Anchoring and Adjustment: Why the First Number Stays

Anchoring is the pull a number already in the room exerts on the number a person arrives at. Adjustment is the move away from it, and the finding is that the move is reliably too small. The anchor does not need to be relevant, or plausible, or offered in good faith, and knowing about it does not remove it.

Anchoring rests on something about estimation that almost nobody notices while they are doing it. Nobody produces a number from nothing. Estimation starts somewhere, and then it moves. Asked what a second-hand scooter is worth, a person does not build the answer from parts and labour: they think of a figure they have heard, then push it up or down for age, condition and how soon they need it gone. The remembered figure has now become part of the answer. Because every estimate begins somewhere, the place it began is always one of its inputs, and it is the one input nobody writes down. Amos Tversky and Daniel Kahneman set this out in Science in 1974, in the paper that named the three heuristics, and anchoring is the one that acts on a number rather than on a category or a feeling.

What is an anchor, and what can act as one?

An anchorA number already present when an estimate is being formed, which pulls the estimate towards itself. is any number that happens to be present while an estimate is being formed and that pulls the estimate towards itself. The definition is that short, and notice how little it demands. Nothing in it says the number must be about the same thing. Nothing says the number must be accurate, or recent, or produced by anybody who knew what they were doing. The definition says only that the number was there.

The shape is clearer outside finance than inside it. A shopkeeper says four hundred for a chair the buyer was vaguely willing to pay two hundred for. The counter will not be two hundred. The counter will be something like two eighty, and two eighty will feel to the buyer like their own idea. An aunt says at the start of a wedding conversation that the catering will run to eight lakh. Every figure argued about for the next four months will orbit eight lakh, and none of the people arguing will be able to say why eight lakh rather than five. The first number spoken did not persuade anybody. Persuasion was never the mechanism: the first number set the place the arithmetic starts from.

In a decision about money, the numbers most often doing this job are dull and easy to miss. A purchase costWhat was paid for a holding. A purchase cost is a fact about the past and about one particular person, not a fact about the thing itself. is one. A figure printed on a statement is another. A round number somebody said out loud in a meeting is a third. A level published in a note that somebody expects a holding to reach is a fourth, and it does its work on every person who reads the note. Because a purchase cost feels like a fact about the holding when it is only a fact about the holder, it is the strongest anchor in most decisions about money.

What is the Adjustment Heuristic, and what is wrong with the adjustment?

The adjustment heuristicProducing an estimate by starting from a number that is already available and then moving away from it, rather than building the answer from scratch. is the second half of the mechanism, and it is where the error actually happens. Anchoring on its own is only a starting point. The adjustment heuristic describes what comes next: the number already present is taken as the start, the estimate moves away from it in whatever direction the evidence points, and then it stops. Three steps, and the whole finding lives in the third one.

Consider how a person stops. Nothing in the room announces that a number has become correct, so correctness is not what ends the search. People stop when the number stops feeling wrong. Correctness and comfort are different tests, and comfort is satisfied much earlier. Because two eighty is still visibly connected to four hundred, it stops feeling wrong long before two hundred does. The adjustment heuristic fails not because people refuse to move but because a plausible answer arrives before a defensible one does, and plausible is the test that actually gets applied. Tversky and Kahneman named this insufficient adjustmentStopping the move away from a starting number too early, so the answer stays nearer that number than the evidence supports. in the same 1974 paper, and it has been the standard account of the second half of the mechanism ever since.

The stopping rule explains why anchoring survives effort. Because the stopping rule has not changed, a person who tries harder moves further and further is still not far enough. Effort adjusts the size of the step. Effort does not touch the rule that decides when the stepping stops.

Put numbers on that too, using the holding set out below. An adjustment covering 25 per cent of the distance from Rs 3,00,000/- stops at Rs 2,73,750/-, one covering 65 per cent stops at Rs 2,31,750/-, and a strenuous 90 per cent stops at Rs 2,05,500/-. Even the strenuous version sits Rs 10,500/- above the evidence it was moving towards.

Why does the movement stop too early?

Three things make the stopping rule fire early, and it is worth naming them separately because they suggest different repairs.

The first is that plausibility is a range, not a point. Asked what a flat two streets away is worth, a person holds not one number in mind but a band. Adjustment stops as soon as it reaches the near edge of that band. The near edge is the first defensible-sounding figure encountered on the way, not the middle of the band and certainly not the far edge. Starting from above, the move stops near the top of the range. Starting from below, it stops near the bottom. Same person, same evidence, two different answers.

Put the case numbers on that. Starting from the purchase cost of Rs 3,00,000/- the estimate stops at Rs 2,52,750/-, and starting from Rs 1,00,000/- the very same evidence and the very same 45 per cent produce Rs 1,42,750/-, so two answers Rs 1,10,000/- apart have come out of one set of facts.

The second is that moving is effortful and stopping is free. Every step away from the starting number costs a little work and produces a little discomfort, and there is no signal that arrives to say keep going. The third is that the starting number is usually the only thing in the conversation that feels solid. Everything else is judgement, and judgement is uncomfortable to stand on, so the mind stays within sight of the one number nobody is arguing about. Insufficient adjustment is not laziness: it is what happens when the only stopping signal available is the feeling that an answer has become defensible enough to say out loud.

Does an anchor have to be relevant, plausible or offered honestly?

No, no, and no, and each of those answers is a separate finding worth holding on to.

An anchor does not have to be relevant. In the original experiments reported by Tversky and Kahneman in Science in 1974, a spinning wheel generated a number in front of the participants. Everybody in the room could see the number had nothing to do with the question being asked, and estimates still moved in the direction of the number that came up. Nothing about that number carried information. The number was simply present.

An anchor does not have to be plausible. An absurd starting figure still pulls, and an opening quote of four hundred for a two hundred chair is not a mistake by the shopkeeper. And it does not have to be offered in good faith. A person who says a figure first in order to move somebody has used exactly the same mechanism as a statement that prints a purchase cost with no intention at all. Because an anchor works without being relevant, plausible or honest, spotting that a number is none of those three does nothing to stop it working. Seeing through a number and escaping it are separate acts, and only the first is available to anybody.

What does one real decision look like when the adjustment falls short?

Take the clearest anchor in the Palash decision log, an invented record of one household's decisions. On 12 October, Meera Sundaram sold Suvarna Chemicals Limited whole and kept Kesari Logistics Limited. The holding she kept had cost Rs 3,00,000/- and stood at Rs 1,95,000/- on the 30 September statement, a fall of Rs 1,05,000/- or 35.0 per cent. Her stated reason was that she would sell Kesari Logistics Limited when it got back to Rs 3,00,000/-.

Now put a number on the shortfall. Suppose the defensible figureThe number the available evidence actually supports, arrived at without reference to any starting point. is the Rs 1,95,000/- the statement itself carries, treated as evidence rather than as a conclusion anybody has reached. Suppose adjustment covers about 45 per cent of the distance from the starting number to that figure, a round proportion standing for the order of magnitude the original experiments described rather than a measured constant. The distance is Rs 1,95,000/- less Rs 3,00,000/-, or minus Rs 1,05,000/-. Forty-five per cent of that is minus Rs 47,250/-. Starting from Rs 3,00,000/- the estimate lands at Rs 2,52,750/-.

Consider where that sits. The estimate lands Rs 47,250/- below the anchor and Rs 57,750/- above the statement value, so it agrees with neither of the two numbers actually printed on the statement. Only one of the two numbers the estimate sits between was ever evidence, and the other was a purchase cost, so Rs 2,52,750/- is not a compromise between two views: the whole of the Rs 57,750/- gap was produced by where the reasoning started.

There is a second way to write that same arithmetic, and it makes the size of the contribution impossible to miss. An estimate that moves 45 per cent of the way is 55 per cent of the starting number plus 45 per cent of the evidence, or Rs 1,65,000/- plus Rs 87,750/-.

Read the rule at three settings and the pattern states itself. At an anchor of Rs 1,00,000/- the estimate is Rs 1,42,750/-, at Rs 3,00,000/- it is Rs 2,52,750/-, and at Rs 5,00,000/- it is Rs 3,62,750/-, leaving gaps to the evidence of Rs 52,250/-, Rs 57,750/- and Rs 1,67,750/- respectively.

How to Check for Anchoring in a Valuation Discussion, and where the check starts

A valuation discussion here means people in a room, or on a call, arriving at a number together. A conversation is not a mechanism in any market, and the check below examines how that conversation ran rather than producing a value. Arriving at a value is a separate subject.

The first of the check's three moves stops working the moment it is skipped, so the three run in a fixed order.

First, before any figure is spoken aloud, everybody writes a figure down along with the reason behind it. Writing before speaking is the only move in the whole procedure that prevents anchoring rather than detecting it, and it costs about ninety seconds. Once a number has been said out loud it is in everybody's estimate, including the estimate of the person who thinks they disagreed with it, and nothing done afterwards takes it back out. Second, the counterfactual is put out loud: what would this figure be if the first number were unavailable? People answer that question surprisingly well when it is asked directly, and badly when it is left implicit. Third, somebody asks who benefits from that particular number arriving first. The question is not an accusation. Sometimes the answer is nobody, and the number arrived first because it was printed at the top of a document.

None of the three asks anybody to be immune. The omission is deliberate, and it is what makes the check a procedural correctionA change to how a discussion is run, rather than a change to how any individual thinks. rather than a resolution to think more carefully.

Confirmation Bias vs Anchoring: which one is acting?

Anchoring and confirmation bias get blurred constantly, and the blurring is understandable because both often appear in the same half hour of the same conversation. The two errors are still distinct, acting at different moments on different objects, and the check for each is different.

Anchoring acts while an estimate is being formed, and what it moves is the number produced. Confirmation biasSeeking out and giving extra weight to evidence that supports a view already held, while treating evidence against it more sceptically. acts after a view already exists, and it moves the evidence a person looks for. Set side by side, the order becomes obvious: anchoring is upstream of a conclusion, confirmation bias is downstream of one. Anchoring distorts an estimate, confirmation bias distorts a search, and a single discussion can carry both because they attack different steps of it.

One test separates them in practice: what would have to be different for the answer to change? If the answer is a different number sitting on the table when the estimate was formed, that is anchoring. If the answer is a different set of documents being read, or the same documents being read with the same scepticism on both sides, that is confirmation bias. Confirmation bias in full is covered separately, and appears here only far enough to draw the line.

Does knowing about anchoring protect against it?

Barely, and the experiment behind that answer was built so that nobody could claim the number carried information. In the original work, participants watched a number being produced in front of them by a device that everybody could see was chance, and their estimates still moved with it. Telling people in advance what is about to happen to them shrinks the effect. The warning does not remove it.

The reason is worth stating. Anchoring is not a belief, so there is nothing to correct. An anchor is the place the arithmetic started from, and a start cannot be un-started. Asking somebody to ignore a number they have already seen is the same kind of instruction as asking them to forget a word: the attempt itself keeps the thing in view. Because awareness reduces anchoring without removing it, every workable correction changes the procedure rather than the person.

The conclusion is unusually practical. A room full of careful people is not what the correction needs. The requirement is a rule about paper rather than about character: figures get written before they get spoken.

When is starting from a number the right thing to do?

Very often, and the point needs no hedging. Taking anchoring to mean that starting numbers are bad is both false and disabling. Almost all competent estimation starts from a number. Last year's electricity bill is the right place to begin an estimate of this year's. The rent a neighbour pays is the right place to begin an estimate of the rent on the flat next door. A business much like it was last year is well estimated by starting from what it did last year and moving for what changed.

A single question separates the sound case from the damaging one: does the starting number carry information about the answer? Last year's bill does. The flat, the appliances and the household have not changed much. A purchase cost does not. Nothing about one person's payment at one moment says what a holding is worth now. The practice of starting somewhere is not the problem; the problem is a starting number that carries no information and is never asked to justify itself.

So the working rule is not to stop starting from numbers. The rule is to say out loud what makes a particular starting number informative, and where that sentence cannot be finished, to treat the figure as a number that happened to be present rather than as a place to begin.

How does somebody working with other people's money actually use this?

Devika Rao, the adviser at Palash Advisory Services Private Limited, does not try to spot anchoring in Meera Sundaram by watching her face. She changes two small things about how the conversation runs, and both are cheap.

The first is the order of the meeting. When a holding comes up for review, the figure on the statement is not read out first. Meera Sundaram is asked what she would want to do about the holding if it had arrived in her hands yesterday at no cost at all, and only then is the statement opened. The question is not a trick. Asking it gets an estimate formed before the two most powerful numbers in the room, the purchase cost and the current value, have been spoken. The second is the log itself. Every decision in the Palash decision log carries a written reason, and a reason that consists only of a number is sent back. A written reason that says back to Rs 3,00,000/- names the anchor in the writer's own handwriting, and no cheaper anchoring detector has been built.

The log gives some idea of how far that second habit travels. Across quarters 5 to 8 of the Palash decision log, the 20 people who had adopted a written checklist recorded a reason on 34 of 41 decisions, being 82.9 per cent, against 19 of 63, being 30.2 per cent, for the other 40. Eight quarters and 60 people cannot carry a claim about returns, so no such claim is made.

A lender does something structurally similar when a valuation is commissioned before the asking figure is disclosed, and an analyst does it when the working is built before the number somebody else expects is looked up. A person deciding alone can run the same procedure with a notebook: the figure and the reason go on paper, then the statement is opened, then the two are compared, with curiosity about the gap rather than embarrassment. Every version of this correction works by controlling when numbers enter the room, and none of them works by asking anybody to think harder.

What does anchoring not explain?

Anchoring explains the number a person arrives at. Anchoring does not explain why somebody wanted that number to be high, and motive is a different mechanism entirely. Why a loss feels heavier than an equivalent gain belongs to loss aversion rather than to estimation, and the evidence somebody chose to read belongs to confirmation bias, drawn above. A person can start from a poor number and still reach a sound answer, and can start from an excellent number and reach a poor one, so a single decision that came out badly is never proof that an anchor caused it.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited, the Palash decision log, Suvarna Chemicals Limited and Kesari Logistics Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.