Research Misconduct: Fabrication, Falsification and Plagiarism

Research integrity is the expectation that a study's data, methods and conclusions are what the author claims they are. The most widely used definition of research misconduct comes from the United States Office of Research Integrity, which oversees federally funded research and defines misconduct under 42 CFR Part 93 as fabrication, falsification or plagiarism in proposing, performing or reviewing research, or in reporting research results.

That three-part structure, shortened to FFP, has become the working definition adopted, with local variation, by journals, universities and funders across most jurisdictions that regulate research conduct.

FFP is deliberately narrow. It does not cover honest mistakes, differences of scientific judgement about how to analyse a dataset, or disputes about who deserves authorship credit.

A researcher who mislabels a variable by accident, or who reasonably chooses one statistical model over another, has not committed misconduct even if the choice later proves wrong. What separates misconduct from error is intent: FFP requires that the act be committed knowingly, intentionally or with reckless disregard for the truth of the record, and that the finding be proven by a preponderance of evidence rather than reasonable doubt.

In criminology this distinction matters because so much of the field's evidence base sits outside laboratory conditions. A researcher coding police case files, running a victimisation survey or interviewing former offenders makes hundreds of judgement calls about coding, exclusion and interpretation. The FFP framework exists to separate that ordinary interpretive latitude, which is a normal feature of

types and purposes of crime research, from the small set of acts that break the basic contract between a researcher and their audience: that the reported record reflects what was actually done.

Coverage varies across jurisdictions even though the underlying FFP concept travels well. In the United States, federal agencies that fund research require institutions to have a misconduct policy consistent with the ORI definition as a condition of receiving grants. In the United Kingdom, individual universities and the UK Research Integrity Office set out their own procedures rather than following a single statutory definition, though most converge on fabrication, falsification and plagiarism as the core categories.

India regulates the closest analogue through the University Grants Commission, which since 2018 has required every higher education institution to adopt a plagiarism policy for theses and papers, though it leaves fabrication and falsification largely to each institution's own disciplinary code. Journals sit above all three national systems and increasingly apply a shared set of editorial standards, most influentially those published by the Committee on Publication Ethics.

Fabrication is the most direct form of misconduct because it invents the evidentiary record outright. A fabricating researcher reports survey responses that no respondent gave, interview transcripts from conversations that never took place, or crime statistics pulled from no actual dataset. Unlike falsification, which distorts real material, fabrication starts from nothing and manufactures the appearance of a study.

The best documented case in the behavioural sciences is Diederik Stapel, a Dutch social psychologist whose work on priming and social cognition was widely cited before an internal investigation at Tilburg University found in 2011 that he had fabricated data for dozens of published studies, often without running the experiments he described at all.

The case is instructive for criminology because Stapel's fabricated findings had already been cited approvingly in downstream research before the fabrication surfaced, showing how quickly invented evidence can be absorbed into a field's accepted knowledge once it clears peer review.

Fabrication is comparatively rare next to falsification and plagiarism because it is also the riskiest form of misconduct to sustain. A fabricated dataset has to remain internally consistent across every table, and it has to survive a request from an editor, a co-author or a replicating team to see the raw material.

Statistical forensics, which checks whether reported numbers show the natural variability of real measurement, has become one of the routine tools used to flag datasets that look too clean to be genuine.

In a criminal justice research setting, fabrication can be especially consequential when the invented material concerns something with legal weight, such as a claimed sample of case files, a fabricated offender interview used to support an expert opinion, or invented recidivism figures used to justify a sentencing or parole policy.

Because such claims often cannot be checked against an independent record, criminological fabrication can persist undetected for longer than fabrication in fields where raw data or specimens are routinely archived and shared.

The Stapel case also shows how fabrication is usually caught. It was junior colleagues, not an external auditor, who first noticed that his datasets arrived unusually quickly and with implausibly clean patterns, and who raised those concerns internally before the university opened a formal inquiry.

That pattern, where the first signal comes from a co-author, student or research assistant close to the work rather than from routine oversight, recurs across documented fabrication cases and is one reason universities now build confidential reporting channels into their research integrity policies rather than relying only on external audit, since an outsider reviewing a finished manuscript rarely has any way to see how the underlying data were actually generated.

Falsification differs from fabrication in that the underlying study is real. Data were collected, an experiment was run, or records were coded, but the researcher then changes, omits or selectively presents that material so the published account no longer matches what actually happened. Common forms include altering values in a dataset, deleting inconvenient outliers without disclosure, cropping or adjusting an image or chart to exaggerate an effect, or dropping cases that do not support the intended conclusion.

A widely cited example of falsification through selective reporting and undisclosed data manipulation is Andrew Wakefield's 1998 paper in The Lancet linking the MMR vaccine to autism in a small case series.

The United Kingdom's General Medical Council found in 2010 that Wakefield had acted dishonestly and irresponsibly in how the study was conducted and reported, and The Lancet retracted the paper the same year; a later investigation by journalist Brian Deer, published in the BMJ in 2011, documented how case details and the timing of children's symptoms had been altered to fit the paper's conclusions.

Although the case sits in medicine rather than criminology, it illustrates a pattern that recurs across the social sciences: a single falsified study, once it reaches media and policy audiences, can shape public behaviour for years after the underlying claim collapses.

Falsification in criminological work often takes quieter forms than a fabricated headline finding. A researcher evaluating a policing intervention might report only the precincts where crime fell and omit the comparison sites where it did not. An analyst working with official crime statistics might recode borderline offences into a category that flatters a preferred conclusion.

Because these choices can be dressed up as ordinary analytic discretion, falsification is often harder to prove than fabrication, and investigations typically require access to the original data, coding sheets and any pre-registered plan to establish that a change was made deliberately rather than reasonably.

Falsification also covers the manipulation of images and figures, a category that has grown as more disciplines rely on scanned documents, digital case files and forensic imagery. Splicing together frames from different sources, adjusting contrast to obscure an inconsistent reading, or duplicating a chart element to imply a second independent measurement all count as falsification even where no numeric value in a table was directly altered, because the manipulation changes what the reader is entitled to believe the image shows.

The consequence for criminology is that falsified findings tend to enter the literature wearing the same methodological language as legitimate research. Distinguishing the two usually requires the kind of scrutiny covered under

statistical methods in crime research, including checks for implausible precision, inconsistent sample sizes across tables, and results that cluster suspiciously close to a conventional significance threshold.

Plagiarism is the presentation of someone else's words, ideas, data or findings as one's own without adequate attribution. It ranges from verbatim copying of another author's sentences, through close paraphrasing that keeps the original structure and argument while swapping surface wording, to the unattributed use of a dataset, a coding scheme or a theoretical argument that another researcher developed. All three forms breach the same underlying norm: that credit follows the person who did the intellectual work.

Self-plagiarism, sometimes called text recycling, occurs when an author reuses substantial portions of their own previously published text, data or findings in a new publication without disclosure, and presents the work as new. It is treated as a form of plagiarism because it misrepresents the originality of the new output and, where the same dataset is published as if it were an independent finding, it can inflate the apparent weight of evidence on a given question.

A related practice, salami slicing, splits one dataset or study into the minimum number of publishable units to maximise a publication count, and is treated as a questionable research practice for the same reason: it distorts how much genuinely new evidence exists.

Universities in several jurisdictions now define plagiarism formally rather than leaving it to informal convention. In India, the University Grants Commission's 2018 regulations on the promotion of academic integrity set out similarity thresholds for theses and papers and require every higher education institution to run submissions through similarity-detection software before evaluation.

In the United Kingdom and the United States, individual universities set their own plagiarism policies, but most converge on the same core test: whether a reasonable reader would believe the words or ideas originated with the person presenting them.

Detection has become largely automated. Text-similarity software compares a submitted manuscript or thesis against indexed publications and prior submissions and flags overlapping passages for a human reviewer to assess in context, since a high similarity score can also come from properly quoted material, standard methodological phrasing or a shared literature review.

Because the software flags a pattern rather than making the final judgement, a finding of plagiarism still depends on an editor or an institutional panel establishing that credit was genuinely withheld.

Self-plagiarism sits closer to a judgement call than the other categories, because reusing a short methods description across two papers from the same long-running study is normal and disclosed practice, while republishing a full results section as though it were a new dataset is not.

Journals typically ask authors to disclose any prior publication that overlaps substantially with a new submission and to cite it directly, which converts a potential self-plagiarism problem into an ordinary, transparent case of building on earlier work.

Between clean research conduct and outright FFP sits a wider set of questionable research practices, often shortened to QRPs, that fall short of the formal definition of misconduct but still distort what a study appears to show. QRPs are typically legal within a researcher's institution and rarely trigger a formal investigation on their own, yet they push published findings toward exaggerated or false-positive effects across an entire field.

P-hacking describes trying multiple analytic choices on the same dataset, different subgroup splits, alternative outlier cutoffs, several outcome measures, additional control variables, until one combination crosses a conventional statistical significance threshold, and then reporting only that combination as though it had been the single planned analysis. Simmons, Nelson and Simonsohn's 2011 paper in Psychological Science demonstrated how much this kind of undisclosed analytic flexibility inflates false-positive rates, even when every individual researcher decision looks defensible on its own.

HARKing, a term coined by Norbert Kerr in 1998, stands for hypothesising after the results are known: presenting a hypothesis that was actually formed after seeing the data as though it had been predicted in advance. HARKing misrepresents quantitative and qualitative methods in crime research as confirmatory when it was really exploratory, which matters because a hypothesis confirmed by the same data that generated it carries far weaker evidential weight than one tested against independent data.

These practices share a structural cause: a publication system that rewards novel, statistically significant, positive findings over null results or replications. Two responses have gained traction as a result. Pre-registration requires a researcher to file their hypotheses, sample size and planned analysis in a public registry before collecting or examining outcome data, which closes off HARKing and most forms of p-hacking after the fact.

Open data policies, which ask authors to deposit anonymised datasets and analysis code alongside a publication, let other researchers check the reported result against the original material directly, rather than relying on trust in the write-up alone.

Salami slicing, splitting one dataset or study into the smallest publishable pieces to maximise the number of resulting papers, belongs in the same category. It rarely misstates any single finding, but it obscures how much genuinely independent evidence actually exists on a question, since several thin papers drawn from one survey can look, to a reader scanning a reference list, like several separate confirmations of the same claim.

Journal editors increasingly ask authors to disclose related submissions from the same dataset for this reason, so that reviewers can judge the novelty of each piece rather than treating overlapping work as independent replication.

Most institutions handle a misconduct allegation in staged fashion rather than a single hearing. An initial inquiry establishes whether the allegation is specific and credible enough to warrant a full investigation. If it clears that bar, a formal investigation panel, usually made up of researchers from outside the accused person's immediate team, examines the original data, laboratory or field notebooks, correspondence and any prior versions of the manuscript, and reaches a finding on the balance of the evidence.

A finding of misconduct can lead to correction or retraction of the published work, restrictions on future funding, and disciplinary action up to dismissal, with the exact sanctions set by the employing institution and funder rather than by a single global standard.

The Committee on Publication Ethics, known as COPE, was founded in the United Kingdom in 1997 as a forum for editors dealing with suspected misconduct and now publishes widely used flowcharts and guidance that journals across most disciplines, including criminology, follow when deciding whether to correct, retract or issue an expression of concern about a published article.

A retraction is the formal step of withdrawing a published article once its findings can no longer be relied on, and it remains attached to the original publication record rather than quietly disappearing, so that anyone who later encounters the retracted study through a citation or a database search can see that it was withdrawn and why.

Much of what surfaces first comes from whistleblowers, meaning colleagues, students or collaborators who notice an inconsistency and report it. Because a whistleblower often reports on someone with more institutional power than themselves, most universities and funders commit to keeping the identity of a good-faith reporter confidential during an inquiry and to protecting them from retaliation such as demotion or exclusion from future work, though the practical strength of that protection varies by institution and jurisdiction.

Reporting in bad faith, meaning an allegation known to be false or made recklessly to damage a colleague, is treated separately and can itself be sanctioned.

The cost of undetected misconduct in criminology is not confined to the academic record. Sentencing frameworks, risk assessment tools and policing strategies are often justified by reference to published evaluation studies, so a fabricated effect size or a falsified crime trend can shape practice long before a retraction catches up with it.

Because criminological findings frequently inform policy faster than the ordinary self-correction cycle of replication and retraction can operate, the field has leaned toward pre-registration, open data and mandatory integrity training as preventive measures rather than relying on detection after publication alone.