Learning from experience in nonlinear environments: Evidence from a competition scenario

•In our experiments people assess probabilities of success in a competition scenario.•Our participants exhibit systematic errors in their judgments.•More experience through outcome feedback leads to little or no learning.•We make the problem “decision-friendly” by reformulating it in a more linear format.•People learn better with the new format, as it is consistent with their prior beliefs.

We test people’s ability to learn to estimate a criterion (probability of success in a competition scenario) that requires aggregating information in a nonlinear manner. The learning environments faced by experimental participants are kind in that they are characterized by immediate, accurate feedback involving either naturalistic outcomes (information on winning and/or ranking) or the normatively correct probabilities. We find no evidence of learning from the former and modest learning from the latter, except that a group of participants endowed with a memory aid performed substantially better. However, when the task is restructured such that information should be aggregated in a linear fashion, participants learn to make more accurate assessments. Our experiments highlight the important role played by prior beliefs in learning tasks, the default status of linear aggregation in many inferential judgments, and the difficulty of learning in nonlinear environments even in the presence of veridical feedback.