The conjunction fallacy is a formal fallacy that occurs when it is assumed that specific conditions are more probable than a single general one.
Linda is 31 years old, single, outspoken, and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.
Which is more probable?
- Linda is a bank teller.
- Linda is a bank teller and is active in the feminist movement.
The majority of those asked chose option 2. However the probability of two events occurring together (in "conjunction") is always less than or equal to the probability of either one occurring alone—formally, for two events A and B this inequality could be written as and
For example, even choosing a very low probability of Linda being a bank teller, say Pr(Linda is a bank teller) = 0.05 and a high probability that she would be a feminist, say Pr(Linda is a feminist) = 0.95, then, assuming independence, Pr(Linda is a bank teller and Linda is a feminist) = 0.05 × 0.95 or 0.0475, lower than Pr(Linda is a bank teller).
Tversky and Kahneman argue that most people get this problem wrong because they use a heuristic (an easily calculated procedure) called representativeness to make this kind of judgment: Option 2 seems more "representative" of Linda based on the description of her, even though it is clearly mathematically less likely.
In other demonstrations they argued that a specific scenario seemed more likely because of representativeness, but each added detail would actually make the scenario less and less likely. In this way it could be similar to the misleading vividness or slippery slope fallacies. More recently Kahneman has argued that the conjunction fallacy is a type of extension neglect.
Joint versus separate evaluation
In some experimental demonstrations the conjoint option is evaluated separately from its basic option. In other words, one group of participants is asked to rank order the likelihood that Linda is a bank teller, a high school teacher, and several other options, and another group is asked to rank order whether Linda is a bank teller and active in the feminist movement versus the same set of options (without Linda is a bankteller as an option). In this type of demonstration different groups of subjects rank order Linda as a bank teller and active in the feminist movement more highly than Linda as a bank teller.
Separate evaluation experiments preceded the earliest joint evaluation experiments, and Kahneman and Tversky were surprised when the effect was still observed under joint evaluation.
In separate evaluation the term conjunction effect may be preferred.
Criticism of the Linda problem
The wording can be considered misleading. Arguably, respondent may have answered a different question: which person is more likely to fit Linda's profile, a bank teller, or a feminist bank teller? Here, given a basic probability of a banker fitting Linda's profile, the extra "feminist" attribute arguably increases the probability of this person fitting Linda's profile. Thus, the question boils down to whether the "feminist" attribute increases a person's chance of fitting Linda's profile, and the answer fits. It also can be viewed from perspective of a person trying to use as much information given as possible, and thus changing the meaning of the question to described above.
Critics such as Gerd Gigerenzer and Ralph Hertwig criticized the Linda problem on grounds such as the wording and framing. The question of the Linda problem may violate conversational maxims in that people assume that the question obeys the maxim of relevance. Gigerenzer argues that some of the terminology used have polysemous meanings, the alternatives of which he claimed were more "natural". He argues that the meaning of probable “what happens frequently”, corresponds to the mathematical probability people are supposed to be tested on, but the meanings of probable “what is plausible”, and “whether there is evidence” do not. The term "and" has even been argued to have relevant polysemous meanings. Many techniques have been developed to control for this possible misinterpretation but none of them has dissipated the effect.
Many variations in wording of the Linda problem were studied by Tversky and Kahneman. If the first option is changed to obey conversational relevance, i.e., "Linda is a bank teller whether or not she is active in the feminist movement" the effect is decreased, but the majority (57%) of the respondents still commit the conjunction error. If the probability is changed to frequency format (see debiasing section below) the effect is reduced or eliminated. However, studies exist in which indistinguishable conjunction fallacy rates have been observed with stimuli framed in terms of probabilities versus frequencies.
The wording criticisms may be less applicable to the conjunction effect in separate evaluation.[vague] The "Linda problem" has been studied and criticized more than other types of demonstration of the effect (some described below).
Policy experts were asked to rate the probability that the Soviet Union would invade Poland, and the United States would break off diplomatic relations, all in the following year. They rated it on average as having a 4% probability of occurring. Another group of experts was asked to rate the probability simply that the United States would break off relations with the Soviet Union in the following year. They gave it an average probability of only 1%.
In an experiment conducted in 1980, respondents were asked the following:
- Borg will win the match
- Borg will lose the first set
- Borg will lose the first set but win the match
- Borg will win the first set but lose the match
On average, participants rated "Borg will lose the first set but win the match" more highly than "Borg will lose the first set".
In another experiment, participants were asked:
Consider a regular six-sided die with four green faces and two red faces. The die will be rolled 20 times and the sequence of greens (G) and reds (R) will be recorded. You are asked to select one sequence, from a set of three, and you will win $25 if the sequence you choose appears on successive rolls of the die.
65% of participants chose the second sequence, though option 1 is contained within it and is shorter than the other options. In a version where the $25 bet was only hypothetical the results did not significantly differ. Tversky and Kahneman argued that sequence 2 appears "representative" of a chance sequence (compare to the clustering illusion).
Consider another example:
Which of the following events is most likely to occur within the next year?
1. The United States will withdraw all troops from Iraq.2. The United States will withdraw all troops from Iraq and bomb Iranian nuclear facilities.
The probability of the conjunctions is never greater than that of its conjuncts. Therefore, the first choice is more probable. No matter how unlikely it is that America will withdraw troops within the year from Iraq, it is even less likely they will do so and bomb nuclear facilities.
In one experiment the question of the Linda problem was reformulated as follows:
There are 100 persons who fit the description above (that is, Linda’s). How many of them are:
- Bank tellers? __ of 100
- Bank tellers and active in the feminist movement? __ of 100
Whereas previously 85% of participants gave the wrong answer (bank teller and active in the feminist movement), in experiments done with this questioning none of the participants gave a wrong answer.
However, in some tasks only based on frequencies, not on stories, that used clear logical formulations, conjunction fallacies continued to occur dominantly when the observed pattern of frequencies resembled a conjunction (only few exceptions).
- Gould, Stephen J. (1988). "The Streak of Streaks". The New York Review of Books.
- Tversky, A. and Kahneman, D. (1982) "Judgments of and by representativeness". In D. Kahneman, P. Slovic & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases. Cambridge, UK: Cambridge University Press.
- Tversky, A. and Kahneman, D. (October 1983). "Extension versus intuitive reasoning: The conjunction fallacy in probability judgment". Psychological Review 90 (4): 293–315. doi:10.1037/0033-295X.90.4.293.
- Kahneman, Daniel. (2000). "Evaluation by moments, past and future". In Daniel Kahneman and Amos Tversky (Eds.). Choices, Values and Frames.
- Kahneman, Daniel (2011). "Linda: Less is More". Thinking, Fast and Slow. New York: Farrar, Straus and Giroux. pp. 156–165.
- Gigerenzer, G. (1996). On narrow norms and vague heuristics: A reply to Kahneman and Tversky. Psychological Review, 103, 592-596.
- Hertwig, Ralph; Gerd Gigerenzer (1999). "The ‘Conjunction Fallacy’ Revisited: How Intelligent Inferences Look Like Reasoning Errors". Journal of Behavioral Decision Making 12: 275–305. doi:10.1002/(sici)1099-0771(199912)12:4<275::aid-bdm323>3.3.co;2-d.
- Mellers, Barbara; Ralph Hertwig & Daniel Kahneman (2001). "Do frequency representations eliminate conjunction effects? An exercise in adversarial collaboration". Psychological Science 12 (4): 269–275. doi:10.1111/1467-9280.00350.
- Moro, R. (2009). On the nature of the conjunction fallacy. Synthese, 171,1–24.
- Tentori, K. & Crupi, V. (2012). On the conjunction fallacy and the meaning of and, yet again: A reply to Hertwig, Benz, and Krauss (2008). Cognition, 122, 123–134.
- See, for example, Tentori, K., Bonini, N., & Osherson, D. (2004). The conjunction fallacy: A misunderstanding about conjunction? Cognitive Science, 28, 467–477. Or Weddell & Moro, 2008.[full citation needed]
- Kahneman, D., & Tversky, A. (1996). On the reality of cognitive illusions. Psychological Review, 103, 582-591.
- "The Conjunction Fallacy". fallacyfiles.org. Retrieved 12/2/14. Check date values in:
- Gigerenzer, G. (1991). "How to make cognitive illusions disappear: Beyond ‘heuristics and biases.’". European Review of Social Psychology 2 (1): 83–115. doi:10.1080/14792779143000033.
- von Sydow, M. (1991). "The Bayesian Logic of Frequency-Based Conjunction Fallacies.". Journal of Mathematical Psychology 55 (2): 119–139. doi:10.1016/j.jmp.2010.12.001.