The First Number in the Room

Tversky and Kahneman spun a wheel of fortune, rigged to stop at either 10 or 65, in front of their participants. Then they asked what percentage of United Nations member states were African. The people who had watched the wheel stop at 10 guessed around 25 per cent. The people who had watched it stop at 65 guessed around 45.

Everybody had watched the wheel being spun. Nobody thought it knew anything about Africa. It moved their answers by twenty points regardless.

That is anchoring: an initial value pulls a subsequent estimate toward itself, even when the value is transparently uninformative. It is among the most replicated findings in the whole of judgement research, which, given the next three chapters of this book, is worth saying plainly at the start. The basic effect is not in doubt.

What is in doubt is the explanation, and the story most people are told is the weaker one. The intuitive account is insufficient adjustment: you start from the anchor, adjust away from it because you know it is wrong, and stop too early. That happens, but mainly for anchors you generate yourself, and it does not explain the wheel of fortune at all; nobody adjusts away from a number they never accepted.

The better-supported account is selective accessibility. Encountering a number makes you test it, and testing it means retrieving whatever is consistent with it. Asked whether Gandhi was older than nine when he died, you briefly summon a picture of a child; asked whether he was older than a hundred and forty, you summon extreme old age. By the time the real question arrives, your memory has been stocked with material that fits the anchor, and you answer from a biased sample of your own knowledge. The anchor does not pull the answer. It changes what you have to answer with.

This explains the thing that makes anchoring uncomfortable, which is that expertise does not protect you. Estate agents shown identical properties with different listing prices valued them differently, and insisted afterwards that the listing price had played no part. German judges given a sentencing recommendation that had been generated by rolling dice in front of them gave sentences that followed the dice. Experienced people have more knowledge to retrieve selectively, not less.

The countermeasures follow from the mechanism. Telling yourself to ignore the number does not work, because the retrieval has already happened. What has better evidence behind it is the "consider the opposite" move: deliberately generating reasons the anchor might be too high, and then too low, which forces you to sample the other side of your own memory. It is clumsy and effortful and it shrinks the effect rather than removing it, which is about as good as debiasing gets.

The practical consequence is a negotiating fact that is well known to negotiators and not to everyone else. Whoever states a number first has set the frame everyone subsequently argues inside, including themselves. A counter-offer is an adjustment from the anchor, and the phrase "meeting in the middle" only means anything relative to two endpoints, one of which was chosen strategically.

Which is why the recommended response to an aggressive opening number is not a counter-number and not an argument about it, but a refusal to let it sit: change the subject, ask how it was arrived at, take a break. The anchor works on whatever is in the room. Getting it out of the room is the only reliable defence, and it is much easier than not being affected by it.