Visualization is Growing Up
Several topics at this year's VisWeek conference have come up because visualization is playing a bigger role in important decisions. When the consequences can be severe, it is important to know whether a visualization actually works, whether we can trust it, and what biases it might present.
Verification and Validation
While the definitions differ somewhat, the common understanding of verification is that it means checking whether a system is working correctly, whereas validation asks whether the system performs the function it is supposed to.
Neither question is usually addressed in visualization research. One example that was brought up in the discussion on the panel on this topic at VisWeek are some implementations of Marching Cubes that have a bug that leads to small creases and cuts in the isosurfaces produced, which is obviously a problem.
User studies theoretically provide a way of validating visualization software, but are almost never done in that way in practice. It is often not entirely clear what the correct answer to a question is (or if there even is one), and studies are not typically rigorous and thorough enough to be called validations. Showing that a new system works better than an existing one is often sufficient for publication, but validation needs to be much more specific.
All of these things are clearly important, though, if we expect people to trust visualization to make decisions. As long as we cannot demonstrate that our systems perform the intended function and are working correctly, we cannot expect to be taken as seriously as other, more rigorous fields.
When Rainbow Colormaps Kill
Michelle Borkin presented fascinating work on the use of colormaps in the diagnosis of coronary heart disease. Their 2D representations using perceptually uniform colormaps led to significantly fewer errors than the usual 3D representations and/or the dreaded rainbow colormap.
When the choice of visualization type and parameters makes such a big difference, we need a much better understanding of the potential consequences before we can recommend the use of visualization for important decisions. Similar results have been reported when making decisions about continuing or aborting clinical studies, and Caroline Ziemkiewicz's work has also shown the impact of surface features on people's perception of data.
Visualization and Policy-Making
Another panel that caught my attention was the one on visualization and policy-making. The panelist...
0 comments:
Post a Comment