Using precision as guidance for visualization design is powerful and yet limited in many different ways. Expressiveness may help.
Visualization is about mapping data values to visual channels;
While precision is a useful factor to keep in mind, it’s neither sufficient nor necessary to create effective visualizations.
First, it’s not always true that visualizations that use position are “better” than those that do not. Second, position can be expressed in multiple ways, so the guidelines leave us unable to discern between visualizations that use position differently.
The problem with guidelines based on precision is that visualization is not really about precision.
But visualization is less about precision, and much more about what the visual representation expresses.
data contains information that we want to communicate, and that information is what we want to “express” with visual representations
Good visualizations stem from a good matching between what we want to express and what the visual representation expresses (something highlighted in the foundational work of Jock Mackinlay, who I believe introduced the term in the first place).
Expressiveness is much more about finding a good match between visual properties and “concepts” than precision or accuracy.
diverging color scale is not a matter of precision or accuracy (in fact a diverging color scale may even be less precise!), it’s a matter of good semantic matching; a matter of expressiveness.
The book divides visual channels into two large groups, magnitude channels and identity channels, according to whether they are more appropriate for quantitative or categorical information. For example, if I want to express a quantity, I shouldn’t use color hue, because humans do not perceive hues as ordered). And if I want to express categories, I shouldn’t use symbol size, because we naturally associate sizes with quantities, not categories.
Let’s look at other things one may want to express. Without any pretense to be exhaustive, here are some examples off the top of my head:
Directionality: can the values be positive or negative? Is there a zero value or a meaningful threshold?
Part-to-whole: do the data objects represent the part of a whole?
Order: are the objects organized in a meaningful order?
Grouping: are the objects organized around a set of meaningful groups
Space/time: do the objects represent space or time or space-time phenomena?
But position expresses information in so many other important ways! Think about position on a single axis versus an orthogonal pair of axes; position in polar coordinates; position on a map; position to express grouping and alignment; etc.
Pursuing expressiveness may ultimately lead to inelegant or irrelevant findings, who knows. But it may also open the doors to a much richer description of how visualization works and to much better guidance.
we have to keep in mind that expressiveness is not only useful in figuring out how to express the information and ideas we want to communicate, but it is equally useful in making sure we do NOT accidentally convey information we did not intend to express.
We can work towards having a better and more complete characterization of what information and concepts we typically want to communicate with data and, similarly, develop a better understanding of what visual properties exist and what they express.
The more granular we go, the harder it is to generalize. In addition to that, the impulse to be able to describe every possible situation may backfire by making things more complicated than they need be.
Once we have a better description of data concepts and visual features, how do we build guidelines that can help people think more productively about visualization design?
Finally, if we want to approach this problem scientifically, we will also need to verify that guidelines and models based on expressiveness have a measurable impact on visualization effectiveness. For instance, it would be useful to explore if, among two competing visual representations, the one having a better match in terms of expressiveness leads to a difference in comprehension (which is not easy to measure, but that’s a different story).
We do not visualize data, we visualize ideas based on data, and we need a better mental model (and maybe even a theory) to design visualizations.
