Thinking and Deciding

172 HYPOTHESIS TESTING
hypothesis. The direct way is to try the white door. The indirect way is
to try the black door.
Similarly, when subjects use the successive-scanning strategy, they choose examples
consistent with their current hypothesis, rather than those that are inconsistent.
Inspired by this finding, Wason (1960, 1968b) told subjects that the sequence 2 4
6 followed a rule, which they were to discover by providing additional three-number
sequences as tests. The experimenter would tell the subject whether each of these
sequences followed the rule or not. One difference between this procedure and that
of Bruner and his colleagues is that both the number of possible hypotheses and the
number of possible sequences that could be used to test them were infinite. This fact
makes the situation more like those typically faced by scientists.
Typically, a subject would hypothesize that the rule was “numbers ascending by
2” and give sequences such as 3 5 7, 1.5 3.5 5.5, 100 102 104, and −6 −4 −2 as
a series of tests. If the subject followed this strategy — basically the successivescanning
strategy with direct tests — the experimenter would keep saying yes in
response to each test. Subjects always got the yes answer they expected, and they
never questioned their favored hypothesis. Many subjects, after several trials like
this, concluded that their hypothesis must be correct, and they announced that they
had discovered the rule. The “correct” rule (the rule that the experimenter followed in
giving feedback) was simply “ascending numbers.” Many subjects did discover this
rule. To do so, however, they had to use an indirect test of their original hypothesis.
Wason’s point was not just that subjects had difficulty discovering his rule (which,
because of its simplicity, might be considered a kind of trick). Instead, he suggested
that many subjects failed to consider alternative hypotheses, such as “numbers ascending
by a constant” or even “ascending numbers.” Wason’s conclusion is not
clearly established, however. Subjects may have been trying to test alternative hypotheses
for which the answer would be no (for example, ascending even numbers;
numbers ascending by 2 but less than 100; positive numbers ascending by 2; and so
forth), but they may have failed to mention these hypotheses. Also Wason did not
show that that subjects were violating a normative model. Perhaps, given what the
subjects were told, they were making reasonable assumptions and doing as well as
they could, given those assumptions. Nonetheless, Wason’s experiments inspired a
host of similar efforts.
The term “confirmation bias” has been used for this kind of behavior (Mynatt,
Doherty, and Tweney, 1977). The idea is that subjects were trying to confirm their
hypothesis rather than to test it or falsify it. This suggestion cannot be correct, if
taken literally. Most responses in these tasks could potentially falsify the hypothesis.
In the 2 4 6 problem, the test sequence 100 102 104 would falsify the hypothesis
“ascending even numbers” if the experimenter’s answer were no (meaning, “No, this
does not follow my rule”).
The subjects’ error is supposed to be that they fail to give sequences that don’t fit
their hypothesis, such as 2 3 5. This may be an error, but producing such sequences
is no more “trying to falsify” than is producing a sequence that fits. The sequence 2 3