Representativeness Heuristic

Representativeness Heuristic is the shortcut of judging likelihood by similarity to a mental prototype. If a story "looks like" a librarian, a fraud, or a winning stock, you treat that fit as evidence it probably is one - even when the base rate says otherwise.

In a now-famous experiment, people were told that personality sketches had been drawn from a group of 70 engineers and 30 lawyers, then asked to guess each person's profession. When they saw a sketch that matched their image of an engineer, they confidently guessed engineer and almost entirely ignored the 70/30 split. When the description was deliberately uninformative, they often defaulted toward 50/50 instead of the base rate. Kahneman and Tversky's 1973 work showed resemblance answering the question the statistics should have.

The mind loves coherent types. A quiet, careful person feels more like an accountant than a salesperson, so you guess accountant and feel confident, even though salespeople may be far more common in the population you are actually drawing from. Vivid detail that matches a stereotype weighs heavily; dull base-rate facts weigh lightly.

The same machinery produces the conjunction fallacy. In the "Linda" study (Tversky and Kahneman, 1983), people read about a bright, outspoken woman deeply concerned with social justice, then judged it more probable that she was "a bank teller and active in the feminist movement" than simply "a bank teller." That is logically impossible - a subset can never be more likely than the whole - yet the richer, more representative story won for the large majority who answered. Adding detail made the claim feel more likely even as it made it mathematically less likely.

Representativeness fuels hiring snap judgments, medical guesses, investing stories, and crime fear. Small samples that "feel like" a trend get treated as representative of the whole market, team, or group, and a three-spin streak at a roulette table starts to feel like it must correct itself.

Prototypes are not useless. When data is thin and the type is grounded in real structure, resemblance can be a reasonable first read, and stereotypes sometimes do correlate weakly with reality. The trouble starts only when similarity quietly replaces the questions that actually govern the odds - how common the category is here, how large the sample is, and what independent evidence says once the story is set aside.

• "She seems like a nurse, so she probably is one."
• "This startup feels like the last unicorn - I'm all in."
• "He doesn't look like an engineer, so he can't be on the team."
• "Three red spins in a row - black is due now."
• "That email reads exactly like a scam, so it must be one."
• "The quiet kid fits the profile I read online, so I'm sure what's going on."
• "This coin feels fair because the last flips looked random."
• "The candidate talks like our best hire, so they'll perform the same."

The typecast hire: a hiring manager meets a confident, fast-talking candidate who matches an internal picture of "what a great salesperson sounds like," and rates them a strong fit within minutes. A quieter applicant with a documented track record of closing larger deals reads as a poor match for the type, so the harder evidence on the page gets skimmed. The prototype did the ranking before the resume did.

The lookalike fund: an investor hears a pitch that rhymes with a legendary past winner - same founder energy, same disruptive framing, same hockey-stick chart - and feels the déjà vu as conviction. Because the story is so representative of success, the base rate that most startups in this category quietly fail never enters the decision, and capital chases the resemblance.

The dramatic diagnosis: a clinician sees a cluster of symptoms that fits a vivid, memorable rare disease from a textbook or a recent case, and locks onto it. A far more common condition with overlapping symptoms is the better bet given prevalence, but it lacks the narrative punch, so the workup tilts toward the exotic answer that "looks like" the story.

The gambler's streak: at a table or in a trading account, a short run of the same outcome starts to feel like a pattern that owes a reversal. A handful of results is treated as representative of the underlying odds, so a player raises the bet because red is "due," confusing what a fair process should look like over thousands of trials with what it happens to do over five.

The viral single case: one clip of someone behaving badly matches an outrage template already circulating, so it gets shared as proof that "everyone" in some group acts this way. No one checks how often the behavior actually occurs, because the clip is so representative of the story people expected that its frequency feels settled.

Representativeness Heuristic produces confident wrong guesses. You bet on stories that feel right and miss categories that are common but visually dull, which means your errors are not random - they lean predictably toward whatever matches a vivid prototype.

Stereotypes harden when representativeness replaces individual evidence. People get sorted by type, and fair assessment of skill, risk, or intent never gets a chance. Because the shortcut feels like insight rather than bias, the judgments arrive with unearned certainty and resist correction.

In markets and projects, narrative fit can crowd out due diligence. Teams chase lookalike wins, treat small samples dressed up as patterns as proof, and repeat gambler-style errors that a glance at frequency would have caught. Capital, attention, and effort flow toward the best-told story rather than the best-supported one.

Public fear and policy skew the same way. Vivid, prototype-matching risks - a shark, a plane crash, a stranger abduction - draw funding and worry, while ordinary hazards that kill far more people stay underweighted because they do not match a dramatic template. The result is a society that is well defended against its scariest images and poorly defended against its actual odds.

Similarity is fast to compute. When uncertainty is high, the brain asks "what is this like?" before it asks "how often does this happen here?" A match to a familiar type produces an immediate feeling of recognition, and that feeling is easy to mistake for evidence.

Stories also beat statistics in memory and attention. One detailed stereotype-match sticks because it comes with faces, scenes, and stakes, while base rates arrive as abstract numbers with no built-in drama. Unless you deliberately fetch the frequency, the vivid case is simply the information you have on hand, so it wins by default.

Kahneman and Tversky introduced representativeness in "On the Psychology of Prediction" (1973), showing across the engineer-lawyer studies that people predict category membership by resemblance to a stereotype and neglect how common that category is in the relevant population. Even explicit, salient base rates were largely discarded once a personality sketch invited a similarity judgment.

Tversky and Kahneman's 1983 paper on extensional versus intuitive reasoning pinned down the conjunction fallacy with the Linda problem: a representative description drove most respondents to rate a conjunction as more probable than one of its own components, a direct violation of probability theory. Together the two findings established that resemblance can override both prior frequency and basic logic, and later work traced the same signature through medical, financial, and legal judgment.

• You feel sure because someone or something "fits the type" before you check any frequency data.
• A single vivid example carries more weight than the size of the group it came from.
• Your probability answer would change if you swapped in a different stereotype with the same facts.
• A short streak feels like a pattern because it resembles momentum or randomness you have seen before.
• You know the base rate in theory but do not use it in this specific guess.
• Individual evidence that breaks the prototype gets dismissed as an exception without numbers.

When a story feels like a perfect fit, treat that feeling as a prompt to look up the odds, not as the answer. Ask how common the category actually is in this pool before you decide anything.

• State the base rate out loud before you look at vivid details.
• Separate "similar to X" from "likely to be X" in writing.
• Require independent evidence beyond the stereotype for high-stakes calls.
• Blind or structure reviews so prototypes cannot steer early cuts.
• Count samples instead of narrating streaks when randomness is possible.
• When stuck, run the Base-Rate Neglect checks: priors first, story second.

If Representativeness Heuristic is loud, treat the strong "type match" as your cue to go find the base rate, then let the numbers, not the resemblance, set your confidence.

Meeting a stranger

Hot investment pitch

Small sample streak

Practice this pattern in the Reframing App - capture the trigger, label it (like Representativeness Heuristic), check evidence, and write a more balanced thought.