arxiv: 2604.27035 · v1 · submitted 2026-04-29 · 💰 econ.EM

Recognition: unknown

Doubly robust local projections difference-in-differences

Daniel de Abreu Pereira Uhr, Guilherme Valle Moura

Pith reviewed 2026-05-07 11:41 UTC · model grok-4.3

classification 💰 econ.EM

keywords doubly robust estimationlocal projectionsdifference-in-differencesstaggered adoptionaverage treatment effectcausal inferenceeconometrics

0 comments

The pith

A doubly robust local projections difference-in-differences estimator stays consistent for the ATT if either the outcome regression or the treatment probability model is correct.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops DRLPDID as a doubly robust extension of local projections difference-in-differences for staggered absorbing treatments. This estimator targets the same local-stack average treatment effect on the treated as the original LP-DiD approach. It remains consistent whenever at least one of the two auxiliary models is correctly specified. The method also supplies influence-function inference for summary measures and multiplier-bootstrap bands for dynamic paths. Monte Carlo results show it performs like regression-adjusted LP-DiD when the outcome model is right and outperforms propensity weighting alone when that model is wrong, while an application to no-fault divorce laws produces results aligned with other robust estimators.

Core claim

The paper establishes that the doubly robust local projections difference-in-differences estimator, DRLPDID, preserves the LP-DiD local-stack ATT target and is consistent when either the local untreated-outcome regression or the local treatment-probability model is correctly specified.

What carries the argument

DRLPDID, the estimator that augments local projections DiD with both regression adjustment for untreated outcomes and inverse probability weighting for treatment assignment.

Load-bearing premise

At least one of the two local models for untreated outcomes or treatment probabilities is correctly specified, along with standard staggered DiD assumptions such as no anticipation and appropriate covariate conditioning.

What would settle it

A Monte Carlo design or real-data setting in which both the untreated-outcome regression and the treatment-probability model are misspecified, yet the estimator still recovers the true local ATT without bias.

Figures

Figures reproduced from arXiv: 2604.27035 by Daniel de Abreu Pereira Uhr, Guilherme Valle Moura.

**Figure 1.** Figure 1: Dynamic event-study comparisons in the no-fault-divorce application view at source ↗

read the original abstract

This paper develops a doubly robust extension of local-projections difference-in-differences (LP-DiD) for staggered absorbing treatments. The resulting estimator, DRLPDID, preserves the LP-DiD local-stack ATT target and is consistent when either the local untreated-outcome regression or the local treatment-probability model is correctly specified. It also delivers influence-function-based inference for post-treatment summaries and multiplier-bootstrap bands for dynamic paths. In Monte Carlo designs with covariate-driven selection, DRLPDID matches regression-adjusted LP-DiD under outcome-model alignment and clearly outperforms the IPT-only variant under propensity-score misspecification. In the no-fault-divorce application, DRLPDID tracks robust staggered-adoption estimators and is less negative than unadjusted LP-DiD.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

DRLPDID adds double robustness to LP-DiD for staggered treatments while keeping the original local ATT target and delivering influence-function inference.

read the letter

DRLPDID keeps the LP-DiD local-stack ATT target but becomes consistent if either the local untreated-outcome regression or the local treatment-probability model is correct. That is the central claim, and the Monte Carlo results back it up under covariate-driven selection: the estimator matches the regression-adjusted version when the outcome model is right and beats the IPT-only version when the propensity model is wrong. The no-fault-divorce application produces estimates that line up with other robust staggered DiD methods and are less negative than plain LP-DiD. The inference tools—influence functions for summaries and multiplier bootstrap for paths—are standard but useful additions for dynamic settings. The double-robustness property follows the usual regression-plus-weighting cancellation, with no visible extra assumptions beyond the standard staggered DiD conditions of no anticipation and proper covariate conditioning. The local stacking and absorbing-treatment structure do not appear to break the argument in the reported designs. One practical limitation is that practitioners still need to specify the two local models carefully, especially with many post-treatment horizons, and the paper does not explore how sensitive results are to those choices in finite samples. The work is aimed at applied economists who run staggered-adoption designs and want dynamic treatment-effect estimates that are less fragile to one model being off. Readers who already use LP-DiD or other robust DiD estimators will see the most direct value. The paper shows clear, incremental thinking that builds on existing LP-DiD and doubly robust literature without internal contradictions. It deserves a serious referee.

Referee Report

2 major / 2 minor

Summary. The paper develops a doubly robust extension of local-projections difference-in-differences (LP-DiD) for staggered absorbing treatments. The resulting DRLPDID estimator preserves the original LP-DiD local-stack ATT target and is consistent when either the local untreated-outcome regression or the local treatment-probability model is correctly specified. It supplies influence-function-based inference for post-treatment summaries and multiplier-bootstrap bands for dynamic paths. Monte Carlo designs with covariate-driven selection and an application to no-fault divorce laws are used to illustrate performance relative to regression-adjusted LP-DiD and IPT-only variants.

Significance. If the double-robustness property and target preservation hold, the estimator would provide a practical robustness upgrade for LP-DiD users facing potential misspecification in either the outcome or propensity model under standard staggered DiD assumptions. The Monte Carlo results and application supply concrete evidence of behavior under covariate selection, and the influence-function plus bootstrap inference is a clear implementation strength.

major comments (2)

[§3] §3 (estimator construction): the explicit form of the DRLPDID influence function and the bias-cancellation argument for double robustness should be derived in full, showing how the regression residual and IPW terms combine to yield consistency under either correct model without additional assumptions beyond those stated for LP-DiD.
[Application section] Application section: the local untreated-outcome regression and treatment-probability specifications appear chosen after inspecting the data; pre-specification or a sensitivity table varying the covariate sets and functional forms would strengthen the claim that DRLPDID is less negative than unadjusted LP-DiD.

minor comments (2)

[Introduction] The notation distinguishing the local-stack ATT from the standard ATT could be introduced earlier and used consistently in the abstract and introduction.
[Monte Carlo] Monte Carlo tables would benefit from reporting the exact functional forms used for the outcome and propensity models in each design to allow direct replication of the misspecification cases.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading, positive assessment, and constructive suggestions. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [§3] §3 (estimator construction): the explicit form of the DRLPDID influence function and the bias-cancellation argument for double robustness should be derived in full, showing how the regression residual and IPW terms combine to yield consistency under either correct model without additional assumptions beyond those stated for LP-DiD.

Authors: We agree that greater explicitness in the derivation will strengthen the section. In the revised manuscript we will expand §3 to present the full closed-form expression for the DRLPDID influence function. We will also walk through the bias-cancellation argument in detail, showing algebraically how the regression-residual term and the IPW term jointly eliminate first-order bias when either the local untreated-outcome regression or the local treatment-probability model is correctly specified, under no assumptions beyond those already maintained for the original LP-DiD estimator. revision: yes
Referee: [Application section] Application section: the local untreated-outcome regression and treatment-probability specifications appear chosen after inspecting the data; pre-specification or a sensitivity table varying the covariate sets and functional forms would strengthen the claim that DRLPDID is less negative than unadjusted LP-DiD.

Authors: We acknowledge that the covariate sets and functional forms used in the application were informed by economic reasoning and preliminary data inspection. To address the concern, the revised application section will include a sensitivity table that reports DRLPDID estimates under alternative covariate specifications (different subsets, with and without interactions, linear versus flexible forms). This table will confirm that the qualitative finding—DRLPDID estimates are less negative than unadjusted LP-DiD—remains stable across these choices. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper's central result is the construction of a doubly robust estimator DRLPDID that combines an outcome regression term with an inverse-probability-weighted residual to achieve consistency for the LP-DiD local-stack ATT target whenever at least one of the two nuisance models is correctly specified. This is the standard double-robustness property obtained by algebraic cancellation of bias terms under the maintained staggered DiD assumptions (no anticipation, covariate conditioning). No equation reduces the target ATT or the consistency claim to a fitted quantity by construction, no self-citation supplies a load-bearing uniqueness theorem, and no ansatz is smuggled in. The Monte Carlo and application sections compare the estimator to its components without redefining the target parameter. The derivation therefore stands on independent statistical arguments rather than tautological re-labeling of inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard econometric assumptions for consistency in difference-in-differences and local projections; no free parameters or invented entities are mentioned in the abstract.

axioms (1)

domain assumption Standard no-anticipation and parallel trends assumptions conditional on covariates for staggered absorbing treatments
Invoked to preserve the LP-DiD local-stack ATT target under double robustness.

pith-pipeline@v0.9.0 · 5421 in / 1205 out tokens · 34921 ms · 2026-05-07T11:41:34.524773+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

15 extracted references · 1 canonical work pages

[1]

Girardi, \`O

Dube, A., D. Girardi, \`O. Jord\`a, A. M. Taylor, 2025. A local projections approach to difference-in-differences. Journal of Applied Econometrics 40, 741--758

2025
[2]

diff-diff: Difference-in-differences in Python

diff-diff contributors, 2026. diff-diff: Difference-in-differences in Python. Python package documentation. https://diff-diff.readthedocs.io/

2026
[3]

Difference-in-differences with variation in treatment timing

Goodman-Bacon, A., 2021. Difference-in-differences with variation in treatment timing. Journal of Econometrics 225, 254--277

2021
[4]

Graham, B. S., C. C. Pinto, D. Egel, 2012. Inverse probability tilting for moment condition models with missing data. Review of Economic Studies 79, 1053--1079

2012
[5]

Sant'Anna, P. H. C., J. Zhao, 2020. Doubly robust difference-in-differences estimators. Journal of Econometrics 219, 101--122

2020
[6]

Wolfers, 2006

Stevenson, B., J. Wolfers, 2006. Bargaining in the shadow of the law: Divorce laws and family distress. Quarterly Journal of Economics 121, 267--288

2006
[7]

Jaravel, J

Borusyak, K., X. Jaravel, J. Spiess, 2024. Revisiting event-study designs: Robust and efficient estimation. Review of Economic Studies 91, 3253--3285

2024
[8]

Callaway, B., P. H. C. Sant'Anna, 2021. Difference-in-differences with multiple time periods. Journal of Econometrics 225, 200--230

2021
[9]

D'Haultf uille, 2020

de Chaisemartin, C., X. D'Haultf uille, 2020. Two-way fixed effects estimators with heterogeneous treatment effects. American Economic Review 110, 2964--2996

2020
[10]

D'Haultf uille, 2023

de Chaisemartin, C., X. D'Haultf uille, 2023. Two-way fixed effects and difference-in-differences with heterogeneous treatment effects: A survey. The Econometrics Journal 26, C1--C30

2023
[11]

Two-stage differences in differences

Gardner, J., 2022. Two-stage differences in differences. arXiv preprint arXiv:2207.05943

work page arXiv 2022
[12]

Discussion of ``An Essay on the Logical Foundations of Survey Sampling, Part I'', by D

H \'a jek, J., 1971. Discussion of ``An Essay on the Logical Foundations of Survey Sampling, Part I'', by D. Basu. In: Godambe, V.P., Sprott, D.A. (Eds.), Foundations of Statistical Inference. Holt, Rinehart and Winston, Toronto, p. 326

1971
[13]

Estimation and inference of impulse responses by local projections

Jord \'a , \`O., 2005. Estimation and inference of impulse responses by local projections. American Economic Review 95, 161--182

2005
[14]

Sant'Anna, A

Roth, J., M. Sant'Anna, A. Bilinski, J. Poe, 2023. What’s trending in difference-in-differences? A synthesis of the recent econometrics literature. Journal of Econometrics 235, 2218--2244

2023
[15]

Abraham, 2021

Sun, L., S. Abraham, 2021. Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics 225, 175--199

2021