Guide to replicate the quantitative analyses of the article "Positioning among International Organizations: Shifting Centers of Gravity in Global Health Governance".
Network analysis consists of a network of IOs and self-declared practices involving another IO. This information can be found in the document Data3.csv, where "Source" is the IO declaring the practice, "Year" the year of the report, "Target" the IO concerned by the declaration and "Weight" the number of times this practice takes place in the reporting year. This CSV file can be imported into Gephi, an open source network visualization program (gephi.org), specifying that it is the Edges File.
For practices to appear in the network (and not just IOs), however, it is necessary to create an identifier for each of the lines in this file, as in Data4.csv. In the latter, we find a list of relationships between an IO (Source) and a practice (Target) whose identifier aggregates the names of the IOs it connects. This CSV is the source of Figures 1 and 2.
Download Data4.csv. Open Gephi and create a new project. In the Data Laboratory, click on Import spreadsheet and select Data4.csv. Make sure the software understand that this is an Edges table and import it in the current workspace (append to existing workspace). Go to the Overview interface, where the Context panel shows that you now have 1027 nodes and 2038 edges. In the Statistics panel, run the Avg. Weighted Degree metric, then use this metric in the Appearance panel to change the size of the nodes (click on "nodes", then "size" - the triple circle logo -, then "ranking", then select "Weighted degree" in the drop-down menu): 5-100 for Min/Max size, then click Apply. Then, go to the Layout panel and chose "ForceAtlas 2" in the drop-down menu: change the Scaling to 50 and check Prevent Overlap, then click on Run. With these settings, you have a perfectly legible network. To obtain a result similar to figure 1, you then need to work on the details. For example, to display the node names, duplicate the ID column in the Label column in the Data Laboratory Nodes table. Or by giving different colours to the IOs to distinguish them from the practices (this can be done manually, since there are few of them and they are the most connected nodes, or by adding an attribute to these nodes in the node table, or by automatically detecting the bipartition of the graph with a plugin like Multimode Network Transformation).
To simplify replication and subsequent analysis, we also provide this network data in GEXF (network.gexf), a file that can be immediately opened by software such as Gephi. In Gephi, simply go to File/Open and select the GEXF file. Then, click on the black T at the bottom of the Overview panel to display the names of the IOs. By checking the Context panel, you see that 1027 nodes and 2038 edges is exactly the result of the procedure described in Option 1.
Figure 2 simply consists of versions of Figure 1 filtered by date to give 5-year intervals. Simply open network.gexf as described above and head to the filters panel on the right side of the Overview interface. Double click on Attributes, then on Range, then on year Integer (Edge). This opens a little Range (year) Settings selection tool at the bottom where you can move the slider to cover 1970-1974 for the first interval (and then click Filter), or 1975-1979, 1980-1984, etc. This will remove all the edges that are not occurring during the selected period. This does not remove all the nodes that are not connected anymore (practices that occurs during another period), you can easily remove them by running a new Avg. Weighted Degree calculation in the Statistics panel and then filter the Nodes table in the Data Laboratory according to this attribute and delete all the nodes that scores 0. Removing nodes is not recommended if you still want to experiment with these filters (it is recommended to save the Gephi project before this step, to be able to repeat the operation). Also note that you'll need to recalculate the size of the nodes based on the filtered situation, and that the edges will also be automatically resized when exporting the SVG in the Preview interface.
The Data5.xlsx file aggregates all the lines in the Data3.csv file at IO level and for two-year periods. The visualizations in Figure 3 were produced directly in Microsoft Excel. Just take the 2 columns of a specific IO and create a line or surface chart.
The Data6.csv file is used to produce Figure 4, which aggregates practices at the IO level but retains the details of their classification. The latter information comes from the data produced for a previous paper (Bahr et al. 2021), which focused on practice categories. You can access the raw data about the practices in Data1.csv and Data2.csv, but we prepared Data6.csv in a way that makes it easier to visualise as a Sankey diagram, including the IOs. Once again, therefore, this is a bipartite edge file (IOs and practice categories), which can be visualized with Gephi or, as in the case of Figure 4, with an online Sankey-producing tool. Go to sankey-diagram-generator.acquireprocure.com and click on Load -> Your own data on the top left of the interface. Then select CSV and copy-paste the content of Data6.csv in the field and click Build Sankey. You now have an interactive but rather raw version of Figure 4, that was then exported in SVG and edited in Inkscape to match the colors of the previous figures. Note that the bubbles in the center (meant to show that these categories also have sub-categories) have been added directly in Inkscape and are exactly the bubbles displayed in the figures of our 2021 paper.
