Testing and interpreting the effectiveness of causal machine learning - an economic theory approach - HEC Paris

Information Systems and Operations Management Department

Speaker: Kevin Bauer (Goethe University Frankfurt)

Room Bernard Ramanantsoa

When individual treatment effects vary, personalized treatment assignments (targeting) can substantially enhance the effectiveness of interventions. This paper analyzes if causal machine learning (CML) methods applied to basic web data (digital footprints) can be used for targeting and why they are effective. We combine data from a loss framing treatment within a large-scale field experiment at an online fashion store with experimental measures of consumers' individual loss aversion. We demonstrate that CML based targeting could transform an otherwise ineffective marketing campaign into one that generates a six-figure revenue increase. Our findings also reveal that CML-based targeting is consistent with economic theory, as the predicted individual treatment effects show a significant correlation with measures of loss aversion. Our results demonstrate the effectiveness of CML methods for targeting treatment assignment and how (behavioral) economic theory and the measurement of structural preference parameters can improve their interpretability and transparency.