What Counts as Fabricating Research Data?

When Does Research Data Become Fabricated?

The clearest examples of fabrication are dramatic: a researcher invents participants who never existed or reports experiments that were never performed. But fabrication does not have to involve an entirely fictional study.

What if only five participant records are invented? What if an experiment failed to produce a measurement and someone types in the value they expected? What if a researcher adds plausible responses to complete an otherwise genuine dataset?

The central issue is not how much of the study is false. It is whether purported data or results were made up and recorded or reported as though they genuinely arose from the research.

Fabrication Means Creating Data or Results That Were Never Obtained

The Defining Feature of Fabrication Is Invention

The Data or Result Has No Genuine Research Event Behind It

The current Public Health Service regulation defines fabrication succinctly: making up data or results and recording or reporting them. That definition focuses on provenance. Where did the purported observation come from?

If a dataset says Participant 084 completed a survey, there should be a genuine research event behind that record. If an instrument file reports a measurement, there should ordinarily be a corresponding measurement process. If an interview transcript contains a participant's response, that response should derive from the underlying research rather than the researcher's imagination.

When the purported observation was simply invented and then entered or presented as genuine research data, the basic fabrication problem appears.

Inventing Participants Can Be Fabrication

One straightforward example is creating records for people who never participated. NIH illustrates fabrication with a study coordinator who creates enrollment forms and data for nonexistent participants.

The same principle can apply beyond clinical research. Inventing survey respondents, interviewees, classroom observations, experimental subjects, specimens, cases, or other research units can create purported evidence without the underlying research events those records claim to document.

Inventing Measurements or Responses Can Also Be Fabrication

The participant or research unit itself does not have to be fictional. Suppose a real participant completed most of a questionnaire but left several items blank. If a researcher simply invents answers and records them as the participant's actual responses, those purported responses were not observed.

Likewise, a real experiment may have been conducted while a particular measurement was never obtained. Creating a plausible measurement afterward and recording it as an observed instrument reading can raise the same basic concern.

This is why fabrication is broader than “fake participants.” The object being invented can be a data point or result within an otherwise genuine study.

Inventing Results Without Inventing Raw Data Can Still Matter

The regulatory definition refers to “data or results.” A researcher therefore cannot avoid the issue merely because no raw-data spreadsheet was fabricated.

For example, reporting an experimental result for an experiment that never produced that result can potentially constitute fabrication even if the researcher never creates a complete underlying dataset. Likewise, presenting made-up summary results as though they came from an analysis does not become acceptable simply because the fictional underlying observations were never entered into a file.

Recording or Reporting Is Part of the Definition

The definition does more than say “making up.” It specifies making up data or results and recording or reporting them. This matters because the research record extends well beyond published journal articles.

Under the current PHS regulation, the research record can include raw and processed data, clinical and laboratory records, study records, laboratory notebooks, progress reports, manuscripts, abstracts, theses, records of oral presentations, online content, lab meeting reports, and journal articles.

Fabrication therefore need not wait for publication. An invented observation entered into research records may already be relevant even if it never reaches the final manuscript.

Fabrication and Falsification Are Related but Different

If a researcher invents observations for participants who never existed, the problem points toward fabrication. If genuine observations existed but the researcher improperly changes or omits them so that the record becomes inaccurate, the issue points toward falsification. The distinction between fabrication and falsification concerns how the research record became unreliable.

Fabrication Is Not Defined by Whether the Invented Value Looks Plausible

A fabricated value does not need to be outrageous or statistically impossible. Indeed, an invented value may look entirely ordinary.

Suppose most participants score between 60 and 80. Typing “72” for a nonexistent observation does not make the value authentic merely because 72 is plausible. The question is not whether the value could have occurred. The question is whether it actually arose from the research as represented.

Legitimate Imputation Is Not the Same as Pretending a Value Was Observed

Missing-data methods create an important distinction. Statistical imputation can produce estimated values for missing observations using an explicitly defined analytical procedure. Such values are not automatically fabrication simply because they were not directly observed.

The representation matters. An imputed value should remain identifiable conceptually and, where appropriate, technically as an imputed or analytically generated value. It should not be silently transformed into a purported original observation.

This distinction becomes especially important when a researcher considers filling in a missing value from memory. Remembering what a value “probably was” is not equivalent to applying a transparent missing-data method or recovering the value from a reliable source record.

Simulated and Synthetic Data Are Not Inherently Fabricated Research Data

Researchers legitimately generate simulated, synthetic, or hypothetical data for purposes such as method development, statistical demonstrations, model testing, training, and power analyses. Those data are intentionally generated rather than observed.

The critical difference is representation. Data explicitly identified as simulated are not being presented as measurements that actually occurred. Problems arise when invented information is represented as genuine empirical data or results.

An Error Is Not Automatically Fabrication

A mistyped number can produce a value that was never actually observed. That alone does not establish fabrication.

The PHS definition of research misconduct expressly excludes honest error and differences of opinion. Moreover, a formal finding requires evidence satisfying additional criteria, including the applicable culpability standard. An accidental transcription error and deliberately inventing an observation may produce similarly incorrect spreadsheet cells, but they are not ethically or legally equivalent.

How a Mostly Genuine Dataset Can Still Contain Fabricated Data

Common Misunderstandings About Fabricated Data

Never Turn Uncertainty Into an Observation

The safest practical principle is simple: if you do not have evidence that an observation occurred, do not silently create one.

When data are missing, investigate whether the original value can be recovered from an authoritative source. If it cannot, preserve the missingness and address it using a defensible methodological approach. When data are incorrect, trace the problem to the source and correct genuine errors transparently rather than substituting an unsupported value.

Good provenance solves many problems before they become integrity disputes. Preserve enough documentation that another qualified person could determine where important observations came from, what transformations occurred, and why corrections were made. When legitimate changes are necessary, document corrections to the research data rather than allowing the revised file to erase its own history.

Before Entering or Replacing Research Data, Check the Evidence

Frequently Asked Questions About Fabricating Research Data

Fabrication Creates Purported Evidence That Was Never Obtained

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.