Pith. sign in

REVIEW 2 major objections 4 minor 35 references

This paper establishes a proper, tunable prior on correlation matrices that concentrates on any specified target matrix and reduces the number of free parameters to one per edge of a conditional-independence graph.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 13:55 UTC pith:KRRT3CRF

load-bearing objection Useful graph-structured correlation priors with arbitrary reference R0, but the 'Fisher arc-length' framing overstates what is a tangent-space approximation; the density formula flagged by the reader is actually correct. the 2 major comments →

arxiv 2607.18926 v1 pith:KRRT3CRF submitted 2026-07-21 stat.ME

Informative Distance-Based Priors for Correlation Matrices Centred on a Target Reference

Anna Freni-Sterrantino , Janet van Niekerk , Elias Teixeira Krainski , Denis Rustand , Haziq Jamil , H{\aa}vard Rue This is my paper
classification stat.ME MSC 62F1562H20
keywords correlation matrixPenalised Complexity priorFisher information distanceGaussian graphical modelCholesky factorisationprior elicitationBayesian inferenceconditional independence
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Bayesian analysts often need a prior on a correlation matrix that encodes a substantive belief, such as that certain variables are strongly related, rather than the default shrinkage toward the identity matrix. This paper proposes a distance-based prior that assigns mass decaying exponentially in the Fisher arc-length distance from any user-specified reference correlation matrix, built as a Penalised Complexity prior. The key innovation is a parameterisation that constructs the correlation matrix from the Cholesky factor of its inverse subject to a user-supplied graph, so the parameter count drops from one per pair of variables to one per edge, while the prior remains proper for all rate parameters and allows both positive and negative correlations. If the construction holds, it gives practitioners a generic way to combine an arbitrary shrinkage target with conditional-independence structure, something not available from the LKJ prior or earlier latent-variable PC priors.

Core claim

The paper claims that the prior defined by equation (19), a multivariate Laplace on the whitened parameter xi = H(theta0)^(1/2)(theta - theta0) with H(theta0) the Fisher information at the reference, is a proper prior on correlation matrices that concentrates on any user-specified reference R0. Combined with the Graphical Cholesky Parameterisation, which factors the inverse correlation matrix and leaves one free parameter per edge of a user-supplied graph, it is supported exactly on the submanifold of correlation matrices whose precision respects the graph. The prior is proper for every positive rate lambda, accommodates correlations of either sign, reduces to the dense prior when the graph

What carries the argument

The central object is the whitened parameter xi = H(theta0)^(1/2)(theta - theta0), where H(theta0) is the Hessian of the Kullback-Leibler divergence (the Fisher information) evaluated at the reference matrix. The prior is a multivariate Laplace on xi: the radius r = ||xi|| follows an exponential distribution and the direction is uniform on the unit sphere, giving a Cartesian density proportional to lambda exp(-lambda r) / r^(m-1). This is pushed forward to the correlation manifold through a smooth bijection theta maps to R(theta). The Graphical Cholesky Parameterisation constructs R by Cholesky-factorising the inverse correlation matrix Q = LL^T: the diagonal is fixed, edge entries are free,

Load-bearing premise

The identification of the Euclidean norm in whitened coordinates with Fisher arc-length relies on a local quadratic approximation to the Kullback-Leibler divergence, which may break down far from the reference matrix on the curved correlation manifold.

What would settle it

Compute the exact Fisher geodesic distance from a reference R0 to a distant point R on the manifold by numerically integrating the geodesic equations, and compare it with the whitened norm r = ||H(theta0)^(1/2)(theta - theta0)||; if the ratio between the two varies substantially with distance or direction, the prior's purported exponential decay in Fisher arc-length is only approximate and can deviate arbitrarily far from the stated geometry.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Users can specify an informative prior centred on any substantively motivated correlation matrix, not just the identity, with a single rate parameter controlling shrinkage strength in interpretable distance units.
  • The graph-based parameterisation imposes conditional-independence structure with one parameter per edge instead of p(p-1)/2, making problems up to roughly p = 150 tractable in standard MCMC.
  • The prior is proper for every positive lambda, so posterior inference is well-defined for any shrinkage rate.
  • Exact sampling is available via a simple algorithm, exponential radius, uniform direction, affine whitening, enabling prior predictive checks and sensitivity analysis before data are observed.
  • When the graph is complete the prior reduces to the unstructured dense case, and when R0 = I it provides a shrinkage-toward-independence default comparable to the LKJ prior but with a dimension-stable rate.

Where Pith is reading between the lines

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

  • The local quadratic approximation linking the whitened norm to true Fisher arc-length may fail far from the reference on the curved manifold; one testable extension is to compute exact geodesic distances numerically and compare them with the whitened norm to assess how well the exponential decay holds globally.
  • The GCP's dependence on vertex labelling, since fill-in count varies with ordering, suggests that graph reordering heuristics could be used as a tuning step to improve the accuracy of the whitening approximation, an idea the paper notes but does not turn into an algorithm.
  • The prior's density vanishes as correlations approach plus or minus 1 for all lambda, unlike the polynomial boundary decay of LKJ; this could affect sparse-data posteriors that seek to allow near-boundary correlations, and a direct posterior comparison would be informative.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper proposes a prior over correlation matrices centred at an arbitrary user-specified reference matrix R0. The construction works in an unconstrained parameterisation θ: it defines a whitened coordinate ξ = H(θ0)^{1/2}(θ−θ0), where H(θ0) is the Hessian of the Kullback-Leibler divergence at R0, and assigns an exponential prior to r = ||ξ|| with a uniform direction on the sphere. The claimed motivation is a PC prior decaying exponentially in Fisher arc-length distance from R0. To handle conditional-independence structure, the paper introduces a Graphical Cholesky Parameterisation (GCP) based on the Cholesky factor of the precision matrix, reducing the number of free parameters to the edge count |E|. The paper proves propriety, contraction to R0, bijectivity of the GCP map, provides a direct sampling algorithm, and illustrates the prior on a heart-disease dataset.

Significance. If the construction is taken as a parameterisation-dependent prior with a local Fisher-geometric interpretation, it fills a genuine gap: LKJ-type priors only shrink toward the identity and do not incorporate conditional-independence sparsity. The GCP is an elegant and potentially useful parameterisation, and the sampling algorithm is a practical contribution. The paper is also commendably explicit about the compatibility constraint between R0 and G and provides a public implementation. However, the central 'Fisher arc-length distance' claim is overstated: the distance used is a tangent-space Euclidean norm, equal to Fisher arc-length only to second order. Relatedly, the prior is not reparameterisation-invariant, despite statements in the text. These issues are fixable by revision, but they affect how the contribution should be framed.

major comments (2)
  1. [Abstract; §3 Eq. (16); §4.4] The abstract and §4.4 claim that the prior 'assigns mass decaying exponentially in the Fisher arc-length distance' and that 'each unit of r is one Fisher distance unit'. However, r in Eqs. (15)–(19) is the Euclidean norm of ξ = H(θ0)^{1/2}(θ−θ0). Eq. (8) identifies this with Fisher arc-length only to second order near θ0. On the curved manifold R_p, the level sets are tangent-space ellipsoids, not geodesic balls; for finite displacements the exponential decay is in the whitened tangent norm, not the true Fisher arc-length. Consequently, the elicitation formula in Prop. 3.1(iii) is a tail probability for this whitened norm, not for a Fisher distance. Please revise the abstract and Prop. 3.1 wording, and either define the prior explicitly as based on the whitened norm (with Fisher interpretation as local) or derive a prior based on the exact KLD with an appropriate level-set measure.
  2. [§2.4; §3; §4 Remark (Choice of diagonal)] The construction (18)–(19) is not invariant to the choice of parameterisation. The Hessian H(θ0) and the uniform measure on the level sets {ξ : ||ξ||=r} are coordinate-dependent. Thus the statement in §2.4 that the resulting prior is 'invariant to reparameterisation' and the claim in §3 that 'the choice is immaterial' for any smooth bijection of the form (12) are not correct for the actual prior defined. Similarly, the forward map (25) is invariant to the diagonal values d_i only in the sense that the same R can be represented by different θ; the induced prior on R depends on d_i. Since d_i is an arbitrary user choice (default p−i+1), this source of parameterisation dependence should be acknowledged, or the prior should be redefined using a coordinate-free geometric measure (e.g., Riemannian volume on geodesic spheres).
minor comments (4)
  1. [§3, Eq. (18)] The density (18) is correct: the radial Jacobian r^{m-1} appears in the denominator, consistent with the Appendix's expression c_m λ e^{-λ r} r^{-(m-1)}. The sentence 'the hyperspherical Jacobian contributes r^{m-1}' could be misread as a multiplication; suggest 'after dividing by the hyperspherical Jacobian r^{m-1}'.
  2. [§4.3 / §5] For m ≥ 2 the prior density (18) is unbounded at θ0. This is acknowledged as harmless for prior draws, but if the log-density is evaluated on the θ-parameterisation in Stan, an interior infinite log-density can cause HMC numerical issues. Please describe how the graphpcor implementation handles this (e.g., non-centred parameterisation using r and u, or a regularized density) and whether any safeguards are used.
  3. [§4 Remark (Choice of diagonal)] The remark 'The forward map (25) is invariant to the choice of positive diagonal values' is only a statement about representability of the same R; the prior itself changes with d_i. This should be stated explicitly, perhaps as a limitation or as motivation for the suggested default d_i = p−i+1.
  4. [§3, paragraph after Eq. (12)] The sentence 'In the dense case the choice is immaterial, however, and the derivation below applies equally to the LKJ parameterisation, or to any other smooth bijection' is misleading in light of the parameterisation dependence of the prior. Please qualify it.

Circularity Check

0 steps flagged

No significant circularity: the prior is constructed from explicit definitions and its stated properties follow by construction; self-citations are not load-bearing.

full rationale

The derivation chain is self-contained. Equations (13)-(19) define the prior directly: D_KL is computed explicitly for centred Gaussians, H(theta0) is its Hessian at the reference, xi is the whitened parameter, r = ||xi|| is the chosen distance coordinate, and (17)-(19) define a normalized product of an Exp(lambda) radius and uniform direction. Propriety, contraction to R0, and tail elicitation are immediate consequences of these definitions, not empirical predictions obtained from fitted constants. The graph parameterisation (Section 4) is an explicit bijection whose validity and bijectivity are either proved in the paper or cited to external sparse-Cholesky/fill-in literature (Rose et al. 1976; Roverato 2002; Atay-Kayis & Massam 2005), not to the authors' prior results. Citations to Simpson et al. (2017), Rue & Held (2005), and Rue et al. (2009) are standard framework or textbook references; the PC construction is restated in the paper itself, so no load-bearing step reduces to an unverified self-citation. The paper even flags that lambda lacks a direct subject-matter interpretation and recommends prior predictive calibration, showing that lambda is an elicited hyperparameter rather than a fitted value relabelled as a prediction. The only substantive caveat — that r equals Fisher arc-length only via the local approximation in Eq. (8) and may not equal global geodesic distance on the curved manifold R_p — is a correctness/approximation concern about the prior's geometric interpretation, not a circularity in the derivation.

Axiom & Free-Parameter Ledger

4 free parameters · 7 axioms · 0 invented entities

The prior is defined by a user-supplied target R0, graph G, and rate λ. The only hand-chosen internal constant is the diagonal sequence d_i, which affects the parameterization. No new physical entities or model components are introduced. The mathematical foundations are standard, though the global distance interpretation is approximate.

free parameters (4)
  • λ (rate parameter) = user-specified (e.g., via tail probability Pr{r>t}=α)
    Controls the strength of shrinkage toward R0; not fitted to data but chosen by the user.
  • Reference correlation matrix R0 = user-specified p×p correlation matrix
    The target of the prior; must be Markov w.r.t. the graph G.
  • Graph G = user-specified edge set
    Encodes conditional independence assumptions; reduces parameters to |E|.
  • Diagonal sequence d_i = p−i+1 = p−i+1
    Hand-chosen to condition the Cholesky factor; the paper claims the map R(θ) is insensitive to row scaling but does not show the induced prior on R is invariant to d_i.
axioms (7)
  • standard math Kullback-Leibler divergence formula for centered Gaussians (Eq 13)
    Used to define the distance between correlation matrices.
  • standard math Local quadratic approximation of KLD by the Fisher metric (Eq 8)
    Underlies the whitening r = ||H(θ0)^{1/2}(θ−θ0)||; only an approximation globally.
  • standard math Cholesky fill-in theory (Lemma 2.1, Rose et al. 1976, Rue & Held 2005)
    Used to separate edge, fill-in, and zero entries of the Cholesky factor of a G-sparse precision.
  • domain assumption PC prior framework of Simpson et al. (2017)
    Adopted as the construction principle for the prior; the paper extends it to arbitrary base models.
  • standard math The GCP map θ ↦ R(θ) is a bijection onto ℛp(𝒢)
    Proved in Proposition 4.1; relies on uniqueness of Cholesky factor and row rescaling.
  • domain assumption The user-supplied graph G correctly encodes the true conditional independence structure
    The prior support is restricted to ℛp(𝒢); if G is misspecified the posterior cannot represent the true dependence.
  • domain assumption R0 is Markov w.r.t. G (Algorithm 1, Step 1)
    If Q0 has nonzeros at non-edges, R0 cannot be a reference under G; the user must extend G or adjust R0.

pith-pipeline@v1.3.0-alltime-deepseek · 21395 in / 14415 out tokens · 119339 ms · 2026-08-01T13:55:26.019837+00:00 · methodology

0 comments
read the original abstract

Specifying a prior over the space of correlation matrices is a persistent challenge in Bayesian analysis. The space is a curved manifold whose dimension grows quadratically with the number of variables, making substantive prior beliefs difficult to encode.\\ We propose a distance-based prior that assigns mass decaying exponentially in the Fisher arc-length distance from a user-specified reference correlation matrix, enabling shrinkage toward any target correlation structure rather than being confined to the identity matrix. Formally, this is constructed as a Penalised Complexity prior, but its interpretation shifts accordingly: unless the chosen target represents a structurally simpler state, the shrinkage penalises deviation rather than complexity in the usual sense. To accommodate conditional independence constraints, we introduce a parameterisation that constructs the correlation matrix via the Cholesky factor of the inverse correlation matrix with respect to a user-supplied graph, thereby reducing the number of free parameters from one per variable pair to one per graph edge. The prior is proper for every positive value of its rate parameter, accommodates correlations of either sign under any graph structure, and reduces to a fully unstructured prior when the graph is complete. A direct sampling algorithm is provided, enabling prior predictive checks and sensitivity analysis, implemented within the \texttt{graphpcor} package.

Figures

Figures reproduced from arXiv: 2607.18926 by Anna Freni-Sterrantino, Denis Rustand, Elias Teixeira Krainski, H{\aa}vard Rue, Haziq Jamil, Janet Van Niekerk.

Figure 1
Figure 1. Figure 1: Star graph 𝒢 with 𝑝 = 5 nodes and the hub at vertex 1. Shown are the lower￾triangular sparsity patterns for 𝑄 (left) and its Cholesky factor 𝐿 (right); • denotes a structurally non-zero entry and a blank denotes a structural zero. In 𝐿, ∘ denotes generically non-zero fill-in entries at non-edge positions. The filled graph 𝒢 ∗ (right) shows solid edges (ℰ) matching the non-zero entries of 𝑄, and dashed edge… view at source ↗
Figure 2
Figure 2. Figure 2: Same star graph as Figure [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Informative prior on the bivariate correlation under three parameterisations [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Marginal prior densities for the three pairwise correlations [PITH_FULL_IMAGE:figures/full_fig_p030_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Two dependency graphs on the nine covariates of the South African heart data, [PITH_FULL_IMAGE:figures/full_fig_p033_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Posterior summaries for the South African heart data under the proposed prior [PITH_FULL_IMAGE:figures/full_fig_p035_6.png] view at source ↗

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

35 extracted references · 6 canonical work pages

  1. [1]

    Information

    Amari, Shun-ichi , year = 2016, series =. Information. doi:10.1007/978-4-431-55978-8 , urldate =

    doi:10.1007/978-4-431-55978-8 2016

  2. [2]

    Circulation , volume=

    Red and processed meat consumption and risk of incident coronary heart disease, stroke, and diabetes mellitus: a systematic review and meta-analysis , author=. Circulation , volume=. 2010 , publisher=

    2010

  3. [3]

    Diabetes research and clinical practice , volume=

    Sleep duration and smoking are associated with coronary heart disease among US adults with type 2 diabetes: gender differences , author=. Diabetes research and clinical practice , volume=. 2017 , publisher=

    2017

  4. [4]

    BMC Public Health , volume=

    Bayesian network analysis of factors influencing type 2 diabetes, coronary heart disease, and their comorbidities , author=. BMC Public Health , volume=. 2024 , publisher=

    2024

  5. [5]

    The CORIS baseline study , author=

    Coronary risk factor screening in three rural communities. The CORIS baseline study , author=. South African Medical Journal , volume=

  6. [6]

    Biometrika , volume =

    A. Biometrika , volume =. doi:10.1093/biomet/92.2.317 , urldate =

    work page doi:10.1093/biomet/92.2.317

  7. [7]

    Statistica Sinica , volume =

    Modeling Covariance Matrices in Terms of Standard Deviations and Correlations, with Application to Shrinkage , author =. Statistica Sinica , volume =

  8. [8]

    Bartlett, M. S. , year = 1934, journal =. On the. doi:10.1017/S0370164600015637 , urldate =

    work page doi:10.1017/s0370164600015637 1934

  9. [9]

    Wishart Distributions:

    Bekker, Andri. Wishart Distributions:. Journal of Multivariate Analysis , volume =

  10. [10]

    , year = 2003, edition =

    Boothby, William M. , year = 2003, edition =. An Introduction to Differentiable Manifolds and

    2003

  11. [11]

    and Lee, Daniel and Goodrich, Ben and Betancourt, Michael and Brubaker, Marcus and Guo, Jiqiang and Li, Peter and Riddell, Allen , year = 2017, journal =

    Carpenter, Bob and Gelman, Andrew and Hoffman, Matthew D. and Lee, Daniel and Goodrich, Ben and Betancourt, Michael and Brubaker, Marcus and Guo, Jiqiang and Li, Peter and Riddell, Allen , year = 2017, journal =. doi:10.18637/jss.v076.i01 , urldate =

    doi:10.18637/jss.v076.i01 2017

  12. [12]

    and Kass, Robert E

    Daniels, Michael J. and Kass, Robert E. , year = 1999, journal =. Nonconjugate. doi:10.1080/01621459.1999.10473878 , urldate =

    arXiv 1999

  13. [13]

    , year = 2006, publisher =

    Davis, Timothy A. , year = 2006, publisher =. Direct. doi:10.1137/1.9780898718881 , urldate =

    doi:10.1137/1.9780898718881 2006

  14. [14]

    Dempster, A. P. , year = 1972, journal =. Covariance. doi:10.2307/2528966 , urldate =. 2528966 , eprinttype =

    work page doi:10.2307/2528966 1972

  15. [15]

    , year = 2017, journal =

    Drton, Mathias and Maathuis, Marloes H. , year = 2017, journal =. Structure. doi:10.1146/annurev-statistics-060116-053803 , urldate =

    doi:10.1146/annurev-statistics-060116-053803 2017

  16. [16]

    and Borsboom, Denny , editor =

    Epskamp, Sacha and Maris, Gunter and Waldorp, Lourens J. and Borsboom, Denny , editor =. Network. The. doi:10.1002/9781118489772.ch30 , urldate =

    work page doi:10.1002/9781118489772.ch30

  17. [17]

    Statistical Methods & Applications , volume =

    A Graphical Framework for Interpretable Correlation Matrix Models for Multivariate Regression , author =. Statistical Methods & Applications , volume =. doi:10.1007/s10260-025-00788-y , urldate =

    work page doi:10.1007/s10260-025-00788-y

  18. [18]

    Visualization in

    Gabry, Jonah and Simpson, Daniel and Vehtari, Aki and Betancourt, Michael and Gelman, Andrew , year = 2019, journal =. Visualization in. doi:10.1111/rssa.12378 , urldate =

    doi:10.1111/rssa.12378 2019

  19. [19]

    Krainski, Elias Teixeira and Rustand, Denis and

  20. [20]

    Graphical

    Lauritzen, Steffen L , year = 1996, publisher =. Graphical. doi:10.1093/oso/9780198522195.001.0001 , urldate =

    arXiv 1996

  21. [21]

    Journal of Multivariate Analysis , volume =

    Generating Random Correlation Matrices Based on Vines and Extended Onion Method , author =. Journal of Multivariate Analysis , volume =. doi:10.1016/j.jmva.2009.04.008 , urldate =

    doi:10.1016/j.jmva.2009.04.008 2009

  22. [22]

    Differential Geometry and Statistics , author =

  23. [23]

    PeerJ Computer Science , volume =

  24. [24]

    and Tarjan, R

    Rose, Donald J. and Tarjan, R. Endre and Lueker, George S. , year = 1976, journal =. Algorithmic. doi:10.1137/0205021 , urldate =

    doi:10.1137/0205021 1976

  25. [25]

    Roverato, Alberto , year = 2002, journal =. Hyper. doi:10.1111/1467-9469.00297 , urldate =

    arXiv 2002

  26. [26]

    Gaussian

    Rue, H. Gaussian. doi:10.1201/9780203492024 , abstract =

    doi:10.1201/9780203492024

  27. [27]

    Approximate

    Rue, H. Approximate. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =. doi:10.1111/j.1467-9868.2008.00700.x , urldate =

    arXiv 2008

  28. [28]

    Penalising Model Component Complexity:

    Simpson, Daniel and Rue, H. Penalising Model Component Complexity:. Statistical Science , volume =. doi:10.1214/16-STS576 , urldate =

    doi:10.1214/16-sts576

  29. [29]

    Speed, T. P. and Kiiveri, H. T. , year = 1986, journal =. Gaussian. doi:10.1214/aos/1176349846 , urldate =

    arXiv 1986

  30. [30]

    Differential Geometry , author =

  31. [31]

    Journal of Data Science, Statistics, and Visualisation , volume =

    Visualizing Distributions of Covariance Matrices , author =. Journal of Data Science, Statistics, and Visualisation , volume =

  32. [32]

    Wermuth, Nanny , year = 1980, journal =. Linear. doi:10.1080/01621459.1980.10477580 , urldate =

    arXiv 1980

  33. [33]

    Biometrika , volume =

    The Generalised Product Moment Distribution in Samples from a Normal Multivariate Population , author =. Biometrika , volume =. 2331939 , eprinttype =

  34. [34]

    and Rue, H

    Wist, Hanne T. and Rue, H. Specifying a. Communications in Statistics - Simulation and Computation , volume =. doi:10.1080/03610910500416108 , urldate =

    work page doi:10.1080/03610910500416108

  35. [35]

    Computing the

    Yannakakis, Mihalis , year = 1981, journal =. Computing the. doi:10.1137/0602010 , urldate =

    doi:10.1137/0602010 1981