Visualisation as communication: literacy, journalism and the turn to decisions

A visualisation built to help an analyst search a dataset is judged by a different test from one built to tell a reader what the data shows. The ninth unit treated the analyst's tool as judged on whether the analysis finds something true and survives checking. The public chart is judged on something harder to measure: whether a reader who is not an analyst comes away believing what the data supports. This last of ten units, after the ninth weighed the evidence, follows visualisation from the analyst's screen to the news page, the clinic and the public consultation. It argues that the evidence followed the move. The claim is not that charts do not work. Once the outcome changed from perceptual accuracy to belief and decision, the effects the literature can show became smaller, more conditional and more dependent on the reader, than the earlier optimism assumed.

The audience changed, and with it the question

Franconeri and colleagues (2021) synthesise the evidence on data displays for a general audience and state the reframing directly. Their review holds that the visual system extracts broad statistics from a display quickly, and that comparing between subsets of values is slow. It holds that effective graphics avoid a tax on working memory, guide attention and respect familiar conventions. The closing condition matters. Visualisations teach and communicate, they argue, only when designers tailor them to their audience. That moves the audience from a constant to a variable, and its skill and prior belief enter the design problem rather than sitting outside it.

Diakopoulos (2019) examined newsrooms that use algorithms for data gathering, story generation and distribution, including automated drafts of routine stories, agents that push stories to social audiences, and testing of how content performs. His argument is that the choices built into those systems are editorial choices, so the framing a reader meets can be set in software.

What the public takes from a chart

Sanchez-Holgado and colleagues (2023) give the recent experimental version. They ran a two-condition study with seven hundred participants. It compared a text news item against the same item carrying a data visualisation, on scientific and health topics. A moderated mediation analysis found that the visualised item increased comprehension, and that better comprehension produced more positive attitudes to data visualisation. Interest and comprehension carried the effect, and the audience also includes a reader on a metered phone connection, where the chart has to arrive and stay legible rather than sit on a desk. The result is narrower than it looks. The outcome is comprehension of the item and attitude towards the form, not a decision taken afterwards.

Graph literacy is a measured variable, not an assumption

The reader is not a single kind of person. Galesic and Garcia-Retamero (2011) validated a short graph-literacy scale on representative samples of 495 people in Germany and 492 in the United States, who completed about nine of thirteen items, with roughly a third combining low graph literacy with low numeracy. Okan and colleagues (2011) tested what that difference does: in an experiment on denominator neglect, icon arrays improved risk comprehension and confidence more often for high-literacy readers than for low, whichever way the risk was framed. Lee and colleagues (2017) developed a fifty-three-item test over twelve visualisations and eight tasks, tried out on 191 test takers, with a reliability of 0.76. A chart's effect is moderated by a skill that varies widely within one audience and can be measured.

That instrument has since been questioned. Brockbank and colleagues (2025) gave two widely used graph-comprehension assessments to 1,113 participants and found the overall scores correlated while the error patterns showed the tests probing distinct components. Graph literacy is less of a single number than a single instrument implies.

The designer cannot see the reader's problem

Two results explain why professionally made charts still under-communicate. Xiong and colleagues (2020) showed the curse of knowledge inside visual perception. Participants were primed with one backstory for polling data, and believed naive viewers would find the same pattern most salient, despite being told to ignore their own knowledge. If a designer's reading of a chart is invisible to the designer as a choice, the chart is annotated for a reader who already knows. Bylinskii and colleagues (2017) show the useful half. Models trained on human clicks and importance annotations predicted which elements of a design or visualisation matter, and beat saliency models when validated with crowdworkers.

Where the health-communication literature got to first

The strongest evidence about charts changing what people do comes from outside visualisation. Garcia-Retamero and Cokely (2017) systematically reviewed thirty-six publications covering 27,885 participants across sixty countries. They concluded that transparent visual aids robustly improved risk understanding, particularly for people with lower numeracy. The primary behind that pooled result measured behaviour rather than understanding: Garcia-Retamero and Cokely (2011) found visual aids making both message framings equally effective at promoting reported prevention and detection behaviour in 744 young adults. Lipkus (2007) is the caution inside the same literature. He could offer only limited best practice, because few randomised trials compared formats on the same outcomes and the theory behind format choice was thin.

Is the field's own advice evidence-based?

Kosara (2016) makes the case that the field was built from the wrong end. Displays were drawn before the perceptual principles behind them were established, and several widely taught rules rest on thin or misread evidence.

Visualization was created from the other end: by building visual displays without an idea of how or if they worked, and then finding the relevant perceptual and other basics here and there.

Hullman and colleagues (2011) add the counter-case. They argue from cross-disciplinary evidence that efficiency-based design is incomplete, because a controlled visual difficulty can promote active processing and improve comprehension and recall. That makes design a trade-off rather than a maximisation of ease. Haroz and colleagues (2015) tested pictographs against plain bar charts across five experiments. They found that pictographs encoding the data carried no reading cost, and mildly aided long-term memory and engagement, while superfluous images hurt short-term retention and slowed search. Skau and Kosara (2016) found the central angle the least important cue in pie and donut charts, with donuts as accurate as pies.

Contested question: does a chart change what a reader believes or decides?

The first position holds that the effects are established and usable, especially for the readers who need them most. Garcia-Retamero and Cokely (2017) find transparent visual aids robustly improve risk understanding and work best for those with lower numeracy, Okan and colleagues (2011) show a measured graph-literacy difference changing how much an aid helps, and Franconeri and colleagues (2021) treat the communication case as settled enough to teach as guidelines. The second holds that the effects are smaller, more conditional and measured on the wrong outcome. Lipkus (2007) reports that few cross-format comparisons held outcomes constant, Hullman and colleagues (2011) hold that a display's effect depends on the processing it triggers, and Xiong and colleagues (2020) show the designer reliably mispredicts how a reader will see the display. What would settle it is not another comprehension study but a randomised trial in the decision context itself, with the decision as the outcome, graph literacy as a measured moderator, and the design fixed before the result is known. What the field has instead is comprehension and attitude as proxies, the gap Lipkus already named.

The two positions rest on different evidence. The communication position rests on a systematic review of thirty-six studies and 27,885 participants (Garcia-Retamero and Cokely, 2017) and a graph-literacy comparison (Okan et al., 2011). It also rests on a two-condition news experiment (Sanchez-Holgado et al., 2023). The conditional position rests on a review that found few controlled comparisons (Lipkus, 2007) and experiments where form effects were small (Haroz et al., 2015; Skau and Kosara, 2016). It also rests on the curse-of-knowledge study (Xiong et al., 2020). While the dispute is open, measure the decision rather than the comprehension of it, and report the effect by graph literacy.

The strongest counter-argument

The strongest objection is that the thesis discounts a large, pooled body of decision-relevant evidence. Garcia-Retamero and Cokely (2017) cover thirty-six studies, 27,885 participants and sixty countries, and find transparent visual aids improve risk understanding. That is not a small or single-study result. The reply narrows the claim: improvement in risk understanding is not improvement in the decision, as Lipkus (2007) insists, and Okan and colleagues (2011) show the same aid helps the already graph-literate most, so an average can conceal a distribution in which the least literate gain least.

Where the area is heading

The direction is towards decision-level and distribution-aware evaluation of public visualisation. Franconeri and colleagues (2021) name the reframing, treating communication and audience as the design problem, and Garcia-Retamero and Cokely (2017) supply the pooled evidence base a decision-level standard would have to beat. Lee and colleagues (2017) and Boy and colleagues (2014) make literacy a measurable moderator, and Bylinskii and colleagues (2017) make attention predictable, so a display's framing can be tested against measured attention. The study the area is short of takes a display already shown to raise comprehension and runs it in the decision it is meant to change, with graph literacy measured and reported.

The sources this unit rests on

Each note states what the cited document did and found, from the resolved record and abstract.

• Franconeri, S. L., Padilla, L. M., Shah, P., Zacks, J. M. and Hullman, J. (2021). The Science of Visual Data Communication: What Works. doi:10.1177/15291006211051956 Synthesised the evidence on displays for a general audience, covering encodings, attention and literacy, and argued that visualisations communicate only when tailored to their audience.
• Sanchez-Holgado, P., Arcila-Calderon, C. and Frias-Vazquez, M. (2023). The effect of interest and attitude on public comprehension of news with data visualization. doi:10.3389/fcomm.2023.1064184 Ran a two-condition experiment with 700 participants on a news item with and without a data visualisation, finding it raised comprehension and, through it, more positive attitudes.
• Galesic, M. and Garcia-Retamero, R. (2011). Graph literacy: a cross-cultural comparison. doi:10.1177/0272989x10373805 Validated a short graph-literacy scale on representative samples of 495 Germans and 492 Americans, who averaged about nine of thirteen items, with roughly a third of each sample low in both graph literacy and numeracy.
• Okan, Y., Garcia-Retamero, R., Cokely, E. T. and Maldonado, A. (2011). Individual Differences in Graph Literacy: Overcoming Denominator Neglect in Risk Comprehension. doi:10.1002/bdm.751 Tested icon arrays against a common bias and found they improved comprehension and confidence more for high-literacy readers, whichever way the risk was framed, so the aid helps most those who need it least.
• Boy, J., Rensink, R. A., Bertini, E. and Fekete, J.-D. (2014). A Principled Way of Assessing Visualization Literacy. doi:10.1109/tvcg.2014.2346984 Used item response theory to build tests of visualisation literacy, first for line graphs and then for bar charts and scatterplots. It makes the audience's skill testable before a study runs.
• Lee, S., Kim, S.-H. and Kwon, B. C. (2017). VLAT: Development of a Visualization Literacy Assessment Test. doi:10.1109/tvcg.2016.2598920 Developed a fifty-three-item test over twelve visualisations and eight tasks, tried out on 191 test takers, with a reliability of 0.76 and a correlation of 0.64 with learning an unfamiliar visualisation.
• Xiong, C., van Weelden, L. and Franconeri, S. (2020). The Curse of Knowledge in Visual Data Communication. doi:10.1109/tvcg.2019.2917689 Primed participants with a backstory for polling data and found they predicted naive viewers would find the same pattern salient, despite instructions to ignore their own knowledge.
• Bylinskii, Z., Kim, N. W., O'Donovan, P., Alsheikh, S., Madan, S., Pfister, H. and others (2017). Learning Visual Importance for Graphic Designs and Data Visualizations. doi:10.1145/3126594.3126653 Trained models on human clicks and importance annotations to predict which elements of a design or visualisation matter, validated with crowdworkers.
• Diakopoulos, N. (2019). Automating the News: How Algorithms Are Rewriting the Media. doi:10.4159/9780674239302 Examined how news production uses algorithms for data gathering, drafting, distribution and testing, and argued that the choices built in are editorial choices that should be accountable.
• Garcia-Retamero, R. and Cokely, E. T. (2017). Designing Visual Aids That Promote Risk Literacy: A Systematic Review of Health Research and Evidence-Based Design Heuristics. doi:10.1177/0018720817690634 Systematically reviewed thirty-six publications covering 27,885 participants and found transparent visual aids improved risk understanding, especially for lower-numeracy readers.
• Garcia-Retamero, R. and Cokely, E. T. (2011). Effective communication of risks to young adults. doi:10.1037/a0023677 With 744 young adults in a two-phase longitudinal experiment, visual aids made gain- and loss-framed messages equally effective at promoting reported condom use and screening. It is the primary behaviour result behind the pooled review.
• Lipkus, I. M. (2007). Numeric, Verbal, and Visual Formats of Conveying Health Risks: Suggested Best Practices and Future Recommendations. doi:10.1177/0272989x07307271 Reviewed formats for conveying health risks and offered only limited best practice, because few randomised trials compared formats on the same outcomes.
• Kosara, R. (2016). An Empire Built On Sand: Reexamining What We Think We Know About Visualization. doi:10.1145/2993901.2993909 Argued that the field drew displays before establishing the perceptual principles behind them, so several widely taught rules rest on thin or misread evidence.
• Hullman, J., Adar, E. and Shah, P. (2011). Benefitting InfoVis with Visual Difficulties. doi:10.1109/tvcg.2011.175 Synthesised cross-disciplinary evidence to argue that efficiency-based design is incomplete and that a controlled visual difficulty can improve comprehension and recall, so ease is not monotonic in effectiveness.
• Haroz, S., Kosara, R. and Franconeri, S. L. (2015). ISOTYPE Visualization: Working Memory, Performance, and Engagement with Pictographs. doi:10.1145/2702123.2702275 Across five experiments found pictographs encoding the data cost no reading speed and mildly aided long-term memory and engagement, while superfluous images hurt short-term retention and slowed search.
• Skau, D. and Kosara, R. (2016). Arcs, Angles, or Areas: Individual Data Encodings in Pie and Donut Charts. doi:10.1111/cgf.12888 Varied pie and donut charts to test which encoding readers use and found the central angle least important, with donuts as accurate as pies.
• Brockbank, E., Verma, A., Lloyd, H., Huey, H., Padilla, L. and Fan, J. E. (2025). Evaluating convergence between two data visualization literacy assessments. doi:10.1186/s41235-025-00622-9 Administered two widely used graph-comprehension assessments to 1,113 participants and found the overall scores correlated while individual error patterns showed the tests probing somewhat distinct components. It qualifies the unit's treatment of graph literacy as one measured variable.