Tell Me Without Telling Me: Two-Way Prediction of Visualization Literacy and Visual Attention

Andreas Bulling; Cindy Xiong Bearfield; Huichen Will Wang; Minsuk Chang; Yao Wang; Yuanhong Zhou

REVIEW 2 major objections 2 minor 2 references

Tell Me Without Telling Me: Two-Way Prediction of Visualization Literacy and Visual Attention

T0 review · 2 major / 2 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Eye-movement patterns predict visualization literacy with 86% accuracy.

desk verdict The supplied full text is an unrelated tax-reform paper, so the actual visualization-literacy study is unverifiable from this package; the abstract is promising but cannot be reviewed. read the letter →

arxiv 2508.03713 v1 pith:G5ZCVSXG submitted 2025-07-22 cs.HC cs.CV

Minsuk Chang , Yao Wang , Huichen Will Wang , Yuanhong Zhou , Andreas Bulling , Cindy Xiong Bearfield This is my paper · ORCID

classification cs.HCcs.CV

keywords visualizationliteracyvisualattentionsaliencymodeleyetrackingindividualdifferencesLit2SalSal2Lituserstudy

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish a two-way link between visual attention during data exploration and visualization literacy. Analyzing eye-movement data from 235 participants across three literacy tests (mini-VLAT, CALVI, SGL), it reports that high scorers show strong attentional focus while low scorers explore more diffusely. On that basis it builds two models: Lit2Sal predicts an observer's attention map from their literacy level, and Sal2Lit predicts literacy from a single attention map, reaching 86% accuracy in under a minute. The payoff, if the link holds, is fast literacy assessment and visualization designs that adapt to the reader.

What carries the argument

The load-bearing objects are the two named models. Lit2Sal is a visual saliency model that takes a literacy score as input and outputs a predicted attention map, operationalizing the claim that literacy shapes gaze. Sal2Lit is its inverse: it consumes one attention map and returns a predicted literacy level, which is how the paper obtains its 86% accuracy figure. The empirical foundation is the measured gaze contrast between high and low scorers on the three visualization tests.

What would settle it

If high-literacy readers on a held-out set of visualizations, outside the three tests, show diffuse rather than focused attention, or if Sal2Lit's accuracy drops to near chance when tested on a new chart type, the claimed two-way link would fail to generalize.

Watch

Extended reading notes

Core claim

The central discovery is that attention patterns during chart reading encode visualization literacy reliably enough for two-way prediction. On three standard tests, experts' gaze is concentrated while novices scan more broadly, and this contrast serves as the signal: Lit2Sal generates the attention map a reader of a given literacy level would produce, and Sal2Lit inverts that mapping to estimate literacy from gaze. The paper reports that Sal2Lit achieves 86% accuracy from a single attention map and that the literacy-aware saliency model Lit2Sal outperforms existing saliency models that ignore literacy.

Load-bearing premise

The load-bearing premise is that the gaze patterns recorded in the lab with these three tests and 235 participants remain stable enough to predict literacy and attention outside the study.

Editorial extensions

If this is right

Literacy screening could shrink from a full test battery to a single short gaze recording, since one attention map predicts literacy with 86% accuracy in under a minute.
Visualization tools could estimate a reader's literacy on the fly and adjust chart complexity or annotation accordingly.
The expert-focus, novice-explore pattern suggests a concrete instructional target: training novices to allocate attention more selectively may raise literacy.
Saliency models that ignore literacy likely miss a systematic source of individual variation in where people look.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A testable extension the paper does not pursue is whether Sal2Lit transfers from the three study tests to real-world dashboards; if it does not, the 86% figure may describe task-specific gaze rather than a stable literacy trait.
The two-way link raises the possibility of using gaze as a real-time comprehension sensor, flagging when a viewer's attention wanders from the data.
If attention patterns are partly trainable, then gaze training could become an intervention to build visualization literacy, not just a way to measure it; this is a speculation beyond the paper's claims.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Colleague, quick take on arXiv:2508.03713. The review packet is broken: the attached full text is a completely different manuscript about tax reform optimization, not the visualization-literacy and eye-tracking study described in the abstract. That means I can only assess the abstract, and every substantive claim—the 235-participant study, the expert-focus/novice-exploration pattern, Lit2Sal outperforming SOTA, Sal2Lit's 86% accuracy—is unsupported by the materials actually supplied. This is not a generalizability dispute; it is a basic verification failure. The reader's worry about transfer to real visualization contexts is legitimate but premature.

That said, what is on the page is worth taking seriously. The abstract proposes something genuinely new: two models in the reverse direction, Lit2Sal (literacy-conditioned saliency) and Sal2Lit (attention-to-literacy prediction), and claims a practical payoff—a sub-minute literacy screen from a single attention map. If the real paper delivers those models with honest evaluation, it is a solid contribution to personalized visualization and could be useful to the HCI/vis community. The choice of three standard literacy tests (mini-VLAT, CALVI, SGL) is sensible, and the 86% figure, if accompanied by proper cross-validation and confidence intervals, would be meaningful.

The soft spots are about evidence, not speculation. No methods, no participant demographics, no eye-tracking protocol, no model architecture, no train/test split, no baseline table, no error bars. The abstract alone cannot distinguish a well-fit prediction on a held-out set from optimistic in-sample accuracy. And because the supplied full text is not this paper, I cannot check whether the authors cite prior literacy-aware saliency work properly or whether their 'overlooked' claim is fair. Those questions need the actual manuscript.

Recommendation: if this is a metadata mix-up and the correct version of the paper is available, it absolutely deserves a serious peer review—send it to referees. But on this package, the right editorial move is to desk-reject or, better, ask for the correct submission. I wouldn't cite it or bring it to reading group until the real text is in hand.

Referee Report

2 major / 2 minor

Summary. The manuscript under review, arXiv:2508.03713, presents an abstract claiming a 235-participant user study on visualization literacy and eye tracking, and proposes two computational models (Lit2Sal and Sal2Lit) with a reported 86% accuracy for predicting visual literacy from a single attention map. However, the full text supplied for this arXiv ID is an entirely different manuscript on tax reform as a constrained optimization problem, with no connection to visualization, eye tracking, literacy, or the named tests. The review package therefore contains none of the methods, experimental protocol, results, or analyses that would substantiate the abstract's claims.

Significance. If the abstract's claims were supported by an actual study, the work would be significant for personalized visualization design: predicting literacy from a single attention map could yield a fast assessment tool, and literacy-conditioned saliency could adapt visualizations to individual users. The proposed direction of accounting for individual differences in saliency models is well motivated. However, because the submission contains no supporting study, no model details, and no evaluation results, the significance cannot be assessed from this document, and the central claims remain entirely unverified. No credit can be given for reproducible code or machine-checked proofs, as none appear in the submitted text for the claimed topic.

major comments (2)

[Full Text (entire)] The full text supplied for arXiv:2508.03713 is the manuscript "Tax reform as a constrained optimization problem: a piecewise-linear framework and software implementation" by Verhagen, Schellekens, and Garstka. It contains no mention of visualization literacy, eye tracking, saliency models, mini-VLAT, CALVI, SGL, the 235-participant study, or the Lit2Sal/Sal2Lit models. The central claims of the abstract are therefore entirely unsupported by the submitted manuscript. There is no methods section, no experimental protocol, no data analysis, no model architecture, and no evaluation results from which the reported 86% accuracy could be checked. This is a load-bearing failure: the review package does not contain the study it purports to report.
[Abstract] Even taken on its own, the abstract's headline quantitative claim is insufficiently specified: "86% accuracy" is reported without any definition of the classification task (e.g., binary vs. multi-class literacy levels), the chance baseline, cross-validation scheme, or confidence interval. Similarly, the claim that Lit2Sal "outperforms state-of-the-art saliency models" is made without naming baselines or reporting effect sizes. These omissions would need to be addressed in any complete submission; they are noted here because the full text fails to provide them.

minor comments (2)

[Abstract] The phrase "two-way prediction" in the title is evocative but not defined; context suggests it refers to predicting attention from literacy (Lit2Sal) and literacy from attention (Sal2Lit), which could be stated explicitly.
[Abstract] The claim that Sal2Lit "only takes less than a minute" should be accompanied by the measurement procedure (e.g., one trial duration, number of stimuli) to be interpretable.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity identifiable: the supplied full text is an unrelated tax-reform paper, so the claimed derivation chain for the visualization-literacy models is absent rather than circular.

full rationale

The claimed paper (arXiv:2508.03713) proposes Lit2Sal and Sal2Lit and reports 86% accuracy for literacy prediction from a single attention map, based on a 235-participant study. The supplied full text, however, is a completely different manuscript on tax reform as a constrained optimization problem; it contains no eye-tracking data, no saliency model, no literacy tests, and no training/evaluation procedure. Consequently there is no derivation chain to audit: no equation from the claimed study appears, so no prediction can be shown, by the paper's own equations, to be equivalent to its inputs. The missing evidence is a verification failure, not a circularity. Per the hard rules, circularity must be demonstrated by quoting a specific reduction (e.g., fitted parameter renamed as prediction); no such reduction is available in the supplied material. The review package also does not contain the external benchmarks or held-out validation that would be needed to assess the 86% claim, so the honest finding is 'not assessable' rather than 'circular'.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The ledger is based solely on the abstract. No free parameters, axioms specific to a derivation, or invented entities are reported in the abstract. The domain assumptions listed are the load-bearing premises for the empirical claim. The full text supplied for this arXiv ID is a tax-reform paper, so nothing about the actual models could be audited.

free parameters (1)

Model parameters of Lit2Sal and Sal2Lit = not reported in abstract
The abstract does not disclose the fitted parameters of either model. The ledger cannot be completed without the full text.

assumptions (3)

domain assumption Visual attention patterns differ systematically between high and low visualization literacy groups.
This is the central premise of the abstract; without such a difference, neither Lit2Sal nor Sal2Lit has signal to learn.
domain assumption The three visualization tests (mini-VLAT, CALVI, SGL) measure the same underlying literacy construct.
The models pool attention data across tests; if the tests measure different abilities, the reported accuracy may not generalize across tests.
domain assumption The 235-participant sample is representative enough to support the proposed models.
The abstract gives no demographic or sampling details, yet the predictive claims depend on the sample reflecting the broader population of visualization users.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Tell Me Without Telling Me: Two-Way Prediction of Visualization Literacy and Visual Attention." pith.science (2026). https://pith.science/paper/G5ZCVSXG

@misc{pith2026250803713,
  author       = {Pith},
  title        = {Pith review of: Tell Me Without Telling Me: Two-Way Prediction of Visualization Literacy and Visual Attention},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G5ZCVSXG}},
  note         = {Machine review of arXiv:2508.03713}
}

read the original abstract

Accounting for individual differences can improve the effectiveness of visualization design. While the role of visual attention in visualization interpretation is well recognized, existing work often overlooks how this behavior varies based on visual literacy levels. Based on data from a 235-participant user study covering three visualization tests (mini-VLAT, CALVI, and SGL), we show that distinct attention patterns in visual data exploration can correlate with participants' literacy levels: While experts (high-scorers) generally show a strong attentional focus, novices (low-scorers) focus less and explore more. We then propose two computational models leveraging these insights: Lit2Sal -- a novel visual saliency model that predicts observer attention given their visualization literacy level, and Sal2Lit -- a model to predict visual literacy from human visual attention data. Our quantitative and qualitative evaluation demonstrates that Lit2Sal outperforms state-of-the-art saliency models with literacy-aware considerations. Sal2Lit predicts literacy with 86% accuracy using a single attention map, providing a time-efficient supplement to literacy assessment that only takes less than a minute. Taken together, our unique approach to consider individual differences in salience models and visual attention in literacy assessments paves the way for new directions in personalized visual data communication to enhance understanding.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

implementation gap

Introduction In developed countries income taxation is a core pillar of government revenue and a central lever on economic behavior and equity (OECD, 2023; Mirrlees et al., 2011; Tummers, 2019; Slemrod and Gillitzer, 2013). A large theoretical literature, culminating in the Mirrlees review, prescribes how rates should be set to balance efficiency and redi...

work page 2023
[2]

implementation gap

Related work Our work draws on, and contributes to, two literatures: the economics of tax policy, which tells us what tax schedulesshouldlook like in theory, and the operations-research methodology that provides the tools to optimize the schedule a statute actually encodes. We position the paper against each in turn. 2.1. Tax policy background A large bod...

work page 1971

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

implementation gap

Introduction In developed countries income taxation is a core pillar of government revenue and a central lever on economic behavior and equity (OECD, 2023; Mirrlees et al., 2011; Tummers, 2019; Slemrod and Gillitzer, 2013). A large theoretical literature, culminating in the Mirrlees review, prescribes how rates should be set to balance efficiency and redi...

work page 2023

[2] [2]

implementation gap

Related work Our work draws on, and contributes to, two literatures: the economics of tax policy, which tells us what tax schedulesshouldlook like in theory, and the operations-research methodology that provides the tools to optimize the schedule a statute actually encodes. We position the paper against each in turn. 2.1. Tax policy background A large bod...

work page 1971