01 · The Question
You Saw the Outcome and Now Think a Case Was Coded Wrong. Can You Change It?
A participant was coded as “nonadherent,” but after examining the outcome you reconsider the classification. A survey response was placed in Category B, yet Category A now seems plausible. A qualitative excerpt was assigned one code, but later analysis suggests that another interpretation may fit better.
Changing codes during research is not inherently suspicious. Coding frameworks develop, errors are discovered, and ambiguous cases sometimes require reconsideration.
The complication is timing. Once you know how a coding decision affects the result, your preferred outcome can influence what suddenly seems like the “better” coding choice.
03 · What You Need to Know
Coding Is an Interpretive Process, but It Still Needs Rules and Evidence
“Coding” Can Mean Different Things Across Research Methods
In quantitative research, coding may involve translating information into analyzable categories: yes/no responses into binary values, diagnoses into groups, occupational categories into numerical codes, or behavioral observations into predefined classifications.
In qualitative research, coding often involves assigning interpretive labels to text, images, observations, or other material. Codes may evolve as researchers develop a deeper understanding of the dataset.
These processes involve different assumptions. A coding revision that is entirely normal in an iterative qualitative analysis may be more consequential in a confirmatory quantitative analysis where category definitions were fixed in advance.
There is therefore no sensible universal rule that says “changing a code after seeing outcomes equals misconduct.”
Correcting a Coding Error Is Usually Different From Redefining a Category
Suppose the codebook defines “1 = treatment” and “0 = control.” A participant assigned to treatment was accidentally entered as 0. Correcting the value to 1 based on the randomization record restores the dataset to the underlying evidence.
That is different from deciding after analysis that several control participants should now be treated as members of the intervention group because moving them improves the treatment effect.
Coding correction
The existing code conflicts with the source evidence or established coding rule, and the revised code restores consistency.
Outcome-driven recoding
The classification is changed because the researcher knows that the new code produces a more favorable or convenient result.
A Revised Coding Framework Can Be Scientifically Legitimate
Researchers sometimes discover that their original categories are poorly defined. Two categories may overlap. A code may be too broad to capture an important distinction. New cases may expose ambiguities that were invisible when the codebook was developed.
Revising the framework can therefore improve the analysis.
The critical questions are why the framework changed and what happens next. If the definition of a category changes, researchers should ordinarily reconsider all comparable cases under the revised definition rather than changing only the particular cases that improve the desired result.
Apply the New Rule to the Dataset, Not Just to the Problematic Case
Suppose a study codes participants as “high engagement” when they attend at least 75% of sessions. After seeing the outcomes, researchers decide that attendance alone does not capture engagement and add completion of weekly activities as a second criterion.
If there is a genuine theoretical and methodological reason for the revised construct, researchers should apply that definition consistently to everyone to whom it applies.
Changing the classification of one inconvenient participant while leaving comparable participants untouched is much harder to defend as a general improvement to the coding framework.
The Source Evidence Should Still Support the New Code
A coding decision should not float free from the underlying evidence.
If a transcript clearly states that a participant had never used the technology before the study, recoding that participant as an experienced user because doing so improves a subgroup analysis would contradict the source material.
Where evidence is genuinely ambiguous, researchers may have more than one defensible interpretation. That ambiguity should be handled according to an appropriate coding procedure rather than resolved opportunistically according to the outcome.
Blinding Can Reduce Outcome-Driven Coding
When feasible, coding decisions can sometimes be made without knowledge of variables that might bias classification. For example, an adjudicator determining whether an event meets a predefined outcome criterion may be blinded to treatment assignment.
Blinding is not possible or appropriate in every study, especially in many qualitative designs. Still, the underlying principle is useful: if knowing the outcome could influence a subjective classification, consider whether the decision can be made independently of that information.
Qualitative Coding Often Evolves During Analysis
Qualitative researchers should be particularly cautious about importing rigid assumptions from confirmatory statistical research into interpretive methods.
In many qualitative approaches, researchers iteratively develop codes, revisit earlier material, merge categories, split codes, refine interpretations, and move between data and emerging concepts. Such iteration can be part of the method rather than a deviation from it.
That does not mean anything goes. The researcher should still be able to explain how interpretations were developed, how contradictory or deviant cases were treated, and whether material was selectively recoded or omitted merely to create a cleaner narrative.
Inter-Rater Disagreement Is Not Evidence of Misconduct
Two competent coders may classify an ambiguous response differently. Resolving that disagreement through discussion, adjudication, or a predefined procedure is not inherently suspicious.
The concern would be different if disagreements were resolved according to which coding produces the preferred hypothesis, while the stated coding procedure suggests another basis for adjudication.
Changing Codes After Seeing Outcomes Creates a Risk, Not an Automatic Verdict
Outcome knowledge can create motivated reasoning even when researchers are trying to act responsibly. A classification that previously seemed clear may suddenly appear ambiguous when its analytical consequences become visible.
That does not mean every reconsideration is dishonest. It means the decision deserves a stronger evidentiary basis.
Useful questions include: What new information justifies the change? Would the code have been changed if the outcome pointed in the opposite direction? Does the revised definition apply consistently to other cases? Can someone reviewing the source material understand why the new code is preferable?
Recoding Can Become Part of a Broader Outcome-Driven Analysis
Changing a code may alter group membership, participant eligibility, outcome classification, exposure status, or another variable used in the analysis. Consequently, recoding can sometimes function like changing the analysis itself.
If researchers repeatedly alter coding rules, rerun analyses, and retain whichever combination produces the desired conclusion, the concern overlaps with changing analyses after seeing the results.
If recoding effectively removes participants or observations from the analytical population, it may also overlap with outcome-driven participant exclusion.
Could Outcome-Driven Recoding Become Falsification?
Potentially. Under the U.S. Public Health Service definition, falsification includes manipulating research materials, equipment, or processes, or changing or omitting data or results such that the research is not accurately represented in the research record.
Whether a particular recoding constitutes falsification depends on the facts. A legitimate disagreement over an ambiguous classification is not equivalent to intentionally or recklessly changing data so that the record becomes inaccurate.
A formal PHS misconduct finding also requires a significant departure from accepted practices of the relevant research community, intentional, knowing, or reckless conduct, and proof by a preponderance of the evidence. Honest error and differences of opinion are excluded.
Watch Out
If you cannot explain why a code should change without mentioning that the new coding improves the result, you probably do not yet have a methodological justification for the change.
