Can Filling In a Missing Value From Memory Count as Fabrication?

You Remember the Missing Value. Can You Put It Back?

You open a dataset and discover that one value is missing. You remember collecting it. Perhaps you even remember what the number was: “I’m almost certain it was 17.”

Typing 17 into the empty cell can feel very different from inventing data. After all, you believe the measurement actually happened. But the research record now presents a more difficult question: what evidence establishes that 17 was the value actually observed?

The difference between recovering a lost observation and creating an unsupported replacement may depend on the records available, what exactly you remember, how the reconstructed value is documented, and what the new entry is represented to be.

Memory Alone Should Not Be Treated as a Verified Observation

Why Remembering a Value Is Not the Same as Verifying It

The Fabrication Question Is About the Provenance of the Entered Value

Under the U.S. Public Health Service definition, fabrication means making up data or results and recording or reporting them. That makes provenance central to the problem.

Suppose a measurement is missing from the working dataset. If the researcher finds the original instrument output showing a value of 17 and restores 17 to the dataset, the value has documentary support. The researcher is recovering a recorded observation.

If no record can be found and the researcher types 17 solely because “I remember it being about 17,” the evidentiary situation is different. The dataset may now imply a level of certainty that the surviving research record does not support.

There Is a Difference Between Remembering That a Measurement Happened and Remembering Its Exact Value

A researcher may genuinely remember performing an experiment, interviewing a participant, scoring an assessment, or taking a measurement. That recollection does not necessarily establish the exact numerical or textual content of the missing observation.

This distinction is easy to overlook. “I remember measuring this participant” supports one proposition. “The measurement was exactly 17.4” is a much more specific proposition. The second requires evidence sufficient to justify recording 17.4 as the actual value.

Finding Independent Evidence Changes the Situation

Memory can be useful as a clue. It might tell you where to look.

You may remember that the instrument exported a separate file, that the value was copied into a laboratory notebook, that a paper questionnaire still exists, or that an electronic system retains an audit trail. If a reliable contemporaneous source confirms the value, you are no longer relying solely on memory.

That is fundamentally different from choosing a value because it feels familiar.

“I’m Certain” Does Not Create a Source Record

Confidence in memory does not make recollection equivalent to contemporaneous documentation. Human memory can be incomplete or mistaken, particularly when many similar measurements, participants, interviews, or experimental runs are involved.

The research-integrity issue is not that researchers are forbidden from remembering their work. It is that a research dataset ordinarily needs an evidentiary basis that can survive beyond the researcher's private recollection.

This is one reason good research records matter. NIH intramural guidance, for example, emphasizes recording and retaining research records in a form that allows access and reconstruction of the work by others. A research record should not depend unnecessarily on “I remember what happened.”

A Plausible Approximation Is Not the Original Observation

Suppose you remember that a participant's score was “around 80.” Entering 80 does not recover the observation if you cannot establish whether the actual score was 78, 80, 82, or something else.

The same applies when the researcher can infer a likely value from surrounding observations. If temperatures immediately before and after a missing measurement were 21.1°C and 21.3°C, entering 21.2°C may be a mathematical interpolation. It is not automatically the temperature that was actually observed.

If a defensible analytical method estimates a missing value, the estimate should be treated according to that method and represented appropriately. It should not silently acquire the status of raw observed data.

Imputation Is Different From Recollection

Formal missing-data procedures can estimate unobserved values using specified statistical assumptions and methods. Researchers can describe those methods, reproduce them, evaluate their assumptions, and assess how they affect the analysis.

Memory-based replacement usually lacks those properties. “I remember the score was 80” is not a statistical missing-data method.

Neither should imputation be disguised as direct observation. The important point is that a transparent estimate remains an estimate, while a recovered observation should have evidence supporting its status as an observation.

Could Entering the Remembered Value Actually Be Fabrication?

Potentially, but the answer is fact-dependent.

If a researcher knows that the exact value is unknown yet invents or guesses one and records it as an observed result, the conduct may fit the basic concept of making up research data. Calling the guess a “memory” does not necessarily change what occurred.

Other cases may be less straightforward. A researcher might sincerely but mistakenly believe that a recollection is accurate. There may be partial contemporaneous documentation. Accepted practices may differ depending on the type of research record. These facts can matter when evaluating whether conduct satisfies a formal misconduct standard.

A Formal Misconduct Finding Requires More Than a Questionable Cell

Under the current PHS framework, research misconduct does not include honest error or differences of opinion. A finding also 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.

Consequently, discovering that a researcher entered a remembered value should trigger questions about provenance, documentation, representation, and circumstances. It should not automatically produce a final misconduct verdict.

One Missing Score, Three Very Different Responses

Common Mistakes When Trying to Reconstruct Missing Data

Recover What You Can Verify and Preserve What You Cannot

When you discover a missing value, treat the problem as one of evidence rather than memory. Your recollection may guide the search, but the goal is to find a defensible basis for the value.

If a legitimate correction is made, preserve enough information to show what changed and why. The principles for documenting research data corrections are especially useful when the working dataset no longer matches an earlier version.

If more than one value or an entire research record has disappeared, the problem becomes broader than a single missing cell. Attempting to reconstruct missing research records from memory raises additional questions about what can legitimately be recovered, what must remain uncertain, and how the reconstructed record should be represented.

Before Replacing a Missing Value From Memory, Check These First

Frequently Asked Questions About Missing Values and Memory

Memory Can Guide Recovery, but It Should Not Manufacture Certainty

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.