What Is Anchoring? Why the First Number Mentioned Changes Your Answer
By the BrainSnail editorial team. How these articles are written and checked, and how to tell us when one is wrong.
An arbitrary number encountered just before an estimate pulls that estimate towards it, even when the number is obviously irrelevant and even when people are warned. The effect is robust, well replicated, and exploited commercially every day.
The original demonstration
The classic experiment asked participants to estimate a quantity they could not know, after first spinning a wheel that produced an apparently random number and asking whether the true value was above or below it. Estimates were reliably pulled towards whatever the wheel produced, despite everyone having watched it being generated at random, which rules out any explanation based on the number carrying information. The effect appears with numbers that could not possibly be relevant, including the last digits of an identification number and arbitrary values mentioned in passing. It survives warnings, since telling people about it beforehand reduces it only slightly. It appears in experts making judgements in their own field, including estimates of property value and sentencing decisions, which is the finding with the most uncomfortable implications. And the size of the effect is substantial rather than marginal, frequently shifting estimates by a large proportion.
Why it happens
Two explanations account for different versions and both have support:
- •Insufficient adjustment, where a person starts from the anchor and moves away from it, stopping as soon as the answer becomes plausible rather than continuing to the best estimate
- •Selective accessibility, where considering the anchor brings to mind information consistent with it, which then dominates the subsequent judgement
- •The first explanation fits self-generated anchors, where a person knows the starting point is wrong and adjusts
- •The second fits externally provided anchors, where the number is simply evaluated and its associations activated
- •Both predict the effect will be larger when the person is uncertain, which is observed
- •Neither requires the person to be careless, which is why expertise provides so little protection
Where it costs money
Commercial use of the effect is pervasive and mostly deliberate. An initial asking price sets the frame for a negotiation, and evidence from property and salary negotiations indicates that whoever makes the first offer generally does better, which is exactly what the effect predicts and which contradicts the common advice to let the other party speak first. Displaying a high recommended price alongside a sale price makes the sale price appear better regardless of what anything is worth. Menus place an expensive item to make the rest seem moderate. Limits on quantity, even generous ones, raise the amount purchased by supplying a number. Initial requests in fundraising set the size of subsequent donations. In each case the mechanism is the same and the defence is the same, which is to determine a value independently before encountering any number, since once the number is seen the effect has already operated.
Anchors you create yourself
A distinct version of the effect operates without anyone supplying a number. Asked to estimate something unfamiliar, people frequently start from a related value they do know and adjust, and the adjustment is systematically too small, so the estimate stays close to the starting point. Estimating the boiling point of water at altitude by starting from the sea level figure produces an answer too near it. Estimating how long a task will take by starting from an ideal uninterrupted run and adjusting for reality produces the familiar underestimate. The same mechanism appears in forecasting, where a projection built by adjusting last year's figure moves less than conditions warrant. The defence differs from the external case, since the problem is not the anchor but the insufficiency of the adjustment, and deliberately considering reasons the answer might be much further from the starting point measurably reduces the error.
The replication question
The effect is among the more robust findings in its field and it has not escaped scrutiny. Large coordinated replication efforts have confirmed the basic phenomenon across many laboratories and countries, which places it in better standing than several contemporaneous findings that failed to replicate. Effect sizes in those replications are generally somewhat smaller than in the original reports, which is the usual pattern and reflects publication practices at the time rather than a problem with the finding. Boundary conditions have been identified, with the effect weaker when the person has strong prior knowledge, when the anchor is implausible enough to be rejected outright, and when incentives for accuracy are substantial, though even those reduce rather than remove it. Debate continues about which mechanism operates in which circumstances. The practical conclusion is unaffected, since the effect is real, sizeable and cheap to defend against.
The takeaway
An arbitrary number pulls a subsequent estimate towards it even when generated by a visible random process, and warnings barely help. Starting from the anchor and stopping adjustment too early explains some cases, and the anchor activating consistent information explains others. Whoever makes the first offer in a negotiation generally does better, and the only real defence is deciding a value before hearing any number.