WHAT is the best way to display information? The data geeks at The Economist struggle with this question every day. On January 17th we produced a daily chart about Kickstarter, the largest crowdfunding site. It showed projects by three measures—money pledged, average pledge and success rate—using traditional bar charts.
However a draft version was more ambitious, showing the interrelationships among the three categories using a technique called "parallel coordinates" (see the thumbnail chart on the top right). The vertical columns rank the projects; the lines show how ranks change across the categories. Steep vertical lines signal something interesting. Yet we felt critics might grumble that the comparisons were not related: money in dollars, success rate in percentage. Hence, straightforward bar charts ruled the hour.
How could we have adapted the technique to the data a bit better? We decided to experiment. First, we removed the second category, "average pledge," since it simply refined our understanding of "money pledged". The stacked columns were replaced with projects measured in plain numbers. We smoothed out the lines and put in bold the most striking findings: dance projects attracted far less money but had the best success rates, while games hauled in the most yet had among the worst rates (second chart).
But this wasn't ideal either. What we gained with simplicity we lost in detail, and the proportionality in the ranking was stripped out. So we returned to the workbench. The beauty of the data was both the rank and the change in rank—not one or the other. We decided to reinsert two of the original columns, and used tinted, proportionally-filled lines to show changes (third chart). Now the data in each column remained dominant while the rise or fall in rank was shown with the changing volume of the line.
This made the original worry about the different units a non-issue, since the chart showed the relationship between the projects, not simply the metrics in isolation—as did the original bar charts, forcing viewers to make these connections themselves. Yet the final experimental chart is more complicated: data-vis giveth, data-vis taketh away.
Which do you think works best? Let us know in the comments below.
Update:Download the data that we used for the charts here. Share any infographics you create via a link in the comments field.
