When you hear the term “visualization research,” what comes to mind? Many people think of Tufte when they think about deep reflection on visualization. For example, ask any visualization researcher how many times their description of what they do has been met with a comment about how great Tufte’s books are. We agree, Tufte’s books are great resources! His guidelines, like maximizing the data-ink ratio or avoiding chart junk, are helpful maxims when you are starting to design a visualization and become aware of the large space of possibilities, even to show a simple data set. Yet Tufte’s suggestions can break down in many realistic design scenarios. Have you ever tried to follow Tufte’s advice to a T? You might end up with something like the “ghost” chart to the right. Is it really superior to the original chart on the left?
What does visualization research actually cover?
Understanding of the limits of different principles and guidelines for creating effective visualizations is one goal of visualization research. Started in the late 1980’s by computer scientists and others who were inspired by the possibilities for using interactive graphics to amplify cognition about data, visualization research is now a thriving community with a conference (IEEE VIS) that draws ~1200 people annually. Visualization researchers work on a number of problems that can help inform how anyone understands, designs, or evaluates visual representations of data:
Tools that make it easier to create visualizations Have you heard of Tableau Software, Spotfire, or D3? All three tools were proposed and developed by visualization researchers. A common research goal in the visualization community is to develop technology that makes it easier to create effective visualizations, so as to increase the amount of value that people can get from data even without advanced training in design or statistics. A recent trend toward declarative specification of visualizations, as realized in the Vega and Vega-lite grammars, makes specifying a visualization more concise and can reduce the amount of code one must write to develop visualization tools.
New encodings Visualization research produced encodings that are now commonplace in visualization tools, like treemaps, word trees, parallel sets or arc diagrams. Visualization research often considers the limits of conventional methods for representing data and how novel designs might surface new and interesting patterns in our data.
Knowledge from controlled studies of visualization effectiveness Controlled experiments can help establish where maxims don’t work (e.g., chartjunk may help people remember what they saw! Binning 2D continuous data isn’t that bad!), which encodings perform best for different tasks, and how perceptions of what visualized data means or how accurate it is are impacted by different encodings, framings, or user predispositions or prior beliefs, for example.
Supporting visual analytics Often visualization use in the world is motivated by data analysis. Visualizations are used in exploratory data analysis, such as making sense of very large datasets or text corpora, as well as for evaluating complex models. Visualization research develops tools to scaffold the exploratory analysis process and deepens our understanding of how visualizations support meaning-making about data. A recent trend in the community involves understanding how visualizations can facilitate interpretability of deep learning models.
Visualization-based communication Perhaps even more common than strictly analytical uses of visualization are those for communicating data, such as to the public or to decision makers. While most visualization researchers shy away from the idea of “infographics” based on their reputation for low data content, a number of researchers are working on topics related to storytelling and presentation, from understanding how to order a set of visualizations in a presentation to producing tools for drag-and-drop authoring of narrative visualizations, creation of “bespoke” chart layouts, or seemingly hand-drawn storylines, for example.
Studies of visualization use in the world Sometimes called design studies, some visualization research projects involve close collaboration with domain experts who can benefit from visualization in their work. These studies often produce new insights about problems faced in a domain and new encodings or interactive systems to address these problems.
Formal theories of visualization Visualization researchers have proposed formal languages that give rise to the space of possible visualizations, most notably Wilkinson’s Grammar of Graphics, which in turn inspired R’s popular ggplot2 package, as well as algebraic, economic, and other theories of effective visualization.
What should visualization research be about?
At base, visualization is a method for contextualizing data, enabling people to apply their prior experiences and perceptual and cognitive abilities to draw conclusions about phenomena in the real world. If you stop and think about it, this is a pretty ambitious target! Visualization research covers an impressive breadth of topics from perception to memorability, from complex system design to theory about what comprises a graph. But as in all research communities, biases what we think a research contribution should look like in our field can limit the types of questions we consider worth pursuing. For example, while research on perception has played a role in much of the visualization research field’s history, we have not necessarily embraced cognition. Perhaps we are more comfortable with the “bottom-up” nature of perception because its more clearly tied to the visual encodings. But what people take away from a visualization arises from “top-down” forces like what a person wants to believe, what they’ve learned about graphs in the past, or what they know about the domain as much as from patterns emerging as the visual system does its work.
A focus on performance keeps visualization research relevant to the world, where people want to know which chart to use. But, that shouldn’t absolve researchers from trying to explain why a difference was found. This might take the form of proposing and testing for mechanisms in the visual system or cognitive strategies like heuristics. Considering the “why” in addition to the “what” can make it easier to reason about how a difference found between two specific visualizations might also be found under slightly different conditions.
Finally, we often draw a sharp line around representations of abstract data, considering other forms of diagrams or ways of contextualizing data to make it understandable outside our purview. But it is really productive to consider topics like interactive illustration , or satellite imagery, or sketching and analogical reasoning, or understanding aesthetics independent of analytical utility as not “core visualization” enough? Many people think of tables as “not visualization”, but is this distinction useful? From a research perspective, thinking about visualization with less sharp boundaries might allow us to keep our field (and flagship conference!) relevant and exciting to the many people thinking about data presentation in the world. From a practical perspective, thinking of visualization as just one way of contextualizing data may help us realize when a task might be better served with a representation that doesn’t involve mapping data to visual encodings. When visualization is the only answer we see, we fail to acknowledge that sometimes information is better expressed in text or not at all.
Tell us know what you want to know!
One of our goals in starting this blog is to help communicate that yes, visualization research is a thing, and many of the problems it addresses have relevance in everyday use of visualizations! What would you like to know more about? Write us at firstname.lastname@example.org or comment on this post to tell us what you’re interested in!
Written by Jessica, with input from Danielle, Enrico, and Robert.