Knowing and Knowing Well

Nearly 700,000 people

Facebook and Cornell, published June 2014

• For one week in January 2012, Facebook manipulated users’ News Feeds
• Some saw fewer positive posts from friends. Others saw fewer negative posts.
• Researchers then measured what those users posted themselves
• It worked: less positive input produced slightly more negative output
• Published in PNAS as evidence of massive-scale emotional contagion

The reaction

• Nearly 700,000 people were enrolled with no informed consent
• No one was told. No one could decline.
• Facebook argued its terms of service authorized “research”
• Researchers argued the effect was tiny and the knowledge valuable
• PNAS appended an editorial expression of concern, an unusual step

Ethics is a disposition, not a form

• It is tempting to treat ethics as a training module between you and the data
• It shapes how you think about your relationship to the people you study
• It governs how honestly you report what you find
• The protections exist because researchers caused real harm, systematically
• Reasonable people still disagree about the Facebook study. That is the point.

Tuskegee, 1932 to 1972

• The U.S. Public Health Service studied untreated syphilis in 399 Black men
• The men were told they were receiving free treatment for “bad blood”
• Penicillin became standard in the 1940s. The researchers withheld it.
• 28 died directly of syphilis, 100 of complications, 40 wives were infected, 19 children were born with congenital syphilis
• Federally funded, peer-reviewed, and unchallenged for forty years

Milgram and Humphreys

• Milgram, 1961: participants believed they were delivering painful shocks because an authority figure told them to. Most complied.
• The findings were foundational. The distress was severe and sometimes lasting.
• Humphreys, 1970: recorded license plates at public restrooms, traced home addresses through police records, visited in disguise as a health surveyor
• Exposure would have destroyed marriages and careers

The common thread

• Every one of these researchers believed the work served a greater good
• Every one produced participants who were harmed, deceived, or exploited
• Good intentions are not sufficient protection
• Researchers need external accountability and formal principles
• Tuskegee’s exposure in 1972 produced the framework we use today

Belmont: respect for persons

National Commission, 1979

• People are autonomous agents capable of making their own decisions
• Informed consent: understand the research, the risks, and the right to withdraw without penalty
• Protection of vulnerable populations: children, prisoners, people with cognitive impairments, whose voluntary consent is compromised
• Public social media complicates this, and we come back to it

Belmont: beneficence and justice

• Beneficence, do no harm: no physical, psychological, social, or economic injury to participants
• And maximize the benefit-to-risk ratio: the knowledge must justify the risk
• Justice: benefits and burdens distributed fairly across groups
• Tuskegee violated justice directly: all risk on a marginalized population, the knowledge accruing to everyone else

Institutional review

• An IRB reviews a study before data collection: risk minimization, equitable selection, consent, privacy and confidentiality
• Exempt: minimal risk, in categories federal regulation names. Most content analysis of public material qualifies.
• Expedited: no more than minimal risk but not exempt. Many surveys and interview studies with adults.
• Full board: more than minimal risk, vulnerable populations, or deception
• The IRB makes the determination, not you. You submit. The board decides.

The Twitch data through Belmont

• Collected from Twitch’s public IRC interface and public API
• No IP addresses, no emails, no user IDs, no payment data, no whispers
• Usernames are present but pseudonymous: chosen handles, not legal names
• Respect: the case rests on public broadcast, not on consent
• Beneficence: minimal risk. Justice: no group bears disproportionate burden

Most IRBs classify this as exempt.

Public does not mean fair game

• AoIR has long argued that public accessibility is not automatic permission (Markham & Buchanan, 2012)
• So: do not aggregate in ways that re-identify individual users
• Do not republish entire chat logs verbatim
• Do not amplify messages whose senders expected obscurity more than broadcast
• The cleanness of this dataset is a starting point, not a license

Three contrast cases

Same research question, different data, different determination:

• Twitch whispers. Private messages, reasonable expectation of privacy. Expedited or full review.
• A private Discord server, same people. Closed membership, stronger privacy expectation. Expedited or full.
• An internal moderation log with platform flags on named users. Risk of harm to identified individuals. Full board likely.

Consent gets complicated

• Deception is sometimes justified, and always paired with mandatory debriefing plus the option to withdraw your data
• Passive consent, the terms-of-service move, is generally insufficient. Ethicists prefer active consent: an affirmative act.
• Secondary analysis of someone else’s data raises whether the original consent covered your use. Generally acceptable when de-identified.
• This last one is where research ethics is still actively evolving

Honest reporting is an ethical obligation

• Fabrication: inventing data. Falsification: manipulating results. Both are fraud and both are career-ending.
• Selective reporting: running many analyses, publishing only the ones that worked. This distorts the literature.
• HARKing: Hypothesizing After the Results are Known, then writing as though you predicted it. A postdiction carries none of a prediction’s weight.
• “Our results suggest” quietly becomes “our results demonstrate.” Watch for it.

Before next time

• Read Chapter 4, Intelligence Gathering, before Thursday
• Thursday we have a librarian visiting to run database searching with us. Bring a laptop and a topic you are curious about.
• Bring one thing you want to be able to find in the literature
• CITI Ethics Certification is coming. Start it early, it takes real hours.
• Journal entry on Chapter 3 closes the week