pith. machine review for the scientific record. sign in

arxiv: 2605.08034 · v1 · submitted 2026-05-08 · 📊 stat.ML · cs.LG

Recognition: 2 theorem links

· Lean Theorem

Semiparametric Efficient Test for Interpretable Distributional Treatment Effects

Arthur Gretton, Houssam Zenati

Pith reviewed 2026-05-11 02:20 UTC · model grok-4.3

classification 📊 stat.ML cs.LG
keywords distributional treatment effectssemiparametric efficiencykernel mean embeddingsdoubly robust estimationcausal inferencelocal power analysispost-selection inference
0
0 comments X

The pith

DR-ME learns finite outcome locations to test for distributional treatment effects with semiparametric efficiency.

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

Distributional treatment effects often leave means unchanged while shifting tails or modes, so mean-based tests miss them. Global kernel tests can detect overall differences between treated and untreated outcome laws but return only a single rejection without showing where the laws differ. The paper introduces DR-ME, a finite-location test that evaluates an interventional kernel witness at data-driven points and returns explicit causal-discrepancy coordinates. It derives orthogonal doubly robust kernel features from observational data whose centered form is the canonical gradient, ensuring double robustness and semiparametric efficiency. Sample splitting preserves post-selection validity while covariance whitening optimizes local signal-to-noise for the chosen coordinates.

Core claim

DR-ME is the first semiparametrically efficient finite-location test for interpretable distributional treatment effects. It evaluates an interventional kernel witness at learned outcome locations, using orthogonal doubly robust kernel features whose oracle form supplies the canonical gradient. For fixed locations the procedure is chi-square calibrated under the null, attains noncentral chi-square local power, and employs covariance whitening that maximizes local signal-to-noise; the same geometry supplies a principled location-learning criterion.

What carries the argument

Orthogonal doubly robust kernel features evaluated at learned finite outcome locations, with covariance whitening that optimizes local signal-to-noise.

If this is right

  • The test returns explicit coordinates of causal discrepancy rather than a single global p-value.
  • Local power is noncentral chi-square when the alternative is visible through the selected coordinates.
  • Sample splitting keeps the procedure valid after data-driven location selection.
  • The same whitening geometry can be used to compare power across different location choices.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The location-learning step could be replaced by a fixed grid in settings where interpretability is less important than exhaustive coverage.
  • The doubly robust features might be adapted to continuous treatment or multi-arm designs without changing the efficiency argument.
  • The chi-square local-power geometry suggests a natural way to rank candidate location sets before data collection.

Load-bearing premise

The derivation assumes standard causal identification conditions together with correct specification of at least one nuisance model and the validity of sample splitting.

What would settle it

A simulation or real dataset in which the true distributional discrepancy occurs only outside the learned locations yet DR-ME still rejects at the nominal level.

Figures

Figures reproduced from arXiv: 2605.08034 by Arthur Gretton, Houssam Zenati.

Figure 1
Figure 1. Figure 1: Training objective for learned causal locations. Each panel fixes [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Calibration and power under observational confounding. Left: confounded null [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: OCTMNIST image-location experiment. From left to right: sampled potential images [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sharp-null calibration under observational confounding. The dashed line is the nominal level [PITH_FULL_IMAGE:figures/full_fig_p038_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: DR-ME versus the global DR-xKTE baseline. Both methods are close to nominal level under [PITH_FULL_IMAGE:figures/full_fig_p039_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Empirical rejection rates against the noncentral chi-square prediction. The black theoretical b [PITH_FULL_IMAGE:figures/full_fig_p043_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Null distribution of the statistic at the largest sample size. The histogram shows the Monte Carlo b [PITH_FULL_IMAGE:figures/full_fig_p044_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: QQ plot at the largest sample size. Empirical quantiles of the statistic are compared with the [PITH_FULL_IMAGE:figures/full_fig_p044_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Moment diagnostic. The empirical mean of [PITH_FULL_IMAGE:figures/full_fig_p045_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Total runtime versus sample size. DR-xKTE is competitive at small sample sizes, where the [PITH_FULL_IMAGE:figures/full_fig_p046_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Core method runtime versus sample size. This plot removes nuisance fitting and focuses on the [PITH_FULL_IMAGE:figures/full_fig_p047_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Runtime breakdown at n = 10000, J = 2, and M = 80. The finite-dictionary DR-ME implementation keeps the final test low-dimensional and is faster than DR-xKTE, whose cost is dominated by the global kernel test. Gradient-based DR-ME spends most of its time in location learning. and use OCTMNIST images only on the outcome side. For each unit, we first sample synthetic covariates Xi ∈ R dX , Xi ∼ N(0, IdX ), … view at source ↗
Figure 13
Figure 13. Figure 13: Additional diagnostics for the OCTMNIST mean-matched image-location experiment. Top row: [PITH_FULL_IMAGE:figures/full_fig_p050_13.png] view at source ↗
read the original abstract

Distributional treatment effects can be invisible to means: a treatment may preserve average outcomes while changing tails, modes, dispersion, or rare-event probabilities. Kernel tests can detect discrepancies between interventional outcome laws, but global tests do not reveal where the laws differ. We propose DR-ME, to our knowledge the first semiparametrically efficient finite-location test for interpretable distributional treatment effects. DR-ME evaluates an interventional kernel witness at learned outcome locations, returning causal-discrepancy coordinates rather than only a global rejection. From observational data, we derive orthogonal doubly robust kernel features whose centered oracle form is the canonical gradient of this finite witness. For fixed locations, we characterize the local testing limit: DR-ME is chi-square calibrated under the null, has noncentral chi-square local power, and uses the covariance whitening that optimizes local signal-to-noise for discrepancies visible through the selected coordinates. This efficient local-power geometry yields a principled location-learning criterion, with sample splitting preserving post-selection validity. Experiments show near-nominal type-I error, competitive power against global doubly robust kernel tests, and interpretable learned locations that localize distributional effects in a semi-synthetic medical-imaging study.

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.

Referee Report

2 major / 3 minor

Summary. The paper proposes DR-ME as the first semiparametrically efficient finite-location test for interpretable distributional treatment effects. From observational data it derives orthogonal doubly robust kernel features whose oracle form is the canonical gradient of a finite witness function evaluated at learned outcome locations. For fixed locations the method is shown to be chi-square calibrated under the null with noncentral chi-square local power; covariance whitening is used to optimize local signal-to-noise. The efficient local-power geometry induces a location-learning criterion, with sample splitting preserving post-selection validity. Experiments report near-nominal type-I error, competitive power versus global doubly robust kernel tests, and interpretable learned locations in a semi-synthetic medical-imaging study.

Significance. If the central claims hold, the contribution is significant: it supplies the first efficient, interpretable, finite-location procedure for distributional treatment effects that remains valid under standard causal identification and double-robust nuisance estimation. The derivation of the location-learning rule directly from the local-power geometry is a clean application of semiparametric theory and distinguishes the work from purely global kernel tests. The combination of double robustness, chi-square calibration, and post-selection validity via sample splitting is a practical strength for applied causal work.

major comments (2)
  1. [§3.2] §3.2 (canonical gradient derivation): the claim that the constructed features achieve the semiparametric efficiency bound for the finite witness relies on the features coinciding with the canonical gradient; an explicit verification that the doubly robust estimator is orthogonal to the nuisance tangent space (including the precise form of the influence function) would strengthen the efficiency result.
  2. [§4.3] §4.3 (local-power geometry and learned locations): the noncentrality parameter and covariance-whitening optimality are derived for fixed locations; the argument that the data-driven location criterion (induced by the same geometry) preserves the chi-square null limit and the local-power optimality after sample splitting needs a precise statement of the asymptotic expansion under the null.
minor comments (3)
  1. [Abstract] The abstract states 'to our knowledge the first'; a brief sentence contrasting with existing global DR kernel tests and finite-location mean tests would help readers place the novelty.
  2. [Experiments] In the experimental section, the semi-synthetic medical-imaging setup would benefit from an explicit statement of the nuisance estimators used (e.g., which ML methods for propensity and outcome regression) and the number of sample splits.
  3. [§2] Notation for the kernel witness and the finite set of locations could be introduced earlier and used consistently to ease reading of the theoretical sections.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough and constructive review. We appreciate the positive assessment of the contribution and the recommendation for minor revision. We address each major comment below.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (canonical gradient derivation): the claim that the constructed features achieve the semiparametric efficiency bound for the finite witness relies on the features coinciding with the canonical gradient; an explicit verification that the doubly robust estimator is orthogonal to the nuisance tangent space (including the precise form of the influence function) would strengthen the efficiency result.

    Authors: We agree that an explicit verification would strengthen the efficiency result. In the revised manuscript we will add a dedicated appendix deriving the influence function of the doubly robust kernel features and verifying orthogonality to the full nuisance tangent space (including the precise form of the canonical gradient for the finite witness). revision: yes

  2. Referee: [§4.3] §4.3 (local-power geometry and learned locations): the noncentrality parameter and covariance-whitening optimality are derived for fixed locations; the argument that the data-driven location criterion (induced by the same geometry) preserves the chi-square null limit and the local-power optimality after sample splitting needs a precise statement of the asymptotic expansion under the null.

    Authors: We acknowledge that a precise asymptotic expansion under the null is needed to rigorously justify preservation of the chi-square limit after sample splitting. In the revision we will insert an explicit statement of the asymptotic expansion under the null (including the o_p(1) remainder terms) that confirms the chi-square calibration and the retention of local-power optimality for the data-driven locations. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper derives the orthogonal doubly robust kernel features directly as the canonical gradient of the finite witness and obtains the location-learning criterion from the local-power optimality geometry under the chi-square limit characterization. These steps follow from standard semiparametric efficiency theory and local asymptotic analysis applied to the interventional kernel witness, without reducing to a fitted input renamed as prediction or to any self-citation chain. The abstract and description invoke only external causal identification conditions and sample-splitting validity, with no self-definitional loops, ansatz smuggling, or uniqueness theorems imported from the authors' prior work. The central claim of semiparametric efficiency and interpretable finite-location testing therefore remains independent of its own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard causal identification assumptions and semiparametric efficiency theory from prior literature; the paper introduces no new free parameters, axioms, or invented entities beyond the DR-ME procedure itself.

axioms (2)
  • domain assumption Standard causal assumptions (consistency, no unmeasured confounding, positivity) allow identification of interventional outcome distributions from observational data
    Required to derive treatment effects and doubly robust features from observational samples
  • standard math Kernel mean embeddings and reproducing kernel Hilbert spaces are well-defined for the outcome space
    Foundation for the kernel witness function

pith-pipeline@v0.9.0 · 5504 in / 1496 out tokens · 47592 ms · 2026-05-11T02:20:14.055560+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 1 internal anchor

  1. [1]

    Heejung Bang and James M. Robins. Doubly robust estimation in missing data and causal inference models.Biometrics, 61(4):962–973, 2005. doi: 10.1111/j.1541-0420.2005.00377.x

    work page doi:10.1111/j.1541-0420.2005.00377.x 2005

  2. [2]

    Bickel, Chris A

    Peter J. Bickel, Chris A. J. Klaassen, Ya’acov Ritov, and Jon A. Wellner.Efficient and Adaptive Estimation for Semiparametric Models. Johns Hopkins University Press, 1993

    work page 1993

  3. [3]

    Inference on counterfactual distributions

    Victor Chernozhukov, Iván Fernández-Val, and Blaise Melly. Inference on counterfactual distributions. Econometrica, 81(6):2205–2268, 2013

    work page 2013

  4. [4]

    Double/debiased machine learning for treatment and structural parameters.The Econometrics Journal, 21(1):C1–C68, 2018

    Victor Chernozhukov, Denis Chetverikov, Mert Demirer, Esther Duflo, Christian Hansen, Whitney Newey, and James Robins. Double/debiased machine learning for treatment and structural parameters. The Econometrics Journal, 21(1):C1–C68, 2018. doi: 10.1111/ectj.12097

    work page doi:10.1111/ectj.12097 2018

  5. [5]

    Fast two-sample testing with analytic representations of probability measures

    Kacper P Chwialkowski, Aaditya Ramdas, Dino Sejdinovic, and Arthur Gretton. Fast two-sample testing with analytic representations of probability measures. InAdvances in Neural Information Processing Systems, volume 28, 2015. 10

    work page 2015

  6. [6]

    Evans, and Dino Sejdinovic

    Jake Fawkes, Robert Hu, Robin J. Evans, and Dino Sejdinovic. Doubly robust kernel statistics for testing distributional treatment effects.Transactions on Machine Learning Research, 2024

    work page 2024

  7. [7]

    Optimal inference after model selection, 2014

    William Fithian, Dennis Sun, and Jonathan Taylor. Optimal inference after model selection, 2014

    work page 2014

  8. [8]

    A survey of kernels for structured data.ACM SIGKDD explorations newsletter, 5(1): 49–58, 2003

    Thomas Gärtner. A survey of kernels for structured data.ACM SIGKDD explorations newsletter, 5(1): 49–58, 2003

    work page 2003

  9. [9]

    A kernel two-sample test.The Journal of Machine Learning Research, 13(1):723–773, 2012

    Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Schölkopf, and Alexander Smola. A kernel two-sample test.The Journal of Machine Learning Research, 13(1):723–773, 2012

    work page 2012

  10. [10]

    On the role of the propensity score in efficient semiparametric estimation of average treatment effects.Econometrica, 66(2):315–331, 1998

    Jinyong Hahn. On the role of the propensity score in efficient semiparametric estimation of average treatment effects.Econometrica, 66(2):315–331, 1998. doi: 10.2307/2998560

    work page doi:10.2307/2998560 1998

  11. [11]

    Hernán and James M

    Miguel A. Hernán and James M. Robins.Causal Inference: What If. Chapman & Hall/CRC, 2020

    work page 2020

  12. [12]

    Springer, Berlin, 3 edition, 1996

    Reiner Horst and Hoang Tuy.Global Optimization: Deterministic Approaches. Springer, Berlin, 3 edition, 1996

    work page 1996

  13. [13]

    The Likelihood Test of Independence in Contingency Tables,

    Harold Hotelling. The generalization of student’s ratio.The Annals of Mathematical Statistics, 2(3): 360–378, 1931. doi: 10.1214/aoms/1177732979

    work page doi:10.1214/aoms/1177732979 1931

  14. [14]

    Inference on function-valued parameters using a restricted score test, 2021

    Aaron Hudson, Marco Carone, and Ali Shojaie. Inference on function-valued parameters using a restricted score test, 2021

    work page 2021

  15. [15]

    Imbens and Donald B

    Guido W. Imbens and Donald B. Rubin.Causal Inference for Statistics, Social, and Biomedical Sciences: An Introduction. Cambridge University Press, 2015. doi: 10.1017/CBO9781139025751

    work page doi:10.1017/cbo9781139025751 2015

  16. [16]

    Chwialkowski, and Arthur Gretton

    Wittawat Jitkrittum, Zoltán Szabó, Kacper P. Chwialkowski, and Arthur Gretton. Interpretable distribution features with maximum testing power. InAdvances in Neural Information Processing Systems 29, pages 181–189, 2016

    work page 2016

  17. [17]

    Edward H. Kennedy. Semiparametric doubly robust targeted double machine learning: A review, 2022

    work page 2022

  18. [18]

    Kermany, Michael Goldbaum, Wenjia Cai, Carolina C

    Daniel S. Kermany, Michael Goldbaum, Wenjia Cai, Carolina C. S. Valentim, Huiying Liang, Sally L. Baxter, Alex McKeown, Ge Yang, Xiaokang Wu, Fangbing Yan, Justin Dong, Made K. Prasadha, Jacqueline Pei, Magdalene Y. L. Ting, Jie Zhu, Christina Li, Sierra Hewett, Jason Dong, Ian Ziyar, Alexander Shi, Runze Zhang, Lianghong Zheng, Rui Hou, William Shi, Xiao...

    work page 2018

  19. [19]

    Kuchibhotla and John E

    Arun Kumar Kuchibhotla, John E. Kolassa, and Todd A. Kuffner. Post-selection inference.Annual Review of Statistics and Its Application, 9:505–527, 2022. doi: 10.1146/annurev-statistics-100421-044639

    work page doi:10.1146/annurev-statistics-100421-044639 2022

  20. [20]

    Springer Series in Statistics

    Lucien M. Le Cam and Grace Lo Yang.Asymptotics in Statistics: Some Basic Concepts. Springer Series in Statistics. Springer, 2 edition, 2000. doi: 10.1007/978-1-4612-1166-2

    work page doi:10.1007/978-1-4612-1166-2 2000

  21. [21]

    One-step estimation of differentiable hilbert-valued parameters.The Annals of Statistics, 52(4):1534–1563, 2024

    Alex Luedtke and Incheoul Chung. One-step estimation of differentiable hilbert-valued parameters.The Annals of Statistics, 52(4):1534–1563, 2024

    work page 2024

  22. [22]

    Luedtke, Marco Carone, and Mark J

    Alexander R. Luedtke, Marco Carone, and Mark J. van der Laan. An omnibus non-parametric test of equality in distribution for unknown functions.Journal of the Royal Statistical Society: Series B, 81(1): 75–99, 2019. 11

    work page 2019

  23. [23]

    An efficient doubly-robust test for the kernel treatment effect

    Diego Martinez Taboada, Aaditya Ramdas, and Edward Kennedy. An efficient doubly-robust test for the kernel treatment effect. InAdvances in Neural Information Processing Systems, volume 36, pages 59924–59952, 2023

    work page 2023

  24. [24]

    Counter- factual mean embeddings.Journal of Machine Learning Research, 22(162):1–71, 2021

    Krikamol Muandet, Motonobu Kanagawa, Sorawit Saengkyongam, and Sanparith Marukatat. Counter- factual mean embeddings.Journal of Machine Learning Research, 22(162):1–71, 2021

    work page 2021

  25. [25]

    Jerzy Neyman and Egon S. Pearson. On the problem of the most efficient tests of statistical hypotheses. Philosophical Transactions of the Royal Society of London. Series A, Containing Papers of a Mathematical or Physical Character, 231(694–706):289–337, 1933

    work page 1933

  26. [26]

    Wright.Numerical Optimization

    Jorge Nocedal and Stephen J. Wright.Numerical Optimization. Springer, New York, 2 edition, 2006

    work page 2006

  27. [27]

    S. A. Piyavskii. An algorithm for finding the absolute extremum of a function.USSR Computational Mathematics and Mathematical Physics, 12(4):57–67, 1972

    work page 1972

  28. [28]

    Robins, Andrea Rotnitzky, and Lue Ping Zhao

    James M. Robins, Andrea Rotnitzky, and Lue Ping Zhao. Estimation of regression coefficients when some regressors are not always observed.Journal of the American Statistical Association, 89(427):846–866,

    work page

  29. [29]

    doi: 10.1080/01621459.1994.10476818

    work page doi:10.1080/01621459.1994.10476818 1994

  30. [30]

    Biometrika , author =

    Paul R. Rosenbaum and Donald B. Rubin. The central role of the propensity score in observational studies for causal effects.Biometrika, 70(1):41–55, 1983. doi: 10.1093/biomet/70.1.41

    work page doi:10.1093/biomet/70.1.41 1983

  31. [31]

    Nonparametric estimation of distributional policy effects.Journal of Econometrics, 155(1):56–70, 2010

    Christoph Rothe. Nonparametric estimation of distributional policy effects.Journal of Econometrics, 155(1):56–70, 2010

    work page 2010

  32. [32]

    Donald B. Rubin. Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of Educational Psychology, 66(5):688–701, 1974. doi: 10.1037/h0037350

    work page doi:10.1037/h0037350 1974

  33. [33]

    MIT Press, Cambridge, MA, 2004

    Bernhard Schölkopf, Koji Tsuda, and Jean-Philippe Vert, editors.Kernel Methods in Computational Biology. MIT Press, Cambridge, MA, 2004. ISBN 9780262195096

    work page 2004

  34. [34]

    Bruno O. Shubert. A sequential method seeking the global maximum of a function.SIAM Journal on Numerical Analysis, 9(3):379–388, 1972

    work page 1972

  35. [35]

    A hilbert space embedding for distributions

    Alex Smola, Arthur Gretton, Le Song, and Bernhard Schölkopf. A hilbert space embedding for distributions. InInternational conference on algorithmic learning theory, pages 13–31. Springer, 2007

    work page 2007

  36. [36]

    Universality, characteristic kernels and rkhs embedding of measures.Journal of Machine Learning Research, 12(7), 2011

    Bharath K Sriperumbudur, Kenji Fukumizu, and Gert RG Lanckriet. Universality, characteristic kernels and rkhs embedding of measures.Journal of Machine Learning Research, 12(7), 2011

    work page 2011

  37. [37]

    A. W. van der Vaart.Asymptotic Statistics, volume 3 ofCambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, 1998. doi: 10.1017/CBO9780511802256

    work page doi:10.1017/cbo9780511802256 1998

  38. [38]

    Classification of biological sequences with kernel methods

    Jean-Philippe Vert. Classification of biological sequences with kernel methods. In Yasubumi Sakakibara, Satoshi Kobayashi, Kengo Sato, Tetsuro Nishino, and Etsuji Tomita, editors,Grammatical Inference: Algorithms and Applications, volume 4201 ofLecture Notes in Computer Science, pages 7–18, Berlin, Heidelberg, 2006. Springer. doi: 10.1007/11872436_2

    work page doi:10.1007/11872436_2 2006

  39. [39]

    Jean-Philippe Vert, Robert Thurman, and William S. Noble. Kernels for gene regulatory regions. In Yair Weiss, Bernhard Schölkopf, and John Platt, editors,Advances in Neural Information Processing Systems 18, pages 1401–1408, Cambridge, MA, 2005. MIT Press

    work page 2005

  40. [40]

    Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification

    Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, and Bingbing Ni. Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification. Scientific Data, 10(1):41, 2023. 12

    work page 2023

  41. [41]

    Doubly-robust estimation of counterfactual policy mean embeddings

    Houssam Zenati, Bariscan Bozkurt, and Arthur Gretton. Doubly-robust estimation of counterfactual policy mean embeddings. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. URLhttps://openreview.net/forum?id=0GDlX9JFf2

    work page 2025

  42. [42]

    Kernel Treatment Effects with Adaptively Collected Data

    Houssam Zenati, Bariscan Bozkurt, and Arthur Gretton. Kernel treatment effects with adaptively collected data, 2025. URLhttps://arxiv.org/abs/2510.10245. 13 Appendix Appendix organization.Appendix A collects notation and the full assumptions used in the main text. Appendix B gives a compact review of the local asymptotic tools used for the testing-efficie...

    work page internal anchor Pith review Pith/arXiv arXiv 2025