The short version
Somebody says a number. Everything anyone thinks afterwards is measured from it — not agreed with, necessarily, but measured from, which is enough. The argument becomes how far to move away from that number, and the number itself never has to justify its existence.
Whoever speaks first hasn’t won the argument. They’ve drawn the map it happens on.
Where the name comes from
Tversky and Kahneman, 1974. The demonstration is still the best thing about it, because it’s so nakedly unfair.
They spun a wheel of fortune in front of each subject — rigged, though the subjects didn’t know that, to land on either 10 or 65. Then they asked what percentage of United Nations member states were African countries. The wheel has nothing to do with the UN. The subjects had just watched it spin. And the people who saw 10 guessed around 25 per cent, while the people who saw 65 guessed around 45.
A visibly random number, generated in front of them, on a completely unrelated question, moved the answer by twenty points.
What it looks like
- A wheel that stopped on 65
- “Was £200”
- “We were thinking somewhere around forty”
- A placeholder estimate somebody typed into the doc in March
- Six feet apart
- Estimates of UN membership
- What £80 feels like
- Every offer made after it
- The entire budget conversation in October
- Two years of arguing about distance
Someone said a number in a meeting, without deriving it.
Argued about whether it should be a bit higher or a bit lower.
When the anchor becomes the argument
The highest-stakes recent case is the six-foot rule.
The number had a lineage, though not the one usually given for it. The large-droplet picture does go back to the 1930s, but what William Wells actually derived was a size, not a distance: roughly a hundred micrometres, the point below which a droplet evaporates before it has time to fall. The metres in his tables are the height a particle falls from, not the gap between two people. And the bacteriology before him pointed the other way — when Carl Flügge’s laboratory sampled the air around people talking and coughing in the 1890s, plates several metres off came back covered in colonies, and his conclusion was that keeping a certain distance helped little by itself, because the slightest air current carried droplets several metres. ”Large droplets fall within a metre or two” was assembled from those pieces afterwards, and became standard infection-control guidance for decades without the mechanism it claimed ever being established.
Nor did it have a derivation for this particular disease. Different authorities read the same evidence and landed in different places. The WHO said at least one metre. The UK said two metres and printed “6ft” beside it; the United States said six feet and, early on, printed “2 metres” beside that — not two findings but one distance rounded into whichever unit the reader already thought in. Germany, the Netherlands and Australia settled on a metre and a half. A number that doubles depending on which health agency you ask is not a measurement. It also moved without the evidence moving: from July 2020 England’s guidance said two metres — or one, if you took other precautions. A BMJ analysis that summer made the case directly, arguing a single fixed distance was poorly supported — not because distance didn’t matter, but because how much of it you needed depended on ventilation, occupancy, duration, masking and activity. In 2024 Anthony Fauci told a congressional subcommittee that the six-foot figure “sort of just appeared,” and that he was not aware of studies behind it.
Now the anchoring part, which is a separate question from whether distancing worked. Once six feet existed, essentially every argument that followed was conducted in units of six feet. Was three enough? Should it be ten? Could the classroom reopen at three, the restaurant at five, the church at four? An enormous public argument ran for two years with a coordinate system, and the origin of that system was a number nobody had derived.
Notice what mostly never happened: the six itself was rarely the thing being defended. It didn’t have to be. It was the fixed point that everything else was measured from — and that is a far stronger position for a number to occupy than being correct.
There’s a second cost, and it’s the one that lasts. Distance wasn’t the fiction. Closer really was more dangerous and nobody serious disputed it — the BMJ analysis opens by calling physical distancing an important part of controlling the disease. What distance didn’t have was one correct value. The government’s own advisers put the risk at a metre somewhere between two and ten times the risk at two metres, and that range is a factor of five wide because different methods gave answers a factor of five apart. It also assumed two people facing each other; side by side, the question stops being distance and becomes ventilation.
Which is the whole shape of it. How well ventilated the room was, how long you stayed, how many people were in it, indoors or out, whether anyone was masked, which way you were facing — those decided what a safe distance actually was, and the analysis argued for a graded rule rather than a fixed one: more than two metres worth considering in a crowded, badly ventilated bar, less than that needed where the risk was low. A graded rule is hard to paint on a floor. The anchor was one number, so the argument became one-dimensional, and the dimensions left out at the start never made it back in.
None of which means picking a number was a mistake. Acting on incomplete evidence is frequently the only responsible thing available, and a rule people can actually follow beats a more accurate rule they can’t. The failure isn’t choosing a number under uncertainty. It’s that the number stopped being provisional — and became a thing to be defended, by people on both sides who had never once asked where it came from.
Why it doesn’t work
Tversky and Kahneman called it insufficient adjustment: you don’t take the anchor as the answer, you start there and move away from it, and you stop when the number stops feeling wrong — which arrives far too early. That is still the best description of what happens when the anchor is one you came up with yourself.
For a number somebody hands you, which is every case on this page, the account that fits better is less flattering. Checking the anchor against the question — is it higher than that, or lower? — loads your memory with whatever is consistent with it. By the time you come to answer, the evidence available to you has already been filtered. The anchor isn’t a starting point you move away from; it’s the reason some things come to mind and others don’t.
That’s why it doesn’t feel like a bias from inside. You experience yourself reasoning from the anchor, not being moved by it, and those are indistinguishable while it’s happening. There is nothing to catch yourself doing.
Two findings make this worse than the average bias, and they’re the reason it’s worth a page.
The first is that it works on experts, in their own field. Estate agents shown a property with a listing price gave appraisals that tracked the listing — and afterwards said the listing price had played no part in their judgement. They weren’t lying. It genuinely doesn’t feel like an input. By the time they had a number, the comparison had already decided what they could recall about the property.
The second is that being warned barely helps. You can tell people the number is random, tell them it’s designed to influence them, and watch them anchor anyway. This is the practical fact of the matter: attention is not a defence, so anything that depends on you being careful in the moment will fail.
This is not anchoring
- A first number that carries information. Last year’s actual spend, a genuine market rate, a real quote — those should move your estimate. Anchoring is an arbitrary or self-serving number doing work it never earned.
- Opening a negotiation high — as a move. Making an aggressive first offer isn’t a foul. It’s expected, it’s openly adversarial, and nobody is being deceived about what the number is for. What it isn’t is harmless. Opening offers are the best-documented anchors there are, and knowing a number is an opening bid does nothing to stop it dragging the settlement toward itself. The move is fair; the pull is real — which is why the defence below matters most of all here.
- Using a reference point deliberately and saying so. “Let’s start from last quarter and adjust from there” is a stated method, and a stated method can be argued with.
- Being convinced by the case for a number. If somebody showed their working and you found it persuasive, you were persuaded. That’s the system working.
The test is about provenance: where did that number come from, and would I have landed anywhere near it if I hadn’t heard it first?
How to head it off
Because awareness doesn’t work, everything useful here is procedural — something you arrange in advance rather than something you notice in the moment:
- Write your number down before the meeting. Before the email, before the offer, before anyone else’s figure is in the room. This is the entire defence, and nothing else comes close.
- “Before we discuss their figure — what would we have said on our own?” Ask it early. Once the number has been spoken, this question can’t be answered honestly by anybody.
- Ask for the derivation. “How was that arrived at?” An anchor usually can’t answer, and finding that out in front of everyone does more than any objection you could raise.
- Argue against the number — all in one direction. Take the figure on the table and make yourself produce two or three real reasons it’s wrong the same way: if it’s a high opening price, every reason is a reason it’s too high. Don’t balance it with reasons it’s too low. Arguments that push the way the anchor already points don’t loosen its grip, because that half of the case is the half it has already made easy to reach. Considering the opposite is one of the few things shown to reduce the effect, and this is why it works — it’s a quota you fill, not a conclusion you arrive at.
The honest reason to know about this one isn’t that knowing protects you. Mostly it doesn’t. It’s that it tells you the protection has to be a habit you set up beforehand — and decide your number before you hear theirs is a small enough habit to actually keep.