Estimated reading time: 7 minutes

Summary

Encoding information entails translating information into graphical representation.
Cleveland & McGill (1984) and Mackinlay (1986) created lists of graphical representations that human perception is best able to interpret and decode.
More recently, Munzner argues for a complementary and alternate system.
Graphical representations can be thought of as a visual language: expressiveness and effectiveness are two design criteria whereby a practitioner can formulate and evaluate sentences uttered using this particular language.

The section on perception and the deep dive in graphic design principles (in the next module) gives us a theoretical understanding of human perception. This page focuses on the tools we'll need to apply that theory to practice.

The term encoding is frequently used in visualisation contexts. It describes the process of translating information into a graphic representation. To take a simple example; drawing 50% as one half of a circle in a pie chart is the act of visually coding information.

There are many ways of describing visual encoding. We will start by focusing on two seminal articles; the first article is the foundation on which the other builds. Both articles attempt to generate practical tools whereby practitioners can apply theory to practice.

Foundational Text nr. 1: Cleveland and McGill, 1984

One of the foundational texts in the field is William S. Cleveland och Robert McGill's, "Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods" which appeared in the Journal of the American Statistical Association in 1984.

The article's importance is largely due to its list of which graphic representations human perception is best able to interpret and decode.

To describe the basic interpretive process that human perception undergoes, Cleveland and McGill use the concept elementary perceptual task. The term describes how a viewer "performs one or more [...] mental-visual tasks to extract the values of real variables represented on most graphs."

Cleveland and McGill’s list, which focuses on the extraction of quantitative data, consists of ten different awareness-related processes involved in the interpretation of the visualisation. The suitability of the different processes and their expressive ability are empirically verified and ranked (details to be found in the article itself).

The perceptual tasks are ranked as follows:

Position on a common scale
Position on a non-aligned scale
Length
Direction
Angle
Area
Volume
Curvature
Shading
Color saturation

This ranking suggests that position would be a more effective way of representing quantitative data than, for example, area.

Foundational Text nr. 2: Mackinlay, 1986

Two years later, Jock Mackinlay developed Cleveland and McGill's work in "Automating the design of graphical presentations of relational information", published in ACM Transactions on graphics.

Mackinlay's model ranks awareness-related processes with respect to three different data measurement scales: quantitative data (like his predecessors Cleveland and McGill), ordinal data and nominal data.

Considering and comparing both the characteristics of the data, as well as our cognitive ability to decode certain parameters of that data, can be a useful aide in the visualisation process.

In the table below, horisontal lines connect each column, in order to highlight how the ranking of a particular process may vary depending on data scale. The top rows of each column contain encodings that are more suitable for representing data of that type.

Anpassad version av Mackinlays ranking av perceptionsassocierade uppgifter

Table describing perception-associated tasks, organised in accordance with three different data measurement scales. Image created by Karolinska Institutet University Library (Creative Commons BY SA). Based on Mackinlay (1986).

For instance, the table in the picture above shows how the height or length of a bar, in other words its position along an axis, is an effective way of representing quantity. The same is true for slope. Take the line between different measurement points in a line diagram: the steeper the incline, the greater the change in value. By contrast, colour hue is poorly equipped to convey quantity, but easily conveys nominal data, such as different categories in the data.

Cirkeldiagram och stapeldiagram som exemplifierar olika visuella tolkningar av samma dataunderlag

Image by Karolinska Institutet University Library. Creative Commons BY SA

We can see this clearly in a simple pie chart (see example to the left in the image above). The angle of the "pie" describes what portion of the total quantity is being indicated (quantitative data), while colour is used to differentiate each slice of pie from the rest of the pie (nominal data). Now try to compare slice A and slice D. Which of them represents a larger portion: A or D? And how large is that difference?

Then look at the chart to the right. The bars in the diagram next to the pie chart are built using the same basic data set. The small but significant difference in quantity between A and D is clearer here, despite the fact that the pie-slices in the circle diagram show the exact same thing. This illustrates how height, a variation on what Cleveland and McGill call position on a common scale, more readily conveys absolute and comparable values.

That said, keep the strengths and weaknesses of each type of diagram in mind; different diagrams can be used to convey different aspects of any data set. This is connected to what is often termed effectiveness and expressivity, which we'll discuss more below.

A complementary perspective: Munzner, 2015

Cleveland & McGill's initial ranking and Mackinlay's model are recommended tools that can help you get started thinking about the principles of graphic design. They describe why one way of representing data might be perceived as more effective than another. However, neither represents an absolute truth about how to go about doing it.

In her book Visualization Analysis and Design (2015), Tamara Munzner offers a highly controlled approach to working with visual encoding. It is based on a similar ranking system that involves visual and cognitive processes, but Munzner uses a different vocabulary, and her categorisations follow a slightly different logic. Data types are either categorical (nominal) or ordered (ordinal, quantitative) and they are visually encoded using marks, usually a geometric shape or figure (point, line, area). The appearance of the marks are controlled by what Munzner calls visual channels, for example the position, colour, angle or size of the mark. Munzner’s system enables the practitioner to methodically work step by step to create a visualisation while avoiding unfortunate pitfalls and mistakes.

Let’s take a look at how this can be applied. The image below shows a simple generic bar chart to the left and a stacked bar chart to the right.

In the simple bar chart, the position (visual channel) of a bar (mark), along the vertical axis encodes quantitative data. The position (visual channel) of the bar (mark) along the horizontal axis encodes categorical data.
In the stacked bar chart, further encoding is possible by using multiple colours to express another layer of categorical data, as well as varying positions along the vertical axis to express quantitative data.

2 generic bar charts displaying quantitative and categorical data.

Image by Karolinska Institutet University Library. Creative Commons BY SA

Graphical representations as a visual language

In his 1986 article, "Automating the design of graphical presentations of relational information," Jock Mackinlay describes graphical representations as a visual language. A visualisation, such as a graph or a diagram, can therefore be thought of as a sentence expressed in this particular language.

Mackinlay stipulates two design criteria whereby a practitioner can formulate and evaluate sentences uttered using this particular language: expressiveness and effectiveness.

The criteria are defined as follows:

Expressiveness: A set of facts is expressible in a visual language if the sentences (i.e. the visualisations) in the language express all the facts in the set of data, and only the facts in the data.
Effectiveness: A visualisation is more effective than another visualisation if the information conveyed by one visualisation is more readily perceived than the information in the other visualisation. This criteria is connected to the table that ranks cognitive processes.

Formulating an expressive and effective sentence

Once we take these two design criteria into account, expressiveness (or what is being expressed) and efficiency (how it is being expressed), we can begin formulating a sentence. One also affects the other, that is, the content of the data affects how it can and should be presented. It's a useful starting point, since it can be difficult to choose the right diagram, or if we continue the metaphor, formulate the sentence correctly. The process is iterative and the sentence might be reformulated, over and over again, until it is expressive and effective enough.

As demonstrated in the example of the pie chart and the bar chart, different diagrams have different strengths and weaknesses. The use of pie charts is often criticised; they aren't regarded as exact enough; it's difficult to compare the size of the pie slice, etc. These criticisms aren't without merit, but a pie chart also has advantages. If the slices are few and clearly distinguishable, we tend to decode them very quickly. Angles that divide the circle into halves or quarters are easily identifiable. Furthermore, pie charts are very common, and one of the basic chart types that you learn to interpret in school. In other words, pie charts might be easy for people to understand. Especially so when combined with labels or a clear legend (but then you'll want to consider this for most of your visualisations).

Translating Mackinlay

Mackinlay's two design criteria are sometimes perceived as useful but cumbersome. In a course on visualisation, Jeffrey Heer, of the University of Washington, offered the following translation:

Expressiveness: Tell the truth and nothing but the truth (don’t lie, and don’t lie by omission)
Effectiveness: Use encodings that people decode better (where better means faster and/or more accurate)

Not just graphs and diagrams

Of course, a visualisation does not consist only of graphs and diagrams. Instead, the data set might include advanced image data or a comprehensive series of images. We might be talking about graphic material generated by a microscope, through tomography, or via some other equipment. Even here, however, you may need to employ some practical visualisation tools to help you convey your data effectively. Both Mackinlay's table as well as Munzner's system may then be of use in your choice of which visual channels to use and how to do so.

Recommended readings and resources

William S. Cleveland and Robert McGill, "Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods", Journal of the American Statistical Association (1984).
Jock Mackinlay, "Automating the design of graphical presentations of relational information," ACM Transactions on graphics (1986).

For an annotated analysis of Cleveland and McGill's list, including examples:

Nathan Yau, "Graphical perception - Learn the fundamentals first." Flowing Data (2010).

If you're looking for the right diagram to use, check out the list of resources listed on the KI Library's website:

Visualisation resources