Bystander effect / diffusion of responsibility (Latané & Darley, 1968)
The more people present at an emergency, the less likely any one of them is to act. Responsibility diffuses across the group: each person feels only a fraction of it, waits for someone else to move first, and the group as a whole can end up doing nothing that any single member, alone, would have done.
Three mechanisms. (1) Diffusion of responsibility — the felt share of duty shrinks as headcount grows. (2) Ambiguity / pluralistic ignorance — others standing still reads as evidence that nothing is really wrong, so each person misreads the others inaction as information. (3) Evaluation apprehension — fear of acting wrongly in front of an audience.
The experiments. Smoke-filled room (Latané & Darley, 1968): subjects filling out a questionnaire as smoke seeped in — alone, 75% reported it within six minutes; with two passive confederates who ignored it, 10%; with two other naive subjects, 38%. Seizure over intercom (Darley & Latané, 1968): a subject who believed one other person was listening sought help 85% of the time when a confederate appeared to have a seizure; believing five others were listening, 31%, and slower. Subway collapse (Piliavin et al., 1969): a staged collapse in a New York subway car — an apparently drunk victim got slower, less help than an apparently ill one, and help arrived fast overall, which the authors read through a cost-reward lens rather than pure diffusion. Good Samaritan (Darley & Batson, 1973): seminary students hurrying to give a talk passed an injured man by — 10% helped under high time pressure versus 63% under low pressure. The case that started it all, Kitty Genovese (1964), is itself a caution: the "38 witnesses did nothing" version is exaggerated; later accounts put the witness count lower and record calls that were made.
Modern correction. A CCTV study of real public conflicts (Philpot et al., 2019) found at least one bystander intervened in over 90% of cases. The per-person effect is real — each individual probability of acting falls as the crowd grows — but "nobody acts" is not what usually happens. The digital version runs the same way: a help request sent to many recipients gets fewer replies per recipient than the same request sent to one named person.
What is lab-verified vs field practice. See Verified evidence below. Direct address is lab-verified. DRI, on-call, and handoff rules are field practice from incident management (SRE tradition), widely adopted but not lab experiments — do not cite them as experimental findings.
Verified evidence (what the literature actually supports). Meta-analysis (Fischer et al., 2011, Psychological Bulletin; 105 studies, 7,700+ participants; overall g = -0.35, small-to-moderate): the effect is attenuated when the situation is perceived as dangerous, when a perpetrator is present, and when intervention costs are physical; bystanders familiar with each other weaken it further; and bystanders given an active instruction reverse it (g = +0.53, against -0.53 for passive bystanders). CCTV field data (Philpot et al., 2019, American Psychologist; 219 real public conflicts): at least one bystander intervened in 90.9%, averaging 3.76 interveners, and more bystanders raised the likelihood of intervention. Training RCT (Green Dot; Coker et al., 2019, Journal of Family Violence; 5-year cluster RCT, 26 schools, 73,044 students): bystander training reduced violence acceptance, and sexual violence perpetration and victimization.
Children: large-organizations · small-teams · open-teams Kin: spiral-of-silence · third-person-effect · cialdini-principles · models