Judgment bias under uncertainty
Anchoring Effect
Numerical estimates are often pulled toward an initial reference value, even when the anchor is arbitrary or only weakly informative.
estimate = reference + alpha(anchor - reference)
This is an illustrative assimilation model, not a universal law. Anchor influence alpha varies with knowledge, task, incentives, anchor plausibility, elicitation, and whether people generate or receive the starting value.
The reference estimate is fixed at 50 and alpha = 0.35, so the displayed group mean moves partway toward the anchor. This demonstrates assimilation; it is not a calibrated prediction for an individual.
(points)
- CHANGE
- Presented anchor value
- WATCH
- estimate distribution
- MEANING
- The reference estimate is fixed at 50 and alpha = 0.35, so the displayed group mean moves partway toward the anchor. This demonstrates assimilation; it is not a calibrated prediction for an individual.
The anchor moves the center of judgment.
Two otherwise identical estimate distributions begin from different reference numbers. Their overlap shows that anchoring shifts a population tendency rather than determining every response.
What it actually says
Anchoring describes the influence of an initial value on a later numerical judgment. In the classic demonstration, participants first answered whether a quantity was above or below an arbitrary number and then estimated the quantity; estimates remained systematically closer to that number.
Insufficient adjustment is one explanation, especially for self-generated anchors, but not the only one. Selective accessibility, conversational inference, scale interpretation, numeric priming, and task design can contribute. A professional account separates the observed shift from any single proposed mechanism.
"A useful law compresses a pattern. It does not erase the conditions that make the pattern true."
How the idea developed
The modern form emerged through observation, argument, and later refinement. The timeline separates the first insight from the version now used in textbooks and practice.[1]
Psychophysical research uses adaptation-level ideas to study judgments relative to contextual reference points.
Tversky and Kahneman present anchoring and adjustment as a heuristic in judgment under uncertainty.
Research distinguishes self-generated anchors, experimenter-provided anchors, and selective accessibility.
Anchoring is studied in negotiation, valuation, forecasting, legal judgment, medicine, and replication programs.
How the pattern works
The relation becomes useful only when its mechanism, measurement process, and operating range are visible.
People may begin at an accessible value and stop adjustment when an answer enters a plausible range.
Testing whether the true value is near the anchor activates anchor-consistent knowledge.
A provided number may be treated as informative because participants assume it was chosen for a reason.
An anchor can change how the response scale and plausible range are mentally represented.
This is an illustrative assimilation model, not a universal law. Anchor influence alpha varies with knowledge, task, incentives, anchor plausibility, elicitation, and whether people generate or receive the starting value.
Where it earns its keep
Applications are strongest when the law changes a decision, measurement, model, or experiment rather than merely providing an analogy.
Control the first defensible proposal
ApplicationOpening offers can frame the bargaining range when uncertainty is high.
An extreme or unsupported anchor can damage credibility and cooperation.
Separate independent estimates
ApplicationForecasters should record private estimates before seeing a group consensus or prior forecast.
Independence reduces shared anchoring but does not remove common data errors.
Use multiple reference models
ApplicationComparable prices, list prices, and previous valuations can anchor professional estimates.
Triangulate with base rates, cost, cash flow, and blind review where feasible.
Where it stops working
Effect size is not constant. Expertise can reduce some anchors but does not guarantee immunity; relevant anchors can rationally carry information, while implausible anchors may be discounted or trigger contrast rather than assimilation.
Laboratory anchoring effects do not automatically imply large real-world harm. Stakes, feedback, repeated markets, incentives, information search, accountability, and measurement choices can change both mechanism and magnitude.
"The first number always wins"
Better: Anchors shift distributions probabilistically; they do not determine every response."Any correlation with a prior value proves anchoring"
Better: Rational use of relevant information and regression can create similar patterns."Experts are completely immune"
Better: Knowledge helps, but effects can persist in specialized judgments."Removing numbers removes bias"
Better: People generate anchors from memory, goals, round numbers, and prior experience.Sources and further reading
Original publications and serious secondary scholarship are prioritized over summaries.
- Tversky and Kahneman - Judgment under Uncertainty: Heuristics and BiasesThe 1974 Science article introducing the classic anchoring demonstration.https://pubmed.ncbi.nlm.nih.gov/17835457/
- Epley and Gilovich - The Anchoring-and-Adjustment HeuristicExperimental work on insufficient adjustment and self-generated anchors.https://doi.org/10.1111/j.1467-9280.2006.01704.x
- Furnham and Boo - A Literature Review of the Anchoring EffectReview of anchoring mechanisms, moderators, and applied domains.https://doi.org/10.1080/13546783.2011.582032
- Klein et al. - Investigating Variation in ReplicabilityMulti-lab replication project including anchoring and other classic effects.https://doi.org/10.1177/1745691612460688