03 · What You Need to Know
Recognizing When Ethical Constraints Require Substantive Research Reformulation
Why Are Research Questions and Methods Not Interchangeable?
A research question specifies what investigators seek to understand or establish. A method specifies how evidence will be generated and analyzed to address that question.
Although researchers sometimes develop questions and methods iteratively, the relationship between them must remain logically defensible.
For example, the question "Does an AI-supported intervention improve students' writing performance?" concerns a potential causal effect. The question "How do students perceive the influence of AI writing tools on their writing?" concerns reported experiences and interpretations.
Both questions may be valuable, but they require different forms of evidence and support different conclusions.
Replacing an experiment with interviews does not necessarily preserve the original causal question. Interviews may reveal perceived mechanisms or experiences without establishing the intervention's effect on performance.
The central issue is whether the revised method still supports the inference that the question requires.
When Is Changing the Method Enough?
A methodological change may be sufficient when the revised procedure preserves the essential research question and can generate adequately valid evidence.
For example, a researcher studying the prevalence of a sensitive experience may replace identifiable interviews with an appropriately designed confidential survey. If the revised sampling and measurement procedures remain suitable, the original descriptive question may still be answerable.
Similarly, a researcher investigating the effect of an educational intervention might replace an ethically problematic no-support control condition with an acceptable active comparator.
However, the latter change may also alter the causal contrast. Estimating the effect of an intervention relative to standard support is not necessarily equivalent to estimating its effect relative to receiving no support.
Researchers should therefore examine whether the change affects only the procedure or also the quantity, relationship, or phenomenon being estimated.
When Does the Research Question Itself Need to Change?
Reformulation becomes necessary when the ethically acceptable evidence cannot support the original question's central claim.
This may occur when researchers cannot ethically manipulate the exposure of interest, observe a sensitive process directly, collect essential measurements, recruit the intended population, or access information required for the planned inference.
The limitation must be substantive rather than merely inconvenient. Ethical restrictions should not automatically lead investigators to abandon questions that remain answerable through appropriate alternatives.
Nevertheless, retaining the original question while collecting evidence incapable of answering it risks producing misleading conclusions.
| Ethical constraint |
Possible methodological adjustment |
When question reformulation may be necessary |
| Experimental manipulation is unacceptable. |
Use an ethically permissible comparator or natural experiment. |
When available designs cannot support the intended causal inference. |
| Sensitive information cannot be collected. |
Use less intrusive indicators or appropriately protected records. |
When the available indicators measure a materially different construct. |
| The intended population cannot be recruited ethically. |
Revise recruitment procedures or establish appropriate safeguards. |
When the revised population changes the group to which the question applies. |
| Repeated observations impose excessive burden. |
Reduce measurement frequency or use existing records. |
When temporal changes central to the question can no longer be examined adequately. |
| Direct observation creates unacceptable risks. |
Use simulations, interviews, or secondary evidence. |
When the alternative examines perceptions or simulated behavior rather than the original real-world phenomenon. |
These distinctions require methodological judgment. The presence of a constraint does not automatically establish that reformulation is necessary.
How Can You Tell Whether the Original Question Remains Answerable?
Begin by identifying the essential components of the question.
In quantitative research, these may include the target population, exposure or intervention, comparison, outcome, time frame, and intended inference. For causal questions, the estimand should be specified where appropriate.
An estimand describes the quantity the study intends to estimate. For example, the average effect of offering an AI tutoring intervention relative to standard support is a different estimand from the effect of actual voluntary AI tutoring use.
In qualitative research, the essential components may include the phenomenon of interest, participants' experiences, setting, context, and interpretive purpose.
Ask whether the ethically permissible evidence still addresses these components.
If the population changes, the question's scope may need revision. If the outcome becomes a proxy for a different construct, the wording should reflect that difference. If causal identification is no longer defensible, the question may need to focus on associations or another scientifically appropriate objective.
This assessment should be based on the actual capabilities of the design, not merely on the labels researchers assign to it.
Why Is a Causal Question Particularly Sensitive to Methodological Changes?
Causal questions require evidence that supports a comparison between outcomes under different exposure or intervention conditions.
Randomized experiments can provide strong causal evidence when implemented appropriately. However, causal inference is not limited to randomized research. Certain observational and quasi-experimental designs can also support causal claims under defensible assumptions.
The difficulty arises when ethical constraints remove the conditions necessary for the proposed identification strategy.
Suppose researchers want to know whether AI writing assistance reduces independent writing performance. If the available data contain only students' self-reported AI use and perceived writing confidence, the analysis cannot directly establish the effect on objectively assessed writing performance.
Statistical adjustment cannot compensate for an outcome that was never measured or automatically eliminate unmeasured confounding.
Hernán and Robins' framework for causal inference emphasizes the importance of defining causal questions and the assumptions under which observational evidence may address them.
Researchers should therefore distinguish between an ethically permissible observational design that can credibly identify a causal effect and one that supports only descriptive or associational conclusions.
Can Simulations Preserve the Original Research Question?
Simulations can provide valuable evidence when direct investigation would expose participants to unacceptable risks.
For example, students might evaluate fictional AI-generated feedback rather than receive deliberately misleading advice during consequential coursework.
However, simulated decisions may differ from real decisions because participants know that the consequences are limited or hypothetical.
The simulation may therefore investigate recognition of misleading information, perceived credibility, or decision-making under controlled conditions rather than the actual educational consequences of deception.
Researchers should examine ecological validity and the correspondence between simulated conditions and the real-world phenomenon.
A simulation does not necessarily require a different research question, but it may require narrowing the question to what the simulated environment can credibly establish.
What if Ethical Restrictions Change the Population Being Studied?
Population changes can affect the meaning and applicability of research findings.
Suppose a study initially seeks to understand the experiences of students who have experienced severe academic exclusion. An ethics committee determines that the proposed recruitment procedure creates unacceptable risks and requires changes.
If safer recruitment procedures allow the same population to participate, the original question may remain intact.
However, if the researcher instead recruits only students without such experiences, the revised study cannot answer the original question about academic exclusion.
Similarly, replacing a population experiencing substantial vulnerability with a more accessible population may change the social and scientific relevance of the research.
Researchers should avoid presenting findings from a substantially different population as though they directly represent those originally targeted.
The ethical consequences of exclusion should also be examined, particularly when protective restrictions may unnecessarily limit participation.
How Should Researchers Handle the Loss of Essential Measurements?
Ethical restrictions sometimes prevent collection of information central to the proposed construct.
For example, a study of academic dishonesty may originally require access to identifiable disciplinary records. If those records cannot lawfully or ethically be used, researchers may consider anonymous self-reports.
However, self-reported behavior and officially documented disciplinary incidents are not interchangeable measures.
The former may be affected by recall, social desirability, and willingness to disclose. The latter may reflect institutional detection and reporting practices rather than the full occurrence of misconduct.
Researchers should assess whether the alternative measure adequately represents the intended construct.
If it does not, the question should be revised to specify what is actually being measured rather than retain terminology implying access to unavailable evidence.
Why Should Researchers Avoid Reframing Questions Merely to Obtain Approval?
Changing a research question can be a responsible response to ethical constraints. It can also become problematic when the revision is superficial.
For example, changing the wording of a proposal from "measuring the harmful effects of deception" to "examining learning experiences" does not resolve the ethical concern if the same deceptive intervention remains in place.
Ethics review evaluates the actual procedures, risks, consent arrangements, and intended uses of information, not merely the language used to describe the project.
A legitimate reformulation changes the substantive objective or procedures where necessary and accurately represents the evidence the study will produce.
Researchers should explain how the revised question relates to the original one and why the new design is ethically and scientifically defensible.
When Should Question Reformulation Happen?
Ideally, substantial changes occur before data collection, when researchers can revise the protocol, obtain necessary approvals, and align instruments and analysis plans with the revised objective.
However, ethical concerns may emerge during recruitment, fieldwork, analysis, or dissemination.
If the original question becomes ethically or scientifically untenable after data collection has begun, researchers should document the circumstances and determine whether the revised objective requires additional review or consent.
Changes made after inspecting results require particular transparency.
Researchers should not present a post hoc question as though it were prospectively specified. Doing so may misrepresent the evidentiary status of the findings and obscure selective analytical decisions.
Exploratory reformulation can be legitimate when clearly reported, but it should not be disguised as confirmatory testing.
Can an Ethically Reformulated Question Still Make an Important Contribution?
Yes. A narrower question may produce valuable evidence when it is well matched to the available data and addresses a meaningful knowledge gap.
For instance, a researcher unable to establish the causal effects of AI-assisted writing may still investigate how students interpret AI feedback, how they decide whether to accept suggestions, or how institutional policies shape their use of writing tools.
These questions may contribute to educational technology research without requiring investigators to claim evidence they do not possess.
The scientific value of the revised question should be assessed independently rather than justified solely by its relationship to the original proposal.
Where the ethically acceptable design is methodologically weaker, the consequences for scientific validity should be explicitly evaluated.
When Should Researchers Stop Trying to Preserve the Original Question?
Researchers may become attached to a question because it has already received funding, institutional endorsement, or substantial preparatory work.
However, previous investment does not establish that the original objective remains feasible or ethically justified.
When all reasonable alternatives fail to support the intended inference, continuing to preserve the question may encourage methodological overstatement or increasingly problematic procedures.
At that point, researchers should consider whether a different question can produce valuable knowledge without exposing participants to unnecessary burdens.
If no defensible alternative remains, discontinuing the research project may be more appropriate than forcing the available evidence to answer an unanswerable question.
