Mapping Philanthropic Grants with Big Data and Network Science

• In Event: Big Data, Civic Networks, and Creative Collaboration

Abstract

Philanthropic grants connect non-profit organizations in a complex web of financial support. To quantitatively map this ecosystem and uncover its statistical organizing principles, we need a comprehensive network capturing grant-giving between non-profit organizations [1, 2]. One of the largest sources for philanthropic finances comes from the tax records of Form 990 e-filers, released by the Internal Revenue Service (IRS). Here, we detail our efforts to disambiguate over 10 million grants made between 685,397 non-profit organizations in the United States from 2010-2019. We then build the funding network for organizations involved in scientific research and higher education (695,917 grants, totaling over $190 billion) [3], and the arts (149,921 grants totaling over $36B) [4]. Our analysis reveals many similarities between supporters of these spaces and some differences. For one, supporters of both tend to select geographically nearby recipients and both have a high rate of repetition in supporting the same recipients from one year to the next. We also explore the possibility of developing a network-based recommendation system to help both recipients and funders identify new opportunities, and highlight the importance of prestige in arts funding.

We began by collecting the IRS 990 filings of e-filers from AWS (since moved to the IRS home website) including 3,910,398 tax forms for 685,397 organizations for the 2010-2019 tax years. We then identified employer identification numbers (EINs) for recipient organizations, either based on what was listed in the tax form or by matching with an outside file. Ultimately, we obtained the recipient for 8,186,055 of 10,388,779 total grants (79%). Many of the unmatched recipients included non-US organizations, individuals, and entities that are not non-profits e.g., cities.

We find that philanthropic funding is strongly biased by geography. If grants were distributed randomly across the nation (preserving the number of recipients in each state), about 5% of grants would be awarded in the donor’s home state. In contrast, we find that within science 35% of grants go to the donor’s state and in art 56% of grants are within the donor’s same state. Furthermore, we find that in science, the likelihood for a donor’s largest recipient to be in the same state is even higher, occurring in 50% of cases, whereas smaller recipients are less likely to be local. In contrast, in art both larger and smaller recipients are local in nearly 60% of cases. Beyond locality, we find that in both spaces, funding is quite stable, with a one-year repeat rate of nearly 70%.

Further analysis explored the possibility of recommending new grants in science using the Adamic-Adar Index, and obtained a remarkably good AUC (area under the curve) of 0.87 [3], where 0.5 represents predictions that are no better than random and 1 would represent perfect predictions. This high level of predictability raises the possibility of using Big Data and network science to help recipients identify new foundation funders, improving the overall allocation of resources. More broadly, this study represents one use-case in mapping grants using computational social science [5] applied to philanthropic studies.

References

1. Fiennes, C., We need a science of philanthropy. Nature News, 2017. 546(7657): p. 187.
2. Ostrower, F., Why the wealthy give. 1997: Princeton University Press.
3. Shekhtman, L.M.G., A.J., Barabasi, A.L., Mapping Philanthropy in Science. arXiv preprint arXiv:2206.10661, 2022.
4. Shekhtman, L.M. and A.-L. Barabási, Philanthropy in art: locality, donor retention, and prestige. Scientific Reports, 2023. 13(1): p. 12157.
5. Ma, J., et al., Computational social science for nonprofit studies: Developing a toolbox and knowledge base for the field. VOLUNTAS: International Journal of Voluntary and Nonprofit Organizations, 2021: p. 1-12.

Author

• Louis Shekhtman