Teaching by examples: The case of number series

Subjects' performance in Wason's rule discovery task has long been known to deviate from the norms advocated by philosophers of science. For a better understanding of people's rule discovery behaviour we employed a teaching‐by‐ examples paradigm, in which subjects were told to generate examples in order to teach someone else a rule. Subjects were aware of the rule and of the tutee's hypothesis; it was therefore possible to classify each example they provided as either a positive or a negative instance of the rule, and as either a confirmation or a refutation of the tutee's hypothesis. The set‐theoretical relationship between the tutee's hypothesis and the actual rule (embedded, surrounding) was manipulated in order to assess subjects' use of different types of disconfirmation. The task was arranged in such a way that the discrepancy between the tutee's hypothesis and the actual rule in the embedded and the surrounding conditions consisted of an identical set. The results (N = 96) revealed a high incidence of negative instances and of disconfirmations. Disconfirmations were more readily provided in the embedded condition (through negative tests of the hypothesis, using positive examples of the rules) than in the surrounding condition (through positive tests, using negative examples), thus lending support to positivity bias theory (Evans, 1989). The results indicate that subjects' poor performance in rule discovery tasks does not stem from inability to appreciate or use disconfirmations and negative examples, but from failure to consider alternatives. 1995 The British Psychological Society