Difference-in-Differences Estimators of Intertemporal Treatment Effects

Causal Question / Estimand

Dynamic treatment effects when the treatment may be non-binary, non-absorbing (can switch on and off), and outcomes respond to treatment lags. The target is the effect of having been exposed to a weakly higher treatment dose for periods — an event-study/intertemporal effect (Causal-Estimand).

Identification Strategy

Under a Parallel-Trends assumption, proposes event-study estimators for the effect of extra periods of (weakly higher) exposure, plus normalized estimators that recover a weighted average of the current treatment’s effect and its lags. Designed for general treatment paths where the cohort/absorbing-treatment estimators (Callaway–Sant’Anna, Sun–Abraham) do not directly apply.

Key Assumptions

Parallel-Trends (for the relevant exposure contrasts), No-Anticipation, a stable comparison group of units whose treatment does not change, and SUTVA.

Threats to Validity

The paper’s foil: standard TWFE regressions are biased under heterogeneous effects; worse, a local-projection version of those regressions is biased even with homogeneous effects. Requires “stayers” (units with unchanged treatment) to serve as controls; thin in some designs.

Setting / Data

n/a — econometric methodology (the did_multiplegt_dyn toolkit). Illustrated with panel applications featuring switching treatments.

Key Claims

  • Provides DiD estimators valid for non-binary, non-absorbing treatments with dynamic effects.
  • TWFE and especially its local-projection variant are biased; the proposed event-study/normalized estimators are robust to Treatment-Effect-Heterogeneity.
  • Distinguishes “exposure for periods” effects from contemporaneous effects.

Connections

Citation

de Chaisemartin, C., & D’Haultfœuille, X. (2024). Difference-in-Differences Estimators of Intertemporal Treatment Effects. The Review of Economics and Statistics (first version 2020). https://doi.org/10.1162/rest_a_01414