A buyer sees a price of 8,000 before estimating what a used bicycle might be worth. The number may be implausibly high, yet the estimate that follows can move toward it. Anchoring is the influence of an initial value, comparison point, or reference cue on a later judgment. The effect appears in many estimation tasks, but it has several proposed mechanisms and boundaries. It does not mean that every number people encounter controls them, or that an informed decision can be reduced to removing one tempting figure.
Starting points in uncertain judgment
Tversky and Kahneman introduced anchoring and adjustment in their 1974 account of heuristics under uncertainty. In classic demonstrations, people made numerical estimates after considering a number that should have had little or no bearing on the answer. Estimates shifted with that starting point. The observation matters because a judgment can seem personally generated while retaining a trace of a preceding cue. The paper does not establish that all anchoring arises from one mental operation or that all professional estimates are equally affected.
An anchor can be supplied by someone else or generated by the judge. A seller’s asking price is externally provided. A person estimating the year of an event may start from a date they know and adjust. These situations differ. A known reference can contain valid information, while an arbitrary cue should not. The practical question is not whether a starting point exists, but whether the final estimate gives it appropriate weight given its relevance and the available evidence.
The word anchoring is sometimes used too broadly. A prior price may be a rational clue to market value if it reflects recent transactions; it may also be a strategic opening bid. A budget ceiling can be an actual constraint rather than a cognitive bias. To identify an anchoring effect experimentally, researchers vary the initial value while holding other relevant information as constant as possible. A changed estimate then supports a causal claim about the cue in that task.
Even when the anchor is arbitrary, the judge may not simply copy it. The number can alter what examples come to mind, the range considered plausible, or the point from which an adjustment begins. Different experiments favor different explanations. Treating the effect as a single mental shortcut hides this variation. A more precise account asks who provided the anchor, whether the person compared the target with it, and what knowledge they had to correct or reject it.
Adjustment from a familiar reference
Epley and Gilovich investigated judgments that begin with self-generated anchors. A person estimating the freezing point of an unfamiliar liquid might start from the familiar freezing point of water and move outward. Their studies showed that adjustment from a self-generated starting point can be insufficient. The final answer may stop once it enters a plausible range rather than reaching the best-supported estimate. This account is useful for some tasks, but the authors also distinguished it from processes following experimenter-provided anchors.
In later work, Epley and Gilovich tested why adjustment can stop early. Their evidence suggested that people may end the search when an answer feels plausible unless they can and will seek a more accurate one. That is a bounded explanation, not a promise that effort always removes anchoring. A person may lack the relevant data, misunderstand the direction of adjustment, or face an anchor embedded in a social negotiation. The mechanism should be tied to the task that tested it.
Suppose an engineer estimates the duration of a repair from a previous job. Starting from that known duration is sensible, but the new task may involve different access, parts, and safety checks. A disciplined estimate identifies which features differ and adjusts each with evidence. Merely telling the engineer to “think harder” is less useful than providing comparable cases and a way to inspect assumptions. The practical lesson is about calibrating a reference, not trying to think without any reference at all.
Adjustment can also be asymmetric. People may know that an initial value is too high without knowing how much lower to go. If the valid range is wide, they can move away from the anchor and still remain influenced by it. A better estimate may require external data or a second independent method. This is why a warning that an anchor is arbitrary often does not fully solve the problem. Recognition is useful, but numerical correction needs information.
Provided anchors and accessible information
Mussweiler and Strack examined a selective-accessibility account. When people answer a comparative question, such as whether a quantity is above a proposed number, they may search for information consistent with that hypothesis. Related information then becomes more available when they give an absolute estimate. Their experiments support this mechanism in defined comparative judgment tasks. The result does not exclude adjustment in other settings, and it should not be summarized as the sole explanation of all anchors.
An asking price may work partly by making certain property features salient. A high asking price can invite a buyer to ask what would justify such a value, bringing attractive features to mind. A low anchor can prompt a different comparison. The exact effect depends on whether the number is credible, whether the person is motivated to scrutinize it, and what independent information exists. The buyer may also infer something about the seller’s knowledge or bargaining position, which is not the same as a pure cognitive accessibility effect.
Simmons, LeBoeuf, and Nelson challenged a sharp divide between self-generated anchors that invite adjustment and provided anchors that do not. Their experiments examined accuracy motivation and showed that adjustment from provided anchors can become visible when people know which direction to move. This finding matters because two proposed mechanisms may both contribute, depending on conditions. It does not imply that incentives alone eliminate bias. If a person does not know the correct direction or lacks evidence, motivation may have little leverage.
Expertise can help, but it is not an absolute shield. An expert may possess better reference data and recognize an absurd number, yet a professional setting can still present persuasive cues. Conversely, a beginner with access to clear market comparisons may produce a reasonable estimate. The relevant variables are knowledge, feedback, incentives, task structure, and the authority granted to the anchor. Claims that experts are immune or that everyone is equally susceptible overstate the evidence.
Anchors in public and digital settings
Prices, suggested donations, salary ranges, and default quantities can all serve as starting points. Some are informative; some are strategic. An online store that displays a crossed-out “original” price may change how a discount feels. The consumer needs to know whether that original price was actually charged and how comparable products are priced. The classic experiments justify attention to the cue, but they do not prove a particular retailer’s display caused a particular purchase.
In negotiations, the first number may define a bargaining range. But negotiation is more than individual estimation: parties have interests, alternatives, power, and information. A high opening offer can be a signal, an attempted anchor, or both. Treating all outcomes as anchoring can obscure legitimate differences in reservation prices and the consequences of refusing a deal. A fair analysis includes actual market evidence and the positions of both parties.
Decision-support systems can introduce anchors through their first recommendation. If an AI tool proposes a cost estimate, a human reviewer may begin revisions from that value. The proposal may contain useful information or an error. A good workflow records the tool’s assumptions and asks for an independent estimate where feasible. That is an Institute design inference from anchoring research, not a measured guarantee that one interface change prevents automation bias.
Anchors can appear in social judgments too, but numerical evidence should not be stretched carelessly. Hearing one person’s salary may influence a pay comparison; seeing a ranking may shape impressions. Yet claims about moral or social perception need their own studies, because the meaning and validity of the cue differ from a numeric estimation puzzle. The anchor concept is most precise when the starting value and the judged quantity are specified.
Checking an estimate
A practical audit begins by identifying the first number encountered. Where did it come from, and how trustworthy is that source? Then gather reference information independently: comparable cases, known constraints, and a plausible range. If possible, ask another person for an estimate before showing them the anchor. Compare methods, not only final answers. A disagreement can reveal assumptions that were hidden when everyone began from the same suggested value.
Considering an alternative starting point can expose sensitivity. What estimate would seem plausible if the initial value had been much lower or higher? This question does not magically produce truth, but it can show whether the original number is doing unearned work. The next step is to replace arbitrary adjustment with evidence. In a purchase, that may mean recent comparable prices; in a project, historical durations and known dependencies; in medicine, validated ranges and patient-specific facts.
Research on debiasing should be read with the same caution as research on bias. A technique that changes a laboratory estimate immediately may not persist in complex environments. Warnings, training, and incentives may help under some conditions and fail under others. The specific outcome should be measured. A public article should not promise that knowing the word anchoring will make a reader immune. Awareness is a starting point for better information gathering, not a substitute for it.
It is also possible to overcorrect. A person may reject a relevant price simply because it arrived first. If the cue comes from a reliable measurement, discarding it wastes information. The goal is proportional weight. Ask how the value was generated, how uncertain it is, and whether independent evidence points the same way. An anchor becomes a bias when its influence exceeds what the evidence warrants, not merely because the number appeared early.
Anchoring can be especially hard to isolate when a number carries social meaning. A seller’s price is not just a random digit; it may imply confidence, quality, or a willingness to negotiate. A judge’s previous sentence or a clinician’s first estimate may reflect case information that others also saw. Researchers seeking a causal estimate must separate the cue’s informational content from its position and magnitude. Randomized anchors are powerful precisely because they make the cue independent of the target, although that experimental purity can limit direct generalization to ordinary markets.
The ethical issue is not restricted to deliberately inflated numbers. A well-meaning organization can establish a low anchor by publishing an outdated benchmark, causing later estimates to appear excessive. A school can prime expectations with an early score that reflects a temporary difficulty. People affected by such labels may have little power to choose another reference. Responsible systems therefore document where starting values came from, how they are revised, and how a person can contest a misleading first estimate. This is a practical implication, not a single empirically validated correction method.
Another safeguard is to preserve uncertainty. A point estimate such as “the repair will take three days” may imply precision that the evidence lacks. A range with explicit assumptions can show which parts are known and what could move the answer. Ranges can themselves anchor judgment, so they should be grounded in comparable data rather than selected for persuasion. If the evidence is weak, saying so may be more useful than offering a confident number that everyone subsequently adjusts around.
What the evidence does not show
The evidence does not show that every numerical reference is irrelevant, that anchoring has one universal mechanism, or that expertise and motivation never matter. The original laboratory demonstrations cannot establish the size of an effect in every market, workplace, or clinical setting. Studies of self-generated and provided anchors reveal different processes that may overlap. Recognizing a cue does not supply the correct estimate, and a high or low outcome is not by itself proof that anchoring caused it.
Anchoring is a reminder that estimates are made from somewhere. A starting point can be useful when it is relevant and misleading when it is arbitrary or strategic. The buyer considering the bicycle should inspect comparable sales, condition, and repair costs before allowing a displayed price to define value. The aim is not to erase all reference points; it is to choose and weight them with evidence that can be explained to another person.
Research references
- Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. View source
- Epley, N., & Gilovich, T. (2001). Putting adjustment back in the anchoring and adjustment heuristic: Differential processing of self-generated and experimenter-provided anchors. Psychological Science, 12(5), 391–396. View source
- Epley, N., & Gilovich, T. (2006). The anchoring-and-adjustment heuristic: Why the adjustments are insufficient. Psychological Science, 17(4), 311–318. View source
- Mussweiler, T., & Strack, F. (1999). Hypothesis-consistent testing and semantic priming in the anchoring paradigm: A selective accessibility model. Journal of Experimental Social Psychology, 35(2), 136–164. View source
- Simmons, J. P., LeBoeuf, R. A., & Nelson, L. D. (2010). The effect of accuracy motivation on anchoring and adjustment: Do people adjust from provided anchors? Journal of Personality and Social Psychology, 99(6), 917–932. View source