Emergent Structure within the Event Sponsorship Networks of British Parliament

Recently, the fantastic DataBlog at The Guardian cleaned and released data on the associations Members of Parliament were hosting for lunch and dinner events. As DB explains:

Besides the Legg report into specific MP claims from the Members’ estimates committee, there’s also a report into members’ use of dining room facilities.

This immediately struck me as a natural network, where Members are connected to these associations by hosting events, i.e., a two-mode network of Members and who they are hosting. As such, I decided to download, parse, and examine the properties of these networks.

Using NetworkX for the data generation and Network Workbench for the secondary analysis and visualization, we are able to observe interesting emergent structure within this network. Before I describe the results; however, it will be useful to describe the raw data and how I converted it.

First, the data contains over 8,000 events from 2004 to 2009. Each entry contains; among other things, the sponsoring Member, date and the association. The create the two-mode network I collapsed reoccurring Member-to-association observation to obtain a count of the number of times a Member sponsored an event for a unique association. The resulting network; therefore, is a weighted two-mode network, wherein the edge weight represents the number of events a member sponsored for the connecting association. Second, within the raw data there are several association titles that are very closely related, and all likelihood the same association. For the sake of time—and my sanity—I decided not to address these duplicates, and treat “3M” and “3M Company” as separate edges.

The resulting network has 6,560 nodes and 6,831 edges, which is large but relatively sparse. The vast majority of these connection have a weight of one; therefore, in an attempt to reduce the complexity of the network and reveal the core structure, I decided to dichotomize the data to retain only those connections with a weight greater than one. Also, there was a wonderfully ambiguous association called “Association,” which acted as a very large hub in the network but also washed out other emergent structure—so I removed it.

The final network is more manageable, with 726 nodes and 512 edges. Next, I created bipartite projections for both the Member-by-Member and Association-by-Association networks. This generated several disconnected components within each projection, from which I extracted the main components and visualized below using Seadragon.

Unfortunately, I know almost nothing about the sociological dynamics of British Parliament, and therefore can offer no substantive interpretation of the above network. I would point out the obvious tightly connected cluster with Richard Spring and Tim Boswell acting as gatekeepers to other clusters. Are there any readers from the U.K. that care to offer up some armchair analysis?

Moving to the associations network, we can clearly see the centrality of the Third Term group, with close connections to the One Nation and SSB groups. Also, note how the emergent clustering is able to reveal several candidate duplicates within the data. Overall; however, there does appear to be a natural clustering, with medical related associations and apparent research groups clustering. Finally; as a side note, I was intrigued by the clustering of the BBC with the Parliamentary Beer Group.

As always, I have uploaded the data, Python code, and the above networks to the ZIA Code Repository. I am hoping that industrious readers will be inclined to download it and perform their own analysis. One “low hanging fruit” would be to add political attribute data to the nodes, such as political party to the Member network, and what more that data reveals in the structure. Also, as each observation is time stamped a network evolution analysis may reveal additional emergent dynamics.

Automatically Generated Related posts:

1. Visualizing the Structure of Venture Capital Co-Investments
2. The “Exploded Network View” in 3D, Part I
3. The Royal Society’s Take on ‘Stochastic Networks’
4. 3G + Loopt + NetworkX = Geospatial Analysis of Dynamic Networks
5. UPDATED: Network Decomposition Visualization