Overview and State-of-the-Art of Uncertainty Visualization - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 6

ent component models may make different assumptions or use different parameters,
illustrate the potential variability in even the best models.
The output from such a model may include information about estimated error
in the form of a single error measure, ranges for expected values, or predicted dis-
tributions for values or errors. These measures are applicable to numeric quanti-
ties. Alternatively, a model that makes nominal or categorical predictions may also
indicate the degree of confidence in its predictions by producing multi-value predic-
tions, where each possible value or classification is associated with a likelihood.
1.1.1.3 Uncertainty from the Visualization Process
Finally, we should understand how the visualization process impacts the propaga-
tion, magnification, perception, and impact of uncertainty. In order to do this, we
must understand computational sources and magnifiers of error and uncertainty in
input values, perceptual and cognitive influences on the understanding of uncertainty
visualization, effects of differences in audience abilities and cultures, requirements
imposed by different application tasks and goals, and competing positive and negative
consequences of showing uncertainty.
1.2 Perceptual Uncertainty
Logically, it seems sensible to display information about uncertainty in a manner
consistent with our cognitivemodels of which perceptual elements contain variability
or uncertainty. A number of approaches to uncertainty visualization seem to build
on this principle, representing uncertainty with such visual elements as blur, flicker,
reduced saturation, sketched outlines, or transparency.
There have been relatively few careful evaluations of the effectiveness of uncer-
tainty visualization and its impact on the decision-making process that have appeared
in the visualization literature. In some cases, researchers have used quantitative eval-
uations or user studies to evaluate the ability of subjects to understand uncertain infor-
mation [ 33 , 109 ]. Zuk and Carpendale [ 111 ] present a framework for the heuristic
evaluation of uncertainty visualizations from the perceptual and cognitive principles
described by Bertin [ 6 ], Tufte [ 101 ], andWare [ 105 ]. They use this framework to ana-
lyze eight uncertainty visualizations of different types and from different domains.
They propose this sort of heuristic evaluation as a rough substitutewhenmore specific
evaluations are not practical.
Additional insight into the perceptual and cognitive elements of effective
uncertainty representations can be found in the GIS literature. Harrower surveys
a collection of evaluations of methods for representing uncertainty in map-based
visualizations [ 38 ]. He observes that the most common characteristics used to judge
a technique are its effects on confidence, speed, and accuracy of judgements. Two
principles which may be derived from that set of evaluations are the superiority of