Abstract

Forty years of experimentation on class inclusion and its probabilistic relatives have led to inconsistent results and conclusions about human reasoning. Recent research on the conjunction "fallacy" recapitulates this history. In contrast to previous results, we found that a majority of participants adhere to class inclusion in the classic Linda problem. We outline a theoretical framework that attributes the contradictory results to differences in statistical sophistication and to differences in response mode-whether participants are asked for probability estimates or ranks-and propose two precise cognitive algorithms for ranking probabilities. Our framework allows us to make novel predictions about when and why people adhere to class inclusion. Evidence obtained in several studies supports these predictions and demonstrates that the proposed ranking algorithms can account for about three-quarters of participants' inferences in the Linda problem.