4. Human Participants, Consent, and Digital Research

Introduction and Context

Research involving human participants occupies a distinctive position in information technology. A study may appear technically modest—observing online interaction, testing an interface, analysing digital traces, or interviewing software users—yet it can expose identities, preferences, vulnerabilities, social relationships, and behavioural patterns. The ethical difficulty is not limited to medical or psychological experimentation. Whenever researchers collect information from, about, or in interaction with people, questions arise concerning autonomy, privacy, risk, fairness, accountability, and the legitimate use of data.

Digital research complicates these questions because the boundaries between public and private information are unstable. A message posted in an open forum may be technically accessible to anyone but still be experienced by its author as part of a bounded community. A dataset may be anonymised at collection and later become identifiable when linked with other sources. A consent form may be signed once, while the research project evolves through secondary analysis, algorithmic processing, and data sharing. Human–computer interaction studies introduce additional concerns: participants may be deceived by experimental interfaces, harmed by stressful tasks, or influenced by researchers who control access to technology, employment, education, or services.

Ethically defensible research therefore requires more than obtaining a signature. It requires a continuing relationship of explanation, choice, protection, and accountability. This chapter examines how doctoral researchers can design recruitment, informed and dynamic consent, deception, debriefing, compensation, safeguards for vulnerable participants, online observation, social-media research, and human–computer interaction studies. It also considers governance, data protection, emerging technologies, and the relationship between formal approval and genuine ethical practice.

Learning Objectives

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

1. Explain autonomy, informed consent, privacy, confidentiality, anonymity, vulnerability, and minimal risk in digital research.

2. Design recruitment procedures that are fair, transparent, and resistant to undue influence.

3. Develop proportionate consent and dynamic-consent processes for changing digital studies and secondary data use.

4. Evaluate the ethical acceptability of deception, online observation, social-media research, and HCI experiments.

5. Identify risks to vulnerable participants and construct appropriate safeguards.

6. Distinguish compensation from coercive inducement and design ethically defensible payment schemes.

7. Apply ethical review, data governance, debriefing, and incident-response practices to research projects.

8. Critically assess how artificial intelligence, automated profiling, and re-identification affect participant protection.

From Human Subjects to Data Relationships

The phrase “human participant” includes people who provide information, perform tasks, interact with a system, contribute biological or behavioural data, or are represented in records analysed by researchers. In digital research, the relevant person may not be the person who directly interacts with the investigator. For example, a study of a social-media platform may affect both recruited users and individuals mentioned in their posts. A machine-learning project may use historical service records concerning people who never agreed to research use.

Four ethical ideas provide a foundation. Respect for persons requires recognition of individuals as autonomous agents and protection for those whose autonomy is limited. Beneficence requires researchers to seek valuable knowledge while reducing foreseeable harm. Justice concerns the fair distribution of research burdens and benefits. Respect for privacy and confidentiality requires careful control over what is collected, who can access it, how it is used, and whether people can be identified.

These principles must be interpreted rather than applied mechanically. A study may involve no physical intervention but still create serious informational or reputational harm. Conversely, a study involving a minor inconvenience may be ethically acceptable if the social value is substantial, risks are low, and protections are strong. Ethical review is therefore a process of reasoned proportionality: the more intrusive, uncertain, or consequential the research, the stronger the justification and safeguards required.

Privacy is not identical to secrecy. Privacy concerns a person’s ability to control access to themselves, their information, or their social space. Confidentiality concerns the researcher’s obligation to protect information entrusted to the research team. Anonymity means that identity is not known or cannot reasonably be linked to the data; pseudonymity replaces direct identifiers with codes but does not eliminate the possibility of re-identification. Digital datasets are often pseudonymous rather than anonymous.

Recruitment as an Ethical Design Problem

Recruitment determines who enters a study, who is excluded, and how potential participants understand the invitation. A defensible recruitment plan specifies the population, inclusion and exclusion criteria, recruitment channels, initial contact, screening procedures, and safeguards against pressure. Selection should be scientifically justified rather than based merely on convenience, availability, or the researcher’s ease of access.

Researchers should avoid exploiting institutional relationships. A lecturer recruiting current students, a manager recruiting employees, or a clinician recruiting patients may unintentionally create perceived obligation. Where such recruitment is necessary, an independent recruiter, anonymous expression of interest, or separation between participation and institutional services can reduce pressure. Potential participants should be told clearly that refusal will not affect grades, employment, healthcare, immigration status, benefits, or access to technology.

Recruitment materials should state the purpose in accurate but accessible language, the activities involved, expected duration, foreseeable risks, eligibility requirements, compensation, contact details, and the voluntary nature of participation. Researchers should not conceal material information merely to increase enrolment. Targeted recruitment can be appropriate when a population is underrepresented, but researchers should avoid framing a group as deficient, abnormal, or merely useful for data extraction.

Digital recruitment creates additional risks. Online advertisements may use behavioural targeting that identifies people with sensitive interests. Automated screening can exclude people through biased classifications. Snowball recruitment may expose a participant’s association with a sensitive study to their acquaintances. Recruitment through online communities should respect community norms and, where appropriate, involve moderators without treating moderator permission as a substitute for individual consent.

Informed, Ongoing, and Dynamic Consent

Informed consent is a communication process through which a competent person voluntarily agrees to participate after receiving relevant information and having a reasonable opportunity to ask questions. A signature or electronic click is evidence of the process, not its entire substance. Information should normally include the study purpose, procedures, duration, risks, benefits, alternatives where relevant, privacy arrangements, compensation, data retention, future use, withdrawal procedures, contacts for questions or complaints, and the limits of confidentiality.

Consent must be understandable. Technical vocabulary, excessive legal language, and long forms can undermine comprehension. Layered consent is often useful: a brief explanation presents essential facts, while expandable sections provide detail about data flows, security, secondary use, and governance. Comprehension checks may be appropriate for high-risk studies. Participants should know which activities are required, which are optional, whether recordings are made, and whether their data may be combined with external sources.

Consent is voluntary only when refusal is realistically possible. Undue influence may arise from excessive payment, authority relationships, dependency, urgent financial need, or misleading claims. Consent can also be invalid if participants lack decision-making capacity, if information is materially incomplete, or if withdrawal is presented as practically impossible without justification.

Digital studies often require dynamic consent, a continuing process through which participants can review and modify choices as data uses evolve. A participant might separately consent to the primary study, audio recording, data linkage, future research, controlled data sharing, commercial use, or receipt of incidental findings. A secure portal can display current permissions, notify participants of material changes, and record decisions. Dynamic consent is not automatically superior: frequent requests may cause fatigue, and digital interfaces may privilege technically confident participants. It should therefore be proportionate, accessible, and supported by non-digital alternatives.

Withdrawal requires careful explanation. Participants should know whether they can withdraw their identifiable data, whether already anonymised or aggregated results can be removed, and what happens to data already shared with collaborators. Researchers should avoid promising deletion where technical or legal constraints make it impossible. A transparent limitation is ethically preferable to an attractive but inaccurate guarantee.

Deception, Debriefing, and the Integrity of Experimental Research

Deception occurs when researchers deliberately provide false information or withhold information that might influence participation or behaviour. In HCI and behavioural research, deception may be used to prevent demand characteristics—for example, to avoid telling participants that a system is measuring error concealment or trust calibration. Deception is ethically controversial because it interferes with autonomous decision-making and may damage trust.

A proportionate justification should address four questions: Is the research question important? Is deception scientifically necessary rather than merely convenient? Is there a non-deceptive alternative? Are the risks low and reversible? Deception should not conceal risks that a reasonable person would consider important, nor should it be used to expose participants to humiliation, discrimination, or significant psychological distress.

Debriefing should occur as soon as scientifically feasible. It should explain what was concealed, why deception was used, what data were collected, how risks were managed, and how participants may raise concerns or request withdrawal where possible. A meaningful debrief is a dialogue, not a scripted apology. Researchers should assess whether the explanation has caused distress, correct false beliefs, provide support information when needed, and report any unexpected adverse reactions. In some studies, delayed debriefing may be necessary to preserve validity, but the delay requires explicit justification and oversight.

Compensation, Vulnerability, and Fair Participation

Compensation recognizes time, inconvenience, travel, effort, and expenses. It should normally be calibrated to the burden of participation rather than to the risk of the study. Payment must not be so high that people accept risks they would otherwise reject, particularly where participants are economically disadvantaged. Pro-rata payment is often fairer than paying only upon completion, because it does not pressure participants to continue after discomfort or to conceal problems.

Vulnerability is contextual and multidimensional. It may involve age, cognitive or communicative disability, illness, institutional dependency, poverty, language barriers, legal insecurity, social marginalisation, or limited digital literacy. A person should not be labelled permanently vulnerable; vulnerability may arise from the study setting itself. For example, an online platform’s terms may be incomprehensible to a person with limited literacy, while a workplace hierarchy may constrain an otherwise autonomous employee.

Protection should be specific rather than paternalistic. Possible measures include adapted information sheets, interpreters, accessible interfaces, legally authorised representatives where appropriate, assent from minors alongside permission from a responsible adult, independent advocates, additional comprehension checks, private participation spaces, and stopping rules. Researchers should include vulnerable groups when their exclusion would produce unjust knowledge gaps, but inclusion must be accompanied by credible safeguards.

Online Observation and Social-Media Research

The ethical status of online material cannot be determined solely by whether it is publicly accessible. Researchers should consider the reasonable expectations of the community, the sensitivity of the content, the likelihood of harm from exposure, the size and accessibility of the forum, the vulnerability of authors, and whether quotations can be searched to identify individuals. A public post about political preferences may still be sensitive if reproduced in a searchable publication.

Ethically responsible online observation begins with a clear distinction between observation, interaction, intervention, and data extraction. Passive observation of a genuinely open, non-sensitive environment may sometimes present low risk. Entering a restricted group, collecting posts at scale, contacting users, or manipulating visibility changes the ethical profile. Community or moderator permission may be valuable, but it does not necessarily replace individual consent, especially when data are sensitive or identifiable.

Researchers should minimise collection, avoid unnecessary usernames and images, paraphrase quotations where exact wording creates searchability, remove metadata, and consider whether direct publication is justified. Scraping technically accessible data may violate platform rules, user expectations, or privacy interests even when it is not technically difficult. Terms of service are not a complete ethical framework, but they are relevant to responsible governance.

Social-media research also raises issues of consent by association. A participant may post a photograph or conversation involving other people who have not agreed to research use. Researchers should consider whether third parties can be identified and whether their interests require redaction, paraphrase, or exclusion. Automated sentiment analysis and language models can infer health, political, or psychological attributes that users never explicitly disclosed. Inference is a form of data processing with ethical consequences; it should not be treated as harmless because the output is probabilistic.

Human–Computer Interaction Studies

HCI research ranges from usability testing and accessibility evaluation to persuasive interfaces, biometric authentication, immersive environments, recommender systems, and adaptive AI. Ethical review should examine both the participant’s direct experience and the system’s possible downstream effects.

A usability test may involve screen recording, keystroke logging, eye tracking, facial analysis, or collection of interaction histories. Each instrument should have a stated purpose, retention period, access control, and deletion or anonymisation plan. Participants should not be surprised by sensors or recordings. If an interface displays personalised or emotionally manipulative content, researchers should assess whether the intervention creates distress, loss of agency, or unfair treatment.

HCI studies involving children, older adults, people with disabilities, or individuals with low digital literacy require accessible design and realistic testing conditions. Accessibility is not only an outcome variable; it is an ethical requirement of the research process. Interfaces used for consent, withdrawal, and debriefing should support assistive technologies and alternative communication methods.

Researchers should also distinguish laboratory performance from real-world impact. A prototype that improves task completion may increase surveillance or shift decision-making power to an opaque algorithm. Ethical HCI therefore includes questions of contestability, explainability, accessibility, inclusion, environmental burden, and institutional accountability—not merely efficiency or user satisfaction.

A Framework for Ethical Decision-Making

A useful doctoral-level approach is to evaluate a project through six connected questions:

1. Purpose: What knowledge or social value justifies involving people?

2. Necessity: Is human participation, identifiable data, deception, or intrusive measurement genuinely necessary?

3. Risk: What physical, psychological, social, economic, legal, privacy, and group-level harms are foreseeable?

4. Agency: Can people understand the study, refuse freely, withdraw meaningfully, and influence data-use choices?

5. Justice: Who bears the burdens, who benefits, and who is excluded from the knowledge produced?

6. Accountability: Who monitors the study, responds to incidents, controls data, and answers complaints?

This framework should be applied iteratively. Risks may change after recruitment, when a new dataset is linked, when a model is retrained, or when findings are prepared for publication. A data protection impact assessment, threat model, protocol for adverse events, and version-controlled consent record can support the process. Formal approval by an institutional review board or research ethics committee is necessary in many settings, but approval does not transfer moral responsibility from the research team.

Case Study: An Illustrative Study of an AI Well-Being Assistant

Background

A university research group develops an AI chatbot intended to help postgraduate students identify academic stress and locate support services. The proposed study recruits students through university email and social-media advertisements. Participants will use the chatbot for four weeks. The system records conversations, generates stress scores, and adapts its prompts. The research team proposes a single electronic consent form and offers a large completion payment.

Situation and Ethical Problems

Several risks are immediately visible. Students may believe participation affects academic support or that the chatbot is a clinical service. Conversations may contain highly sensitive information. The model may produce inaccurate or harmful advice. Continuous adaptation means that the initial consent form may not adequately explain later uses. Recruitment through faculty staff may create pressure. The large payment may be particularly influential for financially constrained students. Finally, stress scores could be re-identified and used by the university in ways participants did not expect.

Analysis and Approach

The team redesigns recruitment so that an independent research office sends invitations and participation has no relationship to academic assessment or support eligibility. The payment is divided into modest instalments for completed sessions, with compensation for partial participation and expenses. The information sheet states clearly that the chatbot is experimental, is not an emergency or clinical service, and should not replace professional help.

Consent is layered and dynamic. Participants separately choose whether to permit conversation recording, model improvement, linkage with demographic information, and future secondary research. A dashboard allows them to review permissions and request deletion of identifiable records. The study collects only information necessary for the research question, separates contact details from conversation data, limits access, and establishes a short retention period for raw data.

The protocol includes crisis-response procedures, prominent links to human support, a stop function, safety testing before deployment, and human review of serious incidents. The team conducts accessibility testing and offers a non-chatbot comparison activity. The ethics committee requires a pilot with close monitoring before full recruitment.

Outcome

The revised study produces less data than originally planned but generates more defensible evidence. Several pilot participants report that the chatbot’s confident tone made it appear clinically authoritative, leading the team to revise language and interface cues. The dynamic-consent process reveals that many participants want model-improvement use but not identifiable quotation in publications. These preferences shape the data-management plan and publication strategy.

Learning Points

The case illustrates that ethical design can improve scientific quality. Clear boundaries reduce expectancy effects, safety procedures protect participants, and data minimisation reduces confounding privacy risks. It also shows that consent, interface design, payment, recruitment, and governance cannot be treated as separate administrative tasks. In adaptive AI research, ethical review must continue throughout the system lifecycle.

Governance, Data Protection, and Emerging Technologies

Research teams should assign responsibility for data stewardship, access approval, security monitoring, breach response, retention, deletion, and communication with participants. Technical controls may include encryption in transit and at rest, role-based access, multifactor authentication, audit logs, environment separation, secure backups, and controlled data enclaves. These controls reduce risk but do not make ethically unjustified collection acceptable.

Anonymisation should be tested against realistic linkage threats, including publicly available information, distinctive writing styles, timestamps, geolocation, and model-based inference. Small groups and rare attributes are especially difficult to protect. Differential privacy, federated analysis, synthetic data, and secure multiparty computation may reduce exposure, but each introduces trade-offs in utility, transparency, cost, and interpretability.

Generative AI creates new consent questions. Researchers should determine whether participant data are used for training, prompting, evaluation, retrieval, or fine-tuning; whether external providers retain inputs; and whether generated outputs can reproduce identifiable material. Participants should not be promised that a model can forget information unless the technical and organisational process genuinely supports that claim. Automated transcription and coding also require checks for bias, hallucination, accent discrimination, and inappropriate disclosure.

Current ethical practice increasingly emphasises data sovereignty, community governance, algorithmic impact assessment, participatory design, and researcher reflexivity. These developments challenge the assumption that individual consent alone resolves group-level harms. A dataset may affect a community’s representation, cultural identity, or political standing even when every directly identifiable individual has consented.

Common Failures and Best Practices

Frequent failures include treating a click-through form as sufficient consent, collecting more data “just in case,” describing public data as automatically ethical to use, using payment to conceal burdens, recruiting subordinates without safeguards, publishing searchable quotations, failing to plan for withdrawal, and assuming that anonymisation is permanent. Another failure is ethical compartmentalisation: researchers leave privacy to the security team, accessibility to the interface designer, and participant welfare to the ethics committee.

Strong practice begins early in study design. Researchers should map data flows, conduct a risk assessment, consult relevant communities, pilot consent materials, justify every data field, test the study with representative users, document decisions, monitor adverse events, and revise the protocol when assumptions change. They should maintain a plain-language participant record, provide accessible contact routes, report deviations honestly, and disseminate findings without exposing individuals or stigmatising groups.

Chapter Summary

Ethical digital research treats participants as persons and communities rather than as sources of data. Recruitment must be fair and free from undue influence. Informed consent is a continuing process involving comprehension, voluntariness, realistic withdrawal, and transparent data use; dynamic consent can support evolving projects but must remain accessible and proportionate. Deception requires necessity, low risk, careful justification, and meaningful debriefing. Compensation should recognise burden without becoming coercive, while vulnerability requires contextual safeguards rather than exclusion by default.

Online observation and social-media research demand attention to expectations, community boundaries, identifiability, third-party interests, and the difference between technical accessibility and ethical permission. HCI studies must examine recording, accessibility, manipulation, algorithmic effects, and downstream consequences. Data minimisation, security, accountability, impact assessment, and lifecycle oversight are essential, particularly in AI research. Ethical approval is a foundation, not a substitute for continuing judgement.

Key Terms

Autonomy: The capacity and right of a person to make voluntary, informed decisions.

Beneficence: The ethical obligation to promote benefits and reduce foreseeable harm.

Confidentiality: The duty to protect information entrusted to the research team.

Dynamic consent: An ongoing consent process allowing participants to review and modify data-use choices.

Anonymity: A condition in which participant identity cannot reasonably be linked to the data.

Pseudonymity: Use of codes or substitutes for direct identifiers while retaining possible re-identification.

Informed consent: Voluntary agreement based on relevant, understandable information.

Deception: Deliberate withholding or distortion of information in a research study.

Debriefing: Post-study explanation of procedures, deception, risks, and participant options.

Undue influence: Excessive pressure or incentives that impair genuinely voluntary choice.

Vulnerability: A contextual condition that limits autonomy or increases exposure to harm.

Compensation: Payment or reimbursement recognising participation time, effort, inconvenience, or expenses.

Privacy: Appropriate control over access to a person, activity, or information.

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

Re-identification: Linking supposedly de-identified data to an identifiable person.

Human–computer interaction: Study and design of interactions between people and computational systems.

Social-media research: Systematic study of content, behaviour, or interaction on social platforms.

Ethics review: Formal assessment of a proposed study’s ethical acceptability and safeguards.

Review Questions

Short Answer Questions

1. Distinguish privacy, confidentiality, anonymity, and pseudonymity.

2. Why is a signed consent form not by itself sufficient to establish informed consent?

3. What is dynamic consent, and why might it be useful in digital research?

4. Identify two safeguards for recruiting participants who occupy a subordinate institutional position.

5. Why is publicly accessible online information not automatically ethically unrestricted research data?

Long Answer Questions

1. Explain how autonomy, beneficence, justice, and privacy interact in the design of a digital research study.

2. Discuss the ethical conditions under which deception may be justified and analyse the role of debriefing.

3. Design a consent and data-governance plan for an HCI study involving screen recordings, eye tracking, and adaptive AI.

4. Analyse the ethical risks of social-media scraping, including community expectations, third-party information, and re-identification.

5. Explain how researchers can protect vulnerable participants without unjustifiably excluding them from research.

Analytical / Application Questions

1. A researcher wants to quote verbatim posts from a small private online group after obtaining moderator permission. Evaluate whether individual consent, paraphrasing, or exclusion of quotations is most defensible, and justify your decision.

2. An experimental chatbot withholds its true research purpose to measure trust. Construct a decision analysis addressing necessity, risk, alternatives, debriefing, and withdrawal.

3. A university plans to link student wellbeing-chat records with academic performance data for machine-learning research. Identify the ethical problems and propose a proportionate redesign.

22. Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes confidentiality from anonymity?

A. Confidentiality means data are deleted immediately, while anonymity means data are encrypted.

B. Confidentiality concerns protecting entrusted information, while anonymity means identity cannot reasonably be linked to data.

C. Confidentiality applies only to medical research, while anonymity applies only to online research.

D. Confidentiality requires public disclosure, while anonymity requires moderator approval.

Correct Answer: B. Confidentiality concerns protecting entrusted information, while anonymity means identity cannot reasonably be linked to data.

Question 2

What is the strongest ethical reason to use dynamic consent in an evolving digital study?

A. It eliminates the need for ethics review.

B. It allows researchers to collect unlimited data without further explanation.

C. It enables participants to review and modify choices as data uses change.

D. It guarantees that all participants will understand technical systems.

Correct Answer: C. It enables participants to review and modify choices as data uses change.

Question 3

A lecturer wishes to recruit current students for a study. Which safeguard most directly reduces undue influence?

A. Offer a very large completion payment.

B. Recruit through an independent office and separate participation from grading.

C. Tell students that participation is expected but technically voluntary.

D. Allow the lecturer to know who declined.

Correct Answer: B. Recruit through an independent office and separate participation from grading.

Question 4

Which condition is most important when evaluating proposed deception?

A. Whether deception makes recruitment faster.

B. Whether participants are unlikely to notice it.

C. Whether it is scientifically necessary and no lower-risk alternative exists.

D. Whether the study uses an online platform.

Correct Answer: C. Whether it is scientifically necessary and no lower-risk alternative exists.

Question 5

Why may a verbatim quotation from a public social-media post still create privacy risk?

A. Public posts cannot be analysed scientifically.

B. Search engines may enable identification of the author or community.

C. Researchers are prohibited from studying written language.

D. A quotation automatically becomes anonymous when published.

Correct Answer: B. Search engines may enable identification of the author or community.

Question 6

Which payment arrangement is generally most protective when participants may experience discomfort during a study?

A. Payment only if the participant completes every procedure.

B. A penalty for withdrawing before completion.

C. Pro-rata payment for completed participation, with reasonable expense reimbursement.

D. Payment determined by the participant’s personal income.

Correct Answer: C. Pro-rata payment for completed participation, with reasonable expense reimbursement.

Question 7

Which action best reflects data minimisation?

A. Collecting all potentially useful identifiers for future convenience.

B. Retaining raw recordings indefinitely in case another project needs them.

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

D. Publishing the full dataset so participants can verify it.

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

Question 8

In the illustrative AI well-being assistant case, why was the system’s clinical status made explicit?

A. To increase the chatbot’s authority.

B. To prevent participants from mistaking an experimental system for professional care.

C. To avoid providing any support information.

D. To justify collecting more sensitive conversations.

Correct Answer: B. To prevent participants from mistaking an experimental system for professional care.

Question 9

Which statement about vulnerability is most ethically defensible?

A. Vulnerable groups should always be excluded from digital research.

B. Vulnerability is fixed and applies identically in every research setting.

C. Vulnerability is contextual and should prompt specific safeguards rather than automatic exclusion.

D. Vulnerability is relevant only when participants are minors.

Correct Answer: C. Vulnerability is contextual and should prompt specific safeguards rather than automatic exclusion.

Question 10

Why is pseudonymisation insufficient as a complete privacy guarantee?

A. Pseudonymous data cannot be used for research.

B. Codes may be linked with other information to re-identify individuals.

C. Pseudonymisation requires publishing names.

D. Pseudonymous data are always more identifiable than direct identifiers.

Correct Answer: B. Codes may be linked with other information to re-identify individuals.