Robert Kosara is a Visual Analysis Researcher at Tableau Software, and formerly Associate Professor of Computer Science at UNC Charlotte. He has created visualization techniques like Parallel Sets and performed research into the perceptual and cognitive basics of visualization. Recently, Robert’s research has focused on how to communicate data using tools from visualization, and how storytelling can be adapted to incorporate data, interaction, and visualization.
Robert received his M.Sc. and Ph.D. degrees in computer science from Vienna University of Technology (Vienna, Austria). His list of publications can be found online on his vanity website. He can be found on Twitter, Facebook, LinkedIn, Google+ and Google Scholar.
EagerEyes is Robert Kosara’s place to reflect on the world of information visualization and visual communication of data. The goal is to help digest things that are happening in the field and discuss developments that may be tangential or early, but that are likely to have an impact.
The original idea for the site involved the interplay of art and science in visualization. While the focus has shifted, questions of representation are touched upon regularly. In fact, Robert believes that visualization can be vastly improved by a better understanding issues of representation and reading of data.
Other topics of interest include visualization for the masses, open data, and where the field of visualization is heading. Criticism of visualization techniques and applications, websites, and books is also a regular feature. Discussions of visualization techniques provide insights into the thinking behind them. Around important conferences like VisWeek, the site is also used for updates and pointers about things that are going on there.
Robert points out that this is not a blog. Blogs tend to aim for quick, current commentary. The articles on this website are meant to be of value over a longer time period (except for the ones in the blog category), and are usually much longer than the typical blog posting.
The Bikini Chart
The Obama administration released a chart a while ago that shows job losses during the last year of the Bush administration and the first year after Obama took office. The chart is simple yet effective in the way it communicates a message. It also has some very subtle design elements that communicate a much more negative undertone than is immediately obvious.
I have to say that I have admired this chart since the day it came out. It is clean with just the right amount of decoration to work: scales and legends that explain what we are seeing. The colors are based on the typical colors associated with the Republican Party (red) and the Democrats (blue). The data is also indisputable, coming from the Bureau of Labor Statistics.
The chart shows the number of jobs lost per month over about two years, ending in early 2010. The message is clear: things were getting worse under Bush but have been getting better under Obama. It doesn’t take a lot of skepticism or knowledge of politics to know that things don’t happen that quickly, but the message still comes across quite clearly. (Click image for larger version)
It is interesting that they chose to use bars that are pointing down rather than up. In a way, that makes sense: negative numbers typically are represented by bars that point down. But the number of people who lost their jobs is not negative, it’s only negative if you look at it as “negative job growth.” This was clearly a conscious decision. Since almost all the numbers are negative, it might have still made sense to show them pointing up though, to make the chart look less unusual. Its shape has earned the chart the nickname bikini chart, though.
But the downward-pointing bars communicate something beyond the values: there is something wrong here, these bars should not be pointing down. While longer bars are often better (more income, more votes, etc.), this is not the case here. This choice of direction for the bars explains what the viewer should be looking for.
The inverted version of the chart below shows why bars pointing up would have been much less clear: the shorter bars under Obama look like something is decreasing, which is surely is not a good thing, right?
All of these are good choices and make the chart both attractive and effective. This chart is one of the cleanest examples of political communication I know, and it is based on actual, real data – imagine that!
But there is also something devious going on here. The choice of colors is the only logical one given the political context, but there is more to it. The red is quite a bit darker than the blue. That is not a bad choice in principle, since it makes it easier to tell the colors apart when the difference is not only in hue but also in brightness. Of course, the blue could have been darker than the red as well.
The second design choice is one I only discovered fairly recently. It is a lot more obvious in the inverted image than the original, too: there is a gradient in both colors from light at the top to dark at the bottom. That is not very obvious in the original version, since we expect lighter colors at the tops of things and darker colors at their bases. After all, light tends to come from above, and the lower parts of things are where shadows are cast. Only in this case, the effect makes the brightness differences in the colors even stronger. The dark red is close to black, and the entire red-to-very-dark-red gradient is somewhat suggestive. What else is red and turns black? Drying blood.
In addition to that, I believe that the dark color, especially towards the lower end, makes the red bars appear heavier than the blue ones. Since they are also pointing down, the additional weight might make them appear longer, or at least cause people to remember them as longer. Vertical bars appear longer than horizontal ones of the same length, and it may well be that the combination of bars hanging down from a baseline and the heavier color have a similar effect.
This is unproven at this point, but if I am correct I think it opens up some interesting possibilities. It means that we need to be much more careful with our choice of color, since the perceived weight might influence the way the data is read and remembered. Even if long-term recall is not a goal in visualization, we have to remember what we just saw when we switch between views as we think about our data. Subtle shifts could make a big difference if they make some values appear just a bit larger or smaller than the others.
The bikini chart is a great example of just how strongly simple design choices can change the appearance of a simple bar chart. Even if my speculation about weight is wrong, the other choices communicate and explain what the viewer is supposed to look for, without the need for explanatory text or a “shorter bars are better” annotation. That’s pretty good for a simple bar chart.