The number
that pulls

Spin a carnival wheel numbered 0 to 100. Watch it land. Now: what percentage of UN member states are African nations? The wheel's number cannot possibly be relevant. It is about to be relevant.

Kahneman and Tversky rigged a wheel of fortune to stop only at 10 or 65, spun it in front of University of Oregon students, and had them write down the result. Then two questions: is the percentage of African nations in the UN higher or lower than the number you just wrote? And what's your best guess of the actual percentage?

Students whose wheel said 10 guessed 25% on average. Students whose wheel said 65 guessed 45%. Same lecture hall, same question, same total absence of information in the wheel — and a twenty-point gap in the answers, pointed exactly where the wheel pointed.

Sit with the strange part: the wheel spun in front of them. Nobody believed it encoded UN trivia. A number you know is noise still pulls. In code terms, an anchor is a bad default parameter: you know it's a placeholder, you'd never ship it — and it leaks into the return value anyway.

This pull — a presented number dragging your estimate toward it, regardless of the number's merit — is the anchoring effect. It may be the strangest reliable fact in this book: an obviously irrelevant input moves the output. In everyone. Including you, including after you finish this chapter.

The first story is deliberate. You start from the anchor because it's the only number in the room, and you walk away from it: 65 is too high… 55 is too high… 45… could be? You stop the moment the answer stops feeling wrong — which is the near edge of your plausible range, the edge closest to the anchor. It's a loop with an early-exit bug: while (feelsWrong(x)) x -= step terminates at the boundary of your uncertainty, never at its center. Kahneman and Tversky called this adjustment, and the defect is that it's insufficient — running the loop longer costs attention, and chapter 1 already told you how System 2 feels about spending that budget. Load people up with a memory task, or catch them tired, and they stop even earlier.

The second story involves no trying at all. "Was Gandhi older or younger than 144 when he died?" Nobody entertains 144. But to answer the comparison, your associative machinery has to briefly act as if it were true — and acting-as-if means summoning everything compatible with it: an image of a very, very old man, ancient, frail, historic. That evidence is now activated and sitting in working memory when the real question — so how old was he? — arrives. This is chapter 2's machinery doing exactly its job on exactly the wrong input: the anchor selectively wakes the evidence that agrees with it. Anchoring as priming.

Both mechanisms are real; they own different situations. When you knowingly estimate from a wrong-but-nearby starting point, you're running the stop-early loop (System 2, out of budget). When a number merely passes through your attention, the priming path runs on its own (System 1, never asked). Most real anchors get you both ways at once — which is part of why the effect is so hard to kill.

How hard does an anchor pull? There's a clean unit. Give two groups different anchors, measure how far apart their estimates land, and divide: (difference in estimates) / (difference in anchors). That ratio is the anchoring index. 100% means people moved one-for-one with the anchor — puppets. 0% means immune. The wheel scores (45 − 25) / (65 − 10) ≈ 36%. The redwood question in the widget above — asked of visitors at the San Francisco Exploratorium — scores 55%, the book's flagship number. Arbitrary digits buying a third to half of the whole gap.

Surely professionals are immune? Real-estate agents were walked through an actual house with the full information packet; the only manipulated field was the asking price. Their "independent" assessments tracked it with an index of 41% — and afterward they insisted, with some pride, that the list price had played no part in their judgment. Business students with no real-estate experience scored 48%. Expertise bought seven points of resistance and a complete illusion of immunity. Expertise shrinks the pull far less than experts believe it does.

One thing needs saying plainly, because chapter 2 spent a section burying priming studies that didn't survive the replication crisis: anchoring is not one of the casualties. It is among the most robust, most replicated effects in this entire book — hundreds of studies, in labs, courtrooms, supermarkets, and auction rooms, with stakes hypothetical and painfully real. When this chapter says the pull is real, that claim is load-bearing, and it holds.

Once you know the shape, you see it everywhere — usually placed on purpose:

• List prices. The asking price of a house isn't information, it's an opening move — and the agents' 41% says it works on the professionals paid to see through it.
• "Limit 12 per customer." A supermarket soup display with that sign roughly doubled average purchases — about 7 cans versus 3.3 with no limit. (That classic comes from a marketing lab later disgraced in the replication crisis; the quantity-limit effect itself has been reproduced by others in the field. Hold the decimals loosely, keep the moral: the 12 was never a limit, it was an anchor.)
• Damage caps. A cap on awards is sold as a ceiling; it behaves like a magnet. Cap damages at $1 million and small claims inflate toward it — the number pulls up everything that would have settled low.
• First offers. In salary and price negotiations, the first plausible number sets the field. Every subsequent move is measured as a distance from it — which is precisely why the party who names it first is usually happy to.

The general rule: any number on the table is working on you. Including numbers nobody endorses, numbers everyone agrees are ridiculous, and numbers you've been explicitly warned about — like the one in this card.

Maybe the wheel worked because students didn't care. So raise the stakes and make the anchor personal noise. Dan Ariely and colleagues had MBA students write the last two digits of their own Social Security number, style it as a dollar price, and then bid real money on wine, chocolate, and keyboards. High-digit students bid dramatically more — for some items, up to three times as much. Their own arbitrary digits, which they had just labeled as arbitrary, set their willingness to pay.

And the professionals with the highest stakes of all: German judges, fifteen-plus years on the bench, read a shoplifting case and then rolled a pair of dice — loaded to land on 3 or 9 — before stating a sentence. Dice said 9: eight months on average. Dice said 3: five months. Dice.

You cannot ignore an anchor, because "ignore that number" is an instruction to think about that number — attention is exactly how the priming path gets fed. Debiasing by willpower has the track record you'd expect: none. What you can do is fight it with structure:

• Argue the other side, on purpose. List concrete reasons the anchor is too high (or too low), ways it could mislead, evidence against it. This consider-the-opposite move is the one debiasing technique with decent replicated evidence — because it runs the priming machinery in reverse, deliberately waking the incompatible evidence the anchor suppressed.
• Derive before you look. Form your own estimate from your own inputs before reading theirs. Your number is still an anchor — but it's yours, built from comparables instead of tactics.
• Refuse absurd openings. When a first offer is outrageous, do not counter near it — a counter-offer concedes the range and the negotiation becomes "distance from their number." Kahneman's advice: make it clear, loudly and even theatrically, that this number is not a basis for discussion — then walk, or re-anchor with your own.

Negotiation is where this chapter turns into money: the first plausible number frames everything that follows. And "frames" is a loaded word — chapter 11 is about how the mere wording of an option flips preferences between identical outcomes. Anchoring is framing's numeric cousin, and the bargaining table is where they work as a team.