Joachim Andersen

Accelerated Learning

Representativeness Heuristic: When Resemblance Feels Like Probability

Five ornate antique brass skeleton keys of nearly identical shape lying in a neat row on a dark walnut desk at night, beside a small engraved brass padlock
The key that looks right is not always the one that fits.


The Representativeness Heuristic is the mental shortcut of judging how likely something is by how much it resembles what I think is typical. If a person, a situation or a pattern looks like my picture of a category, I conclude that it probably belongs to it, often without asking how common that category actually is.

The shortcut is usually helpful. Things that look alike often are alike, and I could not get through a day without making quick judgments based on similarity. But resemblance and probability are not the same thing, and when I confuse them, I ignore base rates, overlook how small my sample is and see patterns in what is really chance.

I notice it most when I meet a story that fits a familiar picture: a confident founder who seems destined to succeed, a symptom that sounds like a rare disease I read about, or a quiet person I assume must be an introverted specialist. The better the fit, the less I think about the numbers.

The essence of the Representativeness Heuristic is to ask: “Setting aside how well this fits the picture, how likely is it really, and how much evidence do I actually have?”


Origin: Judgment by Resemblance

The heuristic was described by the psychologists Daniel Kahneman and Amos Tversky in a 1972 paper on subjective probability. They defined representativeness as the degree to which an event is similar in its essential features to the population it comes from, and reflects the process that produced it.

In their 1974 paper in Science, they presented it as one of three basic heuristics, alongside availability and Anchoring, that people use to judge probabilities under uncertainty. Their point was not that people are foolish, but that sensible shortcuts lead to predictable errors.


Key Concepts

  • Base Rate Neglect
    We focus on how well a description fits and forget how common each possibility is to begin with, as I describe in Probabilistic Thinking.

  • Conjunction Fallacy
    A detailed story that fits our expectations can seem more likely than a simpler one, even though adding details can only make it less likely.

  • Insensitivity to Sample Size
    We expect small samples to look like the whole population, even though small samples vary much more.

  • Misperception of Chance
    We expect random sequences to look random, so streaks and clusters seem meaningful when they are not.

  • Stereotypes
    Our picture of what is typical can be outdated or simply wrong, which makes every judgment based on it wrong too.


Some Examples

  • Linda the Bank Teller
    In a 1983 study, Tversky and Kahneman described Linda as 31 years old, single, outspoken and very bright, a philosophy graduate who was deeply concerned with discrimination and social justice. They then asked whether it was more probable that Linda is a bank teller, or that she is a bank teller who is active in the feminist movement. About 85% chose the second option, even though a feminist bank teller is always also a bank teller, so the second option cannot be more likely.

  • The Engineers and the Lawyers
    In a 1973 study, Kahneman and Tversky told participants that short personality descriptions had been drawn at random from a group of engineers and lawyers. In one version, the group was 70 engineers and 30 lawyers, and in another it was 30 engineers and 70 lawyers. The participants’ estimates were almost the same in both versions, because they judged by how much each description sounded like an engineer. When they were given a description that said nothing useful at all, they typically judged the chance to be 50%, whatever the proportions in the group.

  • The Two Hospitals
    In their 1974 paper, Tversky and Kahneman asked students about two hospitals in a town, one with about 45 births a day and one with about 15. Over a year, which would record more days on which over 60% of the babies were boys? Of 95 students, 21 said the larger hospital, 21 the smaller one and 53 about the same. The correct answer is the smaller hospital, because small samples stray further from the average.

  • The New Colleague
    When a new colleague had two excellent weeks, I caught myself concluding that we had hired a star, because the start looked just like what I imagined a star would look like. Two weeks is a tiny sample, and a few months later the picture was more mixed, as it usually is.


Applying the Representativeness Heuristic

I can’t stop noticing resemblance, and I wouldn’t want to. What I can do is pause before treating a good fit as a high probability. This is the simple process I follow:

  1. Start With the Base Rate
    Before I look at the details, I ask how common each possibility is in general, and only then adjust for how well the case fits.

  2. Count Instead of Picturing
    I imagine 100 cases like this one and ask how many of them would fit each option, which makes conjunction errors much easier to see.

  3. Check the Sample Size
    When a small sample looks striking, I ask how much it could vary by chance alone before I draw any conclusion.

  4. Be Suspicious of Perfect Fits
    The more a description matches a stereotype, the more I ask whether the stereotype is doing the work rather than the evidence.

  5. Look for Disconfirming Cases
    I ask what I would expect to see if my first impression were wrong, which helps against Confirmation Bias.


Pitfalls and Limitations

  • Similarity is often a good guide. Many judgments by resemblance are correct, and the heuristic only becomes a problem when it overrides better information.

  • The way a question is asked matters. When the Linda problem is asked in terms of how many out of 100 people fit each description, far fewer people make the error.

  • Not every streak is an illusion. In 1985, Thomas Gilovich, Robert Vallone and Tversky concluded that the “hot hand” in basketball was a misperception of random sequences, but in 2018 Joshua Miller and Adam Sanjurjo showed that the original analysis contained a subtle statistical bias, and that correcting it reveals evidence of a “hot hand.”

  • Base rates are not always available. Sometimes I don’t know how common something is, and a rough estimate from a reasonable reference class is better than ignoring the question.

  • Knowing about the heuristic does not remove it. The pull of a good fit is felt before any reasoning starts, so I rely on habits such as checking base rates rather than on willpower.


Related Mental Models

The Representativeness Heuristic reminds me that something can look exactly like what I expect and still be unlikely. By starting with base rates, counting instead of picturing and respecting the limits of small samples, I can use the speed of my intuition without letting resemblance stand in for evidence.



To learn more about mental models and philosophy,
check out: Wise: How to Attain a Deep Understanding of the World, History, Our Place in It, and How to Live Well



📚 Sources and References