Proceedings of the Annual Meeting of the Cognitive Science Society

Although the causal graphical model framework has achievedconsiderable success accounting for causal learning data, appli-cation of that formalism to multi-cause situations assumes thatpeople are insensitive to the statistical properties of the causesthemselves. The present experiment tests this assumption byfirst instructing subjects on a causal model consisting of twoindependent and generative causes and then requesting them tomake data likelihood judgments, that is, to estimate the proba-bility of some data given the model. The correlation betweenthe causes in the data was either positive, zero, or negative. Thedata was judged as most likely in the positive condition andleast likely in the negative condition, a finding that obtainedeven though all other statistical properties of the data (e.g.,causal strengths, outcome density) were controlled. These re-sults pose a problem for current models of causal learning.