There's probably some cool insights in here, but I'll need some help understanding it. For instance, what does an edge in the graph represent? What can I tell from the boldness of the line?

@DavidChoinard can you give us a key?

The thickness of edges represents the number of times the pair of senators voted the same in the Senate (only edges above a threshold are shown).

The two tightly coiled clusters shows just how weak the relationships are between the two parties.

The closer the nodes are to the center, the more bipartisan those Senators are. In the current Congress, Sens. Murkowski and Collins really stand out as by far the most bipartisan senators and important linchpins in bridging the gap between both parties. (this bar graph of partisanship shows that: http://i.imgur.com/gYCMaxo.png)

Also, it’s clear that in the current Senate, partisanship is a lot more pronounced among Democrats, likely because they hold the majority in the Senate. (ie. they have a lot less incentive to compromise)

While there were ups and downs, it’s clear that there’s a general trend of the Senate becoming a lot more divided over the last 25 years. Comparing 1989 to today is striking — opinions were a lot more intertwined back then (this static image really emphasize it: http://i.imgur.com/4kDokyz.jpg).

The last url looks like some kind of political mitosis. There's probably a joke here related to the genome of government being split into two identical halves…

A more plausible reason for why Democrats would be more "partisan" is because moderate Democrats have been defeated by Republicans in recent elections, particularly in red states. This will likely be more pronounced if people like Mary Landrieu and Mark Pryor are taken out.

It probably pays to note that the house and senate were Democratic majorities in 1989, with a Republican president.

I would expect that to create a somewhat different dynamic than the party alignments of today.

Please describe the visualization (relationships you mentioned, how the nodes are colored) in the webpage. It is hard to understand what's going on there.

It would be great if you can make the data public, if possible. You could probably analyze each senator across the years.

I think your explanation of recent partisanship is perhaps not quite specific enough. I suspect the cause is the increasing reliance on threats of filibuster which make the stakes much higher for "crossing the floor".

David,

This may be asking for a bit much, but would you kindly put together a series of snapshots for a long time, spaced every 2 years? I would be willing to help out in building the automation and so forth if it's a big task.

as someone who is generally familiar with key political figures, i'd say this graph is awesome. it immediately quantifies information that you otherwise would guess at based on what people say about them. the time series is also telling.

Have you considered doing PCA and plotting the first two principle components?

Interesting stuff, thanks for putting it together.

I'm guessing that the thickness of the edges represents how often do two given senators vote the same way on a bill. And then the nodes are laid out in a way to cluster together those that have a history of voting similarly.

1) How are abstentions counted? Does a senator that abstains more get pushed out towards the edge for having less connections in general?

2) Is this correct: If a party member votes 100% along the party line, then they will be towards the center of the blob. If a party member is less partisan in the fashion of voting WITH bipartisan consensus (e.g. voting for hugely consensus PATRIOT act) then that person is gets put in the center. If a party member is less partisan in the fashion of voting AGAINST bipartisan consensus (e.g. voting against hugely consensus PATRIOT act) then they would be on the edge far from the center.

Interesting analysis, David. I actually got strikingly similar results looking at voting matrices and using SVD to turn them into two dimensions: http://vikparuchuri.com/blog/how-divided-is-the-senate/ and http://vikparuchuri.com/blog/the-danger-and-power-of-visuali... . I assume you did something similar.

Either way, it's surprisingly easy to get access to this data, and fun to analyze. I have the scraping code on github (described in my blog) for anyone who is interested.

Nice work David. For anyone interested in this type of analysis, Harvard has a new Data Science course this semester that had this node relationship analysis as a HW using the same data sources. You can check out the course site at: cs109.org

I'm actually a student in the course and my interest in this was spun out of the last homework.

Best thing I've seen all day. Thanks for sharing!

That's a great site, thank's for the link.

Ok, so while I wish I had the time to create a cool web app to analyze political data, I def think it would required not to overload the user's CPU. I launched the site and immediately my MAC slowed down to crawl. I checked top and it at at 100%+

I do not recommend you clicking on any of the dots.

Can someone make a glass app that covers a politicians face with the logos of their largest interests? That would make politics bearable for me.

Now there's a good use for "Big Data" ;)

I'd love to see a comprehensive database of donor-politician relationships along with causes they've championed and bills introduced/voted on. In lieu of the "approval graphs" shown by TV stations during debates, it would be more productive to show their existing voting patterns and who has supported them.

Their suits could have logos on them, sort of like boxers' robes.

People have made the "NASCAR jumpsuit covered with logos from donators" suggestion/joke for years. There was even a White House Petition for it: http://www.huffingtonpost.com/2013/03/20/white-house-petitio...

Yes, it's getting kind of old and is unworthy of this discussion.

Just because it's an old idea doesn't mean it's not a good one.

If you like this, you may also be interested in Simon Jackman's ideal-point estimates for both the Senate and House:

http://jackman.stanford.edu/ideal/currentSenate/x1.pdf http://jackman.stanford.edu/ideal/currentHouse/x2.pdf

Really cool! In addition to the comments others are making, I would recommend making the intensity of the base blue and red colors match. The democrats look more intense than the republicans just because of the color choice when given how interlocked they are, I assume the average blues are probably about the same as the average red. (Also a legend would help.)

Awesome work!

I think it's fascinating watching how the parties have split up more over the years.

It looks like the Democrats are coming together on their votes, and Republicans differences are widening. It does make sense, as the fiscal conservatives and social conservatives really are different breeds.

This would be even cooler if it was somehow combined with the data from opensecrets.

If you're into that kind of things, here is a similar project I did last year: http://friggeri.net/research/senate/

For those interested, I believe the underlying data is from DW-NOMINATE (http://voteview.com/dwnomin.htm).

The data is actually from GovTrack.us, which I suspect is also where DW-NOMINATE gets it. My script for querying GovTrack is on Github: https://github.com/DavidChouinard/congressviz

Very cool!

It would be helpful to see an animation to clarify this, but it's interesting to quantify how increasingly polarized this has become, starting in the early '90s.

+1 If the author could get data from 1980-2013, and make it into a 60 second animation, it would be really interesting to see.

I've definitely considered that, but it's computationally difficult to make it in the browser. This animated gif is as close as it comes: http://i.minus.com/ibi7p66VbETpnU.gif

This is illuminating, also would be if you could add years to each slide that would be even more fascinating. Great work and thanks for sharing!

This algoritm pragmatically computes how far left/right a senater's polical views are by placing them in the corresponding place on the page

I am not sure this is correct.

1) How are abstentions counted? Does a senator that abstains more get pushed out towards the edge for having less connections in general?

2) If a party member votes 100% along the party line, then they will be towards the center of the blob. If a party member is less partisan in the fashion of voting WITH bipartisan consensus then that person is gets put in the center. If a party member is less partisan in the fashion of voting AGAINST bipartisan consensus then they would be on the edge far from the center.

Is data for pre-90s years available? I am really curious to see the 60-70-80s progression.

My data source (GovTrack) only has vote count, but not roll calls, before the 101th Congress. It might be possible to get data before that, but I suspect it's quite hard.

A legend/key would be nice.

I am assuming that the more clustered the circles are, the more like minded the party is? Each line represents a vote. So, a thick line between 2 congressman represents more general agreement, and thinner line means they agree only on a few things. Also, the position of the circles to the left or the right represents the extent of their political leaning, based on their votes.

Above are just informed guesses. A key really would be nice.

This is absolutely fascinating.

what's the story on Miller (D-GA) 2003-2004?

He endorsed Bush in 2004. And he wished settling an argument with a duel was still around. https://www.youtube.com/watch?v=AuBnlNjZq24