03 · What You Need to Know
How Do You Find the Assumptions That Could Break the Study?
A useful starting point is to distinguish assumptions from known requirements. “The study requires 12 months of follow-up” is a design feature. “We can retain enough participants for 12 months” is an assumption about whether that design can actually be executed.
Feasibility research exists partly because important assumptions about a future study sometimes need empirical testing. A widely cited conceptual framework describes feasibility studies as asking whether something can be done, whether researchers should proceed, and how. Pilot studies are treated within that framework as a subset in which part or all of the future study is conducted on a smaller scale.
Map the Study as a Chain of Dependencies
Write the study as a sequence:
research question → participants or evidence source → access → measurement or observation → implementation → usable data → analysis → interpretation → conclusion.
At each transition, ask: What am I assuming will be true?
Perhaps you assume participants understand the instructions. You assume the intervention creates enough contrast between conditions. You assume a record identifier permits linkage. You assume the outcome occurs frequently enough to analyze. You assume the measurement instrument behaves adequately in the new population.
This approach makes assumptions visible because it focuses on what must happen for one stage to support the next.
Separate Scientific Assumptions From Operational Assumptions
Both can make a study fail, but they fail differently.
Scientific assumption
An assumption about measurement, causal structure, model behavior, intervention mechanisms, comparability, interpretation, or another feature affecting what the evidence means.
Operational assumption
An assumption about access, recruitment, retention, staffing, timing, equipment, data delivery, implementation, or another condition affecting whether the evidence can be generated at all.
A project can succeed operationally and still fail scientifically. You can recruit every participant on schedule and still use a measure incapable of representing the intended construct. Conversely, an excellent design on paper produces no evidence if recruitment never reaches the required population.
Look for Assumptions Hidden Inside Confident Verbs
Proposal language often makes uncertain processes sound completed before they have happened.
“Participants will be recruited.” “Records will be linked.” “Teachers will implement the intervention.” “Students will use the platform.” “Follow-up assessments will be completed.” “The model will estimate...”
Replace will temporarily with we assume that. The sentence often becomes more revealing.
“We assume that 300 participants can be recruited within eight weeks” is now something you can investigate.
Prioritize Assumptions by Consequence, Not Just Uncertainty
You may be highly uncertain about a minor administrative detail whose failure would cause little harm. Another assumption may seem reasonably likely but would destroy the study if wrong.
Consider both dimensions.
| Assumption |
Uncertainty |
Consequence if wrong |
Priority |
| A secondary questionnaire takes 8 rather than 12 minutes |
Moderate |
Minor scheduling inconvenience |
Lower |
| Enough eligible participants can be recruited |
Moderate |
Primary analysis cannot be completed as planned |
High |
| The primary outcome is recorded reliably in the available dataset |
Uncertain |
Central research question becomes unanswerable |
Very high |
| A specialist software license remains available |
Low |
Alternative software can be used |
Lower |
The objective is not a mathematically precise risk score. It is to direct attention toward assumptions capable of changing whether the study should proceed.
Ask Whether Failure Is Recoverable
Some assumptions can fail without seriously threatening the project.
If one recruitment channel performs poorly, several appropriate alternatives may exist. If one nonessential software package becomes unavailable, another may substitute adequately.
Other failures are difficult to repair. Discovering after data collection that the primary measure does not represent the intended construct may invalidate the central inference. Learning halfway through a longitudinal study that the required follow-up extends beyond the funding period may leave no good solution.
Prioritize assumptions whose failure becomes expensive or irreversible after the study starts.
Distinguish Assumptions You Can Verify From Assumptions You Must Manage
Some uncertainties can be resolved directly.
You can inspect a data dictionary, obtain written site permission, test equipment, check whether variables can be linked, or verify how many eligible participants exist.
Others cannot be eliminated beforehand. Recruitment rates, attrition, implementation fidelity, and unexpected technical problems may remain uncertain even after careful planning.
For these, establish monitoring, thresholds, contingencies, or progression criteria rather than pretending uncertainty has disappeared.
Use Feasibility Work for Assumptions That Need Empirical Evidence
When a critical assumption cannot be resolved from existing information, preliminary research may be appropriate.
Feasibility and pilot-study methodology explicitly focuses on uncertainties that determine whether and how a future definitive study should proceed. The CONSORT extension for randomized pilot and feasibility trials similarly emphasizes that the rationale for pilot work is to investigate areas of uncertainty about a future definitive trial and that decisions about progression should be built into the design.
Although these frameworks were developed around trials, the reasoning extends more broadly. A small recruitment test, instrument pilot, technical prototype, data audit, cognitive interview, workflow test, or preliminary observation may answer a critical feasibility question without pretending to answer the main substantive question.
Do Not Use a Pilot Study to Test Everything
A pilot should have explicit feasibility objectives.
If your uncertainty concerns recruitment, measure recruitment. If it concerns whether participants understand an instrument, investigate that. If the problem is data linkage, test the linkage pathway.
Do not collect a small convenience sample, run the entire planned analysis, and call every uncertain result a feasibility finding. Preliminary work is most informative when it targets the assumptions that actually determine whether the larger study can succeed.
Challenge Measurement Assumptions Early
Measurement assumptions are particularly dangerous because the resulting dataset can look complete even when the evidence is conceptually inadequate.
Ask whether the instrument, coding scheme, behavioral trace, administrative field, or proxy actually represents the construct in the research question. Does it function appropriately in this population? Has the meaning changed across settings or time?
If the answer is uncertain and the measurement is central, resolve that uncertainty before collecting a large volume of unusable data.
Challenge Assumptions About Participant Behavior
Many designs assume that people will behave in particular ways.
Participants will attend repeated sessions. Teachers will implement the intervention as intended. Students will actually use the tool provided to them. Respondents will complete diaries regularly. Organizations will continue cooperating.
These are empirical assumptions. If the study's inference depends on them, consider how they will be assessed rather than merely hoped for.
Challenge Analytical Assumptions Before the Final Dataset Exists
Some assumptions concern the analytical strategy: sufficient outcome variation, identifiable parameters, adequate observations within clusters, plausible missing-data assumptions, model convergence, or enough cases in relevant categories.
Not all can be verified before collection, but many can be investigated using previous data, simulations, pilot evidence, or sensitivity analyses.
Current NIH research-strategy guidance explicitly asks applicants to discuss potential problems, alternative strategies, benchmarks for success, and strategies for establishing feasibility when projects involve high-risk aspects.
Assumptions About Existing Evidence Matter Too
Your study may depend on prior findings being sufficiently credible to justify the new question.
What if the foundational study used a weak measure? What if the influential effect estimate is based on a small sample? What if the mechanism you intend to extend has never been distinguished empirically from a plausible alternative?
NIH rigor guidance explicitly asks researchers to assess strengths and weaknesses in the prior research that serves as key support for a proposed project. That is a useful general reminder: the assumptions beneath your study can originate in the literature, not just in your protocol.
Write Down What Happens if Each Critical Assumption Fails
For every high-priority assumption, specify the consequence.
Does the study stop? Can the design change? Does the research question need narrowing? Is another data source available? Would the resulting evidence support a weaker but still worthwhile conclusion?
This converts a risk register into a decision tool.
Not Every Assumption Needs to Be Eliminated
Research inherently contains uncertainty. A study that proceeds only after every possible failure has been eliminated may never proceed at all.
The objective is to distinguish assumptions that should be tested from risks that can be monitored and uncertainty that is justified by the prospective value of the research.
Current NIH feasibility criteria make a similar distinction by asking whether an approach is achievable while recognizing that uncertain feasibility can sometimes be acceptable when balanced by the potential for major advances.
Watch Out
The most dangerous assumption is not necessarily the one you feel least certain about. It is often the assumption whose failure would invalidate the study after the point at which changing the design has become difficult, expensive, or impossible.
