i
i
i
i
i
i
i
i
27.4. Visual Encoding Principles 687
In both the discrete and continuous cases, colormaps should take into account
whether the data is sequential or diverging. The ColorBrewer application (www.
colorbrewer.org)is an excellent resource for colormapconstruction(Brewer, 1999).
Another important issue when encoding with color is that a significant fraction
of the population, roughly 10% of men, is red-green color deficient. If a coding
using red and green is chosen because of conventions in the target domain, re-
dundantly coding lightness or saturation in addition to hue is wise. Tools such as
the web site http://www.vischeck.com should be used to check whether a color
scheme is distinguishable to people with color deficient vision.
27.4.3 2D vs. 3D Spatial Layouts
The question of whether to use two or three channels for spatial position has been
extensively studied. When computer-based visualization began in the late 1980s,
and interactive 3D graphics was a new capability, there was a lot of enthusiasm
for 3D representations. As the field matured, researchers began to understand the
costs of 3D approaches when used for abstract datasets (Ware, 2001).
Occlusion, where some parts of the dataset are hidden behind others, is a
major problem with 3D. Although hidden surface removal algorithms such as Z-
buffers and BSP trees allow fast computation of a correct 2D image, people must
still synthesize many of these images into an internal mental map. When peo-
ple look at realistic scenes made from familiar objects, usually they can quickly
understand what they see. However, when they see an unfamiliar dataset, where
a chosen visual encoding maps abstract dimensions into spatial positions, under-
standing the details of its 3D structure can be challenging even when they can use
interactive navigation controls to change their 3D viewpoint. The reason is once
again the limited capacity of human working memory (Plumlee & Ware, 2006).
Another problem with 3D is perspective distortion. Although real-world ob-
jects do indeed appear smaller when they are further from our eyes, foreshorten-
ing makes direct comparison of object heights difficult (Tory et al., 2006). Once
again, although we can often judge the heights of familiar objects in the real world
based on past experience, we cannot necessarily do so with completely abstract
data that has a visual encoding where the height conveys meaning. For exam-
ple, it is more difficult to judge bar heights in a 3D bar chart than in multiple
horizontally aligned 2D bar charts.
Another problem with unconstrained 3D representations is that text at arbi-
trary orientations in 3D space is far more difficult to read than text aligned in the
2D image plane (Grossman et al., 2007).
Figure 27.10 illustrates how carefully chosen 2D views of an abstract dataset
can avoid the problems with occlusion and perspective distortion inherent in 3D