01 · The Question
Can You Change How You Analyze the Data After Ethics Approval?
You planned to use one statistical test, but the data violate its assumptions. An additional subgroup analysis becomes interesting. A qualitative coding framework needs refinement after researchers begin reading transcripts. Perhaps a planned model simply does not work with the data you actually collected.
Analysis plans do change, sometimes for entirely defensible methodological reasons. But once a study has ethics approval, the relevant question is not simply whether another analysis would be statistically or methodologically better. You also need to ask whether the revised analysis remains within the research that was reviewed and approved.
Some analytical adjustments may have little or no effect on participants or the ethical basis of the study. Others can change outcomes, hypotheses, data uses, privacy implications, or the scientific justification on which participant involvement was accepted. Those differences matter.
03 · What You Need to Know
Analysis Changes Differ Greatly in Their Ethical Significance
Start With What the Approved Protocol Actually Says About Analysis
Protocols vary considerably in analytical detail. One may specify the primary outcome, statistical model, covariates, subgroup analyses, handling of missing data, and sensitivity analyses. Another may state only the broad analytical approach. A qualitative protocol may describe thematic analysis while deliberately allowing coding to develop iteratively.
That difference matters. A decision already permitted by the protocol is not necessarily a protocol modification simply because it was made after data collection began.
Before asking whether an amendment is needed, identify the analytical commitments actually contained in the approved protocol. Then compare those commitments with what you now propose to do.
Changing a Statistical Technique Is Not Necessarily the Same as Changing the Research
Suppose a researcher planned an independent-samples t-test but discovers that the data do not satisfy the assumptions required for the planned analysis. Using a defensible alternative may represent a technical response to the observed data rather than a change to the study's substantive purpose.
Likewise, modifying model specifications, using a robust estimator, changing a missing-data technique, or correcting a programming error may sometimes leave the research question, outcomes, participant population, collected data, and privacy arrangements unchanged.
Whether such adjustments require ethics review depends on the specificity of the approved protocol and the applicable institutional rules. Ethics committees are not ordinarily intended to function as statistical software help desks, but methodological changes can become ethically relevant when they alter the research that was approved.
Changing Primary Outcomes Is More Consequential
Changing which outcome is considered primary can alter the scientific meaning of a study. If an outcome performs poorly or produces an inconvenient result, replacing it after examining the data also raises concerns about selective outcome reporting and interpretive bias.
This is not merely an ethics-administration issue. The change should be transparently documented in the research record and, where relevant, in trial registration, analysis documentation, manuscripts, or reports. Prospective and post hoc decisions should not be presented as though they were the same.
If the approved protocol identifies specific primary and secondary outcomes, changing their status may represent a substantive protocol modification and should be assessed under the applicable ethics procedure.
Adding a New Analysis Does Not Automatically Mean New Ethics Review
Researchers routinely explore data beyond the primary analysis. An additional correlation, sensitivity analysis, alternative model, or exploratory subgroup analysis does not automatically create a new human-participants study.
The important questions are what data are being used, whether those data were already legitimately collected for the approved research, whether the proposed analysis remains within permitted uses, and whether it creates new privacy, identifiability, or participant-group concerns.
A useful exploratory analysis may therefore be methodologically permissible without changing the approved research, yet it should still be labeled honestly as exploratory if it was not prespecified. Ethics approval should never be used to convert an analysis invented after seeing the results into a prespecified hypothesis.
New Data Linkage Can Change the Ethics Question
An analytical idea may require information that was never part of the approved dataset. For example, researchers might decide to link survey responses with academic records, clinical records, employment data, geolocation, administrative databases, or publicly available information about individual participants.
At that point, the proposal is no longer merely a different statistical test. New data are being obtained or linked, and the change may affect privacy, confidentiality, identifiability, consent, data security, or institutional permissions.
The amendment question should therefore be framed around the actual new activity rather than described vaguely as “changing the analysis.”
Subgroup Analysis Can Create Privacy and Interpretive Concerns
Analyzing smaller participant groups can be scientifically useful, but increasingly granular subgrouping may make individuals easier to identify, particularly in small populations or sensitive settings.
Subgroup analyses can also generate claims about communities or demographic groups that were not part of the original analytical rationale. That does not make such analyses inherently unethical, but it may affect how data use, disclosure risk, and interpretation should be considered.
Qualitative Analysis Is Often Iterative by Design
Qualitative research requires particular care because its analytical process is often intentionally iterative. Codes may be refined, categories merged, themes reconsidered, and interpretations developed as researchers engage with the data.
If that flexibility is part of the approved methodological approach, every adjustment to a codebook is not necessarily a protocol amendment. The situation changes when the analysis expands into a materially different research purpose, requires new data, introduces new identifiable linkages, or otherwise moves beyond the approved study.
Researchers should therefore distinguish methodological iteration within an approved qualitative design from redesigning the research question after data collection has begun.
Scientific Validity Has Ethical Relevance
Research involving participants should be capable of producing knowledge sufficient to justify the burdens and risks imposed. OHRP guidance describing the criteria for IRB approval notes that risks should be minimized through procedures consistent with sound research design.
An analysis change that repairs a defensible methodological problem may therefore strengthen rather than undermine the ethical basis of the study. Conversely, an analytical change that substantially weakens the study, abandons its original purpose, or introduces avoidable bias may raise questions about scientific validity.
Ethics review and methodological peer review remain different functions, but they overlap where scientific inadequacy affects the justification for exposing participants to research burdens or risks.
Changes to Analysis Can Originate Upstream
A revised analysis plan may be the consequence of another protocol modification. Adding questions to a survey creates new variables. Changing the sample size may affect power and modeling. Removing a research site may change planned site comparisons. Altering an intervention may change the outcomes that need to be analyzed.
In such cases, the analytical change should not be assessed in isolation. If the underlying modification also changes the approved research instrument, consider the implications of changing the research instrument. If enrollment changes, assess the sample-size modification as well.
Prospective Changes and Post Hoc Analyses Should Be Distinguished
If researchers decide before seeing relevant outcome data that the analysis plan should change, they can document the rationale and follow whatever amendment or protocol process applies. Once results are known, the interpretive situation changes.
Post hoc analysis is not inherently illegitimate. Many useful findings emerge through exploration. The problem arises when exploratory decisions are concealed or represented as though they were prespecified.
Good research records should therefore preserve when an analytical decision was made, why it changed, and whether relevant outcome data had already been examined.
Watch Out
Do not rewrite the analysis plan after seeing the results and then present the revised analysis as the plan that existed all along. Ethics documentation, protocol history, registrations, analysis plans, and publications should preserve the distinction between prespecified and post hoc decisions where that distinction matters.
Whether an Amendment Is Required Depends on the Governing Process
Under U.S. HHS requirements, proposed changes to approved research generally require IRB review and approval before implementation, except when necessary to eliminate apparent immediate hazards to participants. OHRP also recognizes expedited review of qualifying minor changes.
This does not mean that every analytical keystroke constitutes a protocol amendment. The key question is whether the proposed analytical change is a change to the approved research under the applicable framework. Institutions should have procedures for determining which changes are minor and which require a different level of review.
