Why You Should (Probably) Not be Doing Uncertainty Visualization

Ross T. Whitaker

The visualization of stochastic or uncertain data is typically referred to "uncertainty visualization". However, this terminology implied associated set of assumptions about the paradigm for visualization, which is typically to display an answer that has been modulated or augmented by an associated uncertainty. This however, asserts the existence of a renderable answer, which defies one of the underlying goals or principles of visualization, which is the exploration of data to obtain a holistic understanding or to discover properties that have no associated, a priori hypothesis. An alternative paradigm, is "variability visualization" where the goal of the visualization is to explore or better understand the set of possible outcomes, or the probability distribution, associated with a set of data. One example of such an approach is the method of contour boxplots, which relies on a generalization of data depth, from descriptive statistics, to render the variability of solutions in an ensemble of isocontours.