The data behind funding impact: evaluating the discrepancies of funding–publication linkages across data sources

Funding agencies and bibliographic databases compile funding information to better understand the relationship between research funding and scientific outputs, and these data are increasingly used in research. For analyses based on funding records to be meaningful, the underlying links between funding and research outputs must be consistent and reliable. However, existing studies have often treated funding data as given, overlooking differences in how databases collect funding metadata, such as through researcher self-reporting or text extraction from publications, which can lead to substantial variation in coverage. This study addresses three questions by comparing the funding records of Korea's National Science and Technology Information (NTIS) with Web of Science and OpenAlex. First, to what extent do these three databases cover publications that received Korean public research funding, and their associated funding information? Second, how and to what extent is the coverage or representational fidelity of funding metadata constrained? Third, how are funding statements constrained or distorted by external factors? The results reveal substantial inconsistencies across data sources, such that relying on a single database can produce markedly different funding–publication landscapes. Publisher restrictions on metadata disclosure, optional acknowledgment policies, and incomplete standards for reporting funding information contribute to missing or inconsistently extracted metadata. In addition, co-author agreements regarding acknowledgments may lead to omitted funding statements, while limited verification of reported outputs by funders may incentivize strategic reporting that links less relevant publications to funding. Methodologically, this study introduces a funding-information-level approach that treats funding information itself as the unit of analysis and provides a framework for assessing the reliability and interpretability of funding–output relationships across data infrastructures.