pith. sign in

arxiv: 2510.19409 · v1 · submitted 2025-10-22 · 🌌 astro-ph.GA · astro-ph.IM

Modeling Globular Cluster Counts with Bayesian Latent Models

Rafael S. de Souza , Ana L. Chies-Santos This is my paper

Pith reviewed 2026-05-18 04:52 UTC · model grok-4.3

classification 🌌 astro-ph.GA astro-ph.IM
keywords globular clustersBayesian latent modelsnegative binomialscaling relationsstellar massmeasurement errorscount processesgalaxy populations
0
0 comments X

The pith

A Bayesian latent model adds a Gaussian observation layer to negative-binomial counts for globular cluster scaling relations.

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

This paper updates an earlier Bayesian framework to model how the number of globular clusters scales with the stellar mass of their host galaxies. It places a Gaussian observation layer atop the negative-binomial count process so that measurement uncertainties can be propagated more efficiently. The change keeps the original count distribution unchanged while making the calculations easier to run in standard probabilistic programming tools. Readers would care because cleaner error handling can produce more trustworthy estimates of how galaxies assemble their star-cluster populations.

Core claim

We present a Bayesian latent model to describe the scaling relation between globular cluster populations and their host galaxies, updating the framework proposed in 2015. GC counts are drawn from a negative-binomial process linked to host stellar mass, augmented with a newly introduced Gaussian observation layer that enables efficient propagation of measurement errors. The revised formulation preserves the underlying NB process while improving computational tractability.

What carries the argument

Bayesian latent model that couples a negative-binomial count process to a Gaussian observation layer for error propagation.

Load-bearing premise

Adding a Gaussian observation layer on top of the negative-binomial count process does not distort the inferred scaling relation between globular cluster number and host stellar mass.

What would settle it

A statistically significant shift in the posterior parameters of the scaling relation when the same globular cluster dataset is fit once with and once without the Gaussian observation layer.

Figures

Figures reproduced from arXiv: 2510.19409 by Ana L. Chies-Santos, Rafael S. de Souza.

Figure 1
Figure 1. Figure 1: Negative–binomial regression with a latent errors-in-variables treatment for the relation between NGC and log10(M⋆/M⊙). Data points are elliptical galaxies from Harris et al. (2013). The solid blue curve shows the posterior me￾dian, and the orange band the 95% credible interval. Points include 1σ uncertainties. Berek, S. C., Eadie, G. M., Speagle, J. S., & Harris, W. E. 2023, ApJ, 955, 22, doi: 10.3847/153… view at source ↗
read the original abstract

We present a Bayesian latent model to describe the scaling relation between globular cluster populations and their host galaxies, updating the framework proposed in de Souza 2015. GC counts are drawn from a negative-binomial (NB) process linked to host stellar mass, augmented with a newly introduced Gaussian observation layer that enables efficient propagation of measurement errors. The revised formulation preserves the underlying NB process while improving computational tractability. The code snippets, implemented in Nimble and PyMC are released under the MIT license at https://github.com/COINtoolbox/Generalized-Linear-Models-Tutorial/blob/master/Count/readme.md

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

1 major / 1 minor

Summary. The paper presents a Bayesian latent model updating the de Souza 2015 framework for the scaling relation between globular cluster (GC) counts and host galaxy stellar mass. GC counts are modeled via a negative-binomial process linked to stellar mass, with a new Gaussian observation layer added to propagate measurement errors; the authors claim this preserves the underlying NB process while improving computational tractability. Public code releases in Nimble and PyMC are provided.

Significance. If the central claim holds, the model offers a tractable way to incorporate measurement errors into count-based scaling relations in extragalactic astronomy, building on standard NB and Gaussian assumptions with reproducible code. This could aid future analyses of GC populations, but the absence of validation metrics, posterior predictive checks, or direct comparison to the 2015 baseline limits the assessed impact.

major comments (1)
  1. [Abstract / Model formulation] Abstract and model description: the claim that the Gaussian observation layer 'preserves the underlying NB process' is not supported by any simulation-based recovery test or comparison to the de Souza 2015 baseline. In low-count regimes, the continuous Gaussian layer on discrete NB counts risks shifting the effective likelihood and biasing the inferred power-law index or normalization; a concrete check (e.g., simulated data recovery of the scaling parameters) is required to substantiate the central claim.
minor comments (1)
  1. [Abstract] The GitHub link in the abstract points to a tutorial repository; confirm that the exact model implementation and example scripts used in the paper are included or linked.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comments and the recommendation for major revision. We address the single major comment below and will incorporate the suggested validation in the revised manuscript.

read point-by-point responses

  1. Referee: [Abstract / Model formulation] Abstract and model description: the claim that the Gaussian observation layer 'preserves the underlying NB process' is not supported by any simulation-based recovery test or comparison to the de Souza 2015 baseline. In low-count regimes, the continuous Gaussian layer on discrete NB counts risks shifting the effective likelihood and biasing the inferred power-law index or normalization; a concrete check (e.g., simulated data recovery of the scaling parameters) is required to substantiate the central claim.

    Authors: We agree that the manuscript currently lacks explicit simulation-based recovery tests and direct comparisons to the de Souza 2015 baseline, and that such checks are needed to substantiate the claim, particularly regarding potential biases in low-count regimes. In the revised version we will add a dedicated validation section that generates synthetic datasets from the negative-binomial process with known scaling parameters, applies the latent Gaussian observation layer, and reports recovery of the power-law index and normalization. These tests will explicitly include low-count regimes and will be accompanied by posterior predictive checks. We will also provide a side-by-side comparison of posterior distributions obtained with and without the observation layer and against the original de Souza 2015 formulation. These additions will directly address the concern while preserving the computational advantages of the model. revision: yes

Circularity Check

0 steps flagged

Minor self-citation to de Souza 2015 framework; central model uses standard NB and Gaussian assumptions with public code

full rationale

The paper updates a prior framework from de Souza 2015 but introduces a new Gaussian observation layer atop the standard negative-binomial count process linked to host stellar mass. No derivation step reduces by construction to a fitted parameter or self-defined quantity from the authors' prior work; the claim of preserving the NB process while adding tractability rests on conventional statistical modeling choices. The release of Nimble and PyMC code under MIT license provides an independent verification path, rendering the overall derivation self-contained rather than circular.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The model rests on the domain assumption that globular-cluster counts follow a negative-binomial distribution whose mean is a function of stellar mass, plus the standard assumption that measurement errors can be treated as Gaussian.

free parameters (1)
  • negative-binomial dispersion parameter
    Dispersion parameter of the NB process is expected to be estimated from data as part of the latent model.
axioms (1)
  • domain assumption Globular cluster counts are generated by a negative-binomial process whose expectation depends on host stellar mass
    Core modeling choice stated in the abstract.

pith-pipeline@v0.9.0 · 5626 in / 1209 out tokens · 42173 ms · 2026-05-18T04:52:05.911357+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

16 extracted references · 16 canonical work pages

  1. [1]

    The Optical and Near-Infrared Properties of Galaxies: I. Luminosity and Stellar Mass Functions

    Bell, E. F., McIntosh, D. H., Katz, N., & Weinberg, M. D. 2003, ApJS, 149, 289, doi: 10.1086/378847 Bayesian Globular Cluster Counts3 0 101 102 103 104 8 10 12 log10 (M★ M☉) NGC Figure 1.Negative–binomial regression with a latent errors-in-variables treatment for the relation betweenN GC and log10(M⋆/M⊙). Data points are elliptical galaxies from Harris et...

    work page Pith review doi:10.1086/378847 2003

  2. [2]

    C., Eadie, G

    Berek, S. C., Eadie, G. M., Speagle, J. S., & Harris, W. E. 2023, ApJ, 955, 22, doi: 10.3847/1538-4357/ace7b7

    work page doi:10.3847/1538-4357/ace7b7 2023

  3. [3]

    C., Eadie, G

    Berek, S. C., Eadie, G. M., Speagle, J. S., & Wang, S. Y. 2024, ApJ, 972, 104, doi: 10.3847/1538-4357/ad6147

    work page doi:10.3847/1538-4357/ad6147 2024

  4. [4]

    Blakeslee, J. P. 1997, ApJL, 481, L59, doi: 10.1086/310653

    work page doi:10.1086/310653 1997

  5. [5]

    A., Chies-Santos, A

    Canossa-Gosteinski, M. A., Chies-Santos, A. L., Furlanetto, C., et al. 2024, MNRAS, 534, 1729, doi: 10.1093/mnras/stae2161 de Souza, R. S., Hilbe, J. M., Buelens, B., et al. 2015a, MNRAS, 453, 1928, doi: 10.1093/mnras/stv1825 de Souza, R. S., Cameron, E., Killedar, M., et al. 2015b, Astronomy and Computing, 12, 21, doi: 10.1016/j.ascom.2015.04.002 de Valp...

    work page doi:10.1093/mnras/stae2161 2024

  6. [6]

    Dornan, V., & Harris, W. E. 2025, ApJ, 988, 70, doi: 10.3847/1538-4357/ade05e

    work page doi:10.3847/1538-4357/ade05e 2025

  7. [7]

    S., Krone-Martins, A., et al

    Elliott, J., de Souza, R. S., Krone-Martins, A., et al. 2015, Astronomy and Computing, 10, 61, doi: 10.1016/j.ascom.2015.01.002

    work page doi:10.1016/j.ascom.2015.01.002 2015

  8. [8]

    Harris, G. L. H., Poole, G. B., & Harris, W. E. 2014, MNRAS, 438, 2117, doi: 10.1093/mnras/stt2337

    work page doi:10.1093/mnras/stt2337 2014

  9. [9]

    E., Harris, G

    Harris, W. E., Harris, G. L. H., & Alessi, M. 2013, ApJ, 772, 82, doi: 10.1088/0004-637X/772/2/82

    work page doi:10.1088/0004-637x/772/2/82 2013

  10. [10]

    W., de Souza, R

    Hattab, M. W., de Souza, R. S., Ciardi, B., et al. 2019, MNRAS, 483, 3307, doi: 10.1093/mnras/sty3314

    work page doi:10.1093/mnras/sty3314 2019

  11. [11]

    M., de Souza, R

    Hilbe, J. M., de Souza, R. S., & Ishida, E. E. O. 2017, Bayesian Models for Astrophysical Data Using R, JAGS, Python, and Stan, doi: 10.1017/CBO9781316459515

    work page doi:10.1017/cbo9781316459515 2017

  12. [12]

    J., Harris, G

    Hudson, M. J., Harris, G. L., & Harris, W. E. 2014, ApJL, 787, L5, doi: 10.1088/2041-8205/787/1/L5

    work page doi:10.1088/2041-8205/787/1/l5 2014

  13. [13]

    2022, MNRAS, 511, 4633, doi: 10.1093/mnras/stac328

    Saifollahi, T., Zaritsky, D., Trujillo, I., et al. 2022, MNRAS, 511, 4633, doi: 10.1093/mnras/stac328

    work page doi:10.1093/mnras/stac328 2022

  14. [14]

    Wiecki, and Christopher Fonnesbeck

    Salvatier, J., Wiecki, T. V., & Fonnesbeck, C. 2016, PeerJ Computer Science, 2, e55, doi: 10.7717/peerj-cs.55

    work page doi:10.7717/peerj-cs.55 2016

  15. [15]

    M., Moster, B

    Valenzuela, L. M., Moster, B. P., Remus, R.-S., O’Leary, J. A., & Burkert, A. 2021, MNRAS, 505, 5815, doi: 10.1093/mnras/stab1701

    work page doi:10.1093/mnras/stab1701 2021

  16. [16]

    2022, MNRAS, 513, 2609, doi: 10.1093/mnras/stac1072

    Zaritsky, D. 2022, MNRAS, 513, 2609, doi: 10.1093/mnras/stac1072

    work page doi:10.1093/mnras/stac1072 2022