Research misconduct consists of serious violations of the standards governing the production, evaluation, and communication of research. In many regulatory systems, its core forms are fabrication, falsification, and plagiarism, commonly abbreviated as FFP. Broader institutional definitions may also encompass manipulation of peer review, serious breaches of confidentiality, or other substantial departures from accepted research practice. The applicable definition matters: conduct may be unethical, professionally sanctionable, or unlawful without falling within a particular regulator’s definition of research misconduct.
The subject lies at the intersection of research ethics, institutional governance, and the philosophy of science. Research depends on records and testimony that other people cannot routinely verify in full. Misconduct exploits this dependence, potentially corrupting the evidence used in subsequent scholarship, public policy, engineering, and clinical practice. At the same time, an incorrect result, unsuccessful replication, or methodological disagreement is not by itself evidence of wrongdoing.
Modern systems therefore distinguish the reliability of a research claim from the responsibility of the people who produced it. They combine procedures for investigating allegations with mechanisms for correcting the literature, protecting participants and complainants, and improving research environments. Their central difficulty is to preserve openness to criticism while applying fair evidentiary standards to allegations with potentially severe professional consequences.
Definitions, scope, and standards of responsibility
A widely influential definition appears in United States federal policy, which identifies research misconduct as fabrication, falsification, or plagiarism in proposing, performing, or reviewing research, or in reporting research results. Fabrication means making up data or results and recording or reporting them. Falsification means 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. Plagiarism means appropriating another person’s ideas, processes, results, or words without appropriate credit. These definitions concern more than published articles: applications, presentations, laboratory records, and review activities can also be relevant.
Under the US federal framework, a finding additionally requires a significant departure from accepted practices of the relevant research community, conduct committed intentionally, knowingly, or recklessly, and proof by a preponderance of the evidence. Honest error and differences of opinion are expressly excluded. These conditions distinguish an administrative finding from a general judgment that a study was poorly conducted. Recklessness can be sufficient even without proof of a deliberate plan to deceive, but an isolated mistake does not automatically establish recklessness.
Other frameworks organize the boundaries differently. The European Code of Conduct for Research Integrity, revised by ALLEA in 2023, identifies reliability, honesty, respect, and accountability as fundamental principles. It addresses FFP alongside a wider range of unacceptable practices, including obstructing investigations and manipulating authorship. Such a professional code does not itself operate as a single statute across Europe. Institutions, funders, disciplinary bodies, and national authorities determine how its standards are incorporated into their procedures.
Research misconduct is consequently narrower than all failures of ethics in research. A failure to obtain informed consent, an undisclosed conflict of interest, misuse of confidential information, or harassment may require investigation under separate rules. Conversely, the same episode may trigger several processes: fabricated clinical records could involve research misconduct, participant-protection violations, employment discipline, and potentially fraud. Precise classification matters because jurisdiction, evidentiary requirements, available sanctions, and rights of appeal may differ. A finding under one system does not automatically settle every question arising under another.
Fabrication, falsification, and the manipulation of evidence
Fabrication substitutes invented observations for research that purportedly occurred. Examples include nonexistent interview participants, invented laboratory measurements, or records of patient visits that never took place. The decisive feature is not that a dataset was artificially generated: simulations and synthetic datasets are legitimate when accurately described. The violation occurs when invented material is represented as observation or evidence of an activity that did not occur. In qualitative research, fabricated field notes or quotations can distort the evidentiary basis just as invented numerical measurements do in laboratory science.
Falsification alters the relationship between what was done and what is reported. It can involve changing instrument settings without disclosure, substituting samples, relabeling experimental groups, or suppressing observations that contradict a desired result. Digital images provide a prominent example. Copying a band from one gel image into another and presenting it as a separate experimental observation misrepresents the evidence. Uniform adjustments to brightness or contrast may be acceptable under journal policies when they do not obscure, eliminate, or misrepresent information. Acceptability therefore depends on both the transformation and the claim made about the resulting image.
Omission is especially difficult to classify because research necessarily involves selection. Instruments produce failed readings, interviews contain irrelevant passages, and analysts must decide how to handle missing observations. Excluding a measurement because an instrument malfunctioned can be legitimate if the reason is adequately documented. Excluding it because it weakens a preferred conclusion, while representing the remaining observations as the complete eligible dataset, may constitute falsification. Investigators must reconstruct the selection rule, its timing, its scientific justification, and the researcher’s knowledge rather than infer wrongdoing from an unfavorable exclusion alone.
Misrepresentation may also occur in computational work. A report can describe a model as tested on independent data when those data were actually used to tune it, or present outputs from one configuration as if they came from another. Such practices can make a system appear more accurate or generalizable than the evaluation supports. Yet an unnoticed programming error is not equivalent to a fabricated benchmark. Assessment requires the underlying code, data versions, records of execution, and evidence about what was known when claims were made. Across disciplines, the common issue is the integrity of the link between evidence, procedure, and representation—not whether the evidence happens to be numerical, textual, visual, or computational.
Attribution, authorship, and publication manipulation
Plagiarism concerns appropriation without adequate acknowledgment, rather than the truth or falsity of the appropriated material. A copied passage may be entirely accurate while still misrepresenting its intellectual origin. Conversely, properly citing a source does not make an unsupported claim reliable. The scope of plagiarism can include ideas and procedures as well as wording, although proving appropriation of an idea is often harder than documenting verbatim copying. Confidential grant proposals and manuscripts are particularly sensitive because reviewers receive access for a limited purpose rather than permission to use unpublished work as their own.
Plagiarism and intellectual property infringement overlap but are not identical. Copying material in the public domain without acknowledgment may be plagiarism even when no copyright is violated. Reproducing a protected figure with full attribution may still require permission or another legal justification. Textual-similarity software identifies matching strings, not the ethical or legal status of their use. Quotations, standard methods descriptions, references, translations, and conventional terminology all require contextual assessment. A similarity percentage is therefore not a finding of plagiarism.
Research authorship raises related questions of credit and accountability. Gift or honorary authorship assigns authorship without sufficient contribution; ghost authorship conceals a contributor whose role should be disclosed. Biomedical authorship standards developed by the International Committee of Medical Journal Editors combine substantial contribution, participation in drafting or critical revision, final approval, and accountability. Other fields follow different conventions, particularly for large collaborations or disciplinary practices of author ordering. Contribution statements can clarify who designed a study, curated data, performed analyses, or drafted a paper, but do not remove the need to assess responsibility for particular acts.
Publication manipulation extends beyond appropriation. Duplicate publication can make a single body of evidence appear to represent several independent studies, distorting reviews and evidence synthesis. Undisclosed reuse of an author’s own text is often termed text recycling or self-plagiarism, but its seriousness depends on context; reuse of a methods description differs from presenting previously published results as new. Citation manipulation can inflate apparent scholarly influence or condition publication on irrelevant citations.
Manipulated peer review may involve fabricated reviewer identities, undisclosed coordination, or unauthorized use of confidential submissions. Commercial “paper mills” can supply manuscripts, authorship positions, fabricated data, or other deceptive publication services. Not every authorship dispute or editorial irregularity meets an FFP definition. Nevertheless, these practices can corrupt the systems by which research is selected, attributed, and trusted, and some institutions expressly include them within broader misconduct rules.
Historical development of research-integrity governance
Concerns about unreliable or deceptive scholarship long predate modern research agencies. In Reflections on the Decline of Science in England (1830), Charles Babbage distinguished practices he called hoaxing, forging, trimming, and cooking. His categories drew attention to both invented results and selective treatment of observations. “Trimming” described adjustments that made measurements appear more consistent, while “cooking” involved selecting observations favorable to a desired result. Although these terms do not map exactly onto contemporary regulations, they show that manipulation of the evidentiary record was recognized before twentieth-century laboratory systems developed.
The growth of publicly funded research after the Second World War changed the scale of institutional responsibility. Large grants, specialized laboratories, commercial partnerships, and research involving human participants created overlapping obligations to sponsors, employers, journals, and the public. Professional trust remained important, but informal disciplinary responses were increasingly regarded as insufficient for allegations involving public funds or serious risks. Debates concerned not only dishonest individuals but also who should investigate universities and how far governments should intervene in scientific self-governance.
In the United States, prominent cases and congressional scrutiny during the 1980s contributed to formal federal oversight. The Health Research Extension Act of 1985 required institutions seeking Public Health Service funding to establish procedures for dealing with scientific fraud. The Office of Research Integrity was established in 1992 within the Department of Health and Human Services. A federal research-misconduct policy issued in 2000 subsequently provided a common FFP definition and core procedural principles for federal research agencies, although agencies retained distinct implementing regulations and responsibilities.
Other countries developed different arrangements, including national advisory bodies, statutory committees, funder-led systems, and predominantly institutional investigation. The resulting landscape is not a single international enforcement regime. A collaboration can involve several employers and funders with overlapping authority, different confidentiality rules, and different definitions of misconduct.
International statements have sought common principles without eliminating these differences. The Singapore Statement on Research Integrity, developed at the second World Conference on Research Integrity in 2010, set out responsibilities relating to honesty, accountability, professional fairness, and stewardship. It is a global guidance document rather than a binding treaty. The Montreal Statement of 2013 addressed integrity in cross-boundary collaborations, where responsibility for records, communication, and reporting can otherwise become fragmented. These developments reflect a shift from responding only to individual scandals toward defining the institutional conditions required for trustworthy research.
Prevalence, incentives, and questionable research practices
The prevalence of research misconduct is difficult to measure because confirmed cases are not a representative sample of all research. Detection depends on access to records, disciplinary scrutiny, institutional willingness to investigate, and the visibility of a publication. Administrative case counts also depend on legal definitions and reporting requirements. A rise in recorded findings may reflect more misconduct, better detection, a change in jurisdiction, or several factors at once. Neither the number of investigations nor the number of retracted papers provides a direct estimate of the proportion of researchers who commit misconduct.
Surveys offer a different but imperfect perspective. A 2009 meta-analysis by Daniele Fanelli found that approximately 2% of respondents in the included surveys admitted having fabricated, falsified, or modified data or results at least once, while roughly one-third admitted other questionable practices. These figures were not annual incidence rates, and the analysis excluded plagiarism from its central estimate of data-related misconduct. Survey wording, sampling, recall, and reluctance to confess serious wrongdoing constrain interpretation. Reports about colleagues introduce additional uncertainty because respondents may describe the same episode or mistake suspicion for established fact.
Questionable research practices occupy a broad area between clearly acceptable conduct and formally established misconduct. Examples include undisclosed flexibility in statistical analysis, presenting an exploratory hypothesis as if it had been specified in advance, inadequate recordkeeping, and selective reporting. Some are negligent, some reflect weak disciplinary conventions, and some may become falsification when accompanied by deceptive representation and the required culpability. The category is therefore descriptive rather than a uniform legal classification.
Statistical flexibility illustrates the mechanism. If researchers repeatedly examine data, try alternative outcomes, or change exclusion rules until they obtain a nominally significant result, ordinary significance thresholds no longer have their usual interpretation. Publication bias can compound this problem when positive findings are more likely to appear in print than null results. Neither phenomenon necessarily involves fabricated observations, but both can make the published record systematically misleading.
Explanations of misconduct include personal decisions, opportunities for concealment, supervisory failures, and organizational incentives. Competition for jobs, grants, publication, and commercial advantage can reward striking results while making careful negative findings less visible. These pressures do not establish that any particular researcher acted improperly, and they do not excuse deliberate deception. They help explain why prevention addresses laboratory organization and assessment systems as well as individual ethical understanding. Institutions that evaluate researchers mainly through publication counts may create risks different from those that examine substantive contributions and the quality of underlying records.
Detection, allegations, and evidentiary assessment
Misconduct allegations arise through many routes: a laboratory colleague notices impossible records, a reviewer recognizes copied material, an independent researcher finds duplicated images, or a replication attempt reveals discrepancies requiring explanation. Reproducibility can expose problems, but failure to reproduce a result is not a diagnostic test for dishonesty. Differences in materials, populations, implementation, or chance can produce divergent outcomes. Equally, a fabricated result might accidentally agree with a genuine effect, so reproducibility alone cannot authenticate the original research process.
Technical screening can identify anomalies that deserve investigation. Image analysis can reveal repeated regions, inconsistent backgrounds, or transformations suggesting that one image has been reused. Statistical examination can identify impossible values, duplicated response patterns, implausibly regular distributions, or inconsistencies between reported summaries and raw observations. Document metadata and version histories may establish when files were created or altered. Such findings are strongest when they support a specific account that can be tested against original materials.
Automated signals require caution. Legitimate procedures can generate unusual distributions, and common templates can produce repeated formatting. Benford’s law, sometimes invoked to detect fabricated numbers, applies only to suitable kinds of numerical data; many bounded experimental measurements do not satisfy its assumptions. Similarity software cannot determine attribution by itself, and an image detector cannot establish who altered a file or whether an alteration was disclosed. Detection tools identify possible discrepancies, whereas a misconduct finding requires contextual evidence and an applicable standard of responsibility.
The most informative evidence often consists of contemporaneous records: laboratory notebooks, instrument exports, sample inventories, interview recordings, source code, correspondence, and submitted manuscript versions. Preserving originals, documenting transfers, and controlling access help establish provenance and reduce disputes about subsequent alteration. Missing records can obstruct assessment, but their absence does not automatically prove the original data were invented. Investigators must distinguish records that never existed from records legitimately destroyed, accidentally lost, or withheld.
Whistleblowing presents an additional institutional challenge because complainants may depend professionally on the people they report. Confidential reporting channels and protection against retaliation can make evidence available that would otherwise remain hidden. Fair handling also requires distinguishing a good-faith allegation that is not substantiated from a knowingly false accusation. Confidentiality is generally limited rather than absolute: a meaningful investigation may require disclosure of relevant facts to respondents, witnesses, regulators, or other institutions.
Investigation, due process, and institutional jurisdiction
Formal procedures commonly separate an initial assessment, a preliminary inquiry, and a full investigation, although terminology and requirements vary. Initial assessment asks whether the allegation is sufficiently specific, potentially within scope, and supported enough to warrant further examination. An inquiry generally determines whether a full investigation is justified; it is not necessarily intended to resolve every factual dispute. A full investigation examines relevant evidence, interviews witnesses, evaluates explanations, and reaches findings under the governing rules.
The institution’s early responsibilities often include securing records and addressing immediate risks. If disputed research supports an ongoing randomized controlled trial, a diagnostic procedure, or an active grant application, protective steps may be necessary before culpability is settled. Such actions should not be confused with a final finding. Suspending use of questionable data can protect participants and funds while preserving the respondent’s opportunity to answer the allegation.
Due process in this setting ordinarily includes adequate notice of the allegations, an opportunity to respond, access to relevant evidence subject to applicable protections, and assessment by people without disqualifying conflicts. Expertise matters because a technically unusual procedure may be legitimate within its field. Independence matters because institutions may face financial, reputational, or supervisory conflicts. External committee members, sponsor oversight, or review by a national body can help address such conflicts, although arrangements vary by jurisdiction.
The standard and burden of proof are central. Under the US federal framework, misconduct must be established by a preponderance of the evidence: the relevant conclusion must be more likely true than not. This differs from the criminal standard of proof beyond a reasonable doubt. A criminal prosecution, employment proceeding, funding decision, and journal correction may therefore reach different outcomes without necessarily being inconsistent, because they address different issues under different rules.
Responsibility must also be individualized. Authorship creates obligations, but it does not establish that every coauthor participated in or knew about a particular falsification. Investigators may need to distinguish the person who generated data, the person who altered an image, supervisors who failed to oversee the work, and collaborators who reasonably relied on supplied materials. Findings should separate established facts, inferences about mental state, and recommendations. Appeal or review mechanisms can address procedural defects or disputed conclusions where the governing system provides them. Confidential personnel measures and public correction of the literature remain distinct tasks, even when they arise from the same investigation.
Correction, sanctions, and consequences for the research record
The immediate scholarly question after a serious problem is discovered is whether the research record remains reliable. Journals may issue a correction for a localized error, an expression of concern while substantial uncertainty persists, or a retraction when findings are seriously unreliable or publication was affected by specified ethical violations. Guidance from the Committee on Publication Ethics treats retraction principally as a means of correcting the literature, not simply punishing authors. Honest error can therefore justify retraction, while a misconduct finding does not automatically imply that every publication by the responsible person is invalid.
A retraction ordinarily leaves the article accessible but clearly marked, with a linked notice explaining the reason. Removing the record altogether can conceal what happened and make it harder to understand citations, disputes, and subsequent corrections. Notices are most informative when they distinguish unreliable findings from attribution problems and identify which conclusions are affected. Retraction does not necessarily mean that every observation was false; nor does the survival of some valid observations make an unreliable central conclusion safe to use.
Downstream correction is more difficult than changing a single article. Retracted findings may already appear in reviews, teaching materials, patents, policy documents, or clinical evidence summaries. Duplicate publications can also cause the same participants or observations to be counted more than once in a meta-analysis. Reassessment may therefore require tracing dependencies, removing invalid inputs, and recalculating conclusions. Citations to retracted work are not uniformly improper—some discuss the retraction itself—but uncritical reuse can perpetuate the original problem.
Institutional and regulatory sanctions serve purposes separate from bibliographic correction. Depending on the governing authority and seriousness of the case, consequences can include reprimand, required supervision, restrictions on research activities, termination of employment, exclusion from advisory service, or temporary debarment from funding. Repayment or recovery of funds and civil or criminal proceedings may arise when additional legal conditions are met. A research-misconduct finding is not itself a criminal conviction.
Consequences also extend to people who did not commit the misconduct. Collaborators may lose years of work, trainees may need to reconstruct projects, and research participants may have accepted burdens without producing dependable knowledge. Institutions may have duties to correct personnel records and communicate the outcome when allegations are not substantiated. These interests can conflict: extensive confidentiality may protect individuals but leave uncertainty about published claims, while premature public accusations can cause harm before the evidence has been tested.
Prevention, emerging technologies, and the limits of transparency
Prevention concerns the organization of research as well as individual instruction. Clear responsibility for data, realistic supervision, documented procedures, and training tied to actual disciplinary work can reduce opportunities for misrepresentation. Generic ethics courses cannot substitute for knowing how to preserve an instrument’s raw output, document changes to an analysis, or distinguish acceptable image processing from alteration of evidence. Supervisors also need workable systems for checking important results without assuming that every member of a team can independently reproduce every specialized task.
Research data management provides an evidentiary foundation. Stable identifiers, access controls, version histories, documented exclusions, and retention policies make it easier to reconstruct what happened. These systems have to respect data privacy, participant agreements, security obligations, and legitimate restrictions on sensitive materials. Public release is not the only form of accountability: controlled access or independent auditing may permit scrutiny when unrestricted sharing would be inappropriate. Conversely, a downloadable dataset can still be fabricated, so availability alone does not establish authenticity.
Open science practices address several vulnerabilities. Preregistration records planned hypotheses and analyses before outcomes are known, helping distinguish confirmatory tests from subsequent exploration. Registered reports move editorial assessment of the research question and methods before results are obtained, reducing some incentives to produce publishable outcomes. Sharing code and materials can expose errors and inconsistencies. These practices are not guarantees: a preregistered protocol can be ignored, a registered study can contain fabricated observations, and shared code may reproduce an analysis of invented data.
Artificial intelligence creates additional possibilities for both deception and detection. Generative systems can produce plausible text, images, references, or numerical records at scale, potentially lowering the effort needed to manufacture research-like outputs. They can also assist screening and comparison. The relevant distinction is not simply whether a tool was used, but whether its use was permitted and transparently represented, whether the evidence is authentic, and whether accountable researchers verified the claims. Fabricated citations or synthetic images presented as experimental observations remain misrepresentations regardless of how they were generated. Automated authorship or image-detection scores are not sufficient proof of misconduct.
The deeper issue belongs to social epistemology: research knowledge depends on a division of cognitive labor, so trust cannot be replaced by universal direct verification. Effective safeguards distribute scrutiny across researchers, institutions, funders, journals, and independent critics without treating disagreement as guilt. Important unresolved questions concern the proportionality of sanctions, cross-border coordination, the effectiveness of training, and how assessment systems can reward dependable work rather than merely conspicuous output. Evaluation of preventive measures must distinguish better detection from a genuine reduction in misconduct, since a more transparent research environment may initially reveal more cases rather than fewer.