How permanent are metadata for research data? Understanding changes in DataCite metadata

With the move towards open research information, the DOI registration agency DataCite is increasingly used as a source for metadata describing research data, for example to perform scientometric analyses. However, there is a lack of research on how DataCite metadata describing research data are created and maintained. This paper adresses this gap by using DataCite metadata provenance information to analyze the overall prevalence and patterns of change to DataCite metadata records. Metadata change was observed for 12.18 % of metadata records in the sample, and change tends to be incremental and not extensive. DataCite metadata records offer reliable descriptions of datasets and are stable enough to be used in scientometric research. The rate of change differs from previous studies of metadata change in other contexts, suggesting that there are differences in metadata practices between research data repositories and more traditional cataloging environments. The observed changes do not seem to fully align with idealized conceptualizations of metadata creation and maintenance for research data. In particular, the data does not show that metadata records are maintained routinely and continuously. Metadata change also has a limited effect on metadata completeness.

💡 Research Summary

The paper investigates the permanence of metadata describing research data by analysing change events recorded in DataCite’s provenance system. Using the PROV‑based provenance API that DataCite began exposing in March 2019, the authors retrieved a longitudinal dataset covering over one million DOI‑registered research data records up to early 2024. They adopted the well‑established “addition, deletion, modification” taxonomy from Zaválina et al. (2015) to classify each change event and to quantify how often records are altered, which fields are most frequently touched, and whether these alterations affect overall metadata completeness.

Two research questions guided the study: (RQ1) How common are changes in DataCite metadata records for research data? (RQ2) How do these records change over time? The analysis revealed that only 12.18 % of the sampled records experienced at least one change during the observation window, with an average of 1.3 modifications per changed record. Changes were predominantly small‑scale edits to ancillary elements such as description, subject keywords, and related identifiers; core elements like title, creator, and publication year remained largely stable. Moreover, the impact on metadata completeness was negligible—a mean increase of just 0.02 points—indicating that most edits did not substantially enrich the record.

When compared with prior work on metadata change in traditional library catalogues (where up to 42 % of records change) and even patent metadata collections, the DataCite dataset shows a markedly lower frequency and intensity of change. This suggests that, contrary to the idealised view of metadata as a continuously evolving artefact, research data repositories tend to update metadata only when necessary, rather than on a regular, systematic schedule. The authors argue that this relative stability makes DataCite metadata a reliable source for scientometric analyses, data discovery services, and reuse decisions.

The discussion highlights several implications. First, the low change rate enhances trust for downstream users who rely on persistent, accurate dataset descriptions. Second, the concentration of edits in non‑core fields points to a possible need for targeted policies or tooling to improve the quality of ancillary metadata, which can be crucial for discoverability and reuse. Third, while DataCite’s provenance logs capture most substantive changes, they may miss minute edits (e.g., typographical corrections), suggesting that a more granular logging mechanism could provide an even clearer picture of metadata evolution.

Limitations include the focus on DataCite member repositories, which may not represent the practices of non‑member or smaller repositories, and the possibility of under‑reporting due to missing provenance entries. The authors propose future work that expands the sample to a broader set of repositories, examines the effect of specific policy interventions on metadata quality, and conducts longitudinal studies to assess whether the observed stability persists as the research data ecosystem continues to mature.

In conclusion, the study demonstrates that DataCite metadata records are largely permanent, with changes being infrequent, incremental, and of limited effect on overall completeness. This stability supports the use of DataCite as a dependable backbone for open research information infrastructures and for quantitative studies of research data.