Notes and Applied Theory Small Data Ethics

By Laini Byfield

Theory-informed program design overview

A structured bridge from theory to policy, governance, and daily practice — using ETHICMAP as the operational backbone.

ContextPolicyGovernancePractice
The translation

How to translate theory into system choices

From contextual integrity

  • Define the context — purpose, norms, and expectations
  • Limit information flows to what is appropriate in that context
  • Publish the “why” and “how” in plain language

From minimisation

  • Reduce fields and retention periods to reduce exposure
  • Use role-based access — treat raw data as sensitive
  • Suppress small-n reporting where re-identification is plausible
The map

Use ETHICMAP as the design map

Most harm in small data is created before anyone looks at the data itself — in the cutoffs, lags, and timing assumptions made at program design.

E + T

Write context and timing assumptions first. Most harm in small data is created by cutoffs, lags, and retroactivity — before anyone has looked at the data itself.

H

Identify who is harmed when errors occur. In small systems, fairness is often about error distribution and visibility — not just aggregate outcomes.

I + C

Align incentive to feasibility and capacity. Avoid coercive participation structures or outcomes that cannot realistically be achieved by the people being measured.

M

Specify measurement with uncertainty: match rates, confidence flags, known failure modes. Point estimates without error bands are not complete specifications.

A + P

Build a runbook and publish learning across cycles. A system without documentation is a system without ethics.

What to produce

Suggested artifacts

  • Program timeline and cutoff map
  • Metric spec sheet — definitions, edge cases, and appealability
  • Data lineage log — source, load date, rule version, and release notes
  • Repair protocol — reprocessing triggers and participant notification guidance
  • Governance charter — decision rights and escalation path