03 · What You Need to Know
Not Every Research Decision Needs the Same Degree of Prespecification
Start by distinguishing uncertainty from flexibility
A decision is not legitimately flexible merely because the research team has not made it yet. Sometimes an unresolved issue simply reflects incomplete planning.
Suppose researchers have not decided whether academic achievement will be represented by examination scores or final course grades. If both measures will be available and the team plans to choose whichever produces a stronger association, that is not a compelling methodological reason for flexibility. The primary outcome could have been defined beforehand.
Contrast that with a qualitative interview study in which later interview questions may be refined as researchers identify concepts requiring further exploration. Here, responsiveness to emerging material may be integral to the methodology rather than evidence that the study was poorly planned.
Unresolved planning
A decision could reasonably be made now, but the research team has simply postponed making it.
Legitimate flexibility
A later decision is methodologically justified because it depends on future information, follows an adaptive design, or is governed by criteria established in advance.
Some decisions can be conditional rather than fixed to one option
Prespecification does not always require choosing a single future action. Researchers can sometimes specify a decision rule instead.
For example, rather than deciding in advance that one particular analytical procedure will be used regardless of the circumstances, a protocol might identify alternative procedures and state the conditions under which each would be appropriate. Similarly, a recruitment plan might specify that an additional approved recruitment channel will be activated if enrolment falls below a defined threshold by a particular point.
This approach preserves flexibility without leaving the later choice entirely unconstrained. The decision is made later, but the logic governing it was established earlier.
Before deciding what may remain conditional, identify the study decisions that should normally be fixed before data collection. Flexibility is most defensible when it is deliberate rather than the default state of unfinished planning.
Qualitative research may require responsiveness to emerging data
Many qualitative methodologies involve forms of iteration that would be inappropriate to eliminate merely for the appearance of prespecification. Researchers may refine interview prompts, pursue newly identified lines of inquiry, seek participants with particular experiences, revisit earlier interpretations, or adjust sampling as analysis develops.
The protocol can still specify the methodological framework, initial sampling strategy, recruitment boundaries, data-generation approach, analytical procedures, ethical safeguards, and rationale for adaptation. What remains flexible should follow the logic of the chosen methodology.
Flexibility does not mean that anything can change without explanation. Researchers should be able to describe what evolved, what prompted the change, and how the adaptation remained consistent with the study's purpose and methodological approach.
Exploratory analyses can legitimately emerge after seeing the data
Data can reveal relationships, subgroups, patterns, anomalies, or questions that researchers did not anticipate. Investigating them is not inherently problematic. Exploration is one of the ways research generates new questions.
The important issue is how those analyses are represented. The National Academies has noted that preregistration can help distinguish analyses planned in advance from exploratory analyses developed later. Preregistration does not prevent researchers from performing additional analyses; it makes their status clearer.
A protocol can operate on the same principle. The primary planned analysis can remain identifiable while additional analyses are developed in response to the data. What should be avoided is quietly replacing the planned analysis with an attractive post hoc result and describing the latter as though it had been the original test.
Operational procedures may need to adapt to circumstances
Some changes concern implementation rather than the scientific question itself. A recruitment site may close. Participant availability may require scheduling changes. A supplier may discontinue a material. Software or equipment may fail. A planned fieldwork location may become inaccessible.
Whether such changes are minor or consequential depends on their effect on the study. Moving an interview from one private room to another may have little methodological importance. Replacing face-to-face interviews with asynchronous written responses could change the nature of the evidence and may require much more careful consideration.
The useful question is not simply whether an operational change occurred. Ask whether it changes participant protection, the population reached, the intervention or exposure, measurement conditions, data quality, comparability, or interpretation.
Sampling can sometimes adapt, but the rationale should be methodological
Adaptive sampling is legitimate in several research traditions. Qualitative researchers may use purposive, theoretical, or iterative sampling to seek cases that develop or challenge emerging interpretations. Some quantitative designs also incorporate prospectively defined adaptive features.
However, changing eligibility or sampling because preliminary results would become more favorable is a different matter. SPIRIT 2025 notes that substantive protocol amendments in randomized trials can affect scientific validity and that data-dependent changes can introduce bias.
When sampling is intended to adapt, explain the logic of adaptation in the protocol. What information can trigger additional sampling? What boundaries remain fixed? Who makes the decision? Those questions convert vague flexibility into a defensible methodological process.
The exact sample size may not always be knowable in advance
Some designs require a fixed or formally calculated sample size before data collection. Others do not.
In qualitative research, for example, final sample size may depend partly on the adequacy of the information generated for the study's analytical purpose. Researchers can still specify the initial sampling rationale, anticipated range where appropriate, recruitment boundaries, and criteria or reasoning used to determine whether further data are needed.
Even in studies with formally planned sample sizes, unforeseen circumstances may affect recruitment or retention. Any resulting changes should be evaluated for their methodological implications rather than treated as merely administrative.
Analytical plans may require modification when assumptions or data structures differ from expectations
A planned analysis can turn out to be inappropriate. Perhaps a statistical model's assumptions are seriously violated, observations have a hierarchical structure that was not anticipated, or a variable cannot be analyzed as originally intended because of an unforeseen measurement problem.
Researchers should not knowingly use an inappropriate analysis simply because it appears in the protocol. Methodological quality takes precedence over ritual adherence.
The change should, however, be documented. Explain what made the original analysis unsuitable, what alternative was chosen, and when that decision was made. If researchers already knew the relevant results, that timing is important context for interpreting the revised analysis.
Safety, ethics, and participant welfare can require changes
No protocol should prevent researchers from responding appropriately to an ethical or safety problem. New evidence, unexpected harms, privacy concerns, participant burden, or other circumstances may require modification or even termination of study procedures.
For randomized trials, SPIRIT 2025 explicitly treats the protocol as a living document that may require formal amendment. It recommends documenting amendment procedures, decision-making authority, and communication of substantive changes to relevant parties such as research ethics committees, registries, and regulatory authorities.
Requirements vary by study and jurisdiction. Where formal approval governs the research, investigators should follow the applicable amendment process rather than independently implementing substantive changes unless immediate action is permitted or required to protect participants.
A flexible decision can still be documented before it happens
One of the most useful protocol-writing techniques is to document the source of flexibility itself.
| Instead of leaving this open |
Specify this in advance |
| "We may recruit more participants." |
Why additional recruitment might be needed and what criteria will inform the decision. |
| "Interview questions may change." |
Why iterative refinement is methodologically appropriate and what kinds of changes are anticipated. |
| "Other analyses may be performed." |
Which analysis is primary and how later exploratory analyses will be distinguished. |
| "The statistical method may change." |
Foreseeable conditions under which an alternative method would be considered. |
| "Procedures may be modified if necessary." |
Who can authorize changes, what constitutes a consequential change, and how modifications will be documented. |
This is often more scientifically useful than pretending that the future is fully predictable.
Changes should leave an audit trail rather than rewrite history
A protocol can evolve without erasing its earlier versions. SPIRIT 2025 recommends a transparent audit trail for randomized trial protocols, including dates and descriptions of changes. Its accompanying explanation notes that protocol amendments are common and that some are unavoidable, while changes based on trial data can create interpretive problems or introduce bias.
The broader lesson is straightforward: preserve what was originally planned. If a decision changes, document the new decision alongside the reason and timing. A clear protocol version and change history makes adaptation visible rather than suspiciously invisible.
Watch Out
Flexibility should not become a blanket permission to choose outcomes, exclusions, analyses, or interpretations after discovering which options produce desirable results. The more directly a decision can change the headline finding, the stronger the need for advance specification, explicit decision rules, or transparent reporting of the later change.
