Every year the National Institutes of Health allocates $10.7 billion (one-third of its funds) for clinical science research while the pharmaceutical companies spend $52.9 billion (90% of its annual budget). However, we know little about funder collaborations and the impact of collaboratively funded projects. As an initial effort towards this, we examine the cofunding network, where a funder represents a node and an edge signifies collaboration. Our core data include all papers that cite and receive citations by the Cochrane Database of Systemic Reviews, a prominent clinical review journal. We find that 65% of clinical papers have multiple funders and discover communities of funders that are formed by national boundaries and funding objectives. To quantify success in funding, we use a g-index metric that indicates efficiency of funders in supporting clinically relevant research. After controlling for authorship, we find that funders generally achieve higher success when collaborating than when solofunding. We also find that as a funder, seeking multiple, direct connections with various disconnected funders may be more beneficial than being part of a densely interconnected network of co-funders. The results of this paper indicate that collaborations can potentially accelerate innovation, not only among authors but also funders.

The data was collected from https://app.dimensions.ai/.

We first extracted all publications from the Cochrane Database of Systemic Reviews journal. Then we got all the papers that cite this journal and all the papers cited by this journal. This gives us an entire corpus of ~700k clinical science papers. 

Accompanied with these papers is funding information. This only includes a subset of ~160k papers and 1761 funding organizations. After this, we filter funders who have funded at least 25 papers, which gives data of 653 funders from 143904 papers.

Then, we use this data to create the co-funding network, where a node is a funder and a link indicates collaboration on a project. The constructed network is of 653 nodes and 26165 edges, with an average degree of 80.13 - we publicly release this network along with the paper. This is the primary source of analysis of our work.

Additionally, for ego-network and heterophily analysis, we further filter this network to indicate strong collaborative ties. To do this, we remove edges that have less than 10 papers together. Beyond this, we observe a network of 512 nodes and 3312 edges with an average degree of 12.93 - we also publicly release this constructed data for further analysis.

The publication data can be accessed using the Dimensions portal using appropriate credentials. We are unable to release the entire publication corpus without prior permission.