Stronger funder mandates and international teams push authors toward permissive Creative Commons licences in top hybrid journals, but high‑prestige journals and the medical field more often keep restrictive terms, constraining reusable scholarly material for AI training.

Main Finding

Stronger funder open-access mandates and international collaboration increase the likelihood that authors choose more permissive Creative Commons (CC) licenses for articles in hybrid Q1 journals, while higher journal prestige and discipline-level norms (notably in medicine) lead authors to select more restrictive CC licenses.

Key Points

• Sample: 122,085 open‑access articles published in Q1 hybrid journals in 2024.
• Method: multinomial logistic regression predicting CC license choice.
• Key predictors examined: funding agency mandate strength, presence of international collaboration, subject category, and journal impact-factor percentile.
• Directional results:
  • Stronger funding agency policies → higher probability of choosing permissive licenses (e.g., CC BY).
  • International collaboration → associated with more open licensing.
  • Higher journal prestige (higher impact-factor percentiles) → associated with more restrictive licenses.
  • Discipline effects: medicine tends toward more restrictive licensing relative to humanities.
• Interpretation: both institutional pressure (funders, collaborators) and journal prestige norms shape license selection decisions.

Data & Methods

• Data: 122,085 OA articles from hybrid Q1 journals, year 2024 (dataset covers CC license metadata, funding acknowledgements, author affiliations, subject categories, and journal metrics).
• Dependent variable: categorical CC license chosen at publication (ordered from permissive to restrictive; typical categories include CC BY, CC BY-SA, CC BY-NC, CC BY-ND, CC BY-NC-ND, and possibly CC0/no-CC).
• Independent variables:
  • Funding agency mandate level (coded by strength/requirement of open-license policy).
  • International collaboration (binary or count of foreign coauthor countries).
  • Subject category (discipline fixed effects; medicine and humanities highlighted).
  • Journal impact-factor percentile (continuous or percentile bins as prestige measure).
• Statistical approach: multinomial logistic regression to estimate relative probabilities of selecting each license category conditional on predictors; controls likely included to account for confounders (not fully enumerated in the summary).
• Robustness: implied but not detailed — typical checks would include alternative specifications, subsample analyses by field, and sensitivity to prestige measures.

Implications for AI Economics

• Availability and legal permissibility of training data:
  • More permissive CC licenses (e.g., CC BY) reduce legal frictions for using scholarly text and figures in AI model training and downstream services; restrictive licenses (NC/ND) limit reuse and derivative model outputs.
  • Heterogeneous licensing by journal prestige and field implies uneven access to high-prestige and medical content for AI training, affecting model coverage and potential biases.
• Market and pricing effects:
  • Stronger funder mandates that promote permissive licensing can lower transaction costs for AI firms and public actors that rely on large-scale scholarly corpora, potentially increasing competition and lowering prices for AI-enabled scholarly tools.
  • Publishers of high-prestige journals retaining more restrictive licensing may preserve exclusive control and premium pricing for value-added products, affecting platform competition and vertical integration incentives.
• Incentives and welfare:
  • Institutional levers (funders, international consortia) are effective policy tools to shift licensing toward openness; economists can model welfare trade-offs between publisher revenues, author incentives, and social returns from broader reuse.
  • Field-specific norms (e.g., medicine’s restrictiveness) reflect non-price considerations (privacy, liability, clinical sensitivity) that should factor into policy design and economic models assessing optimal openness.
• Policy design and regulation:
  • To increase usable open data for AI, policymakers should couple OA mandates with explicit license requirements (e.g., require CC BY for funded outputs) and consider enforcement/monitoring mechanisms.
  • Encourage international collaboration policies that indirectly promote more open licensing.
• Directions for further AI-economics research:
  • Quantify how license heterogeneity changes the effective supply of trainable scholarly data and impacts model performance, innovation, and entry by AI startups.
  • Model publisher pricing strategies when faced with stronger funder mandates and potential loss of licensing rents.
  • Estimate welfare implications of different licensing regimes accounting for spillovers to downstream AI products and societal benefits from reproducibility and access.