4. Process

1. We downloaded the data, imported them into Google Refine, checked them for consistency and trimmed the strings.
2. We imported them into a Tulip perspective via a Python script, in the form of a bipartite graph (organisations connecting to projects).
3. We removed one outlier project (Implementation of activities described in the Roadmap to Fusion during Horizon 2020 through a Joint programme of the members of the EUROfusion consortium). With 31 partners and a budget of over 400 million EUR, it was much larger than others. Also, its partnership was probably less subject to competition than most others in Horizon 2020.
4. The resulting network is a organisations-to-projects network with 16,592 organisations and 11,068 projects, connected by 47,196 edges. The edge from organisation O to project P encodes the contribution (ecContribution) allocated to O under P, and so represents the value of that participation to O. The total EU contribution to the project is a property of the node P.
5. We projected this network onto an organisations-to-organisations partnership network, with 16,592 organisations connected by 1,077,218 edges. Each edge P1 => P2 encodes the co-participation between P1 and P2 in one project, seen from the point of view of P1. It encodes the ecContribution property referred to the funding received by P1. Conversely, there will be an edge P2 => P1 that encodes the co-participation of P1 and P2 in the same project, but now seen from the point of view of P2. The ecContribution property now encodes the funding received by P2. If P1and P2 co-participated in n projects, there are n P1 => P2 edges and n P2 => P1 edges
6. We next calculated a measure of bargaining power by each participant in each project (encoded in each edge), as well as a measure of each participant's bargaining power across the whole of her participation in Horizon 2020 projects. Formulas are provided and discussed in a separate wiki page.

In order to reduce the graph, and so make it more amenable to visual analysis, we:

1. "Stacked" edges on top of each other. This means imposing that there is at most one P1 => P2 edge. If P1and P2 co-participated in n projects, we attribute value n to the projectsTogether property. This yields a graph with the same 16,592 nodes and 493,014 edges.
2. Filtered the edges of the stacked graph that encode exactly one project (projectsTogether = 1) to obtain a stable partnership graph. Nodes that have no edges in the resulting graph (that is, that have no stable partnership), are next dropped. The resulting graph represents partnerships that have been considered satisfactory enough to be carried through from one project to another. This graph has 3,414 nodes, and 46,632 edges.
3. Further reduced the stable partnership graph to its largest component. This is a network representation of the Horizon 2020 "scene". This graph has 3,138 nodes and 45,698 edges.