Base rate neglect
Also known as: Base rate fallacy
We fixate on specific detail and forget the underlying odds.
What it means
Base rate neglect is the tendency, when individuating information is available, to underweight or ignore the prior probability — the base rate — of an event in a population. The mechanism is closely tied to representativeness: a vivid description that 'fits' a category captures attention and crowds out the colorless statistical fact of how common that category actually is. The result is badly miscalibrated probability judgments, most dangerously when a rare condition is assessed with an imperfect diagnostic test, where ignoring the low prior makes false positives feel like true ones. An important boundary condition is that base rates are not always neglected: people use them more when the base rate is causally relevant, framed in natural frequencies rather than percentages, or made salient and easy to combine with the evidence. This matters acutely in medical screening, criminal forensics, hiring, and any setting where a 'positive' result from an accurate-sounding test is mistaken for near-certainty.
Origins and the classic demonstration
Kahneman and Tversky (1973) sketched a person supposedly drawn at random from a pool described as either seventy engineers and thirty lawyers, or the reverse, then asked the odds the person was an engineer. The base rate barely moved judgments: a vaguely engineer-ish description produced nearly the same guess in both pools, even though the composition was printed in the instructions. Only when the sketch was deliberately uninformative did people fall back on the prior. Maya Bar-Hillel (1980) sharpened the account. People rank information by perceived relevance and let the item that feels most relevant dominate, so concrete individuating detail swamps the abstract prior. The base rate is not so much forgotten as quietly judged beside the point.
What the evidence shows
Casscells, Schoenberger and Graboys (1978) put a textbook question to staff and students at Harvard teaching hospitals: a test with a five percent false-positive rate for a disease affecting one person in a thousand returns a positive result, with no other information. The correct answer is about two percent; the most common reply was ninety-five percent, and only eighteen percent answered correctly. Yet universal neglect is a caricature. Koehler's (1996) review concluded that base rates are almost always used to some degree, and that their weight rises when the prior is reliable, causally relevant, or more diagnostic than the individuating cue. Neglect is a matter of degree, and how much depends heavily on how the problem is posed.
The format fix and its limits
Gigerenzer and Hoffrage (1995) argued the difficulty lies partly in representation rather than in the mind. Recast the same problem in natural frequencies, telling people that ten of every thousand have the disease and that of the nine hundred and ninety without it roughly fifty still test positive, and correct Bayesian answers rise sharply, because the format does the normalizing that percentages hide. But the fix is only partial. McDowell and Jacobs's (2017) meta-analysis of more than two hundred estimates found natural frequencies lifted correct solutions to around twenty-four percent, against four percent for probability formats: a large relative gain that still leaves three in four people wrong. Format helps; it does not make Bayesians of us.
Using it in practice
The lesson for anyone reporting a test result is to state the number people actually need, the positive predictive value, rather than the sensitivity that sounds reassuring, and to express the underlying counts as frequencies over a concrete reference class. Show the full picture: true positives, false positives, and the base rate that generates both. The danger concentrates wherever a rare target meets an imperfect detector, in disease screening, fraud and intrusion alerts, and forensic database matches, because there a confident-sounding hit is dominated by the false positives that a low base rate guarantees. The discipline is simple to state and hard to keep: ask how common the thing is before you ask how good the test is.
Examples
A test that is '95% accurate' for a disease only 1 in 1,000 people have still flags mostly false positives — a fact most people, including clinicians, get badly wrong.
A fraud filter that wrongly flags one honest payment in a hundred sounds excellent until you notice only one payment in ten thousand is fraudulent, so most blocked cards are innocent.
A manager rejects a candidate because they do not seem like an engineer, never asking how many of the applicants, most of whom were engineers, fit that mental picture anyway.
Told that a soft-spoken stranger who loves poetry is either a literature professor or a delivery driver, most people say professor, overlooking that delivery drivers vastly outnumber literature professors, so the fitting description swamps the far larger prior.
A jury hears that a suspect's DNA matches the crime sample 'one in a million' and reads near-certain guilt, forgetting that trawling a database of millions makes one innocent coincidental match unsurprising.
First described in Kahneman & Tversky (1973).
Key references
- McDowell, M., & Jacobs, P. (2017). Meta-analysis of the effect of natural frequencies on Bayesian reasoning. Psychological Bulletin, 143(12), 1273-1312. doi.org/10.1037/bul0000126
- Koehler, J. J. (1996). The base rate fallacy reconsidered: Descriptive, normative, and methodological challenges. Behavioral and Brain Sciences, 19(1), 1-53. doi.org/10.1017/S0140525X00041157
- Gigerenzer, G., & Hoffrage, U. (1995). How to improve Bayesian reasoning without instruction: Frequency formats. Psychological Review, 102(4), 684-704. doi.org/10.1037/0033-295X.102.4.684
- Bar-Hillel, M. (1980). The base-rate fallacy in probability judgments. Acta Psychologica, 44(3), 211-233. doi.org/10.1016/0001-6918(80)90046-3
- Casscells, W., Schoenberger, A., & Graboys, T. B. (1978). Interpretation by physicians of clinical laboratory results. New England Journal of Medicine, 299(18), 999-1001. doi.org/10.1056/NEJM197811022991808
- Kahneman, D., & Tversky, A. (1973). On the psychology of prediction. Psychological Review, 80(4), 237-251. doi.org/10.1037/h0034747