Defusing the Anchor

You now know what anchoring is, how it works, and how many places it hides. Time for the payoff: what to actually do about it. Spoiler — gritting your teeth and “being more objective” is not on the list. The good news is that a handful of process changes genuinely move the needle, and they’re all things you can rehearse before the anchor ever shows up.

You can’t out-willpower it

Here’s the deflating part first. Telling people about anchoring, warning them it’s coming, even paying them for accuracy — all of it barely helps. In study after study, forewarned judges still get dragged toward the anchor almost as much as everyone else.

Why? Because, as we saw in lesson 2, anchoring isn’t a motivation problem. It runs on selective accessibility: the moment you consider an anchor, your brain quietly retrieves the facts, memories, and comparisons that fit it, and those are the pieces sitting on your mental desk when you form a judgment. You don’t feel biased. You feel informed — by a lopsided sample of evidence you never chose. The other mechanism, insufficient adjustment (starting from the anchor and stopping the moment your estimate feels merely plausible), is similarly automatic.

So effort is aimed at the wrong target. You can’t willpower your way out of a slanted memory search. Every defense that follows works by changing the process — what gets retrieved, what number you start from, whether you commit to a point at all — rather than asking you to try harder against your own recall.

Consider the opposite

The single best-supported debiasing move has a boring name and a powerful effect. Consider the opposite: before you commit to a judgment, deliberately generate reasons the anchor is wrong, or arguments for a value far away from it.

Mussweiler, Strack, and Pfeiffer (2000) showed this cleanly. Participants estimating anchored quantities were prompted to think of arguments against the anchor — reasons the true value might be much higher or lower. That single instruction sharply reduced the anchoring effect, where warnings alone had done nothing.

Why it works maps exactly onto the mechanism. Selective accessibility loaded your mind with anchor-consistent evidence. Consider-the-opposite forces the opposite pool — anchor-inconsistent knowledge — to become accessible too. Now both sides are on the desk, and your judgment averages over a fairer sample instead of a rigged one. You’re not fighting the retrieval; you’re running a second, deliberately biased-the-other-way search to balance it.

Worked example. A house is listed at $1,200,000 and you’re deciding what it’s worth. Instead of adjusting down from the list price, you write three reasons it might be worth far less: the kitchen is dated, comparable homes two streets over sold for $900,000, and it’s been on the market for ninety days. Suddenly those facts — which the list price had crowded out — are front of mind, and your estimate settles far below where a passive “adjust down a bit” would have left it.

Its limits. It takes deliberate effort and a moment of friction, so it only helps when you remember to do it — build it into a checklist. And generating implausible opposites (“maybe it’s worth ten dollars”) doesn’t help; the counter-arguments have to be genuine and relevant to shift what’s accessible.

Generate your own anchor first

The best time to beat an anchor is before you ever hear it. Generate your own anchor first: compute an independent estimate — from a reference class, a base rate, your own reasoning — before the other side’s number lands. Then their figure arrives against your prior instead of onto blank paper.

This ties straight to thinking in probabilities. If you’ve already asked “what does a typical case like this run?” and built a rough number from a base rate, the incoming anchor has to argue with something. On blank paper, an anchor is the only landmark in view, and selective accessibility builds your whole estimate around it. With a prior in place, you notice the gap and ask why.

Worked example. You’re negotiating a used-car price. Before you walk onto the lot, you look up that the model’s fair private-party value is about eleven thousand, and you set a walk-away of twelve thousand five hundred. The dealer opens at seventeen thousand nine hundred. Because you arrived with a number, that opener reads as “aggressive opening move,” not “the neighborhood we’re negotiating in.” Without your prior, seventeen-nine would have quietly become the center of gravity.

There’s a flip side worth naming: when you are the better-informed party, consider making the first offer. Anchoring is a weapon as well as a trap — a well-placed, defensible first number pulls the final deal toward your end. Prepare your target and your walk-away first, then decide who should anchor whom.

Its limits. Your self-generated anchor is only as good as the reference class behind it. A prior built from a bad base rate just anchors you on your own mistake. And if you don’t know the domain, making the first offer can leave money on the table — anchor first only when you’re the informed side.

Question the anchor’s relevance and provenance

Not every number deserves a seat at the table. A sharp habit: ask where did this number come from, and does it have any business informing this estimate?

Recall that anchors work even when they’re transparently arbitrary — random spins of a wheel, the last digits of an ID. So a number’s mere presence is not evidence it’s relevant. When an anchor is planted — the “was $1,200, now $799” tag, the extreme opening bid, the “most people give around two hundred” nudge — its job is to move you, not to inform you. Recognizing that is grounds to reject it and re-anchor on something you actually trust.

Even the simple act of naming the anchor out loud weakens it. “That opening bid is deliberately high to pull me up” converts a silent center of gravity into an object you can inspect and set aside. You can’t scrutinize what you don’t notice; provenance-checking drags the anchor into the light.

Worked example. A vendor quotes a “standard rate” of forty thousand and then “generously” drops to thirty. You ask what the standard rate is based on — and it turns out to be a list price no customer has paid in years. Provenance revealed, the thirty-thousand “discount” stops feeling like a deal and starts feeling like the anchor it always was.

Its limits. Provenance-checking tells you whether to trust an anchor, not how much residual pull it still has once you’ve seen it — even discredited anchors keep tugging, so pair this with consider-the-opposite. And it can tip into cynicism: sometimes the number is relevant, and dismissing every figure is its own error (see the caveat below).

Think in ranges, not points

Anchoring loves a single number. So stop giving it one. Think in ranges: instead of committing to a point estimate, state a low end and a high end — and, crucially, derive each end from independent reasoning rather than by nudging outward from the anchor.

This aims squarely at insufficient adjustment. Left alone, you start at the anchor and drift until the estimate feels “not crazy,” which usually stops at the near edge of the plausible range. Forcing yourself to actually construct the far end — “what would have to be true for this to be way higher?” — pushes past that early stopping point instead of parking against it.

Worked example. Asked how long a software project will take, don’t say “about six weeks” (which quietly hugs the sponsor’s hopeful “can we get it by next month?”). Build the ends separately: best case, everything goes right — four weeks; worst case, list what could go wrong — eleven weeks. The honest interval is four to eleven, and it reveals a risk the tidy point estimate hid.

A close cousin: aggregate independent estimates. Ask two colleagues who haven’t heard the anchor for their own numbers and average them. The key word is independent — a “second opinion” from someone who saw the same anchor just launders the same bias. Estimates only cancel each other’s anchors when they weren’t exposed to the same one.

Its limits. People game ranges by making them absurdly wide to look safe — a range you’re not honestly 90% confident in is theater, not calibration. And ranges still need genuinely independent endpoints; if you “spread” symmetrically around the anchor, you’ve just decorated it.

The honest caveat: when the anchor is a gift

Now the twist that separates a sophisticate from a cynic. Sometimes the number you’re handed is good, and starting from it is exactly the right thing to do.

An anchor that is genuinely informative — a relevant comparable sale, an expert’s calibrated estimate, a real base rate — is a rational starting point. Starting there and adjusting for what’s different is just Bayesian updating with a decent prior. That’s not a bug; that’s how good estimation is supposed to work. If a seasoned appraiser tells you similar houses fetch $900,000, ignoring that number to feel “unbiased” would make you worse, not better.

So what actually makes anchoring a bias? Two conditions: the anchor is irrelevant, arbitrary, or manipulative (a random spin, a planted list price, an extreme opening bid with no basis), or your adjustment is insufficient even from a fine starting point. Absent both — relevant anchor, adequate adjustment — you’re just reasoning from evidence.

The real skill, then, isn’t distrusting all numbers. It’s telling an informative anchor from a planted one — and that’s precisely what provenance-checking, consider-the-opposite, and independent estimation are for. They don’t ask you to throw every anchor away; they let you keep the good ones and disarm the rest.

Worked example. Two “starting points” for pricing your car: (a) the dealer’s off-the-cuff seventeen-nine, and (b) the private-party value guide’s eleven thousand from thousands of real sales. Same shape — a number that frames your estimate — but one is a manipulation to resist and the other is a prior to lean on. Same tools tell them apart.