Extracting Statistical Relationships from Observational Data: Predicting with Full or Partial Information
Abstract
Decision-makers sometimes rely on past data to learn statistical relationships between variables. However, when predicting a target variable, they must adjust how they aggregate past information depending on the observables available. If agents have information on all observables, it is optimal to understand how the observables jointly predict the target, while with only one observable, they should focus on the unconditional correlation. An experiment examining this process shows that predictions that require the use of unconditional correlations are more challenging for decision-makers.
Many UC-authored scholarly publications are freely available on this site because of the UC's open access policies. Let us know how this access is important for you.