Visualizing the U.S. Senate Social Graph, 1991 - 2009 [Part 1]

By Andrew Odewahn
May 4, 2009 | Comments: 3

Last week's party switch by Arlen Specter provoked a lot of discussion and analysis. One of the more interesting was an article in Slate called A Facebook-style visualization of the Senate. It presented a graph visualization technique to reveal voting patterns by connecting senators if they vote together over 65% of the time. The picture is really worth 1,000 words -- you can immediately see that Specter (as well as Olympia Snowe and Susan Collins of Maine) are more aligned with the Democrats than other Republicans.

I was curious if this is a relatively new phenomenon, and what the social graphs have looked like in the past. Using data from Govtrack, one of the many great sites devoted to increasing transparency in government (a theme being explored at the Government 2.0 Summit in September), I was able to pull this information fairly quickly. The results are shown in the following Slideshare document:

First, a words about the charts. Each Senator is represented by a labeled dot. (Labels change from year to year, so Senator 60 in the 106th congress is different than Senator 60 in the 107th.) Like the Slate article, Senators get connected if they vote together 65% of the time across the 2-year legislative session. Red dots indicate Republicans, Blue dots represent democrats, Green dots are independents, and a smattering of yellow dots indicate when my program couldn't identify a party. (More on this in the next post). Finally, it's worth noting that people come and go in the Senate, so the various cluster in the diagram change over time both from within (as people shift allegiances within the Senate) and without (as new people are voted in).

In looking at the patterns that emerge, a few things stand out:

  • While the party switch is certainly big news, I think the much more interesting phenomenon are the clear voting blocks that form within the parties at various points . (Although there is some ebb and flow, this appear to be much more pronounced on the Democratic party.) The notion that Specter has unleashed uniform liberal hegemony because of this switch is nonsense -- there are clear and distinct moderate blocks within both parties, and Specter is aligned with the generally more moderate wing.

  • While the 102nd session (from Jan. 1991 to Jan. 1993) saw considerable cross-party voting (note how the chart has an oblong shape), subsequent charts have a "dumbell" appearance, with tight voting blocks connected by tenuous threads of Senators that are largely consistent from session to session.

  • The level of partisan rancor is clearly evident in the 104th session, the year of the famous Republican Revolution; not a single Senator voted with the other side over 65% of the time. Finally, the group of Senators that cross party lines has remained consistent. Conventional wisdom has it that this group is mainly moderate, New England Republicans, but the final slide shows a new splinter pulling away from the core Republican block.

There are obviously lots of other conclusions (well, speculations, really) from these charts (and I'd love to hear them in the comments!), but the coolest thing about them is that the patterns emerge directly from the data, without and prompting or spin.

In the next post, I'll give the code I used to pull the data from GovTrack site and create the charts using Graphviz.

You might also be interested in:


Any interest in combining this graph with donor/contributor data? It would be good to correlate voting behavior with financial support.

Some folks at the University of Michigan have done this with the data from OpenSecrets -- check out

It definately depicts two distinct voting blocks, with various sub blocks and very little bipartisanship over time. What it doesn't show is a reference to any kind of fixed outside points, like the constitution for example. In other words, it doesn't really show if they're disagreeing over beef vs. pork or omnivore vs. vegetarian. It would be interesting to see the data shown in light of constitutional fidelity. I'll bet the visualization would come out very differently.

News Topics

Recommended for You

Got a Question?