Anchoring, and how one number moves the next

An everyday example

A shop tag reads "was 90 dollars, now 45." The 90 may or may not reflect what anyone actually paid before, but once it is seen, 45 looks like a bargain against it. Shown the same 45-dollar item with no earlier number attached, the same shopper often judges the price quite differently, even though nothing about the item has changed.

The 90 is doing work it has no real claim to: it was simply the first number in view, and the mind measures the second number against it rather than against some independent sense of what the item is worth.

The classic experiment

Amos Tversky and Daniel Kahneman reported the demonstration that gave anchoring its name in their 1974 Science paper. Participants watched a wheel of fortune, secretly rigged to stop on either 10 or 65, spin and land. They were then asked two questions about a topic the wheel obviously had nothing to do with: what percentage of the countries in the United Nations were African nations. First, participants said whether they thought the true percentage was higher or lower than the number the wheel had just shown; then they gave their actual best estimate. The group who saw the wheel stop on 10 gave a median estimate around 25%; the group who saw it stop on 65 gave a median estimate around 45%. The wheel's number carried no information whatsoever about African Union membership, and participants knew that, yet the two groups answered a genuine factual question quite differently depending only on which arbitrary number they had just seen.

Jacowitz and Kahneman extended the demonstration in 1995 to a wide range of everyday quantities: the height of the tallest redwood, the population of Chicago, the freezing point of vodka, and many more. For each quantity they set a low and a high anchor from an earlier pilot survey, then asked new participants to say whether the true value was above or below the anchor before estimating it directly, exactly as in the wheel study. Across the whole set of questions, estimates given a high anchor consistently and substantially exceeded estimates given a low anchor, showing the wheel result was not a one-off: anchoring shows up reliably across very different kinds of quantities, familiar and unfamiliar alike.

Does it replicate?

Replication grade: Strong: one of the more robust effects in the multi-lab Many Labs project

Anchoring is among the effects that were retested directly in the large, multi-site Many Labs replication project, which reran a set of classic psychological findings in dozens of laboratories at once specifically to check whether they held up outside their original setting. Anchoring came through as one of the more robust and consistent effects in that collection: it reproduced clearly across the great majority of the participating labs, with comparatively little variation in its direction or presence from site to site, which is not true of every classic finding retested there.

The size of the pull varies with the question and the anchor, and expert knowledge of a topic can reduce it somewhat, but the basic pattern, a number you have just been shown pulling your subsequent estimate toward it, is one of the most consistently observed effects in the judgement and decision-making literature.

A simulated pull toward the anchor

The diagram on this page simulates, rather than measures, a set of estimates: seven invented answers given a low anchor and seven given a high anchor, both compared to a single true value. It is built to show the shape of the pull, not to report a real result; the anchoring game further down this page produces your own actual answers, which the game's own result compares directly.

Try it: the anchoring game

Eight estimation questions, each preceded by a comparison number set far above or far below the truth. Answer normally, then see whether your answers after high numbers ran higher than your answers after low numbers.

How to catch it

Anchoring works even on people who know about anchoring, which is why the fixes below are about changing the estimate, not just knowing the effect exists.

• Before hearing any number, form your own independent estimate first, even a rough one, and write it down.
• When a number is presented as a comparison, ask where it came from and whether it has any real connection to the quantity you are estimating.
• Deliberately consider reasons the true value could be far from the number you were just given, not only reasons it could be close to it.
• In a negotiation, notice who states a number first; making the first offer sets the anchor other people will be pulled toward, and receiving one means the same is being tried on you.
• Seek out an independent source, such as a published figure or a second, differently anchored estimate, before settling on a final number.