Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects

Key Points

• Determine whether linear regressions with group and period fixed effects yield valid treatment effect estimates under heterogeneous effects, and propose a robust alternative estimator.
• Decomposed the standard two-way fixed effects linear regression coefficient into a weighted sum of cell-specific average treatment effects.
• Developed a novel econometric estimator designed to eliminate negative weighting issues.
• Re-evaluated two published empirical applications to benchmark the new estimator against traditional linear fixed effects models.
• Demonstrated that standard two-way fixed effects regressions assign weights to subgroup average treatment effects that can be negative, allowing the overall coefficient to be negative even when all individual average treatment effects are positive.
• Showed that estimates from the proposed alternative estimator differed significantly from standard linear regression estimates across both empirical applications.