5. Research Design, Risk–Benefit Analysis, and Ethical Protocols

Introduction and Context

Ethical research is not created by adding a consent form to an otherwise completed study. Ethics is embedded in the way a research problem is framed, the evidence is produced, participants are selected, technologies are configured, and findings are communicated. Research design and research ethics are therefore mutually dependent. A methodologically weak study may expose people, communities, institutions, or environments to burdens without producing reliable knowledge. Conversely, a scientifically valuable study may still be unethical if it distributes risks unfairly, excludes relevant groups, conceals conflicts of interest, or treats data subjects merely as sources of information.

This relationship is especially important in information technology research. Studies involving artificial intelligence, cybersecurity, digital health, learning analytics, biometric systems, online communities, human–computer interaction, and large-scale data frequently involve indirect observation, automated decision-making, re-identification risks, and complex power relationships. Researchers may not physically intervene in participants’ lives, yet their work can influence employment, access to services, reputation, privacy, safety, and democratic participation.

An ethical protocol is consequently more than an administrative submission to a research ethics committee. It is a reasoned account of how a study will generate valid knowledge while respecting persons, promoting welfare, distributing burdens fairly, protecting information, and remaining accountable to affected stakeholders. This chapter develops an integrated approach to scientific validity, proportionality, stakeholder analysis, conflicts of interest, harm mitigation, inclusion, cultural sensitivity, and ethics-by-design.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Explain why scientific validity is an ethical requirement.

2. Distinguish risks, burdens, benefits, uncertainties, and vulnerabilities in research.

3. Apply proportionality and stakeholder analysis to research planning.

4. Identify and manage financial, professional, institutional, and intellectual conflicts of interest.

5. Design protocols that incorporate inclusion, accessibility, cultural sensitivity, privacy, and security.

6. Evaluate ethical issues in automated, data-intensive, and AI-enabled research.

7. Construct a defensible risk–benefit assessment and mitigation plan.

8. Critically assess ethical protocols and recognize their limitations.

Scientific Validity as an Ethical Obligation

Scientific validity concerns whether a study can answer its stated question using appropriate concepts, methods, evidence, and analysis. It includes a coherent research question, suitable design, credible sampling, reliable measurement, appropriate analysis, transparent reporting, and a realistic implementation plan. Validity is not only a methodological virtue. It is an ethical condition because participants assume burdens—such as time, inconvenience, exposure of information, or possible distress—in the expectation that their contribution may produce useful knowledge.

A study that is incapable of producing interpretable findings cannot justify those burdens. For example, an intervention study with an inadequate comparison group may expose participants to experimental conditions while making it impossible to determine whether the intervention worked. A machine-learning study based on biased or unusable data may generate apparently precise outputs that are not fit for deployment. A survey with a leading instrument may waste participants’ time and produce misleading conclusions that affect policy.

Scientific validity should be assessed at several levels:

• Conceptual validity: Do the constructs being studied correspond to the research question?
• Design validity: Can the chosen design support the proposed inference?
• Measurement validity: Are variables measured accurately and consistently?
• Statistical or computational validity: Are the analytical procedures appropriate, reproducible, and sufficiently powered?
• Ecological and contextual validity: Do the findings have meaning in the setting and population under study?
• Reporting validity: Are limitations, missing data, uncertainty, and deviations disclosed honestly?

At doctoral level, researchers must also distinguish exploratory, descriptive, predictive, and causal claims. A predictive model may identify correlations without establishing causes. A technically high-performing system may perform poorly for underrepresented groups. A statistically significant result may have little practical or social importance. Ethical review should therefore examine not only whether a method is technically competent, but also whether the proposed claims are proportionate to the evidence.

From Risk–Benefit Assessment to Proportionality

Risk is the possibility that a research activity will produce harm, together with the seriousness and likelihood of that harm. Harm may be physical, psychological, social, economic, legal, reputational, informational, cultural, or environmental. In IT research, informational and inferential harms are particularly important. A dataset may reveal sensitive characteristics even when direct identifiers have been removed; a model may infer health status, political preference, or vulnerability from apparently ordinary behavioral data.

Benefits may accrue directly to participants, to a participating organization, to a community, to future users, or to society. Researchers should distinguish anticipated benefits from speculative promises. Access to a study intervention is not necessarily a benefit if its effectiveness is unknown. Nor should a researcher describe publication, academic recognition, or commercial gain as a participant benefit unless this is relevant and transparently presented.

A useful risk–benefit assessment asks:

1. What could go wrong, and who could be affected?

2. How severe would each harm be?

3. How probable is it, given the design and context?

4. How long might the harm last, and can it be reversed?

5. Who receives the benefits, and who bears the burdens?

6. Could the same knowledge be obtained with less risk?

7. What uncertainties remain, and how will they be monitored?

Proportionality is the principle that the expected value and social importance of a study must justify its residual risks and burdens. Proportionality does not mean that risk must be eliminated; many valuable studies involve some uncertainty. It means that risks should be reasonable in relation to the importance of the question, the quality of the design, the availability of safeguards, and the distribution of benefits and burdens.

A practical assessment can classify risk as minimal, moderate, substantial, or high, but labels alone are insufficient. “Minimal risk” is context-dependent. A data breach involving an ordinary preference may be inconvenient, whereas disclosure of immigration status, medical information, or political activity may have severe consequences. The assessment should document both severity and likelihood, rather than relying on a single aggregate score.

Risk management follows a hierarchy:

1. Avoidance: Remove unnecessary data collection or eliminate a hazardous procedure.

2. Substitution: Use a less intrusive measure, simulated environment, or synthetic data.

3. Reduction: Minimize sample exposure, collect fewer identifiers, or shorten participation.

4. Protection: Apply access controls, encryption, monitoring, training, and support.

5. Response: Establish procedures for adverse events, complaints, breaches, withdrawal, and incident reporting.

6. Review: Reassess risk when the protocol, technology, population, or context changes.

Residual risk is the risk remaining after safeguards have been applied. Ethical approval should concern residual risk, not merely the risk of the original design.

Stakeholders, Power, and the Distribution of Burden

Stakeholder analysis extends ethical review beyond enrolled participants. Stakeholders may include direct participants, non-participant data subjects, families, communities, employees, students, service users, researchers, institutions, funders, technology vendors, regulators, future users, and groups affected by deployment. In digital research, a person can be affected without ever consenting to participate. For example, a workplace monitoring study may use employee records that influence colleagues who are not included in interviews; a public social-media dataset may contain information about bystanders; an algorithm evaluated in one hospital may later affect patients outside the study.

Stakeholder analysis should identify interests, vulnerabilities, decision-making power, exposure to harm, and capacity to benefit. A simple matrix can classify stakeholders according to influence and impact, but researchers should avoid assuming that powerful institutions represent less powerful groups. Consultation with a university, company, or community leader does not automatically constitute consent from individuals.

Power relationships can distort voluntariness. Employees may feel unable to refuse a manager’s invitation. Students may believe that participation affects grades. Patients may fear that refusal will affect care. Researchers should separate recruitment from authority where possible, use neutral communication, make non-participation genuinely consequence-free, and provide private channels for questions and withdrawal.

Justice requires attention to the distribution of research benefits and burdens. Convenience sampling can repeatedly place burdens on accessible or disadvantaged groups while the benefits flow to institutions, commercial partners, or affluent users. Inclusion does not mean enrolling every possible population. It means providing a defensible reason for inclusion and exclusion, ensuring that exclusion is not caused by avoidable barriers, and adapting procedures to accessibility, language, technology access, and cultural context.

Inclusion, Cultural Sensitivity, and Respectful Engagement

Inclusive design begins before recruitment. Researchers should examine whether eligibility criteria unnecessarily exclude people with disabilities, limited digital access, minority-language backgrounds, older age, or different cultural practices. Instruments should be tested for linguistic and cultural appropriateness rather than translated mechanically. Accessibility may require screen-reader compatibility, captions, alternative input methods, flexible scheduling, low-bandwidth options, or assistance that does not compromise privacy.

Cultural sensitivity is more than politeness. It concerns the validity and ethical meaning of research in a particular social context. Concepts such as privacy, family authority, individual consent, data ownership, and acceptable disclosure may vary across communities. Researchers must avoid both cultural imposition and cultural romanticization. Community consultation can reveal risks invisible to an external research team, but consultation does not replace individual consent when individual participation is involved.

Some communities may possess collective interests in data, knowledge, images, language, or cultural practices. Researchers should clarify who may authorize access, how findings will be returned, whether secondary use is acceptable, and whether commercial exploitation is contested. Community engagement should be continuous where feasible, including during interpretation and dissemination. Reciprocity may involve accessible findings, training, infrastructure, compensation, or opportunities for community review, but it should not become a form of inducement that undermines voluntary participation.

Compensation must reimburse time and expenses fairly without becoming so large that it pressures economically vulnerable participants. Researchers should distinguish compensation from benefits and disclose payment schedules, tax implications where relevant, and whether partial participation receives partial payment.

Ethical Protocols as Living Documents

An ethical protocol translates ethical commitments into operational decisions. It should describe:

• the research question, rationale, design, and anticipated contribution;
• participant groups, recruitment channels, inclusion and exclusion criteria;
• consent procedures, comprehension checks, withdrawal arrangements, and compensation;
• data categories, collection methods, ownership, access, retention, deletion, and sharing;
• foreseeable risks, benefits, vulnerabilities, and mitigation measures;
• privacy, confidentiality, cybersecurity, and breach-response arrangements;
• conflicts of interest and independence of analysis;
• procedures for adverse events, complaints, incidental findings, and protocol deviations;
• responsibilities of investigators, partners, vendors, and data processors;
• dissemination, authorship, intellectual property, and return of results.

Consent should be a process rather than a single signature. Meaningful consent requires understandable information, adequate time, freedom from undue influence, and an opportunity to ask questions. Digital consent interfaces should not conceal important information in lengthy legal text or use manipulative defaults. For longitudinal or adaptive studies, researchers may need re-consent when the purpose, data uses, risk level, or technology changes materially.

Privacy and confidentiality are related but distinct. Privacy concerns the person’s control over access to themselves and their information; confidentiality concerns the researcher’s duty to protect information entrusted to them. De-identification reduces risk but does not guarantee anonymity. Re-identification may occur through linkage, rare attributes, timestamps, facial images, location traces, or model outputs. Data minimization, pseudonymization, separation of identifiers, role-based access, encryption, secure deletion, logging, and controlled sharing are therefore parts of the ethical design, not merely technical enhancements.

Conflicts of Interest and Research Independence

A conflict of interest exists when a secondary interest could unduly influence professional judgment or create a reasonable perception of compromised independence. Conflicts may be financial, institutional, personal, relational, professional, ideological, or intellectual. Examples include equity in a company whose product is being evaluated, funding from a vendor, a supervisor assessing a subordinate’s performance, or a researcher seeking confirmation of a favored theory.

A conflict is not necessarily misconduct. The ethical requirement is recognition, disclosure, assessment, and management. Management may involve independent recruitment, blinded analysis, separation of sponsor and investigator decisions, independent data monitoring, preregistration, audit trails, independent statistical review, publication guarantees, or recusal from particular decisions. Disclosure alone is inadequate when the conflict is substantial and controllable safeguards are available.

Contracts deserve careful scrutiny. Sponsors should not have unrestricted control over data interpretation, adverse-event reporting, or publication. Researchers should identify who owns data, who can access source code, whether negative results will be reported, and what happens if the sponsor disputes the findings. Institutional prestige and career incentives can also threaten independence, even when no money is involved.

Ethics-by-Design in Computer Applications Research

Ethics-by-design integrates ethical analysis into system architecture, workflow, and evaluation from the beginning. It is analogous to security-by-design and privacy-by-design, but it also addresses fairness, explainability, accessibility, contestability, sustainability, and human oversight.

For an AI or automated decision system, an ethics-by-design process may include:

1. Defining the legitimate purpose and prohibited uses.

2. Mapping affected stakeholders, including non-users and vulnerable groups.

3. Assessing data provenance, consent, representativeness, and quality.

4. Selecting performance and fairness measures appropriate to the context.

5. Minimizing collection and retaining only necessary data.

6. Designing human review, appeal, correction, and override mechanisms.

7. Testing under realistic conditions and across relevant subgroups.

8. Documenting assumptions, limitations, model changes, and uncertainty.

9. Monitoring drift, disparate impact, misuse, and unexpected consequences.

10. Establishing clear accountability when the system fails.

No single fairness metric resolves every ethical problem. Equal error rates, demographic parity, calibration, and individual fairness can conflict, especially when base rates differ. Metric selection must therefore be justified in relation to the decision’s purpose, stakes, and affected communities. Technical fairness is also insufficient if the underlying decision should not be automated or if people lack meaningful opportunity to challenge it.

Generative AI introduces additional concerns: fabricated outputs, leakage of confidential material, opaque training data, automated profiling, authorship ambiguity, and overreliance by researchers or participants. Protocols should specify whether AI tools may be used, what data may be entered, how outputs will be verified, and how human responsibility will be maintained. AI should not be treated as an ethical reviewer capable of replacing contextual judgment.

An Integrated Protocol Development Process

A robust protocol can be developed through the following sequence.

First, formulate the knowledge claim. State what the study will establish, what it will not establish, and why the question matters. This prevents inflated claims from justifying intrusive methods.

Second, map the research system. Identify participants, data subjects, institutions, technologies, vendors, decision-makers, and downstream users. Trace data flows from collection to deletion or publication.

Third, identify vulnerabilities and power. Ask who can refuse freely, who may be exposed indirectly, whose language or access needs are overlooked, and who controls resources or outcomes.

Fourth, compare design alternatives. Consider whether the study can use less data, a smaller sample, synthetic data, simulation, aggregation, staged deployment, or secondary data with stronger protections.

Fifth, assess risks and benefits. Record severity, probability, duration, reversibility, affected stakeholders, uncertainty, and mitigation. Separate direct benefits from societal or speculative benefits.

Sixth, build safeguards into procedures and systems. Assign responsibility for each safeguard and define how compliance will be checked.

Seventh, plan consent, communication, and remedy. Participants should know what will happen, how to withdraw, whom to contact, and what support is available if harm occurs.

Eighth, establish monitoring and stopping rules. Define conditions for pausing recruitment, modifying the protocol, reporting incidents, or terminating the study.

Ninth, obtain independent review. Ethical review should occur before implementation and whenever material changes occur. Independence is particularly important where investigators, institutions, or funders have strong interests in continuation.

Tenth, report responsibly. Publish methods, limitations, exclusions, missingness, adverse events, conflicts, and relevant negative findings. Responsible dissemination prevents technically correct results from being misleading in practice.

Illustrative Case Study: Predictive Analytics in Student Support

Background

A university proposes an early-warning system that uses learning-management-system activity, assessment results, attendance records, and demographic information to predict which students may discontinue their studies. Advisers will receive risk scores and may contact students with offers of academic and financial support. The project is presented as a way to improve retention and allocate scarce support resources.

Situation and Ethical Problem

The research team wants to train a model using five years of historical data. Students were not originally told that their activity would be used for predictive profiling. Some demographic variables improve predictive performance, but their use may reproduce historical inequality. Students may also feel stigmatized if contacted because a system classifies them as “at risk.” The university’s technology partner requests access to identifiable data and hopes to commercialize a similar product.

Analysis

The study has potential social benefit, but its validity and ethical acceptability are uncertain. Historical withdrawal patterns may reflect institutional barriers rather than individual deficiencies. A high-performing model may therefore predict disadvantage without identifying a fair or effective intervention. False positives could burden students with unwanted attention, while false negatives could leave vulnerable students unsupported.

Stakeholders include students, advisers, instructors, the university, the vendor, future cohorts, and groups historically underserved by the institution. Students have limited ability to avoid data generation if learning systems are compulsory. The vendor’s commercial interest is a conflict requiring disclosure and strict data governance.

The research team should determine whether all proposed variables are necessary, whether consent or another legitimate governance basis is appropriate, and whether students can meaningfully object. It should examine subgroup performance, missing data, proxy discrimination, and the consequences of intervention. Advisers must not treat a score as a fact, and students should have access to understandable explanations and a route to challenge inaccurate information.

Approach and Safeguards

The team redesigns the project in stages. First, it conducts a retrospective feasibility analysis using de-identified and aggregated data. It excludes highly intrusive features, such as private message content and precise location traces. It evaluates subgroup error rates and consults student representatives, disability services, advisers, and data-protection specialists.

A prospective pilot then uses a limited set of validated indicators. Risk scores are visible only to trained advisers, are treated as prompts for supportive conversation rather than automated decisions, and cannot affect grades, discipline, or eligibility for services. Students receive clear information about the pilot, available support, data use, retention, and complaint procedures. The vendor receives only the minimum data necessary under a contract prohibiting unrelated use, re-identification, and commercialization.

The protocol includes independent evaluation, a stopping rule for evidence of disparate harm, regular model review, secure access logs, deletion schedules, and a commitment to publish limitations and negative results. Support is offered universally through accessible channels so that students are not required to accept a label in order to obtain help.

Outcome

The pilot demonstrates that the model’s overall accuracy is moderate and that performance differs across student groups. The university decides not to automate eligibility decisions. Instead, it uses the findings to improve general advising capacity and address structural barriers identified through consultation. The predictive model remains a limited research tool rather than a gatekeeping mechanism.

Learning Points

The case illustrates that predictive accuracy does not establish ethical acceptability. Scientific validity requires examining what the model predicts and whether the predicted outcome is actionable. Proportionality favors staged evaluation and data minimization. Inclusion requires subgroup analysis and participation by affected communities. Ethics-by-design requires human judgment, contestability, governance, and the possibility of not deploying a technically feasible system.

Challenges, Limitations, and Common Errors

Ethical protocols can become bureaucratic documents detached from actual practice. A carefully written plan may fail if researchers misunderstand the technology, if staff are inadequately trained, or if incentives reward rapid recruitment and positive results. Risk assessments may also underestimate harms that are indirect, cumulative, delayed, or distributed across populations.

Common errors include treating anonymization as absolute, obtaining broad consent without meaningful explanation, using vulnerable groups because they are easy to access, confusing institutional permission with participant consent, promising confidentiality that cannot be guaranteed, ignoring non-participant data subjects, and disclosing conflicts only after publication. Another error is applying generic ethical checklists without examining the social meaning of the particular technology.

Ethical review also has limitations. Committees may lack specialized expertise in machine learning, cybersecurity, disability, Indigenous data governance, or local cultural practice. Review takes place under uncertainty, while digital systems may evolve after approval. Continuous governance, community engagement, technical auditing, and post-deployment monitoring are therefore necessary complements to initial review.

Researchers should resist ethical overconfidence. A safeguard can reduce one risk while creating another: identity verification may improve data quality but increase surveillance; model transparency may reveal sensitive information; extensive data retention may support reproducibility but expand breach exposure. Ethical judgment requires examining trade-offs rather than declaring a design simply “ethical.”

Current Developments and Professional Practice

Contemporary research governance increasingly emphasizes data stewardship, algorithmic impact assessment, reproducibility, responsible innovation, and participatory approaches. Researchers are expected to document datasets and models, justify secondary use, evaluate distributional effects, and monitor systems beyond the point of publication. Open science can strengthen accountability, but open release must be balanced against privacy, security, intellectual property, and the risk of enabling misuse.

Doctoral researchers should cultivate an ethics vocabulary that extends beyond compliance. They should be able to explain why a design is valid, whose interests it serves, what uncertainty remains, and how affected people can influence or challenge the work. Supervisors and institutions should create environments in which researchers can report ethical concerns without retaliation and can pause a project when conditions change.

Best Practices for Researchers

• Link every participant burden to a clearly justified knowledge claim.
• Prefer the least intrusive design capable of answering the question.
• Analyze direct, indirect, cumulative, and downstream harms.
• Involve affected stakeholders before finalizing the protocol.
• Treat consent as ongoing and understandable.
• Minimize data and specify retention, access, deletion, and breach procedures.
• Test technologies across relevant populations and document uncertainty.
• Disclose and actively manage conflicts of interest.
• Provide accessible recruitment, participation, feedback, and complaint processes.
• Build human oversight, correction, appeal, and stopping mechanisms into automated systems.
• Reassess ethics when methods, data, partners, populations, or uses change.
• Report limitations, negative results, exclusions, and incidents honestly.

Chapter Summary

Ethical research design integrates scientific validity with respect for persons, welfare, justice, accountability, and social responsibility. A study must be capable of producing meaningful knowledge before its risks can be justified. Risk–benefit assessment should consider severity, probability, duration, reversibility, uncertainty, and the distribution of burdens and benefits. Proportionality supports the least intrusive design that can answer an important question.

Stakeholder analysis reveals power relationships and identifies people affected indirectly by data or technology. Inclusion and cultural sensitivity improve both fairness and validity when they shape recruitment, measurement, interpretation, and dissemination. Conflicts of interest require disclosure and active management, not merely a statement in a publication. Ethical protocols should operate as living documents with monitoring, incident response, and independent review.

In computer applications research, ethics-by-design incorporates privacy, security, fairness, accessibility, explainability, human oversight, contestability, and accountability into system development. AI and predictive systems must be evaluated not only for accuracy but also for purpose, distributional effects, actionability, and the consequences of error. Ethical research is therefore a continuing practice of justified design, critical reflection, stakeholder engagement, and responsible governance.

Key Terms

Scientific validity: The extent to which a study can reliably answer its stated research question.

Risk: The possibility and significance of harm resulting from research activity.

Benefit: A valuable outcome reasonably expected from research for participants or society.

Proportionality: The requirement that justified benefits and knowledge outweigh residual risks and burdens.

Residual risk: Risk remaining after safeguards have been implemented.

Stakeholder: A person, group, institution, or entity affected by or able to influence research.

Vulnerability: A condition that may limit autonomy or increase exposure to harm.

Justice: Fair distribution of research burdens, benefits, opportunities, and protections.

Informed consent: Voluntary agreement based on understandable and sufficient information.

Privacy: A person’s interest in controlling access to themselves and their information.

Confidentiality: The duty to protect information entrusted to the researcher.

Data minimization: Collection and retention of only the data necessary for a justified purpose.

Conflict of interest: A secondary interest that may improperly influence professional judgment.

Ethics-by-design: Incorporating ethical requirements into system and research design from the outset.

Contestability: The capacity of affected persons to question, challenge, or appeal a decision.

De-identification: Modification of data to reduce links to identifiable individuals.

Adverse event: An unexpected or harmful occurrence associated with research participation.

Protocol deviation: A departure from an approved research procedure.

Stakeholder engagement: Structured involvement of affected or interested parties in research decisions.

Algorithmic impact assessment: A systematic examination of the effects and risks of an automated system.

Review Questions

Short Answer Questions

1. Why is scientific validity considered an ethical requirement?

2. Distinguish privacy from confidentiality.

3. What is meant by residual risk?

4. Identify two ways in which power relationships can affect voluntary consent.

5. Why is disclosure alone insufficient for managing a serious conflict of interest?

Long Answer Questions

1. Explain proportionality in research ethics and describe how it should guide the selection of research methods.

2. Discuss how stakeholder analysis can reveal harms that are not visible when researchers focus only on enrolled participants.

3. Examine the relationship between inclusion, cultural sensitivity, and scientific validity.

4. Describe the contents of a strong ethical protocol for an IT research project involving personal data.

5. Explain how ethics-by-design should shape the development and evaluation of an AI-based decision system.

Analytical / Application Questions

1. A researcher proposes to analyze publicly available social-media posts to infer users’ mental-health status. Apply stakeholder analysis and risk–benefit reasoning to determine what safeguards are required.

2. A company funding a cybersecurity study asks to review results before publication and suppress findings that may damage its reputation. Identify the conflict-of-interest issues and propose governance measures.

3. An automated university admissions tool has high overall accuracy but substantially higher false-negative rates for applicants with disabilities. Analyze whether deployment is ethically defensible and recommend a revised protocol.

22. Multiple Choice Questions with Answers

Question 1

Why is scientific validity an ethical requirement in research?

A. It guarantees that every participant receives a direct benefit.

B. It ensures that participant burdens are connected to a credible prospect of producing useful knowledge.

C. It eliminates all possible research risks.

D. It permits researchers to avoid independent ethical review.

Correct Answer: B. It ensures that participant burdens are connected to a credible prospect of producing useful knowledge.

Question 2

Which action best represents data minimization?

A. Collecting every available variable for possible future analysis.

B. Retaining identifiable records indefinitely for reproducibility.

C. Collecting and retaining only data necessary for the justified research purpose.

D. Sharing raw data publicly before analysis.

Correct Answer: C. Collecting and retaining only data necessary for the justified research purpose.

Question 3

A residual risk is best defined as:

A. A risk that exists only before consent is obtained.

B. A risk remaining after safeguards have been applied.

C. A risk created only by financial sponsors.

D. A risk that cannot affect non-participants.

Correct Answer: B. A risk remaining after safeguards have been applied.

Question 4

Which situation most clearly creates a power-related concern about consent?

A. A researcher invites volunteers through a public advertisement.

B. A university lecturer recruits students and implies that participation may influence grades.

C. A community organization distributes a neutral study notice.

D. A participant contacts a researcher independently.

Correct Answer: B. A university lecturer recruits students and implies that participation may influence grades.

Question 5

Which is the strongest response to a substantial conflict of interest?

A. Mentioning it only in an unpublished internal document.

B. Assuming professional integrity makes safeguards unnecessary.

C. Disclosing it and introducing measures such as independent analysis or recusal.

D. Allowing the sponsor to control publication decisions.

Correct Answer: C. Disclosing it and introducing measures such as independent analysis or recusal.

Question 6

Why does de-identification not guarantee anonymity?

A. De-identified data cannot be analyzed.

B. Identifiers may be reconstructed through linkage or rare attributes.

C. Encryption automatically reverses de-identification.

D. Anonymity applies only to paper records.

Correct Answer: B. Identifiers may be reconstructed through linkage or rare attributes.

Question 7

In the illustrative student-support case, why was high predictive accuracy insufficient to justify automated decisions?

A. Accuracy is never relevant to ethical evaluation.

B. The model could predict disadvantage without identifying fair or effective interventions and could distribute errors unequally.

C. Students are never permitted to receive support based on research.

D. Predictive models cannot be used with historical data.

Correct Answer: B. The model could predict disadvantage without identifying fair or effective interventions and could distribute errors unequally.

Question 8

Which feature is most consistent with ethics-by-design for an automated decision system?

A. Removing all human review to improve efficiency.

B. Using one fairness metric without considering the decision context.

C. Providing monitoring, explanation, appeal, correction, and human override mechanisms.

D. Expanding data collection whenever performance decreases.

Correct Answer: C. Providing monitoring, explanation, appeal, correction, and human override mechanisms.

Question 9

What is the most ethically appropriate reason to compare less intrusive research designs?

A. To avoid producing any scientifically significant result.

B. To select the design capable of answering the question while reducing unnecessary risk.

C. To ensure that the research requires no participant consent.

D. To transfer all responsibility to the technology vendor.

Correct Answer: B. To select the design capable of answering the question while reducing unnecessary risk.

Question 10

Which statement best describes an ethical protocol?

A. A one-time administrative form that remains unchanged after approval.

B. A living operational plan covering design, consent, data governance, risks, monitoring, and response procedures.

C. A contract that protects only the research sponsor.

D. A technical description that excludes social and cultural issues.

Correct Answer: B. A living operational plan covering design, consent, data governance, risks, monitoring, and response procedures.