What is the intrinsic number of dimensions (not necessarily spatial) in your data, and how many do you want to show at once?

Most people immediately think of spatial dimensions when they hear “2D” or “3D,” but a statistical view treats a “dimension” as any one of many measured properties (or ‘attributes’) in a data set. For a simple data format like a table, a “dimension” can be thought of as any column of the table. A good mantra to keep in mind when choosing how to visualize high-dimensional data  is “Data-Dimensions-Display,” in that understanding both the intrinsic dimensionality of your data and the dimensionality made available by various kinds of visualizations and tangible and online display modes (Question 10: Display Modes) all interact. To make a bad pun, think outside the box(es).

Example: Arterial Visualization
The two-paneled figure below, from Borkin et al. 2011’s work on Artery Visualization, shows the value of presenting 3D data in 2D for some purposes. Surgeons can benefit from the realistic 3D view of the arteries (left) when operating, but in Borkin’s user study, the 2D view at right was better for identifying trouble spots according to a measure of “Shear Stress” shown in the color bar. Borkin’s 2011 paper shows that not only is the accuracy of diagnosis greatly improved with the 2D view at right, the time to correct diagnosis is also dramatically faster.

For the best diagnostic+surgical combination, Borkin’s study and others find that linked views of the 2D and 3D data work best (as discussed in Question 4: Patterns) , especially when viewers can control the colors and viewing angle.

Source: Borkin, M. A., Gajos, K. Z., Peters, A., Mitsouras, D., Melchionna, S., Rybicki, F. J., … Pfister, H. (2011). Evaluation of Artery Visualizations for Heart Disease Diagnosis. IEEE Transactions on Visualization and Computer Graphics, 17(12), 2479–88. http://doi.org/10.1109/TVCG.2011.192

Last revised: 12th of July 2018, Alyssa Goodman & Arzu Çöltekin