A Test of Probability Model in Hypothesis Testing

• Main Authors: Yi-Chi Lin, 林奕祺
• Other Authors: Ruey-Yun Horng
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/12787553108399687989

碩士 === 國立交通大學 === 工業工程與管理系 === 90 === While testing a hypothesis, people have difficulty in using falsifying evidence. Kirby’s signal detection theory（1994）assumes that people can falsify, but the probability of P and Q in a rule “if P then Q” would affect hypothesis-testing. He predicts that tending to falsify would increase if the probability of P is low. In contrast, Oaksford and Chater’s optimal data selection theory（1996）doesn’t assume that people can falsify, they predict that people would use whatever information that more expected information gain. In this study, the probability of P or Q in Wason’s selection task was manipulated and the Subject’s selection of P or ~Q card in 20 selection tasks were compared with-or-without rule violation hint. The results show that subjects select more P cards when probability of P is low, and more ~Q card when probability of Q is high. This pattern of result lend the optimal data selection theory, namely, the hypothesis-testing process involves a inductive reasoning process：P and ~Q are regarded as samples of a population and are evaluated according to expected information gain of testing the sample. Besides, the rule violation hint also increased ~Q selection significantly. It indicates that people can falsify, but do not use it automatically. The lack of interaction between rule violation hint and probability of P or Q suggests that the heuristic regarding information gain and the falsification logic may affect hypothesis-testing independently.