SeqDesk · original analysis

Do published papers actually share their data?

Two whole-index desk-checks of open-data practice, re-counted weekly: the share of open-access papers that link at least one dataset (Europe PMC), and how much of DataCite's 130-million-DOI corpus carries a machine-readable licence. Aggregate counts only.

Each point is the share of that year’s open-access papers that Europe PMC can link to at least one dataset — a text-mined accession or a data-availability statement pointing at a repository. The share has climbed from 35% in 2010 to 51% in 2025: data a detectable data link is now present for the majority of open papers, though still far from universal. The denominator is deliberately the open-access corpus — the papers whose full text can actually be inspected for data links; counting paywalled abstracts as “no data” would conflate access with a measurement limit. The most recent year is still in progress (marked * in the table below).

A licence is only half the story. Across DataCite’s 130M DOIs, 70.6% carry some rights statement — but only 30.1% attach a standard SPDX identifier a machine can resolve and act on. The 41-point gap is free-text rights that a human can read but a reuse pipeline cannot, the open-data analogue of the “present but not usable” metadata gap.

Two licences dominate the machine-readable set: CC0 (an explicit public-domain dedication) and CC BY. The long tail of NonCommercial and NoDerivatives variants adds reuse conditions that restrict large-scale text-mining and redistribution. Shares are of the 39M DOIs that carry an SPDX identifier.

The open-science counterpart to the archive counts: not how much data exists, but whether published work actually shares it and licenses it for reuse — re-counted weekly straight from the EMBL-EBI / Europe PMC and DataCite REST API. Only aggregate percentages and counts are published. · back to all research data