The Representativeness Heuristic

Opening Hook

Picture two candidates for a senior financial management role. The hiring panel has spoken to both.

The first has an unremarkable track record but presents impeccably. She speaks with measured authority, wears the right suit, went to the right school, and carries herself in a way that the panel privately agrees feels like a senior executive. She does not have exceptional results, but nothing about her raises flags.

The second has outperformed every benchmark set for him over nine years in progressively senior roles, has a string of measurable achievements, and received the strongest written references the panel has seen. He is also visibly nervous, speaks quickly, and does not particularly look like what anyone in the room would draw if asked to sketch a successful financial executive.

You know what happens next. The first candidate gets the job.

This is not a hypothetical. It is a description of a cognitive process that plays out thousands of times a day in interview rooms, investment committees, and consulting engagements. The panel is not making a probability assessment about which candidate is more likely to perform well. They are making a similarity assessment: which candidate most resembles their mental prototype of a successful senior executive? And they are reporting the result of that similarity check as if it were a probability judgement.

That substitution, swapping probability for similarity, is the representativeness heuristic. And once you know it is there, you will see it everywhere.

The Concept

The representativeness heuristic is the tendency to judge the probability of something by how much it resembles a typical member of a category, rather than by the actual statistical evidence. When people are asked “how likely is X to belong to category Y?”, they frequently answer a different question: “how much does X look like a typical Y?”

Amos Tversky and Daniel Kahneman identified this as one of the central shortcuts in human probabilistic reasoning, first described in their 1972 paper on subjective probability and developed fully in their landmark 1974 paper in Science, “Judgment under Uncertainty: Heuristics and Biases.”

The heuristic is not always wrong. In many everyday settings, the thing that most resembles a category really is the most common member of it. If something quacks and swims on water and has feathers, it probably is a duck, and the representativeness shortcut gets you there faster than a formal probability calculation would. The trouble begins in three specific places.

The conjunction fallacy. In Unit 4.3, you met Linda: a former philosophy student, politically active, concerned with social justice. When asked whether Linda is more likely to be a bank teller, or a bank teller who is active in the feminist movement, 85 percent of people choose the second option. This is a formal impossibility. The probability of two things being simultaneously true cannot exceed the probability of either one alone. But Linda’s description matches the mental prototype of a feminist activist, and it matches the prototype of a bank teller rather badly. When representativeness is doing the work, the more specific description wins the similarity contest and gets reported as the more probable outcome. Probability has been quietly replaced by narrative fit.

Base rate neglect. In Unit 4.2, you saw how people fail to use background frequency when assessing a specific case. The witness says it was a blue cab. Most people conclude it probably was blue. But if 85 percent of cabs in the city are green, and the witness is only 80 percent reliable, the mathematics says the cab was more likely green than blue. The witness testimony is specific and vivid; the city-wide cab distribution is abstract and statistical. The specific feels like the real evidence. The base rate slides past without registering.

Both errors share the same engine. The mind generates a similarity judgement, attaches it to the specific case, and discards the prior probability. Representativeness is the mechanism that drives base rate neglect: the specific description is representative of something, and that something is what feels probable, regardless of how common it actually is.

The small sample fallacy, which Tversky and Kahneman called “belief in the law of small numbers,” is the third form of the same error. The idea is that people expect small samples to be just as representative of their parent population as large ones. A coin flipped five times should come up heads five times? That feels wrong. But consider how often people draw confident conclusions from small samples in daily life, treating a run of results as if it revealed the underlying truth. Three good months, and the fund manager is a genius. Four cases in a row, and the junior doctor has found the pattern.

The statistical reality is that small samples vary enormously. A hospital delivering fifteen babies in a day will have many more days on which 60 percent of births are boys than a hospital delivering 1,500 babies, simply because of chance variation. The large hospital’s numbers will cluster close to the true population proportion because large samples are reliably representative. The small hospital’s numbers bounce around widely. If you expect a small sample to behave like a large one, you will see patterns in noise, read streaks as signals, and attribute to cause what is merely variation.

Together, these three failures form a single family. They all arise from treating similarity as a proxy for probability, from the expectation that the world will look like its prototypes at every scale.

Why It Matters

In hiring and leadership selection, the consequences are well-documented. Researchers have repeatedly shown that people judge leadership potential substantially by how closely a candidate matches a culturally dominant leadership prototype, rather than by objective performance evidence. Alexander Todorov and colleagues at Princeton demonstrated in 2005 that snap judgements of facial competence, made after seeing a photograph for one second, predicted the outcomes of US congressional elections at rates better than chance. People were not assessing policy platforms; they were running a representativeness check: does this face match my prototype of a competent leader?

In a job interview, the same process operates with more information available, but the underlying heuristic often dominates anyway. The candidate who “looks right” benefits from a tailwind that has nothing to do with their actual probability of performing well.

In investment analysis, the heuristic leads to what finance researchers call the glamour stock problem. Companies with impressive recent histories and compelling narratives attract investment at high valuations. They feel like successful companies, and successful companies are expected to generate high returns. But research by Josef Lakonishok, Andrei Shleifer, and Robert Vishny, published in the Journal of Finance in 1994, showed that so-called value stocks, those that look like they have failed or stagnated, consistently outperformed glamour stocks over long time horizons. Investors had priced glamour stocks as if recent strong performance were representative of future performance, while ignoring the base rate tendency of high-growth periods to mean-revert. The company’s story matched the prototype of a winner. The probability of continued outperformance did not.

In medical diagnosis, the pattern-matching speed of the representativeness heuristic is genuinely valuable most of the time, and experienced clinicians rely on it deliberately for common, clearly presented conditions. The danger arrives with atypical presentations. Research on myocardial infarction (heart attack) diagnosis has found that women show significantly more variation in symptom patterns than men, and that clinicians sometimes fail to identify MI in female patients because the presentation does not fit the prototypical pattern of chest pain, radiating arm pain, and sweating, which was established primarily from studies conducted on male patients. The symptom cluster does not match the mental prototype. The probability judgement follows the mismatch, and diagnoses are delayed.

How to Spot It

The documented case is from Tversky and Kahneman’s original 1974 research. Participants were presented with a description of a man described as conservative, careful, and meticulous, with a passion for detail and little interest in people or politics. They were told the description came from a pool that contained a mix of engineers and lawyers, and were asked to estimate the probability that the man was an engineer.

In one version of the experiment, participants were told the pool was 70 percent engineers and 30 percent lawyers. In another, the proportions were reversed. Participants’ probability estimates barely shifted between the two conditions. The description was so clearly representative of their mental prototype of an engineer that the actual composition of the pool, the base rate, was almost completely ignored. This is the signature. When the similarity match is strong enough, the background frequency becomes invisible.

The tell is a probability judgement that moves in the wrong direction, or not at all, when base rate information changes. In the interview room, it sounds like: “She just has the right presence.” In investment analysis: “This is exactly the kind of company that becomes a ten-bagger.” In medicine: “The presentation is classic.” In each case, the speaker has a clear mental image of what the category looks like. They are reporting how well the case matches that image. They are calling it a probability assessment.

The corrective is to force the prior back into the picture. Ask what the base rate is. Ask how many people who “look like” this candidate actually perform in the top quartile. Ask how many companies with this narrative profile outperform over a five-year period. Ask how often this pattern of symptoms turns out to be the serious condition rather than a more common mimic.

Your Challenge

A large hospital delivers approximately 1,500 babies per month. A small rural hospital in the same country delivers approximately 15 babies per month. In a given year, each hospital records the number of days on which more than 60 percent of the births were boys.

Which hospital would you expect to record more such days?

Most people say the hospitals would record roughly the same number, or that it is impossible to know. Some say the large hospital, reasoning that larger numbers give more reliable results.

Work through the logic of sample size and chance variation and see whether your intuition was right.

There is no answer on this page.

References

Tversky, A. and Kahneman, D., “Judgment under uncertainty: heuristics and biases,” Science, 185(4157), 1124–1131 (1974). The foundational paper cataloguing the representativeness, availability, and anchoring heuristics and the biases they produce. URL: https://www.science.org/doi/10.1126/science.185.4157.1124

Kahneman, D. and Tversky, A., “Subjective probability: A judgment of representativeness,” Cognitive Psychology, 3(3), 430–454 (1972). The original paper introducing the representativeness heuristic. Available via ScienceDirect: https://www.sciencedirect.com/science/article/abs/pii/0010028572900163

Tversky, A. and Kahneman, D., “Belief in the law of small numbers,” Psychological Bulletin, 76(2), 105–110 (1971). The paper defining the small sample fallacy and demonstrating that people expect small samples to be as representative as large ones. URL: http://www.stats.org.uk/statistical-inference/TverskyKahneman1971.pdf

Kahneman, D., Thinking, Fast and Slow, Farrar, Straus and Giroux (2011). Chapters 14–16 cover representativeness, base rate neglect, and the law of small numbers in accessible detail, with the engineer-lawyer experiment and the hospital baby problem discussed directly.

The hospital baby problem (boys born in large vs. small hospitals) is from the original Kahneman and Tversky (1972) paper on representativeness cited above. The classic formulation is also documented in Kahneman (2011), Chapter 10.

Todorov, A., Mandisodza, A.N., Goren, A., and Hall, C.C., “Inferences of competence from faces predict election outcomes,” Science, 308(5728), 1623–1626 (2005). The Princeton study showing that snap judgements of facial competence predicted US congressional election outcomes at better-than-chance rates. URL: https://www.science.org/doi/10.1126/science.1110589

Lakonishok, J., Shleifer, A., and Vishny, R.W., “Contrarian investment, extrapolation, and risk,” Journal of Finance, 49(5), 1541–1578 (1994). The study demonstrating that glamour stocks, those with impressive recent histories that match investors’ prototypes of winning companies, systematically underperform value stocks over long holding periods. URL: https://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.1994.tb04772.x

Fitzgerald Health Education Associates, “Cognitive errors in clinical diagnosis: Representativeness.” An accessible clinical overview of how the representativeness heuristic produces diagnostic errors, including atypical MI presentation in women. URL: https://www.fhea.com/resource-center/cognitive-errors-in-clinical-diagnosis-representativeness/

Wikipedia overview of the representativeness heuristic with further citations: https://en.wikipedia.org/wiki/Representativeness_heuristic