Confirmation and Matching Biases in Hypothesis Testing

Abstract

A number of investigators have claimed that subjects show confirmation bias on a variety of reasoning tasks. However, subjects who were classified as “confirming” had often selected evidence that could have falsified their hypothesis. In this study we present a novel task in which each piece of evidence is either falsifying or confirming, but not both. This task is similar in structure to Wason's (1966) 4-card task and its negated variant (Evans & Lynch, 1973); hence subjects were asked to perform these two tasks for comparison purposes. Some subjects also provided thinking-aloud protocols, allowing a test of the Johnson-Laird and Wason (1970) information processing model. Subjects were found who showed severe confirmation bias by selecting only evidence that could corroborate, but not falsify, their hypotheses. Several subjects even retained their hypotheses when presented with clearly falsifying evidence. The tendency to show confirmation bias was significantly greater in first-year undergraduates than in more educated subjects. A revised definition of confirmation bias is offered to clarify subjects’ reasons for their selections. In addition to evidence of confirmation bias, support was found for the matching bias model (Wason & Evans, 1975; Evans, 1977), emphasizing the need for sensitivity to individual differences.