Normalized Co-Mobilization Network of Antiwar Groups

Data is from

Michael T. Heaney, Fabio Rojas, "Partisans, Nonpartisans, and the Antiwar Movement in the United States" AMERICAN POLITICS RESEARCH, Vol. 35, No. 5 (September 2007)
2529 respondents to surveys of antiwar activists attending large-scale public demonstrations in 2004-2005:

Apologies if the acronyms are unfamiliar. Thanks to Michael and Fabio for letting me play with their data!

Normalized Co-Contact

Layout is multi-component KK in SoNIA, edge weight co-occurences are normalized in UCINET to control for group sizes. Area of nodes proportional to number of respondents who said they were contacted by that organization. (Michael points out that the "no contact" node I've added should really be "indirect contact" via friends, internet, mass media, etc.) Mouseover unlabled nodes to show labels

Raw Count Co-Contact

After re-reading and thinking again, I've convinced myself that since the mobilizations are drawing on a pretty large population (metro area of the cities involved?) probably it is not correct to normalize the tie weights. In other words, since the membership of ANSWER and UFPJ are still pretty small relative to the population, co-membership is still statistically interesting. (Anybody have numbers on this?) So this plot uses the counts of contacts directly a weights. But since the range is fairly large, it was very hard to get it to layout well, and the big nodes are crunched together

Raw Count Co-Contact, line widths only

Another version, this ignores the weights for the layout, but still shows them as edge thickness. (I'm beginning to think this is a good technique, that changing both the line widths and distances of nodes messes with the viewer)

Co-Membership Network

Since it is a bi-partite graph, it is possible to flip it the other way, and draw it so that each node is a person and the people have links to everyone they share an organization with. This makes for very dense clique-y graphs that function kind of like Venn diagrams. All the structurally equivalent nodes are pretty much grouped. You can see the core in the middle (UFPJ?) and the three other slightly small groups (CodePink, ANSER MoveOn, and little clumps of nodes at the boundaries that were contacted by both groups. Really, to be consistent I should add a disconnected clique of 1766 nodes who were not directly contacted... Sorry I left the labels off. Also should color the nodes by their partisian affiliation, would highlight the claims in the paper

- Skye Bender-deMoll