Skip to content

Agreement Groups in US Senate and Dynamic Clustering

Adrien Friggeri has a lovely visualization of US Senators movement between clusters:

You have to click the image and play with it to appreciate it. The methodology isn’t yet published – but I can see how this could be very illuminating. The dynamic clustering aspect hasn’t been researched much – one of the notable pieces is the Blei and Lafferty dynamic topic model of Science.

I did a static analysis of the US Senate back in 2005 with Wray Buntine and coauthors. Some additional visualizations and the source code are here. We did a dynamic analysis of US Supreme Court on this blog but there’s also a paper.

My knowledge on this topic is out of date, however. Who has been doing good work in this area? I’ll organize the links.

[added 4/29/12, via Edo Airoldi]: Visualizing the Evolution of Community Structures in
Dynamic Social Networks by Khairi Reda et al (2011) [PDF].

[added 4/29/12, via Allen Riddell] Joint Analysis of Time-Evolving Binary Matrices and Associated Documents by Eric Wang et al (2010) [PDF] [Video]


  1. Andrew says:

    Cool. I just have some quick comments on the display, which looks clean but I think could be further improved.

    1. It is conventional to put left-wing views on the left of a graph and right-wing views on the right. Thus, the graph should be mirror-reflected.

    2. I recommend that the time axis be shown in years rather than in numbered congresses. Everybody knows what 2005 means; not everybody knows when the 111th congress is.

    3. I’m suspicious of the lines being right next to each other. This would seem to be misleadingly implying some sort of equal spacing.

    On the substance of the matter, the above graph disagrees with the standard findings of ideal-point models in which the two parties are completely separated. It would be good to understand this discrepancy, otherwise it’s hard to know how much to trust these numbers.

    • Well, Zell Miller was voting like a true Republican even though he was a Democrat. But I agree that there have been few exceptions. I’m not sure they’re using anything like ideal point in this work. To examine the discrepancy between two groups, I’ve used tabulations, such as those contrasting Democrats with Republicans at — this way, one can contrast the fringe and the majority.

    • Thanks for you suggestions, I have followed your advice and updated the visualization.

      About 3. there is no ranking on the edges, their position inside the group is totally arbitrary, if this is what your were wondering about.

      They do disagree with the ideal-points models, but I suspect this is due to the fact that we take absenteism into account differently (but I’ll have to check the math).

  2. This made me think of Eric Wang and collaborators’ work in “Joint Analysis of Time-Evolving Binary Matrices and Associated Documents” NIPS 2010 which looks at votes and legislation texts between 1989 and 2008. The paper is available here: Additional visualizations at