Heuristics in Judgment and Decision Making

Core Idea

Heuristics are mental shortcuts that reduce the cognitive effort required for judgment and decision making. The three canonical heuristics identified by Kahneman and Tversky — availability (frequency judgments based on ease of retrieval), representativeness (probability judgments based on typicality to a category), and anchoring-and-adjustment (estimates insufficiently adjusted from an initial anchor) — explain a wide range of systematic judgment errors. Gigerenzer's ecological rationality framework argues that many of these fast-and-frugal heuristics perform well in natural environments and should not be uniformly condemned as biases.

How It's Best Learned

Use the availability heuristic: ask participants to estimate causes of death and compare their estimates to actuarial data. Dramatic or media-covered events are overestimated because they are highly available in memory. Then identify domains where availability is genuinely diagnostic to show heuristics can be accurate.

Common Misconceptions

• Heuristics are not simply bad reasoning — fast-and-frugal strategies often match or exceed the accuracy of complex statistical models in real-world data environments.
• Anchoring effects persist even when subjects know the anchor is arbitrary and random, showing that cognitive awareness does not neutralize the bias.

Explainer

From your prerequisite study of cognitive biases, you know that human judgment departs systematically from the predictions of classical rational-agent models. Heuristics are the cognitive mechanisms that produce these departures — mental shortcuts that simplify complex judgments by substituting a hard question with an easier one. The question "How likely is it that this event will occur?" gets replaced with "How easily can I recall examples of this event occurring?" Same surface question, completely different computational process, often very different answer.

The availability heuristic exploits the fact that memory retrieval ease correlates with actual frequency in many real-world environments. If plane crashes are frequent, you'd expect to be able to retrieve many examples quickly. The problem is that availability is also driven by vividness, emotional salience, and media coverage — factors uncorrelated with actual frequency. Dramatic events are over-retrieved; mundane common events are under-retrieved. This produces the classic finding: people dramatically overestimate death rates from shark attacks, plane crashes, and homicide (vivid, covered in media) while underestimating death rates from stroke, diabetes, and suicide (common but unglamorous). The heuristic is fast and often accurate, but systematically biased when retrievability is decoupled from frequency.

The representativeness heuristic substitutes a different question: instead of "What is the probability of this?" it asks "How much does this resemble the prototype of category X?" This produces base rate neglect: if someone is described as quiet, thoughtful, and precise, people rate them as more likely to be a librarian than a salesperson — even when told there are 100 salespeople for every librarian in the sample. The description matches the librarian prototype, and prototype matching overrides base rate information. Representativeness also produces the conjunction fallacy: the famous "Linda problem" where people rate "Linda is a bank teller and a feminist" as more probable than "Linda is a bank teller" — which is logically impossible, because every conjunction is at most as probable as either conjunct. The rich description made the conjunction more representative, even though it made it less probable.

Anchoring is mechanistically different: a starting value — even a random or arbitrary one — exerts disproportionate influence on subsequent estimates, because adjustment from an anchor is typically insufficient. In experiments, subjects who first see a high random number on a spinning wheel give higher estimates of "percent of African nations in the UN" than those who see a low number. The anchor contaminates the estimate even when it is demonstrably uninformative. This matters enormously in negotiation, medical diagnosis, and legal sentencing — first offers, initial diagnoses, and recommended sentences all function as anchors that constrain subsequent reasoning. Gigerenzer's countervailing argument is that in natural environments with ecologically valid cues, these heuristics often outperform complex statistical models — not because they're lucky, but because they exploit genuine regularities in structured environments. The lesson is not that heuristics are always wrong, but that they are tuned to particular environmental conditions and fail predictably when those conditions are violated.