Two-Way Fixed Effects and Difference-in-Differences Estimators with Heterogeneous Treatment Effects and Imperfect Parallel Trends DOI

Clément de Chaisemartin,

Xavier D’Haultfœuille

SSRN Electronic Journal, Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Two-way fixed effects (TWFE) regressions with period and group are widely used to estimate policies' effects: 26 of the 100 most cited papers published by American Economic Review from 2015 2019 such regressions. Researchers have long thought that TWFE estimators equivalent differences-in-differences (DID) estimators, rely on a partly testable parallel trends assumption. In two-groups two-periods designs where treatment is untreated at both dates becomes treated second period, coefficient in indeed DID. Motivated this fact, researchers also estimated more complicated many groups periods, variation timing, treatments switching off, and/or non-binary treatments, confident there as well, was giving them an estimation method only relied Two recent strands literature shattered confidence. First, it has recently been shown even if holds, may produce misleading estimates, policy's effect heterogeneous between or over time, often case. The realization one commonly empirical methods quantitative social sciences relies often-implausible assumption spurred flurry methodological papers. Some diagnosed issue analyzed its origins. Other proposed alternative relying conditions, like but robust effects, unlike estimators. Hereafter, those referred heterogeneity-robust DID Second, paper, Roth (2022) tests lack statistical power, fail detect differential control locations large enough account for significant share effect. This growing interest among practitioners strand literature, weaker assumptions than trends. Examples include conditional (see, e.g., Abadie, 2005), assuming bounded Manski Pepper, 2018; Rambachan Roth, 2023), factor model interactive Bai, 2003) synthetic Abadie et al., 2010), grouped patterns heterogeneity (see,e.g., Bonhomme Manresa, 2015).This textbook aims provide overview these two well other panel data routinely causal inference practitionners.

Language: Английский

From Access to Wellness: Early Life Exposure to Abortion Legalization and the Next Generation's Health DOI
Hamid Noghanibehambari, David Slusky, Hoa T. Vu

et al.

SSRN Electronic Journal, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

Language: Английский

Citations

0

Two-Way Fixed Effects and Difference-in-Differences Estimators with Heterogeneous Treatment Effects and Imperfect Parallel Trends DOI

Clément de Chaisemartin,

Xavier D’Haultfœuille

SSRN Electronic Journal, Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Two-way fixed effects (TWFE) regressions with period and group are widely used to estimate policies' effects: 26 of the 100 most cited papers published by American Economic Review from 2015 2019 such regressions. Researchers have long thought that TWFE estimators equivalent differences-in-differences (DID) estimators, rely on a partly testable parallel trends assumption. In two-groups two-periods designs where treatment is untreated at both dates becomes treated second period, coefficient in indeed DID. Motivated this fact, researchers also estimated more complicated many groups periods, variation timing, treatments switching off, and/or non-binary treatments, confident there as well, was giving them an estimation method only relied Two recent strands literature shattered confidence. First, it has recently been shown even if holds, may produce misleading estimates, policy's effect heterogeneous between or over time, often case. The realization one commonly empirical methods quantitative social sciences relies often-implausible assumption spurred flurry methodological papers. Some diagnosed issue analyzed its origins. Other proposed alternative relying conditions, like but robust effects, unlike estimators. Hereafter, those referred heterogeneity-robust DID Second, paper, Roth (2022) tests lack statistical power, fail detect differential control locations large enough account for significant share effect. This growing interest among practitioners strand literature, weaker assumptions than trends. Examples include conditional (see, e.g., Abadie, 2005), assuming bounded Manski Pepper, 2018; Rambachan Roth, 2023), factor model interactive Bai, 2003) synthetic Abadie et al., 2010), grouped patterns heterogeneity (see,e.g., Bonhomme Manresa, 2015).This textbook aims provide overview these two well other panel data routinely causal inference practitionners.

Language: Английский

Citations

5