The accuser under suspicion

A healthy person presents at a psychiatric hospital and says he hears a voice. It says only three words: “empty”, “hollow”, “thud”. Once admitted he behaves unremarkably, states that the voice has gone, and waits to be recognised as sane. That is how the psychologist David Rosenhan, writing in Science in 1973, described the experiment that made him famous.

What Rosenhan actually put on record at his own admission to Haverford State Hospital has been established since 2019 — and it was not three harmless words. He reported voices that had been growing louder for six months and that he attributed to a radio. He said he had held a copper pot to his ear to shield himself from the signals. He described ideas of reference, pronounced ideas of persecution, and two months of suicidal thoughts. The paper’s central charge — that a trivial symptom is enough to disappear into psychiatry — fails for his own case above all.

What the paper claimed

Eight pseudopatients, the report said, had presented at twelve US hospitals and had been admitted without exception. Eleven of the twelve admissions carried a diagnosis of schizophrenia, one that of a manic-depressive psychosis. The stays lasted between seven and 52 days, 19 days on average. None of the pseudopatients was unmasked by staff — though some were, Rosenhan wrote, by fellow patients.

The paper’s force lay less in this tally than in its scenes. Ordinary behaviour suddenly appeared in the chart as a symptom: taking notes as compulsive writing, turning up early for meals as a pathological fixation on oral needs. Rosenhan’s point was not merely that diagnoses can be wrong. A label changes the frame within which every further observation is read — and inside a closed institution it carries forward from one observer to the next. His most famous sentence accordingly reaches far beyond the design meant to carry it:

The paper became one of the most cited indictments of psychiatric diagnosis and was invoked again and again in the reform debates of the 1970s.

The objection of 1975

The first contradiction came not from historians but from within the field — and it was aimed at the logic, not the data. In 1975 Robert Spitzer, who a few years later would take charge of DSM-III, replied that Rosenhan’s finding showed something other than what he claimed. Someone who reports a hallucination ought to be examined; a hospital admitting him is not a diagnostic failure but the expected response. And the discharge diagnosis Rosenhan cited as proof that the label could not be shaken — “schizophrenia in remission” — proved the opposite. It is an unusual formulation, and it records precisely what the physicians had seen: a person without symptoms. Spitzer’s mocking diagnosis for the paper was “logic in remission”.

For a long time, then, the dispute ran as an argument about method: what does this procedure measure at all? Rosenhan’s design did not test whether professionals given the same information arrive at the same judgement. It tested whether professionals notice that they are being lied to. That is a different question, and its answer says little about the reliability of diagnoses. The time frame came on top of that: the institutions examined belonged to an era of care whose criteria were vague, whose wards were overcrowded, and whose discharge decisions were shaped in part by administrative logic.

What the check of the records found

In 2019 the journalist Susannah Cahalan moved the criticism from method to data. For The Great Pretender she searched Rosenhan’s papers and tried to find the other pseudopatients. Two of the eight could be established: Rosenhan himself, who appeared under the name “David Lurie”, and his doctoral student Wilburn Underwood. The remaining six stayed untraceable after years of searching. The psychiatry historian Andrew Scull, who examined the material in 2023, considers it likely that they never existed.

Underwood’s case, the one that could be checked, could not be squared with the paper. The paper speaks of seven days in an institution with 8000 beds; in fact it was eight days in one with 1500 beds. The discharge diagnosis “schizophrenia in remission”, on which a considerable part of the argument rested and against which Spitzer’s rejoinder had worked, does not appear in his record.

A ninth participant does not appear in the paper at all. Harry Lando spent 19 days on an open ward, by his own account had well over an hour of contact with medical staff each day, and in a publication of his own in 1976 described his stay as largely positive. Rosenhan struck him from the analysis in a footnote on the grounds that he had falsified his biography — which was true of every pseudopatient, and of Rosenhan most of all.

The most concrete evidence of falsification lay in Rosenhan’s own papers. Among them is a draft version with nine pseudopatients whose tables on staff contact carry numbers identical to those in the published version with eight. Two samples of different size cannot produce the same means; and since Lando reported hours of staff contact daily, both data sets are demonstrably doctored. Then there is what is missing: the announced book remained a fragment, Rosenhan never produced the full documentation, and the files contain neither the identities of the pseudopatients nor the names of the hospitals. At Science, too, the manuscript and reviewer records no longer exist.

In 2023 Scull called the study a successful scientific fraud. That is a harsh verdict, and it does not prove that every single observation was invented. But the more cautious wording carries far enough as well: central parts of the report cannot be independently verified today, and the surviving evidence does not support the certainty of the famous story.

What the damaged study does not undo

A second error would be to discard the whole criticism of the hospitals of that period along with Rosenhan’s credibility. Overcrowding, coercion, little privacy and often demeaning treatment are documented by numerous other sources — Rosenhan was never needed for that. The labelling effect is just as real as a research question: prior information directs attention, shapes interpretation and can write itself forward through a file. It is only that Rosenhan’s paper does not demonstrate the effect; it illustrates it.

Modern diagnosis responded to the problems of that era with more explicit criteria, structured interviews and reliability studies — Spitzer’s own programme. That has not made it infallible. Reliability means that assessors more often reach the same conclusion; validity additionally requires that the category capture something real and useful. And what a label does to the person and to those around them is a third, ethical question that no reliability coefficient settles. Rosenhan’s narrative blended these three levels into a single dramatic test — that was its rhetorical strength and its scientific weakness.

What remains

Little of what the paper claims is secure, and much of what it set in motion is. Two of the eight pseudopatients have been identified, six have not; the one case account that can be checked departs from the text in duration, hospital size and discharge diagnosis; a ninth case that went well was struck out on grounds that applied to all of them; and two versions of the same data set show identical figures for different numbers of people. Only the name for this remains in dispute: whether one calls it fraud, as Scull does, or stops at “not verifiable”. The choice of word changes nothing about how much weight the study can bear.

How to recognise the case in everyday life: a study whose raw data, locations and participants are never disclosed is not evidence, even when its cause is just. Rosenhan was often right on the substance — about labels, about asylums, about the power of the file. He simply did not show it; he told it. Precisely because psychiatric diagnoses decide questions of liberty, treatment and self-image, their critics are held to the same standard as the diagnoses themselves: reasoned, reviewable, revisable. The greater the consequence, the higher the threshold — for the diagnosis as for the accusation.