Abstract

In testing hypotheses, people often perform inefficiently because of confirmation bias . In the present research, various ways to reduce confirmation bias and improve hypothesis-testing performance were investigated by having subjects perform an analog scientific inference task under different instructional conditions that were modeled after various philosophies of science. ;In the task , subjects tested hypotheses in order to ascertain predetermined rules governing sequences of card plays. In Experiment 1, subjects performed under instructions to disconfirm their hypotheses , under instructions to generate multiple hypotheses , under both of the previous instructions , or in the presence of a partner who was instructed to attempt to disprove the subject's hypotheses . All groups except the Multiple-Hypothesis group showed reduced rates of confirmatory plays when compared to a no-strategy control group, but only the combined Disconfirm/Multiple-Hypothesis group and the Kibitzer group discovered significantly more rules than the control group. ;Experiment 2 tested the effect of combining the kibitzer manipulation with the disconfirmatory instructions. The Disconfirm, Kibitzer, and combined Disconfirm/Kibitzer groups all showed reduced rates of confirmatory plays and discovered more rules than a no-strategy group, but did not differ from one another. All subjects in this experiment performed the task with a partner, ruling out interpretation of the kibitzer effect as due to coaction or audience effects. Post-hoc data analyses suggested that kibitzers enhanced subjects' performance by playing cards that were more informative about the underlying rule, and perhaps by reducing subjects' confidence in their hypotheses. ;In both experiments, the presence of a kibitzer more than doubled the rates of rule discovery, supporting the pluralist notion that the presence of multiple, mutually critical research traditions in science aids the search for knowledge, in part by providing a corrective for confirmation bias. The results also suggest that the strategies advocated by Popper and Platt can be beneficial