Dispatch: effective visualization
Question of the day: How do you present social network findings to an audience?
I think about this a lot and I encourage you to post positive examples or your own solutions in the "comments" section. Especially after attending a few social network conferences and producing my fair share of circle and line network diagrams. Most of the time, these do little more than demonstrate the researcher's ego (insert bad egonetwork joke here). This is the hallmark "look at all the data I collected!" slide for social network analysis. It is the prettier and more dizzing version of the size 8 font regression table that also appears in many a scientific talk. So, what's the solution? There are many out there, but I think a recent article in SEED magazine (http://seedmagazine.com/content/article/getting_past_the_pie_chart/) presents a good synopsis of the great visualization debate.
The article, after presenting the history of the pie chart and its rise to glory, goes on to warn about the lack of values in some varieties of visualization. Among them: social network diagrams. Social network diagrams as they are produced in UCINET or other programs with graphical display are good for illustrating connectivity, but poor at giving an audience much insight into patterns and processes associated with connections unless they are given more thought and attention. The following two paragraphs (take directly from the article) do a nice job summing up the problem and alerting us all to a solution that also helps with data exploration:
One way to solve the problem of overly complicated diagrams is to introduce interactivity. For an expert using visualization to analyze data, interactive displays may be much more useful than inert maps. A flat picture of a network does not suffice. “A network diagram will be completely unusable at about 30 nodes,” data visualizer Ware says. “But if you have simple interaction techniques, you can work up to a few thousand.” For those who work with networks, zooming in on a node to observe its connections with others nearby can mean the difference between a useless tangle and a successful tool.
Iterated simplicity, however, may turn out to be an even bigger breakthrough. Cleveland finds that an effective way of detecting patterns in massive data sets is to make a simple chart for each subset and view the hundreds of charts in quick succession. The human perceptual system is phenomenally good at spotting patterns, he notes: “We take our vision processes for granted. But there’s nothing that humans have created that’s more complex and amazing than the visual perceptual system in the first 100 milliseconds.” Skimming through these visual databases, he’s found, can be much more effective than complicated visualization; Cleveland is working on a protocol to share with others soon. Yet even as he advocates the use of visualization databases, he emphasizes that numerical tools — statistical tests of variance and significance — are just as important in assessing trends. Current enthusiasm for putting numbers into pictures sometimes obscures the fact that science is, after all, a quantitative pursuit, and an image alone cannot replace numbers.