Stereotyping: How a Sketch of a Person Overrides the Odds

A stereotype is a stored picture of a group, and stereotyping is the act of reading an individual through it. The picture does work that is useful in general and wrong in particular: it lets a person form an impression of a stranger in seconds, and it does so by substituting the group's supposed traits for facts about the person that have not been gathered yet.

The demo shows the mechanism at its barest. Dick's description was written to say nothing about whether he is an engineer or a lawyer; it fits either equally well. The only information that bears on the question is the makeup of the group he was drawn from, and the right answer is simply that share, 30 percent in one version and 70 percent in the other. Kahneman and Tversky found that once any description of a person was present, people judged how well the person resembled their picture of an engineer and more or less stopped using the odds. A description that resembled neither picture came out at 50 percent regardless of the group. They called the shortcut representativeness: a judgment of probability is replaced by a judgment of similarity.

The pictures come partly from experience, partly from what other people say, and partly from a quirk of memory. Hamilton and Gifford showed in 1976 that people form a stereotype about a smaller group with no basis at all: shown statements about two groups that behaved identically in proportion, participants came away believing the smaller group had done more of the rare, undesirable things, because rare pairings stand out and are overcounted. Once a picture exists, Devine's 1989 experiments suggest it is activated automatically on meeting a group member, in people who endorse it and people who reject it alike; what differs between them is whether the controlled, slower process that follows corrects for it.

The base-rate part of the story has been argued over. Bar-Hillel showed in 1980 that base rates are neglected across many versions of the problem, while Koehler argued in 1996 that people do use base rates in some formats and settings, especially when they are learned from experience rather than stated, and that the laboratory results had been over-generalized.

Daniel Kahneman and Amos Tversky published "On the psychology of prediction" in Psychological Review in 1973. One of its experiments has become the standard demonstration of how a description crowds out the odds.

Participants were told that a panel of psychologists had interviewed and tested a group of 100 professionals, all of them engineers or lawyers, and had written a short personality sketch of each. One group of participants was told the hundred consisted of 30 engineers and 70 lawyers; another group was told 70 engineers and 30 lawyers. Everyone then read five sketches and, for each, gave the probability that the man described was an engineer. Some sketches were written to fit the stereotype of an engineer, such as Jack, a careful and conservative man with no interest in politics whose hobbies included carpentry, sailing and mathematical puzzles; Dick's sketch, reproduced in the demo above, was written to fit neither profession.

The two base-rate groups gave nearly the same probabilities for every sketch. A sketch that sounded like an engineer was judged likely to be an engineer whether engineers were 30 or 70 percent of the group, and the median judgment for Dick was 50 percent in both conditions. Participants who were given no sketch at all, only the group makeup, answered with the base rate, which shows that they understood the odds and could use them; a few lines of personality were enough to set them aside. The authors took this as evidence that intuitive prediction runs on representativeness, how much the case resembles the category, rather than on the probability rules that the problem calls for.

The shortcut appears whenever a person is read through a group before they are known. The rows below are illustrations, not findings from the studies cited on this page.

Stereotypes activate quickly and correction is slow, so the counters below work by building in the slow step.

• Ask for the base rate first. Before reading a description, estimate how common the category is in the group you are drawing from, write it down, and treat the description as an adjustment from that number rather than a replacement for it.
• Ask what the description would look like if the person belonged to the other group. Dick's sketch fits a lawyer as well as an engineer, which is the test of whether it carries any information.
• Replace resemblance with records. Structured interviews, work samples and blind review all put facts about the person in front of the picture of the group.
• Notice the activation and treat it as a prompt. Devine's work found that the stereotype comes to mind in people who reject it too; the difference is what they do in the next second.
• Get more individuating information before a judgment that matters. Stereotypes are strongest when a person is known by one or two cues, and they lose ground as specific knowledge about the person accumulates.

Stereotyping sits inside a cluster of related terms, and several of them have their own pages on this site.

• It is not prejudice, which is an attitude toward a group, usually a negative one, with feeling attached. A stereotype is a belief about what members of a group are like; it can be held without hostility and can concern groups a person feels nothing about, such as the engineers and lawyers in the demo. Prejudice usually draws on stereotypes, but the two are measured and studied separately.
• It is not the base-rate fallacy in general, which is any neglect of how common something is in favor of case-specific evidence. The engineer-lawyer problem is a base-rate problem in which the case-specific evidence happens to be a stereotype; base rates are also neglected in medical testing and other settings where no group picture is involved.
• It is not implicit stereotyping, the term for stereotype associations that operate without awareness and are measured indirectly, for example by reaction times. This page is about the explicit judgment a person makes when asked; implicit measures and how they relate to behavior are a separate and contested literature.
• It is not out-group homogeneity bias, the tendency to see members of groups one does not belong to as more alike than members of one's own. That bias is about how varied a group seems; stereotyping is about reading one person through the group's picture.