A House Divided? A Network Modelling Approach
http://snap.stanford.edu/class/cs224w-2013/projects2013/cs224w-054-final.pdf
This paper detailed network analysis of the U.S. House of Representatives and Senate and sought to devise a method for detecting whether a certain congress member would vote for a certain bill, or more importantly whether a congress member would betray their party’s vote on a certain bill. This analysis is important because it could be useful in predicting whether a bill will get a majority vote, which greatly increases the probability that bill gets passed. To do this, the authors described a fully-connected network in which nodes represent congressmen, and all edges have weights corresponding to how many bills two congress members voted differently for subtracted from the number of bills they voted similarly for. From this, they created a signed graph based on the sign of these edges and analyzed the proportion of balanced and unbalanced triangles for the whole House, Democrats, and Republicans. This revealed that for the house there were ~83% balanced triangles in House, ~97% balanced triangles in House Democrats, and 99% balanced triangles in House Triangles. They then go on to describe a machine learning model they used to predict whether a congress member would vote against their party. A list of the 5 candidates most likely to defect from their parties are
Democrat: Nathan Deal (GA), Ralph Hall (TX), Richard Shelby (AL), Virgil Goode (VA), and Rodney Alexander (LA)
Republican: Thomas Massie (KY), Arlen Specter (PA), Craig Thomas (WY), Jo Ann Davis (VA), and Luis Fortuno (Puerto Rico)
At its core, this project sought to analyze the extent to which the Structural Balance Property held for the U.S. Congress, and while 83% balanced triangles are much better than random, it indicates that not everyone’s vote is set in stone. When also considering the comparatively high proportion of balanced triangles within the Democrats and Republicans, we can deduce the majority of the unbalanced triangles are across party lines. So now we know there are structural incentives for congress members to change their voting patterns. In conclusion, the House is divided, but it can still stand.