Embracing Uncertainty and Constraints
Knowledge is an unending adventure at the edge of uncertainty.
— JACOB BRONOWSKI
Some of the most important considerations about any visualization are the limits, constraints, and assumptions behind the data that drives what we see. The aesthetic purity of a graphical representation can belie the fact that what it’s conveying doesn’t really represent the truth. The power of the visual image in general — and the credibility we might unconsciously invest in polished charts and graphs — can overshadow the fact that the underlying data or assumptions are incorrect. An essential but often easily overlooked question is: what about any potentially problematic data behind the chart? How do we address that? The data might not be bad in and of itself. It might be meticulously collected and categorized but still be problematic. Can the graphical perfection of a visualization be counterproductive for the purposes of knowledge, discovery, or even simply communicating an idea?
In neurobiology and cosmology, grappling with uncertainty in measurements, assumptions, and knowledge is a fundamental concern. This uncertainty can come in many forms and combinations, including various kinds of gaps in the data, limitations of statistical approaches, artifacts from the tools and methods of collection, and other issues. Visually representing uncertainty is another crucial element for helping people ask the right questions and draw the most accurate conclusions about the data. But showing uncertainty requires visual techniques that can jostle with the other encoding of data dimensions. In the visualization, what represents the data and what represents the level of uncertainty about them?[—]
We can solve some of the practical problems of displaying uncertainty by using interactivity. For example, we can provide sliders or buttons so that users can select the amount of uncertainty for a given visualization. For clarity purposes, they might begin with an idealized presentation and then adjust the controls to see how the picture changes when various statistical factors, sample sizes, or competing assumptions are introduced into the mix.