Ask a stranger how many keys are on a piano and they will not conjure a number from the void. They will grab whatever number is lying around — the one you just said, the one on the wall, the one they saw five seconds ago — and shuffle away from it until the guess feels “close enough.” That grab-and-shuffle is one of the most reliable moves in the entire human mind, and it has a name.
Before you read — take a guess
A game-show wheel spins and lands on 7 — obviously random. Then someone asks you to guess how many countries are in Africa. Compared to a person who saw the wheel land on 65, your guess will most likely be:
Anchoring, defined precisely
Anchoring is the tendency to lean on an initial value — the anchor — when estimating an unknown quantity, so that your final estimate is pulled, or assimilated, toward that anchor. The anchor can be a number you were just handed, a first offer, a partial calculation, or a figure you plucked yourself.
The second half of the mechanism is adjustment: once you have an anchor, you move away from it toward what you actually believe. In principle you keep adjusting until you reach your best estimate. In practice you don’t. You stop early — a pattern called insufficient adjustment — quitting as soon as your guess reaches the near edge of the range that seems plausible, rather than pressing on to the middle of it.
Think of it as dropping anchor in a harbor and then rowing toward open water. The rope is elastic, and you get tired. You row a bit, feel the pull, and call it a day far short of where you meant to be. Your final position is a compromise between where you dropped anchor and where you wanted to go.
The general shape, in plain words:
estimate ≈ anchor plus only a fraction of the distance from the anchor to the truth.
If the anchor is 10 and the truth is 30, you don’t land on 30. You land somewhere like 22 — you closed most of the gap but not all of it. Shift the anchor to 65 and you’ll land around 45. Same truth, different anchors, estimates that never fully converge.
Two moving parts, one bias
Anchoring is not one error but two working together: an anchor that shouldn’t matter, and an adjustment that stops too soon. Fix either half — ignore the anchor, or adjust all the way — and the bias disappears. The trouble is that both halves fail at once.
Anchoring lab
Feel the pull
An anchor is any number in view when you estimate. Drag your anchor and watch the estimate get dragged with it — then set how far people adjust and watch two arbitrary anchors pull two groups apart.
What percentage of the world’s countries are in Africa?
- Your estimate
- 42%
- Still off by
- 14%
- Anchoring gap
- 36%
Start from an anchor of 50% and the estimate lands at 42% — dragged most of the way from the anchor, not toward the truth (28%). With people adjusting only 35%, the low-anchor group guesses 16% and the high-anchor group 52%: a 36% gap conjured out of two meaningless numbers. Adjusted a little, then stopped far short — textbook insufficient adjustment.
Two groups, two arbitrary anchors
The wheel of fortune
The founding demonstration comes from Amos Tversky and Daniel Kahneman in 1974. They sat participants in front of a wheel of fortune marked 0 to 100 and spun it. Unknown to the participants, the wheel was rigged to stop on either 10 or 65. Then came the question: what percentage of countries in the United Nations are African?
First, people said whether the true figure was higher or lower than the number the wheel had shown. Then they gave an actual estimate. The results were almost comic. People who saw the wheel land on 10 guessed about 25% on average. People who saw 65 guessed about 45%.
Read that again: a wheel. Everyone watched it spin. Nobody thought a roulette wheel had secret knowledge of African geopolitics. And the anchor moved their answers by roughly twenty percentage points anyway.
The worked example. Suppose the honest answer someone would give with no anchor at all is around 35%. Give them the anchor 10 and they adjust upward — but insufficiently — landing near 25. Give them the anchor 65 and they adjust downward, again insufficiently, landing near 45. The anchor sets the direction of travel and the fatigue sets the stopping point.
Pitfall: 'but I knew it was random'
The most seductive misconception is that awareness is a cure. Participants could see the wheel was rigged theater, and it changed almost nothing. Knowing an anchor is arbitrary does not unhook you from it. The pull operates below the level where “that’s irrelevant” lives.
The multiplication trick
Anchors don’t have to be handed to you by a wheel. You can manufacture one yourself, mid-calculation, without noticing. Tversky and Kahneman showed this with a stopwatch and a product.
They gave one group five seconds to estimate:
8 × 7 × 6 × 5 × 4 × 3 × 2 × 1
and another group five seconds to estimate:
1 × 2 × 3 × 4 × 5 × 6 × 7 × 8
Same eight numbers, same true product — 40,320 — just written in opposite order. Five seconds isn’t enough to actually multiply, so people did a few steps and extrapolated from there. The first few steps became the anchor.
The descending group, starting 8 × 7 × 6…, hit big early partial products and guessed a median of about 2,250. The ascending group, starting 1 × 2 × 3…, saw tiny early products and guessed a median of about 512.
Two lessons hide in this one experiment. First, order flips the anchor: the same numbers produce a four-fold difference in guesses purely because of which partial product you see first. Second, both groups were catastrophically low — 2,250 and 512 are nowhere near 40,320. That’s insufficient adjustment in its purest form: people started from a small partial product and adjusted up, but nothing like far enough.
Ready? You have five seconds. Estimate, do not compute: 8 × 7 × 6 × 5 × 4 × 3 × 2 × 1. Whatever you blurted, the true answer is 40,320. If you guessed a few thousand, congratulations — you are a perfectly normal human who anchored on an early partial product and quit adjusting way too soon. Doing the ascending order first would have made you guess even lower.
Why did the group estimating 1×2×3×4×5×6×7×8 guess lower (~512) than the group estimating 8×7×6×5×4×3×2×1 (~2,250)?
Coherent arbitrariness
Here’s where it gets unsettling. In 2003 Dan Ariely, George Loewenstein, and Drazen Prelec ran a study that weaponized the most meaningless number in your wallet. They asked participants to write down the last two digits of their Social Security number, then to treat those two digits as a dollar price — “would you pay that many dollars for this bottle of wine?” — and then to state the maximum they’d actually bid on wine, gadgets, chocolate, and other goods.
The Social Security digits have exactly zero relevance to what a bottle of wine is worth. And yet participants whose two digits fell in the top fifth bid up to about three times more than those in the bottom fifth. The garbage number became the anchor for the whole auction.
But now the twist that names the phenomenon. Within each person, the relative prices stayed perfectly sensible. Everyone — high-anchor and low-anchor alike — still paid more for the better bottle than the worse one, more for the fancier keyboard than the plain one. The absolute level was arbitrary, yanked around by a random anchor; the relative ordering was coherent and rational.
Ariely and colleagues called this coherent arbitrariness: preferences that are internally consistent (coherent) while resting on an anchor that is essentially random (arbitrary). You reliably know that A is worth more than B, and you have almost no idea what either is worth in dollars — so the first number in the room fills the vacuum.
Where you'll meet this in the wild
Coherent arbitrariness is why the first price quoted in any negotiation matters so much, why “manufacturer’s suggested retail price” exists at all, and why a menu’s absurdly expensive dish makes the second-most-expensive one feel reasonable. Your sense of relative value is sharp; your sense of absolute value is a hostage to whatever anchor showed up first.
Pick a term, then click its definition.
Anchoring is not stupidity
It would be easy to file all of this under “people are dumb.” That reading is wrong, and it will mislead you.
Using a starting value is often the correct move. If a real estate agent tells you the last three houses on the street sold for around 400,000 dollars, anchoring on that figure is not a bug — it is a prior, a rational starting point built from genuine evidence. Estimating from a relevant reference point is exactly what a good reasoner should do. This is the same instinct as thinking in probabilities: you start from a sensible base rate and update from there.
The bias is narrower and sneakier than “people are irrational.” It has two specific failure modes:
- The pull persists even when the anchor is irrelevant — a rigged wheel, a Social Security number, a made-up first offer.
- The adjustment is too small even when the anchor is relevant — you start from a decent reference point and still quit before reaching your best estimate.
So the skill to build isn’t “never use starting points.” It’s knowing which starting points have earned the right to move you, and how far you should have traveled from them.
The one-sentence version
Anchoring is smart reasoning (start from a reference and adjust) running two millimeters too far off the rails: the reference doesn’t deserve the influence, and you stop adjusting too soon.
Lock it in
A car dealer opens with a sticker price of 48,000 dollars on a car worth about 34,000. You talk them down to 41,000 and feel victorious. What just happened?
Check your answer to continue.
You now have the whole machine in view: an anchor that grabs your estimate, an adjustment that fatigues too early, and a mind that stays coherent about relative value while its absolute judgments blow around in whatever wind arrives first. Next we’ll look at just how strong and stubborn this pull is — including the moment it survives even when people are paid to beat it.