A Meta-learner for Heterogeneous Effects in Difference-in-Differences

Eleanor Dillon; Haoge Chang; Hui Lan; Vasilis Syrgkanis

REVIEW 2 major objections 4 minor 1 cited by

A Meta-learner for Heterogeneous Effects in Difference-in-Differences

T0 review · 2 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that heterogeneous difference-in-differences effects can be estimated by minimizing a single convex propensity-weighted loss, with errors from the two auxiliary models entering only as a product.

desk verdict Useful, mostly-correct DiD meta-learner with a patchable gap in the main rate theorem (missing clipping of estimated propensity); deserves serious review. read the letter →

arxiv 2502.04699 v2 pith:S7IPR4OF submitted 2025-02-07 stat.ML cs.LG

Hui Lan , Haoge Chang , Eleanor Dillon , Vasilis Syrgkanis This is my paper · ORCID

classification stat.MLcs.LG MSC 62D2062G0562G0862G2068T05

keywords difference-in-differencesconditionalaveragetreatmenteffectonthetreateddoublyrobustestimationNeymanorthogonalityheterogeneouseffectsmeta-learningconvexriskminimizationcovariateshift

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that the heterogeneous effect of a binary treatment in panel data, the conditional average treatment effect on the treated (CATT), can be estimated by minimizing one convex loss with generic machine learning in the second stage. It builds a pseudo-outcome from a control-group growth model and the treatment propensity, and proves the resulting loss is Neyman orthogonal, so bias from auxiliary-model errors enters only as a product of those errors. If this is right, practitioners get a plug-in recipe for treatment-effect heterogeneity along any subset of covariates, with fast statistical rates, in a setting where previous DiD tools mostly delivered average or group-average effects. The same recipe extends to general conditional functionals under covariate shift and to instrumented DiD with two-sided non-compliance.

What carries the argument

The load-bearing object is the propensity-weighted pseudo-outcome $\tilde{Y}(\pi, g) = \frac{D-\pi(W)}{1-\pi(W)}(\Delta Y - g(W))$ inserted into the incomplete quadratic loss $L(\theta) = \mathbb{E}[D\theta(X)^2 - 2\tilde{Y}\theta(X)]$. The associated conditional moment is Neyman orthogonal: first-order errors in $g$ and $\pi$ vanish at the truth, so the loss minimizer matches best projection onto $\theta_0$ over the treated population, and the excess-risk bound inherits only the product of the nuisance errors. For the general conditional functional problem, the same machinery uses a conditional Riesz representer $\alpha(W)$ in place of the constant $-1$ that the CATT delivers.

What would settle it

Run a large simulation with a subgroup whose true propensity equals 0.98, fit $\hat{\pi}$ and $\hat{g}$ with deliberately misspecified but reasonable models, and compare the treated-population MSE of $\hat{\theta}$ to the product-rate bound in Theorem 3.6; if the MSE scales much worse than the product of the two nuisance error rates, the double-robustness claim is not holding in that regime.

Watch

Extended reading notes

Core claim

Under conditional parallel trends and no anticipation, the true CATT solves a conditional moment equation involving the treatment indicator $D$, the propensity $\pi_0(W)$, the control-group trend $g_0(W) = \mathbb{E}[\Delta Y \mid D=0, W]$, and the target $\theta(X)$. Minimizing the convex loss $L(\theta; \pi, g) = \mathbb{E}[D\theta(X)^2 - 2\tilde{Y}\theta(X)]$ with $\tilde{Y} = \frac{D-\pi(W)}{1-\pi(W)}(\Delta Y - g(W))$ is, at the true nuisance functions, equivalent to the best $L^2$ projection of $\theta_0$ onto the hypothesis class among treated units. The main rate theorem states that if the second-stage learner has excess risk $R^2_{n,\delta}$, then $\|\hat{\theta} - \theta^*\|^2_{D=1} \le \frac{4}{\rho} R^2_{n,\delta} + \frac{2}{\rho^2 c^2} \mathbb{E}[\mathbb{E}[(\hat{g}-g_0)(\pi_0-\hat{\pi})/(1-\hat{\pi}) \mid X]^2]$, so nuisance bias is only the product of the two estimation errors. This is why the authors call the learner doubly robust for heterogeneous DiD effects and why it needs no treated-outcome regression model.

Load-bearing premise

The treatment propensity must be bounded strictly away from zero and one for all covariate values, because the pseudo-outcome divides by $1-\pi(W)$ and both the finite-sample constants and practical stability degrade as the propensity approaches one.

Editorial extensions

If this is right

Generic machine learning methods, including linear models, lasso, ridge, tree ensembles, and neural networks, can be used in the final stage with a single convex objective, and the out-of-sample loss serves as a model-selection score.
When $W$ is high-dimensional but interest lies in a small interpretable subset $X$, the learner remains consistent, whereas a CATE DR-learner is biased under conditional parallel trends.
With few treated units, the method avoids fitting a treated-outcome regression, which is particularly hard when treated data are scarce.
The multi-period and staggered-treatment extension treats distance-to-treatment and cohort as covariates, so heterogeneity over event time can be learned data-adaptively rather than by prespecified group averages.
The covariate-shift generalization yields a doubly robust learner for any conditional linear functional, covering prediction-powered inference and heterogeneous long-term effect estimation as special cases.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the product-rate bound is tight in practical regimes, then modest accuracy in both nuisance models suffices: errors of order $n^{-1/4}$ each give near-$n^{-1/2}$ CATT error, so the method should scale to large panels where the propensity and control trend are both learnable.
Because the weights divide by $1-\hat{\pi}(W)$, applications with near-universal treatment, such as minimum-wage counties where almost all units are treated, may require propensity clipping or a switch to the lagged-outcome identifying assumption; the paper's empirical case study uses low-dimensional covariates where this is less pressing.
The equivalence between loss minimization and best projection implies that held-out validation on treated units using the same loss can rank competing CATT models, a direct practical use that the paper does not spell out as a procedural recommendation.
Under violations of conditional parallel trends the learner can be badly biased; the paper's own violated-assumption experiments show the lagged-outcome variant is more robust, suggesting the identification assumption matters more than the doubly robust structure itself.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Here's the short version: this is a genuinely useful paper. The core contribution—a doubly robust, convex-loss meta-learner for the CATT that works when you project onto a subset X of the covariates—is new in this combination and mostly correct. I re-checked the identification, the orthogonal moment, and the loss equivalence; they hold. The bias diagnosis for CATE learners when X ⊂ W is also right and worth having on record. The general covariate-shift theorem (Theorem 4.7) is a nice generalization, and the IV extension is a bonus.

The main soft spot is in Theorem 3.6. The rate bound contains the factor (π0−π̂)/(1−π̂), and the proof converts L2 norms using the overlap constant c. But Assumption 3.1 only bounds the true propensity. Nothing in the assumptions or the proof keeps π̂ away from 1. If π̂ gets close to 1 in a region while π0 is moderate, the denominator blows up and the stated bound is not valid. The same gap propagates to Theorem A.12 for the IV case. This is fixable—clip π̂ to [c, 1−c] or otherwise require it—but as written the theorem is not internally valid. It doesn't kill the approach; the product-of-errors structure is still what double robustness should give you. But the theorem needs an explicit condition.

Two smaller things. First, no code or data are shipped, so the empirical superiority claims rest on the tables alone. Second, the real-data claim that the heterogeneity is statistically significant is based on calibration intervals with no formal test and no multiplicity correction; the out-of-sample validation is a good instinct, but the language overstates what the evidence supports. There are also a bunch of typos and notation slips (g0(x) vs g0(W), CATT vs CLATE) that a careful pass should clean up.

Who's this for? Applied economists using DiD with ML learners, and methodologists working on orthogonal statistical learning. The paper deserves a serious referee despite the gap; the central idea is sound and the fix is local. My recommendation: send it to review, but flag the clipping condition as a required revision, not just a suggestion.

Referee Report

2 major / 4 minor

Summary. The paper proposes a doubly robust meta-learner for the conditional average treatment effect on the treated (CATT) in two-period difference-in-differences designs. It derives identification under conditional parallel trends, introduces a Neyman-orthogonal moment condition, converts the moment into a convex loss whose minimizer equals the best projection of the CATT onto the treated population, and proves a rate bound in which nuisance estimation bias enters only through a product of propensity and outcome-regression errors. The framework is extended to general conditional functionals under covariate shift, to multiple time periods, and to an instrumented DiD setting with two-sided non-compliance. Experiments cover fully synthetic data, imbalanced data, a semi-synthetic minimum-wage exercise, and a real minimum-wage case study with a calibration check.

Significance. The paper addresses a relevant gap: flexible estimation of heterogeneous DiD effects projected onto a low-dimensional set X without estimating a high-dimensional treated-outcome model. The identification argument, the orthogonal-moment derivation, and the loss-equivalence result are carefully presented, and the algebra in Propositions 2.3, Lemma 3.3, and Proposition 3.5 is sound. The convex-loss formulation is clean and should allow generic ML second stages. However, the central rate theorem as stated is missing a condition on the estimated propensity, and this is load-bearing for the claimed double robustness. The IV extension has an additional strong-convexity smallness condition that is not checked or discussed empirically. These issues are fixable within the scope of the manuscript, so I recommend major revision.

major comments (2)

[Theorem 3.6 / Appendix D.4] The rate theorem is not valid as stated because no assumption is imposed on the estimated propensity pi_hat. Assumption 3.1 bounds only the true propensity pi_0 between c and 1-c. The nuisance-error term in Theorem 3.6, and the product term E[(g_hat - g0)(pi_0 - pi_hat)/(1 - pi_hat) | X] in Lemma D.2, divide by 1 - pi_hat. If pi_hat(W) is near (or equal to) 1 on a set of positive measure, this term is unbounded and the displayed coefficient beta = 2/(rho^2 c^2) cannot be justified from c alone. The proof in Appendix D.4 uses c to convert L2 norms to ||.||_{D=1}, but it never uses a bound on pi_hat. Please add an explicit condition such as pi_hat in [c, 1-c] almost surely, enforced by clipping or another stated mechanism, and include it in Theorem 3.6 and in the plug-in loss results in Section 4. The same issue propagates to Theorem A.12 through the Error(pi, g_D) and Error(pi, g_Y) terms.
[Theorem A.12] The IV rate theorem depends on the smallness condition Error(pi, g_D) <= c h k / (8 B^2), which is needed for strong convexity of the loss. The paper does not discuss when this condition holds, how to verify it, or how to enforce it in practice, and no IV simulations or data applications are reported. Because the condition is a premise of the theorem rather than a checked or enforced property, the practical scope of the IV extension is unclear. At minimum, please provide an implementable diagnostic or enforcement mechanism and, if possible, a small simulation illustrating the condition.

minor comments (4)

[Definition 3.2 / Proposition A.11] There are small terminology slips: Definition 3.2's title misspells 'Orthogonality', and Proposition A.11 refers to 'the CATT among the treated' where the estimand is the CLATT.
[Theorem 3.6 / Appendix D.4] The constants in the theorem statement do not match the proof. With beta_2 = 3/c and lambda = rho, Theorem D.7 gives coefficient 9/(rho^2 c^2) on the product term; even using the tighter Lemma D.1 bound gives 4/(rho^2 c^2), not the stated 2/(rho^2 c^2). Please reconcile the constants or state explicitly that they are not optimized.
[Section 6.2] The section contains typos ('minium wage', 'some extend of positive trends', 'tresholds') and the calibration figure captions should say whether the displayed confidence intervals are pointwise or simultaneous.
[Section 6 / Appendix E] The experiments do not report whether estimated propensities were clipped or otherwise constrained. If clipping is not used, the empirical implementation can suffer from the same near-one-propensity instability identified in Theorem 3.6, especially in the minimum-wage application where treatment is widespread.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the CATT learner's loss, moment, and rate bound are derived from stated identifying assumptions and an independent published theorem; self-citations are not load-bearing.

full rationale

The paper does not exhibit circularity. Proposition 2.3 derives the CATT identification formula from Assumptions 2.1 and 2.2, and the proof in Appendix D.1 is a direct calculation rather than an assumption of the target. Lemma 3.3 constructs the conditional moment equation from that identification and then verifies Neyman orthogonality by explicit directional-derivative computations in Appendix D.2; the pseudo-outcome is not defined to contain the target CATT, and the moment is shown to hold for the true θ0. Proposition 3.5 proves the equivalence between the squared loss and the best-projection problem algebraically from the moment equation, again without fitting any constant to the target. Theorem 3.6 is an application of Theorem 1 of Foster and Syrgkanis (2023), a published general orthogonal statistical learning result; the paper checks the required assumptions (D.3-D.6) in Appendix D.4 rather than assuming the theorem's conclusion. The self-citations, including Lan and Syrgkanis (2024) for model selection and Foster and Syrgkanis (2023) for the rate machinery, are either non-central or provide an independently published and generally applicable theorem, so they do not make the derivation circular. The general conditional-functionals extension in Theorem 4.7 follows the same pattern: the Riesz representer is defined as the representer of the moment functional, and orthogonality is proven directly. The only substantive concern raised in the proof is technical rather than circular: Theorem 3.6's bound contains the factor (π0(W) - πhat(W))/(1 - πhat(W)), and the stated assumptions bound the true propensity π0 but do not explicitly require the estimated πhat to be bounded away from 1. This is a potential validity gap in the theorem's statement, not a case of the result reducing to its inputs by construction. Empirical sections use synthetic and semi-synthetic data with known ground truth, so the reported comparisons are not defined in terms of the estimates being evaluated.

Assumptions & free parameters 0 free parameters · 7 assumptions · 0 invented entities

The central claim rests on untestable DiD identification (parallel trends, no anticipation), the overlap condition that makes the DR weights bounded, and the standard sample-splitting/excess-risk conditions borrowed from orthogonal statistical learning. No fitted constants are introduced in the theory; the simulation DGP parameters are illustrative, and the county-size finding is read from data, not encoded. The imported rate theorem (Foster and Syrgkanis 2023) is external published support, so these axioms carry ordinary identification burden, not circularity burden. The paper introduces no new entities: the conditional Riesz representer is imported from the debiased ML literature, and the estimands (CATT, CLATE) are standard.

assumptions (7)

domain assumption Conditional parallel trends: E[Y1(0)-Y0(0)|D=1,W] = E[Y1(0)-Y0(0)|D=0,W] (Assumption 2.1)
Untestable identification premise; if false, the CATT is not identified by the DiD contrast.
domain assumption No anticipation: E[Y0(0)-Y0(1)|D=1,W] = 0 (Assumption 2.2)
Needed to replace the treated potential outcome at t=0 with the observed outcome; standard in DiD.
domain assumption Sufficient overlap: c <= P(D=1|W) <= 1-c for some c>0 (Assumption 3.1)
Keeps the DR weight (D-pi)/(1-pi) bounded and enters the rate bound via 1/(rho^2 c^2).
domain assumption No concept drift in the covariate-shift framework: g0(W)=Es[Y|W]=Et[Y|W] (Assumption 4.1)
Needed for Theorem 4.7; the regression learned on the source distribution must transfer to the target distribution.
domain assumption IV-DiD assumptions: exclusion restriction, monotonicity, no carryover, CPTA in treatment and outcome, strong instrument (Assumptions A.2-A.6, A.8)
Identify the CLATE ratio in Proposition A.9 and are also needed for the Neyman orthogonality of the IV moment.
standard math Second-stage excess risk bound with nuisances from an auxiliary sample: L(theta-hat; pi-hat, g-hat) - inf_theta L(theta; pi-hat, g-hat) <= R^2_{n,delta} (hypothesis of Theorem 3.6)
Standard sample-splitting and empirical-risk condition inherited from the orthogonal statistical learning framework of Foster and Syrgkanis (2023).
standard math Units are i.i.d. draws from a superpopulation; T=2 panel (Section 2)
Sampling model for the probability statements; the multi-period extension is a reduction that treats distance to treatment as a covariate.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Meta-learner for Heterogeneous Effects in Difference-in-Differences." pith.science (2026). https://pith.science/paper/S7IPR4OF

@misc{pith2026250204699,
  author       = {Pith},
  title        = {Pith review of: A Meta-learner for Heterogeneous Effects in Difference-in-Differences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S7IPR4OF}},
  note         = {Machine review of arXiv:2502.04699}
}

read the original abstract

We address the problem of estimating heterogeneous treatment effects in panel data, adopting the popular Difference-in-Differences (DiD) framework under the conditional parallel trends assumption. We propose a novel doubly robust meta-learner for the Conditional Average Treatment Effect on the Treated (CATT), reducing the estimation to a convex risk minimization problem involving a set of auxiliary models. Our framework allows for the flexible estimation of the CATT, when conditioning on any subset of variables of interest using generic machine learning. Leveraging Neyman orthogonality, our proposed approach is robust to estimation errors in the auxiliary models. As a generalization to our main result, we develop a meta-learning approach for the estimation of general conditional functionals under covariate shift. We also provide an extension to the instrumented DiD setting with non-compliance. Empirical results demonstrate the superiority of our approach over existing baselines.

Figures

Figures reproduced from arXiv: 2502.04699 by the authors.

**Figure 2.** Calibration plot for CATT of minimum wage with respect to log county population. As a preliminary evaluation, we tested the performance of our methods on semi-synthetic data generated from this dataset. The semi-synthetic data was generated by bootstrapping the samples in the dataset and applying a chosen function for treatment assignment and treatment effect to compute the post-treatment outcomes. Since this datas… view at source ↗

**Figure 1.** Predicted CATT with respect to log county population. We applied our proposed approach to the minimum wage dataset that is also studied in Callaway & Sant’Anna, 2021 and Callaway, 2023. This dataset studies the effect of minimum wage changes on teen employment during the period 2001–2007. The outcome variable of interest is the log of county-level teen employment, while the treatment variable is defined as a binar… view at source ↗

**Figure 3.** Calibration plot for CATT w.r.t log county population for the XGBoost doubly robust learner. 31 [PITH_FULL_IMAGE:figures/full_fig_p031_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Calibration plot for CATT w.r.t log county population for the linear doubly robust learner. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Orthogonal Survival Learners for Estimating Heterogeneous Treatment Effects from Time-to-Event Data
cs.LG 2025-05 conditional novelty 6.0 of 10

The paper proposes a general toolbox of orthogonal survival learners with custom weighting functions to estimate heterogeneous treatment effects robustly under treatment, censoring, and survival overlap violations.

Reference graph

Works this paper leans on

37 extracted references · 31 canonical work pages · cited by 1 Pith paper

[1]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...
[2]

N., Bates, S., Fannjiang, C., Jordan, M

Angelopoulos, A. N., Bates, S., Fannjiang, C., Jordan, M. I., and Zrnic, T. Prediction-powered inference. Science, 382 0 (6671): 0 669--674, 2023

work page 2023
[3]

Angrist, J. D. and Pischke, J.-S. Mostly harmless econometrics: An empiricist's companion. Princeton university press, 2009

work page 2009
[4]

Autoregressive models for panel data causal inference with application to state-level opioid policies

Antonelli, J., Rubinstein, M., Agniel, D., Smart, R., Stuart, E., Cefalu, M., Schell, T., Eagan, J., Stone, E., Griswold, M., et al. Autoregressive models for panel data causal inference with application to state-level opioid policies. arXiv preprint arXiv:2408.09012, 2024

work page arXiv 2024
[5]

Ashenfelter, O. C. and Card, D. Using the longitudinal structure of earnings to estimate the effect of training programs, 1984

work page 1984
[6]

Difference-in-differences for policy evaluation

Callaway, B. Difference-in-differences for policy evaluation. Handbook of Labor, Human Resources and Population Economics, pp.\ 1--61, 2023

work page 2023
[7]

and Sant’Anna, P

Callaway, B. and Sant’Anna, P. H. Difference-in-differences with multiple time periods. Journal of econometrics, 225 0 (2): 0 200--230, 2021

work page 2021
[8]

and Krueger, A

Card, D. and Krueger, A. B. Minimum wages and employment: A case study of the fast-food industry in new jersey and pennsylvania. The American Economic Review, 84 0 (4): 0 772--793, 1994. ISSN 00028282. URL http://www.jstor.org/stable/2118030

work page arXiv 1994

Show all 37 references

[9]

Double/debiased/neyman machine learning of treatment effects

Chernozhukov, V., Chetverikov, D., Demirer, M., Duflo, E., Hansen, C., and Newey, W. Double/debiased/neyman machine learning of treatment effects. American Economic Review, 107 0 (5): 0 261--265, 2017

2017
[10]

Double/debiased machine learning for treatment and structural parameters, 2018

Chernozhukov, V., Chetverikov, D., Demirer, M., Duflo, E., Hansen, C., Newey, W., and Robins, J. Double/debiased machine learning for treatment and structural parameters, 2018

2018
[11]

K., Singh, R., and Srygkanis, V

Chernozhukov, V., Newey, M., Newey, W. K., Singh, R., and Srygkanis, V. Automatic debiased machine learning for covariate shifts. arXiv preprint arXiv:2307.04527, 2023

2023
[12]

What to do (and not to do) with causal panel analysis under parallel trends: Lessons from a large reanalysis study

Chiu, A., Lan, X., Liu, Z., and Xu, Y. What to do (and not to do) with causal panel analysis under parallel trends: Lessons from a large reanalysis study. arXiv preprint arXiv:2309.15983, 2023

2023
[13]

Daw, J. R. and Hatfield, L. A. Matching and regression to the mean in difference-in-differences analysis. Health services research, 53 0 (6): 0 4138--4156, 2018

2018
[14]

Dimick, J. B. and Ryan, A. M. Methods for evaluating changes in health care policy: the difference-in-differences approach. Jama, 312 0 (22): 0 2401--2402, 2014

2014
[15]

P., Allen, H., Baicker, K., and Oregon Health Study Group, t

Finkelstein, A., Taubman, S., Wright, B., Bernstein, M., Gruber, J., Newhouse, J. P., Allen, H., Baicker, K., and Oregon Health Study Group, t. The oregon health insurance experiment: evidence from the first year. The Quarterly journal of economics, 127 0 (3): 0 1057--1106, 2012

2012
[16]

Foster, D. J. and Syrgkanis, V. Orthogonal statistical learning. The Annals of Statistics, 51 0 (3): 0 879--908, 2023

2023
[17]

Evaluation of effectiveness of china's carbon emissions trading scheme in carbon mitigation

Gao, Y., Li, M., Xue, J., and Liu, Y. Evaluation of effectiveness of china's carbon emissions trading scheme in carbon mitigation. Energy Economics, 90: 0 104872, 2020

2020
[18]

and Green, D

Gerber, A. and Green, D. Field Experiments: Design, Analysis, and Interpretation. W. W. Norton, 2012. ISBN 9780393979954. URL https://books.google.com/books?id=yxEGywAACAAJ

2012
[19]

J., Ichimura, H., and Todd, P

Heckman, J. J., Ichimura, H., and Todd, P. E. Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme. The review of economic studies, 64 0 (4): 0 605--654, 1997

1997
[20]

Kennedy, E. H. Towards optimal doubly robust estimation of heterogeneous causal effects. Electronic Journal of Statistics, 17 0 (2): 0 3008--3049, 2023

2023
[21]

R., Sekhon, J

K \"u nzel, S. R., Sekhon, J. S., Bickel, P. J., and Yu, B. Metalearners for estimating heterogeneous treatment effects using machine learning. Proceedings of the national academy of sciences, 116 0 (10): 0 4156--4165, 2019

2019
[22]

and Syrgkanis, V

Lan, H. and Syrgkanis, V. Causal q-aggregation for cate model selection. In International Conference on Artificial Intelligence and Statistics, pp.\ 4366--4374. PMLR, 2024

2024
[23]

Instrumented difference-in-differences with heterogeneous treatment effects

Miyaji, S. Instrumented difference-in-differences with heterogeneous treatment effects. arXiv preprint arXiv:2405.12083, 2024

2024
[24]

and Wager, S

Nie, X. and Wager, S. Quasi-oracle estimation of heterogeneous treatment effects. Biometrika, 108 0 (2): 0 299--319, 2021

2021
[25]

L., Rotnitzky, A., and Robins, J

Ogburn, E. L., Rotnitzky, A., and Robins, J. M. Doubly robust estimation of the local average treatment effect curve. Journal of the Royal Statistical Society Series B: Statistical Methodology, 77 0 (2): 0 373--396, 2015

2015
[26]

Oprescu, M., Syrgkanis, V., and Wu, Z. S. Orthogonal random forest for causal inference. In International Conference on Machine Learning, pp.\ 4932--4941. PMLR, 2019

2019
[27]

Pierce, J. R. and Schott, P. K. The surprisingly swift decline of us manufacturing employment. American Economic Review, 106 0 (7): 0 1632--1662, 2016

2016
[28]

J., and Waldfogel, J

Rossin-Slater, M., Ruhm, C. J., and Waldfogel, J. The effects of california's paid family leave program on mothers’ leave-taking and subsequent labor market outcomes. Journal of Policy Analysis and Management, 32 0 (2): 0 224--245, 2013

2013
[29]

H., Bilinski, A., and Poe, J

Roth, J., Sant’Anna, P. H., Bilinski, A., and Poe, J. What’s trending in difference-in-differences? a synthesis of the recent econometrics literature. Journal of Econometrics, 235 0 (2): 0 2218--2244, 2023

2023
[30]

Sant’Anna, P. H. and Zhao, J. Doubly robust difference-in-differences estimators. Journal of econometrics, 219 0 (1): 0 101--122, 2020

2020
[31]

and Chernozhukov, V

Semenova, V. and Chernozhukov, V. Debiased machine learning of conditional average treatment effects and other causal functions. The Econometrics Journal, 24 0 (2): 0 264--289, 2021

2021
[32]

D., and Sontag, D

Shalit, U., Johansson, F. D., and Sontag, D. Estimating individual treatment effect: generalization bounds and algorithms. In International conference on machine learning, pp.\ 3076--3085. PMLR, 2017

2017
[33]

Adapting neural networks for the estimation of treatment effects

Shi, C., Blei, D., and Veitch, V. Adapting neural networks for the estimation of treatment effects. Advances in neural information processing systems, 32, 2019

2019
[34]

and Abraham, S

Sun, L. and Abraham, S. Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of econometrics, 225 0 (2): 0 175--199, 2021

2021
[35]

Machine learning estimation of heterogeneous treatment effects with instruments

Syrgkanis, V., Lei, V., Oprescu, M., Hei, M., Battocchi, K., and Lewis, G. Machine learning estimation of heterogeneous treatment effects with instruments. Advances in Neural Information Processing Systems, 32, 2019

2019
[36]

C., Rebeiro, P

Thome, J. C., Rebeiro, P. F., Spieker, A. J., and Shepherd, B. E. Understanding difference-in-differences methods to evaluate policy effects with staggered adoption: an application to medicaid and hiv. arXiv preprint arXiv:2402.12576, 2024

2024 arXiv
[37]

Wang, G., Hamad, R., and White, J. S. Advances in difference-in-differences methods for policy evaluation research. Epidemiology, 35 0 (5): 0 628--637, 2024

2024

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [1]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

[2] [2]

N., Bates, S., Fannjiang, C., Jordan, M

Angelopoulos, A. N., Bates, S., Fannjiang, C., Jordan, M. I., and Zrnic, T. Prediction-powered inference. Science, 382 0 (6671): 0 669--674, 2023

work page 2023

[3] [3]

Angrist, J. D. and Pischke, J.-S. Mostly harmless econometrics: An empiricist's companion. Princeton university press, 2009

work page 2009

[4] [4]

Autoregressive models for panel data causal inference with application to state-level opioid policies

Antonelli, J., Rubinstein, M., Agniel, D., Smart, R., Stuart, E., Cefalu, M., Schell, T., Eagan, J., Stone, E., Griswold, M., et al. Autoregressive models for panel data causal inference with application to state-level opioid policies. arXiv preprint arXiv:2408.09012, 2024

work page arXiv 2024

[5] [5]

Ashenfelter, O. C. and Card, D. Using the longitudinal structure of earnings to estimate the effect of training programs, 1984

work page 1984

[6] [6]

Difference-in-differences for policy evaluation

Callaway, B. Difference-in-differences for policy evaluation. Handbook of Labor, Human Resources and Population Economics, pp.\ 1--61, 2023

work page 2023

[7] [7]

and Sant’Anna, P

Callaway, B. and Sant’Anna, P. H. Difference-in-differences with multiple time periods. Journal of econometrics, 225 0 (2): 0 200--230, 2021

work page 2021

[8] [8]

and Krueger, A

Card, D. and Krueger, A. B. Minimum wages and employment: A case study of the fast-food industry in new jersey and pennsylvania. The American Economic Review, 84 0 (4): 0 772--793, 1994. ISSN 00028282. URL http://www.jstor.org/stable/2118030

work page arXiv 1994