Thinking and Deciding

CLASSIFICATION 383
more eggs. ‘How many of these schools are approved for veterans benefits?’
The announcer continues until only the walnut remains. The
announcer cracks the nutshell, which reveals the name of the company,
and concludes: ‘This is all you need to know in a nutshell.’ (Tversky,
1972, p. 287)
Other simple heuristics
A great many other heuristics have been proposed, and sometimes people have been
shown to use them. One simple example is the recognition heuristic (Goldstein and
Gigerenzer, 2002), which is to decide on the basis of sheer recognition. If you are
asked which of two cities is larger, and you have heard of one but not the other, you
do well to pick the former.
Classification
Suppose we change the task slightly. Instead of looking at a policy already formed
over years of experience (as our personal method of assessing personality is), we
require the subject to learn to make judgments during the experiment. On each trial,
subjects make a prediction and are then given feedback about the actual value of
the criterion they were trying to predict. The cue and criterion variables can now
be entirely artificial, with labels such as x and y instead of GPA and SAT. When
this procedure is used, subjects still make their judgments in terms of linear rules
(Klayman, 1984). In general, if the system in question obeys linear rules, subjects
will learn them about as well (or as badly) as they do in real life.
Suppose we make one more change. Instead of the variables being continuous
quantities, with each variable taking several possible values, we use variables with
only two possible values each. For example, let us assume that from such variables
as a person’s hair color (light or dark), shirt color (light or dark), hair length (short
or long), and smile (open or closed) we are supposed to guess whether the person’s
name is Asch or not (Medin, Dewey, and Murphy, 1983). (Clearly, this is a very
artificial task.) Once again, the results do not seem to change much (Smith and
Medin, 1981). Subjects’ judgments can be predicted fairly well on the basis of a
linear combination of features.
In this kind of task, though, certain results suggest that something else is going
on. These results may have implications for judgment research as a whole. The fictional
findings from a medical case study given in Table 15.3 were used in an experiment
by Medin, Altom, Edelson, and Freko (1982, p. 39). Subjects were presented
with these cases as examples, one by one, and were told that the initials represented
patients with “burlosis,” a fictitious disease. In Table 15.3, 1 represents the presence
of a symptom, and 0 represents its absence. Notice that the last two symptoms, discolored
gums and nose bleed, are perfectly correlated. Otherwise, all symptoms are
equally useful. The same number of patients has each of the four symptoms.