CSEG3060 · Research Methodology in Computer Science
Informed consent · confidentiality · fabrication and falsification · authorship norms · conflict of interest
Research rests on trust. Every rule in this lecture exists because trust can be broken at a specific point.
Unit III · Lecture 7 of 8 · 60 minutes · CO3
Dr. Mohsin Furkh Dar · School of Computer Science, UPES
→ / Space next · ← back · O overview · N speaker notes · F full screen
Learning objectives · CO3
Define and apply informed consent in human-subjects research
Explain the mechanisms that preserve confidentiality and protect data
Distinguish fabrication, falsification and legitimate data transformation
Identify criteria for legitimate authorship and the cost of honorary or ghost authorship
Recognise, disclose and manage conflicts of interest
Connect each principle to concrete computer science scenarios
Orientation
| Domain | Core principle | Primary risk if violated | Who is protected |
|---|---|---|---|
| Informed consent | Voluntary, informed, capacitated agreement | Coercion, exploitation | Participants |
| Confidentiality | Privacy protection and controlled disclosure | Privacy breach, identity exposure | Data subjects |
| Fabrication / falsification | Honest data collection and reporting | Loss of scientific validity | Readers and the record |
| Authorship norms | Credit for substantial intellectual contribution | Misattribution of credit | Collaborators |
| Conflict of interest | Transparent disclosure of competing interests | Compromised objectivity | The public and peer review |
CS research now involves human participants (HCI, user studies, AI fairness), large-scale personal datasets, and multi-institution collaborations — each activating a different ethical regime.
Belmont principles · Declaration of Helsinki · ICMR National Ethical Guidelines (2017) · DPDPA 2023 and GDPR · ICMJE and COPE · ORI misconduct definitions
Pillar 1 · part 1
Purpose, duration, procedures, foreseeable risks, expected benefits, compensation.
Language the participant comprehends — free of jargon and undue influence.
Free from coercion or penalty for refusal — a live concern when you recruit classmates or juniors.
Legal and cognitive capacity; surrogate or parental consent for minors and incapacitated individuals.
Normally a signed form; click-through or recorded verbal consent is acceptable in some digital contexts.
A consent form written at postgraduate reading level fails the understanding criterion no matter how complete its disclosure.
Pillar 1 · part 2
Consent must be obtained before task exposure. IP addresses and metadata may themselves constitute identifiable information even when no name is collected.
Reuse beyond the original consent's scope may violate the agreement under which data were collected — however public the dataset now appears.
Many firms experiment on users without explicit consent, raising unresolved concerns about autonomy and transparency. A practice being routine does not settle whether it is ethical.
A usability study of a new authentication mechanism, 100 participants. Each must be told:
Pillar 2 · part 1
Identifiers present. Directly attributable to a person.
Identifiers replaced by keys held separately — still re-identifiable, so not anonymisation under GDPR.
Link destroyed. Cannot be traced back to an individual.
India — DPDPA 2023 · EU — GDPR · US — HIPAA for medical data
Minimisation is the strongest safeguard: data never collected cannot be breached, subpoenaed, re-identified or leaked.
Pillar 2 · part 2
TLS 1.3 in transit, AES-256 at rest. The baseline expectation, not an advanced measure.
Cost: negligible.
Calibrated noise so the presence or absence of any individual cannot be inferred — bounding re-identification risk mathematically rather than by assertion.
Cost: accuracy, against the privacy budget.
Training across devices or institutions without transferring raw data — the architecture used where data cannot lawfully be pooled.
Cost: efficiency; leakage via model updates.
Joint computation over inputs that no party reveals to the others.
Cost: computationally expensive.
Most relevant in health informatics, social-network analysis and behavioural analytics — the subfields where confidentiality obligations are strictest and research questions most demanding.
Pillar 3 · part 1
Making up data, results or records and reporting them as real.
Manipulating materials, equipment or processes, or changing, omitting or selectively reporting data so the research is not accurately represented.
Appropriating another's ideas, processes, results or words without credit — Lectures 4 and 5.
| Practice | Ethical status |
|---|---|
| Removing outliers with documented statistical justification | Generally acceptable |
| Applying anonymisation or aggregation | Acceptable, often required |
| Cherry-picking supportive results while omitting null findings | Unethical — selective reporting |
| Generating synthetic data to fill missing records without disclosure | Unethical — fabrication |
| Re-running an experiment until the result is significant | Unethical — manipulation |
| Image manipulation that alters figure content | Unethical — falsification |
Pillar 3 · part 2
Cherry-picking benchmarks, omitting unfavourable datasets and tuning hyperparameters on the test set are all falsification — notwithstanding that each step looks like ordinary engineering iteration.
Once the test set has informed a design decision, the reported number is a training-set estimate presented as a generalisation estimate — a misrepresentation of what the experiment measured.
Figures in computational biology and imaging can be altered in general-purpose editors — which is why journals now require raw image data alongside submitted figures.
Many published ML results cannot be replicated, substantially because of undisclosed preprocessing and selective reporting rather than outright invention.
Jan Hendrik Schön (2002) — fabricated data across semiconductor studies at Bell Labs.
Hwang Woo-suk (2004–05) — fabricated stem-cell data; retraction and criminal conviction.
Contemporary cases are catalogued by Retraction Watch.
Pillar 4 · part 1
Contributors meeting fewer than four should be acknowledged, not listed.
Criterion 4 is the one students overlook: accepting authorship means accepting answerability for the whole paper, including parts you did not produce — which is why a gift author is exposed, not merely flattered.
| Misconduct | Description |
|---|---|
| Gift authorship | Listing senior colleagues without their meaningful contribution |
| Guest authorship | Adding a well-known researcher to enhance credibility |
| Ghost authorship | Omitting a contributor who made substantive contributions |
| Coercive authorship | Senior authors pressuring juniors to include unrelated contributors |
| Mutual-benefit authorship | Reciprocal name-swapping across papers without genuine contribution |
COPE and ICMJE are the standards journals apply when an authorship dispute is escalated.
Pillar 4 · part 2
CRediT — 14 roles: conceptualisation · methodology · software · validation · formal analysis · investigation · data curation · writing – original draft · writing – review and editing · visualisation · supervision · project administration · funding acquisition · resources
Submissions at venues such as NeurIPS may carry 30 or more authors, at which point a byline conveys almost no information and per-role contribution statements become the only meaningful transparency.
A student who is the primary contributor should be listed as first author. The power asymmetry means the student rarely raises it — so the remedy is a documented authorship agreement before work begins, not a negotiation after submission.
Software, data curation and validation are the roles this cohort most often fills uncredited. Naming them explicitly is how the contribution becomes visible.
Pillar 5 · practical application
Stock, consulting fees, royalties, honoraria from parties who benefit from the outcome.
Evaluating a self-driving system funded by its manufacturer.
Close relationships with individuals or organisations.
Benchmarking against a competing library you maintain.
Pressure from a funder whose product is under evaluation.
Strongly held theoretical commitments that bias interpretation.
Acutely relevant in AI ethics, algorithmic fairness and policy research.
IEEE, ACM, NeurIPS and ICML all require declaration of funding sources, employment, equity interests, advisory memberships and any other relevant relationship within roughly the past three years.
Disclose → blind → independent oversight → recompose the team → recuse. Disclosure is the floor.
Case studies
| Scenario | Domain | Lesson |
|---|---|---|
| A fairness paper reports a 2% parity gap computed on a favourable dataset subset; retracted | Falsification | Selective reporting and benchmark manipulation are falsification |
| A platform runs a large-scale A/B test manipulating emotional valence of feeds; published without consent | Informed consent | Scale does not suspend the consent obligation |
| A company's medical writer drafts a manuscript; only academic authors listed; writer later untraceable | Authorship (ghost) | Ghost authorship destroys accountability; breaches ICMJE |
| A researcher praises an AI hiring tool; stock options in the vendor emerge later | Conflict of interest | Non-disclosure undermines the credibility of the finding |
Retraction · funding blacklist · termination · degree revocation · civil or criminal liability where data-protection law is breached. Confirmed cases are documented publicly by Retraction Watch.
Scenarios are illustrative composites for teaching. The interdependence point: case 2 implicates consent and confidentiality; case 1 implicates falsification and authorship accountability.
Consolidation
A PhD student wants to scrape user reviews from a public website for sentiment analysis. Identify the ethical concerns and propose safeguards.
Three students collaborate; one writes the paper, the other two did coding and analysis, all three are listed. Evaluate against ICMJE.
You are reviewing a paper and notice the lead author co-founded the startup whose product is evaluated. What do you do?
A dataset contains personal data and its terms prohibit re-identification. Discuss the ethical and legal implications.
Exam preparation
Definition → principle → CS example → consequence. Always ground the example in computing — HCI, datasets, benchmarking, AI fairness — rather than medical research. State the institutional and legal ramifications.
Given a scenario: identify the violated domain, name the governing standard, prescribe the remedy.
Further reading: Belmont Report (1979) · Declaration of Helsinki (rev. 2013) · ICMR National Ethical Guidelines (2017) · DPDPA (2023) · GDPR · ICMJE Recommendations · COPE Guidelines · ORI Introduction to RCR · Retraction Watch.