arxiv: 2605.13439 · v1 · submitted 2026-05-13 · 📊 stat.ME

Recognition: unknown

Median Radial Function: A Robust, Covariance-Free Framework and Applications

Elsayed Elamir

Authors on Pith no claims yet

Pith reviewed 2026-05-14 19:11 UTC · model grok-4.3

classification 📊 stat.ME

keywords median radial functiondepth functionrobust statisticsmultivariate centralitycovariance-freeradial dispersionskewness detectionhigh-dimensional data

0 comments

The pith

A median radial depth function measures centrality in multivariate data without covariance or moment assumptions.

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

The paper introduces a framework that uses median distances to a center to define both a scale-invariant radial dispersion and a depth function. This approach aims to provide a robust way to assess how central each point is in a dataset. A sympathetic reader would care because it works without assuming the data follows a normal distribution or has finite variances, and it handles skewness and multiple modes naturally. The method is shown to be convex with computable subgradients that can indicate asymmetry. It is tested empirically to match existing methods in simple cases but offer more in complex ones.

Core claim

The author establishes a median-radius framework for multivariate centrality based on median distances, from which a scale-invariant radial dispersion is defined and used to construct a depth function. This depth function is robust to outliers and independent of covariance structure. It requires no moment assumptions and adapts to skewness, multimodality, and heavy-tailed distributions, making it suitable for high-dimensional data. The underlying functionals are shown to be convex, and their subgradients encode directional imbalances, suggesting a radial method for detecting skewness and asymmetry.

What carries the argument

The median radial function, which computes median distances from a data center to measure centrality and dispersion in a scale-invariant way.

Load-bearing premise

That the median of distances from a suitable center provides a valid and useful foundation for a depth function without requiring any moment conditions or specific distributional forms.

What would settle it

Observing that the proposed depth values do not decrease for points farther from the center in a simulated multivariate dataset with heavy tails, or that the depth depends on the sample covariance in controlled experiments, would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.13439 by Elsayed Elamir.

read the original abstract

A median-radius framework for assessing centrality in multivariate data using median distances is proposed. Based on the proposed framework, a scale invariant measure of radial dispersion is defined and used to establish a depth function that is robust to outliers and independent of covariance structure. The depth function does not depend on moment assumptions and naturally adapts to skewness, multimodality, and heavy-tailed distributions, which make it effective for high-dimensional data structures. We demonstrate fundamental characteristics of the underlying functionals such as subgradient and convexity. The subgradients provide additional insight and encode the imbalance in directional contributions of the data. This suggests a new approach to detect skewness and structural asymmetry through a purely radial construction. Empirical studies demonstrate that the method agrees with classical approaches under symmetry while providing a more flexible and informative characterization in complex settings.

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 median radial depth gives a covariance-free framing for non-elliptical data but leaves the center choice underspecified, which undercuts uniqueness and robustness in d>1.

read the letter

The paper defines a median radial function from distances to a center, builds a scale-invariant dispersion measure from it, and turns that into a depth function. The core idea is to stay radial and avoid any covariance or moment structure so the method can handle skewness, multimodality, and heavy tails without the usual assumptions. That framing is new relative to the classical depth literature and could be useful where elliptical methods break down. The subgradient construction for spotting directional imbalance is a concrete addition that might give a simple way to flag asymmetry. The empirical checks showing agreement with standard methods under symmetry are a fair baseline test. The main weakness is the center. The text refers to an “appropriately chosen center” without supplying a canonical, unique selection rule that itself stays covariance-free and moment-free. In dimensions greater than one the multivariate median is typically a set, so any fixed choice (componentwise, geometric, or otherwise) is arbitrary and can change the resulting depth values. This makes the claimed automatic adaptation and outlier robustness depend on an extra, unstated decision. The abstract asserts convexity and subgradient properties but gives no derivations, so those need direct verification in the full text. The work is aimed at people already working on robust multivariate depth or outlier tools who are looking for moment-free options. It is coherent enough on its own terms to warrant referee time, provided the center rule and the missing proofs are checked.

Referee Report

2 major / 2 minor

Summary. The paper proposes a median radial function framework for assessing centrality in multivariate data based on median distances to a chosen center. It defines a scale-invariant measure of radial dispersion and constructs a depth function claimed to be robust to outliers, independent of covariance structure, and free of moment assumptions. The depth is said to adapt naturally to skewness, multimodality, and heavy-tailed distributions. The manuscript demonstrates properties including subgradients and convexity of the underlying functionals, with subgradients interpreted as encoding directional imbalance for skewness detection, and includes empirical studies comparing to classical methods under symmetry.

Significance. If the central claims hold, the framework could provide a useful covariance-free and moment-free alternative to existing depth notions for high-dimensional and non-elliptical data, with the subgradient-based asymmetry detection offering a distinct radial perspective on structural features.

major comments (2)

[§2.1] §2.1 (definition of median radial function): The construction relies on distances to an 'appropriately chosen center,' but no canonical, unique selection rule is supplied for d>1 where the multivariate median is typically a set. Any fixed choice (e.g., componentwise or geometric median) is arbitrary; without a covariance-free and moment-free mechanism for selecting the center, the claimed uniqueness, scale-invariance, and automatic adaptation to skewness/multimodality inherit dependence on that choice, undermining the central robustness claim.
[§3] §3 (subgradient and convexity properties): The abstract asserts that subgradients and convexity are demonstrated and that subgradients encode directional imbalance, yet the provided derivations appear to assume a fixed center. If the center is not uniquely defined, these properties are shown only conditionally and do not establish the distribution-adaptive character without additional assumptions.

minor comments (2)

[§5] The empirical section would benefit from explicit statements of the center-selection rule used in the reported simulations.
[§2] Notation for the radial dispersion measure and depth function should be introduced with a clear equation reference early in §2 to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the role of center selection and the scope of the derived properties. We address each major comment below and indicate the revisions planned for the next version of the manuscript.

read point-by-point responses

Referee: [§2.1] §2.1 (definition of median radial function): The construction relies on distances to an 'appropriately chosen center,' but no canonical, unique selection rule is supplied for d>1 where the multivariate median is typically a set. Any fixed choice (e.g., componentwise or geometric median) is arbitrary; without a covariance-free and moment-free mechanism for selecting the center, the claimed uniqueness, scale-invariance, and automatic adaptation to skewness/multimodality inherit dependence on that choice, undermining the central robustness claim.

Authors: We agree that the original text leaves the center selection implicit. In the revision we will explicitly adopt the geometric median as the canonical center in §2.1. This choice is uniquely defined under standard conditions (the distribution is not supported on a lower-dimensional flat), requires neither covariance nor moments, and is itself outlier-robust. Because the geometric median is determined solely by the data geometry, the scale-invariance, uniqueness (up to the mild non-degeneracy condition), and adaptation to skewness or multimodality of the resulting radial function remain intact. We will add a short paragraph justifying this selection and noting that the framework is still well-defined for any other fixed center, but the geometric median supplies the required canonical, covariance-free rule. revision: yes
Referee: [§3] §3 (subgradient and convexity properties): The abstract asserts that subgradients and convexity are demonstrated and that subgradients encode directional imbalance, yet the provided derivations appear to assume a fixed center. If the center is not uniquely defined, these properties are shown only conditionally and do not establish the distribution-adaptive character without additional assumptions.

Authors: The derivations in §3 are correctly stated for a fixed center; this is the natural setting once a center has been chosen. With the geometric median now designated as the explicit center in the revised §2.1, the convexity, subgradient existence, and directional-imbalance interpretation hold unconditionally for this data-determined point. The subgradients therefore reflect imbalance relative to a robust, moment-free location estimator that itself adapts to the underlying distribution. We will insert a clarifying sentence in §3 that ties the fixed-center assumption to the geometric-median choice and reiterates that no moment or covariance conditions are invoked. revision: partial

Circularity Check

0 steps flagged

No significant circularity; definition-based construction stands independently

full rationale

The paper defines the median radial function directly from median distances to a chosen center and derives the depth function, scale-invariant dispersion, subgradients, and convexity properties from those definitions. No equations reduce a claimed prediction or uniqueness result back to a fitted parameter or self-citation by construction. The central claims rest on the explicit radial construction rather than on any tautological renaming or imported ansatz. External benchmarks or moment-free properties are asserted from the definitions themselves without circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; the central construction rests on the existence of a well-defined median distance in multivariate space and on the claim that this distance yields a convex depth function. No explicit free parameters, axioms, or invented entities are stated.

pith-pipeline@v0.9.0 · 5423 in / 1015 out tokens · 17984 ms · 2026-05-14T19:11:28.192095+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

Boente, G., & Salibián-Barrera, M. (2021). Robust functional principal components for sparse longitudinal data. METRON, 79(2), 159–188. https://doi.org/10.1007/s40300- 020-00193-3 Boyd, S., & Vandenberghe, L. (2004). Convex Optimization. Cambridge University Press. https://doi.org/10.1017/CBO9780511804441 Capezza, C., Centofanti, F., Lepore, A., & Palumbo...

work page doi:10.1007/s40300- 2021
[2]

On the Generalised Distance in Statistics

https://doi.org/10.1007/s42519-021-00236-6 Pokotylo, O., Mozharovskyi, P., & Dyckerhoff, R. (2019). Depth and Depth-Based Classification with R Package ddalpha. Journal of Statistical Software, 91(5). https://doi.org/10.18637/jss.v091.i05 R Core Team. (2026). R: A Language and Environment for Statistical Computing. https://www.R-project.org/. Reprint of: ...

work page doi:10.1007/s42519-021-00236-6 2019