Funding landscape of the corpus: funder attribution, policies, and the Gates 2025 natural experiment

[larnsce]

Background

Extending the corpus methods work (#38) to the funding landscape. Starting point: a detailed analysis of the Gates Foundation 2025 open-access policy and its implications for data-availability measurement (provided by @larnsce, reproduced below). Before building on it, this issue grounds the funder angle in what our corpus actually carries, and the empirical finding materially reshapes the plan.

Related: #14 (FAIR scoring of DAS), #35 (DAS-accessibility), #38 (harvest-existing methods), #37 (WASH-relevance).

Empirical finding: funder metadata in our corpus is sparse and era/field-dependent

Before assuming we can identify funded papers, I measured funder-metadata coverage on our own scraped corpus (samples, both APIs):

Slice	Source	Funder coverage	Gates found
jwh matched works (all eras)	OpenAlex grants	0 / 60 (0%)	0
jwh matched works (all eras)	Crossref funder	4 / 60 (7%)	0
Water Supply, 2021+	Crossref funder	23 / 50 (46%)	0

Takeaways:

1. OpenAlex grants is effectively empty for our corpus (0% on the sample); it is biased toward funders that deposit structured award data (NIH, ERC, major national funders). Not a usable funder source here.
2. Crossref funder is the better source but is strongly era-dependent: ~7% for older/mixed articles, ~46% for 2021+. A historical funder analysis is not feasible from metadata; a recent-years one partially is.
3. The dominant funders in the well-covered recent slice are Chinese/national (National Natural Science Foundation of China leads), reflecting the engineering journals' authorship. Gates appears 0 times across all samples.
4. Gates-funded WASH work likely does not concentrate in our corpus. Gates WASH funding flows to global-health / development / LMIC venues (PLOS, tropical-medicine, WASH-development titles), not water-supply engineering. Our scraped corpus (jwh/aqua/ws + the 46 filtered-slice) is engineering-heavy. The Gates natural experiment may require a Gates-enriched corpus slice we have not scraped, not the current one.

Implication: funder attribution is itself a research problem here, not a metadata join. The metadata route covers a minority of articles and misses Gates; the remainder would need full-text funding/acknowledgement extraction (same full-text-access constraint as #35; rtransparent already extracts funding statements, see #38).

Funder-policy landscape sources (the "required" layer, funder edition)

Parallel to the journal/publisher policy layer in #38, funders have their own OA + data policies, with registries:

• Crossref Funder Registry / Open Funder Registry (OFR) - https://www.crossref.org/services/funder-registry/ - CC0 registry of funder names + IDs; the vocabulary behind Crossref funder. Transitioning into ROR (Crossref deprecating OFR). Query via Crossref REST API (no key).
• ROR (Research Organization Registry) - the successor identifier for funders too; API for name->ID matching.
• ROARMAP - https://roarmap.eprints.org/ - registry of 280+ OA/data mandates by funders and institutions.
• Sherpa Juliet -> Jisc Open Policy Finder - funder OA + data-archiving policies; 178 funder policies in the merged Open Policy Finder (already the OA-policy source in Measure data-availability and OA practice: harvest existing open datasets before scraping #38, joins funders too). Legacy endpoints die April 2026.
• cOAlition S / Plan S - the mandate framework UKRI, Wellcome, Gates(historically), and others align to; useful for classifying funder stringency.
• Major individual funder data policies for comparison (all with WASH-relevant grants): Gates, Wellcome, NIH (DMS policy 2023), UKRI, Horizon Europe ("as open as necessary, as closed as required", FAIR by default).

The Gates policy note (provided starting point)

Announced March 2024, effective for submissions from 1 Jan 2025:

• All Gates-funded manuscripts posted as preprints on a recognised server.
• Foundation will not pay APCs (fees fall to grantees).
• All funded manuscripts CC-BY 4.0.
• Accelerated open-data timeframe; DAS required.
• Rationale: APCs relocate the barrier from reader to author, excluding LMIC researchers. Funds redirected to open infrastructure, incl. a $3.3M PLOS partnership (2025-2027) for APC-free PLOS publishing by Gates authors.

Implications the note identifies (retained, still valid where funder ID is available)

1. Version fragmentation - preprint + accepted manuscript + (fewer) version-of-record. Same study may enter the corpus twice; DAS depends on which version was harvested. Actions: add a version field (preprint/accepted_manuscript/version_of_record); dedup on preprint->DOI links (Crossref relation, bioRxiv/medRxiv APIs).
2. Preprint DAS is a different object - ODDPub/rtransparent validated on published PMC full text; preprint servers don't enforce structured DAS. Detection recall drops for measurement reasons, not behaviour. Action: stratify any tool-accuracy validation by version type.
3. Natural experiment - accelerated open-data + CC-BY should improve Gates-funded data sharing from 2025. Design: tag by Crossref funder ID, compare Gates vs non-Gates pre/post Jan 2025. CAVEAT (new): not executable on the current corpus - Gates is absent from it; needs a Gates-enriched slice.
4. PLOS Water as higher-value source - Gates WASH content concentrates in PLOS given the APC waiver; PLOS OSI (Measure data-availability and OA practice: harvest existing open datasets before scraping #38) gives a free per-article data-sharing measure to validate the FAIR rubric against. Action: prioritise PLOS Water (New data source: PLOS Water via the PLOS API (pure R, no scraping) #14), join on DOI to PLOS OSI.

Causal-claims caveat (from the note, reinforced by the data)

Author behaviour under the policy is messy: preprint-only compliance, APCs paid from other funds, venue changes, "green-via-gold". Venue-composition shifts confound any clean "policy improved FAIRness" story. Treat funder status as one stratification variable among several, not a treatment assignment. The corpus finding above adds: for this corpus, funder status is mostly unobserved, which is a stronger limitation than confounding.

Planned work

1. Funder-coverage assessment (corpus-wide, not sample): run the Crossref funder lookup across all corpus DOIs; report coverage by journal and by year; quantify how many articles are attributable to any funder, and to Gates specifically.
2. Full-text funder extraction feasibility: on the OA subset, test rtransparent's funding-statement extraction against Crossref metadata to see how much the funder signal improves from full text.
3. Funder-policy layer: join corpus funders (where identified) to Open Policy Finder / ROARMAP / cOAlition S status, mirroring the journal-policy layers in Measure data-availability and OA practice: harvest existing open datasets before scraping #38.
4. Gates-experiment scoping: decide whether a Gates natural experiment needs a separate Gates-enriched corpus (e.g. Gates grant outputs via their reporting + PLOS OSI), rather than the current engineering-heavy corpus. Likely a distinct sub-project.
5. Version/dedup handling: add the version field and preprint->DOI dedup step to the processing pipeline regardless, since preprints will grow across all funders, not just Gates.

Honest limits

• Funder attribution from metadata covers a minority of our articles and misses Gates entirely; the historical corpus is largely unattributable.
• The Gates effect is best studied where Gates funds concentrate (global-health/PLOS venues), which is not what we scraped.
• Full-text funder extraction inherits the OA-only full-text constraint from Manual DAS-accessibility check for the 46 filtered-slice journals #35/Measure data-availability and OA practice: harvest existing open datasets before scraping #38.

Sources

• Nature news: https://www.nature.com/articles/d41586-024-00996-8
• Science/AAAS: https://www.science.org/content/article/bold-bid-avoid-open-access-fees-gates-foundation-says-grantees-must-post-preprints
• Scholarly Kitchen: https://scholarlykitchen.sspnet.org/2024/04/15/gates-2025-open-access-policy/
• Gates OA policy hub: https://openaccess.gatesfoundation.org/
• Gates-PLOS partnership: https://openaccess.gatesfoundation.org/unlocked-issue-5-april-2025/
• PLOS OSI: https://plos.figshare.com/articles/dataset/PLOS_Open_Science_Indicators/21687686
• Crossref Funder Registry: https://www.crossref.org/services/funder-registry/
• OFR->ROR transition: https://www.crossref.org/blog/roring-ahead-using-ror-in-place-of-the-open-funder-registry/
• ROARMAP: https://roarmap.eprints.org/

Refs #14, #35, #37, #38