Visualising uncertainty is often vital to the accuracy of a visualisation, as it allows viewers to understand the relative fuzziness of certain factors of the model. Possible approaches to visualising uncertainty include transformations such as blurring visual entities, or introducing a way of displaying multiple possible predictions of a model, similarly to how a user might choose to highlight a certain path or visual element.
As an example, Harris introduces the concept of fuzzygrams, in which “columns [of a histogram] are replaced by blurred or fuzzy areas to indicate the probability of the values plotted. The degree of fuzziness is inversely proportional to the sample size”.
When developing the visual vocabulary for TREsPASS we composed a set of symbols or graphics that function as building elements for describing larger, more complex visual entities. A strong visual language forms an important basis for representing security models as it provides a suitable mapping from the model’s language to the visual vocabulary. This vocabulary should also be extensible, allowing one to highlight and visualise particular parts of interest, including uncertainty. Although uncertainty may seem to go against highlighting important aspects in a security model, both are critical for the understanding and evaluation of information, and should be developed in concert with the remainder of the vocabulary. Uncertainty also has different interpretations, for instance it can mean how uncertain future events will be, but also how certain one is of the data that is represented. The visual language should be able to be useable for both instances.
Alex Krusz describes it clearly:
“Traditionally, uncertainty is conveyed graphically with error bars denoting the top and bottom of a ‘significant region’, generally the central 95% of the probability distribution of the data. The shortcomings of this are that the finer details about the probability distribution are lost, and the significance cutoff value is arbitrary. Other options like box-and-whisker plots are slightly more informative, but are visually clunkier and still suffer from the same problems. The most common depiction of a probability distribution is as a curve, but this requires an extra graphical dimension and would add clutter. We can accurately depict a distribution without adding dimensions by rendering it as a cloud. Conceptually, we’ll plot a horizontal bar that spans each height where the distribution has a non-zero value, and shade it in proportion to the probability density. In practice, this creates a rectangle whose opacity varies with height.”