The Linda problem is Daniel Kahneman and Amos Tversky’s best-known experiment on judgment under uncertainty, designed to place intuitive representativeness in direct conflict with the logic of probability. A vivid portrait makes Linda seem unmistakably like an active feminist; when participants must compare “bank teller” with “bank teller and active in the feminist movement,” many choose the more detailed description, even though every feminist bank teller is necessarily a bank teller. The experiment therefore became the defining demonstration of the conjunction fallacy and a compact illustration of how a coherent story can overpower a rule that the mind is capable of understanding.

The Portrait and Its Logic

Linda was introduced as thirty-one, single, outspoken, very bright, and a philosophy major. As a student, she had been deeply concerned with discrimination and social justice and had taken part in antinuclear demonstrations. The description was deliberately constructed to evoke a recognizable social and political profile rather than a neutral statistical case. Audiences in the 1980s often laughed because they immediately inferred that Linda had attended the University of California at Berkeley, whose students were then associated with radical political engagement.

Participants received a list of possible descriptions of Linda’s life. The alternatives included elementary-school teacher, bookstore worker who took yoga classes, feminist activist, psychiatric social worker, member of the League of Women Voters, bank teller, insurance salesperson, and bank teller who was active in the feminist movement. The list reflected the period in which the experiment was conducted: the League of Women Voters was more prominent, and the language of a feminist “movement” now sounds historically dated. Yet the central judgment remained easy to anticipate. Linda seemed an excellent fit for an active feminist, a plausible fit for a bookstore worker who took yoga classes, and a poor fit for a bank teller or insurance salesperson.

The decisive comparison concerned the two bank-teller descriptions. Linda appeared more like a feminist bank teller than like the stereotype of a bank teller alone, because the additional feminist detail made the portrait more coherent. But probability imposes a different requirement. The set of feminist bank tellers is wholly contained within the set of bank tellers; the conjunction cannot be more probable than one of its components. Adding a condition can only narrow the set of cases. The experiment thus separated two forms of judgment that usually cooperate: representativeness asks which description resembles the portrait, while probability asks which event is likelier.

The Experimental Discovery

The initial experiment used a between-subjects design. Some participants ranked outcomes by resemblance, while others ranked them by likelihood. Each participant saw only one of the two critical alternatives, either “bank teller” or “feminist bank teller.” The average rankings were nevertheless identical across the two tasks: “feminist bank teller” was placed above “bank teller” both as a resemblance judgment and as a probability judgment.

Kahneman and Tversky then created a within-subject questionnaire containing both alternatives, with “bank teller” in the sixth position and “feminist bank teller” as the final item. They expected participants to notice the logical relation and answer consistently. Because they were so confident that a special test was unnecessary, an assistant gave the questionnaire to people completing another experiment while signing out of the laboratory before payment.

About ten completed questionnaires accumulated on the assistant’s gray metal desk before Kahneman casually looked at them. Every respondent had ranked “feminist bank teller” as more probable than “bank teller.” The discovery was so unexpected that he retained a vivid flashbulb memory of the desk’s gray color and of where everyone was standing, then called Amos in excitement. The experiment had placed logic against representativeness, and representativeness had won.

The result was not confined to a small or inexperienced sample. Eighty-nine percent of the undergraduates in the sample violated the logic of probability. The researchers expected doctoral students in decision science at the Stanford Graduate School of Business, all of whom had taken advanced courses in probability, statistics, and decision theory, to perform better. Instead, 85 percent of those respondents also ranked “feminist bank teller” as more probable than “bank teller.”

The Conjunction Fallacy

In increasingly determined attempts to eliminate the error, the researchers presented large groups with the stripped-down question asking which alternative was more probable: Linda is a bank teller, or Linda is a bank teller and active in the feminist movement. At several major universities, roughly 85 to 90 percent of undergraduates selected the second option. When Kahneman asked an indignant class whether the students realized that they had violated an elementary logical rule, a person in the back row answered, “So what?” A graduate student who had made the same mistake explained that she thought she had merely been asked for her opinion.

Kahneman and Tversky called this error the conjunction fallacy. It occurs when people judge a conjunction of two events—in this case, being a bank teller and being active in the feminist movement—to be more probable than one of those events considered alone. The error is especially revealing because the relevant rule is not hidden. Participants are shown both alternatives in a direct comparison and therefore have a fair opportunity to recognize that the conjunction is a subset of the single event. The failure lies not in an absence of information but in the failure to deploy an available logical operation.

The experiment also showed that the fallacy is attractive even after it has been recognized. Stephen Jay Gould understood the correct answer, yet described an insistent “little homunculus in my head” that kept shouting that Linda could not merely be a bank teller and that the description should be read again. In the later language of the two-systems framework, that homunculus represented System 1: a fast, associative process responding to the portrait’s social meaning while System 2 struggled to impose the probability rule.

Comparison, Attention, and Expertise

The proportion of correct answers changed when the task made the comparison more explicit. In one study, 64 percent of graduate students in the social sciences at Stanford and the University of California, Berkeley correctly judged “feminist bank teller” to be less probable than “bank teller.” Yet in the original eight-outcome version, only 15 percent of a similar group of graduate students made the correct choice.

The difference arose from the structure of attention. In the longer list, the two critical outcomes were separated by “insurance salesperson,” so participants evaluated each item independently and apparently relied on its fit with Linda’s portrait. The shorter version placed the alternatives together and required an explicit comparison. That arrangement mobilized System 2 and enabled many statistically sophisticated students to avoid the fallacy. The result did not show that expertise erased intuition; it showed that a carefully designed task could sometimes force the relevant logical relation into working attention.

The Linda problem therefore distinguishes knowledge from performance. Participants could know probability theory and still produce an answer governed by representativeness. The doctoral students’ 85 percent error rate made the point especially sharply: formal training did not automatically protect them when the problem was encountered in a psychologically compelling form. Even when the correct rule was elementary, the portrait supplied a more immediate and satisfying answer, and the mind treated coherence as evidence.

Significance in Judgment Research

The Linda problem became famous because it made an abstract conflict visible in a single human portrait. The description invites readers to construct a rich story about Linda’s political commitments, education, and likely occupations. The conjunction adds detail and improves the story’s fit, but that narrative improvement is precisely what makes the probability judgment worse. The experiment demonstrates how a representative description can substitute for the question actually asked: instead of asking which event includes more possible cases, respondents ask which description sounds more like Linda.

Its broader significance lies in the relation between intuitive judgment and normative standards. The fallacy is not merely a computational slip that disappears once the arithmetic is explained. Like the Müller-Lyer illusion, it remains psychologically compelling even when people know it is wrong. The subject can recognize the logical answer and still feel the force of the representative one. The experiment thus became central evidence that judgment is shaped by heuristics, that System 2 often fails to correct System 1, and that people may need carefully structured comparisons—not simply more knowledge—to make judgments consistent with probability.