Anchoring bias
Anchoring bias is what makes a claim about a quantity settle near the first number you were shown, even when that number is irrelevant, arbitrary or obviously random. Amos Tversky and Daniel Kahneman described it in 1974, and the uncomfortable part of their result is not that people are influenced by a starting figure. It is that they are influenced by a starting figure they know to be meaningless.
What anchoring bias is
Anchoring bias is a distortion of a quantitative estimate caused by exposure to a reference value before the estimate is made. Tversky and Kahneman described the mechanism as anchoring and adjustment: the mind starts from whatever value is in front of it and adjusts away, and the adjustment stops too early. Two people given different starting points therefore finish at different answers, having reasoned in the same direction from different places.
What distinguishes anchoring from the rest of the cognitive bias family is how little it requires. Confirmation bias needs you to hold a prior belief. The Dunning Kruger effect needs a skill to misjudge. Anchoring needs only that a number was recently visible. It works on strangers, on experts in their own field, and on people who have just been warned that it is about to happen.
Anchoring in one sentence, and the size of the pull
In one sentence: the first number sets the range, and the answer lands inside it. The size of the pull is what makes the finding worth knowing rather than merely amusing. In the original demonstration, groups given a low starting figure and a high starting figure produced median estimates that differed by roughly twenty percentage points on a question neither group had any way to answer precisely.
That gap was produced by a number the participants had watched being generated at random in front of them. No argument was made for it, no source was cited, and it still moved the answers.
The wheel of fortune experiment, and the multiplication task
The wheel of fortune experiment and the multiplication task are the two demonstrations that established anchoring, both from Tversky and Kahneman's 1974 paper in Science, "Judgment under Uncertainty: Heuristics and Biases", and both worth reproducing exactly.
In the first, a wheel marked from zero to one hundred was spun in front of each participant. The wheel was rigged to stop at either 10 or 65. Participants were then asked whether the percentage of African countries in the United Nations was higher or lower than the number the wheel had given, and then asked to state their own estimate. The group that saw 10 gave a median estimate near 25. The group that saw 65 gave a median estimate near 45. The wheel had no information in it, everybody could see it was a wheel, and the answers moved anyway.
In the second, participants had five seconds to estimate a product. One group saw it written as 8 x 7 x 6 x 5 x 4 x 3 x 2 x 1 and the other as 1 x 2 x 3 x 4 x 5 x 6 x 7 x 8. The two expressions have the same value, 40320. Estimates from the descending version ran far higher than estimates from the ascending version, and both groups came in well under the true figure. The first few terms served as an anchor, and there was no time to adjust far enough away from it.
Where anchoring bias shows up, from price tags to forecasts
Anchoring bias shows up wherever a number arrives before a judgment does. Six settings account for most of it:
- Retail pricing. A list price shown next to a discounted price gives the discount something to be measured against, and the list price need never have been charged.
- Negotiation. The first offer on the table shapes the range for the rest of the conversation, which is why the side that opens is not always at a disadvantage.
- Project estimates. A manager who says "should this take about three weeks?" has just set the answer, and the team's revised figure will cluster near it.
- Salary discussions. A stated current figure anchors the offered figure, independently of the work being priced.
- Surveys and forms. The default option and the highest listed value both act as anchors on what respondents choose.
- Forecasting. Last year's number is the anchor for this year's number, and adjustment is usually too small to catch a real change in conditions.
What anchoring bias is confused with
Anchoring bias is confused with three things in particular: framing, confirmation bias and cognitive dissonance. Yes, confirmation bias is a cognitive bias, and so is anchoring; both are members of the same broad category, and they are not two names for one thing. Cognitive dissonance is not a bias at all in the same sense. It is the discomfort described by Leon Festinger in 1957, felt when a person holds two inconsistent beliefs or acts against a belief, together with the changes people make to reduce that discomfort.
Framing, which changes the words rather than the number
Framing is the closest neighbor and the one most often mislabeled as anchoring. Both arrive before you think, and both come from how the question was presented. The difference is the mechanism. Anchoring supplies a quantity that your estimate drifts toward. Framing supplies a description that changes which option looks better while every quantity stays identical, as when the same batch is called ninety percent defect free or ten percent defective.
The question that separates them: was I given a number to adjust away from, or the same number in different clothes? If deleting the number would remove the effect, it is anchoring. If the effect survives because of the wording, it is framing.
| Effect | What it acts on | Trigger | Typical symptom |
|---|---|---|---|
| Anchoring bias | A numerical estimate | Exposure to any reference number | Answers cluster near a figure nobody defended |
| Framing | Which option is preferred | The wording of an identical quantity | The choice flips when the sentence is rewritten |
| Confirmation bias | The search for evidence | Holding a prior belief | Every test could only confirm |
| Cognitive dissonance | A belief, after the fact | Inconsistency between belief and action | The belief quietly changes to fit what was done |
The practical difference: anchoring hits before you think, framing hits at the moment of choosing, confirmation bias shapes how you think, and dissonance rewrites what you thought once the thinking is over. All four can operate on the same decision in sequence. The question that separates anchoring from the last two: did this reach me before I had an opinion, or after?
What actually reduces anchoring
To reduce anchoring, generate your own number before you see anyone else's. Five measures, roughly in order of effectiveness:
- Estimate first, in writing, then look at the quoted figure. An anchor cannot pull an estimate that already exists on paper.
- Build the number from components, since a figure assembled from parts you can each defend is harder to drag than a single intuition.
- Consider the opposite anchor deliberately, asking what argument would support a value far above and far below the one offered.
- Collect estimates independently, because a group that hears one member's figure first is a group with one estimate in it.
- Ask where the number came from, and treat a figure with no derivation as decoration rather than data.
Simply knowing about anchoring helps less than you would hope. Warning people in advance reduces the effect but rarely removes it, which is why the fix is procedural: change the order in which numbers appear, not the attitude of the person reading them.
The test to run before you accept a number
Ask one question of any figure that arrives before your own estimate: if this number had been twice as large, or half as large, would my answer have moved with it? If yes, your answer is partly a report of what you were shown.
Then apply the second test to the number itself: what would I have to believe for this figure to be exactly right, and is there a derivation I could check? A quoted price, a benchmark or a target that nobody can derive is an anchor wearing the clothes of evidence. Deductive reasoning offers no protection here, because an argument can be perfectly valid and still take its starting figure from a spinning wheel.
The psychology of anchoring: anchoring and adjustment, and what has been re-examined
In the psychology literature anchoring bias is not usually called anchoring bias. It is called anchoring and adjustment, and it was introduced by Amos Tversky and Daniel Kahneman in 1974 as one of three heuristics, the other two being availability and representativeness. The difference between the two names is not cosmetic. The popular name labels the error. The research name states a proposed mechanism: that people start from the anchor and adjust away from it, and stop adjusting too soon.
That mechanism has been argued over ever since, which is the honest part of this section. Fritz Strack and Thomas Mussweiler proposed in the 1990s that externally supplied anchors work by selective accessibility instead, making anchor consistent information easier to bring to mind, so the anchor changes what you think of rather than where you start counting from. Nicholas Epley and Thomas Gilovich argued that genuine adjustment happens mainly with anchors people generate for themselves. The upshot is that anchoring names a solid finding attached to a contested explanation, and a page that reports the effect as settled science with a settled cause is reporting one of those two things wrongly.
On how firmly established the effect is, the position is unusually good for this field. Anchoring was among the effects examined in the large multi laboratory replication projects of the 2010s and was reproduced reliably, at a time when several better known results were not. What varies is size rather than existence: the pull is larger for quantities people have no independent handle on and smaller where the person has real knowledge, though expertise does not abolish it.
Three things the popular version overstates:
- That any number moves any judgment equally. The wheel of fortune result is dramatic because the estimate was one almost nobody holds an informed view about.
- That the first offer therefore always wins. An anchor shifts a range; it does not determine an outcome, and a party who has done the arithmetic in advance is much harder to move.
- That knowing about it fixes it. Warnings reduce anchoring without removing it, which is why the countermeasures on this page are about the order in which numbers reach you.
Anchoring bias in finance and behavioral finance
In behavioral finance, anchoring bias is one of the standard named effects, and it takes three recognizable forms. The first is the purchase price: a holder treats what they paid as the reference point for whether an asset is doing well, although the market has no memory of that figure. The second is the recent range, such as a fifty two week high or low, which becomes the frame for whether a current price counts as cheap or expensive. The third is forecast stickiness, where a revised estimate stays close to the previous published estimate because that figure, rather than the underlying situation, is the starting point for the revision.
None of that is advice about what to do, and this page gives none. It is a description of a measurement error: the reference point in each case is a historical artifact, and treating a historical artifact as a standard is exactly the move anchoring makes. The related trap is holding on because of what has already been spent, which is the sunk cost fallacy, and it is a different mistake with a similar feel.