Stepping in for someone else

Most of the time, when something goes wrong in public, you aren't the person it's happening to. You're the one three seats back, or across the car park, or behind someone at a counter, with a few seconds to decide whether to do anything and what. This lesson is about that decision: five ways of helping that one widely used training teaches, what researchers watching real incidents have seen bystanders do, how often the people who step in get hurt, whether any of that holds in a robbery, and a real disagreement about when to call the police.

The responsibility sentence in that box covers everyone in this lesson. The person harassing or attacking someone is responsible for what they do. A bystander who froze, looked away or couldn't think what to say didn't cause it, and nor did one who stepped in and was hurt for it: that harm is the doing of whoever caused it. If you have been in that moment, harassed while a carriage full of people studied their phones, or one of the people studying a phone and thinking about it since, or the one who spoke up and got hurt, nothing below is a verdict on what you did.

Where First Aid and CPR left off

First Aid and CPR lesson 1 covered the ground this lesson starts from, so here it is in four lines rather than again. The 1964 story of 38 witnesses who watched a murder and did nothing was wrong, as the New York Times itself later wrote. The bystander effect is real in laboratory studies, and modest, and it weakened when people saw a situation as dangerous. When Philpot and colleagues watched 219 CCTV clips of public conflicts in three countries, at least one bystander stepped in in 90.9% of them. And that lesson's practical advice was to give people jobs: point at one person and name the task.

What it left open is what this lesson takes up. What does stepping in look like when the emergency isn't a collapse? Do the people who step in get hurt? Does the 90.9% hold for every kind of emergency? And who decides whether the police come?

What watching real incidents shows

Several research groups have gone past asking people what they would do and looked at what people did.

That second finding needs setting beside lesson 8, which gave you NICE's and Project BETA's rule of one voice: when several people talk at an angry person at once, it confuses and escalates. Levine's finding sounds like the opposite. This course read only the abstract of Levine's paper, which does not list what counted as a conciliatory action and gives no reason why several people did better than one, so the two can't be matched point for point. This course's reading, and it's only a reading: the one-voice rule is about how many people do the talking, and Levine's finding is about how many people act to calm a conflict. Several people can do that with one of them talking and the rest standing near, fetching help or keeping watch, which was roughly Kwame's job at the pharmacy counter in lesson 8.

The second study came from nights of observation in bars rather than from cameras. Parks and colleagues, in a 2013 paper, studied 860 incidents of verbal and physical aggression in large bars and clubs in Toronto. Third parties were more likely to get involved when the aggression was more severe, and the features that predicted their involvement predicted calm, non-aggressive involvement more strongly than aggressive involvement.² The authors read that as bystanders responding to the perceived danger of serious harm. It fits what First Aid and CPR lesson 1 reported from Fischer's pooled laboratory studies, that the bystander effect weakened when a situation looked dangerous. This course read the abstract only.

Do the people who step in get hurt?

The study was first posted as a preprint in 2020 and published in the journal Psychology of Violence in 2021. This course read the abstract of each, and both give the same 3.6%. Both also say three more things. The harm was no more likely in the more severe conflicts. Whether the person stepping in was connected to one side of the conflict predicted being hurt, a link the authors rate as only moderately dependable, and the abstract does not say which way it ran. And the intervener's gender was, in their words, a highly fragile risk factor, which this course reads as: not a result to rely on.³

Two cautions, and they are this course's reading rather than the authors'. First, the abstract does not say how serious the harm was, or which countries the footage came from. Second, the people on those recordings stepped in at moments of their own choosing. So 3.6% describes those moments, not a promise about stepping into any fight at all. Lesson 10, "What fear does to you, and the robbery decision", meets the same problem in the robbery data, where the people who resist are the ones who judged they could.

Even so, it answers the worry that stepping in usually gets you hurt: in the disputes these cameras recorded, the people who stepped in were rarely hurt.

The same question in a shop robbery

Philpot's 90.9% came from public conflicts: aggression, arguments and fights. Lindegaard and colleagues asked whether it holds somewhere else.

The study was published online in The Journal of Social Psychology on 15 September 2026, and this course read its abstract only. The abstract describes commercial robberies as a setting with fewer bystanders and more weapons than the street assaults.⁴ It doesn't test why people helped less there, and this course won't supply a reason the authors did not give. It does tell you that a number about street fights is not a number about every emergency: measured on whom, and measured as what, again.

Right To Be's five Ds

One widely taught scheme for stepping in comes from Right To Be, a US organisation formerly called Hollaback!, which trains people to respond to harassment. Its page on the 5Ds of Bystander Intervention opens on harassment in public, including harassment about someone's race, religion, immigration status, gender presentation, sexual orientation, size, age or disability. The five are "Distract, Delegate, Document, Delay, and Direct", and the page says "four of them are indirect methods of intervention."⁵

They grew in stages. In the organisation's account: "In 2012, we partnered with the bystander program Green Dot, which pioneered the Three Ds ... In 2015, we expanded those to include Delay, and in 2017, we expanded them again to include Document."⁵ Green Dot comes back in the next section, because it is the part of this family that has been tested.

All quotations in the table are from Right To Be's page.⁵

Distract has two keys: "Ignore the person who is harassing, and engage directly with the person who is being harassed", and "Don't talk about or refer to the harassment that's happening. Instead, talk about something completely unrelated."⁵ Its examples include "Pretend to be lost and ask the person being harassed to give you directions", and simply "Get in the way." The page calls Distraction "a subtle and creative way to intervene".⁵

Delegate starts with finding someone "ready and willing to help. Often, a great choice is the person right next to you." Then: "try to tell them as clearly as possible what you're witnessing and how you'd like them to help."⁵ In public places, the page says, that could be someone with authority in the space, "a store supervisor, bus driver, or a transit employee".⁵ That's First Aid and CPR lesson 1's named job again, arrived at from a different direction: not "can someone do something?", but "you, can you stand with me while I go and talk to her?". What the page says about delegating to the police is the last section of this lesson.

Document comes with two rules, and the first is the one people skip. "Assess the situation. Is anyone helping the person being harassed? If not, use another of the 5Ds." If someone else is helping, the page goes on, "assess your own safety, and if you are safe, begin documenting." The second: "ALWAYS ask the person who was harassed what they want to do with your recording and/or notes. NEVER post it online or use it without their permission."⁵ The page gives its reasons at more length: filming someone's harm while nobody helps them can add to it, and footage posted without consent can make a person visible in a way they never chose.

Delay is for when the moment has passed, or you could not act in it. "Even if we can't act in the moment, we can still make a difference for someone who's been harassed by checking in on them after the fact." Its examples include "Ask them if they're okay, and let them know you saw what happened and it wasn't okay", and "Offer to accompany them to their destination or sit with them for a while."⁵

Direct is the one the page hedges most. "Use this one with caution, because Direct intervention can be risky", and it gives the reason: the person harassing may turn on the bystander, or escalate in another way.⁵ It gives four questions to ask first: "1. Are you physically safe? 2. Is the person being harassed physically safe? 3. Does it seem unlikely that the situation will escalate? 4. Can you tell if the person being harassed wants someone to speak up?" And: "If you can answer yes to all of these questions, you might choose a direct response."⁵ Even then, "keep it short and succinct", as in its example "Please stop right now.", and the page adds that you shouldn't get drawn into an argument with the person doing it. The part about not arguing is lesson 8's advice too. But Direct itself, naming the behaviour to the person's face, is the kind of challenge lesson 8's sources, Project BETA and Quinn as MacYoung passes him on, warn against. On this course's reading, that fits the page's own caution about it.

The page ends on a note about safety: "We don't ever want you to get hurt while trying to help someone out. Always prioritize safety".⁵ The claim that the 5Ds are "designed to be safe and not to escalate situations" is Right To Be's account of what they were built to do, not a measured result.⁵ In this course's labels, the 5Ds are practitioner opinion: a framework from people with long experience of teaching it, which the next section shows nobody has put to a controlled test.

One boundary, because the page does not draw it. The 5Ds are written for harassment in public. If what you see looks like someone being controlled or hurt by a partner or family member, the lesson "Domestic abuse and stalking" has the services' own advice for people who want to help, and Refuge's helpline in England takes calls from friends, family and colleagues worried about someone.

What has been measured

Here the lesson slows down, because one claim made for bystander training is that it is proven, and the evidence is uneven.

The 5Ds themselves. This course found no controlled evaluation of them. The nearest thing is a 2022 press release from Right To Be about its workplace training, reporting a nine-month study of it, with "67% of attendees reporting applying the bystander intervention skills" after seeing harassment.⁶ Run that through the course's question. Measured on whom: people who took one workplace training. Measured as what: what they said, afterwards, that they had done. There was no comparison group, so nobody knows how many would have stepped in anyway, and this course read the release, not the report behind it. The release calls the 5Ds proven; this study cannot show that.

Green Dot, the programme the first three Ds came from, has the strongest test. Coker and colleagues ran a cluster randomised trial in 26 Kentucky high schools, published in 2017: schools, not pupils, were randomly assigned to get Green Dot or to wait for it. Green Dot there meant school-wide talks plus bystander training for students who were influential with their peers, and the outcome was violence that students reported in anonymous surveys, counted school by school. For sexual violence perpetration the prevalence rate ratios, intervention schools against control schools, were "0.83 (95% CI=0.70, 0.99) in Year 3 and 0.79 (95% CI=0.67, 0.94) in Year 4".⁷

A rate ratio of 0.83 means the intervention schools reported 17% less sexual violence perpetration than the control schools that year, and the brackets are the range the true figure probably lies in. Both ranges stay below 1, so in both years the intervention schools reported less than the control schools, though the Year 3 range comes close to 1 at its top. The abstract says similar patterns showed up for victimisation, sexual harassment, stalking and dating violence. An earlier study of three US college campuses, one with Green Dot and two without, found victimisation "17% lower" on the Green Dot campus, but that study was not randomised, so the campuses may have differed in other ways.⁸

Bystander programmes in general. Kettrey and Marx pooled 15 high-quality studies of bystander programmes for preventing sexual assault among college students and found "a significant, desirable effect" on bystander efficacy (how capable people felt of stepping in), on their intentions, and on intervening itself.⁹ A second meta-analysis by the same two authors, of 14 studies, 12 of them randomised trials, found the programmes raised bystander intervention but had no significant effect on the sexual assault perpetration the participants reported.¹⁰ This course read both abstracts only.

Put together, and sorted the way this course sorts every claim:

• Observed on camera or by observers: bystanders usually step into public conflicts, usually to calm them, and the ones who stepped in were rarely hurt, in the footage studied. In shop robberies they helped far less often.
• Tested: Green Dot's whole programme, in Kentucky high schools, cut the violence students reported. That is a result for a school-wide programme, not for one person on a bus. Pooled trials of bystander programmes for young people found they raised intervening, and, in one meta-analysis of 14 studies, made no significant difference to the sexual assault perpetration participants reported.
• Practitioner opinion: the 5Ds, and the claim that indirect options are the safer ones.
• Not known: whether reading about the 5Ds makes anyone more likely to help, or safer when they do. Nobody has measured it.

The late bus

The first case, worked through. It's nearly eleven at night and you're on a bus in Southampton, in England, two rows behind a young woman who is looking at her phone. A man across the aisle has been making comments to her for several stops. She moved seats once, towards the front, and he moved too. She hasn't answered him. Nobody else has said anything.

None of this is her doing. The question is what you can do now.

Direct first, because it is the one people picture. Run Right To Be's four questions. Are you physically safe? Probably, but you don't know him. Is she? You don't know. Does it seem unlikely to escalate? He has already followed her once. Can you tell whether she wants someone to speak up? Not yet. Four yeses are the page's condition for Direct, and you have none you're sure of, so on Right To Be's own test this isn't the moment for it.⁵

Distract costs least. You move to the empty seat across from her, which also puts you between them, and talk to her about something unrelated, ignoring him. That is both of Right To Be's keys for Distract, and "Get in the way" as well.⁵ The risk is that he turns his attention to you. Right To Be's view is that the indirect Ds carry less of that risk than Direct; nobody has measured it. In this course's labels, practitioner opinion.

Delegate is there too. The driver is exactly the person Right To Be names: someone with authority in the space.⁵ If Distract doesn't end it, you could go forward at the next stop and tell the driver plainly what is happening and what you would like, which is the page's second key for Delegate.

Document is not the first choice. Nobody else is helping her, and the page's first rule says that in that case you use another D.⁵

So you choose Distract, then Delay: once he has gone, and before her stop, you check in.

When you would switch. If he hits her, grabs her or threatens to hurt her, someone may be in immediate danger, and GOV.UK's page on contacting the police says: "Call 999 if you are reporting a crime that is in progress or if someone is in immediate danger."¹¹ Whether to check with her first, as Right To Be asks for harassment, is the question the last section of this lesson sets out. If a call is made, the driver is the person with authority on the bus; if you make it yourself, a call he cannot hear is the safer one, and if speaking would draw him to you, the box at the top of this lesson has the silent 999 call. That last part is this course's reading, sensible, untested. And if you once did nothing on a bus like this and still think about it, this lesson is not marking that moment either.

Third in the queue

The second case is a different kind of emergency. Theo is third in the queue at a convenience store in Albany, New York, late on a weekday evening, holding a carton of milk. The man at the front puts a knife on the counter and tells the cashier, Joel, to empty the till. There's one other customer, between Theo and the man.

Everything the bus taught pulls Theo towards doing something. Here the sources point the other way, for three reasons.

The evidence is different here. Lindegaard's footage of shop robberies found help in 35% of them, not 90.9%, in a setting its authors describe as having fewer bystanders and more weapons.⁴ The abstract does not say what happened to the people who helped, and this course will not guess.

The cashier may have been told what to do. In the US, the Occupational Safety and Health Administration's recommendations for late-night retail say staff training should include "guidance to limit workers from intervening in workplace altercations whenever possible", and list, among the topics it may include, "Specific instructions on how to respond to a robbery such as turning over money or valuables without resistance".¹³ Those are recommendations to employers, from 2009, written for staff, not customers. This course reads them as a reason for a customer to stay out of the way too: if Joel's plan is to hand the money over and let the man go, a customer who steps in cuts across that plan and puts themselves inside a situation built to end quickly. Sensible, untested, in this course's labels.

What the man wants is property. Lesson 10, "What fear does to you, and the robbery decision", takes up the case of a weapon shown and property demanded, where the three official sources this course read, London's Metropolitan Police, the Washington, DC police and the OSHA guidance above, all point towards handing it over. London's Metropolitan Police put it plainly in its own advice on street robbery: "Property can be replaced, you can't."¹⁴ Here the property is not even Theo's.

So what helps? This course's reading of the sources, labelled sensible, untested:

• Staying out of it, and leaving if he can do that quietly, without moving towards the man or drawing his attention. If leaving would mean walking past him, staying still is also staying out of it.
• Calling when it's safe to. Once the man has gone, or once Theo is outside, it's a 911 call. 911.gov lists what the call-taker may ask, including the location and details such as a physical description of a person who may have committed a crime.¹⁵ Noticing what the man did, what he took and which way he went is useful. Lesson 10 explains why memory under that kind of stress is less exact than it feels, so nobody should expect a perfect description of themselves.
• Delay. Afterwards, checking on Joel is the same D as on the bus. Being robbed at work is frightening whatever the training said.

Should a bystander call the police?

Back to the bus, and to Delegate. Many people would say the obvious third party is the police. Right To Be doesn't, and this is a question people genuinely disagree about. It's partly about facts and mostly about values: whose decision it should be. The course sets out both positions and takes neither.

Right To Be's position

On its 5Ds page, Right To Be answers the question directly: "you should not contact the police unless you've checked with the person being harassed and they've explicitly asked you to call the police on their behalf."⁵ Its reasons, in summary, are these: some people are not comfortable or safe with police involvement, and for many people and communities a history of mistreatment and violent escalation by law enforcement has led to fear and mistrust. It names communities of colour and undocumented people among those who "may rightfully feel unsafe in the hands of police".⁵ Put in one line, as this course reads it: the person being harassed is the one who lives with the consequences of a police response, so the choice should be theirs. The same page's Delay section suggests offering to help the person make a report, if they want to.

That is an organisation's stated position, and the course reports it as one. Right To Be's page is written about harassment. It does not say how its rule applies when someone is being violently assaulted in front of you.

Put at its strongest, as this course reads it, the case for asking first runs like this. A police response brings its own risks, and on Right To Be's reasons above they fall on the person being harassed, not on the caller, and harder on some communities than on others. The caller walks away afterwards; the person harassed is the one who lives with what follows. And asking first doesn't leave them without help, since the other Ds are still there, and Delay includes offering to help with a report if they want one.

The official pages' position

Official public pages start from the other end: the situation, not the permission. GOV.UK's page says: "Call 999 if you are reporting a crime that is in progress or if someone is in immediate danger."¹¹ In the US, 911.gov describes an emergency as any situation that needs immediate help from the police, fire service or ambulance, and says to dial 911 straight away.¹⁵ The Office for Victims of Crime tells victims themselves: "If at any time you feel that you are in immediate danger, we strongly urge you to call your local police department's emergency number (911)."¹⁶ None of these pages is written about bystanders asking permission, and none says a bystander should ask the victim first. They are guidance to callers, not an argument about bystanders, and the course reports them as that. Put in one line, as this course reads them: a crime in progress, or someone in danger, is itself the reason to call, whoever is on the receiving end.

Put at its strongest, as this course reads it, the case for calling runs like this. The person being harassed may not be able to say what they want in that moment, and asking them in front of the person doing it can put them on the spot. Stopping a crime in progress is the police's job, not a passer-by's. And a report can matter to the person harmed later, in some places for compensation, which lesson 16, "Afterwards", takes up.

What does not depend on the argument

Two things don't depend on which position you take. First, where someone is being assaulted or is in immediate danger, GOV.UK's own words make it a 999 call. Second, both positions put safety first: Right To Be's "Always prioritize safety", and the official pages' "immediate danger".⁵¹¹ The clearest difference is over harassment where nobody is in immediate danger, and who gets to decide about the police then. That is a question about values, which this course describes and does not answer.

The call itself is First Aid and CPR lesson 2's: location first, and stay on until the call handler says you can go. That lesson put the phone on speaker to keep your hands free for CPR; with the person causing the danger still nearby, a call he cannot hear is the safer one, which is this course's reading, sensible, untested.

What people get wrong

"Crowds mean nobody helps." First Aid and CPR lesson 1 corrected this for public conflicts: someone stepped in in 90.9% of them. Add the exception: in shop robberies, help came in 35% of cases, and more people did not mean more help.⁴

"Filming is always helping." Right To Be's rule is to film only if someone else is already helping and you are safe, and never to post or use a recording without the person's permission.⁵

"Stepping in means confronting the person." Four of the five Ds involve no confrontation at all, and Right To Be treats the fifth, Direct, as the one to use with caution.⁵

"People who step in usually get hurt." In the public disputes Liebst and colleagues studied, 3.6% of people who stepped in were physically harmed.³ They stepped in at moments of their own choosing, and the rate is not zero.

"More people means more talking." Several people acting to calm a conflict did better than one in Levine's footage,¹ but lesson 8's one-voice rule still applies to who does the talking.

"Calling the police is always right", and its mirror, "calling the police is never a bystander's decision". Each is one side of a disagreement about values. Where someone is in immediate danger, the official pages are clear; over harassment that is not violent, Right To Be and the official pages start from different places, and the course takes neither side.

Practice

Connections

Back. First Aid and CPR lesson 1 taught the bystander effect, the correction to the Genovese story, Philpot's 90.9% and "give people jobs"; this lesson added what happens when people step in, and the exception for robberies. First Aid and CPR lesson 2 taught the call. Lesson 8's advice against arguing is part of what Direct needs, and its one-voice rule sits beside Levine's finding. "Domestic abuse and stalking" has the services' advice for helping someone you know.

Forward. Lesson 10, "What fear does to you, and the robbery decision", takes up the moment the queue in Albany left off: what fear does to the people in it, and the decision to hand property over or resist. Lesson 16, "Afterwards", covers reporting a crime and the support available whether or not you do.

Go deeper

• Right To Be, "The 5Ds of Bystander Intervention", free. Read here in full. Short, with more examples for each D than this lesson had room for, and the organisation's own account of its police stance. It also lists free online trainings.
• Coker and colleagues, "RCT Testing Bystander Effectiveness to Reduce Violence", American Journal of Preventive Medicine, 2017, free. Read here: the abstract only. The randomised trial of Green Dot this lesson reports; even the abstract sets out what the programme involved and what was counted.
• Lindegaard and colleagues, "The High Bystander Helping Rate for Public Assaults Does Not Generalize to Commercial Robberies", The Journal of Social Psychology, 2026, open access. Read here: the abstract only. The newest study in this lesson, and the one that most changes what First Aid and CPR lesson 1's number means.

Sources

1. Levine, Taylor and Best, "Third parties, violence, and conflict resolution: the role of group size and collective action in the microregulation of violence", Psychological Science 22(3), 2011, pp. 406 to 412. Read: abstract only.
2. Parks, Osgood, Felson, Wells and Graham, "Third party involvement in barroom conflicts", Aggressive Behavior 39(4), 2013, pp. 257 to 268. Read: abstract only.
3. Liebst, Philpot, Levine and Lindegaard, "Cross-national CCTV footage shows low victimization risk for bystander interveners in public conflicts", Psychology of Violence 11(1), 2021, pp. 11 to 18; first posted as a PsyArXiv preprint in 2020. Read: the abstracts of the 2020 preprint and of the published version only.
4. Lindegaard, Liebst, Bernasco, Philpot, Thomas and Ejbye-Ernst, "The High Bystander Helping Rate for Public Assaults Does Not Generalize to Commercial Robberies: A Video Observational Study", The Journal of Social Psychology, published online 15 September 2026. Read: abstract only.
5. Right To Be, "The 5Ds of Bystander Intervention". Read: in full, on the live page and in the Internet Archive capture 20260827212950.
6. Right To Be, "Press Release: 67% of People Have Used Bystander Intervention Strategies in the Workplace After Taking Right To Be's Training", 28 October 2022. Read: in full, from Internet Archive capture 20251205135030. The study it reports was not read.
7. Coker and colleagues, "RCT Testing Bystander Effectiveness to Reduce Violence", American Journal of Preventive Medicine 52(5), 2017, pp. 566 to 578. Read: abstract only.
8. Coker and colleagues, "Multi-College Bystander Intervention Evaluation for Violence Prevention", American Journal of Preventive Medicine 50(3), 2016, pp. 295 to 302. Read: abstract only.
9. Kettrey and Marx, "The Effects of Bystander Programs on the Prevention of Sexual Assault across the College Years: A Systematic Review and Meta-analysis", Journal of Youth and Adolescence 48(2), 2019, pp. 212 to 227. Read: abstract only.
10. Kettrey and Marx, "Does the Gendered Approach of Bystander Programs Matter in the Prevention of Sexual Assault Among Adolescents and College Students? A Systematic Review and Meta-Analysis", Archives of Sexual Behavior 48(7), 2019, pp. 2037 to 2053. Read: abstract only.
11. GOV.UK, "Contact the police". Read: in full, re-read 25 September 2026.
12. Support lines in the content note, each read on its service's own page on 25 September 2026: Victim Support (England and Wales), Victim Support Scotland, Victim Support NI (home and contact pages) and VictimConnect. Read: each page in full, re-read 25 September 2026.
13. Occupational Safety and Health Administration, Recommendations for Workplace Violence Prevention Programs in Late-Night Retail Establishments, OSHA 3153, 2009. Read: the PDF, searched.
14. Metropolitan Police, "How to protect yourself from street robbery". Read: in full, from Internet Archive capture 20250812180728.
15. 911.gov, "Calling 911", last updated 24 January 2023. Read: in full, 25 September 2026.
16. Office for Victims of Crime, "Help for Victims". Read: in full, re-read 25 September 2026.