Pith. sign in

REVIEW 1 major objections 5 minor 35 references

An association measure for mixed-type variables

T0 review · 1 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A new coefficient, ξ′, measures how much a continuous variable X concentrates a categorical variable Y: 0 means independence, 1 means Y is a function of X.

desk verdict A solid, useful mixed-type association paper with careful asymptotics and one real gap: the plug-in variance estimator used for confidence intervals is not proven consistent for the discontinuous conditional-probability cases the paper itself features. read the letter →

arxiv 2607.26508 v1 pith:ZZXHS75A submitted 2026-07-29 stat.ME math.STstat.TH

Yongjae Kim , Haeun Moon , Sungkyu Jung This is my paper · ORCID
classification stat.MEmath.STstat.TH MSC 62H2062G3062G10
keywords associationmeasuremixed-typedatanominalcategorieslabelinvarianceindependencetestrunsstatisticasymptoticnormalityrankcorrelation
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 establishes a single normalized coefficient, written ξ′, that measures how strongly a real-valued variable X predicts a categorical variable Y. The value is 0 exactly when X and Y are independent and 1 exactly when Y is a deterministic function of X; in between, it reports the fraction of the categorical variable's dispersion explained by conditioning on X. The sample version sorts the observations by X and counts how often neighboring observations carry the same label, then rescales that count to remove the baseline chance agreement; the computation is O(n log n) and needs no arbitrary integer coding of categories. The paper's main technical contribution is proving that this sample statistic converges almost surely and is asymptotically normal both under independence and under general dependence, which yields a permutation-free independence test and asymptotic confidence intervals. A sympathetic reader would care because existing rank-based approaches give different answers depending on how category labels are encoded, while this measure is invariant to relabeling.

What carries the argument

The load-bearing identity is the ratio (A−B)/(1−B), where A is the expected conditional concentration of Y given X, A=E[Σ_j P(Y=j|X)²], and B=Σ_j p_j² is the marginal concentration. The numerator is the reduction in impurity gained by conditioning, and the denominator normalizes it to [0,1]. On the sample side, the estimator counts adjacent matches: after sorting by X (with random tie-breaking), A_n is the average of 1{Y_(i+1)=Y_(i)}, and B_n is the empirical version of B. Under independence the match indicators form a 1-dependent sequence, giving the null variance B(1+B)−2ρ over (1−B)². Under dependence the proof decomposes the fluctuation of (A_n,B_n) into an X-measurable part and a label

What would settle it

Take the paper's wiggly block design — a conditional class probability that switches its preferred category across 40 narrow intervals — with a strong signal and large n, then compute the plug-in 95% confidence intervals over many replicates. If true ξ′ falls inside at the nominal rate despite the discontinuities and rapid oscillation, the total-variation requirement is not a practical obstacle; if coverage degrades as the number of switches grows, that requirement is doing load-bearing work and the general-dependence theorem's scope is confirmed to exclude such cases.

Watch

Extended reading notes

Core claim

The central claim is that the population quantity ξ′(X,Y) = (E[Σ_j P(Y=j|X)²] − Σ_j p_j²)/(1 − Σ_j p_j²) is a valid directed association measure. Convexity of the sum-of-squares map puts it in [0,1]; equality with 0 forces the conditional class probabilities to equal the marginal probabilities almost surely, which is independence; equality with 1 forces one conditional probability to be 1 almost surely, which is functional dependence; and permuting category labels leaves both sums unchanged. The sample estimator ξ′_n sorts by X, averages indicators that adjacent observations share the same label, and rescales by the empirical marginal concentration. The paper proves strong consistency for ar

Load-bearing premise

The load-bearing premise for the confidence-interval results is an oscillation budget: the conditional class probabilities must have cumulative wobble (total variation) growing slower than √n over intervals containing all but a vanishing fraction of X; a second fragile link is that the variance-estimator consistency proof cites a condition mainly established for continuous conditional probabilities, leaving the discontinuous case less covered.

Editorial extensions

If this is right

  • Nominal labels no longer need an arbitrary integer encoding: the same data give the same ξ′_n under every permutation of category labels, removing a source of unstable and unreproducible results.
  • Independence can be tested without permutations: the centered statistic has a standard normal limit under the null, so genome-style screening of many continuous–categorical pairs becomes practical at O(n log n).
  • Confidence intervals for the strength of association become available, allowing effect-size statements rather than only p-values, which most alternative dependence measures do not offer.
  • In the binary case the population coefficient coincides with the established rank-based coefficient, so the new measure inherits that theory while extending it to multiple nominal categories.
  • Because ξ′_n is an affine transform of the number of runs in the X-sorted label sequence, classical runs intuition transfers directly: fewer runs means stronger dependence.

Reading between the lines

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

  • The deliberate choice of unweighted aggregation across categories means the measure is especially sensitive to associations carried by rare categories; for a prevalence-weighted effect size, the paper's alternative ξ′′ would behave differently, often much smaller when minority classes carry the signal.
  • The general-dependence theorem relies on one-dimensional ordering and a total-variation budget; extending the idea to multivariate predictors via nearest-neighbor graphs, as the paper anticipates, will require a new way to keep conditional probabilities close along the graph edges.
  • The paper's comparison suggests the ordinary bootstrap should not be used with adjacent-match statistics; any future implementation or user should prefer the plug-in variance estimator or m-out-of-n subsampling, and the finite-sample centering term −1/(n−1) matters at modest sample sizes.
  • The consistency proof for the plug-in variance estimator invokes a sufficient condition that the paper cites mainly for continuous conditional probabilities, even though the main theorem allows bounded-variation discontinuities; a targeted simulation with step-function conditional probabilities would tell whether this is a real limitation or just a proof gap.
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

1 major / 5 minor

Summary. The paper proposes a population association measure ξ′(X,Y) = (A−B)/(1−B) for a real-valued X and a nominal categorical Y, where A = E[Σ_j P(Y=j|X)²] and B = Σ_j p_j², together with an O(n log n) sample estimator ξ′_n based on adjacent label matches after sorting by X. It proves that ξ′ is normalized in [0,1], equals 0 only under independence, equals 1 only under functional dependence, is invariant to label permutations and monotone transformations of X, and reduces to Chatterjee's ξ in the binary case. The paper further establishes strong consistency of ξ′_n for arbitrary X (Theorem 3.1), asymptotic normality under independence with a consistent plug-in variance estimator (Theorem 3.2), and asymptotic normality under general dependence under a bounded-variation oscillation condition (Theorem 3.6). A plug-in variance estimator is used to construct Wald tests and confidence intervals, with simulations and a TCGA breast-cancer application illustrating coding stability, power, and computational speed.

Significance. If the results hold, the paper provides a practically useful, label-invariant alternative to integer-coded rank correlations and permutation-calibrated kernel methods for mixed-type data. The population measure has a clean Gini-impurity interpretation, the estimator is simple and fast, and the main asymptotic proofs are technically careful: Theorem 3.1 uses Lusin's theorem in a genuinely nonparametric way, the null CLT treats the 1-dependent adjacent-match indicators correctly, and Theorem 3.6 employs an appropriate dependency-graph CLT. The paper is also commendably transparent about the empirical choice of the m-out-of-n resampling exponent and about the finite-sample bias correction. The main weakness is a gap between the general-dependence asymptotic-normality theorem and the consistency proof of the plug-in variance estimator used for confidence intervals; this is a load-bearing issue for the confidence-interval claim, though it appears patchable.

major comments (1)
  1. [§3.4, Proposition 3.7 and Corollary 3.8] Proposition 3.7 establishes consistency of the plug-in variance estimator κ̂_D² only under the empirical L1 condition (**). The paper's stated sufficient condition for (**) — via Györfi et al. (2002) — requires X absolutely continuous and each g_j bounded and continuous. However, Assumption 3.5 explicitly permits bounded-variation discontinuous conditional probability functions, and the simulation study in Section 4 uses a piecewise-constant block design in which the g_j are discontinuous. Thus, for the class of distributions covered by Theorem 3.6 and actually used in the confidence-interval simulations, the asymptotic validity of the Wald intervals is not established by Proposition 3.7 as stated. The formal theorem is conditional on (**), so this is not a contradiction of Theorem 3.6 itself, but it is a genuine gap between the advertised confidence-interval procedure for 'general depen
minor comments (5)
  1. [§2.3.2, Example 2, Eq. (7)] The displayed formula for ξ′(ϵ,δ,k) has ε² in the numerator, but the text immediately after states that ξ′ vanishes at a linear rate as ε→0. The derivation in Appendix B.1 and the numerical table values correspond to the corrected formula δ(k−1)(k−2)ε/(2−kε). Please fix this typo.
  2. [§3.4, Assumption 3.5] The assumption is stated as 'there exists a choice of the Borel versions g_1,…,g_k'. Since total variation is a property of the representative, this is acceptable but should be clarified: the user of the theorem may need to know whether the condition must hold for the versions used in the k-NN estimator. A short remark on this point would prevent ambiguity.
  3. [§4.2.4, Table 4] The parameter γ=0.65 in the m-out-of-n bootstrap is selected empirically in the same experimental design used for evaluation. The paper acknowledges this, but reporting a small sensitivity table over γ∈{0.50,0.65,0.75} at n=400 would strengthen the comparison.
  4. [General presentation] The proof of Theorem 3.6 is long and dense. Some intermediate steps, such as the comparison of µ_iµ_{i+1} with h_i², are written correctly but are hard to follow. A few signposts in Appendix C.3.1 would help readers verify the covariance calculations.
  5. [Reproducibility] The simulations and TCGA application are described in enough detail to be reimplemented, but no code or data-access instructions are given. Providing a supplementary R/Python implementation would make the computational claims and the numerical results fully reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the population measure, estimator, and limit theorems form a self-contained derivation chain.

full rationale

The paper defines ξ′ directly as (A−B)/(1−B), a functional of the conditional class probabilities, and the estimator ξ′_n is a separate adjacent-match statistic. Theorem 3.1 proves A_n → A using Lusin’s theorem; the independence CLT (Theorem 3.2) and general-dependence CLT (Theorem 3.6) are proved from the same definition rather than assumed. The plug-in variance estimator (Proposition 3.7) is ancillary: its consistency is conditional on condition (**), with Györfi et al. cited only as a sufficient condition for continuous g_j. The manuscript explicitly acknowledges the empirical choice γ = 0.65 and the standard bootstrap’s failure, but these affect simulation comparisons, not the definition of the target or the main asymptotic claims. The runs-statistic identity (Eq. 9) is presented explicitly as a connection, not as the source of the population measure. The only self-citation (Moon and Chen 2022) appears in a literature list and is not load-bearing. The technical gap flagged by the skeptic—that condition (**) is not verified for the discontinuous g_j admitted by Assumption 3.5 and used in the block-design simulations—is a proof gap affecting confidence-interval coverage, not a circularity, because the k-NN variance estimator is not used to define ξ′ or to fit the population measure. The apparent typo in Eq. (7) is an algebraic slip, not a circular reduction. No step reduces a prediction to an input by construction, so the circularity score is 0.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The central measure is a functional of the joint distribution of (X,Y) and the estimator is a deterministic function of the sample plus random tie-breaking; no new physical or model entities are postulated. The only free quantities are standard nonparametric tuning choices (k_nn for the variance plug-in) and one empirically selected resample exponent (γ) that affects a secondary bootstrap comparison, not the main estimator or its target. The main axioms are standard probability results plus the regularity condition Assumption 3.5 and the unproven k-NN L1 consistency condition for discontinuous g_j.

free parameters (2)
  • k_nn (neighborhood size for plug-in variance estimator) = ⌊√n⌋
    Neighborhood size for the k-NN plug-in estimator of conditional class probabilities used in the general-dependence variance estimator and confidence intervals (§3.4). Chosen by default; a standard nonparametric choice that is not fitted to the target measure.
  • γ (m-out-of-n resample exponent) = 0.65
    Chosen empirically from {0.50, 0.65, 0.75} because it performed best in the block design; used only in the bootstrap comparison (§4.2.4), not in the main method.
assumptions (6)
  • standard math The conditional class probability functions g_j are Borel versions with Σ_j g_j = 1 and are bounded; used throughout Section 2.
    Standard conditional-probability regularization; the paper chooses such versions and keeps them fixed.
  • domain assumption Assumption 3.5: there are expanding intervals I_n with nP(X∉I_n)→0 and max_j V_{I_n}(g_j)=o(√n).
    Needed for Theorem 3.6 general asymptotic normality; permits bounded variation and many discontinuities but excludes total variation growing faster than √n.
  • domain assumption For Proposition 3.7, the k-NN estimators satisfy the empirical L1 consistency condition (**); the paper states this follows if X is absolutely continuous and g_j is bounded and continuous (citing Györfi et al., 2002).
    Needed for a consistent plug-in variance estimator and valid confidence intervals; for the discontinuous g_j allowed by Assumption 3.5 and used in simulations, the condition is asserted rather than proved.
  • domain assumption For Corollary 3.8, the asymptotic variance κ_D² > 0.
    Needed for nondegenerate studentization; excludes pathological cases where the dependence is not visible in the asymptotic variance.
  • domain assumption Ties in X are broken independently at random; this is part of the estimator definition and is used in the proof of the null distribution.
    Without random tie-breaking, the permutation-independence argument under H0 can fail for discrete or mixed X distributions.
  • standard math Lusin's theorem, the Hoeffding–Robbins CLT for 1-dependent sequences, the delta method, and Baldi–Rinott Berry–Esseen bounds for dependency graphs are used in the proofs.
    Standard cited results used in the appendices; no new mathematical foundation is introduced.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An association measure for mixed-type variables." pith.science (2026). https://pith.science/paper/ZZXHS75A

@misc{pith2026260726508,
  author       = {Pith},
  title        = {Pith review of: An association measure for mixed-type variables},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZZXHS75A}},
  note         = {Machine review of arXiv:2607.26508}
}
abstract

Quantifying the association between a real-valued variable and a categorical variable is a fundamental task in data analysis. Existing methods often rely on parametric assumptions or arbitrary integer encoding, which may lead to unstable results. We propose a label-invariant population measure of association, $\xi'$, specifically designed for the mixed real-valued-categorical setting. The proposed measure is normalized between 0 and 1; it equals 0 if and only if the variables are independent and 1 if and only if the categorical variable is a measurable function of the real-valued one. We also introduce a corresponding sample estimator, $\xi_n'$, computable in $O(n \log n)$ time. These measures are invariant to permutations of category labels and strictly monotone transformations of the real-valued variable. We establish the strong consistency and asymptotic normality of the estimator $\xi_n'$, enabling a computationally efficient, permutation-free Wald test for independence, and an asymptotic confidence interval for the population measure $\xi'$. Extensive simulations and an application to The Cancer Genome Atlas (TCGA) data demonstrate that the proposed method provides coding stability, competitive power, and substantial computational advantages in nominal mixed-type settings.

Figures

Figures reproduced from arXiv: 2607.26508 by the authors.

Figure 1
Figure 1. Illustration of the integer coding problem. Applying two different arbitrary [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The left panels display the conditional class probabilities [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the rare signal setting described in Example [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Illustration of the dominant noise setting described in Example [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Scatterplots of (X, Y ) under the block design model. As the signal θ increases from 0 to 1, the label homogeneity within each block becomes more noticeable, transitioning from random noise to a deterministic step function. label Y is sampled from: Pr(Y = jb | X ∈ bloc…
Figure 6
Figure 6. Figure 6: Null calibration of Z c n = √ n{ξ ′ n + 1/(n − 1)}/κb for k = 6. Histograms are overlaid with the standard normal density (solid blue) and the empirical Gaussian fit (red dashed) [PITH_FULL_IMAGE:figures/full_fig_p023_6.png]
Figure 7
Figure 7. Figure 7: Distribution of the standardized statistic under dependence ( [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Finite-sample performance of ξ ′ n around the detection threshold (n = 500). The plot zooms in on the interval θ ∈ [0.20, 0.25]. The lower confidence bound (dotted blue line) crosses the zero at θ ≈ 0.233, corresponding to a true association of ξ ′ ≈ 0.054. population …
Figure 9
Figure 9. Figure 9: Stability comparison under 6! label permutations. While the proposed measure ξ ′ n (pink solid) remains invariant, ξn (histogram) exhibits high vari￾ability, crossing the rejection threshold (dashed line) purely due to relabeling. Y (where δ(y, y′ ) = 1 if y = y ′ and …
Figure 10
Figure 10. Figure 10: Comparison of dependence measures across three scenarios. While [PITH_FULL_IMAGE:figures/full_fig_p034_10.png]
Figure 11
Figure 11. Figure 11: (a) Density plots show distinct variance differences between subtypes [PITH_FULL_IMAGE:figures/full_fig_p036_11.png]
Figure 12
Figure 12. Figure 12: Instability of Chatterjee’s ξn versus the stability of ξ ′ n . Blue dots repre￾sent ξn values computed under all 24 possible label permutations. In the resid￾uals setting, ξn ranges from 0.288 to 0.664, demonstrating severe sensitivity to arbitrary coding. In contrast…
Figure 13
Figure 13. Figure 13: Decomposition of contributions. In the residuals scenario, the contribu [PITH_FULL_IMAGE:figures/full_fig_p038_13.png]
Figure 14
Figure 14. Figure 14: Density plots of ESR1 expression within the ER-positive group (n = 887), stratified by PR status. The substantial overlap between PR+ (blue) and PR- (red) distributions explains why adding PR status to the classification dilutes the dependence measured by ξ ′ n [PIT…
Figure 15
Figure 15. Figure 15: Comparison between ξ ′ (solid) and ξ ′′ (dashed) in the rare-class con￾struction with k = 3. The prevalence-weighted measure ξ ′′ is smaller because the contribution of the rare class is scaled by its marginal probability, even though that class is perfectly predicted…
Figure 16
Figure 16. Figure 16: Comparison between ξ ′ (solid) and ξ ′′ (dashed) in the rare-class construc￾tion with k = 10. The discrepancy becomes more pronounced as the number of categories increases, illustrating the effect of prevalence weighting on the con￾tribution of rare classes as k grows…
Figure 17
Figure 17. Figure 17: Visualization of the high-frequency wiggly block design ( [PITH_FULL_IMAGE:figures/full_fig_p048_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 5 canonical work pages

  1. [1]

    A simple measure of condi- tional dependence.The Annals of Statistics49.https://doi.org/10.1214/ 21-AOS2073

    Azadkia, M.andChatterjee, S.(2021). A simple measure of condi- tional dependence.The Annals of Statistics49.https://doi.org/10.1214/ 21-AOS2073

    2021

  2. [2]

    On Normal Approximations of Distributions in Terms of Dependency Graphs.The Annals of Probability171646–1650

    Baldi, P.andRinott, Y.(1989). On Normal Approximations of Distributions in Terms of Dependency Graphs.The Annals of Probability171646–1650

    1989

  3. [3]

    E.andDavid, F

    Barton, D. E.andDavid, F. N.(1957). Multiple Runs.Biometrika44168– 178.https://doi.org/10.1093/biomet/44.1-2.168

    work page doi:10.1093/biomet/44.1-2.168 1957

  4. [4]

    A consistent test of independence based on a sign covariance related to Kendall’s tau.Bernoulli201006–1028

    Bergsma, W.andDassios, A.(2014). A consistent test of independence based on a sign covariance related to Kendall’s tau.Bernoulli201006–1028

    2014

  5. [5]

    J.andOlshen, R

    Breiman, L.,Friedman, J.,Stone, C. J.andOlshen, R. A.(1984).Clas- sification and Regression Trees. Taylor & Francis

    1984

  6. [6]

    Chatterjee, S.(2021). A New Coefficient of Correlation.Journal of the Amer- ican Statistical Association1162009—2022.https://doi.org/10.1080/ 01621459.2020.1758115 Cramér, H.(1946).Mathematical Methods of Statistics. Princeton University Press

    arXiv 2021

  7. [7]

    A simple bootstrap for Chatterjee’s rank correlation.Biometrika112.https://doi.org/10.1093/biomet/asae045 An association measure for mixed-type variables81

    Dette, H.andKroll, M.(2025). A simple bootstrap for Chatterjee’s rank correlation.Biometrika112.https://doi.org/10.1093/biomet/asae045 An association measure for mixed-type variables81

    work page doi:10.1093/biomet/asae045 2025

  8. [8]

    F.andStoimenov, P

    Dette, H.,Siburg, K. F.andStoimenov, P. A.(2012). A Copula-Based Non-parametric Measure of Regression Dependence.Scandinavian Journal of Statistics4021-–41.https://doi.org/10.1111/j.1467-9469.2011. 00767.x

    arXiv 2012

Show all 35 references
  1. [9]

    A Log Log Law for Maximal Uniform Spacings.The Annals of Probability10.https://doi.org/10.1214/aop/1176993799

    Devroye, L.(1982). A Log Log Law for Maximal Uniform Spacings.The Annals of Probability10.https://doi.org/10.1214/aop/1176993799

    1982

  2. [10]

    Polychoric and polyserial correlations

    Drasgow, F.(1986). Polychoric and polyserial correlations. InEncyclopedia of Statistical Sciences, (S. Kotz and N. L. Johnson, eds.)768–74. John Wiley & Sons, New York

    1986

  3. [11]

    A.(1921)

    Fisher, R. A.(1921). Studies in crop variation. I. An examination of the yield of dressed grain from Broadbalk.The Journal of Agricultural Science11107- –135.https://doi.org/10.1017/s0021859600003750

    1921 doi

  4. [12]

    B.(1999).Real Analysis: Modern Techniques and Their Applica- tions, 2nd ed

    Folland, G. B.(1999).Real Analysis: Modern Techniques and Their Applica- tions, 2nd ed. John Wiley & Sons, New York

    1999

  5. [13]

    Variabilità e mutabilità.Studi economico-giuridici della R

    Gini, C.(1912). Variabilità e mutabilità.Studi economico-giuridici della R. Università di Cagliari31–158

    1912

  6. [14]

    V.andHopkins, K

    Glass, G. V.andHopkins, K. D.(1995).Statistical Methods in Education and Psychology, 3rd ed. Allyn & Bacon

    1995

  7. [15]

    A.andKruskal, W

    Goodman, L. A.andKruskal, W. H.(1954). Measures of Association for Cross Classifications*.Journal of the American Statistical Association49 732—764.https://doi.org/10.1080/01621459.1954.10501231

    1954

  8. [16]

    Springer Berlin Heidelberg.https://doi.org/10

    Gretton, A.,Bousquet, O.,Smola, A.andSchölkopf, B.(2005).Mea- suring Statistical Dependence with Hilbert-Schmidt NormsInAlgorithmic Learning Theory63–77. Springer Berlin Heidelberg.https://doi.org/10. 1007/11564089_7

    2005

  9. [17]

    A Kernel Statistical Test of Independence

    Smola, A.(2007). A Kernel Statistical Test of Independence. InAdvances in Neural Information Processing Systems(J. Platt,D. Koller,Y. Singer andS. Roweis, eds.)20. Curran Associates, Inc. Györfi, L.,Kohler, M.,Krzyżak, A.andW alk, H.(2002).A Distribution- Free Theory of Nonpar...

    2007 doi

  10. [18]

    Measures of concentration.Journal of the American Statistical Association62162–168

    Hall, M.andTideman, N.(1967). Measures of concentration.Journal of the American Statistical Association62162–168

    1967

  11. [19]

    The Central Limit Theorem for Dependent Random Variables.Duke Mathematical Journal15773–780

    Hoeffding, W.andRobbins, H.(1948). The Central Limit Theorem for Dependent Random Variables.Duke Mathematical Journal15773–780

    1948

  12. [20]

    J.(2016)

    Huo, X.andSzékely, G. J.(2016). Fast Computing for Distance Covariance. Technometrics58435-–447.https://doi.org/10.1080/00401706.2015. 1054435

    2016

  13. [21]

    G.(1938)

    Kendall, M. G.(1938). A New Measure of Rank Correlation.Biometrika30 81.https://doi.org/10.2307/2332226

    1938 doi

  14. [22]

    RTCGA: The Cancer Genome Atlas Data Integration R package version 1.41.0

    Kosinski, M.(2025). RTCGA: The Cancer Genome Atlas Data Integration R package version 1.41.0

    2025

  15. [23]

    Extension of Process Convergence With Application to Chatterjee’s Rank Correlation

    Kroll, M.(2026). Extension of Process Convergence With Application to Chatterjee’s Rank Correlation

    2026

  16. [24]

    Limit theorems of Chatterjee’s rank correlation

    Lin, Z.andHan, F.(2022). Limit theorems of Chatterjee’s rank correlation. 82Y. Kim et al. https://doi.org/10.48550/ARXIV.2204.08031

    2022 doi arXiv

  17. [25]

    On boosting the power of Chatterjee’s rank correla- tion.Biometrika110283–299.https://doi.org/10.1093/biomet/asac048

    Lin, Z.andHan, F.(2023). On boosting the power of Chatterjee’s rank correla- tion.Biometrika110283–299.https://doi.org/10.1093/biomet/asac048

    2023 doi

  18. [26]

    On the failure of the bootstrap for Chatterjee’s rank correlation.Biometrika1111063–1070

    Lin, Z.andHan, F.(2024). On the failure of the bootstrap for Chatterjee’s rank correlation.Biometrika1111063–1070

    2024

  19. [27]

    Interpoint-Ranking Sign Covariance for the Test of Independence.Biometrika109165–179.https://doi.org/10.1093/ biomet/asab011 The Cancer Genome Atlas Network(2012)

    Moon, H.andChen, K.(2022). Interpoint-Ranking Sign Covariance for the Test of Independence.Biometrika109165–179.https://doi.org/10.1093/ biomet/asab011 The Cancer Genome Atlas Network(2012). Comprehensive molecular portraits of human breast tumours.Nature49061–70.https://doi.o...

    2022 doi

  20. [28]

    J.(1982)

    Olsson, U.,Drasgow, F.andDorans, N. J.(1982). The Polyserial Corre- lation Coefficient.Psychometrika47337–347.https://doi.org/10.1007/ BF02294164

    1982

  21. [29]

    Pearson, K.(1900). On the Criterion That a Given System of Deviations from the Probable in the Case of a Correlated System of Variables is Such That It Can Be Reasonably Supposed to Have Arisen from Random Sampling. Philosophical Magazine Series 550157–175.https://doi.org/10.1...

    1900

  22. [30]

    On the Partial Correlation Ratio.Proceedings of the Royal Society of London

    Pearson, K.(1915). On the Partial Correlation Ratio.Proceedings of the Royal Society of London. Series A, Containing Papers of a Mathematical and Phys- ical Character91492–498.https://doi.org/10.1098/rspa.1915.0041

    1915

  23. [31]

    A.,Pollack, J

    Rees, C. A.,Pollack, J. R.,Ross, D. T.,Johnsen, H.,Akslen, L. A. et al. (2000). Molecular portraits of human breast tumours.Nature406747– 752.https://doi.org/10.1038/35021093 Rényi, A.(1959). On Measures of Dependence.Acta Mathematica Hungarica 10441–451.https://doi.org/10.100...

    2000 doi

  24. [32]

    The Proof and Measurement of Association Between Two Things.The American Journal of Psychology1572–101.https://doi

    Spearman, C.(1904). The Proof and Measurement of Association Between Two Things.The American Journal of Psychology1572–101.https://doi. org/10.2307/1412159

    1904 doi

  25. [33]

    F.(2024)

    Strothmann, C.,Dette, H.andSiburg, K. F.(2024). Rearranged depen- dence measures.Bernoulli (Andover.)30. Székely, G. J.,Rizzo, M. L.andBakirov, N. K.(2007). Measuring and Testing Dependence by Correlation of Distances.The Annals of Statistics35 2769–2794.https://doi.org/10.121...

    2024 doi

  26. [34]

    F.(1954)

    Tate, R. F.(1954). Correlation between a discrete and a continuous variable. Point-biserial correlation.Ann. Math. Stat.25603–607. van der V aart, A. W.(1998).Asymptotic Statistics.Cambridge Series in Sta- tistical and Probabilistic Mathematics3. Cambridge University Press, Ca...

    1954 doi

  27. [35]

    Coverage correlation: detect- ing singular dependencies between random variables.https://doi.org/10

    Yang, X.,Azadkia, M.andW ang, T.(2025). Coverage correlation: detect- ing singular dependencies between random variables.https://doi.org/10. 48550/arXiv.2508.06402

    2025 doi arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.