Researchers from Harvard and MIT have teamed up to address an important question: what makes a data visualization memorable? The conventional opinion is that it’s easy to identify “bad” data visualization: tacky renderings with too much text, excessive ornamentation, distracting colors, and kitschy clip art.
Top twelve most memorable visualizations from the experiment (Image courtesy of Michelle Borkin, Harvard SEAS.)
Design expert Edward Tufte refers to these pieces as “chart junk” classifying them as redundant at best, and useless at worst. The visualization community, however, is divided. Some say these seemingly extraneous elements actually serve a purpose by creating a lasting impression in the viewer’s mind.
The debate over “chart junk” became the impetus for a scientific study, which was then documented in a research paper by computer scientists at Harvard and cognitive scientists at MIT. These experts of design call into question the usefulness of a perfectly-executed graphic that hardly anyone remembers. They conclude that the very design elements that attract so much criticism can also make a visualization more memorable.
The authors write that “knowing what makes a visualization memorable is a step towards answering higher level questions like ‘What makes a visualization engaging’ or ‘What makes a visualization effective?'”
Results of this study were presented on October 15 at the IEEE Information Visualization (InfoVis) conference in Atlanta, Georgia. The work was also highlighted on Harvard’s School of Engineering and Applied Sciences website.
For lead author Michelle Borkin, a doctoral student at the Harvard School of Engineering and Applied Sciences (SEAS), memorability is a key metric. “I spend a lot of my time reading these scientific papers, so I have to wonder, when I walk away from my desk, what am I going to remember?” she says. “Which of the figures and visualizations in these publications are going to stick with me?”
Borkin and her team performed the largest-scale visualization study of its kind, collecting 5,693 visualizations, categorized by visualization type (e.g., bar chart, line graph, etc.), from news media sites, government reports, scientific journals, and infographic sources. After eliminating multiple images (i.e., ones that were grouped rather than stand-alone) the initial pool was winnowed to 2,070 single-panel visualizations. A further subset of 410 images were selected as “target” visualizations. Each of these was annotated with additional attributes, including ratings for data-ink ratios and visual densities.
The experiment was set up as a game on Amazon’s Mechanical Turk, which compensates participants, called workers, for performing HITs (“Human Intelligence Task”). Workers were presented with a sequence of images and asked to press a key if they saw an image for the second time in the sequence. At the end of the testing, each image was given a memorability score. What the researchers discovered was that observers are consistent in which visualizations are most memorable and which are most forgettable.
Out of the 410 target images, 145 contained either photographs or cartoons, humanly recognizable objects, which the scientists refer to as pictograms. The study showed that visualizations that used pictograms had on average higher memorability scores.
Borkin’s adviser, Hanspeter Pfister, a Wang Professor of Computer Science at Harvard SEAS, adds this commentary: “A visualization will be instantly and overwhelmingly more memorable if it incorporates an image of a human-recognizable object – if it includes a photograph, people, cartoons, logos – any component that is not just an abstract data visualization,” she says. “We learned that any time you have a graphic with one of those components, that’s the most dominant thing that affects the memorability.”
Visualizations that were more dense or used more color also had higher memorability scores, but other results proved a bit more surprising:
“You’d think the types of charts you’d remember best are the ones you learned in school – the bar charts, pie charts, scatter plots, and so on,” Borkin says. “But it was the opposite.” Charts with more unusual shapes – tree diagrams, network diagrams, grid matrices and such – were actually more memorable.
Audra Oliva, a principal research scientist at MIT’s Computer Science and Artificial Intelligence Lab, has been studying visual memory for about six years now. Research performed by her team demonstrates that human memory responds better to human-centric images rather than landscapes.
Without this similarity across human responses, asking what makes an image or visualization more memorable than another would be pointless.
“All of us are sensitive to the same kinds of images, and we forget the same kind as well,” Oliva says. “We like to believe our memories are unique, that they’re like the soul of a person, but in certain situations it’s as if we have the same algorithm in our heads that is going to be sensitive to a particular type of image. So when you find a result like this in photographs, you want to know: is it generalizable to many types of materials – words, sound, images, graphs?”
The scientists who performed the study are excited about the potential for advancing the science of visualization, but they are also quick to point out that memorability is just one parameter. Accuracy is always the highest priority with the best visualizations also being easy to comprehend, engaging and aesthetically-pleasing. But there’s no reason they can’t also be memorable.