Visualizing Uncertainty and Making Comparisons on Maps Using Simple Uncertainty Classes
Daniel B. Carr, (George Mason University), firstname.lastname@example.org
Analysts in some settings have remained reluctant to provide numerical values for uncertainty. The considerations of source credibility, the absence of data and other factors can make uncertainty assessment daunting.
Given numerical values or intervals for uncertainty there are numerous possible visual encoding. These include confidence intervals about estimates, color, translucence, texture, and partitioning. Icons with labels from the National Weather Service illustrate the need for careful work in the visual encoding process. There are issues to address for each alternative representation.
This talk focuses on illustrations using dynamic shareware developed to show three variables for either arcs in network or regions on a map. The arcs can be tangible such as road or stream segments, or abstract such as in social networks. Each of the three selected variables is attached to its own two-value slider. One or more of the selected variables can be estimates of uncertainty. The sliders allow the data analyst to dynamically set the thresholds defining low, middle and high classes. One slider controls the color for a region (or arc). The two remaining sliders define the rows and columns in a 3 x 3 grid of maps in which regions are highlighted. This design conveying for three variables and geospatial coordinates is cognitively simple and enables comparison.
The examples shown are based on publicly available data. The concepts discussed are general in nature and easily applicable to defense and national security contexts. A possible application of these methods could address the security status of road segments or regions in a city. The uncertainty of road/region security status (or length of time since a status update) and an assessment of risk can be represented and studied simultaneously. The real time response of shifting thresholds combined with capabilities such as zooming and recording of annotated views help the analyst to see and document the implications in the geospatial context.
Perhaps the reluctance to provide numerical values for uncertainty would be diminished when the interpretation is reduced to simple classes of low, middle and high and the numerical values are not tied to an absolute scale.