pith. sign in

arxiv: 2606.29842 · v1 · pith:ADU5MCSZnew · submitted 2026-06-29 · ⚛️ physics.data-an · astro-ph.CO· stat.ME

The Squealer: Sensification of model exploration and model misfit

Andrew Gelman , Andrew H. Jaffe , Eliot Carlson , Philip Greengard This is my paper

Pith reviewed 2026-06-30 03:57 UTC · model grok-4.3

classification ⚛️ physics.data-an astro-ph.COstat.ME
keywords model misfitauditory feedbackinteractive explorationcurve fittingdata visualizationstatistical modelingsquealer methodmodel checking
0
0 comments X

The pith

Dragging a model curve emits a squeal that grows louder and harsher as the fit to data worsens.

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

The paper presents a method that adds sound to the process of checking how well a curve matches observed data points. A user adjusts the curve by hand and hears an increasingly unpleasant noise whenever the adjustment moves the curve away from the points. The approach is shown on four different data sets, ranging from simple two-parameter fits to nonparametric models. If the method works as intended, it turns an often silent visual inspection into an immediate sensory signal that highlights discrepancies without requiring separate diagnostic plots.

Core claim

The central claim is that auditory feedback, implemented as a squeal whose volume and unpleasantness increase with the discrepancy between a user-adjusted curve and the data, can be combined with visual display to support interactive exploration and detection of model misfit.

What carries the argument

The squealer: an auditory signal whose intensity and character are driven directly by a quantitative measure of curve-data discrepancy.

If this is right

  • Interactive adjustment of two-parameter curves, such as those for golf-putting data, immediately signals worsening fit through sound.
  • Four-parameter models fitted to dilution-assay data become easier to tune because large residuals produce an audible cue.
  • Cosmological parameter fits sensitive to Big Bang model values gain real-time auditory confirmation of alignment with observations.
  • Nonparametric Gaussian process fits to temperature series allow users to hear when local adjustments create excess discrepancy.

Where Pith is reading between the lines

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

  • The same principle could be applied to other sensory channels, such as vibration or color shifts, when auditory output is impractical.
  • Embedding the feedback in standard statistical software might lower the barrier for non-statisticians to perform informal model checks.
  • The method might be extended to higher-dimensional parameter spaces by mapping multiple discrepancy measures to different sound attributes.

Load-bearing premise

That the generated squeal will be noticeable and informative enough to help users detect misfits during real-time curve adjustment.

What would settle it

A user study in which participants adjust curves to minimize misfit with and without the squeal, then measure whether the squeal version produces systematically better final fits or faster detection of obvious mismatches.

Figures

Figures reproduced from arXiv: 2606.29842 by Andrew Gelman, Andrew H. Jaffe, Eliot Carlson, Philip Greengard.

Figure 1
Figure 1. Figure 1: The basic Squealer. Left: scatterplot of data and fitted model (dark blue curve representing the point estimate gˆ and light blue curves representing posterior simulations g ∗ ), a pseudo-data point (x ∗ , y∗ ) in red, and the new curve g ∗ in red. Center: dashboard showing, for each parameter θk in the model, the point estimate ˆθk as a blue dot, the posterior density from the simulations θ s in blue, and… view at source ↗
Figure 2
Figure 2. Figure 2: Example of the Squealer for data that are not consistent with the model. In this case there is no way to get the curve close to the data: pulling up the curve to improve the fit for the point at x = 200 degrades the fit elsewhere. The above displays show two attempts, first adding one pseudo-data point and then adding another. In every case, pulling toward the pseudo-data decreases the log posterior densit… view at source ↗
Figure 3
Figure 3. Figure 3: Challenge of shifting the posterior distribution by pulling it toward a pseudo-data point. Left graph: data and fitted curves g(x|θ s ) based on 100 random draws θ s from the posterior distri￾bution, p(θ|y), for the data and model shown in [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Data on the proportion of successful golf putts as a function of distance from the hole, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Top row: first try to adjust the golf model to line up with the data, obtained by dragging the curve upward at one point at x = 20. Bottom row: second try, dragging the curve at two points (x = 10 and x = 30), yields a much improved visual fit to the data. Point-by-point details are shown in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Squealer applied to the golf model. It is possible to pull the fitted curve g(x| ˆθ) toward two pseudo-data points and obtain a new curve g ∗ = g(x|θ ∗ ) that is a better visual match to the data, but the log-posterior density for θ ∗ is much lower. The rightmost plot shows that the poor fit is coming from the data points with the lowest values of x. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: A model for laboratory assays fit to calibration data (top row) and samples with unknown concentrations (remaining rows). For each, data have been gathered at multiple dilutions, and the curves show expected measurement value as a function of the dilution level. The unknown samples are displayed in decreasing order of mean measurements. The curves for samples 23 and 3 show some misfit to the data; we explo… view at source ↗
Figure 8
Figure 8. Figure 8: Applying the Squealer to the data and model from [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Parameter estimates corresponding to the fitted models in Figures 7 and 8. The model has two parameters that vary across the 23 unknown samples: β2, which is proportional to the concen￾tration of the compound of interest within the sample, and σy, the scale of modeling/measurement error. The plots show the posterior estimates ±1 standard error for log β2 and log σy for each sam￾ple. Perturbing the fit alte… view at source ↗
Figure 10
Figure 10. Figure 10: The Squealer applied to the CMB power spectrum as measured by the Planck satellite. The x-axis is on an idiosyncratic scale to more evenly display the data. Here, Dℓ = ℓ(ℓ+ 1)Cℓ/(2π) as is traditionally plotted in cosmology. Panel (a) shows the data (blue error bars), the overall best-fit (posterior mode) spectrum (red curve) and residuals between the two (red points in bottom panel, standard error units)… view at source ↗
Figure 11
Figure 11. Figure 11: Fitted curves corresponding to draws from the GP posterior with pseudo-data, fixing the fitted hyperparameters of the covariance kernel. ℓ α σ original 0.23 0.52 0.10 one pseudo-point 0.23 0.48 0.14 two pseudo-points 0.26 0.58 0.17 [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
read the original abstract

We introduce a method for visual and auditory feedback when exploring the fit of a model to data. Starting with a best-fit curve fit to data, the user can drag the curve to a new position and the computer will emit a squeal, becoming louder and more unpleasant as the discrepancy between curve and data increases. We demonstrate with four examples: a two-parameter curve fit to golf putting data, a four-parameter curve fit to dilution assays, a fit to cosmological data sensitive to the parameters of the Big Bang model, and a nonparametric Gaussian process fit to temperature readings.

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 introduces 'The Squealer', a method for visual and auditory feedback when exploring model fits to data. Starting from a best-fit curve, users drag the curve and receive a squeal whose volume and unpleasantness increase with growing discrepancy to the data points. The approach is illustrated via four examples: a two-parameter fit to golf putting data, a four-parameter fit to dilution assays, a cosmological fit sensitive to Big Bang parameters, and a nonparametric Gaussian process fit to temperature data.

Significance. If the chosen audio mapping can be shown to improve misfit detection, the technique could provide a practical multimodal aid for interactive model exploration in data analysis. The manuscript presents a direct conceptual proposal with no self-referential derivations or fitted quantities, and the four examples serve only as illustrations rather than tests of efficacy.

major comments (1)
  1. [Abstract] Abstract: the central claim that the squealing feedback 'meaningfully aids' interactive exploration and misfit detection rests on an untested assumption; the four examples demonstrate only the mapping from discrepancy to sound properties and supply no quantitative metrics, error analysis, user testing, or visual-only baseline comparisons.
minor comments (1)
  1. The term 'sensification' in the title is not defined or motivated in the provided text and may require a brief explanation for readers outside visualization or HCI communities.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review. The manuscript presents a conceptual proposal for an auditory feedback technique, with the examples serving strictly as illustrations rather than efficacy tests. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the squealing feedback 'meaningfully aids' interactive exploration and misfit detection rests on an untested assumption; the four examples demonstrate only the mapping from discrepancy to sound properties and supply no quantitative metrics, error analysis, user testing, or visual-only baseline comparisons.

    Authors: We agree there is no user testing, quantitative metrics, or baseline comparisons in the manuscript; the four examples illustrate application of the discrepancy-to-sound mapping across contexts (golf putting, dilution assays, cosmology, Gaussian processes) but do not evaluate performance gains. The provided abstract introduces the method and notes demonstration via examples without asserting empirical superiority. We will revise the abstract and introduction to explicitly frame the work as a conceptual proposal and remove any phrasing that could be read as claiming meaningful aid, thereby aligning the text with the illustrative scope. revision: partial

Circularity Check

0 steps flagged

No circularity: direct interface proposal without derivations or self-referential fits

full rationale

The paper introduces an auditory-visual feedback interface for model exploration but contains no equations, parameter fits, predictions, or derivations. Its central contribution is a proposed mapping from curve-data discrepancy to sound intensity, demonstrated via four qualitative examples. No load-bearing step reduces to a self-definition, fitted input renamed as prediction, or self-citation chain; the work is self-contained as a methodological suggestion with no mathematical claims that could be circular.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper proposes a new interface method without introducing fitted parameters, new axioms, or postulated entities beyond standard statistical curve fitting.

pith-pipeline@v0.9.1-grok · 5628 in / 904 out tokens · 47796 ms · 2026-06-30T03:57:23.253975+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

18 extracted references · 5 canonical work pages · 5 internal anchors

  1. [1]

    A unified pseudo-$C_\ell$ framework

    Alonso, David, Javier Sanchez, and Anˇ ze Slosar (2019). “A unified pseudo-C ℓ framework”. In: Monthly Notices of the Royal Astronomical Society484, pp. 4127–4151.doi:10.1093/mnras/ stz093. eprint:1809.09603

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/ 2019

  2. [2]

    Radical Compression of Cosmic Microwave Background Data

    Bond, J. R., A. H. Jaffe, and L. Knox (2000). “Radical compression of cosmic microwave background data”. In:Astrophysical Journal533, p. 19.doi:10.1086/308625. eprint:astro-ph/9808264

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1086/308625 2000

  3. [3]

    The sound of science

    Bornmann, Lutz (2024). “The sound of science”. In:EMBO Reports25, pp. 3743–3747

    2024

  4. [4]

    Two simple putting models in golf

    Broadie, Mark (2018). “Two simple putting models in golf”. In:https://statmodeling.stat. columbia.edu/wp-content/uploads/2019/03/putt_models_20181017.pdf. Day´ e, Christian and Alberto de Campo (2006). “Sounds sequential: sonification in the social sci- ences”. In:Interdisciplinary Science Reviews31, pp. 349–364

    2018

  5. [5]

    San Diego: Academic Press

    Dodelson, Scott (2003).Modern Cosmology. San Diego: Academic Press

    2003

  6. [6]

    Exploratory data analysis for complex models (with discussion)

    Gelman, Andrew (2004). “Exploratory data analysis for complex models (with discussion)”. In: Journal of Computational and Graphical Statistics13, pp. 755–787

    2004

  7. [7]

    Model building and expansion for golf putting

    Gelman, Andrew (2019). “Model building and expansion for golf putting”. In:Stan Case Studies 6.https://mc-stan.org/users/documentation/case-studies/golf.html. 17

    2019

  8. [8]

    The typical set and its relevance to Bayesian computation

    Gelman, Andrew (2020). “The typical set and its relevance to Bayesian computation”. In:Statisti- cal Modeling, Causal Inference, and Social Science. 2 Aug.https : / / statmodeling . stat . columbia . edu / 2020 / 08 / 02 / the - typical - set - and - its - relevance - to - bayesian - computation/

    2020

  9. [9]

    Bayesian analysis of serial dilu- tion assays

    Gelman, Andrew, Ginger Chew, and Michael Shnaidman (2004). “Bayesian analysis of serial dilu- tion assays”. In:Biometrics60, pp. 407–417

    2004

  10. [10]

    From visualization to sensification

    Gelman, Andrew and S. Gwynn Sturdevant (2023). “From visualization to sensification”. In:Amstat News547, pp. 18–19

    2023

  11. [11]

    (2026).Bayesian Workflow

    Gelman, Andrew, Aki Vehtari, Richard McElreath, et al. (2026).Bayesian Workflow. London: CRC Press

    2026

  12. [12]

    MASTER of the CMB Anisotropy Power Spectrum: A Fast Method for Statistical Analysis of Large and Complex CMB Data Sets

    Hivon, E., K. M. G´ orski, C. B. Netterfield, B. P. Crill, S. Prunet, and F. Hansen (2002). “MASTER of the cosmic microwave background anisotropy power spectrum: A fast method for statistical analysis of large and complex cosmic microwave background data sets”. In:Astrophysical Jour- nal567, p. 2.doi:10.1086/338126. eprint:astro-ph/0105302

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1086/338126 2002

  13. [13]

    The Physics of Microwave Background Anisotropies

    Hu, Wayne, Naoshi Sugiyama, and Joseph Silk (1997). “The physics of microwave background anisotropies”. In:Nature386, pp. 37–43.doi:10.1038/386037a0. eprint:astro-ph/9504057

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1038/386037a0 1997

  14. [14]

    Planck 2018 results. V. CMB power spectra and likelihoods

    Liu, Jun S. and Rong Chen (1998). “Sequential Monte Carlo methods for dynamic systems”. In: Journal of the American Statistical Association93, pp. 1032–1044. Planck Collaboration, N. Aghanim, Y. Akrami, M. Ashdown, et al. (2020a). “Planck 2018 results. V. CMB power spectra and likelihoods”. In:Astronomy & Astrophysics641, A5.doi:10.1051/ 0004-6361/2018363...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1051/0004- 1998

  15. [15]

    Rasmussen, Carl Edward and Christopher K. I. Williams (2005).Gaussian Processes for Machine Learning. MIT Press

    2005

  16. [16]

    Delivering data differently

    Sturdevant, S. Gwynn, A. Jonathan R. Godfrey, and Andrew Gelman (2022). “Delivering data differently”. In:https://arxiv.org/abs/2204.10854

    arXiv 2022

  17. [17]

    Exploratory model analysis with R and GGobi

    Wickham, Hadley (2006). “Exploratory model analysis with R and GGobi”. In:https://had.co. nz/model-vis/2007-jsm.pdf

    2006

  18. [18]

    Visualizing statistical models: Re- moving the blindfold

    Wickham, Hadley, Dianne Cook, and Heike Hofmann (2015). “Visualizing statistical models: Re- moving the blindfold”. In:Statistical Analysis and Data Mining8, pp. 203–225. 18

    2015