Working Paper: NBER ID: w31184
Authors: Arindrajit Dube; Daniele Girardi; Oscar Jord; Alan M. Taylor
Abstract: We propose a local projection (LP) based difference-in-differences approach that subsumes many of the recent solutions proposed in the literature to address possible biases arising from negative weighting. We combine LPs with a flexible ‘clean control’ condition to define appropriate sets of treated and control units. Our proposed LP-DiD estimator can be implemented with various weighting and normalization schemes for different target estimands, can be extended to include covariates or accommodate non-absorbing treatment, and is simple and fast to implement. A simulation and two empirical applications demonstrate that the LP-DiD estimator performs well in common applied settings.
Keywords: Local Projections; Difference-in-Differences; Event Studies; Treatment Effects; Bias Mitigation
JEL Codes: C1; C23; C5
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| LP-DiD estimator (C51) | performs well in simulations compared to existing estimators (C51) |
| LP-DiD estimator (C51) | unbiased estimate of treatment effects (C90) |
| LP-DiD estimator (C51) | avoids biases associated with TWFE models (C51) |
| LP-DiD methodology (C69) | convex weighted average of heterogeneous cohort-specific treatment effects (C21) |
| LP-DiD estimator (C51) | valid estimates of impact of banking deregulation on labor share (E69) |
| LP-DiD estimator (C51) | valid estimates of effect of democratization on economic growth (O11) |
| LP-DiD approach (C69) | valid estimates even when parallel trends assumption holds conditionally on pretreatment dynamics (C22) |