How Should Deviations From a Preregistered Plan Be Documented?

How Do You Report Changes Without Turning the Paper Into a Confession?

A preregistered study rarely becomes more transparent simply because the manuscript contains the sentence, “Some deviations from the preregistration occurred.”

Readers need to know what those deviations were.

At the other extreme, researchers sometimes worry that reporting every small procedural difference will overwhelm the methods section or make ordinary adaptations look like methodological failures.

Good deviation reporting avoids both problems. It identifies differences that matter, tells readers what was originally planned and what was actually done, explains why the change occurred, and provides enough timing information to judge whether knowledge of the data could have influenced the decision.

Document the Original Plan, the Change, the Reason, and the Timing

What Transparent Deviation Reporting Should Let a Reader Reconstruct

Start With the Purpose: Make the Research History Visible

Preregistration creates a record of what researchers intended to do before the relevant results were known. Deviations document what happened afterward.

The goal is not to prove that researchers followed every line of the original plan. Nor is it to apologize whenever reality forced the study to change.

The objective is reconstruction.

A reader should be able to determine which consequential decisions were prospectively specified, which changed, why they changed, and whether those revisions occurred before or after researchers had access to information that could have influenced them.

Willroth and Atherton describe a preregistration deviation broadly as a discrepancy between what researchers said they would do in the preregistration and what they report doing in the final article. Their guidance emphasizes making these differences transparent enough for readers to evaluate their implications.

Report What Was Originally Preregistered

A deviation cannot be understood without the baseline.

Do not merely write, “The analysis was changed.” State what the original plan was.

For example:

“The preregistration specified a linear regression predicting the primary outcome from study condition and baseline score.”

This tells readers what the researchers had prospectively committed to and provides a reference point for the revision.

If the original wording is lengthy, summarize it accurately and provide a direct link or citation to the relevant registration. Do not paraphrase in a way that makes the original plan appear more specific than it actually was.

Report What You Actually Did

Next, describe the revised procedure precisely.

“We changed the analysis” is insufficient if the change involved replacing one model with another, adding a covariate, altering the exclusion rule, redefining the outcome, or changing the sample.

The reader should be able to identify the difference between the registered and implemented methods without conducting their own forensic comparison.

Explain Why the Change Occurred

The rationale is often essential for interpretation.

Consider two studies that both replace the preregistered statistical model. In one, the model fails to converge. In the other, researchers try several alternatives and select the one producing the strongest evidence for their hypothesis.

The visible discrepancy is the same: the final model differs from the preregistered model. The reasons are not.

Explain the substantive, methodological, practical, or ethical reason for the change. If the original plan contained an error, say what the error was. If recruitment failed, explain the constraint. If a measurement problem emerged, identify it.

A vague statement such as “the analysis was modified as appropriate” does little to help readers evaluate the deviation.

Report When the Decision Was Made

Timing helps readers assess whether the observed evidence could have influenced the revision.

Useful timing descriptions may include:

• before data collection began;
• after registration but before any participant was recruited;
• during data collection before the primary outcome was examined;
• during data cleaning before inferential analyses;
• after examining distributions but before testing the primary hypothesis;
• after the preregistered primary result was known.

The appropriate level of precision depends on the deviation. The important point is to describe what information was available when the new decision was made.

State What Relevant Information You Had Already Seen

“Before analysis” can be more ambiguous than it sounds.

A researcher may not have run the final inferential model but may already have examined outcome distributions, correlations, group means, preliminary visualizations, or related analyses.

Those observations can influence subsequent decisions.

When relevant, report the extent of prior data exposure rather than relying on broad chronological labels. This is particularly important when the deviation concerns outcomes, exclusions, transformations, model specification, or subgroup analyses.

Explain Whether the Deviation Affects the Interpretation

Deviation reporting should not stop at description when the change materially affects the claim.

Ask whether the revised decision changes which observations are analyzed, which outcome is primary, which hypothesis is being tested, how uncertainty is estimated, or how strongly the finding can be interpreted as confirmatory.

If the deviation is unlikely to affect the substantive conclusion, explain why where useful. If it introduces additional uncertainty or analytical flexibility, acknowledge that.

Transparency is most useful when readers can understand not only that something changed but why the change matters.

Consider Reporting Both the Preregistered and Revised Analyses

Sometimes the clearest way to show the effect of a deviation is to report both analyses.

Suppose you replace a preregistered analysis because an assumption is violated. Reporting the planned analysis alongside the revised one can reveal whether the conclusion depends on that methodological decision.

This should not be treated as a universal requirement. A planned analysis may be meaningless, impossible, or actively misleading under the circumstances. Journal policies also differ.

Nature Human Behaviour's current preregistration policy, for example, requires deviations to be disclosed and justified and states that, when the analysis plan changes, authors should also report the results of the originally planned analyses.

That is a journal-specific policy. Researchers should verify the requirements of their publication venue rather than assuming that one reporting rule applies everywhere.

A Deviation Table Can Make Multiple Changes Much Easier to Understand

If a study has several deviations, distributing them across the methods, results, footnotes, and supplement can make the research history unnecessarily difficult to reconstruct.

A structured table can be more useful.

The table does not need to look exactly like this. The useful fields are those that allow readers to reconstruct the original plan, revised approach, rationale, timing, and consequences.

Do Not Hide Important Deviations Only in Supplementary Material

Supplementary material can accommodate a detailed deviation table, particularly when there are many changes. But a consequential change to the primary outcome, hypothesis, sample, exclusion criteria, or main analysis should generally not become invisible in the main article merely because a complete table exists elsewhere.

The main text should alert readers to deviations that materially affect interpretation. The supplement can provide fuller documentation.

This is an editorial judgment rather than a universal page-layout rule, and journal requirements vary. The principle is discoverability: important deviations should be difficult to miss.

Do Not Make Readers Compare Two Documents Line by Line

Simply linking to a preregistration is not sufficient deviation reporting.

A reader should not need to open the registration, locate the relevant section, compare it with the published methods, identify discrepancies, and infer why those discrepancies occurred.

That work belongs primarily to the authors.

The preregistration provides the source record. The manuscript should explain consequential departures from it.

Document Deviations During the Research, Not Months Later

Researchers often remember that a change occurred but forget exactly when it occurred, what prompted it, or what they had already seen at the time.

A simple contemporaneous decision log can prevent this problem.

For each consequential change, record:

• the date or stage of the study;
• the relevant preregistered decision;
• the revised decision;
• the reason;
• what relevant data had been accessed;
• whether the change was implemented before or after the primary analysis.

This need not become bureaucratic. A modest log maintained during the study can save considerable reconstruction later. Research notebooks have many virtues; deciphering one's own six-month-old shorthand is not always among them.

Registration Amendments Can Complement, but Not Replace, Final Reporting

Some registries provide mechanisms for updating or amending a registration while preserving its history. The Open Science Framework currently supports registration updates for eligible registrations, allowing changes to be documented without replacing the original record.

If a change occurs before the relevant results are known, an amendment can provide useful prospective documentation of the revised plan.

But updating the registration does not necessarily remove the need to explain the change in the final article. Readers still need to know that the completed study differs from the original plan and understand the reason and timing.

Check the registry's current amendment procedures because available functionality and rules may change.

Do Not Call Every Difference Equally Important

Transparent reporting does not require giving a changed file name the same prominence as a changed primary outcome.

Focus on discrepancies that matter for understanding the study, reproducing the analysis, or evaluating the claims. These commonly include changes to hypotheses, sampling or stopping rules, eligibility and exclusions, outcomes, variable construction, procedures, statistical models, covariates, inferential criteria, and the status of analyses as planned or exploratory.

The exact threshold depends on the research design and publication requirements.

The Language Should Be Descriptive Rather Than Defensive

You do not need to frame every deviation as an apology.

Instead of:

“Unfortunately, we were forced to violate our preregistration by changing the analysis.”

A clearer report might state:

“The preregistration specified model A. After inspecting the data structure, we determined that assumption X required by that model was not satisfied. We therefore used model B for the primary analysis. This decision was made before evaluating the focal coefficient.”

The second version gives readers information they can evaluate. It neither hides the deviation nor treats methodological adaptation as a moral failure.

Transparent Reporting Does Not Determine Whether the Deviation Was Appropriate

Once a deviation is clearly documented, readers and reviewers can evaluate its implications.

A well-reported deviation may still weaken a claim. A poorly justified result-contingent analysis does not become confirmatory because the authors disclosed it. Conversely, a deviation that corrects an error may make the analysis more defensible than the original plan.

This is why researchers should distinguish whether a deviation was defensible from whether it was transparently documented. They are related questions, but they are not the same question.

From an Unhelpful Disclosure to a Transparent Deviation Record

Common Mistakes When Reporting Preregistration Deviations

Create a Deviation Record While You Conduct the Study

The easiest time to document a deviation is when you make it.

Use this framework as an internal decision log during the study. When preparing the manuscript, decide which deviations need direct discussion in the main text and which can be presented more fully in a supplementary table.

If a deviation occurred before relevant results were known, consider whether your registry permits a transparent amendment. If it occurred after the data informed the decision, do not attempt to make it appear prospective by simply updating the registration. Preserve the chronology.

The objective is simple: a reader should not have to guess where the preregistered plan ends and the actual research process begins.

Before Reporting a Preregistration Deviation

Frequently Asked Questions About Reporting Preregistration Deviations

Make Every Consequential Difference Easy to Reconstruct

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.