The Thirty-Eight Witnesses Never Existed

On 27 March 1964, a fortnight after the murder of Kitty Genovese in New York, a report in the New York Times carried the line:

The diagnosis therefore stood in the subheading, years before anyone had investigated it. In the body text below it the number was thirty-eight. Posterity remembered the higher figure, and remembered it so thoroughly that for half a century it opened every introduction to social psychology: thirty-eight people who watched for half an hour and did nothing.

Those witnesses never existed. And the field’s second founding story, Solomon Asch’s line experiments, is told just as wrongly — only with the sign reversed. There the number is in the files, but the textbooks turned it around. Both cases show the same thing: how a discipline makes parables out of usable data — and how it gets rid of them again.

What Asch’s numbers actually say

The arrangement stands in file 011 of this archive: two cards, one standard line, three comparison lines, seven briefed confederates who answer wrongly in one voice. On average, participants joined the false majority in a good third of the critical trials. The textbook version made of this: man is a herd animal. Asch’s own reading ran the other way. It shook him that things went as far as they did, but his data showed independence first: two thirds of the answers stayed correct, a quarter of the participants never yielded, and a single ally who told the truth made conformity collapse — down to single-digit percentages. Asch understood his series as a study of the conditions of independence; the reception read it as a study of compliance, because compliance made the better headline.

The fine structure contradicts the herd thesis too. In further series Asch varied the size of the majority, and the relationship is not the expected one: a single opposing voice did almost nothing, two did a fair amount, three came close to the maximum. Beyond that, more staff at the table changed little. It is not mass that presses but unanimity — and precisely for that reason a single dissenter is enough to break it. In the post-interviews Asch also distinguished who yielded from genuine doubt, who from politeness, and who had adjusted the perception itself — the rarest group. In 1955 Deutsch and Gerard separated the two motives every introduction has taught since: informational influence, where the group serves as a source of data, and normative influence, where it sits as a jury.

It is worth taking the task itself more seriously than the headline does. Asch’s comparison was deliberately trivial; the lines differed clearly, and solitary judges in the control condition were wrong in less than one per cent of cases. That is exactly what makes the finding strong: whoever yielded did not yield for lack of information but against their own evidence. At the same time it limits the reach. About judgements under real uncertainty — the normal case in life — the line task says little; there it is often simply rational to give the majority weight. Sherif’s older experiments with the apparently wandering point of light show how group judgements under uncertainty turn into norms that persist long after the group has dispersed. Conformity research is about both, and only the confusion of the two cases yields the caricature of the herd animal.

How the data became a textbook sentence

That this confusion was no accident was demonstrated by Ronald Friend, Yvonne Rafferty and Dana Bramel in 1990. They read the accounts of Asch in social psychology textbooks across several decades and found a directed drift: the more recent the book, the more prominent the compliance and the less that remained of Asch’s own point. A study of the conditions of standing firm became evidence for group pressure — not through new data, but through copying.

And the repetitions? In 1980 Perrin and Spencer found practically nothing among English students: a single conforming answer in 396 critical trials. They declared the Asch effect a “child of its time.” The meta-analysis by Bond and Smith across 133 studies from 17 countries, by contrast, found it almost everywhere, with a declining tendency across the decades and marked differences between cultures. The basic finding stands; its size is a function of place and era.

What was left of the thirty-eight witnesses

The Genovese story became the symbol of the cold metropolis and the founding document of bystander research. The later state of the files, above all the reconstruction by Manning, Levine and Collins in 2007 and the research into the trial, left little of it standing. Most of the “witnesses” had heard screams but seen nothing. The second, fatal attack happened in a stairwell, out of sight. At least one neighbour said he called the police, and a neighbour, Sophia Farrar, ran down and held the dying woman in her arms. Thirty-eight indifferent eyewitnesses did not exist. The Times corrected its own account decades later.

The irony is double, because the research the false story set off is good research. In 1968 Latané and Darley built two controlled situations. In the first, smoke poured into a waiting room: those sitting alone reported it in three cases out of four; with two seemingly unconcerned confederates beside them, only one in ten did. In the second, participants heard over an intercom what appeared to be a fellow participant having a seizure: those who believed themselves the only listener almost always fetched help; those who assumed four further listeners did so in fewer than a third of cases. The effect carries two names — diffusion of responsibility and pluralistic ignorance. The second is the more unsettling: everyone reads from the calm faces of the others that nothing much is happening, and so contributes to the very calm the others are reading.

What bystander research actually shows

This effect has survived decades of testing. But it explains delay in helping under ambiguity, not cold-heartedness — and in exactly the kind of situation where the legend places it, it is at its weakest. A collective analysis by Fischer and colleagues in 2011 separated dangerous from harmless emergencies and found that the clearer the danger, the smaller the bystander effect; under physical threat the presence of others can even make intervention easier, because one expects support from them. Camera research has confirmed this in the field: in 219 video recordings of real confrontations in inner cities in three countries, at least one person intervened in roughly nine cases out of ten. People help, especially when the danger is unambiguous. What makes them hesitate is ambiguity plus an audience.

Why did the Genovese legend last fifty years all the same? For the same reasons as the Stanford legend this journal treats elsewhere. It served a need — the unease about the anonymous city. It had a number that was easy to remember. And it took an early fixed place in textbooks whose chapters copy one another. That the correction gathered speed around the fortieth anniversary and was pressed by social psychologists themselves belongs to the field’s credit; that generations of students learned the wrong number belongs on the bill.

What remains

Little is settled about the parables and a good deal about the data. Asch’s effect is real, but it measures the force of unanimity, not of numbers, and it disappears as soon as one person dissents. The bystander effect is real, but it describes hesitation under ambiguity, not indifference, and it reverses when the situation is clear. What remains contested is how far either finding reaches beyond the laboratory: the line task has little in common with everyday judgements, and the field results of the camera research come from a slice of situations in which somebody happened to be filming.

In daily life both are recognisable by a handful of practical moves. Anyone leading a group to a decision should break the unanimity on purpose — one person who objects first changes the situation more than ten who agree later. And anyone who needs help should address a single person and name the situation: “You in the blue coat, call the emergency services.” The sentence now appears in first-aid courses and descends in a straight line from Latané and Darley’s waiting rooms, because a person who has been addressed is no longer an audience but the one responsible. That both branches of research dismantled their own founding legends — one with trial files, the other with meta-analyses — is not a mishap but the normal operation this archive documents: stories age, files decide.