.png){kind=icon}
.png){kind=icon}
Return
Simple Transformation Approach to Difference-in-Differences Estimation for Panel Data
.png){kind=icon}
PRE
AI
DOI：10.1080/07350015.2026.2683047.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In the case of panel data with staggered interventions, we propose a simple time-series transformation that can be combined with various treatment effect estimators. The transformation is at the unit level, and simply requires computing the average outcome prior to an intervention, subtracting it from a post-treatment outcome, and then carefully selecting the control units in each time period. We show that under no anticipation and parallel trends assumptions, the cohort treatment indicators satisfy the key unconfoundedness assumption with respect to the transformed potential outcome. Given identification, any number of treatment effect estimators, including matching estimators, can be applied for each treated cohort and calendar time pair where the average treatment effects on the treated are identified. The doubly robust method of combining inverse probability weighting with linear regression adjustment works particularly well in terms of bias and efficiency. Importantly, our transformation is easily modified to account for unit-specific trends, allowing relaxation of the parallel trends assumption even after conditioning on covariates.
Keywords:
Difference-in-differences
Doubly robust estimators
Heterogeneous trends
Panel data
Parallel trends
Journal
J
IF:
2.5
Papers:
79
Citations:
0
