A record, re-derived

The Parable of the 38 Witnesses

Everyone knows the story. Thirty-eight people watched a woman murdered and not one lifted a finger. Almost none of it is true — and yet the lab effect it launched is real, and runs exactly backwards to the thing you'd expect a crowd to buy you.

On the night of 13 March 1964, Kitty Genovese was stabbed to death outside her apartment in Kew Gardens, Queens. Two weeks later The New York Times ran the story that would put her name in every psychology textbook for the next fifty years.

Look closely and the story is already arguing with itself: the headline says 37, the first sentence says 38. That small crack runs all the way down. When researchers went back to the police files, the trial record, and the surviving witnesses — most thoroughly in a 2007 review by Rachel Manning, Mark Levine and Alan Collins, pointedly titled The Parable of the 38 Witnesses — they found that the famous number, and almost everything attached to it, does not survive contact with the record.

§1 · The story vs. the recordWhat actually happened

The parable
The record
38 witnesses watched the murder.
No evidence of 38. Perhaps a dozen perceived fragments — a shout, a scream — and the fatal attack happened in a stairwell almost none could see.
They watched from their windows, transfixed.
Manning et al. (2007) found no evidence that witnesses observed the murder, or that any who did remained inactive.
Nobody called the police.
At least two witnesses contacted or tried to contact police; one neighbour's shout drove the attacker off — briefly.
She died alone, unhelped.
A neighbour, Sophia Farrar, ran to the vestibule and held Kitty in her arms. She did not die alone.
Three separate attacks over half an hour.
The corrected account is two attacks at two locations; Winston Moseley returned about ten minutes after fleeing the first.

The story was largely built after the fact — a parable assembled from a real and terrible crime. (Kitty's brother Bill spent a decade re-investigating it; the 2016 documentary The Witness is his account, and even Moseley's 2016 Times obituary conceded the original story "grossly exaggerated the number of witnesses and what they had perceived.")


§2 · The effect is realSo the psychologists built a real crowd

Two young researchers, John Darley and Bibb Latané, were unconvinced that thirty-eight people were simply callous. They suspected the opposite: that it was the very presence of others that froze each one. So in 1968 they built the emergency in a lab.

A student sits alone in a booth, discussing personal problems with other students over an intercom — the others are actually tape recordings. One "student" begins to have a seizure, choking, pleading for help, then falls silent. The only real question is: does the subject leave the booth to get help, and how fast? The single thing Darley and Latané varied was how many other people the subject believed could also hear the emergency.

The diffusion machine — drag the crowd

0
The intuition — surely someone helps
85%
If every witness were as ready as a lone one and acted on their own: chance at least one of them helps.
Measured — chance this witness helps
85%
Darley & Latané 1968, by the end of the fit.

Alone, the witness acted in a mean of 52 s.

● measured (3 tested group sizes)  ·  — intuition, if everyone acted alone

Press to run 50 hypothetical sessions at this crowd size — the tally finds the rate Darley & Latané reported.

Ready.

At every group size Darley and Latané tested, the crowd did the opposite of what intuition promises. Alone, 85% of witnesses went for help. With one other person believed present, that fell to 62%. With four others, to 31% — and those who did act took three times as long. Responsibility, they argued, doesn't add up across a crowd. It divides. Everyone's problem becomes no one's.

The check

The blue line is not data — it is what a crowd should buy you. If each of N witnesses helped independently at the lone-witness rate a = 0.85, the chance someone acts is 1 − (1−0.85)ᴺ: 85% alone, 97.7% with one other, 99.99% with four. Help should become a near-certainty.

The orange dots are data — the measured chance a given witness acts: 85 → 62 → 31%. It falls where the model rises. That opposite sign is the whole finding: the decline cannot be independent choices; it is diffusion of responsibility. Only three group sizes were ever tested (0, 1, 4 others), so there are only three dots — the space between them is left honestly empty.

Every figure here — the 85/62/31 rates, the 52/93/166 s latencies, and the independence arithmetic — is recomputed in research/bystander/verify.mjs (21/21 checks pass).


§3 · Name the uncertaintyWhat we actually know now

Here is where the honest account gets harder than either the myth or the classroom version. The founding story is a parable. But is the effect real? Mostly yes — with edges the textbooks skip.

Meta-analysis · Fischer et al. 2011
g = −0.35  ·  105 studies, 7,700+ people

Across five decades of experiments the bystander effect is real and moderate. But it weakens when the emergency is dangerous and a perpetrator is present — the danger clarifies the situation, and other bystanders become potential backup rather than diffusers. In those cases the effect can shrink toward zero.

Real conflicts on CCTV · Philpot et al. 2020
91%  ·  219 street conflicts · 3 cities

When researchers watched actual public fights caught on camera in Amsterdam, Lancaster and Cape Town, someone intervened in 90.9% of them — and bigger crowds made intervention more likely, not less. The catch: that measures whether anyone in the crowd acts, not the odds for a given person — which can still fall. Aggregate rescue and individual diffusion can both be true at once.

So the fair summary is three-layered, and none of the layers is the bumper sticker: the Genovese story is a myth; the lab effect is real but softens exactly when it matters most, in real danger; and in the messy world outside the booth, help usually comes — often because there is a crowd. The lesson was never "people are callous." It was subtler and more useful: whether you act depends on whether you believe the responsibility is yours. In a crowd, make it yours.