Shuichiro Ogawa
日本語

Notes · updated 2026-10-07

What Does Data Visualization Show When It Departs from Efficiency? Wholes Without Numbers, Readings That Stay Open, and Deliberate Abstraction

Moving away from the earlier notes that measured the value of data visualization by optimization criteria such as speed, accuracy, behavior change, and number of users, this note examines the significance of seeing the whole without numbers, of opening new angles because the way of reading is not fixed, and of abstract…

Contents (8)
  1. The yardstick itself was one choice of value
  2. Seeing the whole without numbers
  3. Because the reading is not fixed, new angles open
  4. Abstracting deliberately to avoid locally optimal views
  5. Feeling and people beyond the numbers
  6. The risks of departing
  7. Why is departing significant?
  8. Where this account does not apply

The yardstick itself was one choice of value

The two previous notes (What Value Does Data Visualization Have in Industry? Market Valuation, Where Value Arises, and What Changes after Generative AI (from Industry Sources) and Where Does the Value of Showing Data as Graphics Lie? Kinds of Value and Why They Arise (from Industry Sources)) measured the value of data visualization by how quickly and correctly charts were read, how they changed behavior, and how many people used them. By that yardstick, the value of a graphic comes down to taking over the viewer’s computation. Is this yardstick, however, the only one by which the value of visualization can be measured?

Within visualization research, measuring value by efficiency has an origin. In 2005, van Wijk adopted a technological viewpoint in which the value of visualization is measured by effectiveness and efficiency, and presented an economic model that expresses value in terms of knowledge gained and costs (van Wijk 2005). The same paper also discusses two alternative views: visualization as an art and visualization as a scientific discipline. In 2014, Stasko argued that evaluation focused on benchmark tasks, which compares speed and correctness, is not sufficient for evaluating visualization systems, and named conveying the essence of the data and spurring insightful questions among the values to consider (Stasko 2014).

There is also a critique of the values embedded in the canon itself. In a 2026 paper (a preprint, not peer reviewed), Saharan and colleagues read canonical texts by Bertin, Tukey, Wilkinson, Ware, and Munzner, argued that this corpus prioritizes the values of universality, objectivity, and efficiency, and called for a more pluralistic range of values to shape visualization tools and guidelines (Saharan et al. 2026). In 2013, Dörk and colleagues wrote that the research community had not given sufficient thought to how values and assumptions pervade information visualization, and used “critical” to mean being aware of the values and assumptions embedded within one’s discipline (Dörk et al. 2013).

Efficiency is not a neutral criterion but one value that comes from the traditions of science and engineering. If so, the significance of departing from efficiency can be asked about in terms other than efficiency.

Seeing the whole without numbers

The yardstick of whether individual values can be read accurately measures only one use of a graphic. Bertini, Correll, and Franconeri took up the argument that, because position is perceived most precisely, all charts might as well be scatter plots; they argued for a more expansive view beyond the perceptually precise encoding and decoding of individual data values and made the case for “inefficient” visualizations (Bertini et al. 2020). The paper notes that, among its examples, the authors’ favorite big-picture view was the heatmap, despite its status at the bottom for the precision of extracting ratios between individual values.

Vision can grasp a whole distribution without reading values one by one. The abstract of Szafir and colleagues states that visual ensemble coding supports rapid extraction of statistics about distributed visual information, and that data visualizations allow observers to use this ability to estimate actual statistics on data (Szafir et al. 2016). Reading individual values and grasping the whole are distinct perceptual processes.

Makers have also removed numbers to show the whole. Ed Hawkins, who created the warming stripes, told BBC News that the lesson learned was that you do not need the numbers to get the message across; the colors will do that (Eco-Business 2019). Numbers left on a graphic invite viewers to read values one by one. Removing them is a choice to give up the precision of reading in exchange for directing the viewer’s eye to the shape of the whole.

Because the reading is not fixed, new angles open

HCI has long treated ambiguity as something to remove. The abstract of Gaver, Beaver, and Benford argues that ambiguity, usually considered anathema in HCI, is a resource for design that can encourage close personal engagement with systems (Gaver et al. 2003). It distinguishes three places where ambiguity arises (in the information itself, in the sociocultural contexts used to interpret it, and in the individual’s interpretive and evaluative stance) and describes tactics for emphasizing each.

Making the familiar strange also opens angles. The abstract of Bell, Blythe, and Sengers argues that because the home is so familiar, it is necessary to make it strange, or defamiliarize it, to open its design space, and notes that home appliances are loaded with cultural associations, such as the gendered division of domestic labor, that are easy to overlook (Bell et al. 2005). When the way to read a graphic is not fixed, viewers cannot fit it into familiar categories, and their attention turns to the assumptions themselves.

Encounters with things one was not looking for also lie outside the efficiency yardstick. Thudt, Hinrichs, and Carpendale described serendipity as an often neglected but still important factor in information seeking, research, and ideation, and made it a goal of information visualization (Thudt et al. 2011). By showing a book collection through abstract, metaphorical, and visually distinct representations to entice curiosity, and by highlighting alternate adjacencies between books, they supported encounters with books that people were not looking for.

Difficulty in reading can also create value. As a counterpoint to efficiency-based design theory, Hullman, Adar, and Shah set out conditions under which visual difficulties benefit comprehension and recall, and characterized effective graph design as a trade-off between efficiency and learning difficulties (Hullman et al. 2011). Whether difficulty is appropriate depends, they argue, on factors such as active processing and engagement.

Makers also ask viewers for effort and doubt. The data designer Nadieh Bremer described making more complex visuals that need a little effort to understand, in order to give a lot of information eventually (Rakotondravony and Forrest 2019). The data journalist Mona Chalabi said that part of the purpose of Chalabi’s hand-drawn illustrations is to make people question the illustration in front of them, because there is a high degree of imprecision in data (Al Jazeera 2019). Chalabi added that for every statistic, the truth lies somewhere in the range around that number. The information designer Federica Fragapane said that working with care on the aesthetics of the works is a way to invite people in, to encourage them to look closely and read the stories told through data (designboom 2025).

Abstracting deliberately to avoid locally optimal views

Numerical indicators readily become objects of optimization themselves. Discussing audit in British universities, the anthropologist Marilyn Strathern wrote that when a measure becomes a target, it ceases to be a good measure, and that the more examination performance becomes an expectation, the poorer it becomes as a discriminator of individual performances (Strathern 1997). Placing indicator values on a dashboard sits next to making those values targets.

Continually refining one view also has a pitfall. In a design experiment, Dow and colleagues wrote that iteration can help people improve ideas but can also give rise to fixation, continuously refining one option without considering others, and reported that creating multiple prototypes in parallel led to better results and more divergent designs (Dow et al. 2010). This was an experiment on web advertisement design, not visualization research. Even so, if watching a graphic of a single indicator resembles refining a single option, abstracting to keep several readings open can be expected to work like holding several options in parallel.

The choice of the level of abstraction is the most upstream decision in visualization design. Munzner divided visualization design into four levels (the domain problem, data and operation abstraction, encoding and interaction, and algorithm) and wrote that an upstream error inevitably cascades to all downstream levels (Munzner 2009). Reconsidering what to show at the level of abstraction opens views that no amount of refinement in encoding below it can reach.

Makers have used abstraction for purposes other than efficiency. Giorgia Lupi and Stefanie Posavec conceived Dear Data, in which they exchanged hand-drawn data postcards, as a personal documentary rather than a quantified-self project (Lupi and Posavec n.d.). They wrote that instead of using data just to become more efficient, data can be used to become more humane and to connect with ourselves and others at a deeper level. The two say they prefer to approach data in a slower, more analogue way. Lupi has said that data is never neutral but an abstraction of reality created through selection, perspective, and human decisions (German Design Council n.d.). The designer Moritz Stefaner said that truth and beauty are equally important, so that a project with only one of them is not done, and that trying to blow up dull data with spectacular visuals is something Stefaner despises (Stefaner 2014). The artist Ryoji Ikeda said that scientists get the work right away and say that it looks beautiful but does not mean anything, and that interpretation is left to the critics (McCallum 2011). Ikeda’s work is an extreme case of abstraction cut loose from fixed meaning.

The /network works on this site also try to draw in abstract forms while keeping the data as the protagonist (The Citation Network of This Wiki: Analysis and Artworks Treating Each Note as a Hyperedge, Drawing Networks as Data Art: Literature on Placement, Bundling, Time, and Flow).

Feeling and people beyond the numbers

The efficiency yardstick captures only part of how people relate to data. Studying everyday engagements with data and its visualization, Kennedy and Hill argued that emotions are vital components of making sense of data, and that in datafied times it is not only numbers but also the feeling of numbers that is important (Kennedy and Hill 2018). Kennedy and colleagues identified six factors that affect engagement with visualizations (subject matter, source and media location, beliefs and opinions, time, emotions, and confidence and skills), and argued that the definition of effectiveness varies depending on how, by whom, and why visualizations are encountered (Kennedy et al. 2016).

Behind the numbers are people. The data artist Jer Thorp said that the realization came that the numbers are not just numbers, that each of them is tethered to something in the real world, and often those things are people (Scola 2021).

The risks of departing

Graphics that depart from efficiency can also be misread. In an experiment on hurricane forecasts, Padilla and colleagues showed that salient features of summary displays of uncertainty can be misunderstood as displaying size information (Padilla et al. 2017). Ensemble displays, which overlay many forecast tracks, supported an accurate interpretation of the distribution, but when participants made point-based judgments, they could overweight individual tracks.

A wrongly chosen abstraction cannot be rescued. In Munzner’s four-level model, a wrong abstraction (showing people the wrong thing) cannot be fixed however refined the downstream encoding is (Munzner 2009).

The freedom of exploration also borders on errors of inference. Hullman and Gelman argued that design philosophies that emphasize exploration over other phases of analysis risk confusing a need for flexibility with the conclusion that exploratory visual analysis is inherently “model free,” and that without grounding in theories of statistical inference, they can lead to representations of uncertainty that discourage valid inferences (Hullman and Gelman 2021).

Ambiguity comes at the expense of clarity. The abstract of Aoki and Woodruff notes that research on face-to-face social interaction often identifies ambiguity as a resource for resolving social difficulties, and discusses two design cases of personal communication systems (Aoki and Woodruff 2005).

Abstraction can also cut data off from people. Thorp said that behind polite words such as collection or gathering lies something more violent, in which data are scraped, abstracted, and mined from individuals, and that placing data in a human context gives it meaning (Cannon 2019). Thorp also said that part of the appeal of data work is that one does not have to talk to people (Scola 2021). The value of abstraction and the danger of abstraction making people invisible are two sides of the same operation.

Makers set limits on how they depart. Chalabi said that the hand-drawn illustrations are digitally aligned pixel for pixel with computer-generated graphics so that they are as accurate as any computer graph (Hahn 2023). Stefaner said that makers are creating views of the world that shape people’s world views, so both the “how” and the “what” of data visualization need continuous, critical investigation (Stefaner 2014).

Why is departing significant?

The literature above supports two hypotheses. Neither has been measured directly; at this stage, several studies and practices point in the same direction.

First, the efficiency yardstick belongs to situations in which the question is already fixed; where it is not, a graphic becomes a place that generates questions rather than a tool that delivers answers quickly. Evaluations that compare speed and correctness presuppose that what the viewer wants to know has been decided in advance. What Stasko found missing from such evaluation, conveying the essence of the data and spurring insightful questions (Stasko 2014), is the value of situations in which the question is not yet fixed. The big picture of Bertini and colleagues (Bertini et al. 2020), the serendipity of Thudt and colleagues (Thudt et al. 2011), and the ambiguity of Gaver and colleagues (Gaver et al. 2003) all create value by leaving room for viewers to bring in their own questions. Strathern’s observation that a measure that becomes a target ceases to be a good measure (Strathern 1997) shows the cost of fixing the question to a single number. This hypothesis would be wrong if, even in exploratory situations without a fixed question, graphics that are efficient on fixed tasks generated more new questions and discoveries.

Second, the value of departing from efficiency arises only when the maker designs the departure. Hullman and colleagues locate the conditions under which visual difficulties help in active processing and engagement, and do not recommend difficulty without conditions (Hullman et al. 2011). Chalabi keeps accuracy even in hand drawing (Hahn 2023), and Hawkins, though removing numbers, keeps the sequence of colors faithful to the temperature data (Eco-Business 2019). Meanwhile, salient features of summary displays are read as unintended quantities (Padilla et al. 2017), and a wrong abstraction cannot be rescued (Munzner 2009). The operations of removing numbers, leaving the reading open, and abstracting open new views when the maker draws the line between accuracy and misreading. Departing without drawing that line produces only misunderstanding or decoration. This hypothesis would be wrong if graphics whose departure was not designed (randomly abstracted graphics, or graphics that do not keep accuracy) opened new views just as well.

Where this account does not apply

Empirical studies that measure the value of departing from efficiency by yardsticks other than efficiency are few. Hullman and colleagues 2011 is a synthesis of earlier research, and the divergence effect in Dow and colleagues 2010 was measured by outcome yardsticks, namely click-through rates and expert ratings. Bertini and colleagues 2020, Saharan and colleagues 2026, and Dörk and colleagues 2013 are argumentative papers, and Saharan and colleagues 2026 has not been peer reviewed. The full texts of Gaver and colleagues 2003, Bell and colleagues 2005, and Szafir and colleagues 2016 could not be obtained, and they were used within the scope of their abstracts. Sengers and Gaver’s 2006 work on staying open to interpretation, D’Ignazio and Klein’s Data Feminism, Bertin’s levels of reading, and the original of Tukey’s exploratory data analysis could not be reached and were not used. The makers’ statements come from interviews and articles, from the position of makers explaining their own work.

References

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