平均人的发明：为什么「正常」既不是自然的，也不是中立的
In 1835 Quetelet constructed a man who never existed — l'homme moyen. The ratio he built was renamed BMI a hundred and thirty-seven years later, then cut at 25 and 30, and a distribution became an identity. The question is not whether "normal" is invented, but who falls on the far side when the cut moves, and whether they have standing to object
1. A man who never existed
In 1835 the Belgian polymath Adolphe Quetelet published a two-volume work, Sur l'homme, in which he set out to give human society what celestial mechanics had given the heavens: laws. If the motion of gas molecules could be described statistically, why not the heights, weights and crime rates of people? He gathered large bodies of anthropometric data and observed that weight grows roughly with the square of height. From this he built a ratio: weight divided by height squared.
The decisive step, though, was not measurement. It was naming. He called the centre of that distribution l'homme moyen — the average man — a person who had never existed in any particular country in any particular year. Quetelet believed the average man was more than a statistical midpoint. It was a type toward which nature tended: deviation from the centre was error, return to the centre was the norm.
A hundred and thirty-seven years later, Ancel Keys renamed the ratio the body mass index in a 1972 paper. Then, in the mid-1990s, the World Health Organization cut it at two lines: 25 for overweight, 30 for obesity. China's current standard uses 24 and 28.
One ruler, two sets of cuts. And today a ratio constructed in 1835 to describe the shape of a population is the basis on which a doctor reaches a conclusion about a single person in a consulting room.
2. Three steps from distribution to identity
The journey from distribution to identity has three steps, and each one looks reasonable on its own.
First: average equals typical. This is statistics used properly. Average height, average income, average lifespan are all meaningful descriptions, and no one is harmed by them.
Second: typical equals normal. Here language does almost all the work, and no argument is required. In French, normal carries both "statistical mean" and "that which ought to be" — norme. English does the same. One word, two jobs, and so the slide from description to prescription makes no sound. The physician-philosopher Georges Canguilhem made the point explicitly in The Normal and the Pathological in 1943: equating the normal with the average smuggles a value judgment in under a statistical fact. Health, he argued, is not conformity to a mean but the capacity to institute new norms in a changing environment.
Third: normal equals ought. Once "normal" is written into diagnostic criteria, insurance rates and drug indications, it stops describing you and starts adjudicating you. A continuous distribution is cut into a binary: sick or well, within limits or over them.
The arbitrariness of that cut has a remarkably clean demonstration. In 1913 the American psychologist Henry Goddard ran intelligence tests on immigrants at Ellis Island. The version of the story that circulates most widely — that eighty per cent of immigrants were classified as feeble-minded — is a misreport. He tested a pre-selected sample: thirty-five Jewish, twenty-two Hungarian, fifty Italian and forty-five Russian immigrants, all of whom had already been judged to fall between "obviously feeble-minded" and "obviously normal". Within that group, 83, 80, 79 and 80 per cent respectively were classified as feeble-minded by the tests. Goddard himself stated in the 1917 publication that the results did not apply to immigrants in general, nor to first- and second-class passengers.
What happened next is more interesting. When the results struck even him as absurd, he did not redesign the test. He lowered the threshold. After the adjustment, only about forty per cent remained "feeble-minded".
The same people, the same measurement, the same researcher. The cut moved, and two in five of them stopped being patients and became normal again. The threshold was not discovered. It was decided.
3. The case against: without a cut, there is no action
Three objections deserve to be taken seriously, and none of them is weak.
First, clinical practice has to be binary. Insurance must be priced, drugs must be approved for indications, public health must set a poverty line. You may say the 25 in BMI is arbitrary, but no hospital can write a prescription for "a BMI somewhere between 24.8 and 25.2". In practice, abolishing thresholds means abolishing decisions.
Second, some norms are hard. Body temperature, serum potassium, blood glucose, glomerular filtration rate — these ranges are not social constructions. Departures from them have definite physiological consequences, reproducible and verifiable across populations. Relativise "normal" wholesale and you drift toward a dangerous nihilism: if nothing can count as abnormal, then pain, failure and death become merely different lifestyles.
Third, averages do predict. Quetelet's average man is a fiction, yet actuarial tables work. An insurer does not need to know whether Mr. Zhang will reach eighty; it needs to know how many of a hundred thousand will. The validity of a population regularity is not cancelled by the unpredictability of an individual.
All three are correct. But they do not show that the constructivist account is wrong. They mark a more precise boundary: the average is reliable as a description of a population and unreliable as a verdict on a person. The danger is not that averages get computed. It is that they get carried across the border — from describing a hundred thousand people to judging one. Gilbert Ryle called this a category mistake in The Concept of Mind: attributing to one kind of thing a property that belongs to another, and thereby generating a question that can never be answered.
4. The real question: who pays for the wrong cut
If "normal" is unavoidably a convention, then the test of a norm cannot be whether it is true. It usually is not, and it does not need to be. The test is three operational questions.
One: is the cut contestable? Who set it, on what data, and is there a route of appeal. In 1841 the English engineer Joseph Whitworth proposed a uniform screw thread, collapsing the mutually incompatible threads of British workshops into a single specification. This was progress not because the specification was closer to some truth, but because it was public and reproducible: any shop could measure it, verify it, and object to it. The authority of a standard comes from its being testable, not from its being unquestionable. Simon Winchester traces exactly this line in The Perfectionists: the modern world was not built by making things more accurate, but by defining accuracy as something transferable and reproducible.
Two: is the cost shared? Whether those who set the threshold are also bound by it. Goddard's tests were dangerous not because they were inaccurate but because the measurer stood outside the measured: he paid nothing for the line he drew. The whole problem of eugenics compresses into that sentence — the measurer was not in the sample.
Three: is there an exit? Whether an individual can be reassessed, or is overruled by the threshold once and for all. The trouble with BMI is not its crudeness but its role as terminus rather than starting point: a heavily muscled athlete is classified as overweight, and no one asks a second question.
These three criteria have theoretical support. Herbert Simon's notion of satisficing in The Sciences of the Artificial — not optimising, but stopping at good enough — changes what a threshold is. If a threshold is a classifier, it must claim to be carving nature at its joints. If it is a stopping rule, it need only say: this is sufficient for now, and it can be revised. The second use requires no ontological correctness, only revisability.
Kuhn's "normal science" in The Structure of Scientific Revolutions describes the other face of the same mechanism. A paradigm specifies not only what counts as a problem but what counts as a solution and what counts as an anomaly. The "normal" is never a list handed over by nature; it is an accepted set of judgment standards applied repeatedly to particular cases. Its force comes from consensus, and consensus can be renegotiated.
5. The judgment, and its limits
My judgment is this: "normal" is a necessary fiction, but it must be a revisable one. Whether a given norm is dangerous is not a question about its truth. It is a question about whether it can be challenged, whether its costs are shared, and whether there is an exit. Missing any one of the three, it stops being a tool and becomes a power.
That judgment has clear limits, and they need stating.
In acute, high-consequence, well-understood settings — sepsis, myocardial infarction, newborn screening, drug toxicity — thresholds save lives and debating them is a luxury. There is no time for individual assessment, and the physiology is clear enough to support a hard cut. These criteria barely apply there. Extending them into such settings is not rigour; it is dogma.
Their real domain is chronic, weakly understood, highly heterogeneous settings: body weight, cholesterol, psychiatric diagnosis, educational assessment, performance grading. What these share is a continuous distribution, an unclear mechanism, and large individual variation — and a cut that, once made, redistributes identity and resources.
One further caution. Anti-standardisation can itself become dogma. If "normal is constructed" becomes a reason to reject every threshold, the people who suffer most are rarely those who set the standards. They are the people who were protected by them: workplace injury determinations, disability assessments, poverty lines. All of these are cuts, and all of them rescue someone. Abolishing them does not return judgment to individuals; it returns it to whoever has the most bargaining power.
So the danger is not that someone defines the normal. The danger is that the definer never has to settle the bill for the definition.
