Anchoring survived the replication wars, but its explanation did not

What the effect is

Anchoring is the finding that an initial value, even an arbitrary one, shifts subsequent numerical judgements towards itself. The standard design gives people a number, asks whether some unknown quantity is higher or lower than it, and then asks for an estimate of that quantity. Estimates track the number they were shown first, in the direction they were shown it.

It is worth being precise about what is claimed here, because a loose version of the idea has been stretched to cover almost anything. The well-supported finding concerns numerical estimation under uncertainty when a candidate value is presented before the judgement. It is not a general theory that people are hypnotised by the first thing they hear about any subject.

Why it matters that this one replicated

Over the past fifteen years or so, coordinated projects have taken famous psychological findings and run them again across many laboratories at once, with sample sizes far larger than the originals and analysis plans fixed in advance. The results were sobering. A substantial fraction of well-known effects came back much smaller than reported, or not at all.

Anchoring was among the effects that came through those tests in reasonable shape. That does not make it a law of nature, and the magnitude varies with the task, but it earns anchoring a different status from the many findings that only ever existed in a handful of small studies. When you are deciding how much to trust an idea from this field, the number of independent, well-powered attempts to reproduce it matters far more than how often it appears in popular books.

The mechanism is still argued about

What is much less settled is why it happens. One account says people start from the anchor and adjust towards their own estimate, but stop too early, because adjustment is effortful and there is no clear signal that you have gone far enough. Another says the anchor makes anchor-consistent information easier to recall, so the mind assembles an estimate from evidence that has already been tilted.

These predict different things in different circumstances, and the experimental literature contains support for both. Anchors the person generated themselves seem to behave differently from anchors supplied by someone else, which suggests more than one process is involved. Honest summary: the effect is real, the explanation is contested, and anybody who tells you the mechanism is settled is going beyond the evidence.

Where it shows up with money attached

Negotiation is the obvious case, and the reason so much negotiating advice concerns who names a figure first. An opening number sets the range within which the rest of the conversation happens, and counter-offers tend to be expressed as movements from it rather than as independent valuations. That is a structural advantage for whoever speaks first with a defensible figure.

It shows up in pricing, where a higher reference figure displayed next to a lower one changes how the lower one reads. It shows up in budgeting, where last year’s number frames this year’s discussion regardless of whether last year was sensible. And it shows up in professional estimation, where the first person to say how long something will take has largely determined what everyone else will say.

There is a defensive version of the same observation, and hardly anybody uses it. If you are about to be handed a number by someone with an interest in where it lands, the valuable moment is the one just before it is spoken, because that is the last point at which your own estimate is uncontaminated. Forming a view then feels premature, since the conversation has not yet declared itself as a negotiation. That is precisely why it works.

What actually helps

Very little of the standard advice survives contact with the evidence, but two things have reasonable support. The first is generating your own estimate before exposure — writing down what you think a thing is worth or how long it will take before you hear anyone else’s number. Once an anchor has landed, you cannot un-hear it, so the only reliable move is to get there first.

The second is deliberately considering reasons the anchor might be wrong in the opposite direction, which works better than simply trying to ignore it. Instructions to disregard an anchor tend to fail, in the same way instructions to disregard inadmissible evidence tend to fail, because suppression is not how the underlying process operates.

In a group, the sequencing rule matters more than any individual’s discipline. Collect estimates in writing before discussion begins. A meeting where the most senior person speaks first has effectively taken one estimate and made it look like consensus.