I think that the between-subjects versions of the conjunction fallacy studies actually provide stronger evidence that people are making the mistake. For instance*, one group of subjects is given the description of Linda and asked to rank the following statements in order, from the one that is most likely to be true of Linda to the one that is least likely to be true of her:
a) Linda is a teacher in elementary school b) Linda works in a bookstore and takes Yoga classes c) Linda is a psychiatric social worker d) Linda is a member of the League of Women voters e) Linda is a bank teller and is active in the feminist movement f) Linda is an insurance salesperson
A second group of subjects does the same task, but statement (e) is replaced by “Linda is a bank teller.”
In this version of the experiment, there should be no ambiguity about what is being asked for, or about the meaning of the words “likely” or “and”—the subjects can understand them, just as statisticians and logicians do. But ‘feminist bank teller’ still gets judged as more likely than ‘bank teller’ relative to the alternatives; that is, subjects in the first group give statement (e) an earlier (more likely) ranking than subjects in the second group.
Some people don’t like the between-subjects design because it’s less dramatic, and you can’t point to any one subject and say “that guy committed the conjunction fallacy”, but I think it actually provides more clear-cut evidence. Subjects who saw a conjunctive statement about Linda being a bank teller and a feminist considered it more likely than they would have if the statement had not been a conjunction (and had merely included “bank teller”). This isn’t a trick, it’s an error that people will consistently make—overestimating the likelihood that a very specific or rare statement applies to a situation, even considering it more likely than they would have considered a broader statement which encompasses it, because it sounds like it fits the situation.
* This study design has been used, with these results, I think in Kahneman and Tversky’s original paper on representativeness. But I didn’t take the time to find the paper and reproduce the exact materials used, I just googled and took the 6 statements from here.
Looking at some of this, I wonder if people are biased towards thinking that the more concrete statement is more likely? Somehow, in my mind, “feminist” is more abstract than “feminist book keeper”. The latter seems closer to being a person, whereas the former seems to be closer to a concept. The more descriptive you are about a subject, the more concrete it sounds, and thus the more likely it is, because it sounds closer to reality. The less descriptive, the more abstract it sounds, and therefore the less likely it is, because it sounds more hypothetical or “theoretical”.
Of course, the more descriptive account is going to have more conjunctions, and therefore has lesser or the same probability. I just wonder if this has been taken into account.
I think that the between-subjects versions of the conjunction fallacy studies actually provide stronger evidence that people are making the mistake. For instance*, one group of subjects is given the description of Linda and asked to rank the following statements in order, from the one that is most likely to be true of Linda to the one that is least likely to be true of her:
a) Linda is a teacher in elementary school
b) Linda works in a bookstore and takes Yoga classes
c) Linda is a psychiatric social worker
d) Linda is a member of the League of Women voters
e) Linda is a bank teller and is active in the feminist movement
f) Linda is an insurance salesperson
A second group of subjects does the same task, but statement (e) is replaced by “Linda is a bank teller.”
In this version of the experiment, there should be no ambiguity about what is being asked for, or about the meaning of the words “likely” or “and”—the subjects can understand them, just as statisticians and logicians do. But ‘feminist bank teller’ still gets judged as more likely than ‘bank teller’ relative to the alternatives; that is, subjects in the first group give statement (e) an earlier (more likely) ranking than subjects in the second group.
Some people don’t like the between-subjects design because it’s less dramatic, and you can’t point to any one subject and say “that guy committed the conjunction fallacy”, but I think it actually provides more clear-cut evidence. Subjects who saw a conjunctive statement about Linda being a bank teller and a feminist considered it more likely than they would have if the statement had not been a conjunction (and had merely included “bank teller”). This isn’t a trick, it’s an error that people will consistently make—overestimating the likelihood that a very specific or rare statement applies to a situation, even considering it more likely than they would have considered a broader statement which encompasses it, because it sounds like it fits the situation.
* This study design has been used, with these results, I think in Kahneman and Tversky’s original paper on representativeness. But I didn’t take the time to find the paper and reproduce the exact materials used, I just googled and took the 6 statements from here.
Looking at some of this, I wonder if people are biased towards thinking that the more concrete statement is more likely? Somehow, in my mind, “feminist” is more abstract than “feminist book keeper”. The latter seems closer to being a person, whereas the former seems to be closer to a concept. The more descriptive you are about a subject, the more concrete it sounds, and thus the more likely it is, because it sounds closer to reality. The less descriptive, the more abstract it sounds, and therefore the less likely it is, because it sounds more hypothetical or “theoretical”.
Of course, the more descriptive account is going to have more conjunctions, and therefore has lesser or the same probability. I just wonder if this has been taken into account.
Note for example the contextual hints in
which bias it.