Anchoring Bias: How The First Number You Hear Controls Your Thinking

Amos Tversky and Daniel Kahneman introduced anchoring in their landmark 1974 Science paper "Judgment Under Uncertainty: Heuristics and Biases." The spinning wheel experiment — technically called the wheel-of-fortune study — was designed to demonstrate that obviously arbitrary numerical information influences subsequent numerical estimates.

The finding was striking not just for its effect size but for what it implied about the nature of the anchoring mechanism. The wheel-of-fortune setup ruled out informational accounts of anchoring (the wheel clearly communicated nothing about UN membership). What remained was a purely psychological mechanism: the first number encountered sets a reference point, and adjustment from that reference point is systematically insufficient.

Subsequent decades of research have confirmed the robustness of anchoring across domains, subject populations, and stimulus types. A partial summary:

Northcraft and Neale (1987): Real estate agents given different listing prices for the same property gave estimates that tracked the listing prices, despite reporting that they used professional appraisal methods and hadn't relied on the listing price. Expert domain knowledge does not eliminate anchoring.

Englich and Mussweiler (2001): Experienced criminal court judges were influenced by anchoring in sentencing recommendations. In one condition, the anchor was set by a randomly rolled die. Judges who rolled higher (and therefore received a higher anchor) gave systematically higher sentencing recommendations.

Ariely, Loewenstein, and Prelec (2003): Participants were asked to write down the last two digits of their Social Security number, then bid on consumer goods. Participants with higher SSN digits bid significantly more. A random, arbitrary, personally irrelevant number anchored willingness-to-pay.

Critcher and Gilovich (2008): People who were shown a jersey with the number 55 rated a football player's abilities higher than people shown a jersey with the number 10. Numerical context bleeds across domains even when there is no rational basis for the transfer.

The breadth of these demonstrations makes anchoring unusually well-supported as a cognitive phenomenon. It's not a laboratory artifact — it appears in field settings, with experts, with explicit warnings, with arbitrary anchors, across cultures.

The mechanism underlying anchoring has generated substantial theoretical debate. Three main accounts:

Insufficient adjustment (Kahneman and Tversky): Estimates start from the anchor and adjust in the correct direction but not far enough, because adjustment is a cognitively effortful, serial process that is terminated prematurely — typically when the first plausible estimate is reached. This account predicts that time pressure and cognitive load would increase anchoring (less adjustment capacity), which is supported.

Confirmatory hypothesis testing (Mussweiler and Strack): When presented with an anchor, people test the hypothesis "is the true value similar to the anchor?" by selectively activating information consistent with the anchor. This anchoring-consistent information then influences the final estimate. This account predicts that anchoring should be reduced when the anchor is so extreme as to be obviously wrong, which is also supported.

Numeric priming: Anchors activate numerically proximate concepts and categories, making anchor-consistent information more accessible for judgment. This is a more purely cognitive account and has less evidence than the above two.

The most plausible current view is that multiple mechanisms are involved. The insufficient adjustment account explains anchoring from starting points in sequential estimation tasks. Confirmatory hypothesis testing explains anchoring in absolute judgment tasks where there is no explicit starting point. Both are probably active in most real-world situations.

The insufficient adjustment account has the clearest practical implications, so it's worth examining closely.

When you adjust from an anchor, adjustment is an effortful, attention-dependent process. It requires generating reasons why the true value differs from the anchor, evaluating how far each reason should push the estimate, and continuing to adjust until the estimate feels appropriate. The "feels appropriate" stopping criterion is the problem: it activates too early.

Epley and Gilovich (2001) showed this by manipulating cognitive resources available for adjustment. Participants under cognitive load showed greater anchoring — they adjusted less from the anchor. Participants instructed to try harder showed less anchoring — they adjusted more. This is direct evidence that adjustment is effortful and terminable — and that the normal stopping point is short of what's warranted.

The practical implication: any cognitive load during estimation (time pressure, multitasking, emotional stress, information overload) will increase the influence of anchors. This is precisely when anchors are most frequently deployed — in negotiations, in high-stakes decisions, under deadline pressure. The conditions that increase anchoring are the same conditions that characterize important decisions.

Salary and Compensation Negotiations

The first-offer advantage in salary negotiations is well-documented. Galinsky and Mussweiler (2001) found that first offers anchored negotiation outcomes, but also found that the magnitude of first-mover advantage depended on who set the anchor and how extreme it was.

Several principles for salary negotiations specifically: - The person who makes the first offer anchors the negotiation. If you're in a strong position, make the first offer. If you're in a weak position, consider eliciting theirs first and then applying an extreme counter-anchor. - Specificity of the anchor matters. Specific numbers ($87,500) are taken more seriously than round numbers ($90,000) because specificity implies knowledge and calculation rather than guessing. - Justification reduces adjustment. An anchor accompanied by a rationale produces more adjustment from the counter-party than a bare anchor. Conversely, if you've received a low anchor, demanding explicit justification for it forces the anchor-setter to engage with its actual basis, which often weakens its influence.

Real Estate

The listing price is the anchor. Buyers adjust downward from the listing price; sellers adjust upward from it. Both adjustments are insufficient. This is why artificially inflated listing prices are a standard real estate strategy — the inflated price shifts the entire negotiation range upward, even when both parties know that the listing price exceeds true market value.

Research by Northcraft and Neale specifically addressed whether expertise eliminates this effect. It doesn't. Experienced real estate agents showed the same anchoring effect as naïve buyers, despite explicitly denying using the listing price as a reference.

The practical corrective: before viewing a listed property, develop an independent estimate of its value using comparables and objective criteria. Commit to this estimate before the listing price is encountered. This generates a competing anchor that reduces (though doesn't eliminate) the listing price's influence.

Legal Settings

Anchoring in legal contexts is especially consequential because the outputs are sentences, damages awards, and verdicts — outcomes with major impacts on real people.

Englich, Mussweiler, and Strack (2006) published a study where experienced judges received a case file and, before rendering sentencing recommendations, either heard a prosecutor's demand for a specific sentence or rolled dice that produced a random number. Even with the randomly generated anchor, sentencing recommendations were significantly influenced.

In civil damages, plaintiff attorneys have known about anchoring for decades. Requesting extremely high damages — even obviously excessive ones — anchors the damages discussion in a range that produces settlements and awards higher than if no anchor had been stated. Defense attorneys have correspondingly learned to generate counter-anchors by explicitly stating low numbers before the discussion proceeds.

Medical Decision-Making

Diagnostic anchoring — or premature closure around an initial diagnosis — has anchoring bias components. Once a provisional diagnosis is made, subsequent information is interpreted through the lens of that diagnosis. Numerical test results are interpreted in the context of previously stated normal/abnormal ranges. Treatment intensity is anchored to the initial assessment of disease severity.

Croskerry's work on cognitive error in emergency medicine identifies anchoring as one of the most common diagnostic errors. The initial impression, the first test result, the triage nurse's assessment — each serves as an anchor that shapes subsequent clinical judgment.

No debiasing strategy eliminates anchoring. Several reduce it:

Self-anchoring before exposure to external anchors. Research by Mussweiler, Strack, and Pfeiffer (2000) found that participants with pre-generated estimates were less influenced by extreme anchors. The mechanism: the self-generated anchor provides a competing reference point that limits how far the external anchor can pull.

Consider-the-opposite strategy. Explicitly generating reasons why the anchor might be too high (or too low) reduces anchoring by activating anchor-inconsistent information, partially countering the confirmatory hypothesis-testing mechanism. This is an active debiasing technique that requires deliberate effort.

Reference class anchoring. Rather than anchoring on a specific number, establish a reference class distribution: "What range of values has characterized similar transactions? What's the distribution?" This replaces the point anchor with a range-based estimate that is harder to distort.

Extreme counter-anchors in negotiations. If an anchor has already been set by the other party, a counter-offer that is extreme — well outside what you expect to accept — shifts the midpoint of the negotiation range. This uses the anchoring mechanism strategically rather than trying to defeat it.

Demand justification. Asking "How did you arrive at that number?" forces the anchor-setter to articulate the basis of their anchor. Unjustified anchors have more influence than justified ones; demanding justification disrupts the anchor's automatic influence.

Anchoring is one of the cleaner demonstrations that human judgment is not a process of reasoning to a conclusion from a blank starting point. It's a process of adjusting from reference points. The reference points are often arbitrary — inherited from whoever spoke first, whatever appeared on the page first, whatever number was salient in the recent environment.

The practical consequence: whoever controls the first number controls a significant part of the subsequent conversation. This is not manipulation in a dark sense — it's a predictable feature of how human judgment works. Understanding it lets you use it deliberately and guard against it being used on you.

The guard is always the same: before encountering any anchor, develop your own estimate. Know your number before you hear theirs.