Thinking and Deciding

HEURISTICS AND BIASES IN PROBABILITY 151
Subjects were then asked to rank the following items in terms of probability:
Linda is a teacher in an elementary school.
Linda works in a bookstore and takes yoga classes.
Linda is active in the feminist movement. [F]
Linda is a psychiatric social worker.
Linda is a member of the League of Women Voters.
Linda is a bank teller. [B]
Linda is an insurance salesperson.
Linda is a bank teller and is active in the feminist movement. [B and F].
The critical items here are those that I have marked F (“feminist”), B (“bank teller”),
and B and F; the other items are fillers, designed to disguise the issue. Of course,
subjects rated F as very probable and B as very unlikely. However, they rated “B and
F” as more probable than B alone. This is called the “conjunction fallacy” because “B
and F” is a conjunction of B and F, and a conjunction cannot be more probable than
either of its components. This result was found even when the subjects who rated B
were not the same as the subjects asked to rate “B and F” or F, so it does not seem
that subjects misunderstood “bank teller” to mean “bank teller but not feminist.” The
subjects who rated B alone heard no mention of feminism.
Of course, the requirement of coherence in probability judgments makes this
choice impossible. The set of people who are both bank tellers and feminists cannot
be larger than the set of female bank tellers. These sets would be the same only if
every female bank teller were an active feminist.
What the subjects do seem to be doing, once again, is judging probability according
to representativeness, or similarity. Although the description given was not
judged (by other subjects) as very representative of women bank tellers, it was judged
to be more representative of women bank tellers who are feminists. (See Tversky and
Kahneman, 1983, for other examples and an interesting discussion.) This error leads
to an incorrect ordering of probabilities. It, therefore, concerns the strengths of beliefs
themselves, not just our ability to assign numbers to beliefs.
A simpler form of the conjunction fallacy can be found in children. Agnoli
(1991) asked children questions such as “In summer at the beach are there more
women or more tanned women?” or “Does the mailman put more letters or more
pieces of mail in your mailbox?” Tanned women and letters are more representative,
and most children, even seventh graders, judged these as more likely than the more
inclusive set.
Is use of the representativeness heuristic irrational? If so, how? In everyday reasoning,
it is not necessarily irrational to use similarity in our judgments. In many
cases, there is little else we can rely on as a guide to the interpretation of a piece of
evidence. In the taxicab problem and the other problems just discussed, however,
other relevant information is either provided or readily accessible in memory. In
these cases, subjects seem to be overgeneralizing a heuristic that is typically useful,
the heuristic of judging probability by representativeness. From a prescriptive point