Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 68 of the 68 references it cites were found in a registry or an index; 16 of the 153 sources its claims cite were checked for support.

Two-Way Fixed Effects and Differences-in-Differences with Heterogeneous Treatment Effects: A Survey

Open access
de Chaisemartin, Clément, D'Haultfoeuille, Xavier · 2022 · National Bureau of Economic Research

Abstract

Linear regressions with period and group fixed effects are widely used to estimate policies' effects: 26 of the 100 most cited papers published by the American Economic Review from 2015 to 2019 estimate such regressions. It has recently been shown that those regressions may produce misleading estimates, if the policy's effect is heterogeneous between groups or over time, as is often the case. This survey reviews a fast-growing literature that documents this issue, and that proposes alternative estimators robust to heterogeneous effects. We use those alternative estimators to revisit Wolfers (2006a).

Overview

CoChat Summary

TL;DR

This survey explains why two-way fixed effects (TWFE) regressions can give misleading treatment effects when effects vary across groups or over time, and it reviews alternative difference-in-differences estimators that are robust to heterogeneity. It also re-analyzes Wolfers (2006a) on unilateral divorce laws and shows that several modern estimators give broadly similar conclusions, though precision and long-run estimates can differ.

Key findings

  • In a survey of the 20 most cited AER papers from each of 2015–2019, 26 of 100 papers estimated at least one TWFE regression for a treatment effect.
  • TWFE coefficients can place negative weights on some group-time treatment effects; the paper gives a minimum-wage example where the coefficient could be positive even if the treatment effect is negative in every group-time cell.
  • In the Gentzkow et al. (2011) newspaper-turnout example, the paper reports DID M = 0.0043 (s.e. = 0.0014), equivalent to a 0.43 percentage point increase in turnout from one more newspaper, and says this is 66% larger than the reported first-difference FE estimate.
  • In the Wolfers (2006a) application on unilateral divorce laws, the event-study estimates imply divorce rates rise in the year of adoption and for seven years thereafter, then become significantly negative 11 years after adoption; the survey reports placebo estimates are small and, in the main specifications, jointly insignificant (e.g. F-test p-value = 0.863 for the eventstudy regression, 0.120 for did_imputation, and 0.427 for did_multiplegt).

Methods

  • The paper is a survey and methodological review of TWFE and DID estimators under heterogeneous treatment effects, focused on parallel-trends-based estimators and on when dynamic effects are or are not allowed.
  • It reviews decomposition results for TWFE regressions, then summarizes alternative estimators including DID M, Callaway and Sant’Anna (2021), Sun and Abraham (2021), Borusyak et al. (2021), and de Chaisemartin and D'Haultfœuille (2021a, 2022).
  • It reanalyzes Wolfers (2006a) using the authors’ own and related Stata/R commands, with estimates weighted by states’ populations and standard errors clustered at the state level.

Limitations

  • The paper explicitly says the literature it reviews is very recent, and that several papers are still working papers, so some findings have not yet been peer reviewed.
  • The authors state that whether researchers should systematically abandon TWFE estimators is still unclear; TWFE can sometimes estimate a convex combination of effects or have lower variance, depending on the setting.
  • Evident from the text: the paper’s empirical reanalysis is only of one application (Wolfers 2006a), so its application-level comparison is illustrative rather than a broad systematic benchmark.

Model-written summary of a shortened version of the paper’s full text (it is longer than we read). Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3386/w29734

Cites

68 of 68 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 A numerical example with three periods, an early and a late treated group
Figure 1 A numerical example with three periods, an early and a late treated groupFrom the PDF
Figure 2 A numerical example with two periods, a more- and a less-treated group
Figure 2 A numerical example with two periods, a more- and a less-treated groupFrom the PDF
Figure 3 Effects of Unilateral Divorce Laws, using the data in Wolfers (2006a)
Figure 3 Effects of Unilateral Divorce Laws, using the data in Wolfers (2006a)From the PDF

Verification

Claims the paper makes

12 Valid · 6 Not valid · 98 Not checked · 118 Own results and interpretations

127 more claims · 5 with their cited sources checked

Create a free account to see all 234 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 68
  • Probably not a substantial claim. 46

Create a free account to see these claims.

Citations

Cites

Cited by

See all 70

Metadata

Authors
de Chaisemartin, C., & D'Haultfoeuille, X.
Venue
National Bureau of Economic Research
Published
2022-02
OpenAlex
W4213318018
Cited by
70
References
57
Citations by year
2026: 8 · 2025: 16 · 2024: 19 · 2023: 20 · 2022: 7
Type
report
Language
en
Open access
Locations
National Bureau of Economic Research (publishedVersion)
Topics
  • Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Spatial and Panel Data Analysis
  • Social Sciences › Public Administration › Evaluation and Performance Assessment
Keywords
two-way fixed effects, difference-in-differences, treatment effect heterogeneity, policy evaluation, robust estimators, causal inference

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.