Machine Learning about Treatment Effect Heterogeneity: The Case of Household Energy Use

We use causal forests to evaluate the heterogeneous treatment effects (TEs) of repeated behavioral nudges toward household energy conservation. The average response is a monthly electricity reduction 9 kilowatt-hours (kWh), but full distribution responses ranges from -40 +10 kWh. Households learn reduce more over time, conditional on having responded in year one. Pre-treatment consumption and home value are most commonly used predictors forest. results suggest ability machine learning techniques for improved targeting tailoring treatment.