Research / method

Accepted research

Method — stratification ethics and governance

Type: method Updated: 2026-10-05 Scale: cross-jurisdictional Path: research/methods/stratification-and-fairness.md

stratification

Sources: SRC-0031SRC-0032SRC-0038SRC-0048SRC-0078SRC-0101SRC-0109SRC-0122SRC-0123SRC-0124

Method — stratification ethics and governance

Updated: 2026-10-05. Stream D — a priority area. Records competing approaches; does not decide Cicada policy.

Who decides which characteristics count

Criterion types and their governance profile

CriterionUsed in corpusGovernance issues
Geographicalllow controversy; needed for territorial legitimacy
Sex/genderallbinary categories contested; low controversy as a stratum
Agealllow controversy
Socioeconomic/deprivationCA UK (invitation weighting + education stratum)proxy choice is political
EducationCA UK, Brusselsdoubles as class proxy; self-report reliability
EthnicityCA UK, Scotlandspecial-category data under GDPR — Brussels prohibition excludes it there (SRC-0032); UK practice allows with consent
Disabilitynot evidenced as a stratumsupport needs handled via inclusion, not quota (gap)
Political/attitudinalCA UK climate-attitude stratum; Brussels permits topic-linked “political (e.g. vote/abstention)” criteria subject to limits (SRC-0031)most contested — see below
BehaviouralBrussels permits topic-linked (SRC-0031)measurement/consent issues

The core tensions (recorded, not resolved)

  1. Statistical representativeness vs equal probability: every quota reduces the chance-structure of pure lot; every unstratified draw tolerates achieved imbalance. Leximin methods manage the trade-off within the respondent pool but cannot remove it (SRC-0109, SRC-0078).
  2. Attitudinal stratification — legitimacy dispute: for (CA UK rationale): prevents a climate-sceptic-free room, mirrors population opinion (SRC-0038). Against: opinions are not demographics — quotas on belief risk engineering the outcome and are manipulable (state the remit, measure attitude, select the answer); self-reported and changeable (SRC-0124 sceptical tradition; contested — no consensus in literature).
  3. Quotas as politically determinative: because strata + weights decide who is in the room, the design of the strata table is the least-scrutinised consequential decision in the pipeline; Brussels’ rules-based criteria + institutional ratification is the corpus’s best-governed answer (SRC-0031).
  4. Data protection: special-category strata (ethnicity, health, political opinion, religion) require explicit legal basis/consent — Brussels solves by prohibition, UK by consent-based collection (SRC-0032, SRC-0038, SRC-0122).
  5. Small-cell privacy: intersecting strata at N≈100 make individuals identifiable — publication of composition should be at marginal level, not cell level (SRC-0122; practice implication).
  6. Manipulation surface: whoever chooses strata can shape the room — transparency of the choice (published criteria + rationale, per SRC-0101 transparency principle) is the main defence.

Corpus comparators (required)

Cicada relevance

Sources