03 · What You Need to Know
Why Changing a Preregistered Study Is Not Automatically a Problem
Preregistration Is a Record of Intentions at a Particular Time
The basic purpose of preregistration is temporal. It allows researchers to document important research decisions before the relevant results are known.
That record can later help readers distinguish decisions made prospectively from decisions influenced by information that emerged during the research. This is one reason why a useful preregistration should specify the consequential decisions that matter for the study's claims.
None of this requires pretending that researchers can foresee everything.
Nosek and colleagues explicitly describe preregistration as a plan rather than a constraint that must be obeyed regardless of circumstances. They argue that deviations may improve research quality when the original plan contains errors or better approaches emerge through learning and methodological development.
Commitment to transparency
You preserve an accurate record of what you originally intended to do and disclose consequential departures from it.
Commitment to literal adherence
You follow the original plan regardless of whether later information shows that a different decision would be methodologically preferable.
Preregistration requires the first. It does not necessarily require the second.
A Preregistered Decision Can Turn Out to Be Wrong
Suppose you preregister an analysis and later discover that it is statistically inappropriate for the structure of your data. Following that analysis merely because it was preregistered would not rescue the study. It could make the inference worse.
The same principle applies when a preregistration contains an error. Researchers are fallible at the planning stage just as they are at every other stage of research. A time stamp does not transform an incorrect decision into a correct one.
In such circumstances, methodological validity should not be sacrificed simply to produce perfect correspondence between the preregistration and the paper.
The more defensible approach may be to explain what was planned, identify why that approach became inappropriate, describe what was done instead, and assess whether the change affects the strength or interpretation of the resulting claim.
Some Changes Are Forced by Circumstances Outside Your Control
Not all deviations are analytical. Research involves people, institutions, technologies, materials, budgets, and timelines, any of which can change.
A participating institution may withdraw. Recruitment may become impossible under the original procedure. Equipment may fail. A commercial instrument may be discontinued. A data source may change its access conditions. An ethical or safety concern may require a procedure to be modified.
These events can make parts of the original plan impossible or inappropriate.
The relevant question is not whether the study changed. It is whether the change is understandable, appropriately documented, and taken into account when interpreting the evidence.
Changes Can Also Result From Information Learned During the Study
This category requires particular care because the researcher may now possess information that was unavailable when the preregistration was created.
Imagine that a researcher preregisters a statistical model but discovers during analysis that a key assumption is seriously violated. An alternative model may provide a more defensible analysis.
Changing the model can be justified. At the same time, the new choice is no longer fully independent of what was learned from the data.
Both facts matter.
The revised analysis should not be treated as illegitimate merely because it was not preregistered, but neither should its post-registration origin disappear from the report. This is precisely why preregistration is useful: it helps distinguish what was planned from what became necessary or desirable later.
When the Change Occurs Matters
Not every post-registration change has the same implications.
Suppose you discover a mistake in your analysis plan immediately after submitting the preregistration but before collecting any data. Correcting it at that point differs from choosing a new analysis after examining the outcome because the first correction was not informed by the study's results.
Likewise, a procedural change forced by equipment failure before outcome data are available raises different concerns from an analytical change selected after comparing several models against the observed results.
| When the change occurs |
What the researcher knows |
Why the timing matters |
| Before relevant data are collected or observed |
The study results are still unknown |
The revision may be documented without having been selected because of the eventual result. |
| During data collection but before relevant outcomes are examined |
Some information about study implementation may be known |
The change may respond to practical realities while remaining uninformed by the primary outcome. |
| After examining relevant data or results |
The researcher has information about the observed evidence |
The decision may have been influenced by the data and should be reported with that chronology clear. |
This is why simply labeling every difference a “deviation” can conceal important distinctions. Readers benefit from knowing not only what changed but when and why it changed.
A Deviation Does Not Automatically Invalidate Confirmatory Research
Researchers sometimes assume that once anything changes, the entire study must be relabeled exploratory. That is too crude.
A minor procedural adjustment unrelated to the primary hypothesis may have little bearing on the confirmatory claim. A correction to an obvious coding error may improve the validity of the analysis. Conversely, changing the primary outcome after seeing which outcome produces the strongest result would have much more serious implications for how the evidence should be interpreted.
The consequences depend on the nature of the deviation.
Ask whether the change could have been influenced by the observed result, whether it alters the hypothesis being tested, whether it changes which observations or outcomes contribute to the analysis, whether it changes the inferential procedure, and whether it creates additional researcher flexibility.
A useful evaluation therefore goes beyond “Did the researchers follow the preregistration?” to “How does the deviation affect what can reasonably be inferred?”
Changing the Plan Is Different From Hiding the Change
This distinction is fundamental.
Changing a preregistered decision can be methodologically justified. Silently reporting the revised decision as though it were the original plan undermines the transparency that preregistration is intended to provide.
Nature Human Behaviour, for example, currently requires authors reporting preregistered non-clinical confirmatory research to disclose deviations from the preregistered protocol and explain their rationale. Its policy recognizes reasons such as flaws, feasibility problems, or suboptimal plans. In some circumstances involving changes to the analysis plan, the journal also requires the originally planned analyses to be reported.
That is a journal-specific policy rather than a universal rule for every publication. The broader principle, however, is widely applicable: a deviation should not be allowed to disappear from the research record merely because the revised approach seems preferable.
Preserving the Original Record Is Important
If a preregistration could simply be rewritten after the results were known, it would lose much of its value as evidence of what was planned earlier.
Registries therefore need ways to preserve the chronology of changes. The Open Science Framework currently supports registration updates intended to reflect necessary changes while maintaining a transparent record rather than silently replacing the original registration.
Exactly how revisions, amendments, or updates work depends on the registry. Check the current documentation before modifying a registration.
Whatever system you use, the underlying objective is the same: someone examining the research later should be able to distinguish the original plan from subsequent revisions.
Registered Reports Require Additional Care
Preregistration and Registered Reports should not be treated as interchangeable. In a Registered Report, the Stage 1 plan has undergone prospective peer review and may have received in-principle acceptance from a journal.
A change can therefore involve not only a departure from your own earlier plan but also a departure from a protocol reviewed and accepted by editors and reviewers.
Journal procedures vary, so researchers should consult the specific Registered Report policy. Depending on the nature and timing of a proposed change, communicating with the editor before implementing it may be appropriate or required.
This additional editorial relationship is one of the important differences between preregistration and Registered Reports.
The Goal Is Not a Perfect Match Between Plan and Paper
Perfect adherence can look reassuring, but it should not become the wrong target.
Research quality depends on making defensible methodological decisions given the information available. Sometimes the best decision is to follow the preregistration. Sometimes it is to depart from it.
The important questions are whether the change was justified, whether its timing and rationale are visible, and whether the resulting claims appropriately reflect any additional uncertainty introduced by the deviation.
Watch Out
Do not preserve a demonstrably inappropriate analysis merely to claim perfect adherence to a preregistration. But do not use “research is flexible” as permission to change outcomes, exclusions, models, or hypotheses after seeing the results without disclosure. Methodological flexibility and transparent reporting need to operate together.
06 · What This Means for You
Change the Plan When Necessary, but Preserve the Research History
When you encounter a reason to depart from a preregistration, do not begin by asking whether deviations are “allowed.” Ask whether the proposed change improves or protects the validity, feasibility, ethics, or integrity of the research and what information you already know when making that decision.
A simple decision framework
If you discover a mistake before seeing the relevant data
Correct it transparently and document when and why the correction was made.
If circumstances make the original procedure impossible
Adopt a defensible alternative and preserve a record of the practical reason for the change.
If the observed data reveal that the planned analysis is inappropriate
Use the methodologically defensible analysis, but state that the decision was made after examining the relevant data and explain its consequences.
If a new analysis is motivated by an unexpected pattern in the results
Conduct it if scientifically useful, but distinguish it from the analysis specified in advance.
If you are working under a Registered Report
Check the journal's policy and communicate consequential proposed changes to the editor when required or appropriate.
Keep a decision log while conducting the study. Recording what changed, when it changed, why it changed, and whether the relevant outcomes had already been examined can make later reporting considerably easier. Few researchers enjoy reconstructing six months of analytical decisions from cryptic filenames and comments in old scripts.
When writing the manuscript, do not make readers perform forensic work to discover discrepancies. Explain consequential departures where they are relevant and use supplementary material or a structured deviation table when the number of changes would otherwise overwhelm the main text.
The next practical issue is identifying which parts of a preregistered study can legitimately change and what different changes may mean for interpretation.
