Ten cognitive biases, told apart

A heuristic is a shortcut for making a judgement quickly. A bias is the systematic error that shortcut produces. The two words are not interchangeable, and swapping them is the single most common mistake in writing about this subject — the availability heuristic is a strategy, and the availability bias is what it costs you.

• Heuristics are usually useful. Tversky and Kahneman said so in the paper that started the field: the shortcuts are economical, and they sometimes lead to severe and systematic errors.
• A bias is systematic, not random. It pushes judgements in a predictable direction, which is what makes it studiable.
• Ten biases are defined below, grouped by what they distort: memory, framing, evidence, or the self.
• Not all of them are equally solid. Anchoring and framing replicated cleanly in a 36-sample project; the Dunning–Kruger interpretation is contested; behavioural priming largely failed.
• Base-rate neglect depends on how the numbers are presented. Give people frequencies instead of percentages and much of the error disappears.

A heuristic is a shortcut. A bias is the error it produces

Amos Tversky and Daniel Kahneman set out the distinction in Science in 1974. People rely on a small number of heuristic principles that reduce complicated judgements — how likely, how frequent, how similar — to simpler operations. Those principles are generally useful, and they sometimes produce errors that are severe and systematic.

That last word is what makes a bias a bias. A random mistake is noise. A bias sends judgements in one direction reliably enough that you can predict which way people will be wrong before you ask them.

So a heuristic is a method and a bias is a result, and the same shortcut can produce both a good answer and a predictable error depending on the situation. Any sentence that treats them as synonyms has lost the distinction the field is built on.

The ten, defined

Biases in what comes to mind

• Availability heuristic. Judging how likely or how common something is by how easily examples come to mind. Named by Tversky and Kahneman in 1973. It is why people overestimate deaths from dramatic causes and underestimate quiet ones.
• Hindsight bias. Once you know the outcome, your estimate of how predictable it was rises. Baruch Fischhoff demonstrated it in 1975 by telling participants how a historical episode ended and asking what probability they would have assigned beforehand. The German name says it exactly: Rückschaufehler, the looking-back error.

Biases in how the question is put

• Anchoring. An initial number pulls a subsequent estimate toward it, even when the number is obviously irrelevant. In the 1974 Science paper, participants watched a wheel of fortune stop on a number and then estimated the percentage of African countries in the United Nations — and the wheel moved the estimates.
• Framing effect. Two logically identical descriptions of the same choice produce different decisions. Tversky and Kahneman’s 1981 paper put an epidemic scenario to participants in terms of lives saved and lives lost; preferences reversed between the two versions.

Biases in weighing evidence

• Confirmation bias. Seeking, noticing and giving weight to evidence that fits what you already believe. Peter Wason’s 1960 task asked participants to work out a rule behind a number sequence, and found that they tested cases they expected to confirm rather than cases that could refute. Raymond Nickerson’s 1998 review remains the standard account of how many different guises it takes.
• Representativeness, and the conjunction fallacy. Judging probability by resemblance to a stereotype rather than by base rates. The 1983 Linda problem is the famous case: told about a woman who is outspoken and concerned with social justice, most people rate “bank teller and active in the feminist movement” as more likely than “bank teller”, which is arithmetically impossible.

Biases about ourselves

• Sunk cost fallacy. Continuing an investment because of what has already been spent, when the remaining decision should turn only on what is left to gain. Arkes and Blumer named and tested it in 1985.
• Fundamental attribution error. Explaining other people’s behaviour by their character while explaining your own by the situation. Jones and Harris showed the core effect in 1967: readers inferred a writer’s real opinion from an essay even after being told the position had been assigned. Lee Ross gave it its name in 1977; many researchers now prefer correspondence bias.
• Dunning–Kruger effect. In 1999 Kruger and Dunning reported that the least competent participants overestimated their performance the most. The finding is widely quoted and the interpretation is disputed, which the section below covers.
• Optimism bias. Expecting your own future to go better than the average person’s. Neil Weinstein measured it in 1980: students rated positive events as more likely to happen to them than to their classmates, and negative events as less likely.

Which of these have survived replication, and which are contested

This is the part most bias listicles leave out, and it is the part that decides whether you should cite an effect in an essay. The honest summary is that the vocabulary is not uniformly solid.

Two entries deserve the detail.

The Dunning–Kruger argument is not settled

Gignac and Zajenkowski argued in 2020 that the classic result is mostly a statistical artefact: plot self-assessed against measured ability properly and the relationship is close to linear, with over-estimation spread fairly evenly rather than concentrated at the bottom. Hiller replied in 2023 that their conclusion depends on a recoding choice, and Gignac and Zajenkowski answered in the same issue. Citing the effect as an established fact is not currently defensible; citing the exchange is.

Priming is two different things

Semantic priming — a word being recognised faster after a related word — is one of the most reliable findings in cognitive psychology. Behavioural or social priming, the claim that subtle cues change complex behaviour, is where the trouble is. Bargh, Chen and Burrows reported in 1996 that people primed with words about old age walked more slowly afterwards; Doyen and colleagues failed to reproduce it in 2012 and found the effect appeared when experimenters knew which condition a participant was in. The Many Labs project tested thirteen effects across 36 samples and 6,344 participants. Ten replicated consistently, one drew only weak support, and the two that failed outright were both priming effects.

The lesson is not that the field is broken. It is that “cognitive bias” names a mixed bag, and the specific effect matters more than the category. Gigerenzer and Hoffrage’s 1995 demonstration makes the point in the other direction: much of what looked like a hard limit on Bayesian reasoning turned out to be a limit of percentages, and switching to natural frequencies made people substantially better at the same problems.

Where this vocabulary sits next to the others

Cognitive biases sit close to two other clusters and are regularly confused with both. Cognitive dissonance is a motivational state, not a bias in judgement, and the two show up together because reducing dissonance often means searching for confirming evidence. Attributional style, the vocabulary that surrounds learned helplessness, describes the habitual way one person explains events, where the fundamental attribution error describes something almost everybody does.

Ten biases is one study list, and the pairs on it — heuristic against bias, availability against representativeness, confirmation against motivated reasoning — are the sort that have to be learned against each other rather than one at a time. Several of the ten are already in the dictionary’s sample terms, including priming, which is on this page for a reason.