Litcius/Paper detail

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

Christopher R. Knittel, Samuel Stolper

2021AEA Papers and Proceedings61 citationsDOIOpen Access PDF

Abstract

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

Topics & Concepts

Nudge theoryConsumption (sociology)Energy (signal processing)ElectricityEnergy conservationValue (mathematics)Average treatment effectDistribution (mathematics)Treatment effectEconomicsEnvironmental economicsComputer sciencePsychologyStatisticsMachine learningMedicineEngineeringMathematicsSocial psychologySociologyTraditional medicineMathematical analysisEstimatorElectrical engineeringSocial scienceEnvironmental Education and SustainabilityEconomic and Environmental ValuationEnergy and Environment Impacts
Machine Learning about Treatment Effect Heterogeneity: The Case of Household Energy Use | Litcius