Research Data

“Research data” is one of the most load-bearing terms in research administration, and one of the most frequently used imprecisely. Getting the definition right matters because it determines what a data management plan (DMP) has to cover, what a repository deposit is expected to include, what a records-retention policy applies to, and — in the case of federally funded work — what a research institution can be legally required to disclose under a Freedom of Information Act (FOIA) request.

The regulatory definition: 2 CFR § 200.315(e)(3)

The most concrete, citable definition of research data in U.S. research administration comes from the Office of Management and Budget’s Uniform Guidance, codified at 2 CFR Part 200.315. This regulation consolidated a set of earlier OMB circulars — including OMB Circular A-110, which contained the original version of this same definition — into a single framework effective 2014. Paragraph (e)(3) defines research data as:

“The recorded factual material commonly accepted in the scientific community as necessary to validate research findings.”

That single sentence does three things at once: it requires the material to be recorded (so it excludes things that were never written down or captured), it requires factual content (as opposed to opinion, interpretation-in-progress, or speculation), and it anchors the scope to what a scientific community would consider necessary to validate a specific set of findings — not everything generated during a project, only what’s needed to check the results.

What counts as research data

Under this working definition, research data typically includes:

• Raw and processed datasets: spreadsheets, structured databases, sensor or instrument output
• Digital records of observations: field notes once transcribed/recorded, survey responses, coded interview transcripts
• Computational outputs and the code that generates them, where that output is what a reader would need to check a published result
• Images, sequences, and other structured digital files (e.g., FASTQ files, MRI scans) that directly underlie a reported finding

What is explicitly excluded

2 CFR 200.315(e)(3) is unusually explicit about what does not count as research data, which is often the more practically useful half of the definition. Excluded categories include:

• Preliminary analyses — work product that hasn’t yet become part of a validated finding
• Drafts of scientific papers and other in-progress written work
• Plans for future research
• Peer reviews and reviewer communications
• Communications with colleagues — ordinary research correspondence
• Physical objects, such as laboratory samples or specimens — because the definition is limited to recorded material, a physical biospecimen itself is not “research data” even though records describing it may be
• Trade secrets, commercial information, and material a researcher is entitled to keep confidential until publication
• Personnel, medical, and other personally identifiable information whose disclosure would constitute an invasion of privacy

These exclusions matter most in two practical contexts: scoping what a data management plan actually needs to describe (it does not need to promise access to draft manuscripts or lab notebooks in the informal sense), and responding to a FOIA request against federally funded research, where the exclusions define the outer limit of what a requester can compel.

How this differs from a broader, working definition

Many funders and institutions use a looser working definition of “research data” in practice — effectively, any digital or physical material a researcher needs to keep in order to substantiate a published or reportable result, without drawing the sharp physical-object exclusion that the federal FOIA-driven definition draws. That broader usage is why physical specimens, biological samples, and instrument calibration records often show up in institutional data management guidance even though they sit outside the strict 2 CFR 200.315(e)(3) text. When precision matters — for example, in a FOIA response or a compliance audit — the regulatory definition governs; for day-to-day DMP scoping, most institutions use the broader sense and say so explicitly in their own policy documents.

Research data vs. types of research data

“Research data” as a term answers the question what qualifies, at all. It does not tell you how to classify a given dataset once you know it qualifies — whether it’s raw or processed, quantitative or qualitative, observational or experimental, and how sensitive it is. That classification work is a separate, necessary step for DMP drafting and repository selection; see CASRAI’s guide to Types of Research Data: A Taxonomy for Data Management Planning for the taxonomies research administrators actually use once a dataset has cleared the threshold of counting as research data at all.

Why the definition matters operationally

Scoping questions come up constantly in practice: does a lab notebook count? What about a pilot dataset that never made it into a publication? Is a spreadsheet of instrument calibration readings research data or just operational metadata? None of these has a single universal answer — it depends on whether you’re applying the strict federal FOIA-context definition or an institution’s broader working definition — but starting from the 2 CFR 200.315(e)(3) text gives a stable, citable anchor for the conversation, which is why it’s the definition most frequently referenced in U.S. research-administration policy documents and DMP guidance.