Theory — sortition and representation
Updated: 2026-10-05. Normative claims and empirical findings are labelled separately.
Concept
Sortition (selection by lot) as a democratic selection method, and the representation claims made for randomly selected bodies.
Definitions / competing meanings
- Descriptive representation: the body “mirrors” the population on observable characteristics (age, gender, ethnicity, class) (SRC-0104, SRC-0113).
- Substantive representation: the body acts “for” the represented — interests/values are advocated, not merely mirrored. The two are linked but not identical: mirroring does not guarantee advocacy, and advocates need not mirror (SRC-0113).
- Statistical representativeness: a technical claim that selected membership approximates population parameters within defined margins — achievable only relative to chosen strata and frame, never exhaustively (SRC-0100, SRC-0108).
- Political equality / equal probability of selection: sortition’s distinctive equality claim — every eligible person has an equal (or fairly weighted) chance of selection, unlike electoral competition (SRC-0108, SRC-0077).
Major theoretical arguments
- Normative: sortition realises political equality differently from elections — Manin’s history shows lot was the democratic marker, election the aristocratic one (SRC-0108); Stone frames selection as an egalitarian fair procedure where reasons should not discriminate among qualified candidates (SRC-0109).
- Epistemic: descriptive diversity imports cognitive diversity; diverse groups can outperform expert-homogeneous groups on complex problems (Landemore’s “numbers trump ability” argument) (SRC-0107).
- Empirical: in practice, equal probability of selection is bounded by the opt-in bottleneck — everyone may have equal chance of invitation, not of sitting (see
selection-bias-and-stratification.md; SRC-0038, SRC-0077).
Practical implications
- Representation claims must specify which dimensions are stratified and why — the strata are a political choice (CA UK chose climate attitudes as a stratum, SRC-0038; Newham added occupation, SRC-0048).
- “Representative” is a verifiable property only against declared benchmarks; CA UK’s published composition table is the model (SRC-0038).
Critiques / limitations
- A mirror is not a mandate: descriptive resemblance does not itself confer authority to decide (Lafont’s critique applies where mini-publics bypass public deliberation — SRC-0104).
- Small N limits mirroring: 25–100 members cannot represent all intersecting groups; over-representation of small strata is a deliberate correction (CA UK +5 seats, SRC-0038).
Operationalisation in the corpus
- CA UK: seven declared strata + published benchmark table + attitudinal stratum (SRC-0038).
- Scotland CA: population-representative civic lottery under statutory duty (SRC-0043).
- Panelot/OpenDLP: equal-probability fairness is a computable objective (leximin), not just rhetoric (SRC-0078, SRC-0079).
Cicada relevance
- Cicada records should distinguish descriptive targets (strata), substantive claims (who speaks for whom — usually no one), and equality of selection chance (the fairness objective the algorithm optimises). These are three separate properties commonly collapsed.