Research Misconduct, Questionable Practices, and Honest Error

A paper can be wrong without being fraudulent. A researcher can make a serious mistake without committing research misconduct. A statistically questionable choice can weaken a result without satisfying the legal definition of falsification.

These distinctions matter because integrity concerns affect both the reliability of the record and the rights and reputations of everyone involved. Loose use of the word “fraud” can bypass evidence. An overly narrow focus on fraud can also ignore routine practices that distort the literature.

US Public Health Service policy provides a formal category for fabrication, falsification, and plagiarism. Research institutions and journals also address a wider range of detrimental or questionable practices. Honest error and scientific disagreement occupy a different category again.

Key points#

• Under 42 CFR Part 93, federal research misconduct is fabrication, falsification, or plagiarism in the research record or research process covered by the rule.
• A finding also requires a significant departure from accepted practices, an intentional, knowing, or reckless state of mind, and proof by a preponderance of the evidence.
• Honest error and honest differences of opinion are expressly excluded.
• Selective reporting and poor documentation can damage research without always meeting the federal misconduct threshold.
• Institutions use staged procedures to assess allegations while protecting evidence, confidentiality, fairness, and nonretaliation.

The formal federal definition#

The Office of Research Integrity, or ORI, oversees the Public Health Service framework in 42 CFR Part 93. The regulation defines three forms of research misconduct.

Fabrication is making up data or results and recording or reporting them. Inventing participants, laboratory measurements, interviews, or observations can fit this category.

Falsification is manipulating research materials, equipment, or processes, or changing or omitting data or results, so that the research is not accurately represented in the research record, and the effect on the record matters. A legitimate, disclosed data-cleaning rule is not the same as removing observations to create a false account.

Plagiarism is appropriating another person's ideas, processes, results, or words without appropriate credit. The current rule contains definitions and boundaries that should be consulted for a specific case. Authorship disputes and ordinary credit disagreements do not automatically become federal plagiarism findings.

The regulation covers misconduct in proposing, performing, or reviewing research, or in reporting research results, within its jurisdiction. Other funders, agencies, employers, publishers, and professional bodies may apply their own policies; the federal definition is not the only standard that governs research behavior, and you should not read it as if it were.

A finding needs more than one of three labels#

It is not enough to point to a changed value and call it falsification. Under the rule the conduct must represent a significant departure from accepted practices of the relevant research community, it must have been committed intentionally, knowingly, or recklessly, and it must be proved by a preponderance of the evidence, meaning the evidence shows it is more likely than not.

These elements separate misconduct from negligence, ambiguity, and ordinary scientific revision. Accepted practice can be field- and context-specific, so institutions may use expert input. Intent is inferred from evidence, chronology, records, explanations, and patterns; it should not be guessed from an unfavorable outcome.

The policy expressly excludes honest error and differences of opinion, and a mistaken formula, mislabeled specimen, failed instrument, or interpretation later rejected by the field can be consequential and require correction without being misconduct. Repeated carelessness may violate other policies even when the federal state-of-mind element is not met.

The 2024 final rule became effective on January 1, 2025, with its general institutional compliance date on January 1, 2026. As of July 15, 2026, discussions of PHS-covered procedures should use the current rule and ORI guidance, not rely only on the superseded 2005 text.

The wider category of detrimental research practices#

The National Academies uses “detrimental research practices” for behavior that damages integrity but may fall outside fabrication, falsification, and plagiarism. Examples can include misleading statistical practices, inadequate data retention, and improper authorship conduct. They can include poor supervision, undisclosed conflicts, and irresponsible communication, depending on facts and applicable policy.

“Questionable research practices” is also widely used, but it is not one fixed legal category. Some choices are plainly unacceptable; others are defensible in one design and misleading in another. The label should not replace a description of what happened.

Selective reporting illustrates the problem. A study may measure many outcomes or run many analyses and publish only favorable ones, and if the omissions make the record inaccurate and were intentional, knowing, or reckless, facts could potentially support falsification. In other cases, selective reporting may violate a protocol, journal policy, or accepted standards without meeting every federal element. The evidence and jurisdiction decide, not the nickname. Other examples include changing an outcome after seeing results without disclosure, stopping data collection according to unreported results, presenting exploratory analyses as prespecified, excluding inconvenient observations without a defensible rule, or reporting p values without accounting for extensive testing, and any of them can inflate the certainty you are shown even when no federal misconduct finding ever follows.

Honest error still requires a response#

Exclusion from misconduct does not mean an error should be ignored. The research record may need a correction, retraction, replacement analysis, or notice. Participants, collaborators, funders, regulators, or clinicians may need timely information when safety or decisions are affected.

A healthy integrity culture makes error correction possible without forcing every admission into an accusation; researchers should preserve original data, document changes, report mistakes promptly, and distinguish corrected results from original results. Institutions should evaluate impact separately from culpability.

An error can be high impact and low culpability. Misconduct can be deliberate even if its measurable impact is limited. Conflating impact and intent produces both underreaction and overreaction.

How a concern is assessed#

Procedures vary with jurisdiction and institutional policy. But the PHS framework separates an assessment, an inquiry, and, when warranted, an investigation. An assessment determines whether an allegation falls within the policy and is sufficiently credible and specific to move forward, and an inquiry conducts an initial review to decide whether a full investigation is warranted. An investigation develops and examines the factual record and reaches findings under the required standards.

This is not a vote on whether a paper is persuasive. It is a documented process that preserves evidence, gives the respondent notice and opportunities to address relevant material, and manages conflicts of interest. Institutions have duties concerning records, reports, timelines, and notification to ORI in covered matters.

Confidentiality has limits but remains important. Premature public identification can harm complainants, respondents, witnesses, and the inquiry itself, and retaliation against people who raise concerns in good faith or cooperate with review undermines reporting and may violate policy. Institutions may also need immediate protective actions before a final finding, for example to secure records or protect participants, and those actions should not be described as proof of guilt.

How to evaluate a public integrity claim#

Begin with the exact alleged act. “The statistics look wrong” is not specific. Was a value invented, altered, omitted, copied, or analyzed under an undisclosed but arguable choice? Which version of the data or protocol are you comparing it against?

Then ask which rule applies. Was the work supported by PHS, another federal agency, a private funder, or an institution? Is the issue before a journal, employer, court, regulator, or professional body? Each can have a different scope and remedy.

Separate evidence about the research result from evidence about intent. A failed replication challenges a claim but does not prove fabrication. An impossible dataset pattern may justify investigation but still requires provenance and explanation.

Use procedural language accurately. An allegation, inquiry, and investigation are different events. So are a finding, appeal, correction, and retraction. A retraction can result from honest error, unavailable data, authorship or ethics problems, or misconduct. It is not a universal synonym for fraud. Avoid naming or accusing people without reliable, appropriately public findings. Scientific critique can focus on methods, data, and claims while formal processes determine individual responsibility.

Prevention is broader than policing#

Good systems reduce ambiguity and opportunity. Prospective protocols, version control, and audit trails all help. So do clear authorship criteria, data retention, and code review. So do conflict disclosure, mentorship, and channels for raising concerns.

Teams should document who can change data, how exclusions are approved, how primary analyses are locked, and how corrections are handled; journals and funders can require complete reporting and make null findings easier to publish. Institutions should train both mentors and trainees in field-specific practices, not only definitions.

Incentives matter. Rewarding only novel, positive, fast results encourages selective narratives. Valuing careful methods, replication, data stewardship, peer review, and correction supports integrity before a case reaches an office.

Limits and cautions#

This article summarizes a US federal framework and general research-integrity concepts. Definitions and procedures can differ across agencies, countries, employers, contracts, and journals.

The boundary between a detrimental practice and federal falsification can be fact-intensive. If you have a real matter, consult the current regulation, ORI guidance, and the responsible institutional office.