Evaluating Encodings for Bivariate Edges in Adjacency Matrices
A new study with 156 participants reveals which visual designs work best for showing two data points per edge.
A team of researchers from the University of Utah and Harvard has published a landmark study in Computer Graphics Forum, providing the first empirical evidence for how to best visualize complex network data. The paper, "Evaluating Encodings for Bivariate Edges in Adjacency Matrices," tackles a common problem in fields like biology, finance, and social network analysis: how to clearly display two statistical values (like a mean and a standard deviation) for each connection in a network graph, all within the compact grid of an adjacency matrix.
In a rigorous, preregistered study with 156 participants, the team tested four candidate visual encodings: a bivariate color palette, embedded micro bar charts, and two types of overlaid marks that map one attribute to color and the other to either area or angle. Participants performed eight distinct analytical tasks, from comparing central tendencies to identifying outliers. The results established clear performance tiers: designs using area-based overlaid marks and embedded bar charts delivered the highest accuracy and speed.
The study's findings offer concrete, evidence-based design rules for developers and data scientists. The consistently poor performance of bivariate color schemes—a common intuitive choice—is a particularly valuable insight, warning against their use for precise analytical work. This research moves network visualization from artistic preference to a science of perception, enabling tools that help professionals accurately interpret multidimensional relationships in their data.
- Area-based overlaid marks and embedded bar charts achieved the highest performance in user testing.
- Bivariate color palettes, a common intuitive choice, consistently underperformed all other encoding methods.
- The preregistered study involved 156 participants performing eight distinct analytical tasks on the visualizations.
Why It Matters
Provides evidence-based design rules for creating clearer, more effective data visualization tools in science and business analytics.