Shuichiro Ogawa
日本語

Notes · updated 2026-10-07

Where Does the Value of Showing Data as Graphics Lie? Kinds of Value and Why They Arise (from Industry Sources)

This note examines the value of showing data as charts and visual displays in themselves, as distinct from data analysis or data science in general.

Contents (11)
  1. Separating the value of graphics from the value of analysis
  2. Immediate understanding: comprehension and speed
  3. Reaching many people: communication and engagement
  4. Changing behavior: what people choose and how much they use
  5. Noticing abnormalities: safety and operations
  6. Continued use: product experience
  7. Maintaining or damaging trust
  8. When graphics create no value, or work against it
  9. Why do graphics work?
  10. Where this account does not apply
  11. Footnotes

Separating the value of graphics from the value of analysis

This note is about industry practice; academic research was used only as far as needed to explain the reasons1. The previous note, What Value Does Data Visualization Have in Industry? Market Valuation, Where Value Arises, and What Changes after Generative AI (from Industry Sources), centered on the market for BI and dashboards and mixed the value of analytical tools with the value of showing data graphically. This note narrows the question to what changes when the same data are shown as charts, colors, and shapes rather than as tables of numbers or text.

Answering that question requires sources that compare graphics with other displays. Public industry sources contain many figures on scale, such as views and shares, but almost none that compare graphics with non-graphic displays. The comparisons that could be found were experiments and surveys that governments and regulators ran to decide the design of displays, and academic research. For each kind of value below, comparisons come first where they exist, followed by figures on scale and the makers’ explanations.

Immediate understanding: comprehension and speed

In 2024, the US FDA compared candidate front-of-package nutrition labels in an online experiment with 9,200 participants (U.S. Food and Drug Administration 2024). Labels that added the interpretive words “Low,” “Med,” and “High” for saturated fat, sodium, and added sugars performed best overall in helping consumers identify healthier foods, while labels showing only amounts and percentages of daily values made it harder to identify nutrient levels correctly. The FDA notes that the literature is clear that interpretive schemes do best for conferring an understanding of nutrient content. However, the experiment compared labels with each other and did not include a no-label condition.

In 2006, the US FTC compared candidate EnergyGuide labels for appliances in a survey of about 4,000 people (Federal Trade Commission 2007). Of the respondents who saw a categorical label showing efficiency in steps, 82% ranked refrigerators correctly by energy efficiency, compared with 72% for the current label and 71% for a label showing operating cost. On questions about the operating cost or energy use of a single model, over 80% answered correctly with every label. Categorical displays can thus be read as strong for questions of “which is better,” and numerical displays for questions of “how much.” The survey also included a control condition that presented the information in tables and text without a label, but the difference from that condition could not be confirmed within the pages read.

In June 2022, the Japan Meteorological Agency revised the colors of Kikikuru, which shows the risk of heavy-rain disasters on a map, merging the two purples for alert level 4 into one and adding black for level 5 (Japan Meteorological Agency 2022). It aligned the Kikikuru colors with the colors of the evacuation alert levels and asks residents to decide to evacuate promptly when purple appears. In a 2024 report, the agency and the Ministry of Land, Infrastructure, Transport and Tourism wrote that the five alert levels had been introduced so that residents could intuitively understand the degree of danger and the action to take (Japan Meteorological Agency and Ministry of Land, Infrastructure, Transport and Tourism 2024). No figures measuring how much the color changes altered evacuation were found in these sources.

Makers also name speed of understanding as a value of graphics. Ed Hawkins, who created the warming stripes, wrote that each stripe represents one year’s temperature and that all other superfluous information is removed so that the changes in temperature are seen simply and undeniably (Hawkins 2018). Hawkins also said that our visual system interprets the stripes without our even thinking about it (NCAS 2018).

Reaching many people: communication and engagement

Graphics were read on a scale that text alone did not reach. According to the charting tool Datawrapper, embedded charts by the German news channel ntv had a record of about 50 million views a month before COVID-19, which climbed to 1.7 billion in March and April 2020 (Datawrapper 2020). John Burn-Murdoch of the Financial Times said that the page holding the paper’s trajectory charts had become, by a large margin, its most viewed page ever.

Graphics also spread beyond their makers. The animated “Flatten the Curve” graphic released in March 2020 by the New Zealand microbiologist Siouxsie Wiles and the cartoonist Toby Morris gained over 10 million impressions in three days, and the prime minister used it at a national press conference on March 14 (Wiles et al. 2023). The two released more than 70 graphics under a license that allowed adaptation and translation, and volunteers translated them into several languages. Morris said that showing someone doing something can be more effective than telling them about it.

Matt Daniels of The Pudding, which tells data stories with graphics, said that bringing in data and visuals lets the publication tell broader, bigger stories than prose-led pieces (Bertini and Stefaner n.d.). The Gapminder Foundation developed Trendalyzer, software that shows statistics as animated bubble charts, which Google acquired in 2007 (Gapminder n.d.).

Changing behavior: what people choose and how much they use

Graphics changed what people chose. A network meta-analysis of 156 studies of color-coded and warning nutrition labels found that traffic light labels were associated with an increased probability of selecting healthier products, with an odds ratio of 1.5 (Song et al. 2021). In the early UK smart meter rollout, households with smart meters and in-home displays used 2.3% less electricity and 1.5% less gas than households with traditional meters (Department of Energy and Climate Change 2015). The report confirms the value of a well-designed in-home display for increasing energy awareness and understanding and promoting reduced consumption, and of calibrating the ambient traffic-light feedback to each household’s range of consumption. In a 2026 consultation document, the UK government puts the sustained savings from smart meters and in-home displays at 3% for electricity and 2.2% for gas (Department for Energy Security and Net Zero 2026). The two figures differ in population and definition and cannot be joined. Neither measures the display graphics alone; both measure the combined effect of the meter, the display, and the advice that came with it.

The effects are small and conditional. A meta-analysis of 42 studies of feedback on energy use found an overall effect of r = .071, with effects ranging from −.080 to .480 and moderated by frequency, medium, comparison message, and duration (Karlin et al. 2015). The feedback in this meta-analysis includes non-graphic formats, so the effect of graphics alone cannot be isolated.

Noticing abnormalities: safety and operations

Toyota explains that in the Toyota Production System, when equipment stops, the andon (a problem display board) lights up to notify workers of the abnormality, so that people need only respond when there is an abnormality and no one needs to watch over the equipment (Toyota Motor Corporation n.d.). The value here lies in not having to look at anything while things are normal.

Continued use: product experience

Experiences that turn personal data into graphics have widened the use of products. Spotify’s Wrapped, which presents a year of listening in graphics and short stories, attracted more than 5 million users in its 2015 predecessor and engaged a record 227 million monthly active users in 2023 (Spotify 2024). By the end of the 2025 campaign, Wrapped had more than 300 million engaged users and more than 630 million shares on social media in 56 languages (Spotify 2026). Jay Blahnik, who led the design of the Apple Watch Activity rings, described the final rings as simple, elegant, easy to view, visually motivating, and fun (Engadget 2015). Apple reports that during the first two weeks of January, over 60% of users increased their daily exercise minutes by over 10% from their December average, and that nearly 80% of them maintained those levels through the second half of January (Apple 2026). These are observational figures, not a comparison with the absence of the rings.

Giorgia Lupi, a partner at Pentagram, wrote that we have reached peak infographics and that the next wave of visualization will be about personalization (Lupi n.d.). Wrapped and the rings generate sharing and continued use by returning data to people as pictures of their own lives.

Maintaining or damaging trust

Graphics can also help maintain trust. Citing research by the Winton Centre at the University of Cambridge, the UK statistics regulator wrote that graphical representations of uncertainty around time series, such as error bars and fan charts, maintained trust in both the numbers and the producer (Office for Statistics Regulation 2022).

The same regulator has corrected misleading graphics by name. It found that a 2023 bar chart of inflation posted by HM Treasury gave a misleading impression of the scale of the deceleration because its y-axis began at 8% (Office for Statistics Regulation 2023). The regulator pointed to the government analysis function’s guidance that the y-axis of a bar chart should always start at zero. In the same year, the chair of the UK Statistics Authority wrote that a political party’s chart of growth forecasts was misleading because neither the piles of coins nor the flags displayed the forecasts to scale (UK Statistics Authority 2023). The letter states that an important role of data visualization is to aid understanding of the data.

When graphics create no value, or work against it

Not everyone can read graphics. In a 2015 Pew Research Center survey, 63% of US adults correctly interpreted a scatterplot of sugar consumption and cavities (Pew Research Center 2015). In a 2013 survey, only 21% correctly chose, from four charts, the one showing the trend in the stock market since 2008, making it the most difficult question in that quiz (Pew Research Center 2013).

Graphics can also be misread. The US National Hurricane Center’s forecast cone represents the probable track of a tropical cyclone’s center, and its size is set so that two-thirds of historical official forecast errors over a five-year sample fall within it (National Hurricane Center n.d.). Yet a literature review commissioned by NOAA described the cone as possibly both the most viewed and the most misinterpreted product among tropical cyclone products (Eastern Research Group 2019). After Hurricane Charley in 2004, over 90% of about 100 people interviewed said that they had heard the hurricane was going to hit Tampa and did not realize they could be affected until it was too late to prepare. The same review cites an experiment in which participants who viewed graphics with the track line had higher levels of preparation than those who saw only the cone. A senior official of the hurricane center said that the cone still has merit for conveying high-level information about where the center is likely to move, but that it is better to wait for evidence than to engineer a quick fix (WLRN 2022).

The claim that graphics increase persuasiveness is also uncertain. A study reported that adding trivial graphs increased belief in a medication’s efficacy (Tal and Wansink 2016, abstract), but Dragicevic and Jansen could not reproduce the finding in replications on two crowdsourcing platforms (Dragicevic and Jansen 2018). In the replications, text with a chart was no more persuasive, and sometimes less persuasive, than text alone. The replication suggests that the chart may have promoted understanding, although the effect is likely very small.

Graphics produced by generative AI are not yet accurate enough. In a 2024 study that had GPT-3.5 Turbo and Llama 3.1 70B produce charts, the LLM-generated charts did not match the accuracy of the original charts as measured by question answering (Ford et al. 2024). A 2026 study reports that vision-language models remain vulnerable to charts distorted by techniques such as truncated axes (Mahbub et al. 2026).

Why do graphics work?

The sources above support two hypotheses. Neither has been measured directly; at this stage, several sources point in the same direction.

First, the value of a graphic arises less from its being a picture than from carrying benchmarks and evaluations in visual codes, thereby taking over the viewer’s computation. In the FDA experiment, what worked was the label that added the evaluations “Low,” “Med,” and “High” to the amounts (U.S. Food and Drug Administration 2024), and in the FTC survey, the label that showed efficiency in categories was strongest for ranking (Federal Trade Commission 2007). Kikikuru aligned its colors with the evacuation alert levels, which are benchmarks for action (Japan Meteorological Agency 2022). The traffic-light colors of in-home displays were also to be calibrated to the benchmark of each household’s range of consumption (Department of Energy and Climate Change 2015).

On the academic side, the abstract of Larkin and Simon states that diagrams organize information by location and often present the needed inferences at single locations, so that a diagram and a text with the same information can differ in computational efficiency (Larkin and Simon 1987). The abstract of the review by Franconeri and colleagues states that extracting global statistics from a graphic is fast, but comparing between subsets of values is slow, and that effective graphics avoid taxing working memory (Franconeri et al. 2021). In health risk communication, the abstract of the systematic review by Garcia-Retamero and Cokely states that well-designed visual aids help diverse decision makers (Garcia-Retamero and Cokely 2017).

From this view, it also follows that graphics without a benchmark, or with a wrong one (a bar chart whose y-axis starts at 8%), lose their value and can cause harm. This hypothesis would be wrong if presenting the same evaluative benchmarks in words or numbers alone made judgments as fast and accurate as graphics do.

Second, the value of a graphic depends both on the maker’s editing and on the reader’s ability to read. Hawkins names the removal of all superfluous information as the source of the value (Hawkins 2018), and Morris and Wiles also advise trimming any element that does not make the point clearer (Wiles et al. 2023). Meanwhile, only 63% could read a scatterplot (Pew Research Center 2015), and the forecast cone was read with a meaning (the area of impact) different from the one intended (uncertainty in the track of the center) (Eastern Research Group 2019). Spiegelhalter and colleagues write that the effectiveness of some graphics clearly depends on the numeracy of the audience, but that there is limited experimental evidence on how different types of visualizations are processed and understood (Spiegelhalter et al. 2011). The abstract of the study by Kong and colleagues states that even when titles contradict the graphic, most people perceive the graphic as impartial, and that their recall of its message tends to follow the title (Kong et al. 2019). The more a graphic leaves interpretation to the reader, the further its reading may drift from the maker’s intent. What secures the value is the maker’s editing that writes the way to read into the graphic, such as evaluative words and reference lines.

Where this account does not apply

Sources comparing graphics with other displays were limited to the experiment and survey on nutrition labels (FDA) and energy labels (FTC), and to academic research. The FDA experiment compared labels with one another and did not report differences from a no-label condition or a table-and-text condition. Figures on views, shares, and users show the scale of value but do not show that the same numbers would not have been reached without graphics. The figures for in-home displays, Apple’s rings, and energy feedback do not isolate the effect of graphics alone. The value of memorability (the abstract of Borkin and colleagues 2013 states that color and human-recognizable objects enhance memorability) was not found in industry sources. For Japan, the only sources were the Japan Meteorological Agency’s records of its color design and its aims; no measurements of effect could be obtained.

Unverified items

No unverified statement remains in the main claims. Sources that could not be reached (the UK Food Standards Agency’s evaluation of traffic-light nutrition labels, the evaluation of the rescaled EU energy label, the Agency for Natural Resources and Energy’s research on Japan’s unified energy-saving label, the FDA’s Communicating Risks and Benefits, NICE guidance on decision aids, the Consumer Affairs Agency’s research on nutrition labeling, and measurements of the effect of the Japan Meteorological Agency’s color changes) are recorded in the ledger with the routes attempted.

References

Footnotes

  1. The sources are sorted into public bodies and regulators (T1), figures published by companies and research organizations (T2), and statements by the makers of graphics (T3, developer voice), with academic research treated separately as support for the reasons. Company figures are self-reported, and some do not disclose definitions or denominators. Academic articles whose full texts could not be obtained were used with attribution within the scope of their abstracts. The ledger is at source/review/dataviz-value-types/industry.md. ↩

Author: Shuichiro Ogawa (Design Researcher / Consultant) About me →