Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Helping and Bystander Intervention

Abstract. Helping and bystander intervention comprise the recognition of another person's need and the selection and implementation of an appropriate response under social uncertainty, practical constraints, and potential risk. This fundamental research monograph examines the origins of experimental emergency intervention research, the five-stage decision framework, diffusion of responsibility, pluralistic ignorance, motivational and identity-based models, and the role of danger and public accountability. It critically synthesizes classic experiments, field studies, a major meta-analysis, naturalistic CCTV observations, developmental evidence, and evaluations of bullying and sexual-violence prevention programs. Findings support a context-dependent relationship between the presence of witnesses and intervention, with important distinctions between individual and group-level helping probabilities, intentions and action, and intervention frequency and effectiveness. The monograph evaluates measurement, causal inference, bias, ethics, cultural generalization, online witnessing, and research priorities for human–artificial support systems. It is a critical narrative synthesis of published evidence, not a newly conducted systematic review or original empirical study.

Keywords: helping behavior; bystander intervention; bystander effect; diffusion of responsibility; pluralistic ignorance; emergency intervention; prosocial behavior; empathy and altruism; social identity; witness behavior; dangerous emergencies; naturalistic observation; bystander education; bullying prevention; sexual violence prevention; public self-awareness; online bystanders; human–artificial relations; Artificial Era

Helping and Bystander Intervention as a Scientific Object

Helping is behavior intended to improve another person's condition, prevent harm, or enable an otherwise obstructed goal. Bystander intervention is the subset of helping in which a person becomes aware of another's need or a potentially harmful event and can choose among direct aid, indirect support, reporting, recruitment of additional assistance, and nonintervention. A bystander may be a stranger, peer, colleague, family member, community witness, or digital participant. The word 'intervention' describes an observable response, not necessarily a morally appropriate or effective outcome: some interventions are mistimed, coercive, or hazardous, while appropriate action may involve contacting trained responders instead of entering a dangerous scene. Darley & Latané (1968) and Latané & Darley (1968) established the experimental importance of others' presence, and Fischer et al. (2011) demonstrated why physical danger and social context modify that relationship.

The central explanatory problem is how potential helpers transform incomplete social information into an action under uncertainty, opportunity constraints, competing obligations, emotional arousal, and group expectations. The process comprises noticing an event, interpreting its significance, accepting responsibility, choosing a feasible response, and performing it. These stages are analytically distinguishable even when real-world cognition is rapid and recursive. A witness may understand that assistance is needed while believing another person has already called emergency services; another may intend to help while lacking a safe means of intervention. Consequently, studies measuring intentions, emergency judgments, actual action, reaction time, or outcomes address overlapping yet distinct dependent variables. Nickerson et al. (2014) operationalized the multistage approach for adolescents responding to bullying and harassment.

This research dossier investigates acute aid, situational prosocial assistance, and witness action in emergencies, peer victimization, interpersonal aggression, organizational environments, and digitally mediated encounters. It considers developmental precursors and cross-context generalization only insofar as they illuminate assistance and intervention. Repeated reciprocity, cooperation in public-goods games, and stable prosocial identity belong to adjacent research objects. The method is a critical narrative synthesis of original experiments, field observations, validated measures, longitudinal and intervention studies, and quantitative reviews; it makes no claim to a new systematic search, new participant data, or a newly computed meta-analysis. The guiding questions concern mechanisms, boundary conditions, real-world frequency, the effectiveness and safety of training, and the conditions under which social environments make helping possible.

Origins of Experimental Helping Research

Scientific work on helping accumulated across moral psychology, sociology, developmental inquiry, and experimental social psychology. In the late 1960s, research on apparently passive witnesses provided a means of studying situational control over helping. Darley & Latané (1968) experimentally varied whether a participant believed that other people heard a simulated seizure. Participants who believed that they alone knew of the emergency generally reported more rapidly than those who thought other listeners were present. The study is an important causal demonstration of a responsibility-related effect under a defined laboratory procedure; its fabricated emergency, student sample, and constrained response options limit extrapolation to all public crises.

In another 1968 experiment, Latané & Darley (1968) placed participants in a room in which smoke appeared. Responses differed substantially across alone, group, and passive-confederate conditions, supporting the proposition that other people's visible calm informs observers' interpretation of an ambiguous event. The construct 'pluralistic ignorance' describes the possibility that several concerned individuals privately perceive danger while publicly appearing unconcerned and interpreting one another's inaction as evidence that nothing serious is occurring. In this design, the social cue was deliberately manipulated; naturally occurring crowds also contain friends, knowledgeable witnesses, aggressors, and individuals already preparing assistance. Laboratory group effects therefore identify a mechanism rather than an immutable behavioral law.

One year later, Piliavin et al. (1969) examined staged collapses aboard New York subway trains. The investigators varied characteristics of the apparent victim, including signs of illness or intoxication, and recorded spontaneous assistance. Helping was common under some conditions, and the expected simple diffusion pattern did not appear. This divergence was theoretically productive. An enclosed carriage, an identifiable victim, a shared time horizon, and the physical ability to act differ from anonymous, uncertain laboratory settings. The comparison initiated a lasting research agenda on the interaction between responsibility, perceived need, victim characteristics, risk, and spatial organization.

The Decision Sequence and its Limits

The influential five-step model, formalized in the classic Latané & Darley (1970) monograph, describes noticing an event, interpreting it as requiring aid, taking personal responsibility, deciding what help to provide, and implementing the decision. The model is useful for mapping points of failure to different mechanisms: inattentiveness can prevent detection, ambiguous social cues can prevent emergency recognition, distributed obligations can inhibit responsibility, and inadequate practical knowledge can prevent action despite willingness. It also helps avoid interpreting all failures to intervene as evidence of low empathy or defective moral character. In a real incident, however, stages overlap, and witnesses may revise their beliefs after another person moves, shouts, or provides information.

Noticing is shaped by selective attention, spatial location, perceptual salience, divided attention, and time pressure. Darley & Batson (1973) demonstrated that participants assigned to hurry were less likely to stop for an apparently distressed individual. The specific experimental setting was seminary students traveling between buildings, which allows a situational inference about urgency while not establishing that time pressure always suppresses all forms of aid. Intervention opportunities are also distributed unequally: one witness sees a collapse directly, another arrives after help has begun, and a third witnesses only a brief online excerpt. Studies should define opportunity and exposure before comparing raw helping rates.

Interpretation requires inference about the event, the victim's wishes, danger, and whether intervention would improve the situation. Shotland & Straw (1976) found that the perceived relationship between a man and a woman in a staged violent encounter affected witnesses' interpretations and responses. Familiarity scripts and mistaken assumptions about private conflict can suppress aid even when objective behavioral risk is similar. Responsibility and implementation then depend on who seems authorized or able to act, the availability of safe procedures, and whether one intervention opens the way for others. The five stages are best treated as a testable conceptual framework with partial empirical support and context-specific transitions rather than a universally fixed serial algorithm.

Diffusion of Responsibility and Social Expectations

Diffusion of responsibility refers to a reduction in an individual's perceived obligation to act when other potential helpers are present. The emergency-hearing manipulation in Darley & Latané (1968) is its foundational demonstration: participants' beliefs about invisible coparticipants altered response behavior. An important distinction concerns actual versus imagined social presence. In five studies, Garcia et al. (2002) showed that experimentally priming a group context could reduce later helping in an unrelated task, extending the analysis to mental representation of collective presence. This evidence does not imply that all group reminders are suppressive; group identity, roles, accountability, and danger change the meaning of others' presence.

Social comparison supplies another mechanism. When an event is ambiguous, a witness may use others' calm as evidence of safety. Such informational influence differs from responsibility diffusion: in one case the witness doubts whether there is a problem, while in the other the witness may recognize the problem but assign the response to someone else. Evaluation apprehension adds concern that acting could look foolish, intrusive, or incompetent. Perceived norms can either support or discourage intervention, and visible initiation by one person may create a new shared definition of the event. Latané & Darley (1968) demonstrated one experimentally restricted form of social inhibition; van Bommel et al. (2012) showed that accountability cues can shift the direction of the effect in online helping scenarios.

The size of a group should not be confused with the size of its functional helping capacity. A group of trained coworkers with assigned roles differs from a group of anonymous commuters; several uncoordinated responders may interfere with one another, whereas coordinated witnesses can divide tasks. The experimentally measured probability that any particular person intervenes may fall even as the probability that somebody intervenes rises with the number of potential helpers. A complete model must distinguish individual helping, at-least-one-helping, number of helpers, speed, appropriateness, and outcome quality. The absence of any single measure from a study restricts what that study establishes.

Arousal, Cost–Reward Evaluation, and the Stakes of Action

Emergency events can produce physiological arousal, fear, empathic concern, confusion, and an urge to escape. Arousal-based accounts propose that witnesses seek to reduce aversive emotional states, while cost–reward accounts specify that the preferred response depends on anticipated consequences of intervening, remaining present, or leaving. Danger may intensify urgency and increase clarity, while simultaneously raising the cost of physical contact. These forces operate in different directions, explaining why a simple relation between danger and intervention is insufficient. Fischer et al. (2011) found that bystander inhibition was weaker in dangerous than nondangerous emergencies, and Hortensius & de Gelder (2018) reviewed evidence and hypotheses relating emotion, defensive reactions, and disposition to witness behavior.

The cost of helping includes physical injury, social sanction, time, money, procedural uncertainty, and risk to the person being helped. Costs of withholding help can include distress, guilt, sanction, or harm to relationships. Some interventions, such as calling an emergency number or alerting nearby staff, carry substantially different costs from physically confronting a violent aggressor. A psychologically complete model therefore represents response alternatives rather than a binary 'help versus do nothing' variable. Ethical intervention studies and practical safety guidance should preserve this distinction, particularly in interpersonal violence and medically consequential incidents.

Emotional activation is also neither identical to empathy nor proof of altruistic motivation. An observer may be highly distressed and withdraw, or relatively calm and organize assistance effectively. Batson et al. (1981) tested the claim that empathic concern can motivate aid even when escape from witnessing suffering is easy. Cialdini et al. (1997) subsequently challenged an exclusively altruistic interpretation by considering self–other overlap. Those studies illuminate motivational accounts but do not settle a universal motive for helping. Explanation requires the observed behavior, the alternatives available, and the experimental manipulation used to distinguish competing motives.

Empathy, Social Identity, and Recipient Recognition

Empathic concern can direct attention to another person's welfare, but it must be distinguished from personal distress, perspective-taking, compassion, and behavioral competence. Batson et al. (1981) reported evidence compatible with an empathy–altruism account under its laboratory conditions; Cialdini et al. (1997) argued that perceived oneness or self–other inclusion may explain related patterns. Both positions depend on how emotional experience and motives are operationalized. Help can arise through care, perceived obligation, reputation, social affiliation, or practical role assignment; multiple motives may coexist in a single response.

Social categorization affects whose suffering is noticed and how readily assistance is offered. In field experiments involving soccer supporters, Levine et al. (2005) found that a shared group identity increased help to a visibly injured person and that making a more inclusive identity salient expanded helping across earlier group boundaries. These findings reveal a malleable social boundary rather than a universal preference for familiar individuals. The practical implication is that community definitions of 'us' can change perceived responsibility, while the research implication is that victim and helper identity need direct measurement in studies of emergency response.

Judgments about recipients may also depend on perceived deservingness and the imagined cause of the difficulty. In Piliavin et al. (1969), an apparently ill victim received help more readily than a victim who appeared intoxicated. Interpretations of controllability and risk can affect whether need is recognized as legitimate, yet such perceptions may reflect stereotypes and systematic social disadvantage. Experiments separating actual need from attributions about the person are crucial for identifying inequity. Assessing helping solely through the observer's moral self-description risks overlooking whose needs were ignored and whether the assistance was welcomed.

Experimental Paradigms and Causal Inference

Bystander research uses laboratory simulations, staged field events, observational video, retrospective surveys, school-based intervention trials, and digital experiments. These approaches differ in the intervention that can be manipulated, the behavior visible to the researcher, and the population to which findings generalize. Laboratory paradigms allow controlled manipulation of the number or apparent behavior of witnesses. The 1968 seizure and smoke studies distinguished an individual's belief about others from visible passive witnesses, allowing different inferences about responsibility and informational influence. However, a contrived emergency may produce task-specific expectations; a participant who suspects an experiment is underway may be responding partly to perceived observation. Darley & Latané (1968) Latané & Darley (1968).

Field experiments preserve more of the structure of real opportunity, although control is usually less complete. The staged subway collapse in Piliavin et al. (1969) varied salient victim characteristics across trains while observing spontaneous aid. The staged pedestrian encounter in Darley & Batson (1973) manipulated hurry as students traversed a campus. Field experiments therefore identify more realistic action constraints while retaining concerns about site, time, visibility, sample composition, and the ethics of exposing witnesses to apparent crises. Manipulated conditions should be contrasted on the assigned unit; repeated episodes involving shared participants require analytic attention to nonindependence.

Direct observation of naturally occurring emergencies avoids fabricating an incident but cannot randomize who happens to be present. CCTV-coded studies such as Levine et al. (2011) and Philpot et al. (2020) generate descriptive evidence about public conflict and the actions of third parties. Observed associations between crowd size and helping may reflect the setting, seriousness of the conflict, density of observers, or selection of visible incidents. Causal identification is therefore stronger in experiments for their particular manipulation, while ecological descriptive coverage is often stronger in naturalistic observational designs. A scientific synthesis integrates the two rather than treating ecological realism as a substitute for causal control.

Meta-Analytic Evidence on the Bystander Effect

The most influential quantitative synthesis is Fischer et al. (2011), which combined 105 independent effect sizes representing more than 7,700 participants across bystander experiments. Its fixed-effects estimate reported an overall inhibition of helping by the presence of others, Hedges' g = −0.35. That standardized contrast is meaningful as an aggregate of included paradigms; it does not denote a thirty-five-percent decline in the probability of rescue. Effect-size interpretation depends on outcome scaling, comparison conditions, model choice, and between-study heterogeneity. Its authors also tested moderators related to the seriousness of the emergency and the role of other witnesses.

The meta-analysis found weaker inhibition in situations perceived as dangerous, when perpetrators were present, and when intervention costs were physical. The aggregate did not support the idea that a large surrounding group always produces the strongest withdrawal. Physical supporters, naïve fellow witnesses, and familiar co-present people could create conditions in which additional bystanders did not depress intervention. The study's broad historical coverage is important, but the included methods were heterogeneous, and laboratory analogues dominated key areas of the evidence. Its conclusions must be read as moderator-sensitive rather than universal statements about all emergencies. Fischer et al. (2011).

A meta-analytic mean cannot replace the analysis of study quality. Publication bias, selective reporting, uneven outcome definitions, use of confederates rather than genuine uninformed observers, and unequal participant demographics can affect a pooled estimate. The older literature was created before current expectations of preregistration and open data, so opportunities for evaluating analytic flexibility are uneven. At the same time, the existence of these limitations is not evidence that the foundational effect is imaginary. The strongest interpretation is conditional: social presence often depresses individual help in specific experimental situations, while danger, supportive group composition, salient accountability, and incident characteristics can weaken or reverse that tendency. Fischer et al. (2011) van Bommel et al. (2012).

Observing Real Public Conflicts

Real-world third-party behavior becomes visible when researchers analyze recorded confrontations. Levine et al. (2011) coded 42 public-space incidents involving 312 people, comparing behaviors that escalated aggression with those that worked toward de-escalation. Third parties more frequently engaged in conciliatory than escalating actions, and constructive third-party involvement was more likely in larger groups. This finding suggests that the social resources represented by a crowd can matter positively in conflicts. The unit of observation, available camera angles, local street and nighttime contexts, and coding of behaviors shape what can be inferred from these episodes.

In a later cross-national investigation, Philpot et al. (2020) analyzed video of public conflicts in Amsterdam, Cape Town, and Lancaster. The authors reported that intervention by at least one witness was common in the sampled incidents, challenging a popular belief that public witnesses usually remain entirely passive. The analysis concerns visible public conflicts accessible to fixed cameras, not every type of emergency or private violence. Moreover, the likelihood that someone helps is a group-level quantity. It should not be substituted for the probability that a randomly chosen member of that group intervenes. Both patterns can coexist without logical contradiction.

These findings change the research question. Rather than asking whether crowds 'help' or 'fail,' investigators can ask how many witnesses act, how they coordinate, when they withdraw, whether they can protect the vulnerable person, and whether intervention escalates or resolves the event. A camera may capture an attempted de-escalation while missing earlier verbal assistance, a call to emergency services, or subsequent harm. Observational studies need transparent coding categories, interrater reliability, and clear eligibility criteria. They also require protections for privacy and against oversimplified claims about ethnicity, gender, or neighborhood based on visually ambiguous footage. Levine et al. (2011) Philpot et al. (2020).

Danger, Clarity, and Physical Support

Danger changes the cognitive and motivational environment of helping. In an ambiguous incident, a passive witness can signal apparent normality, whereas a visible attack or collapse may identify the problem without further interpretation. This is one reason Fischer et al. (2011) found that the classic group-inhibition effect was attenuated in dangerous emergencies. Yet dangerous incidents also increase personal risk and can justify distance, protective withdrawal, or calling qualified responders. A theory that treats every refusal of direct physical confrontation as moral failure confounds sensible risk management with unwillingness to assist.

The presence of a perpetrator introduces another set of variables: perceived physical threat, the possibility of deterring the aggressor, social support for confronting wrongdoing, and opportunity to move the victim to safety. The same number of witnesses may intimidate an aggressor or make individual role assignment more ambiguous. The cost structure is asymmetric: observers can sometimes offer help remotely, while the person at risk has no such alternative. Empirical research should distinguish the moral value of aid from the safety of particular intervention tactics and avoid recommending confrontation as a default response. Fischer et al. (2011) Levine et al. (2011).

Clarity is also produced socially. When one witness plainly names the emergency, requests assistance, or identifies a task, others have less uncertainty about what is required. Such direct communication constitutes a theoretically plausible intervention on interpretation and responsibility. The specific effect of assigning an individual a task, however, must be established in a given applied context rather than inferred as a universally measured effect from the original 1968 experiments. Future trials should compare identification of a specific helper, generalized calls for help, team protocols, and expert-directed instruction using behavioral and safety outcomes.

Group Composition, Identification, and Collective Capability

The social significance of a witness group varies with acquaintance, similarity, roles, and shared identity. Levine et al. (2005) demonstrated that an identity boundary could be experimentally widened so that people previously categorized as rival-group members became eligible for assistance under a broader salient identity. Thus the presence of observers is not merely a count of bodies; it changes what people interpret as common concern and who is understood as responsible. Shared membership may enable coordination, but it may also foster exclusion when outsiders' need is discounted.

Group expertise is another potential moderator. Someone trained in first aid can choose a relevant response faster than an untrained witness, and the presence of a competent peer may reduce uncertainty or enable task distribution. Conversely, a group in which everyone assumes that someone else has appropriate expertise can still delay action. The mechanisms are empirically separable: knowledge can be measured with skill tests, responsibility through belief reports, and action through logged or observed behavior. Simply asking participants whether they would help cannot establish actual skill or emergency competence. Nickerson et al. (2014) Banyard et al. (2007).

Formal institutional roles can alter the attribution of duty. A teacher observing bullying, a workplace supervisor receiving a complaint, and a passerby encountering a public accident face different authorization, reporting, and confidentiality obligations. Social identity and role expectations partly explain why the same act might be experienced as welcome help in one group and as intrusive surveillance in another. Research should therefore specify the participant's structural position, the setting's rules, and the beneficiary's preferences. Those variables can interact with status differences and dependency relationships that are absent from brief anonymous laboratory encounters.

Personality, Individual Differences, and Development

Dispositional tendencies such as empathic concern, conscientiousness, prosocial values, social anxiety, and sensitivity to threat may contribute to helping. These tendencies compete with strong situational influences, and the value of a predictor depends on its incremental explanatory power over exposure, perceived need, and opportunity. The early emergency-hearing experiment found that its personality and background measures were not generally predictive of intervention within that design. That finding cannot establish that stable differences never matter. Darley & Latané (1968) Hortensius & de Gelder (2018) instead suggest that dispositional and reflexive emotional processes should be evaluated jointly with immediate contexts.

Developmental evidence indicates that instrumental helping behaviors emerge early in life. Warneken & Tomasello (2006) examined helping among young human children and chimpanzees, identifying assistance with another person's instrumental goal under experimental arrangements. Such early behaviors demonstrate capacities relevant to prosocial development, while bystander intervention in adolescent peer aggression requires additional skills: interpreting social conflict, anticipating retaliation, and navigating norms. Eisenberg et al. (2015) reviewed the developmental literature on empathy-related responding and prosocial behavior, including family, peer, and contextual contributions. Evidence about early helping should not be used as direct proof of mature emergency intervention competence.

Adolescence introduces distinctive audience pressures, especially where peers can reward aggression, punish intervention, or circulate recordings. Nickerson et al. (2014) developed and assessed a measurement model for bystander stages in bullying and sexual-harassment contexts. A child who notices a harmful act may still need adult protection and credible reporting channels before intervening safely. Developmental models should therefore be tested with age-appropriate tasks and report both supportive and potentially hazardous responses. Longitudinal work is needed to distinguish changes in awareness, attitudes, skills, opportunities, and actual behavior as individuals move through educational and workplace institutions.

Bystander Responses to Bullying and Peer Harassment

Bullying and harassment involve repeated or structurally unequal interactions that can turn an audience into a resource for either perpetration or protection. Witnesses may laugh, reinforce the aggressor, remain silent, comfort the target, challenge the behavior, or report it to responsible adults. Each response has different causal implications, so classifying every person who is physically present as an equivalent inactive bystander obscures processes of reinforcement and defense. Models of adolescent intervention adapt the recognition, interpretation, responsibility, strategy, and action sequence to the social realities of peer groups. Nickerson et al. (2014) gives an empirical example of how these stages can be measured across bullying and sexual harassment.

Direct defense can involve personal risk, including retaliation and status loss. School-level norm change must therefore alter expectations around what happens after a disclosure or defense. Paluck et al. (2016) randomized an anticonflict social-network intervention across 56 schools, using influential student peers to shift norms and evaluating disciplinary reports of student conflict. The design tests a community-level process rather than isolated statements of intention. It also illustrates how selective communication from salient peers can change behavioral climates, although disciplinary records are only one measure of the underlying occurrence of mistreatment.

A scientific program for bullying interventions should distinguish victim safety, observed defense, student reporting, school administrative follow-through, and actual rates of harm. A rise in reporting may initially reflect more trusted channels rather than more aggression; a decrease in official discipline may reflect fewer incidents or changing detection. The evidence therefore calls for multiple measures, long-term follow-up, and careful separation of student influence from staff response. Curriculum-specific effect claims should be attributed to the trial that tested them and not generalized from social psychology's laboratory emergency paradigm alone.

Sexual Violence Prevention and Community Bystander Programs

Bystander education is widely used in sexual-violence prevention because harm often unfolds in a social setting with witnesses before, during, or after a threatening event. A responsible program teaches recognition of coercive or dangerous patterns, safe interruption, support for the person affected, referral, and the use of trained assistance. It cannot treat a witness as responsible for preventing all violence, and it must protect the target's agency and privacy. Banyard et al. (2007) evaluated a community-of-responsibility intervention by randomly assigning 389 undergraduates to program conditions or control. The study reported improvements in attitudes, knowledge, and behavioral indicators and followed participants over subsequent months, furnishing experimental evidence with specific student-population limits.

Program evaluation expanded from individual curricula to campus-wide strategies. Coker et al. (2011) reported associations between exposure to Green Dot training and student-reported bystander behavior and norms in a college survey. Because that study's comparison was observational, selection into training and differences between exposed and unexposed students must be considered before making a causal claim. Katz & Moore (2013) conducted an early meta-analysis of campus bystander education, assembling evidence across programs and outcomes. Inferences should distinguish proximal knowledge and intentions from verified reductions in victimization, and program effectiveness should be tracked across time and institution types.

A program can produce more reported helpful acts while having uncertain effects on assault incidence or survivors' well-being; these are important but separate outcome families. Consent, confidentiality, trauma-informed response, institutional trust, and reliable professional services affect whether encouraged intervention is useful. Reports and whistleblowing channels must have meaningful safeguards against retaliation. Evaluation should include implementation fidelity, adverse events, differential uptake, and equitable access, especially for students whose previous experiences with authority make official help-seeking costly or unsafe.

Visibility, Accountability, and Public Self-Awareness

Visible accountability can transform the meaning of being observed. In two experiments on online support requests, van Bommel et al. (2012) varied group presence along with cues making a participant's identity or public self-awareness more salient. Under greater accountability, the usual negative bystander pattern could reverse: additional observers were associated with more assistance. The manipulation was specific to online contexts and a particular class of requests; it establishes conditional causal effects rather than a universal benefit of surveillance cameras or named identities.

Accountability can support prosocial action when participants believe that a community can identify and evaluate their response, especially if helping is normatively approved. It can also inhibit witnesses when social groups reward the aggressor, when public identification increases risks to a whistleblower, or when the target's privacy is threatened. It is consequently necessary to identify who observes whom, whose approval matters, and what penalties and benefits accompany speaking. Reputational processes complement empathy and responsibility accounts; their effects vary by context and institutional power.

Experimental online platforms offer opportunities to vary visibility, number of prior viewers, response counts, anonymity, and explicit role assignment while tracking actual behavior. However, a participant can click a supportive reaction without providing material aid, and an online silence may reflect a lack of exposure rather than a conscious choice to withhold help. Digital studies should log who saw a request, when they saw it, whether any knowledgeable person had already responded, and whether the response produced a meaningful benefit. These observations are prerequisites for interpreting platform-scale intervention rates.

Online Emergencies, Platform Design, and Distributed Witnessing

Digital witnessing includes responding to a distress message, identifying harassment, reporting exploitation, checking misinformation during an emergency, or contacting a trusted person on another's behalf. Platforms structure intervention through notification systems, default anonymity, public metrics, algorithmic ranking, moderation access, and the availability of verified emergency resources. The central explanatory questions remain familiar—recognition, responsibility, competence, risk, and action—but the material environment changes. The visible number of account followers is not a valid proxy for the number of people who actually viewed a particular request or had the capacity to respond.

The implicit group-priming work of Garcia et al. (2002) makes it plausible that imagined audiences can affect willingness to assist, while van Bommel et al. (2012) provides experimental evidence that accountability cues can reverse some online group effects. Their measures and historical platforms differ from contemporary large-scale social networks, so applying those estimates to current feed ranking or short-form video demands fresh tests. Online emergencies introduce additional risks of amplification, wrongful accusation, doxxing, and repeated exposure of the target to harmful material. Useful aid may consist of private support or responsible reporting rather than public participation.

A responsible platform experiment should measure intervention opportunities at the user–incident level, protect vulnerable subjects, and minimize exposure to manufactured crises. Randomizing false emergency cues may impose unacceptable distress or diminish trust. Researchers can instead use ethical simulations, consent-based help requests, natural experiments following interface changes, and administrative data with strong safeguards. Outcomes should include quality, timeliness, recipient consent, and downstream safety rather than counts of likes, flags, or public comments alone.

Cultural Context, Inequality, and Institutional Trust

Helping norms are situated within cultural expectations about personal obligation, privacy, kinship, authority, and the legitimacy of informal intervention. A bystander may face social sanctions for intervening in an apparently private matter, while in another community intervention is expected as mutual care. Existing laboratory studies, often reliant on Western undergraduate samples, provide incomplete coverage of these differences. International descriptive data such as Philpot et al. (2020) widen the ecological picture across three cities, but comparing those cities does not itself establish a causal effect of national culture. Interpretation must attend to case selection, visibility, and the composition of recorded incidents.

Perceived ethnicity, race, gender, disability, housing status, and intoxication can influence assessments of credibility and deservingness. The subway experiment in Piliavin et al. (1969) showed that signs of apparent illness versus drunkenness affected helping, thereby revealing the explanatory relevance of victim classification. Shotland & Straw (1976) demonstrated that relationship assumptions affected intervention in an enacted assault. Such outcomes invite investigation of social stereotypes and inequality, with careful attention to the difference between participants' perceptions and a researcher's description of the person in need.

Institutional trust changes which response counts as feasible. Calling authorities may be appropriate in some settings but may also carry serious perceived or actual risks to particular groups. In workplaces and schools, formal complaint systems may be inaccessible, slow, or entangled with the very authority relationships that generate harm. Intervention science must examine whether a route to support exists, whether people trust it, and whether the organization protects those who use it. Reporting rates without exposure, context, and outcome measures cannot establish that a community is becoming safer.

Development, Training, and the Transfer of Helping Skills

Helping skills develop within interactional practices. Early instrumental helping documented by Warneken & Tomasello (2006) shows that young children can recognize and respond to another's obstructed goal, while the synthesis of Eisenberg et al. (2015) situates prosocial development among empathy, socialization, and broader relationships. Yet emergency response, peer defense, and organizational reporting demand additional judgment about risk and authority. Training curricula that assume a single generic helping disposition may overlook these necessary context-dependent competencies.

Instruction can target specific barriers: attention to warning signs, interpretation of ambiguous incidents, responsibility acceptance, knowledge of safe options, rehearsal under stress, and confidence in reporting channels. Banyard et al. (2007) demonstrates that an experimentally evaluated bystander program can improve a range of student outcomes; Katz & Moore (2013) summarizes a wider early evaluation literature. The transfer question is whether trained individuals act appropriately when a new, unanticipated incident occurs, particularly after the supportive conditions of training have faded.

Researchers should follow people over time, measure real opportunities, and evaluate behavior rather than relying exclusively on self-report. A program may increase intervention where incidents are visible but fail where organizational incentives punish disclosure. Successful transfer also depends on broader institutional design: explicit policies, approachable staff, credible protection, and responsive professional services. These are measurable parts of a helping system rather than background details to be mentioned after an individual-level experiment.

Operational Definitions and Outcome Measurement

Helping is often operationalized as the occurrence of a target behavior, its latency, duration, frequency, or quality. Typical dependent variables include reporting an apparent emergency, entering a room to assist, responding to a request, defending a bullied peer, calling appropriate services, and providing informational or instrumental support. A behavior must be specified before collecting data because broad categories collapse important differences. Calling emergency services and physically moving an injured person are distinct acts, and the safer behavior may depend on the circumstances. The classic laboratory studies recorded reporting behavior in structured events, whereas naturalistic video work codes third-party action in observed conflicts. Darley & Latané (1968) Philpot et al. (2020).

Researchers must define a denominator of opportunity. Among one hundred people in a public square, perhaps ten witnessed the beginning of an incident, five could see the victim clearly, and two were physically able to act; treating all one hundred as bystanders can distort estimates. Online research adds a comparable problem because potential reach differs from confirmed exposure. Time-to-action measures offer more information than a binary act when delay may worsen outcomes, but waiting for expert help may sometimes be prudent. A rigorous instrument therefore defines who was present, when, what was perceptible, which actions were available, and which interventions were appropriate.

Self-reported intentions, confidence, and knowledge are worthwhile constructs but should be modeled separately from observed helping. Participants can describe themselves as willing to intervene while failing to notice a future event, or they may report action because it is socially desirable. Nickerson et al. (2014) measured the steps of the bystander process for bullying and harassment, and Katz & Moore (2013) synthesized training effects that varied by outcome domain. When using survey scales, researchers should publish item content, dimensional structure, reliability, test–retest evidence where appropriate, and validity across age and cultural groups. A questionnaire's high internal consistency alone does not establish that it predicts real-world protective behavior.

Sampling, Selection, and the Problem of Generalization

Classic helping experiments often used volunteers from universities in the United States, with social and demographic circumstances different from those of the broader population. Emergency analogues may involve one fabricated scenario in a particular building or campus. A statistically well-controlled effect within such a study does not necessarily transfer to violent crime, community care, online distress, or disaster response. Darley & Latané (1968) Darley & Batson (1973) thus provide historically decisive evidence for their manipulations while leaving external-validity questions open.

Naturalistic samples have complementary selection problems. CCTV research captures incidents occurring in camera-covered public spaces and meeting the coders' inclusion criteria. Cases of private abuse, assaults outside camera range, or emergencies resolved before video recording may be absent. Even multi-city analysis does not remove differential camera coverage, observer density, local policing, and event severity from interpretation. Levine et al. (2011) Philpot et al. (2020) provide behavioral evidence beyond student laboratory samples, but their observational settings cannot determine a universal baseline helping rate.

Sampling also affects who appears as the victim and who can act. Research that draws on only healthy young adults may overlook constraints faced by older adults, children, disabled witnesses, and people exposed to institutional coercion. In cultural comparisons, local practices may alter the meaning of public intervention itself. Researchers should describe the setting, recruitment, sample characteristics, event types, and intervention opportunity rather than making broad claims about 'human nature' from a single population.

Statistical Analysis and the Unit of Responsibility

The choice of statistical unit is central. A binary outcome for an individual witness is different from the probability that at least one member of a group helps. For hypothetical independent identical helpers with individual helping probability p and n witnesses, the probability of at least one helper would be 1 − (1 − p)^n. This mathematical expression illustrates the difference between levels of analysis; real witness decisions are usually dependent, p varies with group size, and coordinated action may introduce positive or negative correlations. Therefore the expression is a conceptual benchmark, not an empirically validated formula for any particular emergency.

Repeated incidents within a setting and repeated witnesses within incidents violate ordinary independence assumptions. Suitable analyses may require hierarchical models, incident-level clustering, robust uncertainty estimates, or event-history methods. Video studies should report both incidents and participants, while training studies should respect randomization at the individual, class, or school level. The 56-school network experiment of Paluck et al. (2016) particularly illustrates why institutional-level assignment and social spillovers matter: an individual student's behavior may be influenced by peers selected for the intervention.

Effect-size reporting should preserve the outcome scale and uncertainty. Fischer et al. (2011) used standardized contrasts across heterogeneous studies and reported moderators; program evaluations may report odds ratios, mean changes, frequency of help, or institution-level victimization measures. A standardized mean difference should not be read as a percentage risk reduction, and an association between training exposure and helping should not be described as a randomized causal effect. Confidence intervals, sensitivity analyses, and design-specific estimands communicate more scientific information than binary claims that a study 'proved' intervention effective.

Bias, Replication, and Competing Interpretations

Demand characteristics may affect staged laboratory emergencies: volunteers may suspect deception or search for clues about what the researcher expects. Field studies can improve behavioral realism while still allowing confederates' manner, timing, or body language to vary unintentionally across conditions. Observational video research can suffer from coder ambiguity and selective visibility. Training evaluations can be affected by self-selection, attrition, and reporting changes. Each source of uncertainty concerns a distinct pathway from the research design to the stated conclusion, and none is resolved simply by enlarging a sample.

Competing theories are also frequently difficult to separate. A reduction in helping in a group could reflect diminished personal responsibility, mistaken interpretation of the event, fear of embarrassment, social identity boundaries, personal distress, or the belief that another observer has greater expertise. Darley & Latané (1968) and Latané & Darley (1968) manipulated different social arrangements, while van Bommel et al. (2012) manipulated accountability cues. Future factorial studies should measure mediators without treating post-treatment self-reports as definitive proof of a single causal mechanism. Behavioral timing and independently manipulated cues can help discriminate explanations.

Reproducibility requires more than repeating a headline effect. Modern replication should specify the incident's ambiguity and hazard, group composition, local norms, and outcome definition. Historically important experiments may require ethically updated analogues rather than direct replication of deceptive emergency procedures. The quantitative synthesis of Fischer et al. (2011) and real-conflict findings in Philpot et al. (2020) should be read together as evidence that effect direction is context dependent, not as evidence that one literature invalidates the other.

Ethics and Researcher Responsibility

Creating the impression of medical distress or violence exposes uninformed bystanders to fear and potentially costly action. Modern protocols must assess whether any deception is scientifically necessary, whether alternatives are available, how participants can withdraw, and how debriefing will work. Direct observation of real victims requires strong safeguards for privacy, confidentiality, and data security. The ethical status of research does not follow solely from the societal importance of preventing harm; it depends on the risks imposed by particular data-collection methods.

Programs encouraging witnesses to intervene must protect intended beneficiaries and witnesses alike. An observer should not be encouraged to attempt a dangerous physical intervention merely to satisfy a behavioral metric. In violence, harassment, or acute medical situations, a timely call for professional assistance may be the relevant protective act. Educational evaluations should include any measured adverse effects, escalation, retaliation, and breaches of confidentiality. Ethical assessment is therefore part of the outcome model rather than a compliance statement attached after the research.

Digital monitoring adds questions of consent, permanence, algorithmic amplification, and wrongful identification. Automated detection can flag an apparent emergency, but its errors may produce unnecessary intervention or missed need. Human moderation, escalation rules, and transparent accountability are measurable safeguards. Research on such systems should compare false-positive and false-negative consequences and avoid treating increased reporting volume as direct proof of improved protection.

Interpreting Intervention Evidence Without Overclaiming

The intervention evidence supports a differentiated position. Banyard et al. (2007) used random assignment and showed that a particular campus program could change selected knowledge, attitudes, and behavior-related outcomes. Coker et al. (2011) assessed a Green Dot intervention through a cross-sectional design and reported associations with reported behavior, but exposure was not randomly assigned. Katz & Moore (2013) synthesized twelve studies and identified stronger evidence for improved confidence and intentions than for some distal outcomes. These designs cannot be pooled rhetorically into a claim that all bystander programs prevent violence.

An intervention may work through several mechanisms at once: clarifying the problem, making responsibility salient, offering low-risk response options, improving efficacy, or shifting community norms. To isolate those pathways, studies need comparison groups that separate components, measurement of intermediate processes, and follow-up behavioral data. Academic settings also experience spillovers when trained and untrained students interact; such spillovers may be part of the intended effect but must be considered analytically. Program labels do not identify mechanisms without a causal model.

Practical efficacy has multiple horizons. Immediate knowledge improvement can precede increased intervention; increased intervention may or may not reduce injury, harassment, or victimization; and reduced victimization may depend on institutional policy and broader social support. Research should report each transition using its own outcome measure and inference level. Credible synthesis preserves promising findings alongside uncertainty about transfer, durability, safety, and the comparative effectiveness of specific program elements.

A Multilevel Scientific Synthesis of Helping

Across laboratory, field, observational, and intervention research, helping emerges from interactions among the event, the witness, the surrounding group, and the available institutional response. The event supplies perceptual evidence, urgency, uncertainty, and risk. The witness brings attention, physiological state, acquired competence, moral concerns, and social identity. Other people supply information, influence attribution of responsibility, and sometimes make collective action safer. The environment supplies tools, permissions, reporting channels, and realistic opportunities to help. No single mechanism accounts for all combinations. Darley & Latané (1968) Piliavin et al. (1969) Fischer et al. (2011) Philpot et al. (2020).

A multilevel account makes four distinctions particularly important. First, noticing need differs from acknowledging obligation. Second, intending to intervene differs from having the skill and opportunity to do so. Third, the probability that an individual intervenes differs from the probability that a group produces at least one intervention. Fourth, observed action differs from assistance that is competent, welcomed, and effective. These distinctions resolve several apparent disagreements between classic bystander experiments and naturalistic analyses without treating every divergent finding as measurement error.

The resulting scientific synthesis is a structured interpretation of the existing literature rather than a new empirical model with independently validated parameters. It directs researchers to report the unit of analysis, nature of need, witness exposure, available options, group composition, outcome quality, and institutional support. The decision sequence from Latané & Darley (1970) remains an informative organizing tool, while the moderator evidence of Fischer et al. (2011), identity experiments of Levine et al. (2005), and real-conflict observations of Philpot et al. (2020) broaden the explanatory architecture beyond anonymous passive crowds.

Research Priorities: Mechanisms and Causal Identification

Future causal tests should distinguish recognition, responsibility, competence, fear, and accountability using orthogonal manipulations where feasible. Examples include varying the clarity of an incident while holding group size constant, changing whether a trained responder is visibly present, and specifying whether help can be provided privately. Pre-registered primary outcomes can compare action occurrence, action latency, safe response choice, and effects on the recipient. Mediators measured after an experimental manipulation should be interpreted carefully because they may share unmeasured determinants with action itself.

Experimental designs should also include meaningful counterfactuals. A general prompt to 'be helpful' is an inadequate comparison for a program containing explicit practice, protection procedures, and institutional referral pathways. Component trials can separate training in detection, rehearsal, safe response options, and changes in institutional policy. Longer follow-up is essential because memorized classroom responses may decay, whereas social norm change and organizational trust may develop gradually. Banyard et al. (2007) Katz & Moore (2013) Paluck et al. (2016).

Public-emergency studies need designs capable of integrating observational realism with causal precision. Natural experiments following policy changes, staggered implementation of reporting channels, and ethically authorized intervention simulations may complement camera-based description. Multilevel estimates should recognize interference among witnesses and variation across incident types. Shared data dictionaries, preregistered coding definitions, and multisite collaborations can produce more portable evidence without erasing the heterogeneity that is central to the phenomenon.

Research Priorities: Equity, Safety, and Recipient Outcomes

Research should identify which people are less likely to receive help even when objective need is similar. This requires careful measurement of perceived danger, the victim's self-expressed wishes, socioeconomic and identity cues, the witness's relationship with institutions, and the availability of response pathways. The early subway and staged assault studies provide historical evidence that victim categorization and relationship assumptions influence behavior; contemporary tests should examine whether these patterns persist across settings and populations. Piliavin et al. (1969) Shotland & Straw (1976).

Helping quality should become a more central outcome. A witness may interrupt an assault without increasing the target's later safety, or initiate a public disclosure that unintentionally exposes a survivor. Short-term action counts cannot answer whether the target gained access to appropriate support, experienced further retaliation, or judged the assistance as welcome. Research involving vulnerable populations requires privacy-conscious measures and ethical protections for informed consent, particularly when outcomes extend beyond the immediate encounter.

Equitable bystander interventions depend on trustworthy organizations. A witness's willingness to report misconduct may be rationally limited when retaliation is likely or the receiving office lacks independence. This implies that individual training and institutional safeguards should be evaluated together. The intended endpoint is safer social interaction and improved support for those at risk, not a ceremonial requirement that every observer become visibly heroic.

Human–Artificial Relations and Artificial-Era Research

Artificial Era research adds new configurations of witnessing. A person may ask an artificial assistant for guidance during a crisis, use an automated platform to flag threatening conduct, or depend on software to route an urgent request. These systems modify exposure, response timing, anonymity, coordination, and the apparent distribution of responsibility. The psychological question is how human observers interpret the presence and apparent competence of artificial systems and whether they then delay, delegate, or improve their own action. Existing experiments on imagined audiences and public accountability provide relevant starting points but do not establish these effects for current artificial agents. Garcia et al. (2002) van Bommel et al. (2012).

New research can operationalize artificial participation through system capabilities and observed consequences: detection accuracy, response latency, confidence calibration, escalation behavior, human oversight, false alarms, error correction, and the quality of support delivered. A rigorous study would randomly vary whether a system offers a recommendation, claims to have already notified assistance, or transparently states its limits, and then observe human decisions. The hypothesis that an artificial participant may either concentrate or diffuse responsibility is empirically testable. The presence of a conversational interface alone is insufficient evidence of understanding, moral agency, or competent emergency action.

Postsubjective and relational approaches to artificial systems raise additional conceptual questions about distributed cognition and how a helping episode becomes coordinated across human witnesses, institutions, and artificial processes. Those questions form a research program, not an established empirical verdict about artificial psychology. Existing human bystander effects supply hypotheses about social signals; results must be tested afresh in human–artificial environments with measurable outcomes and ethical governance. This separation allows the Artificial Era perspective to extend psychological investigation while preserving the evidentiary status of its claims.

Open Problems and the Boundaries of Contemporary Knowledge

An unresolved problem concerns the generality of the classic bystander effect across type of emergency, kind of social presence, and response medium. Fischer et al. (2011) found overall inhibition and systematic moderators, while Philpot et al. (2020) documented frequent group-level intervention in selected real public conflicts. The results address different causal and descriptive questions. Research that measures both individual-level action and incident-level rescue probability under comparable conditions could reduce uncertainty about how general these patterns are.

Another problem is identifying which intervention components improve real safety. The campus-training evidence supports changes in efficacy, attitudes, and some helping behaviors, but the pathway from classroom outcomes to reduced incidence of violence requires stronger longitudinal designs and clearer institutional outcome data. Katz & Moore (2013) explicitly noted the need for research on behavioral outcomes and assault incidence. Scalable interventions will also need estimates of implementation burden, harm avoidance, recipient experiences, and effectiveness outside American college campuses.

Finally, new social settings require revised operational definitions. Digital audiences are not equivalent to people who actually see a request; artificial support tools do not supply human attention simply by being present; surveillance may promote accountability while threatening privacy; and cultural interpretations of an apparently private emergency may vary. A durable science of helping should treat these differences as variables to measure. Its evidentiary standard is the demonstrable relation between situational conditions, observed behavior, and welfare outcomes rather than the repetition of a single moral narrative.

Conclusion

Helping and bystander intervention are organized psychological responses to another's perceived need under conditions of incomplete information, social observation, practical constraints, and varying degrees of risk. The foundational experiments of Darley and Latané established that beliefs about other witnesses and visible group behavior can inhibit an individual's action. Subsequent field experiments, motivational studies, social-identity research, meta-analysis, and direct observation demonstrated that social presence also provides information, support, coordination, and accountability. The scientific record supports conditional mechanisms rather than a universal equation between larger crowds and less help. Darley & Latané (1968) Latané & Darley (1968) Fischer et al. (2011) Philpot et al. (2020).

The strongest inference depends on separating the opportunity to witness from the decision to act, individual from group outcomes, immediate behavior from actual benefit, and experimental manipulation from observational association. Empathy, responsibility, identity, arousal, danger, competence, and institutional organization participate in a shared causal landscape. Findings from campus training and school-network experiments demonstrate opportunities for purposeful intervention while preserving important questions about durability, generalizability, and verified reductions in harm. Banyard et al. (2007) Paluck et al. (2016) Katz & Moore (2013).

A mature psychology of intervention therefore studies the whole event: who perceived the problem, how they understood it, which actions were safe and feasible, how others shaped responsibility, which institutions supported assistance, and what happened to the person in need. That framework connects the historical bystander literature to contemporary settings and gives Artificial Era research a clear empirical agenda for human–artificial support. Continued progress will depend on precise operational definitions, ethical experimental design, transparent reporting, and outcomes that capture effective assistance rather than intention alone.

References

  • Banyard, V. L., Moynihan, M. M., & Plante, E. G. (2007). Sexual violence prevention through bystander education: An experimental evaluation. Journal of Community Psychology, 35(4), 463–481. https://doi.org/10.1002/jcop.20159
  • Batson, C. D., Duncan, B. D., Ackerman, P., Buckley, T., & Birch, K. (1981). Is empathic emotion a source of altruistic motivation? Journal of Personality and Social Psychology, 40(2), 290–302. https://doi.org/10.1037/0022-3514.40.2.290
  • Cialdini, R. B., Brown, S. L., Lewis, B. P., Luce, C., & Neuberg, S. L. (1997). Reinterpreting the empathy–altruism relationship: When one into one equals oneness. Journal of Personality and Social Psychology, 73(3), 481–494. https://doi.org/10.1037/0022-3514.73.3.481
  • Coker, A. L., Cook-Craig, P. G., Williams, C. M., Fisher, B. S., Clear, E. R., Garcia, L. S., & Hegge, L. M. (2011). Evaluation of Green Dot: An active bystander intervention to reduce sexual violence on college campuses. Violence Against Women, 17(6), 777–796. https://doi.org/10.1177/1077801211410264
  • Darley, J. M., & Batson, C. D. (1973). “From Jerusalem to Jericho”: A study of situational and dispositional variables in helping behavior. Journal of Personality and Social Psychology, 27(1), 100–108. https://doi.org/10.1037/h0034449
  • Darley, J. M., & Latané, B. (1968). Bystander intervention in emergencies: Diffusion of responsibility. Journal of Personality and Social Psychology, 8(4), 377–383. https://doi.org/10.1037/h0025589
  • Eisenberg, N., Spinrad, T. L., & Knafo-Noam, A. (2015). Prosocial development. In R. M. Lerner & M. E. Lamb (Eds.), Handbook of Child Psychology and Developmental Science (7th ed., Vol. 3, pp. 610–658). Wiley. https://doi.org/10.1002/9781118963418.childpsy315
  • Fischer, P., Krueger, J. I., Greitemeyer, T., Vogrincic, C., Kastenmüller, A., Frey, D., Heene, M., Wicher, M., & Kainbacher, M. (2011). The bystander-effect: A meta-analytic review on bystander intervention in dangerous and non-dangerous emergencies. Psychological Bulletin, 137(4), 517–537. https://doi.org/10.1037/a0023304
  • Garcia, S. M., Weaver, K., Moskowitz, G. B., & Darley, J. M. (2002). Crowded minds: The implicit bystander effect. Journal of Personality and Social Psychology, 83(4), 843–853. https://doi.org/10.1037/0022-3514.83.4.843
  • Hortensius, R., & de Gelder, B. (2018). From empathy to apathy: The bystander effect revisited. Current Directions in Psychological Science, 27(4), 249–256. https://doi.org/10.1177/0963721417749653
  • Katz, J., & Moore, J. (2013). Bystander education training for campus sexual assault prevention: An initial meta-analysis. Violence and Victims, 28(6), 1054–1067. https://doi.org/10.1891/0886-6708.VV-D-12-00113
  • Latané, B., & Darley, J. M. (1968). Group inhibition of bystander intervention in emergencies. Journal of Personality and Social Psychology, 10(3), 215–221. https://doi.org/10.1037/h0026570
  • Latané, B., & Darley, J. M. (1970). The unresponsive bystander: Why doesn't he help? Appleton-Century-Crofts. https://search.worldcat.org/title/103764
  • 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(4), 443–453. https://doi.org/10.1177/0146167204271651
  • Levine, M., Taylor, P. J., & Best, R. (2011). Third parties, violence, and conflict resolution: The role of group size and collective action in the microregulation of violence. Psychological Science, 22(3), 406–412. https://doi.org/10.1177/0956797611398495
  • Nickerson, A. B., Aloe, A. M., Livingston, J. A., & Feeley, T. H. (2014). Measurement of the bystander intervention model for bullying and sexual harassment. Journal of Adolescence, 37(4), 391–400. https://doi.org/10.1016/j.adolescence.2014.03.003
  • Paluck, E. L., Shepherd, H., & Aronow, P. M. (2016). Changing climates of conflict: A social network experiment in 56 schools. Proceedings of the National Academy of Sciences, 113(3), 566–571. https://doi.org/10.1073/pnas.1514483113
  • Philpot, R., Liebst, L. S., Levine, M., Bernasco, W., & Lindegaard, M. R. (2020). Would I be helped? Cross-national CCTV footage shows that intervention is the norm in public conflicts. American Psychologist, 75(1), 66–75. https://doi.org/10.1037/amp0000469
  • Piliavin, I. M., Rodin, J., & Piliavin, J. A. (1969). Good Samaritanism: An underground phenomenon? Journal of Personality and Social Psychology, 13(4), 289–299. https://doi.org/10.1037/h0028433
  • 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), 990–999. https://doi.org/10.1037/0022-3514.34.5.990
  • van Bommel, M., van Prooijen, J.-W., Elffers, H., & van Lange, P. 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. https://doi.org/10.1016/j.jesp.2012.02.011
  • Warneken, F., & Tomasello, M. (2006). Altruistic helping in human infants and young chimpanzees. Science, 311(5765), 1301–1303. https://doi.org/10.1126/science.1121448