What Should You Do When the Best Method for Answering a Question Would Be Unethical?

What If the Methodologically Strongest Study Is One You Should Not Conduct?

Research methods are often taught as a search for the design that can answer a question most convincingly. If you want to estimate a causal effect, for example, an appropriately designed randomized experiment may provide stronger causal evidence than a simple observational comparison. If you want highly detailed information, collecting identifiable individual-level data may seem preferable to working with aggregated records.

But methodological strength is not the only constraint on research design. The method that would provide the cleanest comparison, strongest causal inference, most precise measurement, or richest data may require deliberately exposing people to harm, withholding something they need, invading privacy, deceiving participants without adequate justification, or obtaining information that researchers cannot ethically or legally access.

When that happens, the goal is not to find a clever way around ethical protections. It is to determine what evidence can responsibly be obtained and then make sure the research question and eventual conclusions do not claim more than that evidence can support.

Choose the Strongest Ethical Method, Not the Strongest Method in the Abstract

Ethics Sets Boundaries on Methodological Optimization

The "best" method depends on what you mean by best

Researchers sometimes speak of the "best method" as though research design can be ranked on a single scale. In practice, methods involve trade-offs among internal validity, external validity, measurement quality, feasibility, participant burden, cost, access, and ethical acceptability.

A method may therefore be excellent for one purpose and unacceptable overall.

Consider a causal question: Does prolonged severe sleep deprivation impair complex decision-making? Deliberately assigning participants to substantial sleep deprivation might provide stronger causal leverage than simply comparing people who naturally sleep different amounts. Yet creating the exposure could introduce risks that make the proposed experiment ethically difficult or unacceptable.

The fact that randomization would strengthen causal inference does not settle whether randomization should be used. Ethical acceptability is not an optional characteristic added after the supposedly ideal methodology has been chosen.

For actual research, the second definition is generally more useful.

Sound methodology and research ethics are not opposing goals

It can be tempting to frame the situation as a conflict between scientific rigor and ethics: ethics forces researchers to use an inferior study. That framing is too simple.

Ethical frameworks themselves recognize the importance of scientific validity. Under the U.S. Common Rule, for research to which it applies, risks to participants must be minimized through procedures consistent with sound research design that do not unnecessarily expose participants to risk. Risks must also be reasonable in relation to anticipated benefits, if any, and the importance of the knowledge that may reasonably be expected to result.

These requirements connect scientific quality to ethical justification. If a design is so weak that it is unlikely to produce the knowledge invoked to justify participant risk, methodological weakness becomes ethically relevant too. OHRP's Secretary's Advisory Committee on Human Research Protections has similarly noted that exposing individuals to research risk cannot be justified when the study design is unlikely to answer the research question.

The objective is therefore not simply to make a study safer at any methodological cost. It is to develop a study that is both ethically defensible and capable of producing worthwhile knowledge.

First determine exactly what makes the preferred method unethical

Before replacing a method entirely, identify the ethical problem precisely. "This experiment is unethical" is a conclusion, not yet a diagnosis.

The concern might be that the study would deliberately create a harmful exposure. It might require withholding an established beneficial intervention, collecting unnecessarily identifiable information, placing participants at substantial psychological risk, recruiting people under circumstances that compromise voluntary participation, or creating privacy and confidentiality risks that cannot be adequately controlled.

Different problems have different remedies. If the ethical difficulty comes from one feature of the method rather than its entire logic, modifying that feature may preserve much of the design's evidentiary strength.

Ask whether the risky element is scientifically necessary

A useful next question is deceptively simple: must the ethically problematic element be present for the research question to be answered?

Suppose a researcher wants to study the consequences of exposure to harmful online harassment. Deliberately subjecting participants to severe harassment would create an obvious ethical concern. But the phenomenon already occurs outside the research setting. Depending on the precise question, researchers might study naturally occurring exposure, analyze ethically and lawfully accessible records, conduct retrospective interviews with suitable safeguards, or examine less harmful experimental analogues.

Similarly, if a study seems to require highly identifiable information, ask whether the analysis truly needs every identifier. If the proposed design requires deception, ask whether participants can be given enough information for meaningful consent without invalidating the study. If a control condition appears problematic, ask whether another comparator can answer the scientific question.

The purpose is not to weaken the study automatically. It is to distinguish methodological necessities from features that are merely convenient, conventional, or desirable.

Natural variation can sometimes replace researcher-created exposure

One common ethical redesign is to observe a condition that already occurs rather than deliberately create it.

Researchers cannot ethically assign people to many potentially harmful exposures simply to estimate their effects. Smoking, severe deprivation, traumatic events, dangerous occupational conditions, discrimination, and many diseases are obvious examples. Yet researchers can often investigate their consequences through observational designs.

Depending on the research context and available data, possibilities may include cohort studies, case-control studies, longitudinal studies, interrupted time-series designs, natural experiments, quasi-experiments, or analyses of existing records.

These alternatives do not magically reproduce randomization. Confounding, selection effects, measurement problems, and other sources of bias may remain. The methodological task becomes identifying and addressing those limitations rather than behaving as though the ethical alternative is evidentially identical to the prohibited experiment.

A natural experiment is not the same as creating the exposure yourself

Sometimes circumstances outside the researcher's control create variation that can be studied. A policy changes in one jurisdiction but not another. An institution introduces a program in stages. An environmental event affects some locations more than others. Eligibility rules create a threshold that influences who receives an intervention.

Such situations can sometimes support stronger causal inference than ordinary observational comparisons because exposure is generated by a process external to the researcher's manipulation.

Whether a particular situation genuinely functions as a useful natural or quasi-experiment depends on its assumptions and design. Researchers should not attach the label simply because they did not randomize participants.

Still, the broader principle is valuable: researchers may sometimes study ethically problematic exposures without causing those exposures themselves.

Existing data can reduce some risks, but secondary analysis is not automatically ethical

Another alternative is to use information that already exists. Administrative records, health data, educational records, archived datasets, registries, previous studies, and other sources may permit investigation without asking participants to undergo a new intervention or repeated data collection.

This can reduce certain burdens, but existing data do not come with an ethical free pass. Researchers still need to consider whether they may legitimately access and use the information, whether consent or an appropriate waiver or other authorization is required, whether data are identifiable or re-identifiable, and how privacy and confidentiality will be protected.

If the evidence you need cannot legally or ethically be accessed, the fact that the data already exist does not solve the problem.

Sometimes you can study a safer analogue rather than the harmful phenomenon itself

Researchers may occasionally be able to replace a severe exposure with a lower-risk analogue that captures part of the underlying process.

For example, rather than deliberately creating extreme psychological distress, a study might use a carefully validated task that produces a milder, temporary response within ethically acceptable limits. Whether this works depends on the research question. A safer analogue is useful only if it represents the construct or mechanism relevant to the intended inference.

This introduces a trade-off between ethical risk reduction and construct validity. If the analogue becomes so mild or artificial that it no longer represents the phenomenon of interest, the study may be safe but incapable of answering the original question.

At that point, changing the wording of the question is more scientifically honest than pretending the substitution made no difference.

Changing the population may reduce risk, but it can also change the question

A researcher might consider moving a study from a population requiring additional protections to another population for whom the procedure appears less risky. Sometimes that is appropriate. Sometimes it undermines the purpose of the research.

If the phenomenon specifically concerns children, for example, studying adults simply because they are easier to recruit does not necessarily answer the same question. Likewise, routinely excluding people who are perceived as vulnerable can produce systematic evidence gaps and deny those populations the potential benefits of research knowledge.

The better approach is to ask whether inclusion is scientifically necessary and, if so, whether the research can be designed with adequate safeguards. Planning research with populations who may require additional protections calls for careful attention to both justified inclusion and participant protection.

Method changes often require claim changes

This is one of the most consequential points.

If ethics requires you to replace a randomized experiment with an observational study, you cannot simply retain the same causal language because the original question sounds more compelling. If you replace identifiable individual-level information with aggregated data, you may lose the ability to make individual-level inferences. If you substitute retrospective self-report for direct measurement, different measurement limitations arise.

Ethical redesign can therefore produce a chain of changes:

This is not methodological failure. It is alignment between ethics, design, evidence, and inference.

There may be no ethical substitute that answers exactly the same question

Researchers sometimes assume that every ethically prohibited method must have a harmless substitute capable of generating equivalent evidence. That is not always true.

Some questions are difficult precisely because the strongest evidence would require researchers to create conditions they should not create. Ethical alternatives may provide indirect evidence, weaker causal identification, different measurements, or evidence about only part of the original phenomenon.

When this occurs, there are two intellectually honest responses. You can accept a more limited question that the available method can answer, or retain the broader question while acknowledging that the study provides only partial evidence about it.

In more extreme cases, an important question may remain worth asking even though no ethical study can answer it directly. Ethical constraints can create genuine limits on empirical knowledge. Research design cannot always engineer those limits away.

Formal ethics review remains necessary when applicable

A researcher's own conclusion that an alternative method is safer does not replace formal ethics review when such review is required.

For research governed by the U.S. Common Rule, an institutional review board evaluates criteria including risk minimization, the reasonableness of risks in relation to anticipated benefits and expected knowledge, equitable participant selection, informed consent, privacy and confidentiality protections where appropriate, and additional safeguards for participants vulnerable to coercion or undue influence.

In medical research involving human participants, the current 2024 Declaration of Helsinki requires research protocols to undergo consideration, comment, guidance, and approval by a research ethics committee before the research begins. Other jurisdictions, institutions, disciplines, and types of research may operate under different requirements.

Replacing a Harmful Experiment Without Pretending the Evidence Is Equivalent

The revised question is less causally ambitious, but it accurately reflects what the alternative design can establish. Further evidence from longitudinal patterns, natural experiments, converging studies, or other designs might later strengthen causal interpretation.

The key is not to preserve causal wording at all costs. The key is to obtain the strongest defensible evidence and describe its implications accurately.

Common Mistakes When Ethics Rules Out a Preferred Method

Redesign From the Ethical Constraint Outward

If the preferred method appears unethical, begin by identifying exactly which feature creates the problem. Then work outward from that constraint rather than immediately abandoning the research topic.

Your objective is to preserve as much of the original inferential goal as possible without weakening necessary protections. Each redesign should therefore be evaluated twice: first for whether it adequately addresses the ethical problem, and second for what evidence it can still produce.

It is useful to confront these issues before the research question becomes difficult to revise. Once researchers have invested heavily in a particular hypothesis, design, grant proposal, or protocol, there can be a natural tendency to defend the original method rather than reconsider the underlying assumptions.

You should also distinguish an ethically complicated study from one that is simply too risky to conduct in its proposed form. Some important research legitimately requires extensive safeguards and careful review. The existence of ethical complexity is not itself a reason to retreat from a worthwhile question.

What matters is that the final chain remains coherent: the question justifies the evidence needed, the method obtains that evidence within acceptable ethical boundaries, and the conclusions remain proportionate to what the method can establish.

What to Check When Your Preferred Method Is Ethically Problematic

Frequently Asked Questions About Unethical Research Methods

The Best Research Method Must Also Be Ethically Defensible

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.