Pith. sign in

REVIEW 3 major objections 4 minor 80 references

Conditional Mean Independence and Global Sensitivity Analysis using Nearest Neighbor Graphs

T0 review · 3 major / 4 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read A near-linear nearest-neighbor statistic estimates the fraction of response variance explained by the conditional mean and yields an asymptotic normal test for conditional mean independence without bootstrap or sample splitting.

desk verdict Clean, near-linear NN estimator of multivariate mean-dependence / Sobol' index with a pivotal null CLT and consistent screening; solid package, standard nonparametric caveats. read the letter →

arxiv 2607.04692 v2 pith:CKI3UJZL submitted 2026-07-06 stat.ME math.STstat.TH

Anirban Chatterjee , Ziang Niu , Bhaswar B. Bhattacharya This is my paper · ORCID
classification stat.MEmath.STstat.TH MSC 62G1062G2062H20
keywords conditionalmeanindependencenearestneighborgraphsSobol'indicesglobalsensitivityanalysisvariablescreeningSteinmethodnonparametrictesting
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 introduces a simple nonparametric measure, the normalized conditional mean discrepancy, that equals the proportion of variation in a (possibly multivariate) response explained by its regression function on covariates. That same quantity is also a multivariate Sobol' index used in global sensitivity analysis. The authors construct a nearest-neighbor-graph estimator of the measure that can be computed in near-linear time, prove consistency and rates of convergence, and show that a studentized version of the numerator is asymptotically standard normal under the null of conditional mean independence. The resulting test therefore needs neither bootstrap calibration nor sample splitting, yet is universally consistent against fixed alternatives. The same estimator is turned into a sequential variable-screening algorithm that recovers a sufficient set of covariates with high probability, and the framework extends to second-order Sobol' indices that isolate pure interaction effects.

What carries the argument

The normalized conditional mean discrepancy η (equal to the multivariate Sobol' index Tr(Σ_fX)/Tr(Σ_Y)) estimated by averaging inner products of responses over the directed K-nearest-neighbor graph on the covariates; its studentized numerator supplies the pivotal test statistic.

What would settle it

Under a fixed alternative where E[Y|X] is non-constant, check whether the empirical rejection rate of the studentized nearest-neighbor test tends to 1 as sample size grows while remaining near the nominal level under pure noise or pure heteroskedasticity with constant conditional mean.

Watch

Extended reading notes

Core claim

Under mild moment and continuity conditions, the studentized nearest-neighbor statistic based on the numerator of the normalized conditional mean discrepancy converges in Kolmogorov distance to a standard normal under the null of conditional mean independence, producing a level-α test that is universally consistent without bootstrap or sample splitting.

Load-bearing premise

The conditional mean function must satisfy a local Lipschitz condition whose constant grows at most polynomially with the covariates; without that regularity the estimator's rate can be arbitrarily slow.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper introduces the Normalized Conditional Mean Discrepancy (NCMD) η, the proportion of response variance explained by the conditional mean E[Y|X], equivalently a multivariate Sobol' index. It constructs a K-nearest-neighbor graph estimator η̂n that is consistent (Theorem 3.1) under continuous pairwise distances and 4+δ moments, and obtains rates under local Lipschitz and sub-exponential tails (Theorem 3.2). Under the null of conditional mean independence the numerator Tn is unbiased, and a studentized statistic √n Tn/σ̂n is shown to be asymptotically standard normal via Stein's method on a dependency graph (Theorem 4.1), yielding a level-α, universally consistent test without bootstrap or sample splitting (Corollary 4.1). The same estimator drives a sequential model-free screening algorithm with exponential consistency (Theorem 5.1) and extends to second-order Sobol' indices (Theorem 6.1). Simulations and a California Housing example compare power, Type-I error, and runtime against MDD, pMIT, dCov, and Azadkia–Chatterjee methods.

Significance. If the results hold, the paper supplies a near-linear-time, bootstrap-free, sample-splitting-free test for conditional mean independence that is pivotal under the null and universally consistent, together with a matching model-free screener and a practical estimator of multivariate and second-order Sobol' indices. The combination of Stein CLT + consistent studentization under only continuous-distance and moment assumptions, the explicit near-linear complexity, and the publicly released code are concrete strengths that distinguish the contribution from existing MDD, kernel-smoothing, and ML-based procedures. The work is therefore of clear interest to nonparametric inference, global sensitivity analysis, and high-dimensional screening.

major comments (3)
  1. The central asymptotic-normality claim (Theorem 4.1 / Corollary 4.1) is load-bearing and appears correctly proved: under H0 the bias vanishes (Remark 4.1, eq. (4.2)), so only Assumption 3.1 and 8+δ moments are required; the local-Lipschitz condition of Assumption 3.2(2) is used only for rates and screening. No internal gap that would invalidate the null CLT or universal consistency was found.
  2. Section 5 / Theorem 5.1: the screening guarantee relies on a uniform gap δ > 0 for every insufficient set (condition (a)). This is a strong, non-checkable assumption that is not discussed in the simulations of Section 7.2; a short remark on how the procedure behaves when the gap is small or vanishes would strengthen the claim that the algorithm is 'provably consistent' in practice.
  3. Remark 4.3 and the discussion after Theorem 3.2 correctly note that the method does not yet handle conditional mean independence given additional covariates without further debiasing. Because many competing procedures (MDD, pMIT) already address the partial setting, a clearer statement of this limitation (and whether the same NN graph can be adapted) would help readers assess the scope of the contribution.
minor comments (4)
  1. Typographical slips: 'dominatat' (p. 7), 'mong those' (p. 5), 'V ariable' (section headings), and 'T esting' (Section 7.1). A careful proof-reading pass is needed.
  2. Figures 1–6 report average execution times in milliseconds; stating the hardware and whether times include graph construction would make the computational claims fully reproducible.
  3. The free parameter K is fixed at 5 or 10 throughout the experiments. A short sensitivity check or a data-driven rule of thumb would be useful for practitioners.
  4. Proposition 2.1 cites earlier appearances of the same measure in [65, 71]; making the novelty claim more precise (estimator + CLT + screening, rather than the population index itself) would avoid any impression of over-claiming.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: population target, estimator, null CLT, and screening guarantees are derived independently of one another.

full rationale

The population NCMD η is defined in (2.3) from the L2 discrepancy of the conditional mean and is shown in Proposition 2.1 to satisfy the three Renyi-type axioms without reference to any estimator. The nearest-neighbor estimator η̂n in (3.6) is constructed from the K-NN graph and U-statistic terms; consistency (Theorem 3.1) and rates (Theorem 3.2) are proved from first principles via Efron–Stein, covering arguments, and local Lipschitz control (Assumption 3.2). Under H0 the bias vanishes identically (4.2), so the studentized statistic √n Tn/σ̂n is shown asymptotically N(0,1) by a dependency-graph Stein CLT (Proposition D.1) plus consistent variance estimation (Proposition D.2); the resulting test (Corollary 4.1) is therefore level-α and universally consistent by construction of the limit, not by fitting. The screening algorithm (Algorithm 1) and its exponential-error guarantee (Theorem 5.1) likewise rest on concentration of V̂n about V, not on any circular identification of the sufficient set. Self-citations are to standard external tools (Efron–Stein, Hanson–Wright, Chen–Shao dependency-graph CLT, covering numbers) or to the authors’ own earlier technical lemmas that are re-proved or adapted in the appendices; none of them is load-bearing for the definition of η or for the pivotal null limit. No free parameter is fitted to produce the asymptotic claims, and no uniqueness theorem is imported to forbid alternatives. The derivation chain is therefore self-contained.

Assumptions & free parameters 1 free parameters · 5 assumptions · 1 invented entities

The central claims rest on standard moment and geometric assumptions for nearest-neighbor graphs plus a local Lipschitz condition on the regression function. No free parameters are fitted to obtain the asymptotic statements; K is a fixed user choice. The only invented object is the NCMD measure itself, which is a transparent re-packaging of a multivariate Sobol' index already studied by Gamboa et al.

free parameters (1)
  • K (number of nearest neighbors)
    User-chosen integer that appears in the estimator and in the variance formula; asymptotic statements hold for any fixed K ≥ 1, but finite-sample performance depends on the choice.
assumptions (5)
  • domain assumption Assumption 3.1: the Euclidean distance between two independent copies of X has a continuous distribution (so the K-NN graph is a.s. well-defined).
    Standard non-degeneracy condition for nearest-neighbor graphs; invoked throughout Sections 3–6.
  • domain assumption Assumption 3.2(1): sub-exponential tails on X and on Y − EY.
    Used to control maximal inequalities and bias terms in the rate proof (Theorem 3.2).
  • domain assumption Assumption 3.2(2): local Lipschitz condition on g(x) = E[Y|X=x] with polynomially growing constant.
    Essential for the bias rate; without it the rate can be arbitrarily slow (Remark 3.3).
  • domain assumption Finite moments E‖Y‖^{4+δ} < ∞ (consistency) and E‖Y‖^{8+δ} < ∞ (CLT).
    Standard moment conditions for U-statistic-type arguments and Stein's method.
  • domain assumption Existence of a sufficient set of size at most κ = ⌊ M/δ⌋+1 with a uniform signal gap δ > 0 (screening).
    Assumption (a) of Theorem 5.1; without a uniform gap the sequential algorithm need not stop at a sufficient set.
invented entities (1)
  • Normalized Conditional Mean Discrepancy (NCMD) η independent evidence
    purpose: Population target that unifies conditional-mean dependence measurement and multivariate Sobol' indices.
    Defined in (2.3) as the ratio of explained to total variance; already appears (under different names) in the Sobol' and mMSE-gap literature, so independent evidence is strong.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Conditional Mean Independence and Global Sensitivity Analysis using Nearest Neighbor Graphs." pith.science (2026). https://pith.science/paper/CKI3UJZL

@misc{pith2026260704692,
  author       = {Pith},
  title        = {Pith review of: Conditional Mean Independence and Global Sensitivity Analysis using Nearest Neighbor Graphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CKI3UJZL}},
  note         = {Machine review of arXiv:2607.04692}
}
read the original abstract

Quantifying how well a conditional mean function explains a response is central to many statistical tasks, such as model evaluation and feature screening. A basic nonparametric measure of such dependence is the proportion of variation in the response explained by the regression function, which can also be interpreted as a multivariate Sobol' index, a fundamental notion in global sensitivity analysis. In this paper, we propose a consistent estimator of this measure based on nearest neighbor graphs that can be computed in near-linear time. We also derive its rate of convergence and show that a studentized version of the estimator is asymptotically standard normal under the null hypothesis of conditional mean independence. This leads to a computationally efficient test for conditional mean independence that attains the correct asymptotic level and is universally consistent, without requiring bootstrap calibration or sample splitting. Next, we use the proposed estimator to develop a model-free variable screening algorithm that is provably consistent. We also discuss extensions of the framework to measuring interaction effects using higher-order Sobol' indices. The benefits of the proposed methods are demonstrated through simulation studies and a real-data example.

Figures

Figures reproduced from arXiv: 2607.04692 by the authors.

Figure 1
Figure 1. Empirical Type I error/power and (b) computational time for conditional mean independence testing for the settings in Example 7.1, when X „ Uniformr´1, 1s. pMIT directly test conditional mean independence, while dCov and the Chatterjee correlation test for general statistical dependence. Example 7.1. We consider the following response models with univariate predictors, motivated by the examples in [9, 20]. In each c… view at source ↗
Figure 2
Figure 2. Empirical Type I error/power and (b) computational time for conditional mean independence testing for the settings in Example 7.2, when X „ Uniformpr´1, 1s 10q. ‚ Radial: Y “ cosprq ` λε, where r “ ? 1 5 }XS}2 and S Ă rds is chosen uniformly at random among subsets of size 5. ‚ Nonlinear Interaction: Y “ sinpX1q ` cospX2qX3 ` λε. Note that for the Nonlinear Additive, Interaction, Radial, and Nonlinear Interaction mo… view at source ↗
Figure 3
Figure 3. (a) Empirical Type I error/power and (b) computational time for conditional mean independence testing for the settings in Example 7.1, X „ Np0, 1q. 0.00 0.25 0.50 0.75 1.00 0 .25 .50 .75 1 Rejection rate Circular 0.00 0.25 0.50 0.75 1.00 0 .25 .50 .75 1 Rejection rate Heteroskedastic Type I 0.00 0.25 0.50 0.75 1.00 0 .25 .50 .75 1 Rejection rate Linear 0.00 0.25 0.50 0.75 1.00 0 .25 .50 .75 1 Rejection rate Step 0.0… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: (a) Empirical Type I error/power and (b) computational time for conditional mean independence testing for the settings in Example 7.1, when X “ 2U ´ 1 and U „ Betap0.1, 0.1q [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]
Figure 5
Figure 5. Figure 5: (a) Empirical Type I error/power and (b) computational time for conditional mean independence testing for the settings in Example 7.2, when X „ N10p0, I10q. 0.2 0.4 0.6 0.8 0 .25 .50 .75 1 Rejection rate Noise 0.2 0.4 0.6 0.8 0 .25 .50 .75 1 Rejection rate Heteroskedas…
Figure 6
Figure 6. Figure 6: (a) Empirical Type I error/power and (b) computational time for conditional mean independence testing for the settings in Example 7.2, when X “ 2U ´ 1 and the coordinates of U are i.i.d. Betap0.1, 0.1q. A.2. Variable Screening for California Housing Dataset. In this se…
Figure 7
Figure 7. Figure 7: compares the estimated indices ˆηX1 , ˆηX2 , and ˆη2 with their corresponding population values ηX1 , ηX2 , and η2. Specifically, in [PITH_FULL_IMAGE:figures/full_fig_p025_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

80 extracted references · 7 linked inside Pith

  1. [1]

    Aıt-Sahalia, P

    Y. Aıt-Sahalia, P. J. Bickel, and T. M. Stoker. Goodness-of-fit tests for kernel regression with an application to option implied volatilities.Journal of Econometrics, 105(2):363–412, 2001

    2001

  2. [2]

    Azadkia and S

    M. Azadkia and S. Chatterjee. A simple measure of conditional dependence.The Annals of Statistics, 49(6):3070–3102, 2021

    2021

  3. [3]

    Borgonovo, A

    E. Borgonovo, A. Figalli, P. Ghosal, E. Plischke, and G. Savar´ e. Convexity and measures of statistical association.Journal of the Royal Statistical Society Series B: Statistical Methodology, 87(4):1281–1304, 2025

    2025

  4. [4]

    Borgonovo, A

    E. Borgonovo, A. Figalli, E. Plischke, and G. Savar´ e. Global sensitivity analysis via optimal transport. Management Science, 71(5):3809–3828, 2025

    2025

  5. [5]

    L. Cai, X. Guo, and W. Zhong. Test and measure for partial mean dependence based on machine learning methods.Journal of the American Statistical Association, 120(550):833–845, 2025

    2025

  6. [6]

    Chastaing, F

    G. Chastaing, F. Gamboa, and C. Prieur. Generalized hoeffding-sobol decomposition for dependent variables-application to sensitivity analysis.Electronic Journal of Statistics, 6:2420–2448, 2012

    2012

  7. [7]

    Chatterjee, Z

    A. Chatterjee, Z. Niu, and B. B. Bhattacharya. A kernel-based conditional two-sample test using nearest neighbors (with applications to calibration, regression curves, and simulation-based inference). arXiv:2407.16550, 2024

    arXiv 2024

  8. [8]

    Chatterjee, S

    A. Chatterjee, S. Choudhury, and R. Hore. One-shot conditional sampling: MMD meets nearest neigh- bors.arXiv:2509.25507, 2025

    arXiv 2025

Show all 80 references
  1. [9]

    Chatterjee

    S. Chatterjee. A new coefficient of correlation.Journal of the American Statistical Association, 116 (536):2009–2022, 2021

    2009

  2. [10]

    Chatterjee

    S. Chatterjee. A survey of some recent developments in measures of association. In S. Athreya, A. G. Bhatt, and B. V. Rao, editors,Probability and Stochastic Processes, Indian Statistical Institute Series, pages 109–128. Springer, 2024

    2024

  3. [11]

    L. H. Chen and Q.-M. Shao. Normal approximation under local dependence.The Annals of Probability, 32(3):1985–2028, 2004

    1985

  4. [12]

    Chen and C

    T. Chen and C. Guestrin. Xgboost: A scalable tree boosting system. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 785–794, 2016

    2016

  5. [13]

    Cheng, Z

    K. Cheng, Z. Lu, and K. Zhang. Multivariate output global sensitivity analysis using multi-output support vector regression.Structural and Multidisciplinary Optimization, 59(6):2177–2187, 2019

    2019

  6. [14]

    Cheng and H

    X. Cheng and H. Wang. A generic model-free feature screening procedure for ultra-high dimensional data with categorical response.Computer Methods and Programs in Biomedicine, 229:107269, 2023

    2023

  7. [15]

    Chhaibi, F

    R. Chhaibi, F. Gamboa, and C. Pellegrini. A martingale approach to fluctuations of rank estimators in sensitivity analysis.arXiv:2603.22611, 2026

    2026

  8. [16]

    R. D. Cook and B. Li. Dimension reduction for conditional mean in regression.The Annals of Statistics, 30(2):455–474, 2002

    2002

  9. [17]

    Da Veiga, F

    S. Da Veiga, F. Wahl, and F. Gamboa. Local polynomial estimation for sensitivity analysis on models with correlated inputs.Technometrics, 51(4):452–463, 2009

    2009

  10. [18]

    B. Dai, X. Shen, and W. Pan. Significance tests of feature relevance for a black-box learner.IEEE transactions on neural networks and learning systems, 35(2):1898–1911, 2022

    1911

  11. [19]

    de Rocquigny, N

    E. de Rocquigny, N. Devictor, and S. Tarantola.Uncertainty in industrial practice: a guide to quanti- tative uncertainty management. John Wiley & Sons, 2008. CONDITIONAL MEAN INDEPENDENCE AND GLOBAL SENSITIVITY ANALYSIS 19

    2008

  12. [20]

    N. Deb, P. Ghosal, and B. Sen. Measuring association on topological spaces using kernels and geometric graphs.arXiv:2010.01768, 2020

    2010 arXiv

  13. [21]

    M. A. Delgado and W. G. Manteiga. Significance testing in nonparametric regression based on the bootstrap.The Annals of Statistics, 29(5):1469–1507, 2001

    2001

  14. [22]

    Dette, K

    H. Dette, K. F. Siburg, and P. A. Stoimenov. A copula-based non-parametric measure of regression dependence.Scandinavian Journal of Statistics, 40(1):21–41, 2013

    2013

  15. [23]

    Efron and C

    B. Efron and C. Stein. The jackknife estimate of variance.The Annals of Statistics, pages 586–596, 1981

    1981

  16. [24]

    Efron and R

    B. Efron and R. Tibshirani. On testing the significance of sets of genes.The Annals of Applied Statistics, 1(1):107 – 129, 2007

    2007

  17. [25]

    Fan and J

    J. Fan and J. Lv. Sure independence screening for ultrahigh dimensional feature space.Journal of the Royal Statistical Society: Series B (Statistical Methodology), 70(5):849–911, 2008

    2008

  18. [26]

    J. Fan, Y. Ma, and W. Dai. Nonparametric independence screening in sparse ultra-high-dimensional varying coefficient models.Journal of the American Statistical Association, 109(507):1270–1284, 2014

    2014

  19. [27]

    J. Fan, Y. Feng, and L. Xia. A projection-based conditional dependence measure with applications to high-dimensional undirected graphical models.Journal of Econometrics, 218(1):119–139, 2020

    2020

  20. [28]

    Fan and Q

    Y. Fan and Q. Li. Consistent model specification tests: Omitted variables and semiparametric functional forms.Econometrica, 64(4):865–890, 1996

    1996

  21. [29]

    J. H. Friedman, J. L. Bentley, and R. A. Finkel. An algorithm for finding best matches in logarithmic expected time.ACM Transactions on Mathematical Software (TOMS), 3(3):209–226, 1977

    1977

  22. [30]

    Gamboa, A

    F. Gamboa, A. Janon, T. Klein, and A. Lagnoux. Sensitivity indices for multivariate outputs.Comptes Rendus. Math´ ematique, 351(7-8):307–310, 2013

    2013

  23. [31]

    Gamboa, A

    F. Gamboa, A. Janon, T. Klein, and A. Lagnoux. Sensitivity analysis for multidimensional and func- tional outputs.Electronic Journal of Statistics, 8(1):575–603, 2014

    2014

  24. [32]

    Gamboa, A

    F. Gamboa, A. Janon, T. Klein, A. Lagnoux, and C. Prieur. Statistical inference for sobol pick-freeze monte carlo method.Statistics, 50(4):881–902, 2016

    2016

  25. [33]

    Gamboa, P

    F. Gamboa, P. Gremaud, T. Klein, and A. Lagnoux. Global sensitivity analysis: A novel generation of mighty estimators based on rank statistics.Bernoulli, 28(4):2345–2374, 2022

    2022

  26. [34]

    L. Gan, N. N. Narisetty, and F. Liang. Bayesian regularization for graphical models with unequal shrinkage.Journal of the American Statistical Association, 114(527):1218–1231, 2019

    2019

  27. [35]

    Garcia-Cabrejo and A

    O. Garcia-Cabrejo and A. Valocchi. Global sensitivity analysis for multivariate output using polynomial chaos expansion.Reliability Engineering & System Safety, 126:25–36, 2014

    2014

  28. [36]

    Huang, N

    Z. Huang, N. Deb, and B. Sen. Kernel partial correlation coefficient—a measure of conditional depen- dence.Journal of Machine Learning Research, 23(216):1–58, 2022

    2022

  29. [37]

    Jaffe, Y

    A. Jaffe, Y. Kluger, G. C. Linderman, G. Mishne, and S. Steinerberger. Randomized near-neighbor graphs, giant components and applications in data science.Journal of applied probability, 57(2):458– 476, 2020

    2020

  30. [38]

    Janon, T

    A. Janon, T. Klein, A. Lagnoux, M. Nodet, and C. Prieur. Asymptotic normality and efficiency of two sobol index estimators.ESAIM: Probability and Statistics, 18:342–364, 2014

    2014

  31. [39]

    Z. Jin, X. Yan, and D. S. Matteson. Testing for conditional mean independence with covariates through martingale difference divergence. InProceedings of the Thirty-Fourth Conference on Uncertainty in Artificial Intelligence,, pages 1–12, 2018

    2018

  32. [40]

    Josse and S

    J. Josse and S. Holmes. Measuring multivariate association and beyond.Statistics surveys, 10:132, 2016

    2016

  33. [41]

    T. Lai, Z. Zhang, and Y. Wang. A kernel-based measure for conditional mean dependence.Computa- tional Statistics & Data Analysis, 160:107246, 2021

    2021

  34. [42]

    Lavergne and Q

    P. Lavergne and Q. Vuong. Nonparametric significance testing.Econometric Theory, 16(4):576–601, 2000

    2000

  35. [43]

    C. Lee, X. Zhang, and X. Shao. Testing conditional mean independence for functional data.Biometrika, 107(2):331–346, 2020

    2020

  36. [44]

    R. Li, W. Zhong, and L. Zhu. Feature screening via distance correlation learning.Journal of the American Statistical Association, 107(499):1129–1139, 2012

    2012

  37. [45]

    Z. R. Li and T. H. McCormick. An expectation conditional maximization approach for gaussian graphical models.Journal of Computational and Graphical Statistics, 28(4):767–777, 2019. 20 CHATTERJEE, NIU, AND BHATTACHARYA

    2019

  38. [46]

    Z. R. Li, T. H. McCormick, and S. J. Clark. Bayesian joint spike-and-slab graphical lasso. arXiv:1805.07051, 2018

    2018 arXiv

  39. [47]

    Lin and F

    Z. Lin and F. Han. Limit theorems of Chatterjee’s rank correlation.arXiv:2204.08031, 2022

    2022 arXiv

  40. [48]

    A. R. Lundborg, I. Kim, R. D. Shah, and R. J. Samworth. The projected covariance measure for assumption-lean variable significance testing.The Annals of Statistics, 52(6):2851–2878, 2024

    2024

  41. [49]

    Mai and H

    Q. Mai and H. Zou. The fused kolmogorov filter: A nonparametric model-free screening method.The Annals of Statistics, pages 1471–1497, 2015

    2015

  42. [50]

    R. A. Milton and S. F. Brown. Sobol’matrices for multi-output models with quantified uncertainty. arXiv:2501.04602, 2025

    2025 arXiv

  43. [51]

    M. A. Newton, F. A. Quintana, J. A. den Boon, S. Sengupta, and P. Ahlquist. Random-set methods identify distinct aspects of the enrichment signal in gene-set analysis.The Annals of Applied Statistics, 1(1):85 – 106, 2007

    2007

  44. [52]

    W. Pan, X. Wang, W. Xiao, and H. Zhu. A generic sure independence screening procedure.Journal of the American Statistical Association, 2019

    2019

  45. [53]

    T. Park, X. Shao, and S. Yao. Partial martingale difference correlation.Electronic Journal of Statistics, 9:1492–1517, 2015

    2015

  46. [54]

    A. R´ enyi. On measures of dependence.Acta Mathematica Hungarica, 10(3-4):441–451, 1959

    1959

  47. [55]

    Rudelson and R

    M. Rudelson and R. Vershynin. Hanson-wright inequality and sub-gaussian concentration.Electronic Communications in Probability, 18, 2013

    2013

  48. [56]

    Saltelli

    A. Saltelli. Making best use of model evaluations to compute sensitivity indices.Computer physics communications, 145(2):280–297, 2002

    2002

  49. [57]

    Saltelli, K

    A. Saltelli, K. Chan, and E. M. Scott.Sensitivity analysis: Gauging the worth of scientific models. John Wiley & Sons, 2000

    2000

  50. [58]

    Shao and J

    X. Shao and J. Zhang. Martingale difference correlation and its use in high-dimensional variable screen- ing.Journal of the American Statistical Association, 109(507):1302–1318, 2014

    2014

  51. [59]

    I. M. Sobo ´l. Sensitivity estimates for nonlinear mathematical models.Math. Model. Comput. Exp., 1: 407, 1993

    1993

  52. [60]

    I. M. Sobol. Global sensitivity indices for nonlinear mathematical models and their monte carlo esti- mates.Mathematics and computers in simulation, 55(1-3):271–280, 2001

    2001

  53. [61]

    Subramanian, P

    A. Subramanian, P. Tamayo, V. K. Mootha, S. Mukherjee, B. L. Ebert, M. A. Gillette, A. Paulovich, S. L. Pomeroy, T. R. Golub, E. S. Lander, and J. P. Mesirov. Gene set enrichment analysis: A knowledge- based approach for interpreting genome-wide expression profiles.Proceedings...

    2005

  54. [62]

    Sz´ ekely, M

    G. Sz´ ekely, M. Rizzo, and N. Bakirov. Measuring and testing dependence by correlation of distances. Annals of Statistics, 35(6):2769–2794, 2007

    2007

  55. [63]

    G. J. Sz´ ekely and M. L. Rizzo. Partial distance correlation with methods for dissimilarities.The Annals of Statistics, 42(6):2382, 2014

    2014

  56. [64]

    Y. Tang, H. J. Wang, and E. Barut. Testing for the presence of significant covariates through conditional marginal regression.Biometrika, 105(1):57–71, 2018

    2018

  57. [65]

    Z. Tian, T. Lai, and Z. Zhang. Variation of conditional mean and its application in ultrahigh dimensional feature screening.Communications in Statistics-Theory and Methods, 54(2):352–382, 2025

    2025

  58. [66]

    Tissot and C

    J.-Y. Tissot and C. Prieur. A randomized orthogonal array-based procedure for the estimation of first- and second-order sobol’indices.Journal of Statistical Computation and Simulation, 85(7):1358–1381, 2015

    2015

  59. [67]

    A. W. Van der Vaart.Asymptotic statistics, volume 3. Cambridge university press, 2000

    2000

  60. [68]

    S. D. Veiga, F. Gamboa, B. Iooss, and C. Prieur.Basics and Trends in Sensitivity Analysis: Theory and Practice in R. Computational Science and Engineering. Society for Industrial and Applied Mathematics, 2021

    2021

  61. [69]

    Verdinelli and L

    I. Verdinelli and L. Wasserman. Feature importance: A closer look at shapley values and loco.Statistical Science, 39(4), 2024

    2024

  62. [70]

    H. M. Wagner. Global sensitivity analysis.Operations Research, 43(6):948–969, 1995

    1995

  63. [71]

    B. D. Williamson, P. B. Gilbert, M. Carone, and N. Simon. Nonparametric variable importance assess- ment using machine learning techniques.Biometrics, 77(1):9–22, 2021. CONDITIONAL MEAN INDEPENDENCE AND GLOBAL SENSITIVITY ANALYSIS 21

    2021

  64. [72]

    B. D. Williamson, P. B. Gilbert, N. R. Simon, and M. Carone. A general framework for inference on algorithm-agnostic variable importance.Journal of the American Statistical Association, 118(543): 1645–1658, 2023

    2023

  65. [73]

    Yan and J

    X. Yan and J. Bien. Rare feature selection in high dimensions.Journal of the American Statistical Association, 116(534):887–900, 2021

    2021

  66. [74]

    Zhang, N

    B. Zhang, N. Mohammed, V. S. Dave, and M. Al Hasan. Feature selection for classification under anonymity constraint.Transactions on Data Privacy, 10(1):1–25, 2017

    2017

  67. [75]

    Zhang and L

    L. Zhang and L. Janson. Floodgate: inference for model-free variable importance.arXiv:2007.01283, 2020

    2007 arXiv

  68. [76]

    Zhang, L

    Y. Zhang, L. Huang, Y. Yang, and X. Shao. Testing conditional mean independence using generative neural networks. InProceedings of the 42nd International Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Research, pages 75067–75096. PMLR, 2025

    2025

  69. [77]

    1 for the model in (A.1). AppendixB.Proof of Proposition 2.1 The result inpP1qfollow directly from the decomposition: E

    X. Zhu and L. Zhu. Dimension reduction-based significance testing in nonparametric regression.Elec- tronic Journal of Statistics, 12:1468–1506, 2018. 22 CHATTERJEE, NIU, AND BHATTACHARYA AppendixA.Additional Experimental Results A.1.Conditional Mean Independence.In this sectio...

    2018

  70. [78]

    1 n nÿ u“1 1 K ÿ vPNGpXnqpuq Y J u Yv PÑE

    Thus, recalling the decomposition from (3.2), to complete the proof it is enough to show, Vn :“ 1 n nÿ u“1 1 K ÿ vPNGpXnqpuq Y J u Yv PÑE ” }ErY|Xs} 2 2 ı .(C.1) To establish (C.1), it suffices to show the following: ErVns ÑEr}ErY|Xs} 2 2sand VarrV ns “op1q.(C.2) With this in ...

  71. [79]

    εn 4 4`δ . Then, E „ max 1ďuďn }Yu}4 2 ȷ ďε n `E „ max 1ďuďn }Yu}4 2 1

    and following the arguments in the proof of [7, Lemma B.2] gives, VarrVns À d 1 n E „ max 1ď1ďu‰vďnďn ˇˇY J u Yv ˇˇ2 ȷ ď 1 n E „ max 1ďuďn }Yu}4 2 ȷ ,(C.6) where the last inequality once again follows from the Cauchy-Schwartz inequality. Forεą0 define εn :“εn 4 4`δ . Then, E „...

  72. [80]

    opn ´ δ 8`δ q. Similar arguments show that Varr ˜Qp1q n s “opn ´ δ 8`δ q, completing the proof of (D.8). Next, we show that, EH0

    gives, Varr ˆ˜Qp1q n s À d 1 n E „ max 1ď1ďu‰vďnďn ˇˇY J u Yv ˇˇ4 ȷ ď 1 n E „ max 1ďuďn }Yu}8 2 ȷ . Now the arguments from (C.7) can be easily adapted to show Varr ˆ˜Qp1q n s “opn ´ δ 8`δ q. Similar arguments show that Varr ˜Qp1q n s “opn ´ δ 8`δ q, completing the proof of (D....

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.