Mental Models a field guide

Trap/Meta-Concept/No. 0156

Cognitive Bias

A cognitive bias is a systematic tendency in judgment that departs from a specified standard, such as probability rules. In psychology, Amos Tversky and Daniel Kahneman studied these patterns through heuristics: shortcuts whose accuracy depends on the task and setting.

a trap: easy to walk into
Evidence
Well established
Read
6 min
Links
11 connections
Useful when
Deciding under uncertainty · Evaluating a claim · Forecasting · Judging people and hiring · Reading data and statistics

01You've seen this when…

  1. in life

    You budget $500 for a chair. After browsing a showroom full of $1,200 chairs, a $900 model starts to feel economical.

  2. at work

    An applicant gives a polished introduction. During the work-sample review, the panel explains away mistakes it treats as serious in other candidates.

  3. out in the world

    Several vivid assault stories fill your news feed. You become certain that violence is rising in your city, even as the city’s recorded assault rate falls.

02The idea

People can make predictable errors even when they have the relevant facts and want an accurate answer. A starting number pulls an estimate toward it. A favorable first impression colors later evidence. Information supporting an existing belief gets an easier hearing.

Cognitive bias is the umbrella term for systematic tendencies like these. Systematic means a pattern can be detected across repeated judgments or groups. It doesn’t mean every person makes the error on every occasion.

Calling something a bias requires a benchmark. For a probability judgment, that might be a mathematical rule. For a forecast, it might be accuracy across many outcomes. For a hiring decision, it might be consistent treatment of equivalent evidence. The benchmark needs defending: people can reasonably disagree about which qualities a job requires.

Bias also differs from noise in judgment. Repeatedly underestimating delivery time shows a directional bias. Estimates scattered widely around an accurate average show noise. A process can suffer from both, and a single bad estimate gives limited evidence about either.

03Why it happens

Several processes can produce biased judgments:

  • Attention leaves things out. Vivid, recent, or emotionally charged details occupy more of the evidence people consider. Quiet background information can receive less weight than its relevance warrants.
  • Shortcuts reuse available clues. A heuristic is a practical rule that simplifies a judgment. Familiarity, resemblance, and ease of recall can provide useful clues. Their accuracy depends on how closely those clues track the question being answered.
  • Starting points steer estimates. Anchoring bias occurs when an initial value pulls later judgments toward it. Even a number with no useful connection to the task can influence an estimate.
  • Goals steer the search. People often examine agreeable evidence more readily and challenge threatening evidence more closely. This selective search can produce confirmation bias.

These processes overlap. A label identifies an observed pattern; establishing its cause requires further testing. The same judgment can reflect limited attention, prior knowledge, incentives, or several of these together.

04A worked example

In their 1974 Science article, Tversky and Kahneman described an experiment involving a wheel of fortune marked from 0 to 100. The wheel was arranged to stop at either 10 or 65. Participants then judged whether the percentage of African countries among United Nations members was higher or lower than that number, before giving their own estimate.

What it looks like Participants consider a comparison and then answer a factual question. Everyone is estimating the same quantity.

What’s actually going on The median estimates were 25% in the group that saw 10 and 45% in the group that saw 65—a difference of 20 percentage points. The wheel supplied no information about UN membership, yet its result influenced the answers. The relevant benchmark is independence from an irrelevant number; participants could satisfy that benchmark even while getting the exact percentage wrong.

What would have helped Removing the arbitrary starting value would eliminate that particular source of influence. Consulting membership data would make the factual estimate more reliable. Making an independent estimate before seeing someone else’s number is a practical precaution, though it doesn’t guarantee freedom from anchoring.

05How to spot it

Treat these as prompts for investigation. Establishing a bias takes enough comparable judgments to distinguish a pattern from chance or a change in circumstances.

06What to do instead

  • Choose the judgment rule before seeing the tempting details. For hiring, specify the work-sample criteria before reading résumés. For purchases, record the budget and required features before browsing. This makes later departures easier to notice.
  • Bring comparable cases into view. Use reference-class forecasting to check a project estimate against completed projects of similar scope. Record the range as well as the average, so a convenient comparison doesn’t become another misleading anchor.
  • Require a specific challenge. Use the consider-the-opposite strategy: identify evidence that would support a competing explanation, then look for it. Turn this into a cognitive forcing strategy, a required checkpoint such as examining one plausible alternative before approving a diagnosis or proposal.
  • Keep a record and examine the misses. Write down predictions, reasons, and confidence before outcomes arrive. Review comparable decisions together. Look for repeated errors and adjust the procedure that produces them.

Bias training has improved performance on tested tasks in experiments. Its reach into everyday, high-stakes decisions is less certain. Learning a label can help direct attention, but a concrete change to the decision process gives that knowledge a job to do.

07Where it doesn’t imply a bad decision

A shortcut can earn its place. An experienced clinician working with stable, informative cues and reliable feedback may recognize a pattern quickly and accurately. The conditions supporting that expertise matter when judging the quality of intuition.

A benchmark can also demand more time, memory, or calculation than the task allows. Bounded rationality takes those limits seriously. Ecological rationality asks how well a strategy fits its environment; simple rules can perform well under suitable conditions.

A claim of bias therefore needs both a repeatable pattern and a defensible standard. Different priorities can justify different choices under the same facts. A biased answer on one laboratory task also gives limited grounds for judging someone’s competence across their life.

08Roots

In the early 1970s, Amos Tversky and Daniel Kahneman turned statistical questions back on research psychologists. These were people who used samples and significance tests professionally. Yet answers to questions about sample size and replication suggested more faith in small samples than statistical theory justified. Their 1971 paper described this tendency as belief in the law of small numbers.

The puzzle was compelling: statistical training could coexist with unreliable statistical intuition. Working together in Israel, Tversky and Kahneman developed short problems that exposed predictable departures between intuitive answers and explicit standards. A questionnaire or a wheel of fortune could make the departure visible without requiring an elaborate laboratory.

Their 1974 Science article organized examples around judgments by resemblance, ease of recall, and adjustment from a starting value. It brought the heuristics-and-biases program to a broad scientific audience. Later work carried these questions into economics, medicine, law, and organizational decisions, while researchers also challenged whether particular benchmarks fairly represented the environments in which people normally think.

09How solid is this?

ContestedMixedUsefulEstablished

Many specific judgment biases have strong experimental support, including anchoring. The umbrella term has no single effect size: replication, practical consequences, and the appropriateness of the benchmark must be assessed bias by bias.

10Connections

confused withconfused withcountered bycountered bycountered byfollows fromfollows fromincludesincludesincludesCognitive BiasNot written yetNoise inJudgmentHeuristicConsider-the-OppositeStrategyNot written yetReference-ClassForecastingNot written yetCognitiveForcing StrategyBoundedRationalityDual-ProcessTheoryAnchoring BiasConfirmationBiasHalo Effect

+ 1 more in the list

11Origin and sources

Amos Tversky and Daniel Kahneman developed the heuristics-and-biases research program in the early 1970s. Their 1974 Science article brought it to a broad scientific audience.

  1. [1]Tversky, A., & Kahneman, D. (1971). Belief in the law of small numbers. Psychological Bulletin, 76(2), 105–110.
  2. [2]Tversky, A., & Kahneman, D. (1974). Judgment under Uncertainty: Heuristics and Biases. Science, 185(4157), 1124–1131.
  3. [3]Gigerenzer, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: Models of bounded rationality. Psychological Review, 103(4), 650–669.
  4. [4]Kahneman, D., & Klein, G. (2009). Conditions for intuitive expertise: A failure to disagree. American Psychologist, 64(6), 515–526.
  5. [5]Morewedge, C. K., Yoon, H., Scopelliti, I., Symborski, C. W., Korris, J. H., & Kassam, K. S. (2015). Debiasing Decisions: Improved Decision Making With a Single Training Intervention. Policy Insights from the Behavioral and Brain Sciences, 2(1), 129–140.

Suggest an edit· Updated 2026-10-02