I recently had a discussion with some colleagues about high levels of visual complexity in data visualizations. It started around a few questions:
“Can a graph (in the network diagram sense) have axes?”
“Can a network visualization represent qualitative and/or quantitative information?”
“Is it still a network visualization if it includes other components outside of vertices and edges?”
Some quick notes on these: The former uses the axes as a coordinate system to identify nodes without overlaying labels on an already dense layout. (I actually quite like this method, though it would probably be even more effective if it were interactive, with mouse-over identification, etc.) It is also important to point out that it does not imply additional qualitative or quantitative properties, other than a simple label. The latter is technically a Path Graph, who’s qualities are constrained much more than an average network graph. So these are, in some ways, exceptions, though good fodder for the conversation.
Those questions continue to be debated and I think they are good ones. Our conversation turned at one point into a question of whether visualizations with a large amount of complexity, or unconventional visual structures, are really worth using.
One of the problems of standards used for evaluating visualizations is that there is an assumed single audience. I agree, if you are creating a visualization for a single audience (i.e. “everyone everywhere that might see this”) having to explain the way the visual mechanisms work in long form is kind of ridiculous, and not very effective (the bounce rate probably being very high upon even seeing that a paragraph exists, never mind after the first few sentences).
However there are different tiers of audiences for any given visualization design. High-level might be fine when the structure/point is obvious, and further detail (whether it’s just there, or that you can enable through interactivity) might be useful to better understand why the higher level exists as it does. Maybe there’s an even more refined analytical level or series of levels beyond that. (The raw data would be the basement floor for any visualization.) There are certainly ways to provide different sets of information depending on the interest-level of the viewer.
But also, I think too many visualizations are discarded as too complex or opaque when they might be extremely useful to a small set of people. I don’t think we should write off a particular form because the general public can’t make sense of it in a short amount of time. These debates typically take place in public, but a visualization should be valued by taking into account who the audience is, what the utility of the visualization is, the qualities and nature of what is represented, and how broad or narrow the underlying data and phenomena are. An extremely technical visualization that would boggle the mind of the unfamiliar may have extreme utility and efficiency for someone intimately familiar with the task(s) and data at hand.
What do you think are other criteria to evaluate data visualizations, and how does one provide multi-faceted utility in a single form?