Where does representativeness heuristic collide?

How people judge the chance that something belongs to a category, or was produced by a process, by how closely it resembles a typical case rather than by its actual frequency. Tversky and Kahneman set this out in their 1974 paper Judgment under Uncertainty as one of three shortcuts people fall back on when they are uncertain, and argued that leaning on resemblance leads them to ignore base rates, sample size, and regression toward the average. In their 1983 work the same shortcut produced the conjunction fallacy, where the Linda problem led most readers to rate a combination of traits as more likely than one of its parts alone. The referent here is the general rule of judging by likeness, the parent claim that the other panels either extend or dispute. It is a claim about a mental shortcut, not one tidy error.[1][2]

Unit

A person's judged probability that a case fits a category

Recovery means

Not applicable

Measured by

Elicited by probability or ranking tasks, such as category-membership and Linda-style problems

The same shortcut is called on to explain a whole family of separate mistakes: neglecting base rates when a description sounds typical, the conjunction fallacy, the gambler's fallacy that a run of one outcome makes the other due, and predictions that fail to regress toward the average. Each of these was traced back to judging by resemblance, but they need not share a single mechanism, and the heuristic itself is not stated precisely enough to say in advance which cases it covers and which it does not. This is the heart of the jangle on this page: one under-specified label doing the work of several distinct effects. When a broad name explains everything, it becomes hard to test, because almost any error can be retold as a story about likeness after the fact.[1][2]

Unit

One label applied across several judgment errors

Recovery means

Not applicable

Measured by

Inferred from patterns across base-rate, conjunction, gambler's-fallacy, and prediction tasks

Gigerenzer's reading is that representativeness is too loosely defined to count as an explanation, because it names a resemblance without saying how similarity is computed or when it will mislead. In his 1996 reply he argued that several of the classic errors soften or disappear when the same problem is posed in natural frequencies rather than single-event probabilities, which points to the wording of the question rather than a faulty shortcut. This is the ecological rationality line, that judgments should be weighed against the environments they are used in, not against one narrow probability norm. On this view the label mostly describes the answer we want to explain, not the process that produced it. The dispute is unsettled, and honest use of the term should say plainly that it is contested.[3]

Unit

The same tasks re-posed as natural frequencies

Recovery means

Not applicable

Measured by

Compared across probability and frequency formats to see which errors persist

Kahneman and Frederick later recast representativeness as a special case of a more general move: when a hard question, how probable is this, is met, the mind quietly answers an easier one, how similar is this to a prototype, and treats the second answer as the first. Their 2002 chapter Representativeness Revisited folded the heuristic into this attribute-substitution account and tied it to the fast, intuitive side of a dual-process picture of the mind. Kahneman's 2003 lecture carried the same framing into economics. This is a reinterpretation rather than a fresh finding: the original demonstrations still stand, but the name for what causes them has shifted. It leaves open whether substitution is a real mechanism or one more redescription.[4][5]

Unit

A hard probability question answered by an easy similarity one

Recovery means

Not applicable

Measured by

Tested by whether similarity ratings predict the probability judgments they stand in for