Tuesday, January 29, 2013

Intuitiveness vs. learnability in interactive visualization

We're on the third week of my second Introduction to Infographics and Visualization MOOC, which has around 5,000 participants. We used the first two weeks to discuss several examples of good and just-so-so graphics. One of them (of the interesting ones, I mean) was Moritz Stefaner's series on Information Flows in Science. Among the hundreds of reviews that I have read, the one by Evan Sheehan —a UX designer with a clear interest in visualization; don't miss his website and follow him on Twitter— got me really excited. With is permission, I am reproducing it verbatim, as it addresses many topics I'm interested in, such as the balance between novelty and tradition. Read it below.

"I hesitate to accuse these graphics of not being intuitive. Personally, I don't think intuitiveness is an achievable goal since every individual will have developed slightly different intuitions over the course of their lives. Instead, I strive for learnability in my designs. There are two components to learnability: recognizing that you don't understand something; developing the correct understanding.

I think it was very easy with these not to realize that you didn't understand the graphics completely; I certainly didn't. I thought the clustering was a library classification. This assumption was reinforced by the first graphic which showed me several medical journals grouped together, math and computer science, etc. I rationalized the anomalies away because the actual grouping mostly fit the mental model I had already assumed.

The second graphic made it easier for me to recognize that I didn't fully understand the underlying model, because the mental model I had developed didn't accommodate the notion that a journal could change its cluster. Unfortunately, the design also failed to adequately provide the tools I needed to develop a correct understanding now that I had finally realized that I needed to. The information required to fully understand the models underlying the graphics were buried in papers only linked from prose in the introduction paragraphs to each visualization. I think these descriptions are important enough to warrant explanation as part of the graphics, or at the very least, greater visual weight in the presentation so they were more noticeable.

I think a lot of people tend to think that more common charts like scatterplots and bar charts are more intuitive because you already know how to read them, but there was a time when you didn't. You just learned to read them so long ago that it now seems obvious. Admittedly some—like a bar chart—are really quite simple, so learning to read them is not difficult, but I don't feel I should limit myself to monosyllabic words just because they're easier to learn.

I think it's great when someone introduces new vocabulary into the visual lexicon, but when we do so, we should be careful to provide resources that help people come to an understanding of the new vocabulary. Just because you have to study a graphic to interpret it, doesn't mean it's poorly designed. The question is: after you studied it, were you able to understand it, and was there anything that could have made your studies easier? I think these graphics failed not in their representation of the data, but in their support of viewers learning to interpret the data models. Which seems like I'm splitting hairs, but I really do think that once I grokked both the eigenfactor and the clustering algorithm, the visualizations were easier to read."