3.3 Accessibility Considerations
Data visualization is a misnomer: it would no doubt be preferable to refer to the whole entreprise as data perception. Indeed, were it not for the accident of evolution that gave such prominence to our sense of sight, we might be referring to it as data sonification (sound/music, text to speech), physicalization (texture-based), olfactization (odor/smell), gustification (taste), or other -ation terms related to various senses or combination of senses (ESP?).22
But humans grow old, and the quality of their sight changes over time. Furthermore, a sizeable portion of any data visualization audience is subject to various conditions (not necessarily sight-related) that could make it arduous to get the most out of data presentations. Designers must take this into account in order to produce accessible charts, dashboards, and infographics.
Traditionally, the go-to solution has been to translate text and tables into Braille, a
“[…] system of raised dots that can be read with the fingers by people who are blind or who have low vision. […] Braille is not a language. Rather, it is a code by which many languages—such as English, Spanish, Arabic, Chinese, and dozens of others—may be written and read.
As of 2016 the main code for reading material is Unified English Braille, a code used in seven other English-speaking countries [What is Braille?, American Foundation for the Blind].”
It is a clever system, and some conventions can improve the readability, but it requires a medium on which embossing is possible. Summary tables can definitely be translated to Braille (although even the smallest of tables may end up taking up more physical space than expected), but we cannot always do so for charts (presumably, the chart itself can be embossed into the medium, see data physicalization below).
How, then, do we make data visualizations accesible to non-sighted audiences? We could decide to describe the features and emerging structures in a visualization … assuming that they can be spotted in the first place. This approach is tailor-made for data storytelling, where only a small number of charts is produced, but it is unlikely to be useful with data exploration, especially when chart generation is automatic and so many charts are produced that none of them are visited before an analyst calls them up.
In data storytelling, clarity is key: we produce clear and meaningful visualizations that should stand on their own, but we must also describe what it is that we see in a fashion that allows all (including people for whom charts are akin to an alien script) to “see” the insights.
We purposely steer away from attempting to describe all the insights (assuming that they can even all be “seen”) in order to focus on a few highlights, but sighted individuals will always have the ability to spot information and results that the teller has elected to omit. Not so, then, for a non-sighted audience.23
There are, however, other options, which we discuss below.
3.3.1 Sonifications and Physicalizations
An intriguing approach to data representation is that provided by data sonification, using sounds (and music) to present and interpret data: clock chimes and Geiger counters, which produce a ‘click’ whenever a radioactive decay occur, offer early examples of such a process.
In the next video, the planetary structure of the TRAPPIST-1 star system is represented with different frequencies (for each planet) and drum hits (for occlusions):
This is eerily reminescent of some Radiohead songs.24 The approach could be extended to use different instruments (timbre) for different planet types (perhaps winds for rocky planets and brass for gas giants, say); other elements that could be mapped to data features include pitch, amplitude, and tempo.
It is also easy to imagine how this methodolgy could be applied to different stars (and their attending planets), leading potentially to different sonification “styles”: stars that “sound” the same might have something in common which may be difficult to identify in the physical universe (which is to say, according to measurements such as luminosity, spectral type, etc.).
Another impressive experience is that of Lily Asquit, a non-sighted physicist at CERN. In the video below, she tells her audience how she tracks elementary particles in the Large Hadron Collider, which is traditionally done using droplet diagrmas (a sighted method).
After visualization, sonification is probably the most mature data perception approach, as discussed by Carla Scaletti in her excellent ICAD 2017 keynote address: Why Data Sonification Is a Joke.
Additional examples can be found here:
NASA’s Sounds from Around the Milky Way and A New Cosmic Triad of Sound;
Andrea Polli’s Atmospherics/Weather Works: A Spatialized Meteorological Data Sonification Project;
350+ examples are showcased at the Data Sonification Archive.
Data physicalizations have been pursued throughout history; artists and artisans, in particular, have been representing aspects of their worlds through a variety of media over the years.25
Physical representations have a distinctly analog aura (as opposed to digital, see The Analog/Digital Data Dichotomy for a discussion of this topic); as a result, modern uses can feel rather “gimmicky” at times, but there are plenty of “legitimate” approaches.
As is the case with visualization and sonification, the challenge is to find elements that can be mapped meaningfully to the data features (variables), to allow for comparisons, multivariate links, etc. (as in Principles of Analytical Design and Representing Multivariate Observations).
The Inca used khipus (strings and knots) as a physical data storage and representation device; it is believed that colour, the relative knot positions, knot types, and rope lengths were used to encode the variables.
More recently, tactile infographics have been created on thermoform (heated sheet of plastic sealed on a physical model) or swell paper (thermoform-lite). The tactile variables that can be used to represent data include: vibration, flutter, pressure, temperature, size, shape, texture, grain, orientation, and elevation.
Sonifications and physicalizations can unite in audio tactile maps, for instance: in this case, software with audio files is used to convey information as the user’s fingers rolls over features or symbols of the tactile display.
Various examples can be found at the Data Physicalization Wiki.
In Figure 3.19, the focal point is indicated by a change in colour (a dark yellow bar in a sea of beige).
Aesthetically, this is a pleasing choice, but would you agree that the pink bar in the chart on the right (see below) is easier to spot?
A sizeable proportion of the population (~4% in North America) is colourblind to some degree, due to a genetic defect affecting one (or more) of the eye’s cones (red, green, blue).
Charts that rely solely on colours might fail to convey the full extent of the data story to a significant proportion of the audience: the primary and secondary colours, for instance, might not be as easily differentiable as one would hope, depending on the cone defect.
Part of the solution is to consider using contrast-friendly colour palettes for charts, and to support the signals conveyed by colours with other design elements (shapes, size, annotations, etc.), as in the pallets presented below.
It can be difficult to guess what our colour choices will look like to colorblind individuals; uploading charts to simulators (such as Coblis, but there are other tools available) can provide an idea of how they are perceived in the various colourblind spaces:
3.3.3 General Resources
Colourblindness is far from the only accessibility issue related to data charts, but it is one of the most visible (forgive the pun). Exactly why that should be the case is a hotly debated: some commentators believe that it solely because it is a condition that disproportionately affect biological males.26
While we believe this certainly plays a role, we must point out that the focus on colourblindness could also be partly explained by convenience: of all accessibility solutions, changing colours is probably the easiest to implement.
In a short twitter thread, Frank Elavsky lists additional accessibility resources:
Note that producing high contrast charts seems to solve a sizable number of accessibility issues.
3.3.4 Practical Suggestions
Practically-speaking,  suggests to make data charts more accessible by:
- adding text descriptions to the charts (which are early forms of storytelling with data);
- using colours that are bold and clear enough for people to see both text and graphical elements;27
- minimizing the number of colours used in charts, such as going from 4 (1 per graph) to 2 (1 per chart pattern);
- using white separators to increase the contrast between various chart elements (according to the corresponding Gestalt principles);
- favouring the use of “short” graphs over “wide” graphs to reduce issues associated with attention deficits (compare the two charts below).
Making data charts accessible is difficult, but it is worth the effort: accessible charts benefit all audience members.
The first two of these, in fact, are active areas of research and implementation by practitioners.↩︎
The same objections arise for speech-to-text tools.↩︎
Packt Like Sardines in a Crushd Tin Box, anyone?↩︎
Although the resulting physical products are rarely expected to be “touched” by the general public.↩︎
A reverse situation arises among biological females linking deficiencies in the eye’s cones with issues of depth perception.↩︎
Web Content Accessibility Guidelines (WCAG) suggest meeting the WCAG AA requirements. To check if font colour and size choices are AA accessible, we can use a contrast checker website. For colours to be AA accessible they need to have a contrast ratio of at least 3:1 for graphical elements, and 4.5:1 for normal text.↩︎