Session status: data-availability corpus scraped and merged; methods reframed around reuse

[larnsce]

Status note capturing a working session on the data-availability corpus, so the state and decisions are not lost. Links the pieces together; not a work item itself.

Shipped (merged via #36)

• Whole-journal scrape of three IWA journals: Journal of Water and Health (2,013 articles), AQUA (1,819), Water Supply (4,884) = 8,716 articles. Every article captured, not only WASH-matched ones, so data-availability practice can be compared within each journal.
• Quality: ~100% DOI/keywords/supplements, 99%+ authors, 0 duplicate DOIs, 0 parse failures.
• Reusable parameterized IWA scraper (data-raw/iwa_scraping.R, run_iwa_scrapes.R): discovers issues from the site's own widgets, paces under the Silverchair throttle, resumable, survives headless-Chrome crashes (2 crashes over the ~2-day run, both auto-recovered, 0 aborts).
• Also wired OPENALEX_API_KEY into harvest_corpus.R (raises the OpenAlex daily budget from $0.10 to $1).
• Closes the scraping work for Index Journal of Water and Health for data availability #32, Index AQUA (J. Water Supply Research and Technology) for data availability #33, Index Water Supply (Water Science & Technology Water Supply) for data availability #34.

Headline finding

DAS presence is near-0% before ~2019, rising to ~90-98% by 2021-2022, consistently across all three journals. This temporal shift is exactly the within-journal signal the whole-journal indexing was built to capture.

Methods discoveries (reframed the remaining plan) - three open issues

The larger outcome was realising much of the planned build already exists as open tools/datasets. Shift from "build our own" to "reuse what exists, scrape only the gaps, validate against the overlap."

• WASH-relevance match overestimates share: whole-article screening options #37 - the WASH-relevance measure (title+abstract keyword match) overestimates share, badly for large general journals; screening options documented.
• Measure data-availability and OA practice: harvest existing open datasets before scraping #38 - a three-layer methods design (OA permitted / data required / behaviour observed) built from open sources: PLOS Open Science Indicators, AAAS/Dryad, DataCite, TOP Factor, Jisc Open Policy Finder, plus runnable open R detectors (ODDPub, rtransparent) and the French Open Science Monitor as a template.
• Funding landscape of the corpus: funder attribution, policies, and the Gates 2025 natural experiment #39 - the funding landscape. Empirical finding: funder metadata is missing for most of our corpus (OpenAlex ~0%, Crossref ~7-46% and only for recent years), and Gates funding does not appear in our journals (it concentrates in global-health/PLOS venues we did not scrape). The Gates 2025 OA-policy natural experiment is worth doing but needs a Gates-enriched corpus slice, not this one. Incorporates the Gates policy note and its version/dedup implications.

Open / not done

• minor fix #6-equivalent: build the three datasets into the R package - paused, ready to start (schema + wash_matched join + country re-derivation + dictionary + roxygen + README + devtools::check). Two processing reminders recorded: read raw CSVs with explicit col_types (else DAS text is silently dropped by the type guesser); re-derive first_author_affiliation_country for old-era pages.
• Manual DAS-accessibility check for the 46 filtered-slice journals #35 - DAS-accessibility survey: now reframed by Measure data-availability and OA practice: harvest existing open datasets before scraping #38 (harvest existing before scraping); residual scope only.
• Action for @larnsce: register a Jisc Open Policy Finder API key (store as OPENPOLICYFINDER_KEY in ~/.Renviron, matching the OpenAlex pattern) to unlock the OA-policy layer in Measure data-availability and OA practice: harvest existing open datasets before scraping #38.

One-line: data collection is finished and merged; in the process we found that much of the remaining build is already solved by existing open tools/datasets, so the next phase is reuse-and-validate rather than build-from-scratch.

Refs #32, #33, #34, #35, #37, #38, #39

[larnsce]

Meeting notes (plain summary)

These are the notes for the 5 minute briefing. They cover one working session.

What we finished

We scraped three water journals in full: the Journal of Water and Health, AQUA, and Water Supply. We took every article in each journal, not only the ones about water, sanitation, and hygiene. We did this so we can compare data sharing between the water articles and the other articles in the same journal.

We collected 8,716 articles. The data is clean. Almost every article has its DOI, authors, and keywords, and there are no duplicates.

We found a clear pattern. Almost no articles had a data availability statement before about 2019. After that the share rises fast, to around 90 to 98 percent by 2021 and 2022. All three journals show the same pattern. This is the main thing the full scrape was meant to show.

The scraping code is finished and merged into the main project. It can be reused for other journals on the same platform. It ran on its own for about two days. Twice the browser crashed, and the code restarted and carried on without losing data.

What we learned that changed the plan

The larger result of the session was realising that we were about to build things that already exist. We opened three issues to record what we found.

First, our way of measuring how much of a journal is about water, sanitation, and hygiene overcounts. It matches keywords in the title and abstract only, so large general journals look far more relevant than they are. We wrote down how to fix it.

Second, we do not need to build our own tool to detect data sharing. Open datasets from PLOS and Science already measure it, and there are free R tools that do the detection. There is even a national monitor in France that does the whole thing with open code. So the plan changed to reuse what exists and only scrape what is missing.

Third, the funding angle is harder than we hoped. We checked our own articles and found that funder information is missing for most of them. It is better only for recent years. Gates Foundation funding does not appear in our journals at all, because Gates funds global health journals that we did not scrape. So the Gates open access study is worth doing, but it needs a different set of papers.

Where things stand

The scraping is done and merged. Building the three journals into the R package is paused and ready to start. Three research issues are open with the reuse first plan.

There is one small thing to do. We need to register for a Jisc Open Policy Finder key so we can pull journal and funder open access policies, in the same way we did for OpenAlex.

One line for the meeting: the data collection is finished, and along the way we found that much of the planned work is already solved by open tools and datasets, so the next phase is to reuse and check rather than build from scratch.