Thinking and Deciding

160 DESCRIPTIVE THEORY OF PROBABILITY JUDGMENT
physics discussed in Chapter 1. Evidence of bias in averaging is evidence of a need
for education — but the kind of education in question is specific to probability.
The important aspect of this view is its claim that biases are specific to calculations
with numerical probabilities, as opposed to thinking or belief formation in
general. It is, in my opinion, irrelevant to questions of rationality whether a bias
results from mere ignorance or from something else (such as, perhaps, unwillingness
to use rational methods even after learning how to use them). Rational methods are
those methods of thinking that help people achieve their own goals (or the goals they
ought to have). The most important of these methods are those that can be used in
all thinking. If an irrational bias that affects all thinking — such as the lack of active
open-mindedness — results from ignorance about rational thinking itself, the bias
is still irrational, because it prevents people from achieving their goals. Therefore,
the important part of Von Winterfeldt and Edwards’s argument is that the biases are
specific to numerical calculation.
That view holds for averaging but not for all the biases. Some of the ones that we
have observed reflect two central biases that prevent good thinking in general: insufficient
search, or the failure to consider alternative possibilities, goals, and additional
evidence; and favoritism toward initially favored possibilities. These central biases
seemed to be involved in overconfidence and in hindsight bias. They may be involved
in representativeness bias as well, since information about prior probabilities
could be a neglected source of evidence against initial conclusions. If people thought
more thoroughly about what evidence was needed, they could take prior probabilities
into account, whether or not they received special instruction in probability. Likewise,
if people thought critically about their own heuristics, by looking for cases in
which the heuristics are misleading, they could learn what these cases are and what
other heuristics are more useful. These central biases partly account for other more
specific biases, because they stand in the way of the learning that might correct them.
For most people, however, special instruction, as well as good thinking, is required
to learn about probability theory. Probability is already taught as part of
school mathematics. When our naive theories can have harmful effects, more systematic
instruction may be warranted. Such instruction could include such topics
as the relevance of prior probabilities and the dangers of hindsight, availability, and
extreme confidence.