Fabrication, Falsification, and Selective Reporting in Research

Hello. In the previous lesson, you saw how original records, version histories, and documented decisions make a finding traceable. Those records also help answer a different question: Does the account presented to readers faithfully represent what happened? A dataset can be carefully stored while its published account is still misleading.

This lesson distinguishes fabrication, falsification, and selective reporting. The first two concern invented or misrepresented research; the third draws attention to a subtler problem: readers can be misled by a carefully chosen portion of genuine results.

What has happened to the evidence?

Suppose a study collects responses from 100 people. There are several distinct ways its reported findings could become unreliable. Someone could invent responses that were never collected. Someone could change recorded responses before reporting them. Or someone could leave all 100 responses intact but report only the portion that supports a preferred conclusion. These actions differ, even though each can give readers a false understanding of the study.

Watch the short explanation of falsification and fabrication in Scientific Misconducts – Falsification, Fabrication and Plagiarism (FFP) by Professor Rahul Jain. Stop before the section on plagiarism; that is the subject of the next lesson.

The Committee on Publication Ethics (COPE) gives concise definitions in its Glossary. Read the two adjacent entries under “D” to fix the distinction.

Fabrication means making up data or results and recording or reporting them as genuine. If no interview took place but a researcher creates an interview transcript and presents it as collected material, the problem is fabrication. The invented item could be a measurement, an image, an observation, or an entire study result; it need not be a number.

Falsification means manipulating research materials, equipment, or processes, or changing or omitting data or results, such that the research is not accurately represented. If an instrument recorded a value of 3 and someone deliberately substitutes 9 to obtain a preferred result, that is falsification. So is presenting an altered image as an unaltered observation. The key is not simply that a file changed: researchers routinely correct errors and process data. The key is that the change or omission distorts the account of the research.

This yields a useful first distinction: fabrication introduces evidence that did not exist; falsification misrepresents research that was conducted. In practice, one incident could involve both—for example, invented observations mixed into a genuine dataset. The categories identify what went wrong; they are not guaranteed to be mutually exclusive.

Why genuine data can still tell a misleading story

Now return to the 100 genuine responses. Imagine that 20 came from one subgroup. Sixteen people in that subgroup reported improvement, while 24 of the other 80 did. A report that says “80% reported improvement” describes the selected subgroup accurately, but it would mislead readers if presented as the result for everyone: 40 of the 100 respondents reported improvement.

No response had to be invented or altered. The distortion comes from which results were allowed to represent the study. Readers normally understand a reported finding in relation to its stated population, methods, and research question. If an account silently substitutes a favourable subset for the full set, readers have been denied information needed to interpret the claim.

Read COPE’s entry on selective reporting, which names deliberate suppression of unfavourable or null outcomes.

Selective reporting describes presenting only a chosen part of the available evidence—particularly favourable findings—while withholding relevant results that would change its interpretation. It can occur when reporting outcomes, subgroups, analyses, or entire studies. For example, a researcher might report a positive result but omit a measured outcome that showed no effect. A report can also mislead by failing to explain a consequential change in method or a departure from the planned analysis.

The Office of Research Integrity’s A Guide to Ethical Writing discusses why this matters and how to distinguish deceptive selection from justified analytical choices. Read its “Selective reporting of results” passage rather than the surrounding sections on literature and methodology.

Why is selection so powerful? Readers cannot assess evidence they never learn exists. If only favourable outcomes appear, a finding may look more consistent or convincing than the complete results warrant. This does not require technical statistics to understand: showing the strongest few observations as though they represent all observations changes the question the reader thinks the evidence answers.

A related problem arises when researchers try several analyses and report only the one that supports their expectation, without revealing that alternatives were tried. The reported calculation may be correct. But the account conceals a choice that matters to its interpretation. Likewise, a graph can display real values yet give a distorted impression if the report withholds relevant context about what was measured or included.

Selective reporting and falsification can overlap. Falsification explicitly includes omitting data or results such that the research is inaccurately represented. Selective reporting names a pattern of choosing which genuine findings to disclose. Deliberately withholding a crucial result to create a false account can therefore be selective reporting and fall under a policy’s definition of falsification. Do not assume, however, that every omission has the same classification or consequence: the applicable policy, the reason for omission, what readers were told, and whether the research was misrepresented all matter.

Selection is not the same as choosing a legitimate scope

A research report cannot contain every observation in full. Researchers must decide what question they are answering, which material is relevant, and how to handle invalid or missing data. Selection itself is unavoidable; misleading selection is the problem.

Consider the subgroup in the earlier example. If the research question was explicitly about that subgroup, reporting that 16 of its 20 members improved could be appropriate. The report must make its scope clear and must not imply that the finding describes all 100 participants. If the subgroup was chosen after inspecting the results because it looked unusually favourable, readers need to know that too. An exploratory finding can be worth reporting; presenting it as though it were the originally planned test gives it a different, unjustified weight.

The same principle applies to exclusions. A faulty measurement may reasonably be excluded under a defensible rule. Excluding it becomes suspect when the real reason is that it weakens the desired conclusion and the report conceals that decision. Transparent reporting explains what was excluded, why, and how the decision affects the finding, as appropriate to the study. Plans made before examining the results, where a project has them, and the decision records discussed in the previous lesson help others distinguish a reasoned choice from one tailored to the outcome.

This is why trustworthy reporting depends on both accurate records and accurate representation. Original files and change logs can show that data were not invented or silently altered. They also make it possible to compare what was collected and analysed with what the report chose to present. They cannot, on their own, make an incomplete report complete; the researcher must disclose the decisions and results needed for readers to understand the claim.

What to carry forward

Fabrication creates purported evidence that was not collected. Falsification distorts the representation of research through manipulation, changes, or consequential omissions. Selective reporting highlights how presenting only favourable genuine results can mislead without inventing or altering any underlying data; in some circumstances, such an omission also meets a policy’s definition of falsification.

The central question is not merely, “Are the reported values real?” It is also, “Does this account give readers a fair basis for understanding what the research found?” In the next lesson, that concern for an honest account turns to other people’s ideas and words: how paraphrase, quotation, attribution, and citation help avoid plagiarism.