pith. sign in

arxiv: 2207.11825 · v2 · submitted 2022-07-24 · 📊 stat.ME

Fast convergence rates for dose-response estimation

Matteo Bonvini , Edward H. Kennedy This is my paper

Pith reviewed 2026-05-24 11:57 UTC · model grok-4.3

classification 📊 stat.ME
keywords dose-response estimationcontinuous treatmentshigher-order influence functionspartial meanscausal inferencenonparametric regressionconvergence rates
0
0 comments X

The pith

A higher-order influence function estimator attains the fastest known convergence rate for dose-response curves under continuous treatments.

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

The paper studies estimation of the dose-response function, defined as the expected outcome if the entire population receives a fixed level of a continuous treatment variable. It first analyzes global and local estimators obtained by minimizing empirical risk after plugging in estimated outcomes. The central advance constructs an m-th order estimator from higher-order influence function theory. Under smoothness conditions on the dose-response and sufficiently accurate nuisance estimators, this construction delivers faster mean-square error rates than prior approaches. The work supplies explicit error bounds for all three estimators, demonstrates their behavior in simulations and a real application, and adds a nonparametric sensitivity analysis for the no-unmeasured-confounding assumption.

Core claim

Under standard causal identification assumptions the dose-response function equals a partial mean; an m-th order estimator built from the corresponding higher-order influence function then achieves the fastest rate of convergence known for this problem, provided the dose-response is sufficiently smooth and the nuisance functions are estimated at adequate rates.

What carries the argument

m-th order influence function estimator, which subtracts successive-order bias terms arising from nuisance estimation to raise the achievable convergence rate.

If this is right

  • Mean-square error upper bounds are provided that tighten with larger m when smoothness and nuisance rates hold.
  • The two regression-based estimators remain valid and implementable with standard software even when the higher-order method is impractical.
  • A nonparametric sensitivity analysis for violations of no unmeasured confounding is supplied specifically for continuous treatments.
  • Finite-sample behavior of all three estimators is characterized through simulations and an empirical example.

Where Pith is reading between the lines

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

  • The faster rate could support more reliable individualized dosing or policy recommendations when treatments are measured on a continuous scale.
  • The same higher-order construction may extend to other partial-mean functionals arising in causal inference with continuous exposures.
  • Because the method still requires consistent nuisance estimators, it could be paired with flexible machine-learning nuisance models without losing the rate improvement.

Load-bearing premise

The dose-response equals a partial mean under no unmeasured confounding, positivity and consistency, together with enough smoothness for the m-th order influence function to attain its claimed rate.

What would settle it

A data-generating process satisfying the smoothness and nuisance-rate conditions in which the m-th order estimator's mean-square error does not improve beyond the rate of the first-order or regression-based estimators would falsify the central rate claim.

Figures

Figures reproduced from arXiv: 2207.11825 by Edward H. Kennedy, Matteo Bonvini.

Figure 1
Figure 1. Figure 1: Illustration of the convergence rates in MSE for the estimator considered in this article, as a [PITH_FULL_IMAGE:figures/full_fig_p017_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Examples of estimators of the true nuisance functions [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Estimated MSEs for different estimators of the dose-response across 500 simulations. [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
read the original abstract

We consider the problem of estimating a dose-response curve. Continuous treatments arise often in practice, e.g. in the form of time spent on an operation, distance traveled to a location or dosage of a drug. Letting $A$ denote a continuous treatment variable, the target of inference is the expected outcome if everyone in the population takes treatment level $A=t$. Under standard assumptions, the dose-response function takes the form of a partial mean. Building upon the recent literature on nonparametric regression with estimated outcomes, our first contribution is to study global and local estimators of the dose-response based on empirical risk minimization. Our second and main contribution is to construct a $m^{\text{th}}$-order estimator based on the theory of higher-order influence functions. Under certain conditions, this higher order estimator achieves the fastest rate of convergence that we are aware of for this problem. However, the other two approaches are easier to implement using off-the-shelf software, since they are formulated as two-stage regression tasks. For each estimator, we provide an upper bound on the mean-square error and investigate its finite-sample performance through simulations and an empirical application. Finally, the supplementary material introduces a flexible, nonparametric approach for sensitivity analysis to violations of the no-unmeasured-confounding assumption with continuous treatments.

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 studies estimation of the dose-response curve for a continuous treatment A, which under standard causal assumptions equals a partial mean. It develops global and local estimators via empirical risk minimization (formulated as two-stage regressions) and constructs an m-th order estimator using higher-order influence functions. Upper bounds on mean squared error are derived for each; under suitable conditions on smoothness and nuisance rates the higher-order estimator is claimed to attain the fastest known convergence rate. Finite-sample performance is examined via simulations and a real-data application, with supplementary material on sensitivity analysis for the no-unmeasured-confounding assumption.

Significance. A higher-order influence-function estimator that demonstrably attains the fastest known rate for continuous-treatment dose-response estimation would constitute a meaningful methodological contribution to nonparametric causal inference. The explicit MSE bounds, the comparison with simpler two-stage regression procedures, and the sensitivity-analysis supplement are all positive features that would increase the work's utility if the rate claim is rigorously established.

major comments (2)
  1. [Abstract and §2] Abstract and §2: the central claim that the m-th order estimator 'achieves the fastest rate of convergence that we are aware of' is load-bearing yet stated only conditionally ('under certain conditions') without an explicit theorem statement giving the precise rate (in terms of n, smoothness index, and m) or the exact nuisance-rate requirements; this makes independent verification of the improvement over existing partial-mean estimators difficult.
  2. [§4] §4 (higher-order estimator construction): the MSE upper bound is asserted to follow from m-th order influence-function theory, but the manuscript does not verify that the partial-mean functional satisfies the requisite higher-order differentiability and remainder conditions needed for the rate to materialize; without this check the claimed improvement rests on an external reference rather than a self-contained argument.
minor comments (3)
  1. [Introduction] Notation for the target functional (partial mean) and the estimated outcome regression should be introduced once and used consistently; several places reuse symbols for the treatment variable and the dose level.
  2. [Simulations] Simulation section: the choice of tuning parameters (bandwidths, order m, nuisance estimators) is not described in sufficient detail to allow exact replication of the reported MSE curves.
  3. [Abstract] The supplementary sensitivity analysis is mentioned only in the abstract; a one-paragraph pointer in the main text would help readers locate it.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. We address each major comment below and indicate the revisions that will be incorporated.

read point-by-point responses

  1. Referee: [Abstract and §2] Abstract and §2: the central claim that the m-th order estimator 'achieves the fastest rate of convergence that we are aware of' is load-bearing yet stated only conditionally ('under certain conditions') without an explicit theorem statement giving the precise rate (in terms of n, smoothness index, and m) or the exact nuisance-rate requirements; this makes independent verification of the improvement over existing partial-mean estimators difficult.

    Authors: We agree that an explicit theorem statement would strengthen the presentation. In the revision we will add a theorem in Section 2 that states the precise convergence rate of the m-th order estimator as a function of n, the smoothness index, m, and the required nuisance rates, allowing direct comparison with existing partial-mean results. revision: yes

  2. Referee: [§4] §4 (higher-order estimator construction): the MSE upper bound is asserted to follow from m-th order influence-function theory, but the manuscript does not verify that the partial-mean functional satisfies the requisite higher-order differentiability and remainder conditions needed for the rate to materialize; without this check the claimed improvement rests on an external reference rather than a self-contained argument.

    Authors: The referee correctly notes the reliance on external theory. In the revised §4 we will add an explicit verification that the partial-mean dose-response functional satisfies the higher-order differentiability and remainder conditions under the maintained smoothness assumptions on the outcome regression and propensity score. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation rests on external higher-order IF theory

full rationale

The paper's central contribution constructs an m-th order estimator for the dose-response (partial mean) and derives MSE upper bounds under smoothness and nuisance rate conditions. These bounds follow from standard higher-order influence function theory applied to the identified functional; the abstract and description explicitly condition the fastest-rate claim on external technical conditions rather than deriving the rate from a fitted parameter or self-referential definition. No self-citation is invoked as a uniqueness theorem or load-bearing premise for the rate itself, and the two-stage regression estimators are presented as simpler alternatives without claiming they are predictions of the higher-order result. The derivation chain is therefore self-contained against external benchmarks and does not reduce any claimed prediction to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Abstract-only review; standard causal identification assumptions and rate conditions on nuisances are invoked but not enumerated. No free parameters or invented entities are visible in the abstract.

axioms (2)
  • domain assumption Dose-response function identified as partial mean under no unmeasured confounding, positivity, and consistency.
    Stated in abstract as 'under standard assumptions'.
  • domain assumption Existence of m-th order influence functions that yield the claimed rate under unspecified smoothness and nuisance rate conditions.
    Central to the main contribution described in the abstract.

pith-pipeline@v0.9.0 · 5747 in / 1297 out tokens · 18667 ms · 2026-05-24T11:57:29.805947+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages · 1 internal anchor

  1. [1]

    A Unified Framework for Efficient Estimation of General Treatment Models

    Chunrong Ai, Oliver Linton, Kaiji Motegi, and Zheng Zhang. A unified framework for efficient estimation of general treatment models. arXiv preprint arXiv:1808.04936, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  2. [2]

    Some new asymptotic theory for least squares series: Pointwise and uniform results

    Alexandre Belloni, Victor Chernozhukov, Denis Chetverikov, and Kengo Kato. Some new asymptotic theory for least squares series: Pointwise and uniform results. Journal of Econometrics, 186 0 (2): 0 345--366, 2015

    work page 2015

  3. [3]

    Asymptotically exact minimax estimation in sup-norm for anisotropic h \"o lder classes

    Karine Bertin. Asymptotically exact minimax estimation in sup-norm for anisotropic h \"o lder classes. Bernoulli, 10 0 (5): 0 873--888, 2004

    work page 2004

  4. [4]

    Sensitivity analysis for marginal structural models

    Matteo Bonvini, Edward Kennedy, Valerie Ventura, and Larry Wasserman. Sensitivity analysis for marginal structural models. Working manuscript, 2022

    work page 2022

  5. [5]

    Double debiased machine learning nonparametric inference with continuous treatments

    Kyle Colangelo and Ying-Ying Lee. Double debiased machine learning nonparametric inference with continuous treatments. arXiv preprint arXiv:2004.03036, 2020

    work page arXiv 2004

  6. [6]

    Targeted data adaptive estimation of the causal dose--response curve

    Iv \'a n D \' az and Mark J van der Laan. Targeted data adaptive estimation of the causal dose--response curve. Journal of Causal Inference, 1 0 (2): 0 171--192, 2013

    work page 2013

  7. [7]

    Doubly-valid/doubly-sharp sensitivity analysis for causal inference with unmeasured confounding

    Jacob Dorn, Kevin Guo, and Nathan Kallus. Doubly-valid/doubly-sharp sensitivity analysis for causal inference with unmeasured confounding. arXiv preprint arXiv:2112.11449, 2021

    work page arXiv 2021

  8. [8]

    Conditional density estimation in a regression setting

    Sam Efromovich. Conditional density estimation in a regression setting. The Annals of Statistics, 35 0 (6): 0 2504--2535, 2007

    work page 2007

  9. [9]

    Local polynomial modelling and its applications

    Jianqing Fan and Irene Gijbels. Local polynomial modelling and its applications. Routledge, 2018

    work page 2018

  10. [10]

    Estimation of dose-response functions and optimal doses with a continuous treatment

    Carlos A Flores. Estimation of dose-response functions and optimal doses with a continuous treatment. University of Miami, Department of Economics, November, 2007

    work page 2007

  11. [11]

    Orthogonal statistical learning

    Dylan J Foster and Vasilis Syrgkanis. Orthogonal statistical learning. arXiv preprint arXiv:1901.09036, 2019

    work page arXiv 1901

  12. [12]

    Uniformly semiparametric efficient estimation of treatment effects with a continuous treatment

    Antonio F Galvao and Liang Wang. Uniformly semiparametric efficient estimation of treatment effects with a continuous treatment. Journal of the American Statistical Association, 110 0 (512): 0 1528--1542, 2015

    work page 2015

  13. [13]

    Random rates in anisotropic regression (with a discussion and a rejoinder by the authors)

    M Hoffman and Oleg Lepski. Random rates in anisotropic regression (with a discussion and a rejoinder by the authors). The Annals of Statistics, 30 0 (2): 0 325--396, 2002

    work page 2002

  14. [14]

    Optimal doubly robust estimation of heterogeneous causal effects

    Edward H Kennedy. Optimal doubly robust estimation of heterogeneous causal effects. arXiv preprint arXiv:2004.14497, 2020

    work page arXiv 2004

  15. [15]

    Semiparametric doubly robust targeted double machine learning: a review

    Edward H Kennedy. Semiparametric doubly robust targeted double machine learning: a review. arXiv preprint arXiv:2203.06469, 2022

    work page arXiv 2022

  16. [16]

    Nonparametric methods for doubly robust estimation of continuous treatment effects

    Edward H Kennedy, Zongming Ma, Matthew D McHugh, and Dylan S Small. Nonparametric methods for doubly robust estimation of continuous treatment effects. Journal of the Royal Statistical Society. Series B, Statistical Methodology, 79 0 (4): 0 1229, 2017

    work page 2017

  17. [17]

    Robust inference with higher order influence functions: Parts i and ii

    Lingling Li, Eric Tchetgen, J Robins, and A van der Vaart. Robust inference with higher order influence functions: Parts i and ii. In Joint Statistical Meetings, Minneapolis, Minnesota, 2005

    work page 2005

  18. [18]

    Semiparametric efficient empirical higher order influence function estimators

    Rajarshi Mukherjee, Whitney K Newey, and James M Robins. Semiparametric efficient empirical higher order influence function estimators. arXiv preprint arXiv:1705.07577, 2017

    work page arXiv 2017

  19. [19]

    Nonparametric causal effects based on marginal structural models

    Romain Neugebauer and Mark van der Laan. Nonparametric causal effects based on marginal structural models. Journal of Statistical Planning and Inference, 137 0 (2): 0 419--434, 2007

    work page 2007

  20. [20]

    Kernel estimation of partial means and a general variance estimator

    Whitney K Newey. Kernel estimation of partial means and a general variance estimator. Econometric Theory, pages 233--253, 1994

    work page 1994

  21. [21]

    Higher order influence functions and minimax estimation of nonlinear functionals

    James Robins, Lingling Li, Eric Tchetgen, Aad van der Vaart, et al. Higher order influence functions and minimax estimation of nonlinear functionals. In Probability and statistics: essays in honor of David A. Freedman, pages 335--421. Institute of Mathematical Statistics, 2008

    work page 2008

  22. [22]

    Quadratic semiparametric von mises calculus

    James Robins, Lingling Li, Eric Tchetgen, and Aad W van der Vaart. Quadratic semiparametric von mises calculus. Metrika, 69 0 (2): 0 227--247, 2009 a

    work page 2009

  23. [23]

    Semiparametric minimax rates

    James Robins, Eric Tchetgen Tchetgen, Lingling Li, and Aad van der Vaart. Semiparametric minimax rates. Electronic journal of statistics, 3: 0 1305, 2009 b

    work page 2009

  24. [24]

    Higher order estimating equations for high-dimensional models

    James Robins, Lingling Li, Rajarshi Mukherjee, Eric Tchetgen Tchetgen, and Aad van der Vaart. Higher order estimating equations for high-dimensional models. Annals of statistics, 45 0 (5): 0 1951, 2017

    work page 1951

  25. [25]

    Marginal structural models versus structural nested models as tools for causal inference

    James M Robins. Marginal structural models versus structural nested models as tools for causal inference. In Statistical models in epidemiology, the environment, and clinical trials, pages 95--133. Springer, 2000

    work page 2000

  26. [26]

    Estimating causal effects of treatments in randomized and nonrandomized studies

    Donald B Rubin. Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of educational Psychology, 66 0 (5): 0 688, 1974

    work page 1974

  27. [27]

    Debiased machine learning of conditional average treatment effects and other causal functions

    Vira Semenova and Victor Chernozhukov. Debiased machine learning of conditional average treatment effects and other causal functions. arXiv preprint arXiv:1702.06240, 2017

    work page arXiv 2017

  28. [28]

    Reproducing kernel methods for nonparametric and semiparametric treatment effects

    Rahul Singh, Liyuan Xu, and Arthur Gretton. Reproducing kernel methods for nonparametric and semiparametric treatment effects. arXiv preprint arXiv:2010.04855, 2020

    work page arXiv 2010

  29. [29]

    Introduction to nonparametric estimation

    Alexandre B Tsybakov. Introduction to nonparametric estimation. Springer Science & Business Media, 2008

    work page 2008

  30. [30]

    Statistical inference for variable importance

    Mark J Van der Laan. Statistical inference for variable importance. The International Journal of Biostatistics, 2 0 (1), 2006

    work page 2006

  31. [31]

    Unified methods for censored longitudinal data and causality

    Mark J Van der Laan, MJ Laan, and James M Robins. Unified methods for censored longitudinal data and causality. Springer Science & Business Media, 2003

    work page 2003

  32. [32]

    Higher order tangent spaces and influence functions

    Aad van der Vaart. Higher order tangent spaces and influence functions. Statistical Science, pages 679--686, 2014

    work page 2014

  33. [33]

    High-dimensional statistics: A non-asymptotic viewpoint, volume 48

    Martin J Wainwright. High-dimensional statistics: A non-asymptotic viewpoint, volume 48. Cambridge University Press, 2019

    work page 2019

  34. [34]

    All of nonparametric statistics

    Larry Wasserman. All of nonparametric statistics. Springer Science & Business Media, 2006

    work page 2006

  35. [35]

    A unified study of nonparametric inference for monotone functions

    Ted Westling and Marco Carone. A unified study of nonparametric inference for monotone functions. Annals of statistics, 48 0 (2): 0 1001, 2020

    work page 2020

  36. [36]

    Causal isotonic regression

    Ted Westling, Peter Gilbert, and Marco Carone. Causal isotonic regression. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 82 0 (3): 0 719--747, 2020

    work page 2020