pith. sign in

arxiv: 2605.19017 · v1 · pith:6JJJPUSPnew · submitted 2026-05-18 · 💻 cs.HC

Guardrail Selection in Line Charts to Contextualize Persuasive Visualizations

Khandaker Abrar Nadib , Marina Kogan , Alexander Lex , Maxim Lisnic This is my paper

Pith reviewed 2026-05-20 08:20 UTC · model grok-4.3

classification 💻 cs.HC
keywords guardrailspersuasive visualizationsline chartscontextual comparison linesmisleading chartsuser trustvisualization evaluationtime-series data
0
0 comments X

The pith

Guardrails as contextual lines in line charts raise trust, judgment accuracy, and sense of completeness in persuasive visualizations.

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

The paper tests a set of practical sampling strategies for adding contextual comparison lines, called guardrails, directly into line charts. In a study using real COVID-19 and stock-market scenarios, charts with these guardrails produced higher trust, more accurate estimates of data rank, and stronger perceptions of contextual completeness than identical charts without them. The work matters because persuasive charts on social media often cherry-pick data and arrive with misleading captions, while later fact-checking arrives too late to change minds. By embedding context inside the chart itself, the approach aims to reduce framing effects at the moment of viewing.

Core claim

Across two real-world scenarios, guardrails improved trust, accuracy of performance judgments, and perceived completeness of context compared to the control. The study proposes and evaluates practical guardrail sampling strategies for implementing such contextual lines in real systems and reports participants' subjective preferences for different tasks.

What carries the argument

Guardrail sampling strategies that embed contextual comparison lines directly into line charts to supply missing reference points.

If this is right

  • Guardrails increase user trust in the visualized data relative to unguarded charts.
  • They raise the accuracy with which viewers estimate a data point's rank within the full set.
  • Viewers report higher perceived completeness of context when guardrails are present.
  • Task-specific preferences emerge for which guardrail sampling strategy feels most useful.

Where Pith is reading between the lines

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

  • Interactive data explorers for contested topics could adopt these sampling methods to limit one-sided framing at the source.
  • Designers might experiment with dynamic guardrail selection that adjusts based on the user's current zoom level or highlighted region.
  • The approach could be combined with caption-checking tools to create layered defenses against misleading social-media graphics.

Load-bearing premise

The two tested scenarios and the specific sampling strategies chosen are representative enough that the observed benefits will appear in other persuasive visualization contexts and datasets.

What would settle it

A replication using different datasets or a new set of sampling strategies that finds no measurable gain in trust or rank-estimation accuracy relative to the no-guardrail control.

Figures

Figures reproduced from arXiv: 2605.19017 by Alexander Lex, Khandaker Abrar Nadib, Marina Kogan, Maxim Lisnic.

Figure 1
Figure 1. Figure 1: The five guardrails and the control condition evaluated in our experiment. We test different versions of statistical guardrails: (1) percentile markers, i.e., explicitly showing percentiles of the whole dataset over time; (2) percentile-based exemplars, i.e., showing concrete items that are close to a percentile value, (3) cluster representatives, i.e., showing items central to the clusters of the dataset,… view at source ↗
Figure 2
Figure 2. Figure 2: Examples of our guardrails as seen in the Task 1 stimulus. Figure (a) illustrates the Exemplars with Semantic Similarity strategy, showing items semantically similar to the focal entity (here, countries demographically and geographically similar to Belarus). Figure (b) illustrates the Percentile-based Exemplars strategy, which instead selects context based on their percentile bins in the global distributio… view at source ↗
Figure 3
Figure 3. Figure 3: Across all tasks and scenarios, guardrails improve accuracy, trust, and context. For trust and context, Random Exemplars performs as well as data-driven guardrails, and Exemplars with Semantic Similarity performs better in the COVID scenario. Data-driven guardrails— Percentile-based Exemplars, Percentile Markers, and Cluster Representatives— however, offer a significant improvement in accuracy. Exem￾plars … view at source ↗
Figure 4
Figure 4. Figure 4: When asked to choose, participants prefer Exemplars with Semantic Similarity and Percentile-based Exemplars at roughly the same rate, independent of framing, even though we considered Exemplars with Semantic Similarity the better choice for the Holistic framing (the question targeted judgment relative to peers), and percentiles the better choice for Precise framing (the question targeted estimating absolut… view at source ↗
Figure 5
Figure 5. Figure 5: Final codebook used for open-ended responses. We organize the themes based on the highest level of codes. Each text response may be described by one or more codes. We excluded codes that indicated no preference or a clearly mistaken justification. 6.2. Theme 2: Preference for Chart Aesthetics Moreover, the actual contextual content was not the only reason for participants’ preferences. Many of the response… view at source ↗
read the original abstract

Charts used for persuasion can easily veer into being outright misleading when, for instance, cherry-picked data is paired with a deceptive caption, as is commonly encountered on social media. The rise of interactive time-series data explorers for hotly debated topics makes such framing easy to produce and spread. Post-hoc interventions like fact-checking often arrive too late and suffer from persistence of belief. Prior work suggests that guardrails, in the form of contextual comparison lines embedded directly into charts, can reduce these effects. We propose and evaluate a practical set of guardrail sampling strategies for implementing such contextual lines in real systems. In a preregistered mixed-design study with two real-world scenarios (COVID-19 and Stocks), participants viewed persuasive charts with different sets of guardrails and reported trust, estimated rank in the dataset, expressed their perceived completeness of context, as well as subjective preference for different tasks. Across scenarios, guardrails improved trust, accuracy of performance judgments, and perceived completeness of context compared to the control. Taken together, the study offers practical guardrail sampling methods, evidence of their contextual benefits, and insights into participants' preferences.

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

2 major / 2 minor

Summary. The manuscript proposes practical guardrail sampling strategies for embedding contextual comparison lines directly into line charts to reduce misleading effects in persuasive visualizations, such as those with cherry-picked data or deceptive captions. It evaluates these strategies in a preregistered mixed-design user study with two real-world scenarios (COVID-19 and Stocks), where participants viewed charts with varying guardrails and reported on trust, accuracy of performance judgments (e.g., estimated rank), perceived completeness of context, and subjective preferences. Results indicate that guardrails improved trust, judgment accuracy, and perceived context completeness relative to the control condition across scenarios.

Significance. If the benefits hold beyond the tested cases, the work supplies actionable sampling methods for guardrails in interactive time-series explorers, backed by direct evidence from a preregistered mixed-design study using real scenarios rather than synthetic data. This addresses post-hoc fact-checking limitations by providing in-situ contextualization and could inform design of systems handling hotly debated topics on social media.

major comments (2)
  1. [Study Design and Results] The central claim of improved trust, judgment accuracy, and perceived completeness rests on the two scenarios (COVID-19 and Stocks) and the specific guardrail sampling strategies being representative. The manuscript reports improvements relative to control but provides no additional datasets, cross-domain tests, or sensitivity analysis to other data distributions or persuasive framing types (see Study Design and Results sections).
  2. [Abstract] The abstract claims improvements across scenarios but supplies no details on sample size, exact statistical tests, effect sizes, or exclusion criteria, leaving the empirical support for the central claim only partially verifiable from the available text.
minor comments (2)
  1. [Methods] Clarify the exact guardrail sampling strategies (e.g., how lines are selected or bounded) with pseudocode or additional figures to aid reproducibility.
  2. [Figures] Ensure all figures include clear legends and axis labels for the guardrail lines versus the main data series.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments below and have revised the manuscript to improve transparency and acknowledge scope limitations.

read point-by-point responses

  1. Referee: [Study Design and Results] The central claim of improved trust, judgment accuracy, and perceived completeness rests on the two scenarios (COVID-19 and Stocks) and the specific guardrail sampling strategies being representative. The manuscript reports improvements relative to control but provides no additional datasets, cross-domain tests, or sensitivity analysis to other data distributions or persuasive framing types (see Study Design and Results sections).

    Authors: The COVID-19 and Stocks scenarios were selected as representative real-world cases involving time-series data commonly framed persuasively on social media, spanning public health and financial domains. The preregistered mixed-design study provides direct evidence of benefits relative to control in these contexts. We agree that additional datasets, broader cross-domain tests, and formal sensitivity analyses would strengthen generalizability claims. In the revision we have added a limitations subsection in the Discussion that explicitly discusses the representativeness of the chosen scenarios and data distributions, includes a qualitative sensitivity reflection based on the observed variance in the two datasets, and outlines targeted future work for validation on additional domains and framing types. revision: partial

  2. Referee: [Abstract] The abstract claims improvements across scenarios but supplies no details on sample size, exact statistical tests, effect sizes, or exclusion criteria, leaving the empirical support for the central claim only partially verifiable from the available text.

    Authors: We have revised the abstract to include the requested details: total sample size after exclusions, the primary statistical tests (mixed-design ANOVA and follow-up pairwise comparisons), reported effect sizes, and the exclusion criteria (failed attention checks and incomplete responses). These elements were already reported in the Methods and Results sections; the abstract now summarizes them concisely to make the empirical support more immediately verifiable. revision: yes

Circularity Check

0 steps flagged

No significant circularity in empirical user study

full rationale

The paper reports results from a preregistered mixed-design user study with participant responses measuring trust, judgment accuracy, and perceived context completeness across two scenarios. No derivation chain, equations, fitted parameters, or first-principles results exist that could reduce outcomes to inputs by construction. Claims rest on direct empirical data collection rather than self-definitional steps, self-citation load-bearing premises, or renamed known results. The work is self-contained against external participant benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper introduces no new mathematical constructs or fitted parameters; it rests on standard HCI assumptions about the validity of self-reported trust and judgment measures in controlled settings.

axioms (1)
  • domain assumption Self-reported trust, rank estimates, and perceived completeness from study participants reflect real-world responses to persuasive charts.
    Invoked implicitly when interpreting questionnaire results as evidence of reduced misleading effects.

pith-pipeline@v0.9.0 · 5735 in / 1276 out tokens · 35294 ms · 2026-05-20T08:20:56.078727+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

28 extracted references · 28 canonical work pages

  1. [1]

    and Parker, Douglass S

    Chih, Christine H. and Parker, Douglass S. , year = 2008, month = aug, pages =. The Persuasive Phase of Visualization , booktitle =. doi:10.1145/1401890.1401996 , urldate =

    work page doi:10.1145/1401890.1401996 2008

  2. [2]

    doi:10.48550/arXiv.2508.03876 , urldate =

    Cutler, Zach and Wilburn, Jack and Shrestha, Hilson and Ding, Yiren and Bollen, Brian and Nadib, Khandaker Abrar and He, Tingying and McNutt, Andrew and Harrison, Lane and Lex, Alexander , year = 2026, journal =. doi:10.48550/arXiv.2508.03876 , urldate =. arXiv , keywords =:2508.03876 , primaryclass =

    work page doi:10.48550/arxiv.2508.03876 2026

  3. [3]

    doi:10.48550/arXiv.2508.04679 , urldate =

    Das, Amit Kumar and Mueller, Klaus , year = 2025, month = aug, number =. doi:10.48550/arXiv.2508.04679 , urldate =. arXiv , keywords =:2508.04679 , primaryclass =

    work page doi:10.48550/arxiv.2508.04679 2025

  4. [4]

    Annotating

    Fan, Arlen and Ma, Yuxin and Mancenido, Michelle and Maciejewski, Ross , year = 2022, month = apr, series =. Annotating. Proceedings of the 2022. doi:10.1145/3491102.3502138 , urldate =

    work page doi:10.1145/3491102.3502138 2022

  5. [5]

    Hopkins, M

    Hopkins, Aspen K. and Correll, Michael and Satyanarayan, Arvind , year = 2020, journal =. doi:10.1111/cgf.13975 , urldate =

    work page doi:10.1111/cgf.13975 2020

  6. [6]

    Kim, Dae Hyun and Setlur, Vidya and Agrawala, Maneesh , year = 2021, month = may, series =. Towards. Proceedings of the 2021. doi:10.1145/3411764.3445443 , urldate =

    work page doi:10.1145/3411764.3445443 2021

  7. [7]

    Frames and

    Kong, Ha-Kyung and Liu, Zhicheng and Karahalios, Karrie , year = 2018, pages =. Frames and. Proceedings of the 2018. doi:10.1145/3173574.3174012 , urldate =

    work page doi:10.1145/3173574.3174012 2018

  8. [8]

    and Satyanarayan, Arvind , year = 2021, series =

    Lee, Crystal and Yang, Tanya and Inchoco, Gabrielle D and Jones, Graham M. and Satyanarayan, Arvind , year = 2021, series =. Viral. Proceedings of the 2021

    work page 2021

  9. [9]

    Misinformation and Its Correction:

    Lewandowsky, Stephan and Ecker, Ullrich KH and Seifert, Colleen M and Schwarz, Norbert and Cook, John , year = 2012, journal =. Misinformation and Its Correction:

    work page 2012

  10. [10]

    Ericson, David Weintrop, and Tovi Grossman

    Lisnic, Maxim and Polychronis, Cole and Lex, Alexander and Kogan, Marina , year = 2023, pages =. Misleading. Proceedings of the 2023. doi:10.1145/3544548.3580910 , urldate =

    work page doi:10.1145/3544548.3580910 2023

  11. [11]

    Visualization

    Lisnic, Maxim and Cutler, Zach and Kogan, Marina and Lex, Alexander , year = 2025, month = apr, pages =. Visualization. Proceedings of the 2025. doi:10.1145/3706598.3713385 , urldate =

    work page doi:10.1145/3706598.3713385 2025

  12. [12]

    Lisnic, Maxim and Lex, Alexander and Kogan, Marina , year = 2024, series =. `. doi:10.1145/3613904.3642448 , urldate =

    work page doi:10.1145/3613904.3642448 2024

  13. [13]

    Misinformed by

    Lo, Leo Yu-Ho and Gupta, Ayush and Shigyo, Kento and Wu, Aoyu and Bertini, Enrico and Qu, Huamin , year = 2022, journal =. Misinformed by. doi:10.1111/cgf.14559 , urldate =

    work page doi:10.1111/cgf.14559 2022

  14. [14]

    Linting for Visualization:

    McNutt, Andrew and Kindlmann, Gordon , year = 2018, volume =. Linting for Visualization:

    work page 2018

  15. [15]

    Zhang, N

    Moritz, Dominik and Wang, Chenglong and Nelson, Greg L. and Lin, Halden and Smith, Adam M. and Howe, Bill and Heer, Jeffrey , year = 2019, month = jan, journal =. Formalizing. doi:10.1109/TVCG.2018.2865240 , urldate =

    work page doi:10.1109/tvcg.2018.2865240 2019

  16. [16]

    Preregistration:

    Nadib, Khandaker Abrar and Lisnic, Maxim and Kogan, Marina and Lex, Alexander , year = 2025, doi =. Preregistration:

    work page 2025

  17. [17]

    and Bertini, Enrico , year = 2022, journal =

    Padilla, Lace and Fygenson, Racquel and Castro, Spencer C. and Bertini, Enrico , year = 2022, journal =. Multiple. doi:10.1109/TVCG.2022.3209457 , urldate =

    work page doi:10.1109/tvcg.2022.3209457 2022

  18. [18]

    Taher, J

    Pandey, Anshul Vikram and Rall, Katharina and Satterthwaite, Margaret L. and Nov, Oded and Bertini, Enrico , year = 2015, series =. How. Proceedings of the 33rd. doi:10.1145/2702123.2702608 , urldate =

    work page doi:10.1145/2702123.2702608 2015

  19. [19]

    Pandey, Anshul Vikram and Manivannan, Anjali and Nov, Oded and Satterthwaite, Margaret and Bertini, Enrico , year = 2014, month = dec, journal =. The. doi:10.1109/TVCG.2014.2346419 , urldate =

    work page doi:10.1109/tvcg.2014.2346419 2014

  20. [20]

    Ritchie, Jacob and Wigdor, Daniel and Chevalier, Fanny , year = 2019, month = may, pages =. A. Proceedings of the 2019. doi:10.1145/3290605.3300423 , urldate =

    work page doi:10.1145/3290605.3300423 2019

  21. [21]

    Stokes, V

    Stokes, Chase and Setlur, Vidya and Cogley, Bridget and Satyanarayan, Arvind and Hearst, Marti A. , year = 2023, month = jan, journal =. Striking a. doi:10.1109/TVCG.2022.3209383 , urldate =

    work page doi:10.1109/tvcg.2022.3209383 2023

  22. [22]

    Belief Echoes:

    Thorson, Emily , year = 2016, journal =. Belief Echoes:

    work page 2016

  23. [23]

    Science , author =

    The Spread of True and False News Online , author =. Science , volume =. doi:10.1126/science.aap9559 , urldate =

    work page doi:10.1126/science.aap9559

  24. [24]

    Investigating the Effect of the Multiple Comparisons Problem in Visual Analysis , booktitle =

    Zgraggen, Emanuel and Zhao, Zheguang and Zeleznik, Robert and Kraska, Tim , year = 2018, pages =. Investigating the Effect of the Multiple Comparisons Problem in Visual Analysis , booktitle =. doi:10.1145/3173574.3174053 , urldate =

    work page doi:10.1145/3173574.3174053 2018

  25. [25]

    Controlling

    Zhao, Zheguang and De Stefani, Lorenzo and Zgraggen, Emanuel and Binnig, Carsten and Upfal, Eli and Kraska, Tim , year = 2017, pages =. Controlling. Proc. doi:10.1145/3035918.3064019 , urldate =

    work page doi:10.1145/3035918.3064019 2017

  26. [26]

    When Do Data Visualizations Persuade?

    Markant, Douglas and Rogha, Milad and Karduni, Alireza and Wesslen, Ryan and Dou, Wenwen , year = 2023, month = apr, series =. When Do Data Visualizations Persuade?. Proceedings of the 2023. doi:10.1145/3544548.3581330 , urldate =

    work page doi:10.1145/3544548.3581330 2023

  27. [27]

    Hypothetical

    Hullman, Jessica and Resnick, Paul and Adar, Eytan , year = 2015, month = nov, journal =. Hypothetical. doi:10.1371/journal.pone.0142444 , urldate =

    work page doi:10.1371/journal.pone.0142444 2015

  28. [28]

    Task-Driven Evaluation of Aggregation in Time Series Visualization , booktitle =

    Albers, Danielle and Correll, Michael and Gleicher, Michael , year = 2014, month = apr, series =. Task-Driven Evaluation of Aggregation in Time Series Visualization , booktitle =. doi:10.1145/2556288.2557200 , urldate =

    work page doi:10.1145/2556288.2557200 2014