Negative observations, induction and the generation of hypotheses

In category-based induction tasks, it is a robust finding that positive observations raise the judged likelihood of a conclusion and negative observations lower judged likelihood. We present evidence that negative observations can raise the judged likelihood. In particular, we asked participants to judge the likelihood of a conclusion after introducing them to different sets of premises either containing one positive observation or the same positive observation and a negative observation. We found that when the negative observation is dissimilar to the positive observation, willingness to accept a conclusion is raised. Moreover, results from a simultaneous hypothesis generation task suggest that the rise in judged conclusion likelihood is due to a peculiar shift in the hypothesis space of the reasoner, in that the hypothesis with the largest extension, yet still consistent with all premises gains disproportionate popularity when introducing dissimilar negative observations.