Mick McQuaid

The salary performance exercise revealed numerous perspectives on information and information representation. I’d actually like you to continue your thinking along these lines next week, so it may be helpful to review some of the issues you raised. To do so, let’s look at the whiteboards, and in one case, the online document you used to explore the problem. I went back and took photos of the whiteboards to review them. My photos did not turn out very well, so I photoshopped them to highlight your marker strokes and soften the smudges on our whiteboards. I hope you don’t mind that this distorts the original work, and accept it in the spirit of spotlighting topics for discussion.

The first thing most of you did was wrestle with the nature of the information. The concepts shown above include the notion of seasonality and cyclicality, size of icons as a proxy for quantities of money and symbology familiar to your audience for both time and money.

It seems that the basic entity we’d like to represent is the team. And the main attributes of that entity are its salary and performance. Salary, in this data, is a constant. Performance varies, so the ratio of salary to performance varies. The extremes of that ratio represented by the burning money and the logo piggy banks, whimsical icons that may fly through space (as a proxy for time).

A simple, compelling way to represent a team’s wealth is by a circles sized in proportion to that wealth. This sketch suggests that animation and memory of a path may be enough? Do we need a trail? Can we just show these bubbles progressing from left to right with a slider? What would we make of trails, if any? I like this representation but bear in mind that, at any moment in the animation, only a narrow vertical slice is being used to actually show bubbles. The vast majority of the canvas is used to portray time. We’ll see a solution to this problem later.

This exercise was meant to prompt discussion of possible solutions. You can see that this sketch was the focal point of an active discussion in which a lot of ink was spilled. I consider that a sign of successful brainstorming. Here we see further development of the idea presented in the previous sketch: circles marching across a time field from left to right. This is a very frequent depiction of time. Again, as with the previous example, this devotes a lot of space to the time representation and leaves the possibility of trend lines as a record. A significant question with this approach is that the two lines shown may be much more legible than would lines for all teams appearing at once.

Here is one three space-saving approaches that do not use space to depict time. In this representation, little red balls bounce up and down showing shifts in performance as the background of bars showing salary remains constant. I can imagine refining this display by allowing the user to rearrange the order in which teams appear according to different criteria. It would also be interesting to think about letting the balls leave brief trails as they move or otherwise change. Finally, the bars representing money are muted here. Should they be more muted or more prominent? It seems to me that there are two conflicting desiderata. First, the money is important. Second, the money is constant. Those two characteristics suggests that maybe we want to find some function of money, such as bang / buck, that is not so conflicted and represent that instead.

This is medium shot of a very detailed, fully-realized poster. The students who made this decided on their representation very quickly and spent most of their time creating a very detailed depiction of what they had in mind. This reminds me of a recent Business Week column about Apple’s design process. The first ingredient described by Michael Lopp, senior engineering manager at Apple, was Pixel Perfect Mockups. He claimed that they remove all ambiguity. As a result, they uncover problems very far upstream and have little need to correct mistakes later in the process. There’s quite a lot you can say about this design at this stage. The main one, from my point of view, is that it manages positive and negative space very well. Space is a precious resource for any designer.

Finally, Mike Harmala shared with us the further refinement he made of a space saving design after class. This one shows time explicitly with a very prominent arrow on the slider. It also appears to include ratio between money spent and performance. Position is used to show performance, so a given circle may move back and forth across the surface from one week to the next. Size is used to show raw money spent, so we have one way each to show money and performance independently, as well as one way to show them together.

What are principles? Tufte tells us that we should do certain things. In Tufte (2006), he tells us that we should show comparisons, contrasts, differences. It’s his first principle of analytic design. In what sense is this a principle? How do we know that this is what we should do. I shared this “principle” with one of the most successful scholars I know, Judy Olson, and she replied that we should show what’s surprising. We should show differences where those differences are meaningful, but we should show similarities where those are meaningful. Some trends are surprising; some similarities are surprising.

It is not at all clear that these principles are grounded in any kind of theory. At the beginning of my other class, I offer a lengthy quote from Stephen Hawking about theory in A Brief History of Time on page 7.

A theory is a good theory if it satisfies two requirements. It must accurately describe a large class of observations on the basis of a model that contains only a few arbitrary elements, and it must make definite predictions about the results of future observations. For example, Aristotle believed Empedocles’s theory that everything was made out of four elements, earth, air, fire, and water. This was simple enough, but did not make any definite predictions. On the other hand, Newton’s theory of gravity was based on an even simpler model, in which bodies attracted each other with a force that was proportional to a quantity called their mass and inversely proportional to the square of the distance between them. Yet it predicts the motions of the sun, the moon, and the planets to a high degree of accuracy.

Note that Hawking is giving what we might call a normative definition, telling us what a theory should be. There might be plenty of things parading around under the theory banner that could be classed as bad theory. Note also that Hawking requires that a theory both describe and predict.

Why does theory have value? The theory of gravity can be used to decide what to do in a wide variety of situations. A theory of analytic design could be used to tell us about the causes and effects of different actions we take as designers. But what about the Principles of Analytic Design as articulated in Tufte (2006)? What are they good for? For example, the sixth principle tells us that content counts most of all. I sincerely hope that you don’t believe that and that you instead believe that content alone doesn’t count for much. There would be lines around the block at the Library of Congress otherwise. Clearly, I am not suggesting that you take Tufte’s principles at face value. So, why then, do I recommend reading his work?

Part of the answer can be found in Tufte (2003), where he discusses the Cognitive Style of Powerpoint. Tufte rails against a badly designed presentation tool that makes a lot of choices for you and makes most of the choices poorly. While railing, Tufte does what Tufte does best (and what you can also see on his blog): he demonstrates his ability to carry on a data-laced conversation. I believe that, to do this effectively, it helps to be exposed to exemplary work. For this reason, and this reason alone, I think it’s worthwhile to familiarize yourself with the normative perspective. Another way to familiarize yourself with this kind of work is to look at the Mathematica notebooks of our colleague Eytan Bakshy. In fact, let me use this opportunity to call upon Eytan to share a suitable example with us as another aid to our conversation about what you should and shouldn’t do.

Normative, or prescriptive, ways of looking at infoviz guide us toward a narrow set of fashionable design choices. You may guess from the words “narrow” and “fashionable” in the preceding sentence that I have mixed feelings about this perspective on information visualization. This is certainly a valuable perspective but, before we explore it, I’d like to insert two placeholders into your thinking. The first of these has to with the position of infoviz in relation to other disciplines. The second has to do with users, contexts, and communities.

Above is a Venn diagram based on Table 1.1 of Card (1999), colored by ColorBrewer. Looked at this way, InfoViz may be said to inherit from several disciplines. There’s no reason why InfoViz shouldn’t borrow the principles of these disciplines as far as they offer design guidance. Plenty of guidance is available in the worlds of data graphics, information design, and external cognition. Consider this Venn diagram, an example of external cognition that has been available for about 125 years, with many refinements along the way. The coloring comes from a contemporary cartographer, Cynthia Brewer, and a data graphics aid called ColorBrewer. ColorBrewer asks you to consider whether the variables you’d like to color are sequential (which is what I chose), diverging (two groups with subgroups), or qualitative (nominal or categorical). It shows examples of color schemes reflecting these choices on a map of the Southeastern USA, as well as a variety of encodings such as CMYK and RGB to implement these color schemes. The colors were selected by an experimental process where the researchers displayed maps using various color schemes to participants who answered questions about the maps.

The ColorBrewer work is similar to that of William S. Cleveland, documented in his 1985 book, The Elements of Graphing Data, Hobart Press, Summit, NJ. Cleveland displayed graphics and asked questions about the quantities shown to experimental participants. In addition to his own work, he also collected studies conducted by his colleagues at Bell Labs. His books are responsible, for instance, for a worldwide aversion to pie charts. Why? Because he found that it was more difficult for people to estimate the quantities represented on pie charts than on other graphical artifacts.

This raises the second of the two issues I’d like to consider, but before we’ve finished with the first issue. Let’s just make a note of the fact that we have privileged one issue: accurate estimation of quantities represented on a graphic. It seems fair to do so. If all else is equal, we certainly prefer the technique leading to accurate estimation. Let’s return to this subject after we finish with the first issue.

The first issue with the normative perspective is that plenty of guidance is already available for exemplary practice. Several disciplines offer useful principles. They’ve been doing so for a long time. There are a few writings on graphical principles throughout history and a steady stream in the last hundred years. So what do the prescribers offer?

If we think about design as constrained choice, it’s easy to see many graphical aids as making choices for us. I use the term “graphical aid” loosely when I mention Powerpoint, but this is surely the ubiquitous contemporary “graphical aid”. It, more than any other system, insists that we violate graphical practice as old as Aristotle. The prescribers of InfoViz practice, above all else, rail against chartjunk, such as that created by default Powerpoint settings.

This is not new. Christopher Alexander, in Notes on the Synthesis of Form, 1964, sets forth a compelling example of Slovakian shawls. Their golden age as prized souvenirs in the nineteenth century ended with the introduction of a new technology, aniline dyes. Slovakian shawls, formerly prized as delicate and subtle, then became burdened by a reputation as vulgar and uninteresting. Alexander’s insight into the interruption of this art form was to see that the key gift of the shawlmakers was to be able to recognize a bad shawl in a group of good shawls. Individual bad shawls would appear in the tradition, but they would not be repeated and gradually a high standard developed and could be maintained. The tidal wave of available colors from aniline dyes simply overwhelmed this facility. We can imagine a normative approach to rescue the shawl-makers from new technology just as Tufte has rescued us from Powerpoint.

The second issue has to do with users, contexts, and communities. A recent normative book shows an example of good graphical practice as (a) using high saturation for objects and low saturation for backgrounds. The picture in (b) shows just how bad things are when we reverse these. The problem with this good / bad dichotomy can be seen in a painting by Miro (c) and by a stock photo of peppers. In both cases, the goal is to make the eye restless. In the normative view, tranquility is prized and we can’t simply say that one approach leads to tranquility without identifying that as good. The problem of users extends not only to different contexts, but to different users and to the social construction of the meaning of information artifacts.

Jasper Liu shared a learning experience with me that I’d like to pass on. You may recall that Jasper described some information for his Pure Visibility project and said he planned to depict it in three dimensions. He showed several pictures of labeled 3D axes, listing the information he planned to show as the axis labels. I objected to this on the basis that some of the information was nominal and some was ordinal (see Agresti, 2002), so their locations in three dimensions would introduce some ambiguity. Here’s a sample of the information, scores for terms in three search engines.

Jasper’s response was to first plot some data to demonstrate what he had in mind. This is a crucial step. It’s often the case that our mental picture of a planned representation glosses over some practical limitations. When we draw a concrete picture and supply it with data we often see game-changing details. Such was the case for Jasper’s data. Here’s how it looked, plotted as 3D data.

Jasper was dissatisfied with this picture and, significantly, looked at Card (1999), page 60, where the editors say “Static presentations often use retinal properties such as color to add an additional variable to a visual structure.” Reading this, he was apparently struck by the insight that he could use color to distinguish between search engines and greatly simplify the representation to the following picture. As we discuss the normative perspective, it might be helpful to look back at Jasper’s process.