a man slumped in a doorway may be dying or sleeping. Darley and Latané’s own model made interpretation the second step of five, ahead of responsibility, and their subjects in group conditions looked to one another’s calm faces for the classification—pluralistic ignorance, each witness privately alarmed and publicly composed, each reading the others’ composure as evidence that nothing is wrong. The remedy the training literature converged on follows directly from the mechanism: a specific appeal to a specific person—the assigned task, “you, call 911”—which destroys the ambiguity and the diffusion in one sentence. The unspecific scream leaves both intact.

Diffusion survives on the telephone. The camera studies count physical intervention, which is visible. The act that remains most subject to diffusion is the one the cameras cannot see: the call for help, which each witness can assume another has made. The Genovese parable, stripped of its false witness count, reduces to this residue—a call delayed because no one person owned it. Modern dispatch records and the intervention literature treat the assigned call as the standard countermeasure for the same reason.

The camera in the hand. The smartphone era added a substitute action the classic literature did not have: recording. A witness filming an incident is performing a visible, effortful response that is not intervention—it documents the event for later judgment while leaving the event itself untouched. Researchers and trainers now list filming among the substitute behaviors an emergency can absorb, alongside watching and waiting for a professional. The record here is observational rather than experimental; the desk prints it as the field’s current description, not as a measured rate.

The domestic reading. The sharpest measured suppressor is a classification, not a crowd. In Lance Shotland and Margaret Straw’s 1976 field experiment, confederates staged a fight between a man and a woman within hearing of bystanders. When the woman shouted “Get away from me; I don’t know you,” bystanders intervened 65 percent of the time. When she shouted “Get away from me; I don’t know why I ever married you,” intervention fell to 19 percent. Nothing about the violence changed; only its category. An assault read as a domestic dispute is reclassified as a private matter, and the privacy norm—the rule against intruding on a couple—outweighs the emergency. The finding has been replicated in the literature’s teaching ever since, and it identifies the mechanism precisely: the bystander is not indifferent to the victim; he has filed the event under a heading that forbids him to act.

Weapons change the act, not the intent. The CCTV researchers report the same pattern at the other end of severity: where a weapon is present, as in robberies, bystanders rarely intervene physically during the event itself, and commonly act the moment it ends—pursuing the robber, comforting the victim, giving accounts to police. The desk prints this as the research group’s description of its footage; a settled intervention rate for armed robbery does not yet exist in the published record. Read with the Liebst risk figures, the pattern is consistent: bystanders calibrate the form of help to its danger, substituting aftermath assistance for interposition when a weapon is in hand.

Freezing, stated as firmly as the evidence allows. The freeze response is documented most rigorously in victims, not bystanders. In a 2017 study in Acta Obstetricia et Gynecologica Scandinavica, Anna Möller and colleagues assessed 298 women at a Stockholm emergency clinic after sexual assault; 70 percent reported significant tonic immobility—an involuntary, reflexive paralysis under threat—during the assault, and 48 percent reported extreme immobility. The response is a property of the nervous system, not a decision, and it predicts later post-traumatic stress. Its extension to bystanders—the witness who means to act and finds he cannot—is discussed in the stress literature as part of the same defense cascade, but it has not been measured in bystander populations with anything like the victim research’s precision. The desk prints the distinction because the popular version does not: freezing is a proven fact about people under direct threat, and a plausible, unquantified one about the people watching.

The structure beneath the moment

The capable guardian. In 1979, Lawrence Cohen and Marcus Felson published the routine activity approach in the American Sociological Review: a predatory crime requires the convergence in space and time of three elements—a motivated offender, a suitable target, and the absence of a capable guardian. A guardian, in their formulation, is not necessarily a police officer; it is any person or arrangement whose presence discourages the act—a neighbor on a porch, a passenger in a carriage, a concierge at a desk. The theory relocates protection from character to geometry: the question is not whether a street contains brave people but whether it contains present ones. The bystander of the psychological literature and the guardian of the criminological literature are the same person, studied from opposite ends of the same event.

Collective efficacy. Robert Sampson, Stephen Raudenbush, and Felton Earls measured the neighborhood-level version in Chicago. Their 1997 Science study surveyed 8,782 residents across 343 neighborhood clusters and constructed a measure they called collective efficacy: social cohesion among neighbors combined with their willingness to intervene on behalf of the common good—the likelihood, as the survey asked it, that neighbors would act if children were skipping school, spray-painting a building, or fighting in the street. Collective efficacy was negatively associated with violence, controlling for individual characteristics, measurement error, and prior violence; the associations of concentrated disadvantage and residential instability with violence ran largely through it. Two neighborhoods with the same poverty can carry different protection, and the difference is a property of the neighborhood—its trust and its habits of intervention—not of any individual in it. Protection, in this literature, is infrastructure: built slowly, held in common, and invisible to the person it covers until the day it is spent.

Where the formal coverage lapses. Elijah Anderson’s Code of the Street (1999), from years of ethnographic fieldwork in Philadelphia, describes the system that operates where the formal one is not trusted. In the neighborhoods Anderson studied, police protection is perceived as unreliable and the courts as remote, and an informal code governs public space in their place: respect is the currency, it is hard-won and easily lost, and a credible willingness to answer a challenge is the ordinary person’s shield against predation. Anderson distinguishes “decent” families, who hold mainstream values, from “street” families oriented to the code—and shows that both must navigate the code in public, because the code governs the space regardless of the values of the person crossing it. The code is, in his account, an adaptation: “the code of the street emerges where the influence of the police ends,” a people’s law standing where the state’s law is not believed to run. For the bystander question, the consequence is direct: in such a place, intervention follows the code’s rules—who may step in, on whose behalf, at what cost to standing—and not the rules an outsider imports.

Honor, dignity, and the third party. Bradley Campbell and Jason Manning, in “Microaggression and Moral Cultures” (Comparative Sociology, 2014) and their 2018 book The Rise of Victimhood Culture, sort moral cultures by where they send a grievance. In honor cultures, which arise where legal authority is weak, a person answers an insult or an injury himself, and his reputation for answering is his protection. In dignity cultures, worth is inherent and need not be defended; slights are tolerated or, at the limit, taken to third parties—courts and police empowered to wield violence on the complainant’s behalf. In the victimhood culture Campbell and Manning describe emerging, sensitivity to slight combines with heavy reliance on third parties: conflicts are handled by complaint to authorities, and moral status attaches to the victim who secures the authority’s intervention. The three cultures are, for this report’s subject, three different answers to the question of who guards: oneself, the law as a last resort, or the institution as a first resort. Each answer trains different bystander behavior, and each is invisible to the person raised inside it, who experiences his own answer as the nature of things.

One country, two baselines. Richard Nisbett and Dov Cohen’s Culture of Honor (1996) demonstrated the variation inside a single nation’s borders. In their University of Michigan experiments, a confederate bumped into student subjects in a narrow hallway and called them an insulting name. Insulted white men from the American South showed rises in cortisol and testosterone, judged their masculinity threatened, and thought more aggressively; in a subsequent test of nerve, walking toward a 6-foot-3, 250-pound football player down the same narrow hall, insulted Southerners held their course until roughly three feet of separation remained, against nine feet for un-insulted Southerners; Northerners yielded at about six feet regardless of the insult, and 65 percent of insulted Northern men smiled or laughed the encounter off. Same university, same hallway, same law—two different systems of protection and response, carried in the persons of the men themselves, each acquired in childhood in a different region of one country.

The prevention paradox. The British epidemiologist Geoffrey Rose stated the paradox of population prevention in 1981 and in The Strategy of Preventive Medicine: “a preventive measure which brings much benefit to the population offers little to each participating individual.” Applied to guardianship, the paradox runs as follows. Where collective efficacy is high and guardians are present, assaults are rare—and their rarity is then available as evidence that the guardians were never needed, that the threat was exaggerated, that the caution of the protected is a kind of naivety about a world that is, in their experience, safe. The safety is real and the inference is false; the measure’s success has consumed its own evidence. The desk states the application as its own synthesis, in those terms: Rose formulated the paradox for preventive medicine, and the structure of the argument transfers whole to the prevention of violence, where the protected population likewise cannot point to the events that did not happen to it.

The counterweight: disaster. The strongest evidence against the fragility of all this comes from the events where the formal structure is removed entirely. Disaster sociology’s consistent finding, across a century of case studies, is that the public does not panic and does not prey: it rescues, feeds, and organizes. Rebecca Solnit’s A Paradise Built in Hell (2009) assembled the record from the 1906 San Francisco earthquake to Hurricane Katrina—strangers navigating boats down flooded streets, kitchens and shelters arising without instruction—and set against it the documented failure mode, which the Rutgers sociologists Lee Clarke and Caron Chess named “elite panic” in Social Problems in 2008: authorities in a disaster fearing the public itself, withholding information, protecting property over people, and treating citizen relief as a threat to order. The counterweight matters to the bystander record because it locates the default correctly. Cooperation is the baseline behavior of an unguarded crowd; what varies by place is not the presence of decency but the standing arrangements—trust, codes, law—that either route it or suppress it on an ordinary Tuesday.

The law, jurisdiction by jurisdiction

The United States: no general duty. American common law imposes no general duty to rescue. A bystander who watches a stranger drown from a safe bank, and does nothing, commits no crime and owes no damages in most states; the law treats the failure to act as a moral event, not a legal one. A handful of states have written narrow exceptions. Vermont’s statute, 12 V.S.A. § 519, in force since 1968, requires a person to give reasonable assistance to another exposed to grave physical harm, when it can be done without danger to himself; a willful violation carries a fine of up to $100. Minnesota’s § 604A.01 imposes the same duty, classifies violation as a petty misdemeanor with a maximum fine of $300, and counts obtaining help from law enforcement or medical personnel as reasonable assistance. Rhode Island, Wisconsin, and Louisiana carry provisions of the same family. Separately, every state maintains Good Samaritan protections that run the other direction: they shield a volunteer who does act from liability for ordinary negligence, withholding protection for gross negligence or willful misconduct. The American settlement is thus asymmetric—no duty to act, and a statutory promise that acting in good faith will not be punished.

France: the duty as crime. Article 223-6 of the French Penal Code, non-assistance à personne en danger, makes it an offense, punishable by five years’ imprisonment and a €75,000 fine, willfully to fail to prevent by one’s immediate action a crime against a person’s physical integrity, or to fail to bring assistance to a person in danger, where the action can be taken without risk to oneself or to others. The duty is personal, general, and criminal; the bystander in France is not a volunteer but a conscript of the code, and the protected risklessness is written into the offense itself.

Germany: the duty with the helper protected. Section 323c of the German Criminal Code, unterlassene Hilfeleistung, punishes failure to render assistance in an accident, a common danger, or an emergency with up to one year’s imprisonment or a fine, where assistance is necessary and can reasonably be expected—in particular, where it is possible without substantial danger to oneself and without breaching other important duties. The same section criminalizes obstructing a person who is helping. German law then protects the helper it conscripts: first-aiders are covered by statutory accident insurance, and their liability for damage caused in the attempt is limited to intent and gross negligence under § 680 of the Civil Code. The duty is enforced. In the widely reported Essen case of 2016, bank customers stepped over a man who had collapsed in the lobby; those identified were fined several thousand euros, while the customer who telephoned the emergency services was not indicted—the call, in the German as in the Minnesota formulation, discharges the duty.

The CPR gap: a phantom risk with measured effects. The clearest documented case of a false legal belief suppressing real aid is cardiac arrest. In a 2018 analysis of 19,331 adult out-of-hospital cardiac arrests in the Resuscitation Outcomes Consortium registry (United States and Canada, 2011–2015), published in Circulation: Cardiovascular Quality and Outcomes, Amanda Blewer and colleagues found that in public settings women received bystander CPR in 39 percent of cases against 45 percent for men—an adjusted odds ratio of 1.27 in men’s favor—while in private settings there was no difference, 35 against 36 percent. The gap is a public phenomenon, and the stated reasons are on record: in the survey research of Sarah Perman and colleagues (Circulation, 2019), respondents named fear of being accused of sexual or physical assault, fear of causing injury, and the perceived impropriety of touching and exposing a woman’s chest; in a 2024 follow-up, fear of a sexual-assault or harassment accusation was the leading stated reason in the male-rescuer scenario, named by 36.2 percent. Against these stated fears stands the legal record in full: the American Heart Association’s formulation, repeated across the resuscitation literature, is that no lay rescuer has ever been successfully sued for performing CPR. The risk that measurably reduces the aid women receive in public is a risk with no recorded instance. A legal environment can suppress help twice over—by what it punishes, and by what it is merely believed to punish.

Why the protection does not travel

The conclusion the record supports. Every term in the bystander equation is set locally. Whether a conflict draws intervention depends on the guardians a place keeps present and the collective efficacy its residents have built—Chicago’s measure varied block by block inside one city. Whether an assault is classified as public or private follows the local code, which in Anderson’s Philadelphia is not the state’s code, and in Nisbett and Cohen’s hallway is not the same code at the two ends of one country. Whether a witness is legally a volunteer or a conscript depends on the jurisdiction’s statute: nothing in Vermont, $100; nothing in most of the United States; five years and €75,000 in France; a year in Germany. Whether a rescuer’s hand is stayed by fear of suit depends on beliefs about a legal system, and the beliefs, as the CPR record shows, need not correspond to any case ever decided. Even the traveler’s own dispositions are imports: the Southern undergraduate in Michigan carried his baseline with him into the hallway, which is the one direction in which protection does travel—as habit, not as cover.

What a traveler actually carries. A person who crosses a border, or a state line, or the boundary between two neighborhoods, takes with him his classification habits, his sense of what a crowd will do, his estimate of what the law asks of him, and his confidence that the arrangements he grew up inside are the arrangements of the world. The record assembled here indicates that each of those is a local product. The crowd in Lancaster, Amsterdam, and Cape Town intervened at the same rate; the law in Paris and Berlin makes the same witness a criminal for the omission that New York treats as no offense at all; the Chicago block with high collective efficacy and the block without it can share a precinct, a poverty rate, and a police force. The protection a person feels in a strange place is therefore not a measurement of that place. It is a memory of his own, worn as if it were a fact about his surroundings. The desk’s finding, stated in the register of the record: protection is a property of the place and its people. It is not issued to the traveler, it is not stamped in his passport, and it does not cross the border with him. What crosses is only the belief that it has—and the belief, unlike the protection, works everywhere it is carried, right up to the moment it is tested.

The timeline

1964 — Kitty Genovese is murdered in Kew Gardens, Queens; the New York Times reports 38 witnesses who did nothing. 1968 — Darley and Latané publish the seizure experiments and name diffusion of responsibility; Vermont enacts the first American duty-to-assist statute of the modern group. 1976 — Shotland and Straw’s staged fight shows intervention falling from 65 to 19 percent when the assault is framed as a marriage. 1979 — Cohen and Felson publish the routine activity approach and the capable guardian. 1981 — Geoffrey Rose states the prevention paradox. 1996 — Nisbett and Cohen publish Culture of Honor and the hallway experiments. 1997 — Sampson, Raudenbush, and Earls publish the collective efficacy study of 343 Chicago neighborhoods in Science. 1999 — Elijah Anderson publishes Code of the Street. 2004 — Grossman’s On Combat carries the sheep, wolves, and sheepdogs frame to the training world. 2007 — Manning, Levine, and Collins publish the archival re-examination of the Genovese case in the American Psychologist. 2008 — Clarke and Chess name elite panic in Social Problems. 2009 — Solnit publishes A Paradise Built in Hell. 2011 — Fischer and colleagues’ meta-analysis in the Psychological Bulletin finds the bystander effect attenuated in dangerous emergencies. 2014 — Campbell and Manning publish “Microaggression and Moral Cultures”; American Sniper carries the sheepdog speech to its largest audience. 2017 — Möller and colleagues document tonic immobility in 70 percent of 298 assault victims assessed in Stockholm. 2018 — Blewer and colleagues document the public bystander-CPR gap for women; Campbell and Manning’s The Rise of Victimhood Culture appears. 2019 — Perman and colleagues publish the stated reasons for the CPR gap. 2020 — Philpot and colleagues publish the 219-conflict CCTV study in the American Psychologist: intervention in 90.9 percent. 2021 — Liebst and colleagues publish the intervener-risk figures: 3.6 percent of interveners harmed.

Filed by the desk of Objective Daily News from the public record. Errors of fact brought to the desk’s attention are corrected in the open; a report stands as filed unless the record itself changes.