Daniel Kahneman, answered from the texts and cited to the page.
The phenomenon is simple to describe and surprisingly hard to cure. When people are given both statistical information about a population and a vivid description of a particular case, the description tends to swamp the statistics almost entirely. Amos and I demonstrated this with the Tom W problem and with the engineers-and-lawyers study. In the latter, subjects were shown brief personality descriptions of individuals allegedly sampled from a group of 100 professionals.
One group was told the sample contained 70 engineers and 30 lawyers; another was told the reverse. By Bayes' rule, the ratio of the odds should differ by a factor of 5.44 across the two conditions.1 The subjects produced essentially identical probability judgments in both conditions — the base rates might as well not have been stated.2 The description of the individual simply displaced the statistical information.
The mechanism behind this is representativeness: people assess the probability that Steve is a librarian, say, by how closely he resembles their stereotype of a librarian.3 But similarity to a stereotype is not sensitive to how many librarians there actually are. There are far more farmers than librarians in the population, and that fact should enter into any reasonable estimate — yet it does not, because it has no bearing on whether Steve looks like a librarian.4
What Amos and I originally concluded — that base-rate information will always be neglected when individual information is available — turned out to be too strong.5 Later work showed that instructing people to think like a statistician rather than a clinician increased base-rate sensitivity, and a study with Harvard undergraduates found that even inducing a frowning expression, which tends to engage more deliberate processing, produced some improvement.6
That finding points to something important: the failure is often not ignorance but laziness. System 2 knows base rates are relevant; it simply does not bother to apply that knowledge unless pushed.7
Specifically, it can be shown by applying Bayes' rule that the ratio of these odds should be (.7/.3)2, or 5.44, for each description.Judgment under Uncertainty, Heuristics and Biases, p. 2
In a sharp violation of Bayes' rule, the subjects in the two conditions produced essentially the same probability judgments. Apparently, subjects evaluated the likelihood that a particular description belonged to an engineer rather than to a lawyer by the degree to which this description was representative of the two stereotypes, with little or no regard for the prior probabilities of the categories.Thinking, Fast and Slow
In the representativeness heuristic, the probability that Steve is a librarian, for example, is assessed by the degree to which he is representative of, or similar to, the stereotype of a librarian.Judgment under Uncertainty, Heuristics and Biases, p. 2
In the case of Steve, for example, the fact that there are many more farmers than librarians in the population should enter into any reasonable estimate of the probability that Steve is a librarian rather than a farmer. Considerations of base-rate frequency, however, do not affect the similarity of Steve to the stereotypes of librarians and farmers.Thinking, Fast and Slow
Amos and I originally believed, on the basis of our early evidence, that base-rate information will always be neglected when information about the specific instance is available, but that conclusion was too strong.Thinking, Fast and Slow
Norbert Schwarz and his colleagues showed that instructing people to 'think like a statistician' enhanced the use of base-rate information, while the instruction to 'think like a clinician' had the opposite effect.Thinking, Fast and Slow
Their System 2 'knows' that base rates are relevant even when they are not explicitly mentioned, but applies that knowledge only when it invests special effort in the task.Thinking, Fast and Slow