Designing Difference in Difference Studies with Staggered Treatment Adoption: Key Concepts and Practical Guidelines

Working Paper: NBER ID: w31842

Authors: Seth M. Freedman; Alex Hollingsworth; Kosali I. Simon; Coady Wing; Madeline Yozwiak

Abstract: Difference-in-Difference (DID) estimators are a valuable method for identifying causal effects in the public health researcher’s toolkit. A growing methods literature points out potential problems with DID estimators when treatment is staggered in adoption and varies with time. Despite this, no practical guide exists for addressing these new critiques in public health research. We illustrate these new DID concepts with step-by-step examples, code, and a checklist. We draw insights by comparing the simple 2 × 2 DID design (single treatment group, single control group, two time periods) with more complex cases: additional treated groups, additional time periods of treatment, and with treatment effects possibly varying over time. We outline newly uncovered threats to causal interpretation of DID estimates and the solutions the literature has proposed, relying on a decomposition that shows how the more complex DID are an average of simpler 2X2 DID sub-experiments.

Keywords: Difference-in-Differences; Staggered Treatment Adoption; Causal Inference; Public Health

JEL Codes: I01; I11


Causal Claims Network Graph

Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.


Causal Claims

CauseEffect
TWFE model under certain conditions (C69)consistent estimates of treatment effects (C22)
staggered adoption (L15)confounding comparisons in TWFE regressions (C29)
time-varying treatment effects (C32)confounding comparisons in TWFE regressions (C29)
staggered adoption (L15)differences in treatment effects based on timing of adoption (C22)
nonanticipation assumption + common trends assumption (C22)valid causal inference in DiD (C22)

Back to index