Parallel Sets
In our work with large, complex datasets, we are always looking for new methods to visualize this data. An effective visualization reveals patterns and trends that may otherwise have been obscured in the data. Fortunately, data visualization is currently experiencing a renaissance that is moving us well beyond the constraints of the bar graph and pie chart.
Today, I'd like to highlight one particularly promising new method: Parallel Sets, a method for displaying multi-dimensional categorical data developed by Robert Kosara. We've used this method below to demonstrate the relationship between presidential ballots cast across multiple elections, collected during our post-election survey in November of last year.
The thin horizontal bars show the frequency of votes for each candidate in a particular presidential election, while the colored vertical ribbons show how combinations of ballots are distributed. This allows us to quickly and efficiently see how individuals vary their ballots over time.
Unfortunately, the software does not give us any control over color, so we cannot work with a conventional red=Republican and blue=Democrat color scheme. Another limitation to consider here is that we are by definition looking only at people that have voted in every presidential election since 1992. We could have added non-voters and those that were ineligible, but it would have unduly added to the visual complexity of the chart - we've found that once you get beyond three dimensions with more than three categories, the ribbons become overwhelming. Finally, what you see here is just an exported picture file - in the application itself you can hover over the ribbons to highlight them while also seeing tooltips that tell you the size of the ribbon (number of cases and their overall percentage).
As you can see, there is a sizable proportion of the electorate that are "five-of-five" partisan voters: 35% were a Clinton/Clinton/Gore/Kerry/Obama voter, while 28% were a GHWB/Dole/GWB/GWB/McCain voter. But the ribbons that diverge away from these cores merit attention.
To my eye, the most fascinating ribbons are the aqua-colored ones that represent the 11% of the electorate that admitted to voting for Perot in 1992. Roughly one-third of his supporters remained loyal to him in 1996 while the remainder split evenly between Clinton and Dole, and 3 out of 4 two-time Perot voters remained consistently Republican presidential voters post 1996. Finally, there were also a number of GHWB voters that drifted to Perot in '96 only to comfortably return to the Grand Old Party since then.
Other intriguing ribbons to consider:
- the one time surge to Clinton in '96 among 1992 GHWB voters that then came back to vote Republican again and again.
- the 4% that voted Clinton/Clinton/GWB display an interesting distribution post 2000, with a majority staying Bush in 04, but then roughly splitting between McCain and Obama in '08.
Besides displaying relationships between votes, we've found this visualization method to be especially useful for things like the relationship between party identification and ideology, or the interaction between issue stances like abortion, gay marriage and gun control. I'll be investigating some of these other relationships in future posts. In the meantime, let us know what you think about this new visualization method in the comments.
- Alex Lundry
Comments on this entry