Data Citation Mandates Gain Momentum in Journal Policies

A colleague—call her Dr. R, an ecologist who manages a long-term field site—told me she used to treat data deposit as the last box to tick before submission. Now her journal rejection check includes a separate line: 'Dataset DOI missing from references.' Her experience matches what editors describe across the biological and earth sciences, where data availability statements have been expected for a decade but formal data citations are the new compliance target.

The FORCE11 Joint Declaration of Data Citation Principles, published in 2014, gave this movement its language. It said data should be cited according to scholarly practice, with persistent identifiers, and should appear in reference lists. That's a simple requirement with hard operational consequences: a dataset needs a stable identifier, a landing page, and a recommended citation string before it can appear in a bibliography.

From availability statements to citable objects

For most of the 2010s, a data availability statement was enough. PLOS began requiring them in 2014, and many journals followed. But a statement that says 'data available on request' isn't the same as a citation. The current policy wave closes that gap by requiring authors to cite the exact versioned dataset they used.

Springer Nature's research data policy tells authors to cite any datasets that underlie results and to make those citations resolvable. AGU journals, including Journal of Geophysical Research and Water Resources Research, have gone further by requiring dataset citations in reference lists rather than accepting them as supplementary mentions. PLOS policy applies the same logic: when a figure or analysis depends on a deposited dataset, the dataset gets its own reference entry.

The base rate and the exception

Across scholarly publishing, the base rate remains uneven. Editorial mandates are proliferating faster than author compliance. Repositories still see deposits with incomplete metadata, and journals still find reference lists that mention a database by name without a version, a DOI, or a creator. The gap is most visible in disciplines that lack a strong data repository tradition.

The exception is instructive. Journals that treat data citation as a publication output, not a compliance checkbox, see better results. AGU's requirement puts dataset references through the same proof checks as article references. That approach converts an author's afterthought into something a copyeditor will catch. It also means a data citation survives contact with production staff.

Funders have helped build the infrastructure those policies depend on. The U.S. National Institutes of Health's Data Management and Sharing Policy took effect for grant applications submitted on or after January 25, 2023. It requires a data management and sharing plan and expects deposition in established repositories. In the United States, the 2022 Nelson memo pushed federal agencies toward immediate public access, which accelerated dataset deposit and citation infrastructure. Those policies don't dictate a citation style, but they produce the deposits that journals can then ask authors to cite.

The social sciences lag in some ways. The APA Publication Manual's 7th edition, released in 2019, gave authors clear examples for citing datasets, yet many psychology journals still treat data citations as optional. A field like genomics has cited nucleotide databases for decades, because deposition is an accession-based precondition for publication.

Who feels the pressure first

Early-career researchers feel this shift most directly. A postdoc preparing a manuscript now has to chase repository metadata, version numbers, and author lists for data they may not have collected themselves. Principal investigators who built their careers before dataset DOIs existed are often the least prepared to explain to a journal editor why a shared folder link isn't sufficient. Those mandates are part of a larger enforcement wave we covered in an earlier report on data sharing mandates and enforcement.

Research data librarians are the quiet beneficiaries and the new bottlenecks. They're asked to mint DOIs, clean creator records, and fix landing pages in the week before submission. A lab that treats them as part of the manuscript team gets through the process with fewer last-minute exceptions.

Dataset authors gain something tangible. A well-formed data citation travels through DataCite metadata and appears in DataCite's registry, where it can be linked to articles, grants, and ORCID records. That's a visible research output in a system that has spent decades ranking only papers. The FORCE11 Joint Declaration remains the clearest public statement of why that matters.

What this means for your lab

First, treat dataset citation as part of the manuscript, not a task for a graduate student at submission. Paste the repository's recommended citation string into your reference manager before you write the results section. The DOI should resolve, the version should be explicit, and the author list should reflect who contributed to data collection, not only who wrote the paper.

Second, make sure data outputs appear in annual evaluations and promotion files. Some hiring and tenure committees now ask candidates to list datasets with DOIs separately from articles. If your department's form doesn't have that line, ask why. The question takes five minutes and signals that the department is reading the same journal policies its researchers are being asked to follow.

Third, prepare for metadata friction. Repository platforms use different citation formats, and some journals still accept accession numbers without DOIs. Keep a plain-text file with each dataset's title, creators, DOI, version, and landing page. That file is the difference between a ten-minute reference fix and a lost afternoon.

Persistent identifiers for people are part of the same build-out. ORCID iD mandates now extend to some journals' peer reviewers and editors, as we reported in our coverage of ORCID iD mandates. When the dataset DOI and the researcher's ORCID iD are recorded together, a data citation stops being an anonymous string and becomes an attributable output.

The metric problem isn't solved

Data citations remain harder to count than article citations. Aggregators can count a dataset DOI in a reference list, but they often miss citations buried in text, code, or supplementary files. Make Data Count, a project involving DataCite and California Digital Library, has worked to standardize usage and citation metrics for data. The underlying principle is straightforward: if datasets can't be counted reliably, researchers won't list them.

Publishers have a role here that extends beyond editorial policy. When a production system strips a dataset reference into a free-text note, the citation becomes invisible. Journals that persist DOIs through typesetting and deposit them with Crossref are the ones whose data citations actually register.

There is a base rate worth stating plainly: most research datasets still receive no citation at all, even in journals with clear policies. Mandates change the submission workflow; they don't yet change scholarly habit. The exceptions are fields such as crystallography and genomics, where deposition to a recognized archive is a precondition for publication and citation practice is older than the term 'data citation.'

A next step that costs an afternoon

Take one dataset your group plans to publish this year. Deposit it in a repository, get a DOI, and copy the recommended citation into the manuscript. Then open the DOI in a plain browser with no institutional login. If it resolves and shows the version you expect, you've done more than most labs will do this month. If it doesn't, you've found the problem before a journal editor does.

This is a backlog problem, not a culture problem to be solved by another workshop. Start with the next dataset, not the archive you've been avoiding.

Photo by Brecht Corbeel on Unsplash