Conjunction Fallacy (Linda Problem)

Explanation

The conjunction fallacy occurs when individuals mistakenly evaluate a conjunction (e.g., “Linda is a bank teller and active in the feminist movement”) as more likely than one of its constituents (“Linda is a bank teller”), violating the basic law of probability that a conjunction can never be more probable than any of its parts. The phenomenon was famously demonstrated by Tversky and Kahneman using the “Linda problem”:

• Linda is described as 31, single, outspoken, bright, a philosophy major, and a former activist for social justice.
• Participants rank the likelihood of eight statements, including “Linda is a bank teller” (T) and “Linda is a bank teller and active in the feminist movement” (T&F).

Despite the logical necessity that every feminist bank teller is a bank teller, around 85–90% of undergraduates—and even 85% of statistically sophisticated doctoral students—judged the conjunction T&F as more probable than T alone. A parallel case, the “Bill problem,” used a description of Bill that was representative of an accountant (A) and unrepresentative of a person who plays jazz for a hobby (J). The same pattern emerged: most subjects rated the conjunction “accountant and plays jazz” (A&J) as more probable than “plays jazz” (J).

The fallacy is driven by the representativeness heuristic. When asked to assess probability, people intuitively substitute a judgment of representativeness—how well a scenario matches a mental model or stereotype—for the logic of probability. The conjunction (T&F or A&J) seems more representative of the description because it incorporates more congruent features, forming a more coherent, plausible story. This is not inherently fallacious: in a separate study with 88 undergraduates, when participants were explicitly asked to rank statements by representativeness (resemblance to a typical class member), the same ordering emerged (F > T&F > T for Linda, A > A&J > J for Bill), with 85% and 87% of respondents respectively showing the predicted pattern. As the analogy goes, adding eyebrows to a schematic face makes it more similar to another face with eyebrows—adding shared features increases representativeness. The error arises when this intuitive assessment of representativeness is used to answer questions about probability, where set relations dictate that a conjunction cannot outrank its components.

The bias persists even when one is aware of it. As Stephen Jay Gould noted, “a little homunculus in my head continues to jump up and down,” illustrating the conflict between intuitive System 1 judgments and logical System 2 reasoning.

The conjunction fallacy has practical implications far beyond pencil-and-paper puzzles. In forecasting, adding specific, concrete details to a scenario makes it more persuasive even though adding a required detail cannot increase its probability. For example, participants in one study judged a detailed scenario (“An earthquake in California causing a flood with 1,000 drownings”) as more likely than a broader one (“A massive flood in North America”), even though the latter is a superset of the former. The same principle applies when evaluating companies, plans, or rare events: richness of detail can inflate subjective probability.

Key Insights

• Representativeness can make a conjunction feel more plausible than a constituent, but adding a detail need not have that effect. Comparing “Mark has blond hair” with “Mark has hair” illustrates the nested events; “blond hair” need not make the narrower claim feel more representative.
• Representativeness itself is not a mistaken judgment; it becomes a fallacy only when confused with probability. The same cognitive process that sensibly ranks conjunctions as more representative leads to error when misapplied.
• The bias reflects a tension between intuitive, narrative-driven thinking and logical, set-based reasoning. It can be mitigated by prompting explicit comparison of outcomes, which engages more analytical System 2 processes.
• In scenario planning, compare events that are genuinely nested. A more restrictive event cannot be more probable than the broader event that includes it, even when its story is more compelling.