arxiv: 2605.09264 · v1 · submitted 2026-05-10 · 📊 stat.ME

Recognition: 2 theorem links

· Lean Theorem

Nested Sensitivity Envelopes for Transported Quantile Treatment Effects

Pengyun Wang

Pith reviewed 2026-05-12 04:36 UTC · model grok-4.3

classification 📊 stat.ME

keywords quantile treatment effectstransportabilitysensitivity analysismarginal sensitivitycounterfactual distributionsemiparametric estimationbreakdown frontierNeyman orthogonality

0 comments

The pith

By nesting a source treatment-sensitivity map inside a target outcome-shift map, closed-form sharp envelopes for counterfactual CDFs yield attainable bounds on transported quantile treatment effects.

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

The paper addresses estimation of quantile treatment effects in a target population when a source study may suffer from unmeasured confounding and the populations may differ in ways not captured by observed covariates. It imposes an odds-ratio bound on source treatment assignment and a likelihood-ratio bound on the shift in potential-outcome distributions from source to target. For each treatment arm and threshold, the method produces a closed-form envelope for the target counterfactual cumulative distribution function by nesting the two sensitivity maps while preserving their normalizations. The envelopes are shown to be sharp at the process level, meaning they are attained by some valid counterfactual distribution, and can therefore be inverted to obtain sharp quantile bounds and sharp interval-hull bounds on the quantile treatment effect. Semiparametric estimators and inference tools, including cross-fitted Neyman-orthogonal procedures and subsampling for nonsmooth cases, are developed to make the bounds operational in data.

Core claim

Under an odds-ratio bound Γ on source treatment assignment and a conditional likelihood-ratio bound Λ on source-to-target potential-outcome shift, a closed-form sharp target counterfactual CDF envelope is obtained by nesting a source marginal-sensitivity map inside a target outcome-shift map; the envelope preserves two normalizations, improves on a single product likelihood-ratio relaxation, is attainable as an entire CDF, and inverts to sharp target quantile bounds and sharp interval-hull QTE bounds.

What carries the argument

The nested sensitivity envelope for the target counterfactual CDF, which places a source marginal-sensitivity map inside a target outcome-shift map while preserving the two normalizations required by the marginal restrictions.

If this is right

Inverting the CDF envelopes produces sharp bounds on target quantiles for each treatment arm.
The same inversion yields sharp interval-hull bounds on the quantile treatment effect process.
The canonical gradient includes the source propensity score contribution required for observational data.
Cross-fitted Neyman-orthogonal one-step estimators achieve uniform Gaussian approximation on regular index sets.
Simultaneous monotone CDF bands invert to honest confidence sets for the quantile and QTE processes; the breakdown frontier in the (Γ, Λ) plane is obtained via level-set inference.

Where Pith is reading between the lines

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

The nesting construction may extend to other transported functionals such as means or survival probabilities by replacing the quantile inversion step with the appropriate functional.
The two-dimensional breakdown frontier offers a way to report how much confounding and shift would be needed to overturn a qualitative finding, which could be plotted directly from the level sets of the interval-hull process.
When the target sample is large relative to the source, the procedure effectively reduces to standard marginal sensitivity analysis on the source with only a reweighting adjustment for the target covariate distribution.

Load-bearing premise

The two marginal sensitivity restrictions hold exactly: an odds-ratio bound on source treatment assignment and a conditional likelihood-ratio bound on the source-to-target shift in potential-outcome distributions.

What would settle it

A concrete data-generating process in which the true target counterfactual CDF lies strictly outside the derived envelope for some treatment arm and threshold while still satisfying the stated odds-ratio and likelihood-ratio bounds.

Figures

Figures reproduced from arXiv: 2605.09264 by Pengyun Wang.

read the original abstract

We study target-population quantile treatment effects when a source study may have unmeasured treatment confounding and may not transport to a target population after conditioning on observed covariates. The observed data consist of a source sample with treatment, outcome and covariates, and a target sample with covariates only. We impose two marginal sensitivity restrictions: an odds-ratio bound \(\Gam\) for source treatment assignment and a conditional likelihood-ratio bound \(\Lam\) for source-to-target potential-outcome distribution shift. For each treatment arm and threshold \(y\), we derive a closed-form sharp target counterfactual CDF envelope. The envelope nests a source marginal-sensitivity map inside a target outcome-shift map, preserving two normalizations and generally improving on a single product likelihood-ratio relaxation. We prove process-level sharpness, so the envelopes are attainable as entire CDFs and can be inverted to obtain sharp target quantile bounds and sharp interval-hull QTE bounds. We then develop semiparametric theory for these nonsmooth bound processes. On regular index sets, we give the canonical gradient, including the source propensity contribution required in observational studies, and construct cross-fitted Neyman-orthogonal one-step estimators with uniform Gaussian approximation. On full index sets with active-set ties or mass points, we use Hadamard directional differentiability and subsampling-valid inference, with a primitive finite-support route for the required weak convergence. Finally, we invert simultaneous monotone CDF bands to obtain honest confidence sets for quantile and QTE interval-hull processes, and formulate the two-dimensional \((\Gam,\Lam)\) breakdown frontier as level-set inference for interval-hull non-refutation.

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

The nested source-target sensitivity construction gives sharp, invertible CDF envelopes for transported quantile effects under two marginal bounds.

read the letter

The paper's main advance is nesting an odds-ratio bound Γ on source treatment assignment inside a likelihood-ratio bound Λ on source-to-target outcome shift. This produces closed-form sharp counterfactual CDF envelopes for each arm and threshold y that preserve both normalizations and improve on a plain product relaxation. They prove the envelopes are attainable as entire processes, which lets them invert directly to sharp quantile bounds and interval-hull QTE bounds. The two-dimensional (Γ, Λ) breakdown frontier for non-refutation is a clean practical output from the same setup. On the theory side they supply the canonical gradient (including the source propensity term), cross-fitted Neyman-orthogonal estimators, uniform Gaussian approximation on regular index sets, and Hadamard directional differentiability plus subsampling for the nonsmooth case with ties or mass points. That package looks complete for honest inference on the bound processes. The assumptions are the usual marginal sensitivity restrictions, stated explicitly, with no sign of circularity or data-fitting issues in the central claims. The derivations are claimed to be closed-form, and the stress-test finds no internal inconsistency, so the technical steps appear to hold up. Minor practical question is how the process-level objects behave in moderate samples with continuous outcomes, but that is implementation rather than a flaw in the argument. This is for causal-inference people working on transportability and sensitivity analysis, especially with quantiles in policy or medical settings. A reader who needs sharp partial-identification bounds and valid inference under those restrictions will get direct value. It deserves a serious referee because the nesting construction and the full semiparametric treatment are substantive and new relative to single-parameter or product-bound approaches.

Referee Report

2 major / 3 minor

Summary. The manuscript develops a nested sensitivity framework for bounding quantile treatment effects transported from a source study (with treatment, outcome, and covariates) to a target population (covariates only). Under an odds-ratio bound Γ on source treatment assignment and a conditional likelihood-ratio bound Λ on source-to-target potential-outcome shifts, it derives closed-form sharp envelopes for the target counterfactual CDFs by nesting the source marginal-sensitivity map inside the target outcome-shift map. The paper proves process-level sharpness (attainability as entire CDF processes), inverts these to sharp quantile and interval-hull QTE bounds, and supplies semiparametric theory: canonical gradients (including propensity scores), cross-fitted Neyman-orthogonal estimators with uniform Gaussian approximation on regular index sets, and Hadamard directional differentiability plus subsampling inference on full index sets with ties or mass points. It concludes with honest simultaneous confidence sets obtained by inverting monotone CDF bands and a two-dimensional (Γ, Λ) breakdown frontier.

Significance. If the closed-form derivations and process-level sharpness hold, the work advances partial identification and sensitivity analysis for transported quantile effects, which are policy-relevant yet technically demanding. The nesting construction yields tighter bounds than a single product likelihood-ratio relaxation while preserving normalizations, and the accompanying semiparametric efficiency results, uniform inference procedures, and breakdown-frontier formulation provide a complete, practical toolkit. These features address a genuine gap between existing mean-focused transport sensitivity methods and the needs of quantile analysis.

major comments (2)

[§3] §3, main envelope theorem: the claim that the nested map preserves both normalizations (CDF limits at 0 and 1) and is strictly sharper than the product relaxation requires an explicit algebraic verification or counter-example check for discrete outcomes with mass points; the current argument appears to rely on continuity assumptions that may not be stated.
[§5.3] §5.3, subsampling procedure: when active-set ties occur on the full index set, the primitive finite-support route for weak convergence must specify how the directional derivative is computed at points of non-differentiability to guarantee that the subsampling quantiles remain valid for the interval-hull QTE process.

minor comments (3)

Notation for the sensitivity parameters alternates between Γ/Λ and script letters; a single consistent font throughout would reduce reader confusion.
The abstract and introduction cite the improvement over product relaxations but do not reference the closest prior nested or composite sensitivity models (e.g., recent work on marginal sensitivity for transport); adding two or three targeted citations would clarify novelty.
Figure captions for the envelope plots should explicitly label the source map, target map, and nested envelope curves so that the visual comparison to the product bound is immediate.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We address the two major comments point by point below.

read point-by-point responses

Referee: [§3] §3, main envelope theorem: the claim that the nested map preserves both normalizations (CDF limits at 0 and 1) and is strictly sharper than the product relaxation requires an explicit algebraic verification or counter-example check for discrete outcomes with mass points; the current argument appears to rely on continuity assumptions that may not be stated.

Authors: We agree that an explicit algebraic verification for discrete outcomes with mass points would strengthen the presentation. Theorem 3.1 establishes process-level sharpness under general conditions that encompass atoms, but the normalization preservation and strict improvement over the product relaxation are shown via the nested construction without an isolated discrete-case lemma. In the revision we will add a short algebraic verification (new Lemma 3.2) confirming that the nested envelopes attain the CDF limits 0 and 1 at the boundaries for any finite-support distribution and remain strictly sharper than the product relaxation, with a simple two-point counter-example illustrating the improvement. revision: yes
Referee: [§5.3] §5.3, subsampling procedure: when active-set ties occur on the full index set, the primitive finite-support route for weak convergence must specify how the directional derivative is computed at points of non-differentiability to guarantee that the subsampling quantiles remain valid for the interval-hull QTE process.

Authors: We appreciate the request for additional detail on the directional derivative. Section 5.3 already invokes Hadamard directional differentiability of the interval-hull map and supplies a primitive finite-support route for weak convergence. To address the concern explicitly, the revision will expand the exposition to state that, at points of non-differentiability induced by active-set ties or mass points, the directional derivative is obtained as the support function of the tangent cone to the active constraint set; we will include a brief illustrative calculation for a two-point mass example to confirm that the resulting subsampling quantiles remain valid for the interval-hull QTE process. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The derivation begins from two explicitly imposed user-specified marginal sensitivity bounds (odds-ratio Γ on source treatment assignment and conditional likelihood-ratio Λ on source-to-target shift). From these, the paper constructs closed-form nested envelopes for the target counterfactual CDF, proves process-level sharpness by direct verification that the envelopes are attainable as entire processes, and inverts them to quantile and QTE bounds. Subsequent semiparametric theory supplies canonical gradients and Neyman-orthogonal estimators whose first-order influence functions are independent of nuisance estimation error. No step reduces a claimed result to a fitted parameter, a self-citation chain, or a definitional tautology; all load-bearing objects (envelopes, sharpness, gradients) are derived from the stated assumptions and standard semiparametric arguments without circular reduction.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on the validity of the two marginal sensitivity models (Γ and Λ) and the stated data structure; these are domain assumptions rather than derived quantities. No new entities are postulated.

free parameters (2)

Γ
User-chosen odds-ratio bound on source treatment assignment; not estimated from data.
Λ
User-chosen conditional likelihood-ratio bound on source-to-target potential-outcome shift; not estimated from data.

axioms (2)

domain assumption Observed data consist of a source sample with treatment, outcome and covariates, and a target sample with covariates only.
Explicitly stated as the data-generating structure in the abstract.
domain assumption Standard causal consistency and no-interference assumptions hold for defining potential outcomes.
Implicit in any potential-outcome framework for QTE but not enumerated in the abstract.

pith-pipeline@v0.9.0 · 5580 in / 1729 out tokens · 41087 ms · 2026-05-12T04:36:15.201007+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel echoes
The sharp source-population potential-outcome CDF bounds are C_Γ^−(p,e)=max{ℓ_Γ(e)p,1−u_Γ(e)(1−p)} ... The sharp transport-shift CDF maps are T_Λ^−(q)=max{Λ^{−1}q,1−Λ(1−q)} ... b_a,s^−(y,x)=T_Λ^−{C_Γ^−(p_a(y,x),e_a(x))}
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability echoes
We prove that the sharp target conditional CDF envelope is ... process-level sharpness, so the envelopes are attainable as entire CDFs

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages · 1 internal anchor

[1]

and Huling, J

Asiaee, A., Pal, S., Beck, C. and Huling, J. D. (2026). Sharp bounds for treatment effect generalization under outcome distribution shift.Proceedings of the 5th Conference on Causal Learning and Reasoning (CLeaR), PMLR

work page 2026
[2]

Ben-Michael, E. (2025). Partial identification via conditional linear programs: estimation and policy learning. arXiv:2506.12215

work page arXiv 2025
[3]

and Pearl, J

Bareinboim, E. and Pearl, J. (2016). Causal inference and the data-fusion problem.Pro- ceedings of the National Academy of Sciences113, 7345–7352

work page 2016
[4]

P., Gelbach, J

Bitler, M. P., Gelbach, J. B. and Hoynes, H. W. (2006). What mean impacts miss: Distri- butional effects of welfare reform experiments.American Economic Review96, 988–1012

work page 2006
[5]

and Courville, A

Goodfellow, I., Bengio, Y. and Courville, A. (2016).Deep Learning. MIT Press

work page 2016
[6]

and Kennedy, E

Bonvini, M. and Kennedy, E. H. (2022). Sensitivity analysis via the proportion of unmeasured confounding.Journal of the American Statistical Association117, 1540–1550

work page 2022
[7]

A., Eron, J

Adimora, A. A., Eron, J. J. and Mugavero, M. J. (2018). Generalizing evidence from randomized trials using inverse probability of sampling weights.Journal of the Royal Statistical Society: Series A181, 1193–1209

work page 2018
[8]

and Melly, B

Chernozhukov, V., Fern´ andez-Val, I. and Melly, B. (2013). Inference on counterfactual distributions.Econometrica81, 2205–2268

work page 2013
[9]

and Luo, Y

Chernozhukov, V., Fern´ andez-Val, I. and Luo, Y. (2018). The sorted effects method: Dis- covering heterogeneous effects beyond their averages.Econometrica86, 1911–1938

work page 2018
[10]

Robins, J. (2018). Double/debiased machine learning for treatment and structural param- eters.The Econometrics Journal21, C1–C68

work page 2018
[11]

Cole, S. R. and Stuart, E. A. (2010). Generalizing evidence from randomized clinical trials to target populations: The ACTG 320 trial.American Journal of Epidemiology172, 107–115

work page 2010
[12]

J., Robertson, S

Dahabreh, I. J., Robertson, S. E., Tchetgen Tchetgen, E. J., Stuart, E. A. and Hern´ an, M. A. (2019). Generalizing causal inferences from individuals in randomized trials to all trial-eligible individuals.Biometrics75, 685–694

work page 2019
[13]

and Rose, S

Degtiar, I. and Rose, S. (2023). A review of generalizability and transportability.Annual Review of Statistics and Its Application10, 501–524

work page 2023
[14]

J., Robins, J

Dahabreh, I. J., Robins, J. M., Haneuse, S. J.-P. A., Saeed, I., Robertson, S. E., Stuart, E. A. and Hern´ an, M. A. (2023). Sensitivity analysis using bias functions for studies extending inferences from a randomized trial to a target population.Statistics in Medicine42, 2029–2043. 47

work page 2023
[15]

M., D’Amour, A

Franks, A. M., D’Amour, A. and Feller, A. (2020). Flexible sensitivity analysis for ob- servational studies without observable implications.Journal of the American Statistical Association115, 1730–1746

work page 2020
[16]

and Feuerriegel, S

Frauen, D., Melnychuk, V. and Feuerriegel, S. (2023). Sharp bounds for generalized causal sensitivity analysis.Advances in Neural Information Processing Systems36, 40556–40586

work page 2023
[17]

Huang, M. Y. (2024). Sensitivity analysis for the generalization of experimental results. Journal of the Royal Statistical Society Series A: Statistics in Society187, 900–918

work page 2024
[18]

and Guo, K

Dorn, J. and Guo, K. (2023). Sharp sensitivity analysis for inverse propensity weighting via quantile balancing.Journal of the American Statistical Association118, 2645–2657

work page 2023
[19]

and Yap, L

Dorn, J. and Yap, L. (2024). Sensitivity analysis for linear estimators. arXiv:2309.06305

work page arXiv 2024
[20]

and Kallus, N

Dorn, J., Guo, K. and Kallus, N. (2025). Doubly-valid/doubly-sharp sensitivity analysis for causal inference with unmeasured confounding.Journal of the American Statistical Association120, 331–342

work page 2025
[21]

and Santos, A

Fang, Z. and Santos, A. (2019). Inference on directionally differentiable functions.Review of Economic Studies86, 377–412

work page 2019
[22]

Firpo, S. (2007). Efficient semiparametric estimation of quantile treatment effects.Econo- metrica75, 259–276

work page 2007
[23]

and Parikh, H

Lanners, Q., Rudin, C., Volfovsky, A. and Parikh, H. (2025). Data fusion for partial identi- fication of causal effects. arXiv:2505.24296

work page arXiv 2025
[24]

Manski, C. F. (2003).Partial Identification of Probability Distributions. Springer

work page 2003
[25]

Masten, M. A. and Poirier, A. (2018). Identification of treatment effects under conditional partial independence.Econometrica86, 317–351

work page 2018
[26]

Masten, M. A. and Poirier, A. (2020). Inference on breakdown frontiers.Quantitative Economics11, 41–111

work page 2020
[27]

A., Poirier, A

Masten, M. A., Poirier, A. and Ren, M. (2025). A general approach to relaxing unconfound- edness. arXiv:2501.15400

work page arXiv 2025
[28]

Q., Ackerman, B., Schmid, I., Cole, S

Nguyen, T. Q., Ackerman, B., Schmid, I., Cole, S. R. and Stuart, E. A. (2018). Sensitiv- ity analyses for effect modifiers not observed in the target population when generalizing treatment effects from a randomized controlled trial.PLOS ONE13, e0208795

work page 2018
[29]

and Wager, S

Nie, X., Imbens, G. and Wager, S. (2021). Covariate balancing sensitivity analysis for extrapolating randomized trials across locations. arXiv:2112.04723

work page arXiv 2021
[30]

and Bareinboim, E

Pearl, J. and Bareinboim, E. (2011). Transportability of causal and statistical relations: A formal approach.Proceedings of the Twenty-Fifth AAAI Conference on Artificial Intelli- gence, 247–254. 48

work page 2011
[31]

N., Romano, J

Politis, D. N., Romano, J. P. and Wolf, M. (1999).Subsampling. Springer

work page 1999
[32]

Rothe, C. (2010). Nonparametric estimation of distributional policy effects.Journal of Econometrics155, 56–70

work page 2010
[33]

A., Cole, S

Stuart, E. A., Cole, S. R., Bradshaw, C. P. and Leaf, P. J. (2011). The use of propensity scores to assess the generalizability of results from randomized trials.Journal of the Royal Statistical Society: Series A174, 369–386

work page 2011
[34]

Tan, Z. (2006). A distributional approach for causal inference using propensity scores.Jour- nal of the American Statistical Association101, 1619–1637

work page 2006
[35]

Tsiatis, A. A. (2006).Semiparametric Theory and Missing Data. Springer. van der Vaart, A. W. (2000).Asymptotic Statistics. Cambridge University Press. van der Vaart, A. W. and Wellner, J. A. (1996).Weak Convergence and Empirical Processes: With Applications to Statistics. Springer

work page 2006
[36]

Wang, P. (2026). Efficient transported distributional and quantile treatment effects with surrogate-assisted missing primary outcomes. arXiv:2605.02154

work page internal anchor Pith review Pith/arXiv arXiv 2026
[37]

H., Bodnar, L

Zeng, Z., Kennedy, E. H., Bodnar, L. M. and Naimi, A. I. (2023). Efficient generalization and transportation. arXiv:2302.00092

work page arXiv 2023
[38]

Zhao, Q., Small, D. S. and Bhattacharya, B. B. (2019). Sensitivity analysis for inverse prob- ability weighting estimators via the percentile bootstrap.Journal of the Royal Statistical Society: Series B81, 735–761. Centers for Disease Control and Prevention, National Center for Health Statistics (2024). NHANES I Epidemiologic Follow-up Study. Public-use d...

work page 2019
[39]

and Barrett, M

Huntington-Klein, N. and Barrett, M. (2024).causaldata: Example Data Sets for Causal Inference Textbooks. R package version 0.1.4,https://github.com/NickCH-K/ causaldata. 49

work page 2024