🧵 View Thread
🧵 Thread (12 tweets)
Visualizing relationships between senators in the 116th US Congress (current) based on cosponsorship of bills. Full res image here: https://t.co/Xq6WxP4uqF https://t.co/W6CQUTIcus
It's reassuring as a sanity check that many senators from the same state have such large edge weights https://t.co/fPAzMBli3r
I'm not familiar with the people in the center, but I checked on Wiki and everything seems to indicate them as known moderates in their parties https://t.co/k9liA9DhMM
Technical details: The original graph is bipartite with one side being senators and the other side being bills. Edges are (co)sponsorship. The colors are inferred with the bipartite stochastic block model (BiSBM) from @oneofyen and @DanLarremore
So, I did the sizing and the community detection on the full bipartite network, and then used that information to make the bipartite projection readable. The network without projection is far less readable because it includes all 100 senators plus over 4k bills https://t.co/iVLaKGwK3Y
Oh, and the weight of the edges is coming from the 'surprise' of the number of shared cosponsorships (relative to the null model). 'Surprise' being the negative log of the probability of that many cosponsorships occurring in the null model
I scraped the raw data using https://t.co/mRsGoiqNm0 from @govtrack and @sunfoundation then constructed the bipartite graph and took it from there