pith. sign in

arxiv: 2604.23367 · v1 · submitted 2026-04-25 · 🧮 math.ST · stat.TH

Conway--Maxwell multivariate Bernoulli distribution

H\'el\`ene Cossette , Etienne Marceau , Alessandro Mutti , Patrizia Semeraro This is my paper

Pith reviewed 2026-05-08 07:02 UTC · model grok-4.3

classification 🧮 math.ST stat.TH
keywords Conway-Maxwell distributionmultivariate Bernoullidependence structurestrongly Rayleigh propertynegative dependencemarginal preservationbinary distributions
0
0 comments X

The pith

A Conway-Maxwell multivariate Bernoulli family can be parametrized to keep univariate marginals fixed while spanning the full spectrum of dependence among binary variables.

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

The paper constructs the Conway-Maxwell multivariate Bernoulli distribution from the Conway-Maxwell-binomial. A suitable choice of parameters leaves each univariate marginal exactly Bernoulli with its original success probability. This separation lets the joint distribution control dependence without touching the margins. The construction reaches every possible dependence level between the binary variables. For some parameter ranges the resulting distributions also satisfy the strongly Rayleigh property, a strong form of negative dependence.

Core claim

We investigate the Conway--Maxwell multivariate Bernoulli distributions, a family of multivariate Bernoulli distributions derived from the Conway--Maxwell-binomial distribution. We show that it is possible to set the parametrization such that the Bernoulli marginals remain intact, allowing us to study dependence properties within this family. In particular, we demonstrate that this family spans the full spectrum of dependence. Moreover, for specific ranges of the parameters, these distributions satisfy the strongly Rayleigh property, a negative dependence notion stronger than negative association.

What carries the argument

Conway-Maxwell multivariate Bernoulli distribution obtained from the Conway-Maxwell-binomial, equipped with a parametrization that fixes the univariate Bernoulli marginals while varying the dependence structure.

Load-bearing premise

A parametrization of the Conway-Maxwell-binomial extension exists that simultaneously preserves all univariate Bernoulli marginals and permits arbitrary dependence structures.

What would settle it

A concrete choice of marginal success probabilities together with a target correlation matrix that lies outside the range achievable by any parameter vector in the Conway-Maxwell multivariate Bernoulli family without altering the marginals, or a parameter setting inside the claimed range where the strongly Rayleigh property fails to hold.

Figures

Figures reproduced from arXiv: 2604.23367 by Alessandro Mutti, Etienne Marceau, H\'el\`ene Cossette, Patrizia Semeraro.

Figure 1
Figure 1. Figure 1: Histogram of the values of pmfs from Example 2. view at source ↗
read the original abstract

We investigate the Conway--Maxwell multivariate Bernoulli distributions, a family of multivariate Bernoulli distributions derived from the Conway--Maxwell-binomial distribution. We show that it is possible to set the parametrization such that the Bernoulli marginals remain intact, allowing us to study dependence properties within this family. In particular, we demonstrate that this family spans the full spectrum of dependence. Moreover, for specific ranges of the parameters, these distributions satisfy the strongly Rayleigh property, a negative dependence notion stronger than negative association.

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

3 major / 2 minor

Summary. The paper constructs a multivariate extension of the Conway-Maxwell-binomial distribution to the setting of n-dimensional Bernoulli random vectors. It shows that the parameters can be chosen so that all univariate marginals remain exactly Bernoulli (with prescribed success probabilities), and claims that the resulting family spans the full spectrum of possible dependence structures while also satisfying the strongly Rayleigh property for certain parameter ranges.

Significance. If the parametrization truly preserves exact Bernoulli marginals while allowing the joint to realize arbitrary dependence (i.e., any correlation matrix or higher-order interaction structure consistent with the marginals), the construction would supply a low-parameter, interpretable model for binary data with controllable dispersion and negative dependence. The strongly Rayleigh claim would further link the family to existing literature on negative association and log-concave measures. However, the dimension of the space of joints with fixed marginals is 2^n - n - 1, so any claim of spanning the full spectrum must be demonstrated by exhibiting sufficient free parameters or by an explicit bijection argument.

major comments (3)
  1. The abstract asserts that the family 'spans the full spectrum of dependence' and permits 'arbitrary dependence structures.' For n>2 this requires the parametrization to have at least 2^n - n - 1 free parameters after fixing the n marginal probabilities. The derivation from the univariate Conway-Maxwell-binomial typically introduces only one or two global dispersion parameters; if the multivariate version follows the same pattern, the family traces a low-dimensional manifold and cannot be dense in the full simplex of joints with given marginals. The manuscript must either (a) introduce n-dependent parameters that grow with the interaction order or (b) clarify that 'full spectrum' means only the range of pairwise correlations, not the entire space.
  2. The claim that marginals remain exactly Bernoulli after re-parametrization is load-bearing for all subsequent dependence results. The abstract states this is possible, but without an explicit expression for the normalizing constant or the joint pmf in terms of the univariate CM-binomial parameters, it is impossible to verify that the marginal probabilities are exactly the prescribed p_i and not merely approximately preserved.
  3. The strongly Rayleigh property is asserted for 'specific ranges of the parameters.' The manuscript should identify the precise region in parameter space (e.g., in terms of the dispersion parameter ν) for which the distribution is strongly Rayleigh, and supply a proof or reference to the relevant theorem (e.g., via the generating function or negative lattice condition).
minor comments (2)
  1. Notation for the multivariate pmf and the mapping from univariate CM-binomial parameters to the joint should be introduced earlier and used consistently.
  2. The paper should include a small-n example (n=3) with explicit probability tables showing that both positive and negative dependence extremes are attained while marginals stay fixed.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thorough review and insightful comments on our manuscript. We address each major comment point by point below, indicating where revisions will be made to improve clarity and rigor.

read point-by-point responses

  1. Referee: The abstract asserts that the family 'spans the full spectrum of dependence' and permits 'arbitrary dependence structures.' For n>2 this requires the parametrization to have at least 2^n - n - 1 free parameters after fixing the n marginal probabilities. The derivation from the univariate Conway-Maxwell-binomial typically introduces only one or two global dispersion parameters; if the multivariate version follows the same pattern, the family traces a low-dimensional manifold and cannot be dense in the full simplex of joints with given marginals. The manuscript must either (a) introduce n-dependent parameters that grow with the interaction order or (b) clarify that 'full spectrum' means only the range of pairwise correlations, not the entire space.

    Authors: We agree that the original phrasing in the abstract overstates the flexibility of the model. Our construction employs a single global dispersion parameter ν together with the n marginal probabilities, yielding an (n+1)-dimensional family. This traces a low-dimensional manifold within the (2^n - n - 1)-dimensional space of joints with fixed marginals and cannot realize every possible dependence structure. We will revise the abstract, introduction, and relevant discussion sections to clarify that the family spans a broad spectrum of dependence (including independence at ν = 1, positive dependence for ν < 1, and strong negative dependence for ν > 1 via the strongly Rayleigh property) rather than claiming to cover arbitrary or the full spectrum of structures. Phrases such as 'full spectrum of dependence' and 'arbitrary dependence structures' will be qualified or removed. revision: yes

  2. Referee: The claim that marginals remain exactly Bernoulli after re-parametrization is load-bearing for all subsequent dependence results. The abstract states this is possible, but without an explicit expression for the normalizing constant or the joint pmf in terms of the univariate CM-binomial parameters, it is impossible to verify that the marginal probabilities are exactly the prescribed p_i and not merely approximately preserved.

    Authors: We acknowledge the need for an explicit form to permit verification. The joint pmf takes the form P(X=x) = [∏_i (p_i^{x_i} (1-p_i)^{1-x_i})^ν] / Z(ν, p), where Z(ν, p) is the normalizing constant obtained by summing the unnormalized weights over all 2^n binary vectors. Marginalization over all coordinates except the i-th recovers the univariate Conway-Maxwell-binomial pmf with one trial, which is exactly Bernoulli(p_i). We will insert the explicit joint pmf and the marginalization argument into Section 2 of the revised manuscript so that exact preservation of the marginals can be checked directly. revision: yes

  3. Referee: The strongly Rayleigh property is asserted for 'specific ranges of the parameters.' The manuscript should identify the precise region in parameter space (e.g., in terms of the dispersion parameter ν) for which the distribution is strongly Rayleigh, and supply a proof or reference to the relevant theorem (e.g., via the generating function or negative lattice condition).

    Authors: We appreciate the request for a precise statement and supporting argument. The distribution satisfies the strongly Rayleigh property for ν ≥ 1. This follows because the associated generating function is a multivariate polynomial whose coefficients obey the negative lattice condition when ν ≥ 1, which can be verified by direct expansion or by appealing to known closure properties of log-concave measures under the Conway-Maxwell weighting. We will add an explicit statement of the range ν ≥ 1 together with a short proof sketch (or a reference to the relevant theorem on strongly Rayleigh measures) in the section discussing negative dependence properties. revision: yes

Circularity Check

0 steps flagged

No circularity: construction and claims are independent of inputs

full rationale

The paper defines the Conway-Maxwell multivariate Bernoulli family by extending the existing univariate Conway-Maxwell-binomial distribution via a parametrization that preserves Bernoulli marginals. Dependence properties (including the claimed spectrum) and the strongly Rayleigh property are then derived as consequences of that parametrization. No quoted step equates a derived quantity to a fitted parameter or prior self-result by construction; the central claims rest on explicit algebraic or probabilistic arguments within the new family rather than on renaming or self-referential fitting. This is the normal case of an honest extension paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be extracted. The construction presumably introduces one or more dispersion or dependence parameters whose values are chosen to satisfy the marginal and dependence conditions.

pith-pipeline@v0.9.0 · 5375 in / 1009 out tokens · 57506 ms · 2026-05-08T07:02:10.835678+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

14 extracted references · 14 canonical work pages

  1. [1]

    and Br¨ and´ en, P

    Borcea, J. and Br¨ and´ en, P. (2010). Multivariate P´ olya–Schur classification problems in the Weyl algebra.Proceedings of the London Mathematical Society, 101(1):73–104

    work page 2010

  2. [2]

    Borcea, J., Br¨ and´ en, P., and Liggett, T. (2009). Negative dependence and the geometry of polyno- mials.Journal of the American Mathematical Society, 22(2):521–567. Br¨ and´ en, P. (2007). Polynomials with the half-plane property and matroid theory.Advances in Mathematics, 216(1):302–320

    work page 2009

  3. [3]

    Cossette, H., Marceau, E., Mutti, A., and Semeraro, P. (2025). Extremal negative dependence and the strongly Rayleigh property.arXiv preprint arXiv:2504.17679

    work page arXiv 2025

  4. [4]

    and Jordaan, K

    Driver, K. and Jordaan, K. (2008). Zeros of the hypergeometric polynomial f (-n, b; c; z).arXiv preprint arXiv:0812.0708

    work page arXiv 2008

  5. [5]

    Fontana, R., Luciano, E., and Semeraro, P. (2021). Model risk in credit risk.Mathematical Finance, 31(1):176–202

    work page 2021

  6. [6]

    and Semeraro, P

    Fontana, R. and Semeraro, P. (2025). Geometrical representation and dependence structure of three-dimensional Bernoulli distributions.https://arxiv.org/abs/2512.06786

    work page arXiv 2025

  7. [7]

    Frostig, E. (2001). Comparison of portfolios which depend on multivariate Bernoulli random vari- ables with fixed marginals.Insurance: Mathematics and Economics, 29(3):319–331

    work page 2001

  8. [8]

    Kadane, J. B. (2016). Sums of possibly associated Bernoulli variables: The Conway–Maxwell- binomial distribution.Bayesian Analysis, pages 403–420. 13

    work page 2016

  9. [9]

    Kadane, J. B. and Næshagen, F. L. (2013). The number of killings in southern rural Norway, 1300–1569.The Annals of Applied Statistics, pages 846–859. M¨ uller, A. and Scarsini, M. (2001). Stochastic comparison of random vectors with a common copula.Mathematics of Operations Research, 26(4):723–740. M¨ uller, A. and Stoyan, D. (2002).Comparison Methods for...

    work page 2013

  10. [10]

    Rainville, E. D. (1960).Special Functions. The Macmillan Company, New York

    work page 1960

  11. [11]

    and P´ olya, G

    Schur, J. and P´ olya, G. (1914).¨Uber zwei Arten von Faktorenfolgen in der Theorie der algebraischen Gleichungen.Journal f¨ ur die reine und angewandte Mathematik

    work page 1914

  12. [12]

    and Shanthikumar, J

    Shaked, M. and Shanthikumar, J. G. (2007).Stochastic Orders. Springer

    work page 2007

  13. [13]

    and Tang, F

    Tang, W. and Tang, F. (2022). The Poisson binomial distribution—old & new.Statistical Science, 1(1):1–12. Tur´ an, P. (1950). On the zeros of the polynomials of Legendre. ˇCasopis pro pˇ estov´ an´ ı matematiky a fysiky, 75(3):113–122

    work page 2022

  14. [14]

    Wagner, D. (2011). Multivariate stable polynomials: theory and applications.Bulletin of the American Mathematical Society, 48(1):53–84. 14

    work page 2011