Registered: 1337363986 Posts: 7
Reply with quote #1
I just finished reading
Data visualization: ambiguity as a fellow traveler which appears in Nature. I was hoping you, the community, might read this article and share your thoughts. The article begins with the idea that uncertainty can be hard to visualize outside conventional methods. However, I don't think the article necessarily supports this point. Instead, the article surveys several products that, to me, address the problem of visualizing uncertainty, through varying data points hues and by employing heatmaps--conventional stuff. If anything, it seems the root of the problem is the existence of uncertainty, not that we lack ways to visualize it. For me, the article appears uncertain about what "uncertainty" even means. Does it refer to probability distributions? Does it refer to the potential for inadvertently injecting error into a raw dataset? Does it refer to a lack of quality data? I see why uncertainty might appear hard to visualize if it means so many different things to different people. In the business world, risk and uncertainty are well defined. I actually came to this article from a post on Flowing Data, specifically Nathan Yau writes: This is essentially the same premise of the article, which opens with:
I still struggle with uncertainty and visualization. I haven't seen many worthwhile solutions other than the old standbys, boxplots and histograms, which show distributions. But how many people understand spread, skew, etc? It's a small proportion, which poses an interesting challenge.
I've always thought box and whisker plots were rather easy to understand. It's true that many are unfamiliar with spread and skew, but they aren't difficult concepts to tackle. The source of the latter quote is Martin Krzywinski who developed a software visualization tool called Circos. Nature didn't reproduce a picture of his tool for their article, but you can find his work,
Error bars are also an option, but it is difficult to convey information clearly with them, he says. “It's likely that if something as simple as error bars is misunderstood, anything more complex will be too,” Krzywinski says.
here. Near the end of the article, he says,
Finding missing data offers rewards for researchers working on visualization methods. Perhaps they may even find surprises, which could lead to a new kind of uncertainty metric. “One uncertainty metric might be 'How many published algorithms would disagree?'” says Krzywinski. “In biology, many!” A measure of uncertainty does not have to be a negative factor: it could be a measure more like a “surprise factor,” he says. The less likely something is to happen, the greater the level of surprise. “We want to make sure that we're paying attention to the surprising results,” he says, because they can be important indicators in experimental findings. I think most researchers have a good idea whether their results are consistent or surprising - and whether that's good or bad thing. But here again what uncertainty means hasn't been nailed down. Likelihood is a measure of probability, for instance. Is something that is less likely to happen also less certain to happen? Such a metric sounds absurd to me. So I'm wondering your thoughts on the article. As well, I'm wondering if you find the idea of uncertainty hard to visualize? __________________ Jordan Goldmeier
Registered: 1135986598 Posts: 838
Reply with quote #2
I think your observation is astute. The fundamental problem that we face is not an inability to represent uncertainty visually, but the confusion that people have about uncertainty. What does it mean? (Unfortunately, the term is used to mean various things, often without an awareness that this is so.) For what purpose do we wish to view and understand uncertainty? How is it used? Developing better ways to visualize uncertainty when those who view and attempt to use the information don't understand it will only lead to greater confusion. __________________ Stephen Few