Continuous Difference-in-Differences with Double/Debiased Machine Learning

Difference-in-differences (DiD) is a widely used research design for estimating causal effects, typically in settings with binary or discrete treatments. Many empirical questions, however, involve policies and interventions that are continuous in nature. Individuals may experience different levels of exposure to marketing campaigns, varying benefit amounts, or differing concentrations of environmental pollutants, all of which are better modeled as continuous treatments.

In “Continuous Difference-in-Differences with Double/Debiased Machine Learning,” Lucas Zhang extends the DiD framework to the continuous treatments setting. The paper focuses on identifying and estimating the average treatment effects on the treated at any given treatment intensity, incorporating covariates nonparametrically in both the identification and estimation steps through the double/debiased machine learning framework.

This extension substantially broadens the scope of DiD designs, providing researchers with new tools to examine the impacts of continuous treatments while flexibly accounting for covariates.

The article is available here (subscription required).