Representativeness & Base-Rate Neglect

We often judge how likely something is by how closely it matches a familiar example, ignoring the actual odds. In the cab problem, most people focus on the witness accuracy (80%) but overlook that green cabs are far more common (85% vs 15%). The real chance is only about 41%, not 80%. This pattern appears everywhere: doctors overestimate cancer risk from positive tests, investors chase 'hot streaks' that look like past winners, and people assume someone quiet must be a librarian not a banker despite more bankers existing. Check this: When you hear a specific story or vivid detail, ask yourself what the underlying statistics say first.

How representativeness overrides base rates

Our brains love patterns. When something matches a familiar prototype, we feel it's more likely than statistics would suggest. This mental shortcut—called the representativeness heuristic—works well in many everyday situations, but it fails systematically when base rates conflict with stereotypes. In classic studies, people consistently ignored population statistics when given a personality sketch that seemed representative of one group, even when that group was numerically rare.

Medical decisions: false positives that feel real

Doctors and patients routinely overestimate disease risk from positive test results. Consider a mammogram. If breast cancer affects 0.8% of women and a test has 90% sensitivity and 7% false positive rate, then only a small fraction of positive results indicate actual cancer. Studies have found that doctors frequently overestimate this probability. Similar errors occur with HIV tests, drug screening, and forensic evidence.

Beyond the lab: everyday pattern-matching errors

At work, representativeness makes us expect job candidates to fit familiar archetypes rather than considering what's statistically common. Investors chase funds with recent performance that matches past winners, neglecting that most hot streaks are random variation. In legal settings, jurors give too much weight to eyewitness confidence without considering how often eyewitnesses are wrong in similar situations. Political judgments suffer too: we assume a leader's personality explains events when structural factors matter more.

When does this bias strengthen or weaken?

Base-rate neglect is strongest when individuating information feels vivid and specific. Abstract statistics fade next to concrete stories. The bias weakens when base rates are presented as frequencies rather than percentages (10 out of 1000 rather than 1%), when people learn from direct experience rather than verbal description, and when decision stakes are high enough to motivate careful reasoning. Individual differences matter too: people with stronger statistical training show less neglect, though even experts can fall prey under time pressure.

Research on representativeness and base-rate neglect spans five decades, beginning with Kahneman and Tversky's foundational work on heuristics and biases. Their studies showed systematic departures from normative probability judgment when specific, individuating information was available. The cab problem, first described by Tversky and Kahneman, became a canonical demonstration: participants given base rates (85% green, 15% blue) and witness reliability (80% accurate) produced median estimates of 80% rather than the Bayesian posterior of roughly 41%.

• Kahneman and Tversky (1972): Introduced representativeness as a heuristic for judging probability, showing that subjective probability depends on similarity to a prototype rather than statistical frequency, with sample size having little effect on likelihood judgments [1]
• Tversky and Kahneman (1973): Demonstrated base-rate neglect in the lawyer-engineer problem, where participants' predictions about a person's occupation were dominated by a personality sketch rather than prior probabilities, even when base rates were explicitly provided [2]
• Gigerenzer and Hoffrage (1995): Showed that natural frequencies (10 out of 1000) dramatically improve Bayesian reasoning compared to conditional probabilities, suggesting base-rate neglect partly reflects information format rather than fundamental irrationality [3]
• Stengård et al. (2022): Found individual differences form a bimodal distribution—some participants nearly ignore base rates while others nearly perfectly account for them, with the former group fit by linear-additive models and the latter by Bayesian models [4]

Base-rate neglect is not universal or inevitable. Gigerenzer and colleagues argue that people can reason effectively about probabilities when information is presented as learned frequencies rather than single-event probabilities. Koehler (1996) documented methodological challenges: many studies used ambiguous wording, and base-rate neglect diminishes when problems are clarified and when participants experience random sampling directly. Cosmides and Tooby (1996) found dramatically better performance when information was presented in frequency formats that may match human evolutionary experience. There is also ongoing debate about whether base-rate neglect reflects cognitive limitation versus adaptive strategy—ignoring base rates can be rational when they seem irrelevant to the specific case. Individual differences matter substantially, with statistical training reducing but not eliminating the effect.