Difference-in-differences with stochastic policy shifts of continuous treatments

2025-11-28

AI Summary

This paper addresses the estimation of causal effects under stochastic policy interventions with continuous treatment variables. Departing from conventional assumptions of conditional independence (unconfoundedness), it operates within a difference-in-differences (DiD) framework that relies solely on the parallel trends assumption. The authors introduce the *Average Random Dose Effect* (ARDE)—a novel causal parameter quantifying how shifts in the treatment distribution causally affect outcomes. They model stochastic interventions as data-dependent exponential tilting transformations of the observed treatment distribution—a first in the literature—and develop a nonparametrically efficient, doubly robust estimator by integrating DiD identification with modern causal inference techniques. The estimator adaptively learns nuisance functions via machine learning, achieving √n-consistency, asymptotic normality, and semiparametric efficiency. An empirical application to hydraulic fracturing policies demonstrates the method’s robustness and superior performance in estimating impacts on employment and income.

Technology Category

• Reasoning under Uncertainty: Stochastic Optimization
• Machine Learning: Causal Learning
• Intelligent Robots: State Estimation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Treatment effects under stochastic policy shifts quantify differences in outcomes across counterfactual scenarios with varying treatment distributions. Stochastic policy shifts generalize common notions of treatment effects since they include deterministic interventions (e.g., all individuals treated versus none treated) as a special case. While stochastic policy effects have been examined under causal exchangeability, they have not been integrated into the difference-in-differences (DiD) framework, which relies on parallel trends rather than exchangeability. In this paper, nonparametric efficient estimators of stochastic intervention effects are developed under a DiD setup with continuous treatments. The proposed causal estimand is the average stochastic dose effect among the treated, where the stochastic dose effect is the contrast between potential outcomes under a counterfactual dose distribution and no treatment. Several possible stochastic interventions are discussed, including those that do and do not depend on the observed data distribution. For generic stochastic interventions, the causal estimand is identified under standard conditions and estimators are proposed. Then, we focus on a specific stochastic policy shift, the exponential tilt, that increments the conditional density function of the continuous dose. For the exponential tilt intervention, a nonparametric estimator is proposed that allows for data-adaptive, machine learning nuisance function estimation. Under mild convergence rate conditions, the estimator is shown to be root-$n$ consistent and asymptotically normal with variance attaining the nonparametric efficiency bound. The proposed method is used to study the effect of hydraulic fracturing activity on employment and income.

Problem

• Estimates stochastic policy effects in difference-in-differences with continuous treatments.
• Develops nonparametric estimators for average stochastic dose effects among treated units.
• Applies method to study hydraulic fracturing's impact on employment and income.

Innovation

• Nonparametric efficient estimators for stochastic policy shifts
• Difference-in-differences with continuous treatments integration
• Exponential tilt intervention using machine learning estimation