Trap/Cognitive Bias/No. 0196
Conjunction Fallacy
The conjunction fallacy is judging two events together as more probable than either event alone. Named by Amos Tversky and Daniel Kahneman in 1983, this error in probability judgment violates the rule that a combined event cannot be more likely than any of its parts.
Also called Conjunction Error
- Evidence
- Well established
- Read
- 6 min
- Links
- 6 connections
01You've seen this when…
- in life
Your friend misses your call. Her phone dying and her charger being left at work feels more likely than her phone being dead, whatever the reason.
- at work
An outage report describes a cyberattack backed by a foreign government. You give that detailed account higher odds than a cyberattack, regardless of who is behind it.
- out in the world
A candidate fits your image of a housing activist. You judge it more likely that she wins a council seat and pushes for rent controls than that she wins a council seat at all.
02The idea
A detailed story can feel more believable than a broad possibility. The trouble starts when that feeling becomes a probability judgment.
Suppose you are comparing two forecasts: a company misses its revenue target, or the company misses its revenue target and replaces its chief executive. Every outcome that satisfies the second forecast also satisfies the first. The combined outcome cannot be more likely than the broader one.
The conjunction fallacy is judging A and B more probable than A, or more probable than B. It breaks a simple inclusion rule: the cases where both happen sit inside the cases where either component happens.
This does not require the events to be independent. They might be tightly connected. A missed target could make a leadership change much more likely. But some missed targets will still happen without that change, and adding the requirement cannot create more qualifying outcomes.
The mistake isn’t preferring a detailed explanation. It’s giving that explanation higher odds than an event it necessarily contains.
03Why it happens
- A good match feels like a high probability. A politically active philosophy graduate fits the image of a feminist better than the image of a bank teller. That fit can dominate the judgment even when feminism is an extra requirement. The representativeness heuristic bases the judgment on how well the story matches. A probability judgment requires checking how many outcomes satisfy it.
- One question quietly replaces another. How convincing the account sounds becomes a stand-in for the chance that all the stated events happen. This is attribute substitution: an easier judgment takes over from the one requested.
- A connected story disguises its extra conditions. Details that explain one another make an account feel coherent. You notice the causal connection and overlook that every required detail must hold.
- Ordinary language can change the task. People sometimes read a broad option as excluding the detail mentioned in another option. A bank teller may be understood as a bank teller who is not a feminist. That interpretation changes the comparison. Some apparent errors arise when people who understand probability interpret ambiguous wording differently.
04A worked example
In Tversky and Kahneman’s classic Linda problem, participants read about a 31-year-old former philosophy major who is outspoken and concerned with discrimination and social justice. Among the possibilities they assess are that Linda is a bank teller and that Linda is a bank teller active in the feminist movement.
Many participants rank the combined description as more probable.
What it looks like A sensitive reading of the person. Linda’s interests strongly suggest political involvement, so the richer description seems to capture her better.
What’s actually going on If bank teller means any bank teller, including feminist bank tellers, every Linda who satisfies the combined description already satisfies the broad one. Her political interests may support the feminist part. They cannot make the combination more probable than bank teller by itself.
What would have helped Making the inclusion explicit and asking people to count cases among women who fit the biography. Imagine a group of women fitting Linda’s description. Count all the bank tellers, then count the feminist bank tellers within that group. The second count cannot exceed the first. Research finds that such changes can reduce errors, with mixed results for eliminating them altogether.
05How to spot it
06What to do instead
- Make the broad option inclusive. Rewrite it to include the narrower possibility explicitly: a cyberattack, including attacks backed by a foreign government. This prevents an accidental comparison between mutually exclusive categories.
- Break the account into requirements. List everything that must be true for the forecast to succeed. Watch for added dates, locations, motives and follow-on events. Each required condition narrows, or leaves unchanged, the set of successful outcomes.
- Count cases using one denominator. Imagine 100 possible outages. If 20 are cyberattacks, no more than 20 can be foreign-government-backed cyberattacks. These are illustrative counts; the inclusion relationship holds whatever the real numbers are.
- Check the combination against each component. Estimate the component events separately, using the same information and time frame. The combined estimate must fall at or below each component estimate. Use that comparison as a deliberate checking step, especially when a story feels unusually convincing.
- Separate explanation from prediction. A detail may help explain an event or distinguish between competing causes. That contribution helps with explanation. Estimating whether every part of a scenario will occur is a separate task.
07When it isn’t a conjunction error
Adding information can legitimately raise a probability. The chance of a recession given that a market crash has occurred may be higher than the chance of a recession before learning about the crash. That is conditional probability: the probability of one event given another. The probability that both events will happen is a separate quantity. The evidence and the denominator have changed.
Likewise, Bayesian updating lets new evidence strengthen a hypothesis. The conjunction rule compares outcomes using the same evidence. Updating tracks beliefs before and after an observation.
A combination can be as likely as its component. If the second condition is guaranteed whenever the first happens, their probabilities are equal. Multiplying the separate probabilities is a safe shortcut only when the events are independent.
This also differs from the base rate fallacy. You can detect a conjunction error without knowing the actual frequency of either event. Inclusion alone establishes the upper bound.
08Roots
Amos Tversky and Daniel Kahneman gave their probability puzzle a memorable protagonist: a philosophy graduate who had participated in anti-nuclear demonstrations. Linda’s biography supplied exactly the details readers use to form a vivid impression. Then the researchers asked them to assess possibilities that crossed that impression with an ordinary occupation.
Their 1983 paper made the conjunction fallacy a striking example of the gap between intuitive judgment and the rules of probability. Linda was memorable because readers could understand the inclusion rule and still feel the pull of the richer description. The problem spread through psychology teaching and later through popular accounts of decision-making.
It also started a debate about what researchers were really asking. Ralph Hertwig and Gerd Gigerenzer argued that participants can make reasonable conversational inferences from ambiguous experimental wording. Other researchers tested versions designed to clarify the conjunction and still found errors. The lasting lesson is two-sided: coherent stories can distort probability judgments, and a confusing question can make sensible interpretation look like bad reasoning.
09How solid is this?
The error has been observed repeatedly, including in tasks designed to clarify the inclusion relationship. Wording, frequency formats and participants’ interpretations substantially affect responses, so what the classic Linda result reveals about people’s reasoning ability depends on those factors.
10Connections
- Often confused with Conditional Probability
- Countered byCognitive Forcing Strategy
- Can follow from Representativeness Heuristic, Attribute Substitution
- See also Bayesian Updating, Base Rate Fallacy
11Origin and sources
Amos Tversky and Daniel Kahneman named and systematically described the conjunction fallacy in their 1983 paper on intuitive probability judgment.
- [1]Tversky, A., & Kahneman, D. (1983). Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment. Psychological Review, 90(4), 293–315.
- [2]Hertwig, R., & Gigerenzer, G. (1999). The 'conjunction fallacy' revisited: How intelligent inferences look like reasoning errors. Journal of Behavioral Decision Making, 12(4), 275–305.
- [3]Tentori, K., Bonini, N., & Osherson, D. (2004). The conjunction fallacy: a misunderstanding about conjunction? Cognitive Science, 28(3), 467–477.
Suggest an edit· Updated 2026-10-02